In order to sleep accurately, the message handler needs to know if _any_ node
has more processing that it should do before the entire thread sleeps.
Rather than returning a value that represents whether ProcessMessages
encountered a message that should trigger a disconnnect, interpret the return
value as whether or not that node has more work to do.
Also, use a global fProcessWake value that can be set by other threads,
which takes precedence (for one cycle) over the messagehandler's decision.
Note that the previous behavior was to only process one message per loop
(except in the case of a bad checksum or invalid header). That was changed in
PR #3180.
The only change here in that regard is that the current node now falls to the
back of the processing queue for the bad checksum/invalid header cases.
Previously addnodes were in competition with outbound connections
for access to the eight outbound slots.
One result of this is that frequently a node with several addnode
configured peers would end up connected to none of them, because
while the addnode loop was in its two minute sleep the automatic
connection logic would fill any free slots with random peers.
This is particularly unwelcome to users trying to maintain links
to specific nodes for fast block relay or purposes.
Another result is that a group of nine or more nodes which are
have addnode configured towards each other can become partitioned
from the public network.
This commit introduces a new limit of eight connections just for
addnode peers which is not subject to any of the other connection
limitations (including maxconnections).
The choice of eight is sufficient so that under no condition would
a user find themselves connected to fewer addnoded peers than
previously. It is also low enough that users who are confused
about the significance of more connections and have gotten too
copy-and-paste happy will not consume more than twice the slot
usage of a typical user.
Any additional load on the network resulting from this will likely
be offset by a reduction in users applying even more wasteful
workaround for the prior behavior.
The retry delays are reduced to avoid nodes sitting around without
their added peers up, but are still sufficient to prevent overly
aggressive repeated connections. The reduced delays also make
the system much more responsive to the addnode RPC.
Ban-disconnects are also exempted for peers added via addnode since
the outbound addnode logic ignores bans. Previously it would ban
an addnode then immediately reconnect to it.
A minor change was also made to CSemaphoreGrant so that it is
possible to re-acquire via an object whos grant was moved.
Surprisingly this hasn't been causing me any issues while testing, probably
because it requires lots of large blocks to be flying around.
Send/Recv corks need tests!
- Drop the interruption point directly after the pnode allocation. This would
be leaky if hit.
- Rearrange thread creation so that the socket handler comes first
This fixes one of the last major layer violations in the networking stack.
The network side is no longer in charge of message serialization, so it is now
decoupled from Bitcoin structures. Only the header is serialized and attached
to the payload.
The changes here are dense and subtle, but hopefully all is more explicit
than before.
- CConnman is now in charge of sending data rather than the nodes themselves.
This is necessary because many decisions need to be made with all nodes in
mind, and a model that requires the nodes calling up to their manager quickly
turns to spaghetti.
- The per-node-serializer (ssSend) has been replaced with a (quasi-)const
send-version. Since the send version for serialization can only change once
per connection, we now explicitly tag messages with INIT_PROTO_VERSION if
they are sent before the handshake. With this done, there's no need to lock
for access to nSendVersion.
Also, a new stream is used for each message, so there's no need to lock
during the serialization process.
- This takes care of accounting for optimistic sends, so the
nOptimisticBytesWritten hack can be removed.
- -dropmessagestest and -fuzzmessagestest have not been preserved, as I suspect
they haven't been used in years.
nMaxInbound might very well be 0 or -1, if the user prefers to keep
a small number of maxconnections.
Note: nMaxInbound of -1 means that the user set maxconnections
to 8 or less, but we still want to keep an additional slot for
the feeler connection.
Add getNetworkActive()/setNetworkActive() method to client model.
Send network active status through NotifyNetworkActiveChanged.
Indicate in tool tip of gui status bar network indicator whether network activity is disabled.
Indicate in debug window whether network activity is disabled and add button to allow user to toggle network activity state.
Added the function SetNetworkActive() which when called with argument set to false disconnects all nodes and sets the flag fNetworkActive to false. As long as this flag is false no new connections are attempted and no incoming connections are accepted. Network activity is reenabled by calling the function with argument true.
4630479 Make dnsseed's definition of acute need include relevant services. (Gregory Maxwell)
9583477 Be more aggressive in connecting to peers with relevant services. (Gregory Maxwell)
We normally prefer to connect to peers offering the relevant services.
If we're not connected to enough peers with relevant services, we
probably don't know about them and could use dnsseed's help.
Only allow skipping relevant services until there are four outbound
connections up.
This avoids quickly filling up with peers lacking the relevant
services when addrman has few or none of them.
There are only a few uses of `insecure_random` outside the tests.
This PR replaces uses of insecure_random (and its accompanying global
state) in the core code with an FastRandomContext that is automatically
seeded on creation.
This is meant to be used for inner loops. The FastRandomContext
can be in the outer scope, or the class itself, then rand32() is used
inside the loop. Useful e.g. for pushing addresses in CNode or the fee
rounding, or randomization for coin selection.
As a context is created per purpose, thus it gets rid of
cross-thread unprotected shared usage of a single set of globals, this
should also get rid of the potential race conditions.
- I'd say TxMempool::check is not called enough to warrant using a special
fast random context, this is switched to GetRand() (open for
discussion...)
- The use of `insecure_rand` in ConnectThroughProxy has been replaced by
an atomic integer counter. The only goal here is to have a different
credentials pair for each connection to go on a different Tor circuit,
it does not need to be random nor unpredictable.
- To avoid having a FastRandomContext on every CNode, the context is
passed into PushAddress as appropriate.
There remains an insecure_random for test usage in `test_random.h`.
In principle, the checksums of P2P packets are simply 4-byte blobs which
are the first four bytes of SHA256(SHA256(payload)).
Currently they are handled as little-endian 32-bit integers half of the
time, as blobs the other half, sometimes copying the one to the other,
resulting in somewhat confused code.
This PR changes the handling to be consistent both at packet creation
and receiving, making it (I think) easier to understand.
After #8594 the addrFrom sent by a node is not used anymore at all,
so don't bother sending it.
Also mitigates the privacy issue in (#8616). It doesn't completely solve
the issue as GetLocalAddress is also called in AdvertiseLocal, but at
least when advertising addresses it stands out less as *our* address.
This was broken by 63cafa6329.
Note that while this fixes the settings, it doesn't fix the actual usage of
-maxuploadtarget completely, as there is currently a bug in the
nOptimisticBytesWritten accounting that causes a delayed response if the target
is reached. That bug will be addressed separately.
In the case of (for example) an already-running bitcoind, the shutdown sequence
begins before CConnman has been created, leading to a null-pointer dereference
when g_connman->Stop() is called.
Instead, Just let the CConnman dtor take care of stopping.
CConnman then passes the current best height into CNode at creation time.
This way CConnman/CNode have no dependency on main for height, and the signals
only move in one direction.
This also helps to prevent identity leakage a tiny bit. Before this change, an
attacker could theoretically make 2 connections on different interfaces. They
would connect fully on one, and only establish the initial connection on the
other. Once they receive a new block, they would relay it to your first
connection, and immediately commence the version handshake on the second. Since
the new block height is reflected immediately, they could attempt to learn
whether the two connections were correlated.
This is, of course, incredibly unlikely to work due to the small timings
involved and receipt from other senders. But it doesn't hurt to lock-in
nBestHeight at the time of connection, rather than letting the remote choose
the time.
This behavior seems to have been quite racy and broken.
Move nLocalHostNonce into CNode, and check received nonces against all
non-fully-connected nodes. If there's a match, assume we've connected
to ourself.
Tests if addresses are online or offline by briefly connecting to them. These short lived connections are referred to as feeler connections. Feeler connections are designed to increase the number of fresh online addresses in tried by selecting and connecting to addresses in new. One feeler connection is attempted on average once every two minutes.
This change was suggested as Countermeasure 4 in
Eclipse Attacks on Bitcoin’s Peer-to-Peer Network, Ethan Heilman,
Alison Kendler, Aviv Zohar, Sharon Goldberg. ePrint Archive Report
2015/263. March 2015.
9e9d644 net: fixup nits (Cory Fields)
8945384 net: Have LookupNumeric return a CService directly (Cory Fields)
21ba407 net: narrow include scope after moving to netaddress (Cory Fields)
21e5b96 net: move CNetAddr/CService/CSubNet out of netbase (Cory Fields)
1017b8a net: Add direct tests for new CSubNet constructors (Cory Fields)
b6c3ff3 net: Split resolving out of CSubNet (Cory Fields)
f96c7c4 net: Split resolving out of CService (Cory Fields)
31d6b1d net: Split resolving out of CNetAddr (Cory Fields)
9d4eb9a Do diskspace check before import thread is started (Pieter Wuille)
aa59f2e Add extra message to avoid a long 'Loading banlist' (Pieter Wuille)
0fd2a33 Use a signal to continue init after genesis activation (Pieter Wuille)
1a5a4e6 Randomize name lookup result in ConnectSocketByName (Pieter Wuille)
f9f5cfc Prevent duplicate connections where one is by name and another by ip (Pieter Wuille)
1111b80 Rework addnode behaviour (Pieter Wuille)
6ee7f05 Allow disconnecting a netgroup with only one member in eviction. (Gregory Maxwell)
5d0ca81 Add recently accepted blocks and txn to AttemptToEvictConnection. (Gregory Maxwell)
* Use CNode::addeName to track whether a connection to a name is already open
* A new connection to a previously-connected by-name addednode is only opened when
the previous one closes (even if the name starts resolving to something else)
* At most one connection is opened per addednode (even if the name resolves to multiple)
* Unify the code between ThreadOpenAddedNodeConnections and getaddednodeinfo
* Information about open connections is always returned, and the dns argument becomes a dummy
* An IP address and inbound/outbound is only reported for the (at most 1) open connection
eebc232 test: Add more test vectors for siphash (Wladimir J. van der Laan)
8884830 Use C++11 thread-safe static initializers (Pieter Wuille)
c31b24f Use 64-bit SipHash of netgroups in eviction (Pieter Wuille)
9bf156b Support SipHash with arbitrary byte writes (Pieter Wuille)
053930f Avoid recalculating vchKeyedNetGroup in eviction logic. (Patrick Strateman)
6182d10 Do not increment nAttempts by more than one for every Good connection. (Gregory Maxwell)
c769c4a Avoid counting failed connect attempts when probably offline. (Gregory Maxwell)
This reduces the rate of not founds by better matching the far
end expectations, it also improves privacy by removing the
ability to use getdata to probe for a node having a txn before
it has been relayed.
If a node is offline failed outbound connection attempts will crank up
the addrman counter and effectively blow away our state.
This change reduces the problem by only counting attempts made while
the node believes it has outbound connections to at least two
netgroups.
Connect and addnode connections are also not counted, as there is no
reason to unequally penalize them for their more frequent
connections -- though there should be no real effect from this
unless their addnode configureation is later removed.
Wasteful repeated connection attempts while only a few connections are
up are avoided via nLastTry.
This is still somewhat incomplete protection because our outbound
peers could be down but not timed out or might all be on 'local'
networks (although the requirement for multiple netgroups helps).
The ability to GETDATA a transaction which has not (yet) been relayed
is a privacy loss vector.
The use of the mempool for this was added as part of the mempool p2p
message and is only needed to fetch transactions returned by it.
5d5e7a0 net: No need to export ConnectNode (Cory Fields)
e9ed620 net: No need to export DumpBanlist (Cory Fields)
8b8f877 net: make Ban/Unban/ClearBan functionality consistent (Cory Fields)
cca221f net: Drop CNodeRef for AttemptToEvictConnection (Cory Fields)
563f375 net: use the exposed GetNodeSignals() rather than g_signals directly (Cory Fields)
9faa490 net: remove unused set (Cory Fields)
52cbce2 net: don't import std namespace (Cory Fields)
1475ecf Fix de-serialization bug where AddrMan is corrupted after exception * CAddrDB modified so that when de-serialization code throws an exception Addrman is reset to a clean state * CAddrDB modified to make unit tests possible * Regression test created to ensure bug is fixed * StartNode modifed to clear adrman if CAddrDB::Read returns an error code. (EthanHeilman)
- Ban/Unban/ClearBan call uiInterface.BannedListChanged() as necessary
- Ban/Unban/ClearBan sync to disk if the operation is user-invoked
- Mark node for disconnection automatically when banning
- Lock cs_vNodes while setting disconnected
- Don't spin in a tight loop while setting disconnected
Locking for each operation here is unnecessary, and solves the wrong problem.
Additionally, it introduces a problem when cs_vNodes is held in an owning
class, to which invididual CNodeRefs won't have access.
These should be weak pointers anyway, once vNodes contain shared pointers.
Rather than using a refcounting class, use a 3-step process instead.
1. Lock vNodes long enough to snapshot the fields necessary for comparing
2. Unlock and do the comparison
3. Re-lock and mark the resulting node for disconnection if it still exists
b559914 Move bloom and feerate filtering to just prior to tx sending. (Gregory Maxwell)
4578215 Return mempool queries in dependency order (Pieter Wuille)
ed70683 Handle mempool requests in send loop, subject to trickle (Pieter Wuille)
dc13dcd Split up and optimize transaction and block inv queues (Pieter Wuille)
f2d3ba7 Eliminate TX trickle bypass, sort TX invs for privacy and priority. (Gregory Maxwell)
* CAddrDB modified so that when de-serialization code throws an exception Addrman is reset to a clean state
* CAddrDB modified to make unit tests possible
* Regression test created to ensure bug is fixed
* StartNode modifed to clear adrman if CAddrDB::Read returns an error code.
Some developers clearly don't get this and have been posting
"improvements" that create clear vulnerabilities. It should
have been better explained in the code, since the design
is somewhat subtle and getting it right is important.
DumpBanList currently does this:
- with lock: take a copy of the banmap
- perform I/O (write out the banmap)
- with lock: mark the banmap non-dirty
If a new ban is added during the I/O operation, it may never be persisted to
disk.
Reorder operations so that the data to be persisted cannot be older than the
time at which the banmap was marked non-dirty.
This will avoid sending more pointless INVs around updates, and
prevents using filter updates to timetag transactions.
Also adds locking for fRelayTxes.
By eliminating queued entries from the mempool response and responding only at
trickle time, this makes the mempool no longer leak transaction arrival order
information (as the mempool itself is also sorted)-- at least no more than
relay itself leaks it.
Note: Some seeds aren't actually returning an IP for their name entries, so
they're being added to addrman with a source of [::].
This commit shouldn't change that behavior, for better or worse.
Previously we used the CInv that would be sent to the peer announcing the
transaction as the key, but using the txid instead allows us to decouple the
p2p layer from the application logic (which relies on this map to avoid
duplicate tx requests).
The "feefilter" p2p message is used to inform other nodes of your mempool min fee which is the feerate that any new transaction must meet to be accepted to your mempool. This will allow them to filter invs to you according to this feerate.
027fdb8 When/if the copyright line does not mention Bitcoin Core developers, add a second line to copyrights in -version, About dialog, and splash screen (Luke Dashjr)
cc2095e Rewrite FormatParagraph to handle newlines within input strings correctly (Luke Dashjr)
cddffaf Bugfix: Include COPYRIGHT_HOLDERS_SUBSTITUTION in Makefile substitutions so it gets passed to extract-strings correctly (Luke Dashjr)
29598e4 Move PACKAGE_URL to configure.ac (Luke Dashjr)
78ec83d splashscreen: Resize text to fit exactly (Luke Dashjr)
3cae140 Bugfix: Actually use _COPYRIGHT_HOLDERS_SUBSTITUTION everywhere (Luke Dashjr)
4d5a3df Bugfix: gitian-descriptors: Add missing python-setuptools requirement for OS X (biplist module) (Luke Dashjr)
e4ab5e5 Bugfix: Correct copyright year in Mac DMG background image (Luke Dashjr)
917b1d0 Set copyright holders displayed in notices separately from the package name (Luke Dashjr)
c39a6ff Travis & gitian-osx: Use depends for ds_store and mac_alias modules (Luke Dashjr)
902ccde depends: Add mac_alias to depends (Luke Dashjr)
82a2d98 depends: Add ds_store to depends (Cory Fields)
de619a3 depends: Pass PYTHONPATH along to configure (Cory Fields)
e611b6e macdeploy: Use rsvg-convert rather than cairosvg (Luke Dashjr)
63bcdc5 More complicated package name substitution for Mac deployment (Luke Dashjr)
1a6c67c Parameterise 2009 in translatable copyright strings (Luke Dashjr)
d5f4683 Unify package name to as few places as possible without major changes (Luke Dashjr)
This corrects a bug the case of tying group size where the code may
fail to select the group with the newest member. Since newest time
is the final selection criteria, failing to break ties on it
on the step before can undermine the final selection.
Tied netgroups are very common.
(cherry picked from commit 8e09f914f8ec66301257358b250e9a61befadd95)
With automatic tor HS support in place we should probably not be providing
absolute protection for local peers, since HS inbound could be used to
attack pretty easily. Instead, this counts on the latency metric inside
AttemptToEvictConnection to privilege actually local peers.
(cherry picked from commit 46dbcd4833115401fecbb052365b4c7725874414)
e8600c9 banlist (bugfix): allow CNode::SweepBanned() to run on interval (Philip Kaufmann)
2977c24 banlist: add more banlist infos to log / add GUI signal (Philip Kaufmann)
ce479aa banlist: better handling of banlist in StartNode() (Philip Kaufmann)
57c77fe banlist: update set dirty to be more fine grained (Philip Kaufmann)
We used to have a trickle node, a node which was chosen in each iteration of
the send loop that was privileged and allowed to send out queued up non-time
critical messages. Since the removal of the fixed sleeps in the network code,
this resulted in fast and attackable treatment of such broadcasts.
This pull request changes the 3 remaining trickle use cases by random delays:
* Local address broadcast (while also removing the the wiping of the seen filter)
* Address relay
* Inv relay (for transactions; blocks are always relayed immediately)
The code is based on older commits by Patrick Strateman.
- Avoids string typos (by making the compiler check)
- Makes it easier to grep for handling/generation of a certain message type
- Refer directly to documentation by following the symbol in IDE
- Move list of valid message types to protocol.cpp:
protocol.cpp is a more appropriate place for this, and having
the array there makes it easier to keep things consistent.
aa4b0c2 When not filtering blocks, getdata sends more in one test (Pieter Wuille)
d41e44c Actually only use filterInventoryKnown with MSG_TX inventory messages. (Gregory Maxwell)
b6a0da4 Only use filterInventoryKnown with MSG_TX inventory messages. (Patick Strateman)
6b84935 Rename setInventoryKnown filterInventoryKnown (Patick Strateman)
e206724 Remove mruset as it is no longer used. (Gregory Maxwell)
ec73ef3 Replace setInventoryKnown with a rolling bloom filter. (Gregory Maxwell)
ebb25f4 Limit setAskFor and retire requested entries only when a getdata returns. (Gregory Maxwell)
5029698 prevent peer flooding request queue for an inv (kazcw)
Mruset setInventoryKnown was reduced to a remarkably small 1000
entries as a side effect of sendbuffer size reductions in 2012.
This removes setInventoryKnown filtering from merkleBlock responses
because false positives there are especially unattractive and
also because I'm not sure if there aren't race conditions around
the relay pool that would cause some transactions there to
be suppressed. (Also, ProcessGetData was accessing
setInventoryKnown without taking the required lock.)
The setAskFor duplicate elimination was too eager and removed entries
when we still had no getdata response, allowing the peer to keep
INVing and not responding.
mapAlreadyAskedFor does not keep track of which peer has a request queued for a
particular tx. As a result, a peer can blind a node to a tx indefinitely by
sending many invs for the same tx, and then never replying to getdatas for it.
Each inv received will be placed 2 minutes farther back in mapAlreadyAskedFor,
so a short message containing 10 invs would render that tx unavailable for 20
minutes.
This is fixed by disallowing a peer from having more than one entry for a
particular inv in mapAlreadyAskedFor at a time.
- Force AUTHCOOKIE size to be 32 bytes: This provides protection against
an attack where a process pretends to be Tor and uses the cookie
authentication method to nab arbitrary files such as the
wallet
- torcontrol logging
- fix cookie auth
- add HASHEDPASSWORD auth, fix fd leak when fwrite() fails
- better error reporting when cookie file is not ok
- better init/shutdown flow
- stop advertizing service when disconnected from tor control port
- COOKIE->SAFECOOKIE auth
Starting with Tor version 0.2.7.1 it is possible, through Tor's control socket
API, to create and destroy 'ephemeral' hidden services programmatically.
https://stem.torproject.org/api/control.html#stem.control.Controller.create_ephemeral_hidden_service
This means that if Tor is running (and proper authorization is available),
bitcoin automatically creates a hidden service to listen on, without user
manual configuration. This will positively affect the number of available
.onion nodes.
- When the node is started, connect to Tor through control socket
- Send `ADD_ONION` command
- First time:
- Make it create a hidden service key
- Save the key in the data directory for later usage
- Make it redirect port 8333 to the local port 8333 (or whatever port we're listening on).
- Keep control socket connection open for as long node is running. The hidden service will
(by default) automatically go away when the connection is closed.
* -maxuploadtarget can be set in MiB
* if <limit> - ( time-left-in-24h-cycle / 600 * MAX_BLOCK_SIZE ) has reach, stop serve blocks older than one week and filtered blocks
* no action if limit has reached, no guarantee that the target will not be surpassed
* add outbound limit informations to rpc getnettotals
Nagle appears to be a significant contributor to latency now that the static
sleeps are gone. Most of our messages are relatively large compared to
IP + TCP so I do not expect this to create enormous overhead.
This may also reduce traffic burstyness somewhat.
- to match the peers.dat handling also supply a debug.log entry for how
many entries were loaded from banlist.dat and how long it took
- add a GUI init message for loading the banlist (same as with peers.dat)
- move the same message for peers.dat upwards in the code, to be able to
reuse the timing variable nStart and also just log, if our read from
peers.dat didn't fail
- only start working on/with banlist data, if reading in the banlist from
disk didn't fail
- as CNode::setBannedIsDirty is false (default) when reading fails, we
don't need to explicitly set it to false to prevent writing
banlist.dat in that case either
`nMinPingUsecTime` was left uninitialized in CNode.
The correct initialization for a minimum-until-now is int64_t's max value, so initialize it to that.
Thanks @MarcoFalke for noticing.
027de94 Use network group instead of CNetAddr in final pass to select node to disconnect (Patrick Strateman)
000c18a Fix comment (Patrick Strateman)
fed3094 Acquire cs_vNodes before changing refrence counts (Patrick Strateman)
69ee1aa CNodeRef copy constructor and assignment operator (Patrick Strateman)
dc81dd0 Return false early if vEvictionCandidates is empty (Patrick Strateman)
17f3533 Better support for nodes with non-standard nMaxConnections (Patrick Strateman)
1317cd1 RAII wrapper for CNode* (Patrick Strateman)
df23937 Add comments to AttemptToEvictConnection (Patrick Strateman)
a8f6e45 Remove redundant whiteconnections option (Patrick Strateman)
b105ba3 Prefer to disconnect peers in favor of whitelisted peers (Patrick Strateman)
2c70153 AttemptToEvictConnection (Patrick Strateman)
4bac601 Record nMinPingUsecTime (Patrick Strateman)
ae037b7 Refactor: Move failure conditions to the top of AcceptConnection (Patrick Strateman)
1ef4817 Refactor: Bail early in AcceptConnection (Patrick Strateman)
541a1dd Refactor: AcceptConnection (Patrick Strateman)
When running the rpc tests in Wine, nodes often fail to listen on localhost
due to a stale socket from a previous run. This aligns the behavior with other
platforms.
7b79cbd limit total length of user agent comments (Pavol Rusnak)
557f8ea implement uacomment config parameter which can add comments to user agent as per BIP-0014 (Pavol Rusnak)
While CBloomFilter is usually used with an explicitly set nTweak,
CRollingBloomFilter is only used internally. Requiring every caller to
set nTweak is error-prone and redundant; better to have the class handle
that for you with a high-quality randomness source.
Additionally when clearing the filter it makes sense to change nTweak as
well to recover from a bad setting, e.g. due to insufficient randomness
at initialization, so the clear() method is replaced by a reset() method
that sets a new, random, nTweak value.
a794284 locking: add a quick example of GUARDED_BY (Cory Fields)
2b890dd locking: fix a few small issues uncovered by -Wthread-safety (Cory Fields)
cd27bba locking: teach Clang's -Wthread-safety to cope with our scoped lock macros (Cory Fields)
- rpcwallet: No need to lock twice here
- openssl: Clang doesn't understand selective lock/unlock here. Ignore it.
- CNode: Fix a legitimate (though very unlikely) locking bug.
This sets aside a number of connection slots for whitelisted peers,
useful for ensuring your local users and miners can always get in,
even if your limit on inbound connections has already been reached.
Use a probabilistic bloom filter to keep track of which addresses
we think we have given our peers, instead of a list.
This uses much less memory, at the cost of sometimes failing to
relay an address to a peer-- worst case if the bloom filter happens
to be as full as it gets, 1-in-1,000.
Measured memory usage of a full mruset setAddrKnown: 650Kbytes
Constant memory usage of CRollingBloomFilter addrKnown: 37Kbytes.
This will also help heap fragmentation, because the 37K of storage
is allocated when a CNode is created (when a connection to a peer
is established) and then there is no per-item-remembered memory
allocation.
I plan on testing by restarting a full node with an empty peers.dat,
running a while with -debug=addrman and -debug=net, and making sure
that the 'addr' message traffic out is reasonable.
(suggestions for better tests welcome)
