Merge #16158: Fix logic of memory_cleanse() on MSVC and clean up docs

f53a70ce95 Improve documentation of memory_cleanse() (Tim Ruffing)
cac30a436c Clean up logic in memory_cleanse() for MSVC (Tim Ruffing)

Pull request description:

  When working on https://github.com/bitcoin-core/secp256k1/issues/185, I noticed that the logic in memory_cleanse(), which is supposed to clear memory securely, is weird on MSVC. While it's correct, it's at least a code smell because the code clears the memory twice on MSVC. This weirdness was introduced by #11558.

  This PR fixes the logic on MSVC and also improves the docs around this function. Best reviewed in individual commits, see the commit messages for more rationale. The second commit touches only comments.

ACKs for top commit:
  practicalswift:
    utACK f53a70ce95 :-)
  laanwj:
    code review ACK f53a70ce95

Tree-SHA512: 1c2fd98ae62b34b3e6e59d1178b293af969a9e06cbb7df02a699ce8802f145a336f72edb178c520e3ecec81f7e8083828f90a5ba6367d966a2c7d7c0dd6c0475

src/support/cleanse.cpp

@@ -11,33 +11,25 @@
 #include <Windows.h> // For SecureZeroMemory.
 #endif
 
-/* Compilers have a bad habit of removing "superfluous" memset calls that
- * are trying to zero memory. For example, when memset()ing a buffer and
- * then free()ing it, the compiler might decide that the memset is
- * unobservable and thus can be removed.
- *
- * Previously we used OpenSSL which tried to stop this by a) implementing
- * memset in assembly on x86 and b) putting the function in its own file
- * for other platforms.
- *
- * This change removes those tricks in favour of using asm directives to
- * scare the compiler away. As best as our compiler folks can tell, this is
- * sufficient and will continue to be so.
- *
- * Adam Langley <agl@google.com>
- * Commit: ad1907fe73334d6c696c8539646c21b11178f20f
- * BoringSSL (LICENSE: ISC)
- */
 void memory_cleanse(void *ptr, size_t len)
 {
-    std::memset(ptr, 0, len);
-
-    /* As best as we can tell, this is sufficient to break any optimisations that
-       might try to eliminate "superfluous" memsets. If there's an easy way to
-       detect memset_s, it would be better to use that. */
 #if defined(_MSC_VER)
+    /* SecureZeroMemory is guaranteed not to be optimized out by MSVC. */
     SecureZeroMemory(ptr, len);
 #else
+    std::memset(ptr, 0, len);
+
+    /* Memory barrier that scares the compiler away from optimizing out the memset.
+     *
+     * Quoting Adam Langley <agl@google.com> in commit ad1907fe73334d6c696c8539646c21b11178f20f
+     * in BoringSSL (ISC License):
+     *    As best as we can tell, this is sufficient to break any optimisations that
+     *    might try to eliminate "superfluous" memsets.
+     * This method is used in memzero_explicit() the Linux kernel, too. Its advantage is that it
+     * is pretty efficient because the compiler can still implement the memset() efficiently,
+     * just not remove it entirely. See "Dead Store Elimination (Still) Considered Harmful" by
+     * Yang et al. (USENIX Security 2017) for more background.
+     */
     __asm__ __volatile__("" : : "r"(ptr) : "memory");
 #endif
 }

src/support/cleanse.h

@@ -8,7 +8,8 @@
 
 #include <stdlib.h>
 
-// Attempt to overwrite data in the specified memory span.
+/** Secure overwrite a buffer (possibly containing secret data) with zero-bytes. The write
+ * operation will not be optimized out by the compiler. */
 void memory_cleanse(void *ptr, size_t len);
 
 #endif // BITCOIN_SUPPORT_CLEANSE_H