lbcwallet/wallet/wallet.go

3699 lines
113 KiB
Go
Raw Normal View History

// Copyright (c) 2013-2017 The btcsuite developers
// Copyright (c) 2015-2016 The Decred developers
// Use of this source code is governed by an ISC
// license that can be found in the LICENSE file.
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
package wallet
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
import (
"bytes"
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
"encoding/hex"
"errors"
"fmt"
"sort"
"strings"
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
"sync"
"time"
"github.com/btcsuite/btcd/blockchain"
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
"github.com/btcsuite/btcd/btcec"
"github.com/btcsuite/btcd/btcjson"
"github.com/btcsuite/btcd/chaincfg"
"github.com/btcsuite/btcd/chaincfg/chainhash"
"github.com/btcsuite/btcd/rpcclient"
2015-04-06 21:03:24 +02:00
"github.com/btcsuite/btcd/txscript"
"github.com/btcsuite/btcd/wire"
"github.com/btcsuite/btcutil"
"github.com/btcsuite/btcutil/hdkeychain"
"github.com/btcsuite/btcwallet/chain"
"github.com/btcsuite/btcwallet/waddrmgr"
Refactor wallet transaction creation code. This began as a change to improve the fee calculation code and evolved into a much larger refactor which improves the readability and modularity of all of the transaction creation code. Transaction fee calculations have been switched from full increments of the relay fee to a proportion based on the transaction size. This means that for a relay fee of 1e3 satoshis/kB, a 500 byte transaction is only required to pay a 5e2 satoshi fee and a 1500 byte transaction only need pay a 1.5e3 fee. The previous code would end up estimating these fees to be 1e3 and 2e3 respectively. Because the previous code would add more fee than needed in almost every case, the transaction size estimations were optimistic (best/smallest case) and signing was done in a loop where the fee was incremented by the relay fee again each time the actual size of the signed transaction rendered the fee too low. This has switched to using worst case transaction size estimates rather than best case, and signing is only performed once. Transaction input signature creation has switched from using txscript.SignatureScript to txscript.SignTxOutput. The new API is able to redeem outputs other than just P2PKH, so the previous restrictions about P2SH outputs being unspendable (except through the signrawtransaction RPC) no longer hold. Several new public packages have been added: wallet/txauthor - transaction authoring and signing wallet/txfees - fee estimations and change output inclusion wallet/txrules - simple consensus and mempool policy rule checks Along with some internal packages: wallet/internal/txsizes - transaction size estimation internal/helpers - context free convenience functions The txsizes package is internal as the estimations it provides are specific for the algorithms used by these new packages.
2016-02-28 05:30:56 +01:00
"github.com/btcsuite/btcwallet/wallet/txauthor"
"github.com/btcsuite/btcwallet/wallet/txrules"
"github.com/btcsuite/btcwallet/walletdb"
"github.com/btcsuite/btcwallet/walletdb/migration"
2015-04-06 21:03:24 +02:00
"github.com/btcsuite/btcwallet/wtxmgr"
"github.com/davecgh/go-spew/spew"
)
const (
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
// InsecurePubPassphrase is the default outer encryption passphrase used
// for public data (everything but private keys). Using a non-default
// public passphrase can prevent an attacker without the public
// passphrase from discovering all past and future wallet addresses if
// they gain access to the wallet database.
//
// NOTE: at time of writing, public encryption only applies to public
// data in the waddrmgr namespace. Transactions are not yet encrypted.
InsecurePubPassphrase = "public"
walletDbWatchingOnlyName = "wowallet.db"
// recoveryBatchSize is the default number of blocks that will be
// scanned successively by the recovery manager, in the event that the
// wallet is started in recovery mode.
recoveryBatchSize = 2000
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
)
var (
// ErrNotSynced describes an error where an operation cannot complete
// due wallet being out of sync (and perhaps currently syncing with)
// the remote chain server.
ErrNotSynced = errors.New("wallet is not synchronized with the chain server")
// ErrWalletShuttingDown is an error returned when we attempt to make a
// request to the wallet but it is in the process of or has already shut
// down.
ErrWalletShuttingDown = errors.New("wallet shutting down")
// Namespace bucket keys.
waddrmgrNamespaceKey = []byte("waddrmgr")
2015-04-06 21:03:24 +02:00
wtxmgrNamespaceKey = []byte("wtxmgr")
)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Wallet is a structure containing all the components for a
// complete wallet. It contains the Armory-style key store
// addresses and keys),
type Wallet struct {
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
publicPassphrase []byte
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Data stores
db walletdb.DB
Manager *waddrmgr.Manager
2015-04-06 21:03:24 +02:00
TxStore *wtxmgr.Store
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
chainClient chain.Interface
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
chainClientLock sync.Mutex
chainClientSynced bool
chainClientSyncMtx sync.Mutex
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
lockedOutpoints map[wire.OutPoint]struct{}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
recoveryWindow uint32
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Channels for rescan processing. Requests are added and merged with
// any waiting requests, before being sent to another goroutine to
// call the rescan RPC.
rescanAddJob chan *RescanJob
rescanBatch chan *rescanBatch
rescanNotifications chan interface{} // From chain server
rescanProgress chan *RescanProgressMsg
rescanFinished chan *RescanFinishedMsg
// Channel for transaction creation requests.
createTxRequests chan createTxRequest
// Channels for the manager locker.
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
unlockRequests chan unlockRequest
lockRequests chan struct{}
holdUnlockRequests chan chan heldUnlock
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
lockState chan bool
changePassphrase chan changePassphraseRequest
changePassphrases chan changePassphrasesRequest
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Information for reorganization handling.
reorganizingLock sync.Mutex
reorganizeToHash chainhash.Hash
reorganizing bool
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
NtfnServer *NotificationServer
chainParams *chaincfg.Params
wg sync.WaitGroup
started bool
quit chan struct{}
quitMu sync.Mutex
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// Start starts the goroutines necessary to manage a wallet.
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
func (w *Wallet) Start() {
w.quitMu.Lock()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
select {
case <-w.quit:
// Restart the wallet goroutines after shutdown finishes.
w.WaitForShutdown()
w.quit = make(chan struct{})
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
default:
// Ignore when the wallet is still running.
if w.started {
w.quitMu.Unlock()
return
}
w.started = true
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
w.quitMu.Unlock()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.wg.Add(2)
go w.txCreator()
go w.walletLocker()
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
// SynchronizeRPC associates the wallet with the consensus RPC client,
// synchronizes the wallet with the latest changes to the blockchain, and
// continuously updates the wallet through RPC notifications.
//
// This method is unstable and will be removed when all syncing logic is moved
// outside of the wallet package.
func (w *Wallet) SynchronizeRPC(chainClient chain.Interface) {
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.quitMu.Lock()
select {
case <-w.quit:
w.quitMu.Unlock()
return
default:
}
w.quitMu.Unlock()
// TODO: Ignoring the new client when one is already set breaks callers
// who are replacing the client, perhaps after a disconnect.
w.chainClientLock.Lock()
if w.chainClient != nil {
w.chainClientLock.Unlock()
return
}
w.chainClient = chainClient
// If the chain client is a NeutrinoClient instance, set a birthday so
// we don't download all the filters as we go.
switch cc := chainClient.(type) {
case *chain.NeutrinoClient:
cc.SetStartTime(w.Manager.Birthday())
case *chain.BitcoindClient:
cc.SetBirthday(w.Manager.Birthday())
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.chainClientLock.Unlock()
// TODO: It would be preferable to either run these goroutines
// separately from the wallet (use wallet mutator functions to
// make changes from the RPC client) and not have to stop and
// restart them each time the client disconnects and reconnets.
w.wg.Add(4)
go w.handleChainNotifications()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
go w.rescanBatchHandler()
go w.rescanProgressHandler()
go w.rescanRPCHandler()
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
// requireChainClient marks that a wallet method can only be completed when the
// consensus RPC server is set. This function and all functions that call it
// are unstable and will need to be moved when the syncing code is moved out of
// the wallet.
func (w *Wallet) requireChainClient() (chain.Interface, error) {
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.chainClientLock.Lock()
chainClient := w.chainClient
w.chainClientLock.Unlock()
if chainClient == nil {
return nil, errors.New("blockchain RPC is inactive")
}
return chainClient, nil
}
// ChainClient returns the optional consensus RPC client associated with the
// wallet.
//
// This function is unstable and will be removed once sync logic is moved out of
// the wallet.
func (w *Wallet) ChainClient() chain.Interface {
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.chainClientLock.Lock()
chainClient := w.chainClient
w.chainClientLock.Unlock()
return chainClient
}
// quitChan atomically reads the quit channel.
func (w *Wallet) quitChan() <-chan struct{} {
w.quitMu.Lock()
c := w.quit
w.quitMu.Unlock()
return c
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Stop signals all wallet goroutines to shutdown.
func (w *Wallet) Stop() {
w.quitMu.Lock()
quit := w.quit
w.quitMu.Unlock()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
select {
case <-quit:
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
default:
close(quit)
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.chainClientLock.Lock()
if w.chainClient != nil {
w.chainClient.Stop()
w.chainClient = nil
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.chainClientLock.Unlock()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
}
// ShuttingDown returns whether the wallet is currently in the process of
// shutting down or not.
func (w *Wallet) ShuttingDown() bool {
select {
case <-w.quitChan():
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
return true
default:
return false
}
}
// WaitForShutdown blocks until all wallet goroutines have finished executing.
func (w *Wallet) WaitForShutdown() {
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.chainClientLock.Lock()
if w.chainClient != nil {
w.chainClient.WaitForShutdown()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.chainClientLock.Unlock()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
w.wg.Wait()
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
// SynchronizingToNetwork returns whether the wallet is currently synchronizing
// with the Bitcoin network.
func (w *Wallet) SynchronizingToNetwork() bool {
// At the moment, RPC is the only synchronization method. In the
// future, when SPV is added, a separate check will also be needed, or
// SPV could always be enabled if RPC was not explicitly specified when
// creating the wallet.
w.chainClientSyncMtx.Lock()
syncing := w.chainClient != nil
w.chainClientSyncMtx.Unlock()
return syncing
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// ChainSynced returns whether the wallet has been attached to a chain server
// and synced up to the best block on the main chain.
func (w *Wallet) ChainSynced() bool {
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.chainClientSyncMtx.Lock()
synced := w.chainClientSynced
w.chainClientSyncMtx.Unlock()
return synced
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// SetChainSynced marks whether the wallet is connected to and currently in sync
// with the latest block notified by the chain server.
//
// NOTE: Due to an API limitation with rpcclient, this may return true after
// the client disconnected (and is attempting a reconnect). This will be unknown
// until the reconnect notification is received, at which point the wallet can be
// marked out of sync again until after the next rescan completes.
func (w *Wallet) SetChainSynced(synced bool) {
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.chainClientSyncMtx.Lock()
w.chainClientSynced = synced
w.chainClientSyncMtx.Unlock()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// activeData returns the currently-active receiving addresses and all unspent
// outputs. This is primarely intended to provide the parameters for a
// rescan request.
func (w *Wallet) activeData(dbtx walletdb.ReadTx) ([]btcutil.Address, []wtxmgr.Credit, error) {
addrmgrNs := dbtx.ReadBucket(waddrmgrNamespaceKey)
txmgrNs := dbtx.ReadBucket(wtxmgrNamespaceKey)
var addrs []btcutil.Address
err := w.Manager.ForEachRelevantActiveAddress(
addrmgrNs, func(addr btcutil.Address) error {
addrs = append(addrs, addr)
return nil
},
)
if err != nil {
return nil, nil, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
unspent, err := w.TxStore.UnspentOutputs(txmgrNs)
return addrs, unspent, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// syncWithChain brings the wallet up to date with the current chain server
// connection. It creates a rescan request and blocks until the rescan has
// finished. The birthday block can be passed in, if set, to ensure we can
// properly detect if it gets rolled back.
func (w *Wallet) syncWithChain(birthdayStamp *waddrmgr.BlockStamp) error {
chainClient, err := w.requireChainClient()
if err != nil {
return err
}
// Neutrino relies on the information given to it by the cfheader server
// so it knows exactly whether it's synced up to the server's state or
// not, even on dev chains. To recover a Neutrino wallet, we need to
// make sure it's synced before we start scanning for addresses,
// otherwise we might miss some if we only scan up to its current sync
// point.
neutrinoRecovery := chainClient.BackEnd() == "neutrino" &&
w.recoveryWindow > 0
// We'll wait until the backend is synced to ensure we get the latest
// MaxReorgDepth blocks to store. We don't do this for development
// environments as we can't guarantee a lively chain, except for
// Neutrino, where the cfheader server tells us what it believes the
// chain tip is.
if !w.isDevEnv() || neutrinoRecovery {
log.Debug("Waiting for chain backend to sync to tip")
if err := w.waitUntilBackendSynced(chainClient); err != nil {
return err
}
log.Debug("Chain backend synced to tip!")
}
// If we've yet to find our birthday block, we'll do so now.
if birthdayStamp == nil {
var err error
birthdayStamp, err = locateBirthdayBlock(
chainClient, w.Manager.Birthday(),
)
if err != nil {
return fmt.Errorf("unable to locate birthday block: %v",
err)
}
// We'll also determine our initial sync starting height. This
// is needed as the wallet can now begin storing blocks from an
// arbitrary height, rather than all the blocks from genesis, so
// we persist this height to ensure we don't store any blocks
// before it.
startHeight := birthdayStamp.Height
// With the starting height obtained, get the remaining block
// details required by the wallet.
startHash, err := chainClient.GetBlockHash(int64(startHeight))
if err != nil {
return err
}
startHeader, err := chainClient.GetBlockHeader(startHash)
if err != nil {
return err
}
err = walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
ns := tx.ReadWriteBucket(waddrmgrNamespaceKey)
err := w.Manager.SetSyncedTo(ns, &waddrmgr.BlockStamp{
Hash: *startHash,
Height: startHeight,
Timestamp: startHeader.Timestamp,
})
if err != nil {
return err
}
return w.Manager.SetBirthdayBlock(ns, *birthdayStamp, true)
})
if err != nil {
return fmt.Errorf("unable to persist initial sync "+
"data: %v", err)
}
}
// If the wallet requested an on-chain recovery of its funds, we'll do
// so now.
if w.recoveryWindow > 0 {
if err := w.recovery(chainClient, birthdayStamp); err != nil {
return fmt.Errorf("unable to perform wallet recovery: "+
"%v", err)
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
// Compare previously-seen blocks against the current chain. If any of
2015-04-06 21:03:24 +02:00
// these blocks no longer exist, rollback all of the missing blocks
// before catching up with the rescan.
rollback := false
rollbackStamp := w.Manager.SyncedTo()
err = walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
addrmgrNs := tx.ReadWriteBucket(waddrmgrNamespaceKey)
txmgrNs := tx.ReadWriteBucket(wtxmgrNamespaceKey)
for height := rollbackStamp.Height; true; height-- {
hash, err := w.Manager.BlockHash(addrmgrNs, height)
if err != nil {
return err
}
chainHash, err := chainClient.GetBlockHash(int64(height))
if err != nil {
return err
}
header, err := chainClient.GetBlockHeader(chainHash)
if err != nil {
return err
}
rollbackStamp.Hash = *chainHash
rollbackStamp.Height = height
rollbackStamp.Timestamp = header.Timestamp
if bytes.Equal(hash[:], chainHash[:]) {
break
}
2015-04-06 21:03:24 +02:00
rollback = true
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// If a rollback did not happen, we can proceed safely.
if !rollback {
return nil
}
// Otherwise, we'll mark this as our new synced height.
err := w.Manager.SetSyncedTo(addrmgrNs, &rollbackStamp)
if err != nil {
return err
}
// If the rollback happened to go beyond our birthday stamp,
// we'll need to find a new one by syncing with the chain again
// until finding one.
if rollbackStamp.Height <= birthdayStamp.Height &&
rollbackStamp.Hash != birthdayStamp.Hash {
err := w.Manager.SetBirthdayBlock(
addrmgrNs, rollbackStamp, true,
)
if err != nil {
return err
}
2015-04-06 21:03:24 +02:00
}
// Finally, we'll roll back our transaction store to reflect the
// stale state. `Rollback` unconfirms transactions at and beyond
// the passed height, so add one to the new synced-to height to
// prevent unconfirming transactions in the synced-to block.
return w.TxStore.Rollback(txmgrNs, rollbackStamp.Height+1)
})
if err != nil {
return err
}
// Request notifications for connected and disconnected blocks.
//
// TODO(jrick): Either request this notification only once, or when
// rpcclient is modified to allow some notification request to not
// automatically resent on reconnect, include the notifyblocks request
// as well. I am leaning towards allowing off all rpcclient
// notification re-registrations, in which case the code here should be
// left as is.
if err := chainClient.NotifyBlocks(); err != nil {
return err
2015-04-06 21:03:24 +02:00
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Finally, we'll trigger a wallet rescan and request notifications for
// transactions sending to all wallet addresses and spending all wallet
// UTXOs.
var (
addrs []btcutil.Address
unspent []wtxmgr.Credit
)
err = walletdb.View(w.db, func(dbtx walletdb.ReadTx) error {
addrs, unspent, err = w.activeData(dbtx)
return err
})
if err != nil {
return err
}
return w.rescanWithTarget(addrs, unspent, nil)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// isDevEnv determines whether the wallet is currently under a local developer
// environment, e.g. simnet or regtest.
func (w *Wallet) isDevEnv() bool {
switch uint32(w.ChainParams().Net) {
case uint32(chaincfg.RegressionNetParams.Net):
case uint32(chaincfg.SimNetParams.Net):
default:
return false
}
return true
}
// waitUntilBackendSynced blocks until the chain backend considers itself
// "current".
func (w *Wallet) waitUntilBackendSynced(chainClient chain.Interface) error {
// We'll poll every second to determine if our chain considers itself
// "current".
t := time.NewTicker(time.Second)
defer t.Stop()
for {
select {
case <-t.C:
if chainClient.IsCurrent() {
return nil
}
case <-w.quitChan():
return ErrWalletShuttingDown
}
}
}
// locateBirthdayBlock returns a block that meets the given birthday timestamp
// by a margin of +/-2 hours. This is safe to do as the timestamp is already 2
// days in the past of the actual timestamp.
func locateBirthdayBlock(chainClient chainConn,
birthday time.Time) (*waddrmgr.BlockStamp, error) {
// Retrieve the lookup range for our block.
startHeight := int32(0)
_, bestHeight, err := chainClient.GetBestBlock()
if err != nil {
return nil, err
}
log.Debugf("Locating suitable block for birthday %v between blocks "+
"%v-%v", birthday, startHeight, bestHeight)
var (
birthdayBlock *waddrmgr.BlockStamp
left, right = startHeight, bestHeight
)
// Binary search for a block that meets the birthday timestamp by a
// margin of +/-2 hours.
for {
// Retrieve the timestamp for the block halfway through our
// range.
mid := left + (right-left)/2
hash, err := chainClient.GetBlockHash(int64(mid))
if err != nil {
return nil, err
}
header, err := chainClient.GetBlockHeader(hash)
if err != nil {
return nil, err
}
log.Debugf("Checking candidate block: height=%v, hash=%v, "+
"timestamp=%v", mid, hash, header.Timestamp)
// If the search happened to reach either of our range extremes,
// then we'll just use that as there's nothing left to search.
if mid == startHeight || mid == bestHeight || mid == left {
birthdayBlock = &waddrmgr.BlockStamp{
Hash: *hash,
Height: int32(mid),
Timestamp: header.Timestamp,
}
break
}
// The block's timestamp is more than 2 hours after the
// birthday, so look for a lower block.
if header.Timestamp.Sub(birthday) > birthdayBlockDelta {
right = mid
continue
}
// The birthday is more than 2 hours before the block's
// timestamp, so look for a higher block.
if header.Timestamp.Sub(birthday) < -birthdayBlockDelta {
left = mid
continue
}
birthdayBlock = &waddrmgr.BlockStamp{
Hash: *hash,
Height: int32(mid),
Timestamp: header.Timestamp,
}
break
}
log.Debugf("Found birthday block: height=%d, hash=%v, timestamp=%v",
birthdayBlock.Height, birthdayBlock.Hash,
birthdayBlock.Timestamp)
return birthdayBlock, nil
}
// recovery attempts to recover any unspent outputs that pay to any of our
// addresses starting from our birthday, or the wallet's tip (if higher), which
// would indicate resuming a recovery after a restart.
func (w *Wallet) recovery(chainClient chain.Interface,
birthdayBlock *waddrmgr.BlockStamp) error {
log.Infof("RECOVERY MODE ENABLED -- rescanning for used addresses "+
"with recovery_window=%d", w.recoveryWindow)
// We'll initialize the recovery manager with a default batch size of
// 2000.
recoveryMgr := NewRecoveryManager(
w.recoveryWindow, recoveryBatchSize, w.chainParams,
)
// In the event that this recovery is being resumed, we will need to
// repopulate all found addresses from the database. Ideally, for basic
// recovery, we would only do so for the default scopes, but due to a
// bug in which the wallet would create change addresses outside of the
// default scopes, it's necessary to attempt all registered key scopes.
scopedMgrs := make(map[waddrmgr.KeyScope]*waddrmgr.ScopedKeyManager)
for _, scopedMgr := range w.Manager.ActiveScopedKeyManagers() {
scopedMgrs[scopedMgr.Scope()] = scopedMgr
}
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
txMgrNS := tx.ReadBucket(wtxmgrNamespaceKey)
credits, err := w.TxStore.UnspentOutputs(txMgrNS)
if err != nil {
return err
}
addrMgrNS := tx.ReadBucket(waddrmgrNamespaceKey)
return recoveryMgr.Resurrect(addrMgrNS, scopedMgrs, credits)
})
if err != nil {
return err
}
// Fetch the best height from the backend to determine when we should
// stop.
_, bestHeight, err := chainClient.GetBestBlock()
if err != nil {
return err
}
// Now we can begin scanning the chain from the wallet's current tip to
// ensure we properly handle restarts. Since the recovery process itself
// acts as rescan, we'll also update our wallet's synced state along the
// way to reflect the blocks we process and prevent rescanning them
// later on.
//
// NOTE: We purposefully don't update our best height since we assume
// that a wallet rescan will be performed from the wallet's tip, which
// will be of bestHeight after completing the recovery process.
var blocks []*waddrmgr.BlockStamp
startHeight := w.Manager.SyncedTo().Height + 1
for height := startHeight; height <= bestHeight; height++ {
hash, err := chainClient.GetBlockHash(int64(height))
if err != nil {
return err
}
header, err := chainClient.GetBlockHeader(hash)
if err != nil {
return err
}
blocks = append(blocks, &waddrmgr.BlockStamp{
Hash: *hash,
Height: height,
Timestamp: header.Timestamp,
})
// It's possible for us to run into blocks before our birthday
// if our birthday is after our reorg safe height, so we'll make
// sure to not add those to the batch.
if height >= birthdayBlock.Height {
recoveryMgr.AddToBlockBatch(
hash, height, header.Timestamp,
)
}
// We'll perform our recovery in batches of 2000 blocks. It's
// possible for us to reach our best height without exceeding
// the recovery batch size, so we can proceed to commit our
// state to disk.
recoveryBatch := recoveryMgr.BlockBatch()
if len(recoveryBatch) == recoveryBatchSize || height == bestHeight {
err := walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
ns := tx.ReadWriteBucket(waddrmgrNamespaceKey)
for _, block := range blocks {
err := w.Manager.SetSyncedTo(ns, block)
if err != nil {
return err
}
}
return w.recoverScopedAddresses(
chainClient, tx, ns, recoveryBatch,
recoveryMgr.State(), scopedMgrs,
)
})
if err != nil {
return err
}
if len(recoveryBatch) > 0 {
log.Infof("Recovered addresses from blocks "+
"%d-%d", recoveryBatch[0].Height,
recoveryBatch[len(recoveryBatch)-1].Height)
}
// Clear the batch of all processed blocks to reuse the
// same memory for future batches.
blocks = blocks[:0]
recoveryMgr.ResetBlockBatch()
}
}
return nil
}
// recoverScopedAddresses scans a range of blocks in attempts to recover any
// previously used addresses for a particular account derivation path. At a high
// level, the algorithm works as follows:
//
// 1) Ensure internal and external branch horizons are fully expanded.
// 2) Filter the entire range of blocks, stopping if a non-zero number of
// address are contained in a particular block.
// 3) Record all internal and external addresses found in the block.
// 4) Record any outpoints found in the block that should be watched for spends
// 5) Trim the range of blocks up to and including the one reporting the addrs.
// 6) Repeat from (1) if there are still more blocks in the range.
//
// TODO(conner): parallelize/pipeline/cache intermediate network requests
func (w *Wallet) recoverScopedAddresses(
chainClient chain.Interface,
tx walletdb.ReadWriteTx,
ns walletdb.ReadWriteBucket,
batch []wtxmgr.BlockMeta,
recoveryState *RecoveryState,
scopedMgrs map[waddrmgr.KeyScope]*waddrmgr.ScopedKeyManager) error {
// If there are no blocks in the batch, we are done.
if len(batch) == 0 {
return nil
}
log.Infof("Scanning %d blocks for recoverable addresses", len(batch))
expandHorizons:
for scope, scopedMgr := range scopedMgrs {
scopeState := recoveryState.StateForScope(scope)
err := expandScopeHorizons(ns, scopedMgr, scopeState)
if err != nil {
return err
}
}
// With the internal and external horizons properly expanded, we now
// construct the filter blocks request. The request includes the range
// of blocks we intend to scan, in addition to the scope-index -> addr
// map for all internal and external branches.
filterReq := newFilterBlocksRequest(batch, scopedMgrs, recoveryState)
// Initiate the filter blocks request using our chain backend. If an
// error occurs, we are unable to proceed with the recovery.
filterResp, err := chainClient.FilterBlocks(filterReq)
if err != nil {
return err
}
// If the filter response is empty, this signals that the rest of the
// batch was completed, and no other addresses were discovered. As a
// result, no further modifications to our recovery state are required
// and we can proceed to the next batch.
if filterResp == nil {
return nil
}
// Otherwise, retrieve the block info for the block that detected a
// non-zero number of address matches.
block := batch[filterResp.BatchIndex]
// Log any non-trivial findings of addresses or outpoints.
logFilterBlocksResp(block, filterResp)
// Report any external or internal addresses found as a result of the
// appropriate branch recovery state. Adding indexes above the
// last-found index of either will result in the horizons being expanded
// upon the next iteration. Any found addresses are also marked used
// using the scoped key manager.
err = extendFoundAddresses(ns, filterResp, scopedMgrs, recoveryState)
if err != nil {
return err
}
// Update the global set of watched outpoints with any that were found
// in the block.
for outPoint, addr := range filterResp.FoundOutPoints {
recoveryState.AddWatchedOutPoint(&outPoint, addr)
}
// Finally, record all of the relevant transactions that were returned
// in the filter blocks response. This ensures that these transactions
// and their outputs are tracked when the final rescan is performed.
for _, txn := range filterResp.RelevantTxns {
txRecord, err := wtxmgr.NewTxRecordFromMsgTx(
txn, filterResp.BlockMeta.Time,
)
if err != nil {
return err
}
err = w.addRelevantTx(tx, txRecord, &filterResp.BlockMeta)
if err != nil {
return err
}
}
// Update the batch to indicate that we've processed all block through
// the one that returned found addresses.
batch = batch[filterResp.BatchIndex+1:]
// If this was not the last block in the batch, we will repeat the
// filtering process again after expanding our horizons.
if len(batch) > 0 {
goto expandHorizons
}
return nil
}
// expandScopeHorizons ensures that the ScopeRecoveryState has an adequately
// sized look ahead for both its internal and external branches. The keys
// derived here are added to the scope's recovery state, but do not affect the
// persistent state of the wallet. If any invalid child keys are detected, the
// horizon will be properly extended such that our lookahead always includes the
// proper number of valid child keys.
func expandScopeHorizons(ns walletdb.ReadWriteBucket,
scopedMgr *waddrmgr.ScopedKeyManager,
scopeState *ScopeRecoveryState) error {
// Compute the current external horizon and the number of addresses we
// must derive to ensure we maintain a sufficient recovery window for
// the external branch.
exHorizon, exWindow := scopeState.ExternalBranch.ExtendHorizon()
count, childIndex := uint32(0), exHorizon
for count < exWindow {
keyPath := externalKeyPath(childIndex)
addr, err := scopedMgr.DeriveFromKeyPath(ns, keyPath)
switch {
case err == hdkeychain.ErrInvalidChild:
// Record the existence of an invalid child with the
// external branch's recovery state. This also
// increments the branch's horizon so that it accounts
// for this skipped child index.
scopeState.ExternalBranch.MarkInvalidChild(childIndex)
childIndex++
continue
case err != nil:
return err
}
// Register the newly generated external address and child index
// with the external branch recovery state.
scopeState.ExternalBranch.AddAddr(childIndex, addr.Address())
childIndex++
count++
}
// Compute the current internal horizon and the number of addresses we
// must derive to ensure we maintain a sufficient recovery window for
// the internal branch.
inHorizon, inWindow := scopeState.InternalBranch.ExtendHorizon()
count, childIndex = 0, inHorizon
for count < inWindow {
keyPath := internalKeyPath(childIndex)
addr, err := scopedMgr.DeriveFromKeyPath(ns, keyPath)
switch {
case err == hdkeychain.ErrInvalidChild:
// Record the existence of an invalid child with the
// internal branch's recovery state. This also
// increments the branch's horizon so that it accounts
// for this skipped child index.
scopeState.InternalBranch.MarkInvalidChild(childIndex)
childIndex++
continue
case err != nil:
return err
}
// Register the newly generated internal address and child index
// with the internal branch recovery state.
scopeState.InternalBranch.AddAddr(childIndex, addr.Address())
childIndex++
count++
}
return nil
}
// externalKeyPath returns the relative external derivation path /0/0/index.
func externalKeyPath(index uint32) waddrmgr.DerivationPath {
return waddrmgr.DerivationPath{
Account: waddrmgr.DefaultAccountNum,
Branch: waddrmgr.ExternalBranch,
Index: index,
}
}
// internalKeyPath returns the relative internal derivation path /0/1/index.
func internalKeyPath(index uint32) waddrmgr.DerivationPath {
return waddrmgr.DerivationPath{
Account: waddrmgr.DefaultAccountNum,
Branch: waddrmgr.InternalBranch,
Index: index,
}
}
// newFilterBlocksRequest constructs FilterBlocksRequests using our current
// block range, scoped managers, and recovery state.
func newFilterBlocksRequest(batch []wtxmgr.BlockMeta,
scopedMgrs map[waddrmgr.KeyScope]*waddrmgr.ScopedKeyManager,
recoveryState *RecoveryState) *chain.FilterBlocksRequest {
filterReq := &chain.FilterBlocksRequest{
Blocks: batch,
ExternalAddrs: make(map[waddrmgr.ScopedIndex]btcutil.Address),
InternalAddrs: make(map[waddrmgr.ScopedIndex]btcutil.Address),
WatchedOutPoints: recoveryState.WatchedOutPoints(),
}
// Populate the external and internal addresses by merging the addresses
// sets belong to all currently tracked scopes.
for scope := range scopedMgrs {
scopeState := recoveryState.StateForScope(scope)
for index, addr := range scopeState.ExternalBranch.Addrs() {
scopedIndex := waddrmgr.ScopedIndex{
Scope: scope,
Index: index,
}
filterReq.ExternalAddrs[scopedIndex] = addr
}
for index, addr := range scopeState.InternalBranch.Addrs() {
scopedIndex := waddrmgr.ScopedIndex{
Scope: scope,
Index: index,
}
filterReq.InternalAddrs[scopedIndex] = addr
}
}
return filterReq
}
// extendFoundAddresses accepts a filter blocks response that contains addresses
// found on chain, and advances the state of all relevant derivation paths to
// match the highest found child index for each branch.
func extendFoundAddresses(ns walletdb.ReadWriteBucket,
filterResp *chain.FilterBlocksResponse,
scopedMgrs map[waddrmgr.KeyScope]*waddrmgr.ScopedKeyManager,
recoveryState *RecoveryState) error {
// Mark all recovered external addresses as used. This will be done only
// for scopes that reported a non-zero number of external addresses in
// this block.
for scope, indexes := range filterResp.FoundExternalAddrs {
// First, report all external child indexes found for this
// scope. This ensures that the external last-found index will
// be updated to include the maximum child index seen thus far.
scopeState := recoveryState.StateForScope(scope)
for index := range indexes {
scopeState.ExternalBranch.ReportFound(index)
}
scopedMgr := scopedMgrs[scope]
// Now, with all found addresses reported, derive and extend all
// external addresses up to and including the current last found
// index for this scope.
exNextUnfound := scopeState.ExternalBranch.NextUnfound()
exLastFound := exNextUnfound
if exLastFound > 0 {
exLastFound--
}
err := scopedMgr.ExtendExternalAddresses(
ns, waddrmgr.DefaultAccountNum, exLastFound,
)
if err != nil {
return err
}
// Finally, with the scope's addresses extended, we mark used
// the external addresses that were found in the block and
// belong to this scope.
for index := range indexes {
addr := scopeState.ExternalBranch.GetAddr(index)
err := scopedMgr.MarkUsed(ns, addr)
if err != nil {
return err
}
}
}
// Mark all recovered internal addresses as used. This will be done only
// for scopes that reported a non-zero number of internal addresses in
// this block.
for scope, indexes := range filterResp.FoundInternalAddrs {
// First, report all internal child indexes found for this
// scope. This ensures that the internal last-found index will
// be updated to include the maximum child index seen thus far.
scopeState := recoveryState.StateForScope(scope)
for index := range indexes {
scopeState.InternalBranch.ReportFound(index)
}
scopedMgr := scopedMgrs[scope]
// Now, with all found addresses reported, derive and extend all
// internal addresses up to and including the current last found
// index for this scope.
inNextUnfound := scopeState.InternalBranch.NextUnfound()
inLastFound := inNextUnfound
if inLastFound > 0 {
inLastFound--
}
err := scopedMgr.ExtendInternalAddresses(
ns, waddrmgr.DefaultAccountNum, inLastFound,
)
if err != nil {
return err
}
// Finally, with the scope's addresses extended, we mark used
// the internal addresses that were found in the blockand belong
// to this scope.
for index := range indexes {
addr := scopeState.InternalBranch.GetAddr(index)
err := scopedMgr.MarkUsed(ns, addr)
if err != nil {
return err
}
}
}
return nil
}
// logFilterBlocksResp provides useful logging information when filtering
// succeeded in finding relevant transactions.
func logFilterBlocksResp(block wtxmgr.BlockMeta,
resp *chain.FilterBlocksResponse) {
// Log the number of external addresses found in this block.
var nFoundExternal int
for _, indexes := range resp.FoundExternalAddrs {
nFoundExternal += len(indexes)
}
if nFoundExternal > 0 {
log.Infof("Recovered %d external addrs at height=%d hash=%v",
nFoundExternal, block.Height, block.Hash)
}
// Log the number of internal addresses found in this block.
var nFoundInternal int
for _, indexes := range resp.FoundInternalAddrs {
nFoundInternal += len(indexes)
}
if nFoundInternal > 0 {
log.Infof("Recovered %d internal addrs at height=%d hash=%v",
nFoundInternal, block.Height, block.Hash)
}
// Log the number of outpoints found in this block.
nFoundOutPoints := len(resp.FoundOutPoints)
if nFoundOutPoints > 0 {
log.Infof("Found %d spends from watched outpoints at "+
"height=%d hash=%v",
nFoundOutPoints, block.Height, block.Hash)
}
}
type (
createTxRequest struct {
account uint32
outputs []*wire.TxOut
minconf int32
feeSatPerKB btcutil.Amount
dryRun bool
resp chan createTxResponse
}
createTxResponse struct {
Refactor wallet transaction creation code. This began as a change to improve the fee calculation code and evolved into a much larger refactor which improves the readability and modularity of all of the transaction creation code. Transaction fee calculations have been switched from full increments of the relay fee to a proportion based on the transaction size. This means that for a relay fee of 1e3 satoshis/kB, a 500 byte transaction is only required to pay a 5e2 satoshi fee and a 1500 byte transaction only need pay a 1.5e3 fee. The previous code would end up estimating these fees to be 1e3 and 2e3 respectively. Because the previous code would add more fee than needed in almost every case, the transaction size estimations were optimistic (best/smallest case) and signing was done in a loop where the fee was incremented by the relay fee again each time the actual size of the signed transaction rendered the fee too low. This has switched to using worst case transaction size estimates rather than best case, and signing is only performed once. Transaction input signature creation has switched from using txscript.SignatureScript to txscript.SignTxOutput. The new API is able to redeem outputs other than just P2PKH, so the previous restrictions about P2SH outputs being unspendable (except through the signrawtransaction RPC) no longer hold. Several new public packages have been added: wallet/txauthor - transaction authoring and signing wallet/txfees - fee estimations and change output inclusion wallet/txrules - simple consensus and mempool policy rule checks Along with some internal packages: wallet/internal/txsizes - transaction size estimation internal/helpers - context free convenience functions The txsizes package is internal as the estimations it provides are specific for the algorithms used by these new packages.
2016-02-28 05:30:56 +01:00
tx *txauthor.AuthoredTx
err error
}
)
// txCreator is responsible for the input selection and creation of
// transactions. These functions are the responsibility of this method
// (designed to be run as its own goroutine) since input selection must be
// serialized, or else it is possible to create double spends by choosing the
// same inputs for multiple transactions. Along with input selection, this
// method is also responsible for the signing of transactions, since we don't
// want to end up in a situation where we run out of inputs as multiple
// transactions are being created. In this situation, it would then be possible
// for both requests, rather than just one, to fail due to not enough available
// inputs.
func (w *Wallet) txCreator() {
quit := w.quitChan()
out:
for {
select {
case txr := <-w.createTxRequests:
heldUnlock, err := w.holdUnlock()
if err != nil {
txr.resp <- createTxResponse{nil, err}
continue
}
tx, err := w.txToOutputs(txr.outputs, txr.account,
txr.minconf, txr.feeSatPerKB, txr.dryRun)
heldUnlock.release()
txr.resp <- createTxResponse{tx, err}
case <-quit:
break out
}
}
w.wg.Done()
}
// CreateSimpleTx creates a new signed transaction spending unspent P2PKH
// outputs with at least minconf confirmations spending to any number of
Refactor wallet transaction creation code. This began as a change to improve the fee calculation code and evolved into a much larger refactor which improves the readability and modularity of all of the transaction creation code. Transaction fee calculations have been switched from full increments of the relay fee to a proportion based on the transaction size. This means that for a relay fee of 1e3 satoshis/kB, a 500 byte transaction is only required to pay a 5e2 satoshi fee and a 1500 byte transaction only need pay a 1.5e3 fee. The previous code would end up estimating these fees to be 1e3 and 2e3 respectively. Because the previous code would add more fee than needed in almost every case, the transaction size estimations were optimistic (best/smallest case) and signing was done in a loop where the fee was incremented by the relay fee again each time the actual size of the signed transaction rendered the fee too low. This has switched to using worst case transaction size estimates rather than best case, and signing is only performed once. Transaction input signature creation has switched from using txscript.SignatureScript to txscript.SignTxOutput. The new API is able to redeem outputs other than just P2PKH, so the previous restrictions about P2SH outputs being unspendable (except through the signrawtransaction RPC) no longer hold. Several new public packages have been added: wallet/txauthor - transaction authoring and signing wallet/txfees - fee estimations and change output inclusion wallet/txrules - simple consensus and mempool policy rule checks Along with some internal packages: wallet/internal/txsizes - transaction size estimation internal/helpers - context free convenience functions The txsizes package is internal as the estimations it provides are specific for the algorithms used by these new packages.
2016-02-28 05:30:56 +01:00
// address/amount pairs. Change and an appropriate transaction fee are
// automatically included, if necessary. All transaction creation through this
// function is serialized to prevent the creation of many transactions which
// spend the same outputs.
//
// NOTE: The dryRun argument can be set true to create a tx that doesn't alter
// the database. A tx created with this set to true SHOULD NOT be broadcasted.
Refactor wallet transaction creation code. This began as a change to improve the fee calculation code and evolved into a much larger refactor which improves the readability and modularity of all of the transaction creation code. Transaction fee calculations have been switched from full increments of the relay fee to a proportion based on the transaction size. This means that for a relay fee of 1e3 satoshis/kB, a 500 byte transaction is only required to pay a 5e2 satoshi fee and a 1500 byte transaction only need pay a 1.5e3 fee. The previous code would end up estimating these fees to be 1e3 and 2e3 respectively. Because the previous code would add more fee than needed in almost every case, the transaction size estimations were optimistic (best/smallest case) and signing was done in a loop where the fee was incremented by the relay fee again each time the actual size of the signed transaction rendered the fee too low. This has switched to using worst case transaction size estimates rather than best case, and signing is only performed once. Transaction input signature creation has switched from using txscript.SignatureScript to txscript.SignTxOutput. The new API is able to redeem outputs other than just P2PKH, so the previous restrictions about P2SH outputs being unspendable (except through the signrawtransaction RPC) no longer hold. Several new public packages have been added: wallet/txauthor - transaction authoring and signing wallet/txfees - fee estimations and change output inclusion wallet/txrules - simple consensus and mempool policy rule checks Along with some internal packages: wallet/internal/txsizes - transaction size estimation internal/helpers - context free convenience functions The txsizes package is internal as the estimations it provides are specific for the algorithms used by these new packages.
2016-02-28 05:30:56 +01:00
func (w *Wallet) CreateSimpleTx(account uint32, outputs []*wire.TxOut,
minconf int32, satPerKb btcutil.Amount, dryRun bool) (
*txauthor.AuthoredTx, error) {
req := createTxRequest{
account: account,
outputs: outputs,
minconf: minconf,
feeSatPerKB: satPerKb,
dryRun: dryRun,
resp: make(chan createTxResponse),
}
w.createTxRequests <- req
resp := <-req.resp
return resp.tx, resp.err
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
type (
unlockRequest struct {
passphrase []byte
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
lockAfter <-chan time.Time // nil prevents the timeout.
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
err chan error
}
changePassphraseRequest struct {
old, new []byte
private bool
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
err chan error
}
changePassphrasesRequest struct {
publicOld, publicNew []byte
privateOld, privateNew []byte
err chan error
}
// heldUnlock is a tool to prevent the wallet from automatically
// locking after some timeout before an operation which needed
// the unlocked wallet has finished. Any aquired heldUnlock
// *must* be released (preferably with a defer) or the wallet
// will forever remain unlocked.
heldUnlock chan struct{}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
)
// walletLocker manages the locked/unlocked state of a wallet.
func (w *Wallet) walletLocker() {
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
var timeout <-chan time.Time
holdChan := make(heldUnlock)
quit := w.quitChan()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
out:
for {
select {
case req := <-w.unlockRequests:
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
return w.Manager.Unlock(addrmgrNs, req.passphrase)
})
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
if err != nil {
req.err <- err
continue
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
timeout = req.lockAfter
if timeout == nil {
2016-03-11 23:01:04 +01:00
log.Info("The wallet has been unlocked without a time limit")
} else {
2016-03-11 23:01:04 +01:00
log.Info("The wallet has been temporarily unlocked")
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
req.err <- nil
continue
case req := <-w.changePassphrase:
err := walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
addrmgrNs := tx.ReadWriteBucket(waddrmgrNamespaceKey)
return w.Manager.ChangePassphrase(
addrmgrNs, req.old, req.new, req.private,
&waddrmgr.DefaultScryptOptions,
)
})
req.err <- err
continue
case req := <-w.changePassphrases:
err := walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
addrmgrNs := tx.ReadWriteBucket(waddrmgrNamespaceKey)
err := w.Manager.ChangePassphrase(
addrmgrNs, req.publicOld, req.publicNew,
false, &waddrmgr.DefaultScryptOptions,
)
if err != nil {
return err
}
return w.Manager.ChangePassphrase(
addrmgrNs, req.privateOld, req.privateNew,
true, &waddrmgr.DefaultScryptOptions,
)
})
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
req.err <- err
continue
case req := <-w.holdUnlockRequests:
if w.Manager.IsLocked() {
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
close(req)
continue
}
req <- holdChan
<-holdChan // Block until the lock is released.
// If, after holding onto the unlocked wallet for some
// time, the timeout has expired, lock it now instead
// of hoping it gets unlocked next time the top level
// select runs.
select {
case <-timeout:
// Let the top level select fallthrough so the
// wallet is locked.
default:
continue
}
case w.lockState <- w.Manager.IsLocked():
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
continue
case <-quit:
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
break out
case <-w.lockRequests:
case <-timeout:
}
// Select statement fell through by an explicit lock or the
// timer expiring. Lock the manager here.
timeout = nil
err := w.Manager.Lock()
if err != nil && !waddrmgr.IsError(err, waddrmgr.ErrLocked) {
log.Errorf("Could not lock wallet: %v", err)
} else {
2016-03-11 23:01:04 +01:00
log.Info("The wallet has been locked")
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
}
w.wg.Done()
}
// Unlock unlocks the wallet's address manager and relocks it after timeout has
// expired. If the wallet is already unlocked and the new passphrase is
// correct, the current timeout is replaced with the new one. The wallet will
// be locked if the passphrase is incorrect or any other error occurs during the
// unlock.
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
func (w *Wallet) Unlock(passphrase []byte, lock <-chan time.Time) error {
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
err := make(chan error, 1)
w.unlockRequests <- unlockRequest{
passphrase: passphrase,
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
lockAfter: lock,
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
err: err,
}
return <-err
}
// Lock locks the wallet's address manager.
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
func (w *Wallet) Lock() {
w.lockRequests <- struct{}{}
}
// Locked returns whether the account manager for a wallet is locked.
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
func (w *Wallet) Locked() bool {
return <-w.lockState
}
// holdUnlock prevents the wallet from being locked. The heldUnlock object
// *must* be released, or the wallet will forever remain unlocked.
//
// TODO: To prevent the above scenario, perhaps closures should be passed
// to the walletLocker goroutine and disallow callers from explicitly
// handling the locking mechanism.
func (w *Wallet) holdUnlock() (heldUnlock, error) {
req := make(chan heldUnlock)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
w.holdUnlockRequests <- req
hl, ok := <-req
if !ok {
// TODO(davec): This should be defined and exported from
// waddrmgr.
return nil, waddrmgr.ManagerError{
ErrorCode: waddrmgr.ErrLocked,
Description: "address manager is locked",
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
return hl, nil
}
// release releases the hold on the unlocked-state of the wallet and allows the
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// wallet to be locked again. If a lock timeout has already expired, the
// wallet is locked again as soon as release is called.
func (c heldUnlock) release() {
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
c <- struct{}{}
}
// ChangePrivatePassphrase attempts to change the passphrase for a wallet from
// old to new. Changing the passphrase is synchronized with all other address
// manager locking and unlocking. The lock state will be the same as it was
// before the password change.
func (w *Wallet) ChangePrivatePassphrase(old, new []byte) error {
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
err := make(chan error, 1)
w.changePassphrase <- changePassphraseRequest{
old: old,
new: new,
private: true,
err: err,
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
return <-err
}
// ChangePublicPassphrase modifies the public passphrase of the wallet.
func (w *Wallet) ChangePublicPassphrase(old, new []byte) error {
err := make(chan error, 1)
w.changePassphrase <- changePassphraseRequest{
old: old,
new: new,
private: false,
err: err,
}
return <-err
}
// ChangePassphrases modifies the public and private passphrase of the wallet
// atomically.
func (w *Wallet) ChangePassphrases(publicOld, publicNew, privateOld,
privateNew []byte) error {
err := make(chan error, 1)
w.changePassphrases <- changePassphrasesRequest{
publicOld: publicOld,
publicNew: publicNew,
privateOld: privateOld,
privateNew: privateNew,
err: err,
}
return <-err
}
// accountUsed returns whether there are any recorded transactions spending to
// a given account. It returns true if atleast one address in the account was
// used and false if no address in the account was used.
func (w *Wallet) accountUsed(addrmgrNs walletdb.ReadWriteBucket, account uint32) (bool, error) {
var used bool
err := w.Manager.ForEachAccountAddress(addrmgrNs, account,
func(maddr waddrmgr.ManagedAddress) error {
used = maddr.Used(addrmgrNs)
if used {
return waddrmgr.Break
}
return nil
})
if err == waddrmgr.Break {
err = nil
}
return used, err
}
// AccountAddresses returns the addresses for every created address for an
// account.
func (w *Wallet) AccountAddresses(account uint32) (addrs []btcutil.Address, err error) {
err = walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
return w.Manager.ForEachAccountAddress(addrmgrNs, account, func(maddr waddrmgr.ManagedAddress) error {
addrs = append(addrs, maddr.Address())
return nil
})
})
return
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// CalculateBalance sums the amounts of all unspent transaction
// outputs to addresses of a wallet and returns the balance.
//
// If confirmations is 0, all UTXOs, even those not present in a
// block (height -1), will be used to get the balance. Otherwise,
// a UTXO must be in a block. If confirmations is 1 or greater,
// the balance will be calculated based on how many how many blocks
// include a UTXO.
2015-04-06 21:03:24 +02:00
func (w *Wallet) CalculateBalance(confirms int32) (btcutil.Amount, error) {
var balance btcutil.Amount
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
var err error
blk := w.Manager.SyncedTo()
balance, err = w.TxStore.Balance(txmgrNs, confirms, blk.Height)
return err
})
return balance, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
// Balances records total, spendable (by policy), and immature coinbase
// reward balance amounts.
type Balances struct {
Total btcutil.Amount
Spendable btcutil.Amount
ImmatureReward btcutil.Amount
}
// CalculateAccountBalances sums the amounts of all unspent transaction
// outputs to the given account of a wallet and returns the balance.
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
//
// This function is much slower than it needs to be since transactions outputs
// are not indexed by the accounts they credit to, and all unspent transaction
// outputs must be iterated.
func (w *Wallet) CalculateAccountBalances(account uint32, confirms int32) (Balances, error) {
var bals Balances
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
// Get current block. The block height used for calculating
// the number of tx confirmations.
syncBlock := w.Manager.SyncedTo()
2015-04-06 21:03:24 +02:00
unspent, err := w.TxStore.UnspentOutputs(txmgrNs)
if err != nil {
return err
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
for i := range unspent {
output := &unspent[i]
var outputAcct uint32
_, addrs, _, err := txscript.ExtractPkScriptAddrs(
output.PkScript, w.chainParams)
if err == nil && len(addrs) > 0 {
_, outputAcct, err = w.Manager.AddrAccount(addrmgrNs, addrs[0])
}
if err != nil || outputAcct != account {
continue
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
bals.Total += output.Amount
if output.FromCoinBase && !confirmed(int32(w.chainParams.CoinbaseMaturity),
output.Height, syncBlock.Height) {
bals.ImmatureReward += output.Amount
} else if confirmed(confirms, output.Height, syncBlock.Height) {
bals.Spendable += output.Amount
}
}
return nil
})
return bals, err
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// CurrentAddress gets the most recently requested Bitcoin payment address
// from a wallet for a particular key-chain scope. If the address has already
// been used (there is at least one transaction spending to it in the
// blockchain or btcd mempool), the next chained address is returned.
func (w *Wallet) CurrentAddress(account uint32, scope waddrmgr.KeyScope) (btcutil.Address, error) {
chainClient, err := w.requireChainClient()
if err != nil {
return nil, err
}
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return nil, err
}
var (
addr btcutil.Address
props *waddrmgr.AccountProperties
)
err = walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
addrmgrNs := tx.ReadWriteBucket(waddrmgrNamespaceKey)
maddr, err := manager.LastExternalAddress(addrmgrNs, account)
if err != nil {
// If no address exists yet, create the first external
// address.
if waddrmgr.IsError(err, waddrmgr.ErrAddressNotFound) {
addr, props, err = w.newAddress(
addrmgrNs, account, scope,
)
}
return err
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Get next chained address if the last one has already been
// used.
if maddr.Used(addrmgrNs) {
addr, props, err = w.newAddress(
addrmgrNs, account, scope,
)
return err
}
addr = maddr.Address()
return nil
})
if err != nil {
return nil, err
}
// If the props have been initially, then we had to create a new address
// to satisfy the query. Notify the rpc server about the new address.
if props != nil {
err = chainClient.NotifyReceived([]btcutil.Address{addr})
if err != nil {
return nil, err
}
w.NtfnServer.notifyAccountProperties(props)
}
return addr, nil
}
// PubKeyForAddress looks up the associated public key for a P2PKH address.
func (w *Wallet) PubKeyForAddress(a btcutil.Address) (*btcec.PublicKey, error) {
var pubKey *btcec.PublicKey
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
managedAddr, err := w.Manager.Address(addrmgrNs, a)
if err != nil {
return err
}
managedPubKeyAddr, ok := managedAddr.(waddrmgr.ManagedPubKeyAddress)
if !ok {
return errors.New("address does not have an associated public key")
}
pubKey = managedPubKeyAddr.PubKey()
return nil
})
return pubKey, err
}
// PrivKeyForAddress looks up the associated private key for a P2PKH or P2PK
// address.
func (w *Wallet) PrivKeyForAddress(a btcutil.Address) (*btcec.PrivateKey, error) {
var privKey *btcec.PrivateKey
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
managedAddr, err := w.Manager.Address(addrmgrNs, a)
if err != nil {
return err
}
managedPubKeyAddr, ok := managedAddr.(waddrmgr.ManagedPubKeyAddress)
if !ok {
return errors.New("address does not have an associated private key")
}
privKey, err = managedPubKeyAddr.PrivKey()
return err
})
return privKey, err
}
// HaveAddress returns whether the wallet is the owner of the address a.
func (w *Wallet) HaveAddress(a btcutil.Address) (bool, error) {
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
_, err := w.Manager.Address(addrmgrNs, a)
return err
})
if err == nil {
return true, nil
}
if waddrmgr.IsError(err, waddrmgr.ErrAddressNotFound) {
return false, nil
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
return false, err
}
// AccountOfAddress finds the account that an address is associated with.
func (w *Wallet) AccountOfAddress(a btcutil.Address) (uint32, error) {
var account uint32
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
var err error
_, account, err = w.Manager.AddrAccount(addrmgrNs, a)
return err
})
return account, err
}
// AddressInfo returns detailed information regarding a wallet address.
func (w *Wallet) AddressInfo(a btcutil.Address) (waddrmgr.ManagedAddress, error) {
var managedAddress waddrmgr.ManagedAddress
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
var err error
managedAddress, err = w.Manager.Address(addrmgrNs, a)
return err
})
return managedAddress, err
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// AccountNumber returns the account number for an account name under a
// particular key scope.
func (w *Wallet) AccountNumber(scope waddrmgr.KeyScope, accountName string) (uint32, error) {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return 0, err
}
var account uint32
err = walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
var err error
account, err = manager.LookupAccount(addrmgrNs, accountName)
return err
})
return account, err
}
// AccountName returns the name of an account.
func (w *Wallet) AccountName(scope waddrmgr.KeyScope, accountNumber uint32) (string, error) {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return "", err
}
var accountName string
err = walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
var err error
accountName, err = manager.AccountName(addrmgrNs, accountNumber)
return err
})
return accountName, err
}
// AccountProperties returns the properties of an account, including address
// indexes and name. It first fetches the desynced information from the address
// manager, then updates the indexes based on the address pools.
func (w *Wallet) AccountProperties(scope waddrmgr.KeyScope, acct uint32) (*waddrmgr.AccountProperties, error) {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return nil, err
}
var props *waddrmgr.AccountProperties
err = walletdb.View(w.db, func(tx walletdb.ReadTx) error {
waddrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
var err error
props, err = manager.AccountProperties(waddrmgrNs, acct)
return err
})
return props, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
// RenameAccount sets the name for an account number to newName.
func (w *Wallet) RenameAccount(scope waddrmgr.KeyScope, account uint32, newName string) error {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return err
}
var props *waddrmgr.AccountProperties
err = walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
addrmgrNs := tx.ReadWriteBucket(waddrmgrNamespaceKey)
err := manager.RenameAccount(addrmgrNs, account, newName)
if err != nil {
return err
}
props, err = manager.AccountProperties(addrmgrNs, account)
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
return err
})
if err == nil {
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.NtfnServer.notifyAccountProperties(props)
}
return err
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
const maxEmptyAccounts = 100
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
// NextAccount creates the next account and returns its account number. The
// name must be unique to the account. In order to support automatic seed
// restoring, new accounts may not be created when all of the previous 100
// accounts have no transaction history (this is a deviation from the BIP0044
// spec, which allows no unused account gaps).
func (w *Wallet) NextAccount(scope waddrmgr.KeyScope, name string) (uint32, error) {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return 0, err
}
var (
account uint32
props *waddrmgr.AccountProperties
)
err = walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
addrmgrNs := tx.ReadWriteBucket(waddrmgrNamespaceKey)
var err error
account, err = manager.NewAccount(addrmgrNs, name)
if err != nil {
return err
}
props, err = manager.AccountProperties(addrmgrNs, account)
return err
})
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
if err != nil {
log.Errorf("Cannot fetch new account properties for notification "+
"after account creation: %v", err)
} else {
w.NtfnServer.notifyAccountProperties(props)
}
return account, err
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
2015-04-06 21:03:24 +02:00
// CreditCategory describes the type of wallet transaction output. The category
// of "sent transactions" (debits) is always "send", and is not expressed by
// this type.
//
// TODO: This is a requirement of the RPC server and should be moved.
type CreditCategory byte
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
2015-04-06 21:03:24 +02:00
// These constants define the possible credit categories.
const (
CreditReceive CreditCategory = iota
CreditGenerate
CreditImmature
)
// String returns the category as a string. This string may be used as the
// JSON string for categories as part of listtransactions and gettransaction
// RPC responses.
func (c CreditCategory) String() string {
switch c {
case CreditReceive:
return "receive"
case CreditGenerate:
return "generate"
case CreditImmature:
return "immature"
default:
return "unknown"
}
}
// RecvCategory returns the category of received credit outputs from a
// transaction record. The passed block chain height is used to distinguish
// immature from mature coinbase outputs.
//
// TODO: This is intended for use by the RPC server and should be moved out of
// this package at a later time.
func RecvCategory(details *wtxmgr.TxDetails, syncHeight int32, net *chaincfg.Params) CreditCategory {
2015-04-06 21:03:24 +02:00
if blockchain.IsCoinBaseTx(&details.MsgTx) {
if confirmed(int32(net.CoinbaseMaturity), details.Block.Height,
syncHeight) {
2015-04-06 21:03:24 +02:00
return CreditGenerate
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
2015-04-06 21:03:24 +02:00
return CreditImmature
}
return CreditReceive
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// listTransactions creates a object that may be marshalled to a response result
2015-04-06 21:03:24 +02:00
// for a listtransactions RPC.
//
// TODO: This should be moved to the legacyrpc package.
func listTransactions(tx walletdb.ReadTx, details *wtxmgr.TxDetails, addrMgr *waddrmgr.Manager,
syncHeight int32, net *chaincfg.Params) []btcjson.ListTransactionsResult {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
2015-04-06 21:03:24 +02:00
var (
blockHashStr string
blockTime int64
confirmations int64
2015-04-06 21:03:24 +02:00
)
if details.Block.Height != -1 {
blockHashStr = details.Block.Hash.String()
blockTime = details.Block.Time.Unix()
confirmations = int64(confirms(details.Block.Height, syncHeight))
}
results := []btcjson.ListTransactionsResult{}
txHashStr := details.Hash.String()
received := details.Received.Unix()
generated := blockchain.IsCoinBaseTx(&details.MsgTx)
recvCat := RecvCategory(details, syncHeight, net).String()
2015-04-06 21:03:24 +02:00
send := len(details.Debits) != 0
// Fee can only be determined if every input is a debit.
var feeF64 float64
if len(details.Debits) == len(details.MsgTx.TxIn) {
var debitTotal btcutil.Amount
for _, deb := range details.Debits {
debitTotal += deb.Amount
}
var outputTotal btcutil.Amount
for _, output := range details.MsgTx.TxOut {
outputTotal += btcutil.Amount(output.Value)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// Note: The actual fee is debitTotal - outputTotal. However,
// this RPC reports negative numbers for fees, so the inverse
// is calculated.
feeF64 = (outputTotal - debitTotal).ToBTC()
2015-04-06 21:03:24 +02:00
}
outputs:
for i, output := range details.MsgTx.TxOut {
// Determine if this output is a credit, and if so, determine
// its spentness.
var isCredit bool
var spentCredit bool
for _, cred := range details.Credits {
if cred.Index == uint32(i) {
// Change outputs are ignored.
if cred.Change {
continue outputs
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
2015-04-06 21:03:24 +02:00
isCredit = true
spentCredit = cred.Spent
break
}
}
var address string
var accountName string
2015-04-06 21:03:24 +02:00
_, addrs, _, _ := txscript.ExtractPkScriptAddrs(output.PkScript, net)
if len(addrs) == 1 {
addr := addrs[0]
address = addr.EncodeAddress()
mgr, account, err := addrMgr.AddrAccount(addrmgrNs, addrs[0])
if err == nil {
accountName, err = mgr.AccountName(addrmgrNs, account)
if err != nil {
accountName = ""
}
}
2015-04-06 21:03:24 +02:00
}
amountF64 := btcutil.Amount(output.Value).ToBTC()
result := btcjson.ListTransactionsResult{
// Fields left zeroed:
// InvolvesWatchOnly
// BlockIndex
//
// Fields set below:
// Account (only for non-"send" categories)
// Category
// Amount
// Fee
2015-04-06 21:03:24 +02:00
Address: address,
Vout: uint32(i),
Confirmations: confirmations,
2015-04-06 21:03:24 +02:00
Generated: generated,
BlockHash: blockHashStr,
BlockTime: blockTime,
2015-04-06 21:03:24 +02:00
TxID: txHashStr,
WalletConflicts: []string{},
2015-04-06 21:03:24 +02:00
Time: received,
TimeReceived: received,
}
// Add a received/generated/immature result if this is a credit.
// If the output was spent, create a second result under the
// send category with the inverse of the output amount. It is
// therefore possible that a single output may be included in
// the results set zero, one, or two times.
//
// Since credits are not saved for outputs that are not
// controlled by this wallet, all non-credits from transactions
// with debits are grouped under the send category.
if send || spentCredit {
result.Category = "send"
result.Amount = -amountF64
result.Fee = &feeF64
2015-04-06 21:03:24 +02:00
results = append(results, result)
}
if isCredit {
result.Account = accountName
2015-04-06 21:03:24 +02:00
result.Category = recvCat
result.Amount = amountF64
result.Fee = nil
2015-04-06 21:03:24 +02:00
results = append(results, result)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
}
2015-04-06 21:03:24 +02:00
return results
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
2015-04-06 21:03:24 +02:00
// ListSinceBlock returns a slice of objects with details about transactions
// since the given block. If the block is -1 then all transactions are included.
// This is intended to be used for listsinceblock RPC replies.
func (w *Wallet) ListSinceBlock(start, end, syncHeight int32) ([]btcjson.ListTransactionsResult, error) {
txList := []btcjson.ListTransactionsResult{}
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
rangeFn := func(details []wtxmgr.TxDetails) (bool, error) {
for _, detail := range details {
jsonResults := listTransactions(tx, &detail,
w.Manager, syncHeight, w.chainParams)
txList = append(txList, jsonResults...)
}
return false, nil
2015-04-06 21:03:24 +02:00
}
return w.TxStore.RangeTransactions(txmgrNs, start, end, rangeFn)
2015-04-06 21:03:24 +02:00
})
return txList, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// ListTransactions returns a slice of objects with details about a recorded
// transaction. This is intended to be used for listtransactions RPC
// replies.
func (w *Wallet) ListTransactions(from, count int) ([]btcjson.ListTransactionsResult, error) {
txList := []btcjson.ListTransactionsResult{}
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
// Get current block. The block height used for calculating
// the number of tx confirmations.
syncBlock := w.Manager.SyncedTo()
// Need to skip the first from transactions, and after those, only
// include the next count transactions.
skipped := 0
n := 0
rangeFn := func(details []wtxmgr.TxDetails) (bool, error) {
// Iterate over transactions at this height in reverse order.
// This does nothing for unmined transactions, which are
// unsorted, but it will process mined transactions in the
// reverse order they were marked mined.
for i := len(details) - 1; i >= 0; i-- {
if from > skipped {
skipped++
continue
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
n++
if n > count {
return true, nil
}
jsonResults := listTransactions(tx, &details[i],
w.Manager, syncBlock.Height, w.chainParams)
txList = append(txList, jsonResults...)
if len(jsonResults) > 0 {
n++
}
2015-04-06 21:03:24 +02:00
}
return false, nil
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// Return newer results first by starting at mempool height and working
// down to the genesis block.
return w.TxStore.RangeTransactions(txmgrNs, -1, 0, rangeFn)
2015-04-06 21:03:24 +02:00
})
return txList, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// ListAddressTransactions returns a slice of objects with details about
// recorded transactions to or from any address belonging to a set. This is
// intended to be used for listaddresstransactions RPC replies.
func (w *Wallet) ListAddressTransactions(pkHashes map[string]struct{}) ([]btcjson.ListTransactionsResult, error) {
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
txList := []btcjson.ListTransactionsResult{}
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
// Get current block. The block height used for calculating
// the number of tx confirmations.
syncBlock := w.Manager.SyncedTo()
rangeFn := func(details []wtxmgr.TxDetails) (bool, error) {
loopDetails:
for i := range details {
detail := &details[i]
for _, cred := range detail.Credits {
pkScript := detail.MsgTx.TxOut[cred.Index].PkScript
_, addrs, _, err := txscript.ExtractPkScriptAddrs(
pkScript, w.chainParams)
if err != nil || len(addrs) != 1 {
continue
}
apkh, ok := addrs[0].(*btcutil.AddressPubKeyHash)
if !ok {
continue
}
_, ok = pkHashes[string(apkh.ScriptAddress())]
if !ok {
continue
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
jsonResults := listTransactions(tx, detail,
w.Manager, syncBlock.Height, w.chainParams)
if err != nil {
return false, err
}
txList = append(txList, jsonResults...)
continue loopDetails
2015-04-06 21:03:24 +02:00
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
return false, nil
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
return w.TxStore.RangeTransactions(txmgrNs, 0, -1, rangeFn)
})
2015-04-06 21:03:24 +02:00
return txList, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// ListAllTransactions returns a slice of objects with details about a recorded
// transaction. This is intended to be used for listalltransactions RPC
// replies.
func (w *Wallet) ListAllTransactions() ([]btcjson.ListTransactionsResult, error) {
txList := []btcjson.ListTransactionsResult{}
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
// Get current block. The block height used for calculating
// the number of tx confirmations.
syncBlock := w.Manager.SyncedTo()
rangeFn := func(details []wtxmgr.TxDetails) (bool, error) {
// Iterate over transactions at this height in reverse order.
// This does nothing for unmined transactions, which are
// unsorted, but it will process mined transactions in the
// reverse order they were marked mined.
for i := len(details) - 1; i >= 0; i-- {
jsonResults := listTransactions(tx, &details[i], w.Manager,
syncBlock.Height, w.chainParams)
txList = append(txList, jsonResults...)
}
return false, nil
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
2015-04-06 21:03:24 +02:00
// Return newer results first by starting at mempool height and
// working down to the genesis block.
return w.TxStore.RangeTransactions(txmgrNs, -1, 0, rangeFn)
})
2015-04-06 21:03:24 +02:00
return txList, err
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
// BlockIdentifier identifies a block by either a height or a hash.
type BlockIdentifier struct {
height int32
hash *chainhash.Hash
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
// NewBlockIdentifierFromHeight constructs a BlockIdentifier for a block height.
func NewBlockIdentifierFromHeight(height int32) *BlockIdentifier {
return &BlockIdentifier{height: height}
}
// NewBlockIdentifierFromHash constructs a BlockIdentifier for a block hash.
func NewBlockIdentifierFromHash(hash *chainhash.Hash) *BlockIdentifier {
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
return &BlockIdentifier{hash: hash}
}
// GetTransactionsResult is the result of the wallet's GetTransactions method.
// See GetTransactions for more details.
type GetTransactionsResult struct {
MinedTransactions []Block
UnminedTransactions []TransactionSummary
}
// GetTransactions returns transaction results between a starting and ending
// block. Blocks in the block range may be specified by either a height or a
// hash.
//
// Because this is a possibly lenghtly operation, a cancel channel is provided
// to cancel the task. If this channel unblocks, the results created thus far
// will be returned.
//
// Transaction results are organized by blocks in ascending order and unmined
// transactions in an unspecified order. Mined transactions are saved in a
// Block structure which records properties about the block.
func (w *Wallet) GetTransactions(startBlock, endBlock *BlockIdentifier, cancel <-chan struct{}) (*GetTransactionsResult, error) {
var start, end int32 = 0, -1
w.chainClientLock.Lock()
chainClient := w.chainClient
w.chainClientLock.Unlock()
// TODO: Fetching block heights by their hashes is inherently racy
// because not all block headers are saved but when they are for SPV the
// db can be queried directly without this.
var startResp, endResp rpcclient.FutureGetBlockVerboseResult
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
if startBlock != nil {
if startBlock.hash == nil {
start = startBlock.height
} else {
if chainClient == nil {
return nil, errors.New("no chain server client")
}
switch client := chainClient.(type) {
case *chain.RPCClient:
startResp = client.GetBlockVerboseTxAsync(startBlock.hash)
case *chain.BitcoindClient:
var err error
start, err = client.GetBlockHeight(startBlock.hash)
if err != nil {
return nil, err
}
case *chain.NeutrinoClient:
var err error
start, err = client.GetBlockHeight(startBlock.hash)
if err != nil {
return nil, err
}
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
}
if endBlock != nil {
if endBlock.hash == nil {
end = endBlock.height
} else {
if chainClient == nil {
return nil, errors.New("no chain server client")
}
switch client := chainClient.(type) {
case *chain.RPCClient:
endResp = client.GetBlockVerboseTxAsync(endBlock.hash)
case *chain.NeutrinoClient:
var err error
end, err = client.GetBlockHeight(endBlock.hash)
if err != nil {
return nil, err
}
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
}
if startResp != nil {
resp, err := startResp.Receive()
if err != nil {
return nil, err
}
start = int32(resp.Height)
}
if endResp != nil {
resp, err := endResp.Receive()
if err != nil {
return nil, err
}
end = int32(resp.Height)
}
var res GetTransactionsResult
err := walletdb.View(w.db, func(dbtx walletdb.ReadTx) error {
txmgrNs := dbtx.ReadBucket(wtxmgrNamespaceKey)
rangeFn := func(details []wtxmgr.TxDetails) (bool, error) {
// TODO: probably should make RangeTransactions not reuse the
// details backing array memory.
dets := make([]wtxmgr.TxDetails, len(details))
copy(dets, details)
details = dets
txs := make([]TransactionSummary, 0, len(details))
for i := range details {
txs = append(txs, makeTxSummary(dbtx, w, &details[i]))
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
if details[0].Block.Height != -1 {
blockHash := details[0].Block.Hash
res.MinedTransactions = append(res.MinedTransactions, Block{
Hash: &blockHash,
Height: details[0].Block.Height,
Timestamp: details[0].Block.Time.Unix(),
Transactions: txs,
})
} else {
res.UnminedTransactions = txs
}
select {
case <-cancel:
return true, nil
default:
return false, nil
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
return w.TxStore.RangeTransactions(txmgrNs, start, end, rangeFn)
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
})
return &res, err
}
// AccountResult is a single account result for the AccountsResult type.
type AccountResult struct {
waddrmgr.AccountProperties
TotalBalance btcutil.Amount
}
// AccountsResult is the resutl of the wallet's Accounts method. See that
// method for more details.
type AccountsResult struct {
Accounts []AccountResult
CurrentBlockHash *chainhash.Hash
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
CurrentBlockHeight int32
}
// Accounts returns the current names, numbers, and total balances of all
// accounts in the wallet restricted to a particular key scope. The current
// chain tip is included in the result for atomicity reasons.
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
//
// TODO(jrick): Is the chain tip really needed, since only the total balances
// are included?
func (w *Wallet) Accounts(scope waddrmgr.KeyScope) (*AccountsResult, error) {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return nil, err
}
var (
accounts []AccountResult
syncBlockHash *chainhash.Hash
syncBlockHeight int32
)
err = walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
syncBlock := w.Manager.SyncedTo()
syncBlockHash = &syncBlock.Hash
syncBlockHeight = syncBlock.Height
unspent, err := w.TxStore.UnspentOutputs(txmgrNs)
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
if err != nil {
return err
}
err = manager.ForEachAccount(addrmgrNs, func(acct uint32) error {
props, err := manager.AccountProperties(addrmgrNs, acct)
if err != nil {
return err
}
accounts = append(accounts, AccountResult{
AccountProperties: *props,
// TotalBalance set below
})
return nil
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
})
if err != nil {
return err
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
m := make(map[uint32]*btcutil.Amount)
for i := range accounts {
a := &accounts[i]
m[a.AccountNumber] = &a.TotalBalance
}
for i := range unspent {
output := unspent[i]
var outputAcct uint32
_, addrs, _, err := txscript.ExtractPkScriptAddrs(output.PkScript, w.chainParams)
if err == nil && len(addrs) > 0 {
_, outputAcct, err = w.Manager.AddrAccount(addrmgrNs, addrs[0])
}
if err == nil {
amt, ok := m[outputAcct]
if ok {
*amt += output.Amount
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
}
return nil
})
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
return &AccountsResult{
Accounts: accounts,
CurrentBlockHash: syncBlockHash,
CurrentBlockHeight: syncBlockHeight,
}, err
}
// AccountBalanceResult is a single result for the Wallet.AccountBalances method.
type AccountBalanceResult struct {
AccountNumber uint32
AccountName string
AccountBalance btcutil.Amount
}
// AccountBalances returns all accounts in the wallet and their balances.
// Balances are determined by excluding transactions that have not met
// requiredConfs confirmations.
func (w *Wallet) AccountBalances(scope waddrmgr.KeyScope,
requiredConfs int32) ([]AccountBalanceResult, error) {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return nil, err
}
var results []AccountBalanceResult
err = walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
syncBlock := w.Manager.SyncedTo()
// Fill out all account info except for the balances.
lastAcct, err := manager.LastAccount(addrmgrNs)
if err != nil {
return err
}
results = make([]AccountBalanceResult, lastAcct+2)
for i := range results[:len(results)-1] {
accountName, err := manager.AccountName(addrmgrNs, uint32(i))
if err != nil {
return err
}
results[i].AccountNumber = uint32(i)
results[i].AccountName = accountName
}
results[len(results)-1].AccountNumber = waddrmgr.ImportedAddrAccount
results[len(results)-1].AccountName = waddrmgr.ImportedAddrAccountName
// Fetch all unspent outputs, and iterate over them tallying each
// account's balance where the output script pays to an account address
// and the required number of confirmations is met.
unspentOutputs, err := w.TxStore.UnspentOutputs(txmgrNs)
if err != nil {
return err
}
for i := range unspentOutputs {
output := &unspentOutputs[i]
if !confirmed(requiredConfs, output.Height, syncBlock.Height) {
continue
}
if output.FromCoinBase && !confirmed(int32(w.ChainParams().CoinbaseMaturity),
output.Height, syncBlock.Height) {
continue
}
_, addrs, _, err := txscript.ExtractPkScriptAddrs(output.PkScript, w.chainParams)
if err != nil || len(addrs) == 0 {
continue
}
outputAcct, err := manager.AddrAccount(addrmgrNs, addrs[0])
if err != nil {
continue
}
switch {
case outputAcct == waddrmgr.ImportedAddrAccount:
results[len(results)-1].AccountBalance += output.Amount
case outputAcct > lastAcct:
return errors.New("waddrmgr.Manager.AddrAccount returned account " +
"beyond recorded last account")
default:
results[outputAcct].AccountBalance += output.Amount
}
}
return nil
})
return results, err
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
2015-04-06 21:03:24 +02:00
// creditSlice satisifies the sort.Interface interface to provide sorting
// transaction credits from oldest to newest. Credits with the same receive
// time and mined in the same block are not guaranteed to be sorted by the order
// they appear in the block. Credits from the same transaction are sorted by
// output index.
type creditSlice []wtxmgr.Credit
func (s creditSlice) Len() int {
return len(s)
}
func (s creditSlice) Less(i, j int) bool {
switch {
// If both credits are from the same tx, sort by output index.
case s[i].OutPoint.Hash == s[j].OutPoint.Hash:
return s[i].OutPoint.Index < s[j].OutPoint.Index
// If both transactions are unmined, sort by their received date.
case s[i].Height == -1 && s[j].Height == -1:
return s[i].Received.Before(s[j].Received)
// Unmined (newer) txs always come last.
case s[i].Height == -1:
return false
case s[j].Height == -1:
return true
// If both txs are mined in different blocks, sort by block height.
default:
return s[i].Height < s[j].Height
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
2015-04-06 21:03:24 +02:00
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
2015-04-06 21:03:24 +02:00
func (s creditSlice) Swap(i, j int) {
s[i], s[j] = s[j], s[i]
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// ListUnspent returns a slice of objects representing the unspent wallet
// transactions fitting the given criteria. The confirmations will be more than
// minconf, less than maxconf and if addresses is populated only the addresses
// contained within it will be considered. If we know nothing about a
// transaction an empty array will be returned.
func (w *Wallet) ListUnspent(minconf, maxconf int32,
addresses map[string]struct{}) ([]*btcjson.ListUnspentResult, error) {
var results []*btcjson.ListUnspentResult
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
syncBlock := w.Manager.SyncedTo()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
filter := len(addresses) != 0
unspent, err := w.TxStore.UnspentOutputs(txmgrNs)
if err != nil {
return err
}
sort.Sort(sort.Reverse(creditSlice(unspent)))
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
defaultAccountName := "default"
2015-04-06 21:03:24 +02:00
results = make([]*btcjson.ListUnspentResult, 0, len(unspent))
for i := range unspent {
output := unspent[i]
// Outputs with fewer confirmations than the minimum or more
// confs than the maximum are excluded.
confs := confirms(output.Height, syncBlock.Height)
if confs < minconf || confs > maxconf {
continue
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Only mature coinbase outputs are included.
if output.FromCoinBase {
target := int32(w.ChainParams().CoinbaseMaturity)
if !confirmed(target, output.Height, syncBlock.Height) {
continue
}
}
2015-04-06 21:03:24 +02:00
// Exclude locked outputs from the result set.
if w.LockedOutpoint(output.OutPoint) {
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
continue
}
2015-04-06 21:03:24 +02:00
// Lookup the associated account for the output. Use the
// default account name in case there is no associated account
// for some reason, although this should never happen.
//
// This will be unnecessary once transactions and outputs are
// grouped under the associated account in the db.
acctName := defaultAccountName
sc, addrs, _, err := txscript.ExtractPkScriptAddrs(
output.PkScript, w.chainParams)
if err != nil {
continue
}
if len(addrs) > 0 {
smgr, acct, err := w.Manager.AddrAccount(addrmgrNs, addrs[0])
2015-04-06 21:03:24 +02:00
if err == nil {
s, err := smgr.AccountName(addrmgrNs, acct)
if err == nil {
acctName = s
}
2015-04-06 21:03:24 +02:00
}
}
if filter {
for _, addr := range addrs {
_, ok := addresses[addr.EncodeAddress()]
if ok {
goto include
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
continue
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
include:
// At the moment watch-only addresses are not supported, so all
// recorded outputs that are not multisig are "spendable".
// Multisig outputs are only "spendable" if all keys are
// controlled by this wallet.
//
// TODO: Each case will need updates when watch-only addrs
// is added. For P2PK, P2PKH, and P2SH, the address must be
// looked up and not be watching-only. For multisig, all
// pubkeys must belong to the manager with the associated
// private key (currently it only checks whether the pubkey
// exists, since the private key is required at the moment).
var spendable bool
scSwitch:
switch sc {
case txscript.PubKeyHashTy:
spendable = true
case txscript.PubKeyTy:
spendable = true
case txscript.WitnessV0ScriptHashTy:
spendable = true
case txscript.WitnessV0PubKeyHashTy:
spendable = true
case txscript.MultiSigTy:
for _, a := range addrs {
_, err := w.Manager.Address(addrmgrNs, a)
if err == nil {
continue
}
if waddrmgr.IsError(err, waddrmgr.ErrAddressNotFound) {
break scSwitch
}
return err
}
spendable = true
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
result := &btcjson.ListUnspentResult{
TxID: output.OutPoint.Hash.String(),
Vout: output.OutPoint.Index,
Account: acctName,
ScriptPubKey: hex.EncodeToString(output.PkScript),
Amount: output.Amount.ToBTC(),
Confirmations: int64(confs),
Spendable: spendable,
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// BUG: this should be a JSON array so that all
// addresses can be included, or removed (and the
// caller extracts addresses from the pkScript).
if len(addrs) > 0 {
result.Address = addrs[0].EncodeAddress()
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
results = append(results, result)
}
return nil
})
return results, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// DumpPrivKeys returns the WIF-encoded private keys for all addresses with
// private keys in a wallet.
func (w *Wallet) DumpPrivKeys() ([]string, error) {
var privkeys []string
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
// Iterate over each active address, appending the private key to
// privkeys.
return w.Manager.ForEachActiveAddress(addrmgrNs, func(addr btcutil.Address) error {
ma, err := w.Manager.Address(addrmgrNs, addr)
if err != nil {
return err
}
// Only those addresses with keys needed.
pka, ok := ma.(waddrmgr.ManagedPubKeyAddress)
if !ok {
return nil
}
wif, err := pka.ExportPrivKey()
if err != nil {
// It would be nice to zero out the array here. However,
// since strings in go are immutable, and we have no
// control over the caller I don't think we can. :(
return err
}
privkeys = append(privkeys, wif.String())
return nil
})
})
return privkeys, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// DumpWIFPrivateKey returns the WIF encoded private key for a
// single wallet address.
func (w *Wallet) DumpWIFPrivateKey(addr btcutil.Address) (string, error) {
var maddr waddrmgr.ManagedAddress
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
waddrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
// Get private key from wallet if it exists.
var err error
maddr, err = w.Manager.Address(waddrmgrNs, addr)
return err
})
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
if err != nil {
return "", err
}
pka, ok := maddr.(waddrmgr.ManagedPubKeyAddress)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
if !ok {
return "", fmt.Errorf("address %s is not a key type", addr)
}
wif, err := pka.ExportPrivKey()
if err != nil {
return "", err
}
return wif.String(), nil
}
// ImportPrivateKey imports a private key to the wallet and writes the new
// wallet to disk.
//
// NOTE: If a block stamp is not provided, then the wallet's birthday will be
// set to the genesis block of the corresponding chain.
func (w *Wallet) ImportPrivateKey(scope waddrmgr.KeyScope, wif *btcutil.WIF,
bs *waddrmgr.BlockStamp, rescan bool) (string, error) {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return "", err
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// The starting block for the key is the genesis block unless otherwise
// specified.
if bs == nil {
bs = &waddrmgr.BlockStamp{
Hash: *w.chainParams.GenesisHash,
Height: 0,
Timestamp: w.chainParams.GenesisBlock.Header.Timestamp,
}
} else if bs.Timestamp.IsZero() {
// Only update the new birthday time from default value if we
// actually have timestamp info in the header.
header, err := w.chainClient.GetBlockHeader(&bs.Hash)
if err == nil {
bs.Timestamp = header.Timestamp
}
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Attempt to import private key into wallet.
var addr btcutil.Address
var props *waddrmgr.AccountProperties
err = walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
addrmgrNs := tx.ReadWriteBucket(waddrmgrNamespaceKey)
maddr, err := manager.ImportPrivateKey(addrmgrNs, wif, bs)
if err != nil {
return err
}
addr = maddr.Address()
props, err = manager.AccountProperties(
addrmgrNs, waddrmgr.ImportedAddrAccount,
)
if err != nil {
return err
}
// We'll only update our birthday with the new one if it is
// before our current one. Otherwise, if we do, we can
// potentially miss detecting relevant chain events that
// occurred between them while rescanning.
birthdayBlock, _, err := w.Manager.BirthdayBlock(addrmgrNs)
if err != nil {
return err
}
if bs.Height >= birthdayBlock.Height {
return nil
}
err = w.Manager.SetBirthday(addrmgrNs, bs.Timestamp)
if err != nil {
return err
}
// To ensure this birthday block is correct, we'll mark it as
// unverified to prompt a sanity check at the next restart to
// ensure it is correct as it was provided by the caller.
return w.Manager.SetBirthdayBlock(addrmgrNs, *bs, false)
})
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
if err != nil {
return "", err
}
// Rescan blockchain for transactions with txout scripts paying to the
// imported address.
if rescan {
job := &RescanJob{
Addrs: []btcutil.Address{addr},
OutPoints: nil,
BlockStamp: *bs,
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// Submit rescan job and log when the import has completed.
// Do not block on finishing the rescan. The rescan success
// or failure is logged elsewhere, and the channel is not
// required to be read, so discard the return value.
_ = w.SubmitRescan(job)
} else {
err := w.chainClient.NotifyReceived([]btcutil.Address{addr})
if err != nil {
return "", fmt.Errorf("Failed to subscribe for address ntfns for "+
"address %s: %s", addr.EncodeAddress(), err)
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
addrStr := addr.EncodeAddress()
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
log.Infof("Imported payment address %s", addrStr)
w.NtfnServer.notifyAccountProperties(props)
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Return the payment address string of the imported private key.
return addrStr, nil
}
// LockedOutpoint returns whether an outpoint has been marked as locked and
// should not be used as an input for created transactions.
func (w *Wallet) LockedOutpoint(op wire.OutPoint) bool {
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
_, locked := w.lockedOutpoints[op]
return locked
}
// LockOutpoint marks an outpoint as locked, that is, it should not be used as
// an input for newly created transactions.
func (w *Wallet) LockOutpoint(op wire.OutPoint) {
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
w.lockedOutpoints[op] = struct{}{}
}
// UnlockOutpoint marks an outpoint as unlocked, that is, it may be used as an
// input for newly created transactions.
func (w *Wallet) UnlockOutpoint(op wire.OutPoint) {
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
delete(w.lockedOutpoints, op)
}
// ResetLockedOutpoints resets the set of locked outpoints so all may be used
// as inputs for new transactions.
func (w *Wallet) ResetLockedOutpoints() {
w.lockedOutpoints = map[wire.OutPoint]struct{}{}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// LockedOutpoints returns a slice of currently locked outpoints. This is
// intended to be used by marshaling the result as a JSON array for
// listlockunspent RPC results.
func (w *Wallet) LockedOutpoints() []btcjson.TransactionInput {
locked := make([]btcjson.TransactionInput, len(w.lockedOutpoints))
i := 0
for op := range w.lockedOutpoints {
locked[i] = btcjson.TransactionInput{
Txid: op.Hash.String(),
Vout: op.Index,
}
i++
}
return locked
}
// resendUnminedTxs iterates through all transactions that spend from wallet
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// credits that are not known to have been mined into a block, and attempts
// to send each to the chain server for relay.
func (w *Wallet) resendUnminedTxs() {
var txs []*wire.MsgTx
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
var err error
txs, err = w.TxStore.UnminedTxs(txmgrNs)
return err
})
2015-04-06 21:03:24 +02:00
if err != nil {
log.Errorf("Unable to retrieve unconfirmed transactions to "+
"resend: %v", err)
2015-04-06 21:03:24 +02:00
return
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
for _, tx := range txs {
txHash, err := w.publishTransaction(tx)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
if err != nil {
log.Debugf("Unable to rebroadcast transaction %v: %v",
tx.TxHash(), err)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
continue
}
log.Debugf("Successfully rebroadcast unconfirmed transaction %v",
txHash)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
}
// SortedActivePaymentAddresses returns a slice of all active payment
// addresses in a wallet.
func (w *Wallet) SortedActivePaymentAddresses() ([]string, error) {
var addrStrs []string
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
return w.Manager.ForEachActiveAddress(addrmgrNs, func(addr btcutil.Address) error {
addrStrs = append(addrStrs, addr.EncodeAddress())
return nil
})
})
if err != nil {
return nil, err
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
sort.Sort(sort.StringSlice(addrStrs))
return addrStrs, nil
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// NewAddress returns the next external chained address for a wallet.
func (w *Wallet) NewAddress(account uint32,
scope waddrmgr.KeyScope) (btcutil.Address, error) {
chainClient, err := w.requireChainClient()
if err != nil {
return nil, err
}
var (
addr btcutil.Address
props *waddrmgr.AccountProperties
)
err = walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
addrmgrNs := tx.ReadWriteBucket(waddrmgrNamespaceKey)
var err error
addr, props, err = w.newAddress(addrmgrNs, account, scope)
return err
})
if err != nil {
return nil, err
}
// Notify the rpc server about the newly created address.
err = chainClient.NotifyReceived([]btcutil.Address{addr})
if err != nil {
return nil, err
}
w.NtfnServer.notifyAccountProperties(props)
return addr, nil
}
func (w *Wallet) newAddress(addrmgrNs walletdb.ReadWriteBucket, account uint32,
scope waddrmgr.KeyScope) (btcutil.Address, *waddrmgr.AccountProperties, error) {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return nil, nil, err
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
// Get next address from wallet.
addrs, err := manager.NextExternalAddresses(addrmgrNs, account, 1)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
if err != nil {
return nil, nil, err
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
props, err := manager.AccountProperties(addrmgrNs, account)
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
if err != nil {
log.Errorf("Cannot fetch account properties for notification "+
"after deriving next external address: %v", err)
return nil, nil, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
return addrs[0].Address(), props, nil
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// NewChangeAddress returns a new change address for a wallet.
func (w *Wallet) NewChangeAddress(account uint32,
scope waddrmgr.KeyScope) (btcutil.Address, error) {
chainClient, err := w.requireChainClient()
if err != nil {
return nil, err
}
var addr btcutil.Address
err = walletdb.Update(w.db, func(tx walletdb.ReadWriteTx) error {
addrmgrNs := tx.ReadWriteBucket(waddrmgrNamespaceKey)
var err error
addr, err = w.newChangeAddress(addrmgrNs, account, scope)
return err
})
if err != nil {
return nil, err
}
// Notify the rpc server about the newly created address.
err = chainClient.NotifyReceived([]btcutil.Address{addr})
if err != nil {
return nil, err
}
return addr, nil
}
// newChangeAddress returns a new change address for the wallet.
//
// NOTE: This method requires the caller to use the backend's NotifyReceived
// method in order to detect when an on-chain transaction pays to the address
// being created.
func (w *Wallet) newChangeAddress(addrmgrNs walletdb.ReadWriteBucket,
account uint32, scope waddrmgr.KeyScope) (btcutil.Address, error) {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return nil, err
}
// Get next chained change address from wallet for account.
addrs, err := manager.NextInternalAddresses(addrmgrNs, account, 1)
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
if err != nil {
return nil, err
}
return addrs[0].Address(), nil
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
2015-04-06 21:03:24 +02:00
// confirmed checks whether a transaction at height txHeight has met minconf
// confirmations for a blockchain at height curHeight.
func confirmed(minconf, txHeight, curHeight int32) bool {
return confirms(txHeight, curHeight) >= minconf
}
// confirms returns the number of confirmations for a transaction in a block at
// height txHeight (or -1 for an unconfirmed tx) given the chain height
// curHeight.
func confirms(txHeight, curHeight int32) int32 {
switch {
case txHeight == -1, txHeight > curHeight:
return 0
default:
return curHeight - txHeight + 1
}
}
// AccountTotalReceivedResult is a single result for the
// Wallet.TotalReceivedForAccounts method.
type AccountTotalReceivedResult struct {
AccountNumber uint32
AccountName string
TotalReceived btcutil.Amount
LastConfirmation int32
}
2015-04-06 21:03:24 +02:00
// TotalReceivedForAccounts iterates through a wallet's transaction history,
// returning the total amount of Bitcoin received for all accounts.
func (w *Wallet) TotalReceivedForAccounts(scope waddrmgr.KeyScope,
minConf int32) ([]AccountTotalReceivedResult, error) {
manager, err := w.Manager.FetchScopedKeyManager(scope)
if err != nil {
return nil, err
}
var results []AccountTotalReceivedResult
err = walletdb.View(w.db, func(tx walletdb.ReadTx) error {
addrmgrNs := tx.ReadBucket(waddrmgrNamespaceKey)
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
2015-04-06 21:03:24 +02:00
syncBlock := w.Manager.SyncedTo()
err := manager.ForEachAccount(addrmgrNs, func(account uint32) error {
accountName, err := manager.AccountName(addrmgrNs, account)
if err != nil {
return err
}
results = append(results, AccountTotalReceivedResult{
AccountNumber: account,
AccountName: accountName,
})
return nil
})
if err != nil {
return err
}
var stopHeight int32
if minConf > 0 {
stopHeight = syncBlock.Height - minConf + 1
} else {
stopHeight = -1
}
rangeFn := func(details []wtxmgr.TxDetails) (bool, error) {
for i := range details {
detail := &details[i]
for _, cred := range detail.Credits {
pkScript := detail.MsgTx.TxOut[cred.Index].PkScript
var outputAcct uint32
_, addrs, _, err := txscript.ExtractPkScriptAddrs(pkScript, w.chainParams)
if err == nil && len(addrs) > 0 {
_, outputAcct, err = w.Manager.AddrAccount(addrmgrNs, addrs[0])
}
if err == nil {
acctIndex := int(outputAcct)
if outputAcct == waddrmgr.ImportedAddrAccount {
acctIndex = len(results) - 1
}
res := &results[acctIndex]
res.TotalReceived += cred.Amount
res.LastConfirmation = confirms(
detail.Block.Height, syncBlock.Height)
}
2015-04-06 21:03:24 +02:00
}
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
return false, nil
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
return w.TxStore.RangeTransactions(txmgrNs, 0, stopHeight, rangeFn)
2015-04-06 21:03:24 +02:00
})
return results, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
// TotalReceivedForAddr iterates through a wallet's transaction history,
// returning the total amount of bitcoins received for a single wallet
// address.
2015-04-06 21:03:24 +02:00
func (w *Wallet) TotalReceivedForAddr(addr btcutil.Address, minConf int32) (btcutil.Amount, error) {
var amount btcutil.Amount
err := walletdb.View(w.db, func(tx walletdb.ReadTx) error {
txmgrNs := tx.ReadBucket(wtxmgrNamespaceKey)
2015-04-06 21:03:24 +02:00
syncBlock := w.Manager.SyncedTo()
2015-04-06 21:03:24 +02:00
var (
addrStr = addr.EncodeAddress()
stopHeight int32
)
if minConf > 0 {
stopHeight = syncBlock.Height - minConf + 1
} else {
stopHeight = -1
}
rangeFn := func(details []wtxmgr.TxDetails) (bool, error) {
for i := range details {
detail := &details[i]
for _, cred := range detail.Credits {
pkScript := detail.MsgTx.TxOut[cred.Index].PkScript
_, addrs, _, err := txscript.ExtractPkScriptAddrs(pkScript,
w.chainParams)
// An error creating addresses from the output script only
// indicates a non-standard script, so ignore this credit.
if err != nil {
continue
}
for _, a := range addrs {
if addrStr == a.EncodeAddress() {
amount += cred.Amount
break
}
2015-04-06 21:03:24 +02:00
}
}
}
return false, nil
}
return w.TxStore.RangeTransactions(txmgrNs, 0, stopHeight, rangeFn)
2015-04-06 21:03:24 +02:00
})
return amount, err
Remove account support, fix races on btcd connect. This commit is the result of several big changes being made to the wallet. In particular, the "handshake" (initial sync to the chain server) was quite racy and required proper synchronization. To make fixing this race easier, several other changes were made to the internal wallet data structures and much of the RPC server ended up being rewritten. First, all account support has been removed. The previous Account struct has been replaced with a Wallet structure, which includes a keystore for saving keys, and a txstore for storing relevant transactions. This decision has been made since it is the opinion of myself and other developers that bitcoind accounts are fundamentally broken (as accounts implemented by bitcoind support both arbitrary address groupings as well as moving balances between accounts -- these are fundamentally incompatible features), and since a BIP0032 keystore is soon planned to be implemented (at which point, "accounts" can return as HD extended keys). With the keystore handling the grouping of related keys, there is no reason have many different Account structs, and the AccountManager has been removed as well. All RPC handlers that take an account option will only work with "" (the default account) or "*" if the RPC allows specifying all accounts. Second, much of the RPC server has been cleaned up. The global variables for the RPC server and chain server client have been moved to part of the rpcServer struct, and the handlers for each RPC method that are looked up change depending on which components have been set. Passthrough requests are also no longer handled specially, but when the chain server is set, a handler to perform the passthrough will be returned if the method is not otherwise a wallet RPC. The notification system for websocket clients has also been rewritten so wallet components can send notifications through channels, rather than requiring direct access to the RPC server itself, or worse still, sending directly to a websocket client's send channel. In the future, this will enable proper registration of notifications, rather than unsolicited broadcasts to every connected websocket client (see issue #84). Finally, and the main reason why much of this cleanup was necessary, the races during intial sync with the chain server have been fixed. Previously, when the 'Handshake' was run, a rescan would occur which would perform modifications to Account data structures as notifications were received. Synchronization was provided with a single binary semaphore which serialized all access to wallet and account data. However, the Handshake itself was not able to run with this lock (or else notifications would block), and many data races would occur as both notifications were being handled. If GOMAXPROCS was ever increased beyond 1, btcwallet would always immediately crash due to invalid addresses caused by the data races on startup. To fix this, the single lock for all wallet access has been replaced with mutexes for both the keystore and txstore. Handling of btcd notifications and client requests may now occur simultaneously. GOMAXPROCS has also been set to the number of logical CPUs at the beginning of main, since with the data races fixed, there's no reason to prevent the extra parallelism gained by increasing it. Closes #78. Closes #101. Closes #110.
2014-07-09 05:17:38 +02:00
}
Refactor wallet transaction creation code. This began as a change to improve the fee calculation code and evolved into a much larger refactor which improves the readability and modularity of all of the transaction creation code. Transaction fee calculations have been switched from full increments of the relay fee to a proportion based on the transaction size. This means that for a relay fee of 1e3 satoshis/kB, a 500 byte transaction is only required to pay a 5e2 satoshi fee and a 1500 byte transaction only need pay a 1.5e3 fee. The previous code would end up estimating these fees to be 1e3 and 2e3 respectively. Because the previous code would add more fee than needed in almost every case, the transaction size estimations were optimistic (best/smallest case) and signing was done in a loop where the fee was incremented by the relay fee again each time the actual size of the signed transaction rendered the fee too low. This has switched to using worst case transaction size estimates rather than best case, and signing is only performed once. Transaction input signature creation has switched from using txscript.SignatureScript to txscript.SignTxOutput. The new API is able to redeem outputs other than just P2PKH, so the previous restrictions about P2SH outputs being unspendable (except through the signrawtransaction RPC) no longer hold. Several new public packages have been added: wallet/txauthor - transaction authoring and signing wallet/txfees - fee estimations and change output inclusion wallet/txrules - simple consensus and mempool policy rule checks Along with some internal packages: wallet/internal/txsizes - transaction size estimation internal/helpers - context free convenience functions The txsizes package is internal as the estimations it provides are specific for the algorithms used by these new packages.
2016-02-28 05:30:56 +01:00
// SendOutputs creates and sends payment transactions. It returns the
// transaction upon success.
Refactor wallet transaction creation code. This began as a change to improve the fee calculation code and evolved into a much larger refactor which improves the readability and modularity of all of the transaction creation code. Transaction fee calculations have been switched from full increments of the relay fee to a proportion based on the transaction size. This means that for a relay fee of 1e3 satoshis/kB, a 500 byte transaction is only required to pay a 5e2 satoshi fee and a 1500 byte transaction only need pay a 1.5e3 fee. The previous code would end up estimating these fees to be 1e3 and 2e3 respectively. Because the previous code would add more fee than needed in almost every case, the transaction size estimations were optimistic (best/smallest case) and signing was done in a loop where the fee was incremented by the relay fee again each time the actual size of the signed transaction rendered the fee too low. This has switched to using worst case transaction size estimates rather than best case, and signing is only performed once. Transaction input signature creation has switched from using txscript.SignatureScript to txscript.SignTxOutput. The new API is able to redeem outputs other than just P2PKH, so the previous restrictions about P2SH outputs being unspendable (except through the signrawtransaction RPC) no longer hold. Several new public packages have been added: wallet/txauthor - transaction authoring and signing wallet/txfees - fee estimations and change output inclusion wallet/txrules - simple consensus and mempool policy rule checks Along with some internal packages: wallet/internal/txsizes - transaction size estimation internal/helpers - context free convenience functions The txsizes package is internal as the estimations it provides are specific for the algorithms used by these new packages.
2016-02-28 05:30:56 +01:00
func (w *Wallet) SendOutputs(outputs []*wire.TxOut, account uint32,
minconf int32, satPerKb btcutil.Amount) (*wire.MsgTx, error) {
// Ensure the outputs to be created adhere to the network's consensus
// rules.
Refactor wallet transaction creation code. This began as a change to improve the fee calculation code and evolved into a much larger refactor which improves the readability and modularity of all of the transaction creation code. Transaction fee calculations have been switched from full increments of the relay fee to a proportion based on the transaction size. This means that for a relay fee of 1e3 satoshis/kB, a 500 byte transaction is only required to pay a 5e2 satoshi fee and a 1500 byte transaction only need pay a 1.5e3 fee. The previous code would end up estimating these fees to be 1e3 and 2e3 respectively. Because the previous code would add more fee than needed in almost every case, the transaction size estimations were optimistic (best/smallest case) and signing was done in a loop where the fee was incremented by the relay fee again each time the actual size of the signed transaction rendered the fee too low. This has switched to using worst case transaction size estimates rather than best case, and signing is only performed once. Transaction input signature creation has switched from using txscript.SignatureScript to txscript.SignTxOutput. The new API is able to redeem outputs other than just P2PKH, so the previous restrictions about P2SH outputs being unspendable (except through the signrawtransaction RPC) no longer hold. Several new public packages have been added: wallet/txauthor - transaction authoring and signing wallet/txfees - fee estimations and change output inclusion wallet/txrules - simple consensus and mempool policy rule checks Along with some internal packages: wallet/internal/txsizes - transaction size estimation internal/helpers - context free convenience functions The txsizes package is internal as the estimations it provides are specific for the algorithms used by these new packages.
2016-02-28 05:30:56 +01:00
for _, output := range outputs {
err := txrules.CheckOutput(
output, txrules.DefaultRelayFeePerKb,
)
if err != nil {
Refactor wallet transaction creation code. This began as a change to improve the fee calculation code and evolved into a much larger refactor which improves the readability and modularity of all of the transaction creation code. Transaction fee calculations have been switched from full increments of the relay fee to a proportion based on the transaction size. This means that for a relay fee of 1e3 satoshis/kB, a 500 byte transaction is only required to pay a 5e2 satoshi fee and a 1500 byte transaction only need pay a 1.5e3 fee. The previous code would end up estimating these fees to be 1e3 and 2e3 respectively. Because the previous code would add more fee than needed in almost every case, the transaction size estimations were optimistic (best/smallest case) and signing was done in a loop where the fee was incremented by the relay fee again each time the actual size of the signed transaction rendered the fee too low. This has switched to using worst case transaction size estimates rather than best case, and signing is only performed once. Transaction input signature creation has switched from using txscript.SignatureScript to txscript.SignTxOutput. The new API is able to redeem outputs other than just P2PKH, so the previous restrictions about P2SH outputs being unspendable (except through the signrawtransaction RPC) no longer hold. Several new public packages have been added: wallet/txauthor - transaction authoring and signing wallet/txfees - fee estimations and change output inclusion wallet/txrules - simple consensus and mempool policy rule checks Along with some internal packages: wallet/internal/txsizes - transaction size estimation internal/helpers - context free convenience functions The txsizes package is internal as the estimations it provides are specific for the algorithms used by these new packages.
2016-02-28 05:30:56 +01:00
return nil, err
}
}
// Create the transaction and broadcast it to the network. The
// transaction will be added to the database in order to ensure that we
// continue to re-broadcast the transaction upon restarts until it has
// been confirmed.
createdTx, err := w.CreateSimpleTx(
account, outputs, minconf, satPerKb, false,
)
if err != nil {
return nil, err
}
txHash, err := w.reliablyPublishTransaction(createdTx.Tx)
if err != nil {
return nil, err
}
// Sanity check on the returned tx hash.
if *txHash != createdTx.Tx.TxHash() {
return nil, errors.New("tx hash mismatch")
}
return createdTx.Tx, nil
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
// SignatureError records the underlying error when validating a transaction
// input signature.
type SignatureError struct {
InputIndex uint32
Error error
}
// SignTransaction uses secrets of the wallet, as well as additional secrets
// passed in by the caller, to create and add input signatures to a transaction.
//
// Transaction input script validation is used to confirm that all signatures
// are valid. For any invalid input, a SignatureError is added to the returns.
// The final error return is reserved for unexpected or fatal errors, such as
// being unable to determine a previous output script to redeem.
//
// The transaction pointed to by tx is modified by this function.
func (w *Wallet) SignTransaction(tx *wire.MsgTx, hashType txscript.SigHashType,
additionalPrevScripts map[wire.OutPoint][]byte,
additionalKeysByAddress map[string]*btcutil.WIF,
p2shRedeemScriptsByAddress map[string][]byte) ([]SignatureError, error) {
var signErrors []SignatureError
err := walletdb.View(w.db, func(dbtx walletdb.ReadTx) error {
addrmgrNs := dbtx.ReadBucket(waddrmgrNamespaceKey)
txmgrNs := dbtx.ReadBucket(wtxmgrNamespaceKey)
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
for i, txIn := range tx.TxIn {
prevOutScript, ok := additionalPrevScripts[txIn.PreviousOutPoint]
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
if !ok {
prevHash := &txIn.PreviousOutPoint.Hash
prevIndex := txIn.PreviousOutPoint.Index
txDetails, err := w.TxStore.TxDetails(txmgrNs, prevHash)
if err != nil {
return fmt.Errorf("cannot query previous transaction "+
"details for %v: %v", txIn.PreviousOutPoint, err)
}
if txDetails == nil {
return fmt.Errorf("%v not found",
txIn.PreviousOutPoint)
}
prevOutScript = txDetails.MsgTx.TxOut[prevIndex].PkScript
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
// Set up our callbacks that we pass to txscript so it can
// look up the appropriate keys and scripts by address.
getKey := txscript.KeyClosure(func(addr btcutil.Address) (*btcec.PrivateKey, bool, error) {
if len(additionalKeysByAddress) != 0 {
addrStr := addr.EncodeAddress()
wif, ok := additionalKeysByAddress[addrStr]
if !ok {
return nil, false,
errors.New("no key for address")
}
return wif.PrivKey, wif.CompressPubKey, nil
}
address, err := w.Manager.Address(addrmgrNs, addr)
if err != nil {
return nil, false, err
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
pka, ok := address.(waddrmgr.ManagedPubKeyAddress)
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
if !ok {
return nil, false, fmt.Errorf("address %v is not "+
"a pubkey address", address.Address().EncodeAddress())
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
key, err := pka.PrivKey()
if err != nil {
return nil, false, err
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
return key, pka.Compressed(), nil
})
getScript := txscript.ScriptClosure(func(addr btcutil.Address) ([]byte, error) {
// If keys were provided then we can only use the
// redeem scripts provided with our inputs, too.
if len(additionalKeysByAddress) != 0 {
addrStr := addr.EncodeAddress()
script, ok := p2shRedeemScriptsByAddress[addrStr]
if !ok {
return nil, errors.New("no script for address")
}
return script, nil
}
address, err := w.Manager.Address(addrmgrNs, addr)
if err != nil {
return nil, err
}
sa, ok := address.(waddrmgr.ManagedScriptAddress)
if !ok {
return nil, errors.New("address is not a script" +
" address")
}
return sa.Script()
})
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
// SigHashSingle inputs can only be signed if there's a
// corresponding output. However this could be already signed,
// so we always verify the output.
if (hashType&txscript.SigHashSingle) !=
txscript.SigHashSingle || i < len(tx.TxOut) {
script, err := txscript.SignTxOutput(w.ChainParams(),
tx, i, prevOutScript, hashType, getKey,
getScript, txIn.SignatureScript)
// Failure to sign isn't an error, it just means that
// the tx isn't complete.
if err != nil {
signErrors = append(signErrors, SignatureError{
InputIndex: uint32(i),
Error: err,
})
continue
}
txIn.SignatureScript = script
}
// Either it was already signed or we just signed it.
// Find out if it is completely satisfied or still needs more.
vm, err := txscript.NewEngine(prevOutScript, tx, i,
2017-04-22 02:26:42 +02:00
txscript.StandardVerifyFlags, nil, nil, 0)
if err == nil {
err = vm.Execute()
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
if err != nil {
signErrors = append(signErrors, SignatureError{
InputIndex: uint32(i),
Error: err,
})
}
}
return nil
})
return signErrors, err
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
// ErrDoubleSpend is an error returned from PublishTransaction in case the
// published transaction failed to propagate since it was double spending a
// confirmed transaction or a transaction in the mempool.
type ErrDoubleSpend struct {
backendError error
}
// Error returns the string representation of ErrDoubleSpend.
//
// NOTE: Satisfies the error interface.
func (e *ErrDoubleSpend) Error() string {
return fmt.Sprintf("double spend: %v", e.backendError)
}
// Unwrap returns the underlying error returned from the backend.
func (e *ErrDoubleSpend) Unwrap() error {
return e.backendError
}
// ErrReplacement is an error returned from PublishTransaction in case the
// published transaction failed to propagate since it was double spending a
// replacable transaction but did not satisfy the requirements to replace it.
type ErrReplacement struct {
backendError error
}
// Error returns the string representation of ErrReplacement.
//
// NOTE: Satisfies the error interface.
func (e *ErrReplacement) Error() string {
return fmt.Sprintf("unable to replace transaction: %v", e.backendError)
}
// Unwrap returns the underlying error returned from the backend.
func (e *ErrReplacement) Unwrap() error {
return e.backendError
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
// PublishTransaction sends the transaction to the consensus RPC server so it
// can be propagated to other nodes and eventually mined.
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
//
// This function is unstable and will be removed once syncing code is moved out
// of the wallet.
func (w *Wallet) PublishTransaction(tx *wire.MsgTx) error {
_, err := w.reliablyPublishTransaction(tx)
return err
}
// reliablyPublishTransaction is a superset of publishTransaction which contains
// the primary logic required for publishing a transaction, updating the
// relevant database state, and finally possible removing the transaction from
// the database (along with cleaning up all inputs used, and outputs created) if
// the transaction is rejected by the backend.
func (w *Wallet) reliablyPublishTransaction(tx *wire.MsgTx) (*chainhash.Hash, error) {
chainClient, err := w.requireChainClient()
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
if err != nil {
return nil, err
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
// As we aim for this to be general reliable transaction broadcast API,
// we'll write this tx to disk as an unconfirmed transaction. This way,
// upon restarts, we'll always rebroadcast it, and also add it to our
// set of records.
txRec, err := wtxmgr.NewTxRecordFromMsgTx(tx, time.Now())
if err != nil {
return nil, err
}
err = walletdb.Update(w.db, func(dbTx walletdb.ReadWriteTx) error {
return w.addRelevantTx(dbTx, txRec, nil)
})
if err != nil {
return nil, err
}
// We'll also ask to be notified of the transaction once it confirms
// on-chain. This is done outside of the database transaction to prevent
// backend interaction within it.
//
// NOTE: In some cases, it's possible that the transaction to be
// broadcast is not directly relevant to the user's wallet, e.g.,
// multisig. In either case, we'll still ask to be notified of when it
// confirms to maintain consistency.
//
// TODO(wilmer): import script as external if the address does not
// belong to the wallet to handle confs during restarts?
for _, txOut := range tx.TxOut {
_, addrs, _, err := txscript.ExtractPkScriptAddrs(
txOut.PkScript, w.chainParams,
)
if err != nil {
// Non-standard outputs can safely be skipped because
// they're not supported by the wallet.
continue
}
if err := chainClient.NotifyReceived(addrs); err != nil {
return nil, err
}
}
return w.publishTransaction(tx)
}
// publishTransaction attempts to send an unconfirmed transaction to the
// wallet's current backend. In the event that sending the transaction fails for
// whatever reason, it will be removed from the wallet's unconfirmed transaction
// store.
func (w *Wallet) publishTransaction(tx *wire.MsgTx) (*chainhash.Hash, error) {
chainClient, err := w.requireChainClient()
if err != nil {
return nil, err
}
// match is a helper method to easily string match on the error
// message.
match := func(err error, s string) bool {
return strings.Contains(strings.ToLower(err.Error()), s)
}
_, err = chainClient.SendRawTransaction(tx, false)
// Determine if this was an RPC error thrown due to the transaction
// already confirming.
var rpcTxConfirmed bool
if rpcErr, ok := err.(*btcjson.RPCError); ok {
rpcTxConfirmed = rpcErr.Code == btcjson.ErrRPCTxAlreadyInChain
}
var (
txid = tx.TxHash()
returnErr error
)
switch {
case err == nil:
return &txid, nil
// Since we have different backends that can be used with the wallet,
// we'll need to check specific errors for each one.
//
// If the transaction is already in the mempool, we can just return now.
//
// This error is returned when broadcasting/sending a transaction to a
// btcd node that already has it in their mempool.
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/mempool/mempool.go#L953
case match(err, "already have transaction"):
fallthrough
// This error is returned when broadcasting a transaction to a bitcoind
// node that already has it in their mempool.
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/validation.cpp#L590
case match(err, "txn-already-in-mempool"):
return &txid, nil
// If the transaction has already confirmed, we can safely remove it
// from the unconfirmed store as it should already exist within the
// confirmed store. We'll avoid returning an error as the broadcast was
// in a sense successful.
//
// This error is returned when sending a transaction that has already
// confirmed to a btcd/bitcoind node over RPC.
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/rpcserver.go#L3355
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/node/transaction.cpp#L36
case rpcTxConfirmed:
fallthrough
// This error is returned when broadcasting a transaction that has
// already confirmed to a btcd node over the P2P network.
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/mempool/mempool.go#L1036
case match(err, "transaction already exists"):
fallthrough
// This error is returned when broadcasting a transaction that has
// already confirmed to a bitcoind node over the P2P network.
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/validation.cpp#L648
case match(err, "txn-already-known"):
dbErr := walletdb.Update(w.db, func(dbTx walletdb.ReadWriteTx) error {
txmgrNs := dbTx.ReadWriteBucket(wtxmgrNamespaceKey)
txRec, err := wtxmgr.NewTxRecordFromMsgTx(tx, time.Now())
if err != nil {
return err
}
return w.TxStore.RemoveUnminedTx(txmgrNs, txRec)
})
if dbErr != nil {
log.Warnf("Unable to remove confirmed transaction %v "+
"from unconfirmed store: %v", tx.TxHash(), dbErr)
}
return &txid, nil
// If the transactions is invalid since it attempts to double spend a
// transaction already in the mempool or in the chain, we'll remove it
// from the store and return an error.
//
// This error is returned from btcd when there is already a transaction
// not signaling replacement in the mempool that spends one of the
// referenced outputs.
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/mempool/mempool.go#L591
case match(err, "already spent"):
fallthrough
// This error is returned from btcd when a referenced output cannot be
// found, meaning it etiher has been spent or doesn't exist.
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/blockchain/chain.go#L405
case match(err, "already been spent"):
fallthrough
// This error is returned from btcd when a transaction is spending
// either output that is missing or already spent, and orphans aren't
// allowed.
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/mempool/mempool.go#L1409
case match(err, "orphan transaction"):
fallthrough
// Error returned from bitcoind when output was spent by other
// non-replacable transaction already in the mempool.
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/validation.cpp#L622
case match(err, "txn-mempool-conflict"):
fallthrough
// Returned by bitcoind on the RPC when broadcasting a transaction that
// is spending either output that is missing or already spent.
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/node/transaction.cpp#L49
case match(err, "missing inputs"):
returnErr = &ErrDoubleSpend{
backendError: err,
}
// Returned by bitcoind if the transaction spends outputs that would be
// replaced by it.
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/validation.cpp#L790
case match(err, "bad-txns-spends-conflicting-tx"):
fallthrough
// Returned by bitcoind when a replacement transaction did not have
// enough fee.
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/validation.cpp#L830
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/validation.cpp#L894
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/validation.cpp#L904
case match(err, "insufficient fee"):
fallthrough
// Returned by bitcoind in case the transaction would replace too many
// transaction in the mempool.
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/validation.cpp#L858
case match(err, "too many potential replacements"):
fallthrough
// Returned by bitcoind if the transaction spends an output that is
// unconfimed and not spent by the transaction it replaces.
// https://github.com/bitcoin/bitcoin/blob/9bf5768dd628b3a7c30dd42b5ed477a92c4d3540/src/validation.cpp#L882
case match(err, "replacement-adds-unconfirmed"):
fallthrough
// Returned by btcd when replacement transaction was rejected for
// whatever reason.
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/mempool/mempool.go#L841
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/mempool/mempool.go#L854
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/mempool/mempool.go#L875
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/mempool/mempool.go#L896
// https://github.com/btcsuite/btcd/blob/130ea5bddde33df32b06a1cdb42a6316eb73cff5/mempool/mempool.go#L913
case match(err, "replacement transaction"):
returnErr = &ErrReplacement{
backendError: err,
}
// We received an error not matching any of the above cases.
default:
returnErr = fmt.Errorf("unmatched backend error: %v", err)
}
// If the transaction was rejected for whatever other reason, then
// we'll remove it from the transaction store, as otherwise, we'll
// attempt to continually re-broadcast it, and the UTXO state of the
// wallet won't be accurate.
dbErr := walletdb.Update(w.db, func(dbTx walletdb.ReadWriteTx) error {
txmgrNs := dbTx.ReadWriteBucket(wtxmgrNamespaceKey)
txRec, err := wtxmgr.NewTxRecordFromMsgTx(tx, time.Now())
if err != nil {
return err
}
return w.TxStore.RemoveUnminedTx(txmgrNs, txRec)
})
if dbErr != nil {
log.Warnf("Unable to remove invalid transaction %v: %v",
tx.TxHash(), dbErr)
} else {
log.Infof("Removed invalid transaction: %v",
spew.Sdump(tx))
}
return nil, returnErr
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
}
// ChainParams returns the network parameters for the blockchain the wallet
// belongs to.
func (w *Wallet) ChainParams() *chaincfg.Params {
return w.chainParams
}
// Database returns the underlying walletdb database. This method is provided
// in order to allow applications wrapping btcwallet to store app-specific data
// with the wallet's database.
func (w *Wallet) Database() walletdb.DB {
return w.db
}
// Create creates an new wallet, writing it to an empty database. If the passed
// seed is non-nil, it is used. Otherwise, a secure random seed of the
// recommended length is generated.
func Create(db walletdb.DB, pubPass, privPass, seed []byte, params *chaincfg.Params,
birthday time.Time) error {
// If a seed was provided, ensure that it is of valid length. Otherwise,
// we generate a random seed for the wallet with the recommended seed
// length.
if seed == nil {
hdSeed, err := hdkeychain.GenerateSeed(
hdkeychain.RecommendedSeedLen)
if err != nil {
return err
}
seed = hdSeed
}
if len(seed) < hdkeychain.MinSeedBytes ||
len(seed) > hdkeychain.MaxSeedBytes {
return hdkeychain.ErrInvalidSeedLen
}
return walletdb.Update(db, func(tx walletdb.ReadWriteTx) error {
addrmgrNs, err := tx.CreateTopLevelBucket(waddrmgrNamespaceKey)
if err != nil {
return err
}
txmgrNs, err := tx.CreateTopLevelBucket(wtxmgrNamespaceKey)
if err != nil {
return err
}
err = waddrmgr.Create(
addrmgrNs, seed, pubPass, privPass, params, nil,
birthday,
)
if err != nil {
return err
}
return wtxmgr.Create(txmgrNs)
})
}
// Open loads an already-created wallet from the passed database and namespaces.
func Open(db walletdb.DB, pubPass []byte, cbs *waddrmgr.OpenCallbacks,
params *chaincfg.Params, recoveryWindow uint32) (*Wallet, error) {
var (
addrMgr *waddrmgr.Manager
txMgr *wtxmgr.Store
)
// Before attempting to open the wallet, we'll check if there are any
// database upgrades for us to proceed. We'll also create our references
// to the address and transaction managers, as they are backed by the
// database.
err := walletdb.Update(db, func(tx walletdb.ReadWriteTx) error {
addrMgrBucket := tx.ReadWriteBucket(waddrmgrNamespaceKey)
if addrMgrBucket == nil {
return errors.New("missing address manager namespace")
}
txMgrBucket := tx.ReadWriteBucket(wtxmgrNamespaceKey)
if txMgrBucket == nil {
return errors.New("missing transaction manager namespace")
}
addrMgrUpgrader := waddrmgr.NewMigrationManager(addrMgrBucket)
txMgrUpgrader := wtxmgr.NewMigrationManager(txMgrBucket)
err := migration.Upgrade(txMgrUpgrader, addrMgrUpgrader)
if err != nil {
return err
}
addrMgr, err = waddrmgr.Open(addrMgrBucket, pubPass, params)
if err != nil {
return err
}
txMgr, err = wtxmgr.Open(txMgrBucket, params)
if err != nil {
return err
}
return nil
})
if err != nil {
return nil, err
}
log.Infof("Opened wallet") // TODO: log balance? last sync height?
w := &Wallet{
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
publicPassphrase: pubPass,
db: db,
Manager: addrMgr,
TxStore: txMgr,
lockedOutpoints: map[wire.OutPoint]struct{}{},
recoveryWindow: recoveryWindow,
rescanAddJob: make(chan *RescanJob),
rescanBatch: make(chan *rescanBatch),
rescanNotifications: make(chan interface{}),
rescanProgress: make(chan *RescanProgressMsg),
rescanFinished: make(chan *RescanFinishedMsg),
createTxRequests: make(chan createTxRequest),
unlockRequests: make(chan unlockRequest),
lockRequests: make(chan struct{}),
holdUnlockRequests: make(chan chan heldUnlock),
lockState: make(chan bool),
changePassphrase: make(chan changePassphraseRequest),
changePassphrases: make(chan changePassphrasesRequest),
chainParams: params,
quit: make(chan struct{}),
}
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.NtfnServer = newNotificationServer(w)
w.TxStore.NotifyUnspent = func(hash *chainhash.Hash, index uint32) {
Modernize the RPC server. This is a rather monolithic commit that moves the old RPC server to its own package (rpc/legacyrpc), introduces a new RPC server using gRPC (rpc/rpcserver), and provides the ability to defer wallet loading until request at a later time by an RPC (--noinitialload). The legacy RPC server remains the default for now while the new gRPC server is not enabled by default. Enabling the new server requires setting a listen address (--experimenalrpclisten). This experimental flag is used to effectively feature gate the server until it is ready to use as a default. Both RPC servers can be run at the same time, but require binding to different listen addresses. In theory, with the legacy RPC server now living in its own package it should become much easier to unit test the handlers. This will be useful for any future changes to the package, as compatibility with Core's wallet is still desired. Type safety has also been improved in the legacy RPC server. Multiple handler types are now used for methods that do and do not require the RPC client as a dependency. This can statically help prevent nil pointer dereferences, and was very useful for catching bugs during refactoring. To synchronize the wallet loading process between the main package (the default) and through the gRPC WalletLoader service (with the --noinitialload option), as well as increasing the loose coupling of packages, a new wallet.Loader type has been added. All creating and loading of existing wallets is done through a single Loader instance, and callbacks can be attached to the instance to run after the wallet has been opened. This is how the legacy RPC server is associated with a loaded wallet, even after the wallet is loaded by a gRPC method in a completely unrelated package. Documentation for the new RPC server has been added to the rpc/documentation directory. The documentation includes a specification for the new RPC API, addresses how to make changes to the server implementation, and provides short example clients in several different languages. Some of the new RPC methods are not implementated exactly as described by the specification. These are considered bugs with the implementation, not the spec. Known bugs are commented as such.
2015-06-01 21:57:50 +02:00
w.NtfnServer.notifyUnspentOutput(0, hash, index)
}
return w, nil
}