Files
bitcoin/src/consensus/merkle.cpp
Ava Chow 8c07800b19 Merge bitcoin/bitcoin#32497: merkle: pre‑reserve leaves to prevent reallocs with odd vtx count
3dd815f048 validation: pre-reserve leaves to prevent reallocs with odd vtx count (Lőrinc)
7fd47e0e56 bench: make `MerkleRoot` benchmark more representative (Lőrinc)
f0a2183108 test: adjust `ComputeMerkleRoot` tests (Lőrinc)

Pull request description:

  #### Summary

  `ComputeMerkleRoot` [duplicates the last hash](39b6c139bd/src/consensus/merkle.cpp (L54-L56)) when the input size is odd. If the caller provides a `std::vector` whose capacity equals its size, that extra `push_back` forces a reallocation, doubling its capacity (causing peak memory usage of 3x the necessary size).

  This affects roughly half of the created blocks (those with odd transaction counts), causing unnecessary memory fragmentation during every block validation.

  #### Fix

  * Pre-reserves vector capacity to account for the odd-count duplication using `(size + 1) & ~1ULL`.
      * This syntax produces [optimal assembly](https://github.com/bitcoin/bitcoin/pull/32497#discussion_r2553107836) across x86/ARM and 32/64-bit platforms for GCC & Clang.
  * Eliminates default construction of `uint256` objects that are immediately overwritten by switching from `resize` to `reserve` + `push_back`.

  #### Memory Impact

  [Memory profiling](https://github.com/bitcoin/bitcoin/pull/32497#issuecomment-3563724551) shows **50% reduction in peak allocation** (576KB → 288KB) and elimination of reallocation overhead.

  #### Validation

  The benchmark was updated to use an odd leaf count to demonstrate the real-world scenario where the reallocation occurs.

  A full `-reindex-chainstate` up to block **896 408** ran without triggering the asserts.

  <details>
  <summary>Validation asserts</summary>

  Temporary asserts (not included in this PR) confirm that `push_back` never reallocates and that the coinbase witness hash remains null:
  ```cpp
  if (hashes.size() & 1) {
      assert(hashes.size() < hashes.capacity()); // TODO remove
      hashes.push_back(hashes.back());
  }

  leaves.reserve((block.vtx.size() + 1) & ~1ULL); // capacity rounded up to even
  leaves.emplace_back();
  assert(leaves.back().IsNull()); // TODO remove
  ```

  </details>

  #### Benchmark Performance

  While the main purpose is to improve predictability, the reduced memory operations also improve hashing throughput slightly.

ACKs for top commit:
  achow101:
    ACK 3dd815f048
  optout21:
    reACK 3dd815f048
  hodlinator:
    re-ACK 3dd815f048
  vasild:
    ACK 3dd815f048
  w0xlt:
    ACK 3dd815f048 with minor nits.
  danielabrozzoni:
    Code review ACK 3dd815f048

Tree-SHA512: e7b578f9deadc0de7d61c062c7f65c5e1d347548ead4a4bb74b056396ad7df3f1c564327edc219670e6e2b2cb51f4e1ccfd4f58dd414aeadf2008d427065c11f
2026-01-20 15:47:17 -08:00

181 lines
7.1 KiB
C++

// Copyright (c) 2015-present The Bitcoin Core developers
// Distributed under the MIT software license, see the accompanying
// file COPYING or http://www.opensource.org/licenses/mit-license.php.
#include <consensus/merkle.h>
#include <hash.h>
#include <util/check.h>
/* WARNING! If you're reading this because you're learning about crypto
and/or designing a new system that will use merkle trees, keep in mind
that the following merkle tree algorithm has a serious flaw related to
duplicate txids, resulting in a vulnerability (CVE-2012-2459).
The reason is that if the number of hashes in the list at a given level
is odd, the last one is duplicated before computing the next level (which
is unusual in Merkle trees). This results in certain sequences of
transactions leading to the same merkle root. For example, these two
trees:
A A
/ \ / \
B C B C
/ \ | / \ / \
D E F D E F F
/ \ / \ / \ / \ / \ / \ / \
1 2 3 4 5 6 1 2 3 4 5 6 5 6
for transaction lists [1,2,3,4,5,6] and [1,2,3,4,5,6,5,6] (where 5 and
6 are repeated) result in the same root hash A (because the hash of both
of (F) and (F,F) is C).
The vulnerability results from being able to send a block with such a
transaction list, with the same merkle root, and the same block hash as
the original without duplication, resulting in failed validation. If the
receiving node proceeds to mark that block as permanently invalid
however, it will fail to accept further unmodified (and thus potentially
valid) versions of the same block. We defend against this by detecting
the case where we would hash two identical hashes at the end of the list
together, and treating that identically to the block having an invalid
merkle root. Assuming no double-SHA256 collisions, this will detect all
known ways of changing the transactions without affecting the merkle
root.
*/
uint256 ComputeMerkleRoot(std::vector<uint256> hashes, bool* mutated) {
bool mutation = false;
while (hashes.size() > 1) {
if (mutated) {
for (size_t pos = 0; pos + 1 < hashes.size(); pos += 2) {
if (hashes[pos] == hashes[pos + 1]) mutation = true;
}
}
if (hashes.size() & 1) {
hashes.push_back(hashes.back());
}
SHA256D64(hashes[0].begin(), hashes[0].begin(), hashes.size() / 2);
hashes.resize(hashes.size() / 2);
}
if (mutated) *mutated = mutation;
if (hashes.size() == 0) return uint256();
return hashes[0];
}
uint256 BlockMerkleRoot(const CBlock& block, bool* mutated)
{
std::vector<uint256> leaves;
leaves.reserve((block.vtx.size() + 1) & ~1ULL); // capacity rounded up to even
for (size_t s = 0; s < block.vtx.size(); s++) {
leaves.push_back(block.vtx[s]->GetHash().ToUint256());
}
return ComputeMerkleRoot(std::move(leaves), mutated);
}
uint256 BlockWitnessMerkleRoot(const CBlock& block)
{
std::vector<uint256> leaves;
leaves.reserve((block.vtx.size() + 1) & ~1ULL); // capacity rounded up to even
leaves.emplace_back(); // The witness hash of the coinbase is 0.
for (size_t s = 1; s < block.vtx.size(); s++) {
leaves.push_back(block.vtx[s]->GetWitnessHash().ToUint256());
}
return ComputeMerkleRoot(std::move(leaves));
}
/* This implements a constant-space merkle path calculator, limited to 2^32 leaves. */
static void MerkleComputation(const std::vector<uint256>& leaves, uint32_t leaf_pos, std::vector<uint256>& path)
{
path.clear();
Assume(leaves.size() <= UINT32_MAX);
if (leaves.size() == 0) {
return;
}
// count is the number of leaves processed so far.
uint32_t count = 0;
// inner is an array of eagerly computed subtree hashes, indexed by tree
// level (0 being the leaves).
// For example, when count is 25 (11001 in binary), inner[4] is the hash of
// the first 16 leaves, inner[3] of the next 8 leaves, and inner[0] equal to
// the last leaf. The other inner entries are undefined.
uint256 inner[32];
// Which position in inner is a hash that depends on the matching leaf.
int matchlevel = -1;
// First process all leaves into 'inner' values.
while (count < leaves.size()) {
uint256 h = leaves[count];
bool matchh = count == leaf_pos;
count++;
int level;
// For each of the lower bits in count that are 0, do 1 step. Each
// corresponds to an inner value that existed before processing the
// current leaf, and each needs a hash to combine it.
for (level = 0; !(count & ((uint32_t{1}) << level)); level++) {
if (matchh) {
path.push_back(inner[level]);
} else if (matchlevel == level) {
path.push_back(h);
matchh = true;
}
h = Hash(inner[level], h);
}
// Store the resulting hash at inner position level.
inner[level] = h;
if (matchh) {
matchlevel = level;
}
}
// Do a final 'sweep' over the rightmost branch of the tree to process
// odd levels, and reduce everything to a single top value.
// Level is the level (counted from the bottom) up to which we've sweeped.
int level = 0;
// As long as bit number level in count is zero, skip it. It means there
// is nothing left at this level.
while (!(count & ((uint32_t{1}) << level))) {
level++;
}
uint256 h = inner[level];
bool matchh = matchlevel == level;
while (count != ((uint32_t{1}) << level)) {
// If we reach this point, h is an inner value that is not the top.
// We combine it with itself (Bitcoin's special rule for odd levels in
// the tree) to produce a higher level one.
if (matchh) {
path.push_back(h);
}
h = Hash(h, h);
// Increment count to the value it would have if two entries at this
// level had existed.
count += ((uint32_t{1}) << level);
level++;
// And propagate the result upwards accordingly.
while (!(count & ((uint32_t{1}) << level))) {
if (matchh) {
path.push_back(inner[level]);
} else if (matchlevel == level) {
path.push_back(h);
matchh = true;
}
h = Hash(inner[level], h);
level++;
}
}
}
static std::vector<uint256> ComputeMerklePath(const std::vector<uint256>& leaves, uint32_t position) {
std::vector<uint256> ret;
MerkleComputation(leaves, position, ret);
return ret;
}
std::vector<uint256> TransactionMerklePath(const CBlock& block, uint32_t position)
{
std::vector<uint256> leaves;
leaves.resize(block.vtx.size());
for (size_t s = 0; s < block.vtx.size(); s++) {
leaves[s] = block.vtx[s]->GetHash().ToUint256();
}
return ComputeMerklePath(leaves, position);
}