Branch data Line data Source code
1 : : // Copyright (c) 2011-2022 The Bitcoin Core developers
2 : : // Distributed under the MIT software license, see the accompanying
3 : : // file COPYING or http://www.opensource.org/licenses/mit-license.php.
4 : :
5 : : #ifndef BITCOIN_NODE_BLOCKSTORAGE_H
6 : : #define BITCOIN_NODE_BLOCKSTORAGE_H
7 : :
8 : : #include <attributes.h>
9 : : #include <chain.h>
10 : : #include <dbwrapper.h>
11 : : #include <flatfile.h>
12 : : #include <kernel/blockmanager_opts.h>
13 : : #include <kernel/chainparams.h>
14 : : #include <kernel/cs_main.h>
15 : : #include <kernel/messagestartchars.h>
16 : : #include <primitives/block.h>
17 : : #include <streams.h>
18 : : #include <sync.h>
19 : : #include <uint256.h>
20 : : #include <util/fs.h>
21 : : #include <util/hasher.h>
22 : :
23 : : #include <array>
24 : : #include <atomic>
25 : : #include <cstdint>
26 : : #include <functional>
27 : : #include <limits>
28 : : #include <map>
29 : : #include <memory>
30 : : #include <optional>
31 : : #include <set>
32 : : #include <span>
33 : : #include <string>
34 : : #include <unordered_map>
35 : : #include <utility>
36 : : #include <vector>
37 : :
38 : : class BlockValidationState;
39 : : class CBlockUndo;
40 : : class Chainstate;
41 : : class ChainstateManager;
42 : : namespace Consensus {
43 : : struct Params;
44 : : }
45 : : namespace util {
46 : : class SignalInterrupt;
47 : : } // namespace util
48 : :
49 : : namespace kernel {
50 : : /** Access to the block database (blocks/index/) */
51 : 338 : class BlockTreeDB : public CDBWrapper
52 : : {
53 : : public:
54 [ + - ]: 338 : using CDBWrapper::CDBWrapper;
55 : : bool WriteBatchSync(const std::vector<std::pair<int, const CBlockFileInfo*>>& fileInfo, int nLastFile, const std::vector<const CBlockIndex*>& blockinfo);
56 : : bool ReadBlockFileInfo(int nFile, CBlockFileInfo& info);
57 : : bool ReadLastBlockFile(int& nFile);
58 : : bool WriteReindexing(bool fReindexing);
59 : : void ReadReindexing(bool& fReindexing);
60 : : bool WriteFlag(const std::string& name, bool fValue);
61 : : bool ReadFlag(const std::string& name, bool& fValue);
62 : : bool LoadBlockIndexGuts(const Consensus::Params& consensusParams, std::function<CBlockIndex*(const uint256&)> insertBlockIndex, const util::SignalInterrupt& interrupt)
63 : : EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
64 : : };
65 : : } // namespace kernel
66 : :
67 : : namespace node {
68 : : using kernel::BlockTreeDB;
69 : :
70 : : /** The pre-allocation chunk size for blk?????.dat files (since 0.8) */
71 : : static const unsigned int BLOCKFILE_CHUNK_SIZE = 0x1000000; // 16 MiB
72 : : /** The pre-allocation chunk size for rev?????.dat files (since 0.8) */
73 : : static const unsigned int UNDOFILE_CHUNK_SIZE = 0x100000; // 1 MiB
74 : : /** The maximum size of a blk?????.dat file (since 0.8) */
75 : : static const unsigned int MAX_BLOCKFILE_SIZE = 0x8000000; // 128 MiB
76 : :
77 : : /** Size of header written by WriteBlockToDisk before a serialized CBlock */
78 : : static constexpr size_t BLOCK_SERIALIZATION_HEADER_SIZE = std::tuple_size_v<MessageStartChars> + sizeof(unsigned int);
79 : :
80 : : // Because validation code takes pointers to the map's CBlockIndex objects, if
81 : : // we ever switch to another associative container, we need to either use a
82 : : // container that has stable addressing (true of all std associative
83 : : // containers), or make the key a `std::unique_ptr<CBlockIndex>`
84 : : using BlockMap = std::unordered_map<uint256, CBlockIndex, BlockHasher>;
85 : :
86 : : struct CBlockIndexWorkComparator {
87 : : bool operator()(const CBlockIndex* pa, const CBlockIndex* pb) const;
88 : : };
89 : :
90 : : struct CBlockIndexHeightOnlyComparator {
91 : : /* Only compares the height of two block indices, doesn't try to tie-break */
92 : : bool operator()(const CBlockIndex* pa, const CBlockIndex* pb) const;
93 : : };
94 : :
95 : 3 : struct PruneLockInfo {
96 : : int height_first{std::numeric_limits<int>::max()}; //! Height of earliest block that should be kept and not pruned
97 : : };
98 : :
99 : : enum BlockfileType {
100 : : // Values used as array indexes - do not change carelessly.
101 : : NORMAL = 0,
102 : : ASSUMED = 1,
103 : : NUM_TYPES = 2,
104 : : };
105 : :
106 : : std::ostream& operator<<(std::ostream& os, const BlockfileType& type);
107 : :
108 : : struct BlockfileCursor {
109 : : // The latest blockfile number.
110 : : int file_num{0};
111 : :
112 : : // Track the height of the highest block in file_num whose undo
113 : : // data has been written. Block data is written to block files in download
114 : : // order, but is written to undo files in validation order, which is
115 : : // usually in order by height. To avoid wasting disk space, undo files will
116 : : // be trimmed whenever the corresponding block file is finalized and
117 : : // the height of the highest block written to the block file equals the
118 : : // height of the highest block written to the undo file. This is a
119 : : // heuristic and can sometimes preemptively trim undo files that will write
120 : : // more data later, and sometimes fail to trim undo files that can't have
121 : : // more data written later.
122 : : int undo_height{0};
123 : : };
124 : :
125 : : std::ostream& operator<<(std::ostream& os, const BlockfileCursor& cursor);
126 : :
127 : :
128 : : /**
129 : : * Maintains a tree of blocks (stored in `m_block_index`) which is consulted
130 : : * to determine where the most-work tip is.
131 : : *
132 : : * This data is used mostly in `Chainstate` - information about, e.g.,
133 : : * candidate tips is not maintained here.
134 : : */
135 : : class BlockManager
136 : : {
137 : : friend Chainstate;
138 : : friend ChainstateManager;
139 : :
140 : : private:
141 [ - - + - : 14455 : const CChainParams& GetParams() const { return m_opts.chainparams; }
+ - ]
142 [ + - - + : 4288 : const Consensus::Params& GetConsensus() const { return m_opts.chainparams.GetConsensus(); }
- - + - ]
143 : : /**
144 : : * Load the blocktree off disk and into memory. Populate certain metadata
145 : : * per index entry (nStatus, nChainWork, nTimeMax, etc.) as well as peripheral
146 : : * collections like m_dirty_blockindex.
147 : : */
148 : : bool LoadBlockIndex(const std::optional<uint256>& snapshot_blockhash)
149 : : EXCLUSIVE_LOCKS_REQUIRED(cs_main);
150 : :
151 : : /** Return false if block file or undo file flushing fails. */
152 : : [[nodiscard]] bool FlushBlockFile(int blockfile_num, bool fFinalize, bool finalize_undo);
153 : :
154 : : /** Return false if undo file flushing fails. */
155 : : [[nodiscard]] bool FlushUndoFile(int block_file, bool finalize = false);
156 : :
157 : : /**
158 : : * Helper function performing various preparations before a block can be saved to disk:
159 : : * Returns the correct position for the block to be saved, which may be in the current or a new
160 : : * block file depending on nAddSize. May flush the previous blockfile to disk if full, updates
161 : : * blockfile info, and checks if there is enough disk space to save the block.
162 : : *
163 : : * The nAddSize argument passed to this function should include not just the size of the serialized CBlock, but also the size of
164 : : * separator fields which are written before it by WriteBlockToDisk (BLOCK_SERIALIZATION_HEADER_SIZE).
165 : : */
166 : : [[nodiscard]] FlatFilePos FindNextBlockPos(unsigned int nAddSize, unsigned int nHeight, uint64_t nTime);
167 : : [[nodiscard]] bool FlushChainstateBlockFile(int tip_height);
168 : : bool FindUndoPos(BlockValidationState& state, int nFile, FlatFilePos& pos, unsigned int nAddSize);
169 : :
170 : : AutoFile OpenUndoFile(const FlatFilePos& pos, bool fReadOnly = false) const;
171 : :
172 : : /**
173 : : * Write a block to disk. The pos argument passed to this function is modified by this call. Before this call, it should
174 : : * point to an unused file location where separator fields will be written, followed by the serialized CBlock data.
175 : : * After this call, it will point to the beginning of the serialized CBlock data, after the separator fields
176 : : * (BLOCK_SERIALIZATION_HEADER_SIZE)
177 : : */
178 : : bool WriteBlockToDisk(const CBlock& block, FlatFilePos& pos) const;
179 : : bool UndoWriteToDisk(const CBlockUndo& blockundo, FlatFilePos& pos, const uint256& hashBlock) const;
180 : :
181 : : /* Calculate the block/rev files to delete based on height specified by user with RPC command pruneblockchain */
182 : : void FindFilesToPruneManual(
183 : : std::set<int>& setFilesToPrune,
184 : : int nManualPruneHeight,
185 : : const Chainstate& chain,
186 : : ChainstateManager& chainman);
187 : :
188 : : /**
189 : : * Prune block and undo files (blk???.dat and rev???.dat) so that the disk space used is less than a user-defined target.
190 : : * The user sets the target (in MB) on the command line or in config file. This will be run on startup and whenever new
191 : : * space is allocated in a block or undo file, staying below the target. Changing back to unpruned requires a reindex
192 : : * (which in this case means the blockchain must be re-downloaded.)
193 : : *
194 : : * Pruning functions are called from FlushStateToDisk when the m_check_for_pruning flag has been set.
195 : : * Block and undo files are deleted in lock-step (when blk00003.dat is deleted, so is rev00003.dat.)
196 : : * Pruning cannot take place until the longest chain is at least a certain length (CChainParams::nPruneAfterHeight).
197 : : * Pruning will never delete a block within a defined distance (currently 288) from the active chain's tip.
198 : : * The block index is updated by unsetting HAVE_DATA and HAVE_UNDO for any blocks that were stored in the deleted files.
199 : : * A db flag records the fact that at least some block files have been pruned.
200 : : *
201 : : * @param[out] setFilesToPrune The set of file indices that can be unlinked will be returned
202 : : * @param last_prune The last height we're able to prune, according to the prune locks
203 : : */
204 : : void FindFilesToPrune(
205 : : std::set<int>& setFilesToPrune,
206 : : int last_prune,
207 : : const Chainstate& chain,
208 : : ChainstateManager& chainman);
209 : :
210 : : RecursiveMutex cs_LastBlockFile;
211 : : std::vector<CBlockFileInfo> m_blockfile_info;
212 : :
213 : : //! Since assumedvalid chainstates may be syncing a range of the chain that is very
214 : : //! far away from the normal/background validation process, we should segment blockfiles
215 : : //! for assumed chainstates. Otherwise, we might have wildly different height ranges
216 : : //! mixed into the same block files, which would impair our ability to prune
217 : : //! effectively.
218 : : //!
219 : : //! This data structure maintains separate blockfile number cursors for each
220 : : //! BlockfileType. The ASSUMED state is initialized, when necessary, in FindNextBlockPos().
221 : : //!
222 : : //! The first element is the NORMAL cursor, second is ASSUMED.
223 : : std::array<std::optional<BlockfileCursor>, BlockfileType::NUM_TYPES>
224 : : m_blockfile_cursors GUARDED_BY(cs_LastBlockFile) = {
225 : : BlockfileCursor{},
226 : : std::nullopt,
227 : : };
228 : 251 : int MaxBlockfileNum() const EXCLUSIVE_LOCKS_REQUIRED(cs_LastBlockFile)
229 : : {
230 : 251 : static const BlockfileCursor empty_cursor;
231 [ + - ]: 251 : const auto& normal = m_blockfile_cursors[BlockfileType::NORMAL].value_or(empty_cursor);
232 [ + + ]: 251 : const auto& assumed = m_blockfile_cursors[BlockfileType::ASSUMED].value_or(empty_cursor);
233 [ + + ]: 251 : return std::max(normal.file_num, assumed.file_num);
234 : : }
235 : :
236 : : /** Global flag to indicate we should check to see if there are
237 : : * block/undo files that should be deleted. Set on startup
238 : : * or if we allocate more file space when we're in prune mode
239 : : */
240 : : bool m_check_for_pruning = false;
241 : :
242 : : const bool m_prune_mode;
243 : :
244 : : const std::vector<std::byte> m_xor_key;
245 : :
246 : : /** Dirty block index entries. */
247 : : std::set<CBlockIndex*> m_dirty_blockindex;
248 : :
249 : : /** Dirty block file entries. */
250 : : std::set<int> m_dirty_fileinfo;
251 : :
252 : : /**
253 : : * Map from external index name to oldest block that must not be pruned.
254 : : *
255 : : * @note Internally, only blocks at height (height_first - PRUNE_LOCK_BUFFER - 1) and
256 : : * below will be pruned, but callers should avoid assuming any particular buffer size.
257 : : */
258 : : std::unordered_map<std::string, PruneLockInfo> m_prune_locks GUARDED_BY(::cs_main);
259 : :
260 : : BlockfileType BlockfileTypeForHeight(int height);
261 : :
262 : : const kernel::BlockManagerOpts m_opts;
263 : :
264 : : const FlatFileSeq m_block_file_seq;
265 : : const FlatFileSeq m_undo_file_seq;
266 : :
267 : : public:
268 : : using Options = kernel::BlockManagerOpts;
269 : :
270 : : explicit BlockManager(const util::SignalInterrupt& interrupt, Options opts);
271 : :
272 : : const util::SignalInterrupt& m_interrupt;
273 : : std::atomic<bool> m_importing{false};
274 : :
275 : : /**
276 : : * Whether all blockfiles have been added to the block tree database.
277 : : * Normally true, but set to false when a reindex is requested and the
278 : : * database is wiped. The value is persisted in the database across restarts
279 : : * and will be false until reindexing completes.
280 : : */
281 : : std::atomic_bool m_blockfiles_indexed{true};
282 : :
283 : : BlockMap m_block_index GUARDED_BY(cs_main);
284 : :
285 : : /**
286 : : * The height of the base block of an assumeutxo snapshot, if one is in use.
287 : : *
288 : : * This controls how blockfiles are segmented by chainstate type to avoid
289 : : * comingling different height regions of the chain when an assumedvalid chainstate
290 : : * is in use. If heights are drastically different in the same blockfile, pruning
291 : : * suffers.
292 : : *
293 : : * This is set during ActivateSnapshot() or upon LoadBlockIndex() if a snapshot
294 : : * had been previously loaded. After the snapshot is validated, this is unset to
295 : : * restore normal LoadBlockIndex behavior.
296 : : */
297 : : std::optional<int> m_snapshot_height;
298 : :
299 : : std::vector<CBlockIndex*> GetAllBlockIndices() EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
300 : :
301 : : /**
302 : : * All pairs A->B, where A (or one of its ancestors) misses transactions, but B has transactions.
303 : : * Pruned nodes may have entries where B is missing data.
304 : : */
305 : : std::multimap<CBlockIndex*, CBlockIndex*> m_blocks_unlinked;
306 : :
307 : : std::unique_ptr<BlockTreeDB> m_block_tree_db GUARDED_BY(::cs_main);
308 : :
309 : : bool WriteBlockIndexDB() EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
310 : : bool LoadBlockIndexDB(const std::optional<uint256>& snapshot_blockhash)
311 : : EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
312 : :
313 : : /**
314 : : * Remove any pruned block & undo files that are still on disk.
315 : : * This could happen on some systems if the file was still being read while unlinked,
316 : : * or if we crash before unlinking.
317 : : */
318 : : void ScanAndUnlinkAlreadyPrunedFiles() EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
319 : :
320 : : CBlockIndex* AddToBlockIndex(const CBlockHeader& block, CBlockIndex*& best_header) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
321 : : /** Create a new block index entry for a given block hash */
322 : : CBlockIndex* InsertBlockIndex(const uint256& hash) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
323 : :
324 : : //! Mark one block file as pruned (modify associated database entries)
325 : : void PruneOneBlockFile(const int fileNumber) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
326 : :
327 : : CBlockIndex* LookupBlockIndex(const uint256& hash) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
328 : : const CBlockIndex* LookupBlockIndex(const uint256& hash) const EXCLUSIVE_LOCKS_REQUIRED(cs_main);
329 : :
330 : : /** Get block file info entry for one block file */
331 : : CBlockFileInfo* GetBlockFileInfo(size_t n);
332 : :
333 : : bool WriteUndoDataForBlock(const CBlockUndo& blockundo, BlockValidationState& state, CBlockIndex& block)
334 : : EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
335 : :
336 : : /** Store block on disk and update block file statistics.
337 : : *
338 : : * @param[in] block the block to be stored
339 : : * @param[in] nHeight the height of the block
340 : : *
341 : : * @returns in case of success, the position to which the block was written to
342 : : * in case of an error, an empty FlatFilePos
343 : : */
344 : : FlatFilePos SaveBlockToDisk(const CBlock& block, int nHeight);
345 : :
346 : : /** Update blockfile info while processing a block during reindex. The block must be available on disk.
347 : : *
348 : : * @param[in] block the block being processed
349 : : * @param[in] nHeight the height of the block
350 : : * @param[in] pos the position of the serialized CBlock on disk. This is the position returned
351 : : * by WriteBlockToDisk pointing at the CBlock, not the separator fields before it
352 : : */
353 : : void UpdateBlockInfo(const CBlock& block, unsigned int nHeight, const FlatFilePos& pos);
354 : :
355 : : /** Whether running in -prune mode. */
356 [ + + # # : 22476 : [[nodiscard]] bool IsPruneMode() const { return m_prune_mode; }
# # ][ - +
+ + - + ]
[ # # # #
# # # # #
# ][ # # #
# # # # #
# # ]
357 : :
358 : : /** Attempt to stay below this number of bytes of block files. */
359 [ - + ]: 168 : [[nodiscard]] uint64_t GetPruneTarget() const { return m_opts.prune_target; }
[ # # # # ]
360 : : static constexpr auto PRUNE_TARGET_MANUAL{std::numeric_limits<uint64_t>::max()};
361 : :
362 [ + - - + ]: 18690 : [[nodiscard]] bool LoadingBlocks() const { return m_importing || !m_blockfiles_indexed; }
363 : :
364 : : /** Calculate the amount of disk space the block & undo files currently use */
365 : : uint64_t CalculateCurrentUsage();
366 : :
367 : : //! Returns last CBlockIndex* that is a checkpoint
368 : : const CBlockIndex* GetLastCheckpoint(const CCheckpointData& data) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
369 : :
370 : : //! Check if all blocks in the [upper_block, lower_block] range have data available.
371 : : //! The caller is responsible for ensuring that lower_block is an ancestor of upper_block
372 : : //! (part of the same chain).
373 : : bool CheckBlockDataAvailability(const CBlockIndex& upper_block LIFETIMEBOUND, const CBlockIndex& lower_block LIFETIMEBOUND) EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
374 : :
375 : : /**
376 : : * @brief Returns the earliest block with specified `status_mask` flags set after
377 : : * the latest block _not_ having those flags.
378 : : *
379 : : * This function starts from `upper_block`, which must have all
380 : : * `status_mask` flags set, and iterates backwards through its ancestors. It
381 : : * continues as long as each block has all `status_mask` flags set, until
382 : : * reaching the oldest ancestor or `lower_block`.
383 : : *
384 : : * @pre `upper_block` must have all `status_mask` flags set.
385 : : * @pre `lower_block` must be null or an ancestor of `upper_block`
386 : : *
387 : : * @param upper_block The starting block for the search, which must have all
388 : : * `status_mask` flags set.
389 : : * @param status_mask Bitmask specifying required status flags.
390 : : * @param lower_block The earliest possible block to return. If null, the
391 : : * search can extend to the genesis block.
392 : : *
393 : : * @return A non-null pointer to the earliest block between `upper_block`
394 : : * and `lower_block`, inclusive, such that every block between the
395 : : * returned block and `upper_block` has `status_mask` flags set.
396 : : */
397 : : const CBlockIndex* GetFirstBlock(
398 : : const CBlockIndex& upper_block LIFETIMEBOUND,
399 : : uint32_t status_mask,
400 : : const CBlockIndex* lower_block = nullptr
401 : : ) const EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
402 : :
403 : : /** True if any block files have ever been pruned. */
404 : : bool m_have_pruned = false;
405 : :
406 : : //! Check whether the block associated with this index entry is pruned or not.
407 : : bool IsBlockPruned(const CBlockIndex& block) const EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
408 : :
409 : : //! Create or update a prune lock identified by its name
410 : : void UpdatePruneLock(const std::string& name, const PruneLockInfo& lock_info) EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
411 : :
412 : : /** Open a block file (blk?????.dat) */
413 : : AutoFile OpenBlockFile(const FlatFilePos& pos, bool fReadOnly = false) const;
414 : :
415 : : /** Translation to a filesystem path */
416 : : fs::path GetBlockPosFilename(const FlatFilePos& pos) const;
417 : :
418 : : /**
419 : : * Actually unlink the specified files
420 : : */
421 : : void UnlinkPrunedFiles(const std::set<int>& setFilesToPrune) const;
422 : :
423 : : /** Functions for disk access for blocks */
424 : : bool ReadBlockFromDisk(CBlock& block, const FlatFilePos& pos) const;
425 : : bool ReadBlockFromDisk(CBlock& block, const CBlockIndex& index) const;
426 : : bool ReadRawBlockFromDisk(std::vector<uint8_t>& block, const FlatFilePos& pos) const;
427 : :
428 : : bool UndoReadFromDisk(CBlockUndo& blockundo, const CBlockIndex& index) const;
429 : :
430 : : void CleanupBlockRevFiles() const;
431 : : };
432 : :
433 : : void ImportBlocks(ChainstateManager& chainman, std::span<const fs::path> import_paths);
434 : : } // namespace node
435 : :
436 : : #endif // BITCOIN_NODE_BLOCKSTORAGE_H
|
