/ src / node / blockstorage.h
blockstorage.h
  1  // Copyright (c) 2011-present 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 <serialize.h>
 18  #include <streams.h>
 19  #include <sync.h>
 20  #include <uint256.h>
 21  #include <util/expected.h>
 22  #include <util/fs.h>
 23  #include <util/hasher.h>
 24  #include <util/obfuscation.h>
 25  
 26  #include <algorithm>
 27  #include <array>
 28  #include <atomic>
 29  #include <cstddef>
 30  #include <cstdint>
 31  #include <functional>
 32  #include <iosfwd>
 33  #include <limits>
 34  #include <map>
 35  #include <memory>
 36  #include <optional>
 37  #include <set>
 38  #include <span>
 39  #include <string>
 40  #include <unordered_map>
 41  #include <utility>
 42  #include <vector>
 43  
 44  class BlockValidationState;
 45  class CBlockUndo;
 46  class Chainstate;
 47  class ChainstateManager;
 48  namespace Consensus {
 49  struct Params;
 50  }
 51  namespace util {
 52  class SignalInterrupt;
 53  } // namespace util
 54  
 55  namespace kernel {
 56  class CBlockFileInfo
 57  {
 58  public:
 59      uint32_t nBlocks{};      //!< number of blocks stored in file
 60      uint32_t nSize{};        //!< number of used bytes of block file
 61      uint32_t nUndoSize{};    //!< number of used bytes in the undo file
 62      uint32_t nHeightFirst{}; //!< lowest height of block in file
 63      uint32_t nHeightLast{};  //!< highest height of block in file
 64      uint64_t nTimeFirst{};   //!< earliest time of block in file
 65      uint64_t nTimeLast{};    //!< latest time of block in file
 66  
 67      SERIALIZE_METHODS(CBlockFileInfo, obj)
 68      {
 69          READWRITE(VARINT(obj.nBlocks));
 70          READWRITE(VARINT(obj.nSize));
 71          READWRITE(VARINT(obj.nUndoSize));
 72          READWRITE(VARINT(obj.nHeightFirst));
 73          READWRITE(VARINT(obj.nHeightLast));
 74          READWRITE(VARINT(obj.nTimeFirst));
 75          READWRITE(VARINT(obj.nTimeLast));
 76      }
 77  
 78      CBlockFileInfo() = default;
 79  
 80      std::string ToString() const;
 81  
 82      /** update statistics (does not update nSize) */
 83      void AddBlock(unsigned int nHeightIn, uint64_t nTimeIn)
 84      {
 85          if (nBlocks == 0 || nHeightFirst > nHeightIn)
 86              nHeightFirst = nHeightIn;
 87          if (nBlocks == 0 || nTimeFirst > nTimeIn)
 88              nTimeFirst = nTimeIn;
 89          nBlocks++;
 90          if (nHeightIn > nHeightLast)
 91              nHeightLast = nHeightIn;
 92          if (nTimeIn > nTimeLast)
 93              nTimeLast = nTimeIn;
 94      }
 95  };
 96  
 97  /** Access to the block database (blocks/index/) */
 98  class BlockTreeDB : public CDBWrapper
 99  {
100  public:
101      using CDBWrapper::CDBWrapper;
102      void WriteBatchSync(const std::vector<std::pair<int, const CBlockFileInfo*>>& fileInfo, int nLastFile, const std::vector<const CBlockIndex*>& blockinfo);
103      bool ReadBlockFileInfo(int nFile, CBlockFileInfo& info);
104      bool ReadLastBlockFile(int& nFile);
105      void WriteReindexing(bool fReindexing);
106      void ReadReindexing(bool& fReindexing);
107      void WriteFlag(const std::string& name, bool fValue);
108      bool ReadFlag(const std::string& name, bool& fValue);
109      bool LoadBlockIndexGuts(const Consensus::Params& consensusParams, std::function<CBlockIndex*(const uint256&)> insertBlockIndex, const util::SignalInterrupt& interrupt)
110          EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
111  };
112  } // namespace kernel
113  
114  namespace node {
115  using kernel::CBlockFileInfo;
116  using kernel::BlockTreeDB;
117  
118  /** The pre-allocation chunk size for blk?????.dat files (since 0.8) */
119  static const unsigned int BLOCKFILE_CHUNK_SIZE = 0x1000000; // 16 MiB
120  /** The pre-allocation chunk size for rev?????.dat files (since 0.8) */
121  static const unsigned int UNDOFILE_CHUNK_SIZE = 0x100000; // 1 MiB
122  /** The maximum size of a blk?????.dat file (since 0.8) */
123  static const unsigned int MAX_BLOCKFILE_SIZE = 0x8000000; // 128 MiB
124  
125  /** Size of header written by WriteBlock before a serialized CBlock (8 bytes) */
126  static constexpr uint32_t STORAGE_HEADER_BYTES{std::tuple_size_v<MessageStartChars> + sizeof(unsigned int)};
127  
128  /** Total overhead when writing undo data: header (8 bytes) plus checksum (32 bytes) */
129  static constexpr uint32_t UNDO_DATA_DISK_OVERHEAD{STORAGE_HEADER_BYTES + uint256::size()};
130  
131  // Because validation code takes pointers to the map's CBlockIndex objects, if
132  // we ever switch to another associative container, we need to either use a
133  // container that has stable addressing (true of all std associative
134  // containers), or make the key a `std::unique_ptr<CBlockIndex>`
135  using BlockMap = std::unordered_map<uint256, CBlockIndex, BlockHasher>;
136  
137  struct CBlockIndexWorkComparator {
138      bool operator()(const CBlockIndex* pa, const CBlockIndex* pb) const;
139      using is_transparent = void;
140  };
141  
142  struct CBlockIndexHeightOnlyComparator {
143      /* Only compares the height of two block indices, doesn't try to tie-break */
144      bool operator()(const CBlockIndex* pa, const CBlockIndex* pb) const;
145  };
146  
147  struct PruneLockInfo {
148      int height_first{std::numeric_limits<int>::max()}; //! Height of earliest block that should be kept and not pruned
149  };
150  
151  enum BlockfileType {
152      // Values used as array indexes - do not change carelessly.
153      NORMAL = 0,
154      ASSUMED = 1,
155      NUM_TYPES = 2,
156  };
157  
158  std::ostream& operator<<(std::ostream& os, const BlockfileType& type);
159  
160  struct BlockfileCursor {
161      // The latest blockfile number.
162      int file_num{0};
163  
164      // Track the height of the highest block in file_num whose undo
165      // data has been written. Block data is written to block files in download
166      // order, but is written to undo files in validation order, which is
167      // usually in order by height. To avoid wasting disk space, undo files will
168      // be trimmed whenever the corresponding block file is finalized and
169      // the height of the highest block written to the block file equals the
170      // height of the highest block written to the undo file. This is a
171      // heuristic and can sometimes preemptively trim undo files that will write
172      // more data later, and sometimes fail to trim undo files that can't have
173      // more data written later.
174      int undo_height{0};
175  };
176  
177  std::ostream& operator<<(std::ostream& os, const BlockfileCursor& cursor);
178  
179  enum class ReadRawError {
180      IO,
181      BadPartRange,
182  };
183  
184  /**
185   * Maintains a tree of blocks (stored in `m_block_index`) which is consulted
186   * to determine where the most-work tip is.
187   *
188   * This data is used mostly in `Chainstate` - information about, e.g.,
189   * candidate tips is not maintained here.
190   */
191  class BlockManager
192  {
193      friend Chainstate;
194      friend ChainstateManager;
195  
196  private:
197      const CChainParams& GetParams() const { return m_opts.chainparams; }
198      const Consensus::Params& GetConsensus() const { return m_opts.chainparams.GetConsensus(); }
199      /**
200       * Load the blocktree off disk and into memory. Populate certain metadata
201       * per index entry (nStatus, nChainWork, nTimeMax, etc.) as well as peripheral
202       * collections like m_dirty_blockindex.
203       */
204      bool LoadBlockIndex(const std::optional<uint256>& snapshot_blockhash)
205          EXCLUSIVE_LOCKS_REQUIRED(cs_main);
206  
207      /** Return false if block file or undo file flushing fails. */
208      [[nodiscard]] bool FlushBlockFile(int blockfile_num, bool fFinalize, bool finalize_undo);
209  
210      /** Return false if undo file flushing fails. */
211      [[nodiscard]] bool FlushUndoFile(int block_file, bool finalize = false);
212  
213      /**
214       * Helper function performing various preparations before a block can be saved to disk:
215       * Returns the correct position for the block to be saved, which may be in the current or a new
216       * block file depending on nAddSize. May flush the previous blockfile to disk if full, updates
217       * blockfile info, and checks if there is enough disk space to save the block.
218       *
219       * The nAddSize argument passed to this function should include not just the size of the serialized CBlock, but also the size of
220       * separator fields (STORAGE_HEADER_BYTES).
221       */
222      [[nodiscard]] FlatFilePos FindNextBlockPos(unsigned int nAddSize, unsigned int nHeight, uint64_t nTime);
223      [[nodiscard]] bool FlushChainstateBlockFile(int tip_height);
224      bool FindUndoPos(BlockValidationState& state, int nFile, FlatFilePos& pos, unsigned int nAddSize);
225  
226      AutoFile OpenUndoFile(const FlatFilePos& pos, bool fReadOnly = false) const;
227  
228      /* Calculate the block/rev files to delete based on height specified by user with RPC command pruneblockchain */
229      void FindFilesToPruneManual(
230          std::set<int>& setFilesToPrune,
231          int nManualPruneHeight,
232          const Chainstate& chain);
233  
234      /**
235       * Prune block and undo files (blk???.dat and rev???.dat) so that the disk space used is less than a user-defined target.
236       * The user sets the target (in MB) on the command line or in config file.  This will be run on startup and whenever new
237       * space is allocated in a block or undo file, staying below the target. Changing back to unpruned requires a reindex
238       * (which in this case means the blockchain must be re-downloaded.)
239       *
240       * Pruning functions are called from FlushStateToDisk when the m_check_for_pruning flag has been set.
241       * Block and undo files are deleted in lock-step (when blk00003.dat is deleted, so is rev00003.dat.)
242       * Pruning cannot take place until the longest chain is at least a certain length (CChainParams::nPruneAfterHeight).
243       * Pruning will never delete a block within a defined distance (currently 288) from the active chain's tip.
244       * The block index is updated by unsetting HAVE_DATA and HAVE_UNDO for any blocks that were stored in the deleted files.
245       * A db flag records the fact that at least some block files have been pruned.
246       *
247       * @param[out]   setFilesToPrune   The set of file indices that can be unlinked will be returned
248       * @param        last_prune        The last height we're able to prune, according to the prune locks
249       */
250      void FindFilesToPrune(
251          std::set<int>& setFilesToPrune,
252          int last_prune,
253          const Chainstate& chain,
254          ChainstateManager& chainman);
255  
256      RecursiveMutex cs_LastBlockFile;
257  
258      //! Since assumedvalid chainstates may be syncing a range of the chain that is very
259      //! far away from the normal/background validation process, we should segment blockfiles
260      //! for assumed chainstates. Otherwise, we might have wildly different height ranges
261      //! mixed into the same block files, which would impair our ability to prune
262      //! effectively.
263      //!
264      //! This data structure maintains separate blockfile number cursors for each
265      //! BlockfileType. The ASSUMED state is initialized, when necessary, in FindNextBlockPos().
266      //!
267      //! The first element is the NORMAL cursor, second is ASSUMED.
268      std::array<std::optional<BlockfileCursor>, BlockfileType::NUM_TYPES>
269          m_blockfile_cursors GUARDED_BY(cs_LastBlockFile) = {
270              BlockfileCursor{},
271              std::nullopt,
272      };
273      int MaxBlockfileNum() const EXCLUSIVE_LOCKS_REQUIRED(cs_LastBlockFile)
274      {
275          static const BlockfileCursor empty_cursor;
276          const auto& normal = m_blockfile_cursors[BlockfileType::NORMAL].value_or(empty_cursor);
277          const auto& assumed = m_blockfile_cursors[BlockfileType::ASSUMED].value_or(empty_cursor);
278          return std::max(normal.file_num, assumed.file_num);
279      }
280  
281      /** Global flag to indicate we should check to see if there are
282       *  block/undo files that should be deleted.  Set on startup
283       *  or if we allocate more file space when we're in prune mode
284       */
285      bool m_check_for_pruning = false;
286  
287      const bool m_prune_mode;
288  
289      const Obfuscation m_obfuscation;
290  
291      /**
292       * Map from external index name to oldest block that must not be pruned.
293       *
294       * @note Internally, only blocks at height (height_first - PRUNE_LOCK_BUFFER - 1) and
295       * below will be pruned, but callers should avoid assuming any particular buffer size.
296       */
297      std::unordered_map<std::string, PruneLockInfo> m_prune_locks GUARDED_BY(::cs_main);
298  
299      BlockfileType BlockfileTypeForHeight(int height);
300  
301      const kernel::BlockManagerOpts m_opts;
302  
303      const FlatFileSeq m_block_file_seq;
304      const FlatFileSeq m_undo_file_seq;
305  
306  protected:
307      std::vector<CBlockFileInfo> m_blockfile_info;
308  
309      /** Dirty block index entries. */
310      std::set<CBlockIndex*> m_dirty_blockindex;
311  
312      /** Dirty block file entries. */
313      std::set<int> m_dirty_fileinfo;
314  
315  public:
316      using Options = kernel::BlockManagerOpts;
317      using ReadRawBlockResult = util::Expected<std::vector<std::byte>, ReadRawError>;
318  
319      explicit BlockManager(const util::SignalInterrupt& interrupt, Options opts);
320  
321      const util::SignalInterrupt& m_interrupt;
322      std::atomic<bool> m_importing{false};
323  
324      /**
325       * Whether all blockfiles have been added to the block tree database.
326       * Normally true, but set to false when a reindex is requested and the
327       * database is wiped. The value is persisted in the database across restarts
328       * and will be false until reindexing completes.
329       */
330      std::atomic_bool m_blockfiles_indexed{true};
331  
332      BlockMap m_block_index GUARDED_BY(cs_main);
333  
334      /**
335       * The height of the base block of an assumeutxo snapshot, if one is in use.
336       *
337       * This controls how blockfiles are segmented by chainstate type to avoid
338       * comingling different height regions of the chain when an assumedvalid chainstate
339       * is in use. If heights are drastically different in the same blockfile, pruning
340       * suffers.
341       *
342       * This is set during ActivateSnapshot() or upon LoadBlockIndex() if a snapshot
343       * had been previously loaded. After the snapshot is validated, this is unset to
344       * restore normal LoadBlockIndex behavior.
345       */
346      std::optional<int> m_snapshot_height;
347  
348      std::vector<CBlockIndex*> GetAllBlockIndices() EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
349  
350      /**
351       * All pairs A->B, where A (or one of its ancestors) misses transactions, but B has transactions.
352       * Pruned nodes may have entries where B is missing data.
353       */
354      std::multimap<CBlockIndex*, CBlockIndex*> m_blocks_unlinked;
355  
356      std::unique_ptr<BlockTreeDB> m_block_tree_db GUARDED_BY(::cs_main);
357  
358      void WriteBlockIndexDB() EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
359      bool LoadBlockIndexDB(const std::optional<uint256>& snapshot_blockhash)
360          EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
361  
362      /**
363       * Remove any pruned block & undo files that are still on disk.
364       * This could happen on some systems if the file was still being read while unlinked,
365       * or if we crash before unlinking.
366       */
367      void ScanAndUnlinkAlreadyPrunedFiles() EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
368  
369      CBlockIndex* AddToBlockIndex(const CBlockHeader& block, CBlockIndex*& best_header) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
370      /** Create a new block index entry for a given block hash */
371      CBlockIndex* InsertBlockIndex(const uint256& hash) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
372  
373      //! Mark one block file as pruned (modify associated database entries)
374      void PruneOneBlockFile(int fileNumber) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
375  
376      CBlockIndex* LookupBlockIndex(const uint256& hash) EXCLUSIVE_LOCKS_REQUIRED(cs_main);
377      const CBlockIndex* LookupBlockIndex(const uint256& hash) const EXCLUSIVE_LOCKS_REQUIRED(cs_main);
378  
379      /** Get block file info entry for one block file */
380      CBlockFileInfo* GetBlockFileInfo(size_t n);
381  
382      bool WriteBlockUndo(const CBlockUndo& blockundo, BlockValidationState& state, CBlockIndex& block)
383          EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
384  
385      /** Store block on disk and update block file statistics.
386       *
387       * @param[in]  block        the block to be stored
388       * @param[in]  nHeight      the height of the block
389       *
390       * @returns in case of success, the position to which the block was written to
391       *          in case of an error, an empty FlatFilePos
392       */
393      FlatFilePos WriteBlock(const CBlock& block, int nHeight);
394  
395      /** Update blockfile info while processing a block during reindex. The block must be available on disk.
396       *
397       * @param[in]  block        the block being processed
398       * @param[in]  nHeight      the height of the block
399       * @param[in]  pos          the position of the serialized CBlock on disk
400       */
401      void UpdateBlockInfo(const CBlock& block, unsigned int nHeight, const FlatFilePos& pos);
402  
403      /** Whether running in -prune mode. */
404      [[nodiscard]] bool IsPruneMode() const { return m_prune_mode; }
405  
406      /** Attempt to stay below this number of bytes of block files. */
407      [[nodiscard]] uint64_t GetPruneTarget() const { return m_opts.prune_target; }
408      static constexpr auto PRUNE_TARGET_MANUAL{std::numeric_limits<uint64_t>::max()};
409  
410      [[nodiscard]] bool LoadingBlocks() const { return m_importing || !m_blockfiles_indexed; }
411  
412      /** Calculate the amount of disk space the block & undo files currently use */
413      uint64_t CalculateCurrentUsage();
414  
415      //! Check if all blocks in the [upper_block, lower_block] range have data available as
416      //! defined by the status mask.
417      //! The caller is responsible for ensuring that lower_block is an ancestor of upper_block
418      //! (part of the same chain).
419      bool CheckBlockDataAvailability(const CBlockIndex& upper_block, const CBlockIndex& lower_block, BlockStatus block_status = BLOCK_HAVE_DATA) EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
420  
421      /**
422       * @brief Returns the earliest block with specified `status_mask` flags set after
423       * the latest block _not_ having those flags.
424       *
425       * This function starts from `upper_block`, which must have all
426       * `status_mask` flags set, and iterates backwards through its ancestors. It
427       * continues as long as each block has all `status_mask` flags set, until
428       * reaching the oldest ancestor or `lower_block`.
429       *
430       * @pre `upper_block` must have all `status_mask` flags set.
431       * @pre `lower_block` must be null or an ancestor of `upper_block`
432       *
433       * @param upper_block The starting block for the search, which must have all
434       *                    `status_mask` flags set.
435       * @param status_mask Bitmask specifying required status flags.
436       * @param lower_block The earliest possible block to return. If null, the
437       *                    search can extend to the genesis block.
438       *
439       * @return A reference to the earliest block between `upper_block`
440       *         and `lower_block`, inclusive, such that every block between the
441       *         returned block and `upper_block` has `status_mask` flags set.
442       */
443      const CBlockIndex& GetFirstBlock(
444          const CBlockIndex& upper_block LIFETIMEBOUND,
445          uint32_t status_mask,
446          const CBlockIndex* lower_block LIFETIMEBOUND = nullptr
447      ) const EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
448  
449      /** True if any block files have ever been pruned. */
450      bool m_have_pruned = false;
451  
452      //! Check whether the block associated with this index entry is pruned or not.
453      bool IsBlockPruned(const CBlockIndex& block) const EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
454  
455      //! Create or update a prune lock identified by its name
456      void UpdatePruneLock(const std::string& name, const PruneLockInfo& lock_info) EXCLUSIVE_LOCKS_REQUIRED(::cs_main);
457  
458      /** Open a block file (blk?????.dat) */
459      AutoFile OpenBlockFile(const FlatFilePos& pos, bool fReadOnly) const;
460  
461      /** Translation to a filesystem path */
462      fs::path GetBlockPosFilename(const FlatFilePos& pos) const;
463  
464      /**
465       *  Actually unlink the specified files
466       */
467      void UnlinkPrunedFiles(const std::set<int>& setFilesToPrune) const;
468  
469      /** Functions for disk access for blocks */
470      bool ReadBlock(CBlock& block, const FlatFilePos& pos, const std::optional<uint256>& expected_hash) const;
471      bool ReadBlock(CBlock& block, const CBlockIndex& index) const;
472      ReadRawBlockResult ReadRawBlock(const FlatFilePos& pos, std::optional<std::pair<size_t, size_t>> block_part = std::nullopt) const;
473  
474      bool ReadBlockUndo(CBlockUndo& blockundo, const CBlockIndex& index) const;
475  
476      void CleanupBlockRevFiles() const;
477  };
478  
479  // Calls ActivateBestChain() even if no blocks are imported.
480  void ImportBlocks(ChainstateManager& chainman, std::span<const fs::path> import_paths);
481  } // namespace node
482  
483  #endif // BITCOIN_NODE_BLOCKSTORAGE_H