fuzz: Add fuzzer for block index

This fuzz target creates arbitrary tree-like structure of indices,
simulating the following events:
- Adding a header to the block tree db
- Receiving the full block (may be valid or not)
- Reorging to a new chain tip (possibly encountering invalid blocks on
  the way)
- pruning
The test skips all actual validation of header/ block / transaction data
by just simulating the outcome, and also doesn't interact with the data directory.

The main goal is to test the integrity of the block index tree in
all fuzzed constellations, by calling CheckBlockIndex()
at the end of each iteration.
This commit is contained in:
Martin Zumsande 2024-10-30 14:59:37 -04:00
parent 477b357460
commit 0e0322767c
5 changed files with 214 additions and 3 deletions

View file

@ -1074,6 +1074,13 @@ bool BlockManager::ReadBlockFromDisk(CBlock& block, const CBlockIndex& index) co
return true;
}
void BlockManager::CleanupForFuzzing()
{
m_dirty_blockindex.clear();
m_dirty_fileinfo.clear();
m_blockfile_info.resize(1);
}
bool BlockManager::ReadRawBlockFromDisk(std::vector<uint8_t>& block, const FlatFilePos& pos) const
{
FlatFilePos hpos = pos;

View file

@ -428,6 +428,8 @@ public:
bool UndoReadFromDisk(CBlockUndo& blockundo, const CBlockIndex& index) const;
void CleanupBlockRevFiles() const;
/** Clear internal state (test-only, only for fuzzing) **/
void CleanupForFuzzing();
};
// Calls ActivateBestChain() even if no blocks are imported.

View file

@ -19,6 +19,7 @@ add_executable(fuzz
block.cpp
block_header.cpp
block_index.cpp
block_index_tree.cpp
blockfilter.cpp
bloom_filter.cpp
buffered_file.cpp

View file

@ -0,0 +1,201 @@
// Copyright (c) 2020-2022 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 <chain.h>
#include <chainparams.h>
#include <cstdint>
#include <flatfile.h>
#include <test/fuzz/FuzzedDataProvider.h>
#include <test/fuzz/fuzz.h>
#include <test/fuzz/util.h>
#include <test/util/setup_common.h>
#include <optional>
#include <ranges>
#include <validation.h>
#include <vector>
const TestingSetup* g_setup;
CBlockHeader ConsumeBlockHeader(FuzzedDataProvider& provider, uint256 prev_hash, int& nonce_counter)
{
CBlockHeader header;
header.nVersion = provider.ConsumeIntegral<decltype(header.nVersion)>();
header.hashPrevBlock = prev_hash;
header.hashMerkleRoot = uint256{}; // never used
header.nTime = provider.ConsumeIntegral<decltype(header.nTime)>();
header.nBits = Params().GenesisBlock().nBits;
header.nNonce = nonce_counter++; // prevent creating multiple block headers with the same hash
return header;
}
void initialize_block_index_tree()
{
static const auto testing_setup = MakeNoLogFileContext<const TestingSetup>();
g_setup = testing_setup.get();
}
FUZZ_TARGET(block_index_tree, .init = initialize_block_index_tree)
{
FuzzedDataProvider fuzzed_data_provider(buffer.data(), buffer.size());
ChainstateManager& chainman = *g_setup->m_node.chainman;
auto& blockman = chainman.m_blockman;
CBlockIndex* genesis = chainman.ActiveChainstate().m_chain[0];
int nonce_counter = 0;
std::vector<CBlockIndex*> blocks;
blocks.push_back(genesis);
LIMITED_WHILE(fuzzed_data_provider.ConsumeBool(), 1000)
{
CallOneOf(
fuzzed_data_provider,
[&] {
// Receive a header building on an existing one. This assumes headers are valid, so PoW is not relevant here.
LOCK(cs_main);
CBlockIndex* prev_block = PickValue(fuzzed_data_provider, blocks);
if (!(prev_block->nStatus & BLOCK_FAILED_MASK)) {
CBlockHeader header = ConsumeBlockHeader(fuzzed_data_provider, prev_block->GetBlockHash(), nonce_counter);
CBlockIndex* index = blockman.AddToBlockIndex(header, chainman.m_best_header);
assert(index->nStatus & BLOCK_VALID_TREE);
}
},
[&] {
// Receive a full block (valid or invalid) for an existing header, but don't attempt to connect it yet
LOCK(cs_main);
CBlockIndex* index = PickValue(fuzzed_data_provider, blocks);
// Must be new to us and not known to be invalid (e.g. because of an invalid ancestor).
if (index->nTx == 0 && !(index->nStatus & BLOCK_FAILED_MASK)) {
if (fuzzed_data_provider.ConsumeBool()) { // Invalid
BlockValidationState state;
state.Invalid(BlockValidationResult::BLOCK_CONSENSUS, "consensus-invalid");
chainman.ActiveChainstate().InvalidBlockFound(index, state);
} else {
size_t nTx = fuzzed_data_provider.ConsumeIntegralInRange<size_t>(1, 1000);
CBlock block; // Dummy block, so that ReceivedBlockTransaction can infer a nTx value.
block.vtx = std::vector<CTransactionRef>(nTx);
FlatFilePos pos(0, fuzzed_data_provider.ConsumeIntegralInRange<int>(1, 1000));
chainman.ReceivedBlockTransactions(block, index, pos);
assert(index->nStatus & BLOCK_VALID_TRANSACTIONS);
assert(index->nStatus & BLOCK_HAVE_DATA);
}
}
},
[&] {
// Simplified ActivateBestChain(): Try to move to a chain with more work - with the possibility of finding blocks to be invalid on the way
LOCK(cs_main);
auto& chain = chainman.ActiveChain();
CBlockIndex* old_tip = chain.Tip();
assert(old_tip);
do {
CBlockIndex* best_tip = chainman.ActiveChainstate().FindMostWorkChain();
assert(best_tip); // Should at least return current tip
if (best_tip == chain.Tip()) break; // Nothing to do
// Rewind chain to forking point
const CBlockIndex* fork = chain.FindFork(best_tip);
// If we can't go back to the fork point due to pruned data, abort and don't do anything. Note that this check does not exist in validation.cpp, where
// the node would currently just crash in this scenario (although this is very unlikely to happen due to the minimum pruning threshold of 550MiB).
CBlockIndex* it = chain.Tip();
bool pruned_block{false};
while (it && it->nHeight != fork->nHeight) {
if (!(it->nStatus & BLOCK_HAVE_UNDO) && it->nHeight > 0) {
assert(blockman.m_have_pruned);
pruned_block = true;
break;
}
it = it->pprev;
}
if (pruned_block) break;
chain.SetTip(*chain[fork->nHeight]);
// Prepare new blocks to connect
std::vector<CBlockIndex*> to_connect;
it = best_tip;
while (it && it->nHeight != fork->nHeight) {
to_connect.push_back(it);
it = it->pprev;
}
// Connect blocks, possibly fail
for (CBlockIndex* block : to_connect | std::views::reverse) {
assert(!(block->nStatus & BLOCK_FAILED_MASK));
assert(block->nStatus & BLOCK_HAVE_DATA);
if (!block->IsValid(BLOCK_VALID_SCRIPTS)) {
if (fuzzed_data_provider.ConsumeBool()) { // Invalid
BlockValidationState state;
state.Invalid(BlockValidationResult::BLOCK_CONSENSUS, "consensus-invalid");
chainman.ActiveChainstate().InvalidBlockFound(block, state);
break;
} else {
block->RaiseValidity(BLOCK_VALID_SCRIPTS);
block->nStatus |= BLOCK_HAVE_UNDO;
}
}
chain.SetTip(*block);
chainman.ActiveChainstate().PruneBlockIndexCandidates();
// ABC may release cs_main / not connect all blocks in one go - but only if we have at least much chain work as we had at the start.
if (block->nChainWork > old_tip->nChainWork && fuzzed_data_provider.ConsumeBool()) {
break;
}
}
} while (node::CBlockIndexWorkComparator()(chain.Tip(), old_tip));
assert(chain.Tip()->nChainWork >= old_tip->nChainWork);
},
[&] {
// Prune chain - dealing with block files is beyond the scope of this test, so just prune random blocks, making no assumptions what must
// be together in a block file.
// Also don't prune blocks outside of the chain for now - this would make the fuzzer crash because of the problem describted in
// https://github.com/bitcoin/bitcoin/issues/31512
LOCK(cs_main);
auto& chain = chainman.ActiveChain();
int prune_height = fuzzed_data_provider.ConsumeIntegralInRange<int>(0, chain.Height());
CBlockIndex* prune_block{chain[prune_height]};
if (prune_block != chain.Tip()) {
blockman.m_have_pruned = true;
prune_block->nStatus &= ~BLOCK_HAVE_DATA;
prune_block->nStatus &= ~BLOCK_HAVE_UNDO;
prune_block->nFile = 0;
prune_block->nDataPos = 0;
prune_block->nUndoPos = 0;
auto range = blockman.m_blocks_unlinked.equal_range(prune_block->pprev);
while (range.first != range.second) {
std::multimap<CBlockIndex*, CBlockIndex*>::iterator _it = range.first;
range.first++;
if (_it->second == prune_block) {
blockman.m_blocks_unlinked.erase(_it);
}
}
}
});
}
chainman.CheckBlockIndex();
// clean up global state changed by last iteration and prepare for next iteration
{
LOCK(cs_main);
genesis->nStatus |= BLOCK_HAVE_DATA;
genesis->nStatus |= BLOCK_HAVE_UNDO;
chainman.m_best_header = genesis;
chainman.m_best_invalid = nullptr;
chainman.nBlockSequenceId = 1;
chainman.ActiveChain().SetTip(*genesis);
chainman.ActiveChainstate().setBlockIndexCandidates.clear();
chainman.m_failed_blocks.clear();
blockman.m_blocks_unlinked.clear();
blockman.m_have_pruned = false;
blockman.CleanupForFuzzing();
// Delete all blocks but Genesis from block index
uint256 genesis_hash = genesis->GetBlockHash();
for (auto it = blockman.m_block_index.begin(); it != blockman.m_block_index.end();) {
if (it->first != genesis_hash) {
it = blockman.m_block_index.erase(it);
} else {
++it;
}
}
chainman.ActiveChainstate().TryAddBlockIndexCandidate(genesis);
assert(blockman.m_block_index.size() == 1);
assert(chainman.ActiveChainstate().setBlockIndexCandidates.size() == 1);
assert(chainman.ActiveChain().Height() == 0);
}
}

View file

@ -771,13 +771,13 @@ public:
{
return m_mempool ? &m_mempool->cs : nullptr;
}
void InvalidBlockFound(CBlockIndex* pindex, const BlockValidationState& state) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
CBlockIndex* FindMostWorkChain() EXCLUSIVE_LOCKS_REQUIRED(cs_main);
private:
bool ActivateBestChainStep(BlockValidationState& state, CBlockIndex* pindexMostWork, const std::shared_ptr<const CBlock>& pblock, bool& fInvalidFound, ConnectTrace& connectTrace) EXCLUSIVE_LOCKS_REQUIRED(cs_main, m_mempool->cs);
bool ConnectTip(BlockValidationState& state, CBlockIndex* pindexNew, const std::shared_ptr<const CBlock>& pblock, ConnectTrace& connectTrace, DisconnectedBlockTransactions& disconnectpool) EXCLUSIVE_LOCKS_REQUIRED(cs_main, m_mempool->cs);
void InvalidBlockFound(CBlockIndex* pindex, const BlockValidationState& state) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
CBlockIndex* FindMostWorkChain() EXCLUSIVE_LOCKS_REQUIRED(cs_main);
bool RollforwardBlock(const CBlockIndex* pindex, CCoinsViewCache& inputs) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
@ -902,7 +902,6 @@ private:
//! most-work chain.
Chainstate* m_active_chainstate GUARDED_BY(::cs_main) {nullptr};
CBlockIndex* m_best_invalid GUARDED_BY(::cs_main){nullptr};
/** The last header for which a headerTip notification was issued. */
CBlockIndex* m_last_notified_header GUARDED_BY(GetMutex()){nullptr};
@ -1064,6 +1063,7 @@ public:
/** Best header we've seen so far (used for getheaders queries' starting points). */
CBlockIndex* m_best_header GUARDED_BY(::cs_main){nullptr};
CBlockIndex* m_best_invalid GUARDED_BY(::cs_main){nullptr};
//! The total number of bytes available for us to use across all in-memory
//! coins caches. This will be split somehow across chainstates.