From 75a76e36199eba228856d59318cb54ca64ca6b57 Mon Sep 17 00:00:00 2001 From: Jim Posen Date: Mon, 27 Aug 2018 17:01:24 -0700 Subject: index: Implement block filter index with write operations. --- src/index/blockfilterindex.cpp | 282 +++++++++++++++++++++++++++++++++++++++++ src/index/blockfilterindex.h | 53 ++++++++ 2 files changed, 335 insertions(+) create mode 100644 src/index/blockfilterindex.cpp create mode 100644 src/index/blockfilterindex.h (limited to 'src/index') diff --git a/src/index/blockfilterindex.cpp b/src/index/blockfilterindex.cpp new file mode 100644 index 0000000000..8fe0ed418b --- /dev/null +++ b/src/index/blockfilterindex.cpp @@ -0,0 +1,282 @@ +// 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 + +#include +#include +#include + +/* The index database stores three items for each block: the disk location of the encoded filter, + * its dSHA256 hash, and the header. Those belonging to blocks on the active chain are indexed by + * height, and those belonging to blocks that have been reorganized out of the active chain are + * indexed by block hash. This ensures that filter data for any block that becomes part of the + * active chain can always be retrieved, alleviating timing concerns. + * + * The filters themselves are stored in flat files and referenced by the LevelDB entries. This + * minimizes the amount of data written to LevelDB and keeps the database values constant size. The + * disk location of the next block filter to be written (represented as a FlatFilePos) is stored + * under the DB_FILTER_POS key. + * + * Keys for the height index have the type [DB_BLOCK_HEIGHT, uint32 (BE)]. The height is represented + * as big-endian so that sequential reads of filters by height are fast. + * Keys for the hash index have the type [DB_BLOCK_HASH, uint256]. + */ +constexpr char DB_BLOCK_HASH = 's'; +constexpr char DB_BLOCK_HEIGHT = 't'; +constexpr char DB_FILTER_POS = 'P'; + +constexpr unsigned int MAX_FLTR_FILE_SIZE = 0x1000000; // 16 MiB +/** The pre-allocation chunk size for fltr?????.dat files */ +constexpr unsigned int FLTR_FILE_CHUNK_SIZE = 0x100000; // 1 MiB + +namespace { + +struct DBVal { + uint256 hash; + uint256 header; + FlatFilePos pos; + + ADD_SERIALIZE_METHODS; + + template + inline void SerializationOp(Stream& s, Operation ser_action) { + READWRITE(hash); + READWRITE(header); + READWRITE(pos); + } +}; + +struct DBHeightKey { + int height; + + DBHeightKey() : height(0) {} + DBHeightKey(int height_in) : height(height_in) {} + + template + void Serialize(Stream& s) const + { + ser_writedata8(s, DB_BLOCK_HEIGHT); + ser_writedata32be(s, height); + } + + template + void Unserialize(Stream& s) + { + char 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 { + uint256 hash; + + DBHashKey(const uint256& hash_in) : hash(hash_in) {} + + ADD_SERIALIZE_METHODS; + + template + inline void SerializationOp(Stream& s, Operation ser_action) { + char 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(hash); + } +}; + +}; // namespace + +BlockFilterIndex::BlockFilterIndex(BlockFilterType filter_type, + size_t n_cache_size, bool f_memory, bool f_wipe) + : 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 = GetDataDir() / "indexes" / "blockfilter" / filter_name; + fs::create_directories(path); + + m_name = filter_name + " block filter index"; + m_db = MakeUnique(path / "db", n_cache_size, f_memory, f_wipe); + m_filter_fileseq = MakeUnique(std::move(path), "fltr", FLTR_FILE_CHUNK_SIZE); +} + +bool BlockFilterIndex::Init() +{ + 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)) { + return error("%s: Cannot read current %s state; index may be corrupted", + __func__, GetName()); + } + + // 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 BaseIndex::Init(); +} + +bool BlockFilterIndex::CommitInternal(CDBBatch& batch) +{ + const FlatFilePos& pos = m_next_filter_pos; + + // Flush current filter file to disk. + CAutoFile file(m_filter_fileseq->Open(pos), SER_DISK, CLIENT_VERSION); + if (file.IsNull()) { + return error("%s: Failed to open filter file %d", __func__, pos.nFile); + } + if (!FileCommit(file.Get())) { + return error("%s: Failed to commit filter file %d", __func__, pos.nFile); + } + + batch.Write(DB_FILTER_POS, pos); + return BaseIndex::CommitInternal(batch); +} + +size_t BlockFilterIndex::WriteFilterToDisk(FlatFilePos& pos, const BlockFilter& filter) +{ + assert(filter.GetFilterType() == GetFilterType()); + + size_t data_size = + GetSerializeSize(filter.GetBlockHash(), CLIENT_VERSION) + + GetSerializeSize(filter.GetEncodedFilter(), CLIENT_VERSION); + + // If writing the filter would overflow the file, flush and move to the next one. + if (pos.nPos + data_size > MAX_FLTR_FILE_SIZE) { + CAutoFile last_file(m_filter_fileseq->Open(pos), SER_DISK, CLIENT_VERSION); + 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; + } + + CAutoFile fileout(m_filter_fileseq->Open(pos), SER_DISK, CLIENT_VERSION); + 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 (!UndoReadFromDisk(block_undo, pindex)) { + return false; + } + + std::pair read_out; + if (!m_db->Read(DBHeightKey(pindex->nHeight - 1), read_out)) { + return false; + } + + uint256 expected_block_hash = pindex->pprev->GetBlockHash(); + if (read_out.first != expected_block_hash) { + return error("%s: previous block header belongs to unexpected block %s; expected %s", + __func__, read_out.first.ToString(), expected_block_hash.ToString()); + } + + 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 value; + value.first = pindex->GetBlockHash(); + value.second.hash = filter.GetHash(); + value.second.header = filter.ComputeHeader(prev_header); + value.second.pos = m_next_filter_pos; + + if (!m_db->Write(DBHeightKey(pindex->nHeight), value)) { + return false; + } + + 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) { + return error("%s: unexpected key in %s: expected (%c, %d)", + __func__, index_name, DB_BLOCK_HEIGHT, height); + } + + std::pair value; + if (!db_it.GetValue(value)) { + return error("%s: unable to read value in %s at key (%c, %d)", + __func__, index_name, DB_BLOCK_HEIGHT, height); + } + + 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 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); + if (!m_db->WriteBatch(batch)) return false; + + return BaseIndex::Rewind(current_tip, new_tip); +} diff --git a/src/index/blockfilterindex.h b/src/index/blockfilterindex.h new file mode 100644 index 0000000000..d2dd3a27bc --- /dev/null +++ b/src/index/blockfilterindex.h @@ -0,0 +1,53 @@ +// 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. + +#ifndef BITCOIN_INDEX_BLOCKFILTERINDEX_H +#define BITCOIN_INDEX_BLOCKFILTERINDEX_H + +#include +#include +#include +#include + +/** + * BlockFilterIndex is used to store and retrieve block filters, hashes, and headers for a range of + * blocks by height. An index is constructed for each supported filter type with its own database + * (ie. filter data for different types are stored in separate databases). + * + * This index is used to serve BIP 157 net requests. + */ +class BlockFilterIndex final : public BaseIndex +{ +private: + BlockFilterType m_filter_type; + std::string m_name; + std::unique_ptr m_db; + + FlatFilePos m_next_filter_pos; + std::unique_ptr m_filter_fileseq; + + size_t WriteFilterToDisk(FlatFilePos& pos, const BlockFilter& filter); + +protected: + bool Init() override; + + bool CommitInternal(CDBBatch& batch) override; + + bool WriteBlock(const CBlock& block, const CBlockIndex* pindex) override; + + bool Rewind(const CBlockIndex* current_tip, const CBlockIndex* new_tip) override; + + BaseIndex::DB& GetDB() const override { return *m_db; } + + const char* GetName() const override { return m_name.c_str(); } + +public: + /** Constructs the index, which becomes available to be queried. */ + explicit BlockFilterIndex(BlockFilterType filter_type, + size_t n_cache_size, bool f_memory = false, bool f_wipe = false); + + BlockFilterType GetFilterType() const { return m_filter_type; } +}; + +#endif // BITCOIN_INDEX_BLOCKFILTERINDEX_H -- cgit v1.2.3