blockfilterindex.cpp

Go to the documentation of this file.

// Copyright (c) 2018 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 <clientversion.h>
#include <common/args.h>
#include <dbwrapper.h>
#include <index/blockfilterindex.h>
#include <logging.h>
#include <node/blockstorage.h>
#include <primitives/blockhash.h>
#include <util/fs_helpers.h>
#include <validation.h>
 
#include <map>
 
constexpr uint8_t DB_BLOCK_HASH{'s'};
constexpr uint8_t DB_BLOCK_HEIGHT{'t'};
constexpr uint8_t DB_FILTER_POS{'P'};
 
// 16 MiB
constexpr unsigned int MAX_FLTR_FILE_SIZE = 0x1000000;
// 1 MiB
constexpr unsigned int FLTR_FILE_CHUNK_SIZE = 0x100000;
 
constexpr size_t CF_HEADERS_CACHE_MAX_SZ{2000};
 
namespace {
 
struct DBVal {
    uint256 hash;
    uint256 header;
    FlatFilePos pos;
 
    SERIALIZE_METHODS(DBVal, obj) { READWRITE(obj.hash, obj.header, obj.pos); }
};
 
struct DBHeightKey {
    int height{0};
 
    DBHeightKey() = default;
    explicit DBHeightKey(int height_in) : height(height_in) {}
 
    template <typename Stream> void Serialize(Stream &s) const {
        ser_writedata8(s, DB_BLOCK_HEIGHT);
        ser_writedata32be(s, height);
    }
 
    template <typename Stream> void Unserialize(Stream &s) {
        const uint8_t prefix{ser_readdata8(s)};
        if (prefix != DB_BLOCK_HEIGHT) {
            throw std::ios_base::failure(
                "Invalid format for block filter index DB height key");
        }
        height = ser_readdata32be(s);
    }
};
 
struct DBHashKey {
    BlockHash hash;
 
    explicit DBHashKey(const BlockHash &hash_in) : hash(hash_in) {}
 
    SERIALIZE_METHODS(DBHashKey, obj) {
        uint8_t prefix{DB_BLOCK_HASH};
        READWRITE(prefix);
        if (prefix != DB_BLOCK_HASH) {
            throw std::ios_base::failure(
                "Invalid format for block filter index DB hash key");
        }
 
        READWRITE(obj.hash);
    }
};
 
}; // namespace
 
static std::map<BlockFilterType, BlockFilterIndex> g_filter_indexes;
 
BlockFilterIndex::BlockFilterIndex(std::unique_ptr<interfaces::Chain> chain,
                                   BlockFilterType filter_type,
                                   size_t n_cache_size, bool f_memory,
                                   bool f_wipe)
    : BaseIndex(std::move(chain),
                BlockFilterTypeName(filter_type) + " block filter index"),
      m_filter_type(filter_type) {
    const std::string &filter_name = BlockFilterTypeName(filter_type);
    if (filter_name.empty()) {
        throw std::invalid_argument("unknown filter_type");
    }
 
    fs::path path =
        gArgs.GetDataDirNet() / "indexes" / "blockfilter" / filter_name;
    fs::create_directories(path);
 
    m_db = std::make_unique<BaseIndex::DB>(path / "db", n_cache_size, f_memory,
                                           f_wipe);
    m_filter_fileseq = std::make_unique<FlatFileSeq>(std::move(path), "fltr",
                                                     FLTR_FILE_CHUNK_SIZE);
}
 
bool BlockFilterIndex::CustomInit(
    const std::optional<interfaces::BlockKey> &block) {
    if (!m_db->Read(DB_FILTER_POS, m_next_filter_pos)) {
        // Check that the cause of the read failure is that the key does not
        // exist. Any other errors indicate database corruption or a disk
        // failure, and starting the index would cause further corruption.
        if (m_db->Exists(DB_FILTER_POS)) {
            LogError(
                "%s: Cannot read current %s state; index may be corrupted\n",
                __func__, GetName());
            return false;
        }
 
        // If the DB_FILTER_POS is not set, then initialize to the first
        // location.
        m_next_filter_pos.nFile = 0;
        m_next_filter_pos.nPos = 0;
    }
    return true;
}
 
bool BlockFilterIndex::CustomCommit(CDBBatch &batch) {
    const FlatFilePos &pos = m_next_filter_pos;
 
    // Flush current filter file to disk.
    AutoFile file{m_filter_fileseq->Open(pos)};
    if (file.IsNull()) {
        LogError("%s: Failed to open filter file %d\n", __func__, pos.nFile);
        return false;
    }
    if (!FileCommit(file.Get())) {
        LogError("%s: Failed to commit filter file %d\n", __func__, pos.nFile);
        return false;
    }
 
    batch.Write(DB_FILTER_POS, pos);
    return true;
}
 
bool BlockFilterIndex::ReadFilterFromDisk(const FlatFilePos &pos,
                                          BlockFilter &filter) const {
    AutoFile filein{m_filter_fileseq->Open(pos, true)};
    if (filein.IsNull()) {
        return false;
    }
 
    BlockHash block_hash;
    std::vector<uint8_t> encoded_filter;
    try {
        filein >> block_hash >> encoded_filter;
        filter =
            BlockFilter(GetFilterType(), block_hash, std::move(encoded_filter));
    } catch (const std::exception &e) {
        LogError("%s: Failed to deserialize block filter from disk: %s\n",
                 __func__, e.what());
        return false;
    }
 
    return true;
}
 
size_t BlockFilterIndex::WriteFilterToDisk(FlatFilePos &pos,
                                           const BlockFilter &filter) {
    assert(filter.GetFilterType() == GetFilterType());
 
    size_t data_size = GetSerializeSize(filter.GetBlockHash()) +
                       GetSerializeSize(filter.GetEncodedFilter());
 
    // If writing the filter would overflow the file, flush and move to the next
    // one.
    if (pos.nPos + data_size > MAX_FLTR_FILE_SIZE) {
        AutoFile last_file{m_filter_fileseq->Open(pos)};
        if (last_file.IsNull()) {
            LogPrintf("%s: Failed to open filter file %d\n", __func__,
                      pos.nFile);
            return 0;
        }
        if (!TruncateFile(last_file.Get(), pos.nPos)) {
            LogPrintf("%s: Failed to truncate filter file %d\n", __func__,
                      pos.nFile);
            return 0;
        }
        if (!FileCommit(last_file.Get())) {
            LogPrintf("%s: Failed to commit filter file %d\n", __func__,
                      pos.nFile);
            return 0;
        }
 
        pos.nFile++;
        pos.nPos = 0;
    }
 
    // Pre-allocate sufficient space for filter data.
    bool out_of_space;
    m_filter_fileseq->Allocate(pos, data_size, out_of_space);
    if (out_of_space) {
        LogPrintf("%s: out of disk space\n", __func__);
        return 0;
    }
 
    AutoFile fileout{m_filter_fileseq->Open(pos)};
    if (fileout.IsNull()) {
        LogPrintf("%s: Failed to open filter file %d\n", __func__, pos.nFile);
        return 0;
    }
 
    fileout << filter.GetBlockHash() << filter.GetEncodedFilter();
    return data_size;
}
 
bool BlockFilterIndex::WriteBlock(const CBlock &block,
                                  const CBlockIndex *pindex) {
    CBlockUndo block_undo;
    uint256 prev_header;
 
    if (pindex->nHeight > 0) {
        if (!m_chainstate->m_blockman.ReadBlockUndo(block_undo, *pindex)) {
            return false;
        }
 
        std::pair<BlockHash, DBVal> read_out;
        if (!m_db->Read(DBHeightKey(pindex->nHeight - 1), read_out)) {
            return false;
        }
 
        BlockHash expected_block_hash = pindex->pprev->GetBlockHash();
        if (read_out.first != expected_block_hash) {
            LogError("%s: previous block header belongs to unexpected "
                     "block %s; expected %s\n",
                     __func__, read_out.first.ToString(),
                     expected_block_hash.ToString());
            return false;
        }
 
        prev_header = read_out.second.header;
    }
 
    BlockFilter filter(m_filter_type, block, block_undo);
 
    size_t bytes_written = WriteFilterToDisk(m_next_filter_pos, filter);
    if (bytes_written == 0) {
        return false;
    }
 
    std::pair<BlockHash, DBVal> value;
    value.first = pindex->GetBlockHash();
    value.second.hash = filter.GetHash();
    value.second.header = filter.ComputeHeader(prev_header);
    value.second.pos = m_next_filter_pos;
 
    m_db->Write(DBHeightKey(pindex->nHeight), value);
 
    m_next_filter_pos.nPos += bytes_written;
    return true;
}
 
static bool CopyHeightIndexToHashIndex(CDBIterator &db_it, CDBBatch &batch,
                                       const std::string &index_name,
                                       int start_height, int stop_height) {
    DBHeightKey key(start_height);
    db_it.Seek(key);
 
    for (int height = start_height; height <= stop_height; ++height) {
        if (!db_it.GetKey(key) || key.height != height) {
            LogError("%s: unexpected key in %s: expected (%c, %d)\n", __func__,
                     index_name, DB_BLOCK_HEIGHT, height);
            return false;
        }
 
        std::pair<BlockHash, DBVal> value;
        if (!db_it.GetValue(value)) {
            LogError("%s: unable to read value in %s at key (%c, %d)\n",
                     __func__, index_name, DB_BLOCK_HEIGHT, height);
            return false;
        }
 
        batch.Write(DBHashKey(value.first), std::move(value.second));
 
        db_it.Next();
    }
    return true;
}
 
bool BlockFilterIndex::Rewind(const CBlockIndex *current_tip,
                              const CBlockIndex *new_tip) {
    assert(current_tip->GetAncestor(new_tip->nHeight) == new_tip);
 
    CDBBatch batch(*m_db);
    std::unique_ptr<CDBIterator> db_it(m_db->NewIterator());
 
    // During a reorg, we need to copy all filters for blocks that are getting
    // disconnected from the height index to the hash index so we can still find
    // them when the height index entries are overwritten.
    if (!CopyHeightIndexToHashIndex(*db_it, batch, m_name, new_tip->nHeight,
                                    current_tip->nHeight)) {
        return false;
    }
 
    // The latest filter position gets written in Commit by the call to the
    // BaseIndex::Rewind. But since this creates new references to the filter,
    // the position should get updated here atomically as well in case Commit
    // fails.
    batch.Write(DB_FILTER_POS, m_next_filter_pos);
    m_db->WriteBatch(batch);
 
    return BaseIndex::Rewind(current_tip, new_tip);
}
 
static bool LookupOne(const CDBWrapper &db, const CBlockIndex *block_index,
                      DBVal &result) {
    // First check if the result is stored under the height index and the value
    // there matches the block hash. This should be the case if the block is on
    // the active chain.
    std::pair<BlockHash, DBVal> read_out;
    if (!db.Read(DBHeightKey(block_index->nHeight), read_out)) {
        return false;
    }
    if (read_out.first == block_index->GetBlockHash()) {
        result = std::move(read_out.second);
        return true;
    }
 
    // If value at the height index corresponds to an different block, the
    // result will be stored in the hash index.
    return db.Read(DBHashKey(block_index->GetBlockHash()), result);
}
 
static bool LookupRange(CDBWrapper &db, const std::string &index_name,
                        int start_height, const CBlockIndex *stop_index,
                        std::vector<DBVal> &results) {
    if (start_height < 0) {
        LogError("%s: start height (%d) is negative\n", __func__, start_height);
        return false;
    }
    if (start_height > stop_index->nHeight) {
        LogError("%s: start height (%d) is greater than stop height (%d)\n",
                 __func__, start_height, stop_index->nHeight);
        return false;
    }
 
    size_t results_size =
        static_cast<size_t>(stop_index->nHeight - start_height + 1);
    std::vector<std::pair<BlockHash, DBVal>> values(results_size);
 
    DBHeightKey key(start_height);
    std::unique_ptr<CDBIterator> db_it(db.NewIterator());
    db_it->Seek(DBHeightKey(start_height));
    for (int height = start_height; height <= stop_index->nHeight; ++height) {
        if (!db_it->Valid() || !db_it->GetKey(key) || key.height != height) {
            return false;
        }
 
        size_t i = static_cast<size_t>(height - start_height);
        if (!db_it->GetValue(values[i])) {
            LogError("%s: unable to read value in %s at key (%c, %d)\n",
                     __func__, index_name, DB_BLOCK_HEIGHT, height);
            return false;
        }
 
        db_it->Next();
    }
 
    results.resize(results_size);
 
    // Iterate backwards through block indexes collecting results in order to
    // access the block hash of each entry in case we need to look it up in the
    // hash index.
    for (const CBlockIndex *block_index = stop_index;
         block_index && block_index->nHeight >= start_height;
         block_index = block_index->pprev) {
        BlockHash block_hash = block_index->GetBlockHash();
 
        size_t i = static_cast<size_t>(block_index->nHeight - start_height);
        if (block_hash == values[i].first) {
            results[i] = std::move(values[i].second);
            continue;
        }
 
        if (!db.Read(DBHashKey(block_hash), results[i])) {
            LogError("%s: unable to read value in %s at key (%c, %s)\n",
                     __func__, index_name, DB_BLOCK_HASH,
                     block_hash.ToString());
            return false;
        }
    }
 
    return true;
}
 
bool BlockFilterIndex::LookupFilter(const CBlockIndex *block_index,
                                    BlockFilter &filter_out) const {
    DBVal entry;
    if (!LookupOne(*m_db, block_index, entry)) {
        return false;
    }
 
    return ReadFilterFromDisk(entry.pos, filter_out);
}
 
bool BlockFilterIndex::LookupFilterHeader(const CBlockIndex *block_index,
                                          uint256 &header_out) {
    LOCK(m_cs_headers_cache);
 
    bool is_checkpoint{block_index->nHeight % CFCHECKPT_INTERVAL == 0};
 
    if (is_checkpoint) {
        // Try to find the block in the headers cache if this is a checkpoint
        // height.
        auto header = m_headers_cache.find(block_index->GetBlockHash());
        if (header != m_headers_cache.end()) {
            header_out = header->second;
            return true;
        }
    }
 
    DBVal entry;
    if (!LookupOne(*m_db, block_index, entry)) {
        return false;
    }
 
    if (is_checkpoint && m_headers_cache.size() < CF_HEADERS_CACHE_MAX_SZ) {
        // Add to the headers cache if this is a checkpoint height.
        m_headers_cache.emplace(block_index->GetBlockHash(), entry.header);
    }
 
    header_out = entry.header;
    return true;
}
 
bool BlockFilterIndex::LookupFilterRange(
    int start_height, const CBlockIndex *stop_index,
    std::vector<BlockFilter> &filters_out) const {
    std::vector<DBVal> entries;
    if (!LookupRange(*m_db, m_name, start_height, stop_index, entries)) {
        return false;
    }
 
    filters_out.resize(entries.size());
    auto filter_pos_it = filters_out.begin();
    for (const auto &entry : entries) {
        if (!ReadFilterFromDisk(entry.pos, *filter_pos_it)) {
            return false;
        }
        ++filter_pos_it;
    }
 
    return true;
}
 
bool BlockFilterIndex::LookupFilterHashRange(
    int start_height, const CBlockIndex *stop_index,
    std::vector<uint256> &hashes_out) const
 
{
    std::vector<DBVal> entries;
    if (!LookupRange(*m_db, m_name, start_height, stop_index, entries)) {
        return false;
    }
 
    hashes_out.clear();
    hashes_out.reserve(entries.size());
    for (const auto &entry : entries) {
        hashes_out.push_back(entry.hash);
    }
    return true;
}
 
BlockFilterIndex *GetBlockFilterIndex(BlockFilterType filter_type) {
    auto it = g_filter_indexes.find(filter_type);
    return it != g_filter_indexes.end() ? &it->second : nullptr;
}
 
void ForEachBlockFilterIndex(std::function<void(BlockFilterIndex &)> fn) {
    for (auto &entry : g_filter_indexes) {
        fn(entry.second);
    }
}
 
bool InitBlockFilterIndex(
    std::function<std::unique_ptr<interfaces::Chain>()> make_chain,
    BlockFilterType filter_type, size_t n_cache_size, bool f_memory,
    bool f_wipe) {
    auto result = g_filter_indexes.emplace(
        std::piecewise_construct, std::forward_as_tuple(filter_type),
        std::forward_as_tuple(make_chain(), filter_type, n_cache_size, f_memory,
                              f_wipe));
    return result.second;
}
 
bool DestroyBlockFilterIndex(BlockFilterType filter_type) {
    return g_filter_indexes.erase(filter_type);
}
 
void DestroyAllBlockFilterIndexes() {
    g_filter_indexes.clear();
}