index: Implement block filter index with write operations.

src/Makefile.am

@@ -133,6 +133,7 @@ BITCOIN_CORE_H = \
   httprpc.h \
   httpserver.h \
   index/base.h \
+  index/blockfilterindex.h \
   index/txindex.h \
   indirectmap.h \
   init.h \
@@ -253,6 +254,7 @@ libbitcoin_server_a_SOURCES = \
   httprpc.cpp \
   httpserver.cpp \
   index/base.cpp \
+  index/blockfilterindex.cpp \
   index/txindex.cpp \
   interfaces/chain.cpp \
   interfaces/handler.cpp \

src/index/blockfilterindex.cpp

@@ -0,0 +1,272 @@
+// 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 <map>
+
+#include <index/blockfilterindex.h>
+#include <util/system.h>
+#include <validation.h>
+
+/* 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.
+ *
+ * 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_FILE_SIZE = 0x1000000; // 16 MiB
+/** The pre-allocation chunk size for fltr?????.dat files */
+constexpr unsigned int FILE_CHUNK_SIZE = 0x100000; // 1 MiB
+
+namespace {
+    struct DBVal
+    {
+        uint256 hash;
+        uint256 header;
+        FlatFilePos pos;
+
+        ADD_SERIALIZE_METHODS;
+
+        template <typename Stream, typename Operation>
+        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<typename Stream>
+        void Serialize(Stream& s) const
+        {
+            ser_writedata8(s, DB_BLOCK_HEIGHT);
+            ser_writedata32be(s, height);
+        }
+
+        template<typename Stream>
+        void Unserialize(Stream& s)
+        {
+            char prefix = ser_readdata8(s);
+            if (prefix != DB_BLOCK_HEIGHT) {
+                throw std::ios_base::failure("Invalid format for DB key");
+            }
+            height = ser_readdata32be(s);
+        }
+    };
+
+    struct DBHashKey
+    {
+        uint256 hash;
+
+        DBHashKey(const uint256& hash_in) : hash(hash_in) {}
+
+        ADD_SERIALIZE_METHODS;
+
+        template <typename Stream, typename Operation>
+        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 DB key");
+            }
+
+            READWRITE(hash);
+        }
+    };
+};
+
+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 == "") 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<BaseIndex::DB>(path / "db", n_cache_size, f_memory, f_wipe);
+    m_filter_fileseq = MakeUnique<FlatFileSeq>(std::move(path), "fltr", FILE_CHUNK_SIZE);
+}
+
+bool BlockFilterIndex::Init()
+{
+    if (!m_db->Read(DB_FILTER_POS, m_next_filter_pos)) {
+        m_next_filter_pos.nFile = 0;
+        m_next_filter_pos.nPos = 0;
+    }
+    return BaseIndex::Init();
+}
+
+bool BlockFilterIndex::Commit(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::Commit(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_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<uint256, DBVal> 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<uint256, 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;
+
+    if (!m_db->Write(DBHeightKey(pindex->nHeight), value)) {
+        return false;
+    }
+
+    m_next_filter_pos += bytes_written;
+    return true;
+}
+
+static bool CopyHeightIndexToHashIndex(CDBIterator& db_it, CDBBatch& batch,
+                                       const std::string& index_name, int start_height,
+                                       const CBlockIndex* stop_index)
+{
+    DBHeightKey key(start_height);
+    db_it.Seek(key);
+
+    for (int height = start_height; height <= stop_index->nHeight; ++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<uint256, DBVal> 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<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)) {
+        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);
+}

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 <blockfilter.h>
+#include <chain.h>
+#include <flatfile.h>
+#include <index/base.h>
+
+/**
+ * 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<BaseIndex::DB> m_db;
+
+    FlatFilePos m_next_filter_pos;
+    std::unique_ptr<FlatFileSeq> m_filter_fileseq;
+
+    size_t WriteFilterToDisk(FlatFilePos& pos, const BlockFilter& filter);
+
+protected:
+    bool Init() override;
+
+    bool Commit(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