/ src / index / base.cpp
base.cpp
  1  // Copyright (c) 2017-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  #include <index/base.h>
  6  
  7  #include <chain.h>
  8  #include <common/args.h>
  9  #include <dbwrapper.h>
 10  #include <interfaces/chain.h>
 11  #include <interfaces/types.h>
 12  #include <kernel/chain.h>
 13  #include <logging.h>
 14  #include <node/abort.h>
 15  #include <node/blockstorage.h>
 16  #include <node/context.h>
 17  #include <node/database_args.h>
 18  #include <node/interface_ui.h>
 19  #include <primitives/block.h>
 20  #include <sync.h>
 21  #include <tinyformat.h>
 22  #include <uint256.h>
 23  #include <undo.h>
 24  #include <util/fs.h>
 25  #include <util/string.h>
 26  #include <util/thread.h>
 27  #include <util/threadinterrupt.h>
 28  #include <util/time.h>
 29  #include <util/translation.h>
 30  #include <validation.h>
 31  #include <validationinterface.h>
 32  
 33  #include <cassert>
 34  #include <compare>
 35  #include <cstdint>
 36  #include <memory>
 37  #include <optional>
 38  #include <span>
 39  #include <stdexcept>
 40  #include <string>
 41  #include <thread>
 42  #include <utility>
 43  #include <vector>
 44  
 45  constexpr uint8_t DB_BEST_BLOCK{'B'};
 46  
 47  constexpr auto SYNC_LOG_INTERVAL{30s};
 48  constexpr auto SYNC_LOCATOR_WRITE_INTERVAL{30s};
 49  
 50  template <typename... Args>
 51  void BaseIndex::FatalErrorf(util::ConstevalFormatString<sizeof...(Args)> fmt, const Args&... args)
 52  {
 53      auto message = tfm::format(fmt, args...);
 54      node::AbortNode(m_chain->context()->shutdown_request, m_chain->context()->exit_status, Untranslated(message), m_chain->context()->warnings.get());
 55  }
 56  
 57  CBlockLocator GetLocator(interfaces::Chain& chain, const uint256& block_hash)
 58  {
 59      CBlockLocator locator;
 60      bool found = chain.findBlock(block_hash, interfaces::FoundBlock().locator(locator));
 61      assert(found);
 62      assert(!locator.IsNull());
 63      return locator;
 64  }
 65  
 66  BaseIndex::DB::DB(const fs::path& path, size_t n_cache_size, bool f_memory, bool f_wipe, bool f_obfuscate) :
 67      CDBWrapper{DBParams{
 68          .path = path,
 69          .cache_bytes = n_cache_size,
 70          .memory_only = f_memory,
 71          .wipe_data = f_wipe,
 72          .obfuscate = f_obfuscate,
 73          .options = [] { DBOptions options; node::ReadDatabaseArgs(gArgs, options); return options; }()}}
 74  {}
 75  
 76  CBlockLocator BaseIndex::DB::ReadBestBlock() const
 77  {
 78      CBlockLocator locator;
 79  
 80      bool success = Read(DB_BEST_BLOCK, locator);
 81      if (!success) {
 82          locator.SetNull();
 83      }
 84  
 85      return locator;
 86  }
 87  
 88  void BaseIndex::DB::WriteBestBlock(CDBBatch& batch, const CBlockLocator& locator)
 89  {
 90      batch.Write(DB_BEST_BLOCK, locator);
 91  }
 92  
 93  BaseIndex::BaseIndex(std::unique_ptr<interfaces::Chain> chain, std::string name)
 94      : m_chain{std::move(chain)}, m_name{std::move(name)} {}
 95  
 96  BaseIndex::~BaseIndex()
 97  {
 98      Interrupt();
 99      Stop();
100  }
101  
102  bool BaseIndex::Init()
103  {
104      AssertLockNotHeld(cs_main);
105  
106      // May need reset if index is being restarted.
107      m_interrupt.reset();
108  
109      // m_chainstate member gives indexing code access to node internals. It is
110      // removed in followup https://github.com/bitcoin/bitcoin/pull/24230
111      m_chainstate = WITH_LOCK(::cs_main,
112          return &m_chain->context()->chainman->GetChainstateForIndexing());
113      // Register to validation interface before setting the 'm_synced' flag, so that
114      // callbacks are not missed once m_synced is true.
115      m_chain->context()->validation_signals->RegisterValidationInterface(this);
116  
117      const auto locator{GetDB().ReadBestBlock()};
118  
119      LOCK(cs_main);
120      CChain& index_chain = m_chainstate->m_chain;
121  
122      if (locator.IsNull()) {
123          SetBestBlockIndex(nullptr);
124      } else {
125          // Setting the best block to the locator's top block. If it is not part of the
126          // best chain, we will rewind to the fork point during index sync
127          const CBlockIndex* locator_index{m_chainstate->m_blockman.LookupBlockIndex(locator.vHave.at(0))};
128          if (!locator_index) {
129              return InitError(Untranslated(strprintf("best block of %s not found. Please rebuild the index.", GetName())));
130          }
131          SetBestBlockIndex(locator_index);
132      }
133  
134      // Child init
135      const CBlockIndex* start_block = m_best_block_index.load();
136      if (!CustomInit(start_block ? std::make_optional(interfaces::BlockRef{start_block->GetBlockHash(), start_block->nHeight}) : std::nullopt)) {
137          return false;
138      }
139  
140      // Note: this will latch to true immediately if the user starts up with an empty
141      // datadir and an index enabled. If this is the case, indexation will happen solely
142      // via `BlockConnected` signals until, possibly, the next restart.
143      m_synced = start_block == index_chain.Tip();
144      m_init = true;
145      return true;
146  }
147  
148  static const CBlockIndex* NextSyncBlock(const CBlockIndex* pindex_prev, CChain& chain) EXCLUSIVE_LOCKS_REQUIRED(cs_main)
149  {
150      AssertLockHeld(cs_main);
151  
152      if (!pindex_prev) {
153          return chain.Genesis();
154      }
155  
156      const CBlockIndex* pindex = chain.Next(pindex_prev);
157      if (pindex) {
158          return pindex;
159      }
160  
161      // Since block is not in the chain, return the next block in the chain AFTER the last common ancestor.
162      // Caller will be responsible for rewinding back to the common ancestor.
163      return chain.Next(chain.FindFork(pindex_prev));
164  }
165  
166  bool BaseIndex::ProcessBlock(const CBlockIndex* pindex, const CBlock* block_data)
167  {
168      interfaces::BlockInfo block_info = kernel::MakeBlockInfo(pindex, block_data);
169  
170      CBlock block;
171      if (!block_data) { // disk lookup if block data wasn't provided
172          if (!m_chainstate->m_blockman.ReadBlock(block, *pindex)) {
173              FatalErrorf("Failed to read block %s from disk",
174                          pindex->GetBlockHash().ToString());
175              return false;
176          }
177          block_info.data = &block;
178      }
179  
180      CBlockUndo block_undo;
181      if (CustomOptions().connect_undo_data) {
182          if (pindex->nHeight > 0 && !m_chainstate->m_blockman.ReadBlockUndo(block_undo, *pindex)) {
183              FatalErrorf("Failed to read undo block data %s from disk",
184                          pindex->GetBlockHash().ToString());
185              return false;
186          }
187          block_info.undo_data = &block_undo;
188      }
189  
190      if (!CustomAppend(block_info)) {
191          FatalErrorf("Failed to write block %s to index database",
192                      pindex->GetBlockHash().ToString());
193          return false;
194      }
195  
196      return true;
197  }
198  
199  void BaseIndex::Sync()
200  {
201      const CBlockIndex* pindex = m_best_block_index.load();
202      if (!m_synced) {
203          auto last_log_time{NodeClock::now()};
204          auto last_locator_write_time{last_log_time};
205          while (true) {
206              if (m_interrupt) {
207                  LogInfo("%s: m_interrupt set; exiting ThreadSync", GetName());
208  
209                  SetBestBlockIndex(pindex);
210                  // No need to handle errors in Commit. If it fails, the error will be already be
211                  // logged. The best way to recover is to continue, as index cannot be corrupted by
212                  // a missed commit to disk for an advanced index state.
213                  Commit();
214                  return;
215              }
216  
217              const CBlockIndex* pindex_next = WITH_LOCK(cs_main, return NextSyncBlock(pindex, m_chainstate->m_chain));
218              // If pindex_next is null, it means pindex is the chain tip, so
219              // commit data indexed so far.
220              if (!pindex_next) {
221                  SetBestBlockIndex(pindex);
222                  // No need to handle errors in Commit. See rationale above.
223                  Commit();
224  
225                  // If pindex is still the chain tip after committing, exit the
226                  // sync loop. It is important for cs_main to be locked while
227                  // setting m_synced = true, otherwise a new block could be
228                  // attached while m_synced is still false, and it would not be
229                  // indexed.
230                  LOCK(::cs_main);
231                  pindex_next = NextSyncBlock(pindex, m_chainstate->m_chain);
232                  if (!pindex_next) {
233                      m_synced = true;
234                      break;
235                  }
236              }
237              if (pindex_next->pprev != pindex && !Rewind(pindex, pindex_next->pprev)) {
238                  FatalErrorf("Failed to rewind %s to a previous chain tip", GetName());
239                  return;
240              }
241              pindex = pindex_next;
242  
243  
244              if (!ProcessBlock(pindex)) return; // error logged internally
245  
246              auto current_time{NodeClock::now()};
247              if (current_time - last_log_time >= SYNC_LOG_INTERVAL) {
248                  LogInfo("Syncing %s with block chain from height %d", GetName(), pindex->nHeight);
249                  last_log_time = current_time;
250              }
251  
252              if (current_time - last_locator_write_time >= SYNC_LOCATOR_WRITE_INTERVAL) {
253                  SetBestBlockIndex(pindex);
254                  last_locator_write_time = current_time;
255                  // No need to handle errors in Commit. See rationale above.
256                  Commit();
257              }
258          }
259      }
260  
261      if (pindex) {
262          LogInfo("%s is enabled at height %d", GetName(), pindex->nHeight);
263      } else {
264          LogInfo("%s is enabled", GetName());
265      }
266  }
267  
268  bool BaseIndex::Commit()
269  {
270      // Don't commit anything if we haven't indexed any block yet
271      // (this could happen if init is interrupted).
272      bool ok = m_best_block_index != nullptr;
273      if (ok) {
274          CDBBatch batch(GetDB());
275          ok = CustomCommit(batch);
276          if (ok) {
277              GetDB().WriteBestBlock(batch, GetLocator(*m_chain, m_best_block_index.load()->GetBlockHash()));
278              GetDB().WriteBatch(batch);
279          }
280      }
281      if (!ok) {
282          LogError("Failed to commit latest %s state", GetName());
283          return false;
284      }
285      return true;
286  }
287  
288  bool BaseIndex::Rewind(const CBlockIndex* current_tip, const CBlockIndex* new_tip)
289  {
290      assert(current_tip->GetAncestor(new_tip->nHeight) == new_tip);
291  
292      CBlock block;
293      CBlockUndo block_undo;
294  
295      for (const CBlockIndex* iter_tip = current_tip; iter_tip != new_tip; iter_tip = iter_tip->pprev) {
296          interfaces::BlockInfo block_info = kernel::MakeBlockInfo(iter_tip);
297          if (CustomOptions().disconnect_data) {
298              if (!m_chainstate->m_blockman.ReadBlock(block, *iter_tip)) {
299                  LogError("Failed to read block %s from disk",
300                           iter_tip->GetBlockHash().ToString());
301                  return false;
302              }
303              block_info.data = &block;
304          }
305          if (CustomOptions().disconnect_undo_data && iter_tip->nHeight > 0) {
306              if (!m_chainstate->m_blockman.ReadBlockUndo(block_undo, *iter_tip)) {
307                  return false;
308              }
309              block_info.undo_data = &block_undo;
310          }
311          if (!CustomRemove(block_info)) {
312              return false;
313          }
314      }
315  
316      // Don't commit here - the committed index state must never be ahead of the
317      // flushed chainstate, otherwise unclean restarts would lead to index corruption.
318      // Pruning has a minimum of 288 blocks-to-keep and getting the index
319      // out of sync may be possible but a users fault.
320      // In case we reorg beyond the pruned depth, ReadBlock would
321      // throw and lead to a graceful shutdown
322      SetBestBlockIndex(new_tip);
323      return true;
324  }
325  
326  void BaseIndex::BlockConnected(ChainstateRole role, const std::shared_ptr<const CBlock>& block, const CBlockIndex* pindex)
327  {
328      // Ignore events from the assumed-valid chain; we will process its blocks
329      // (sequentially) after it is fully verified by the background chainstate. This
330      // is to avoid any out-of-order indexing.
331      //
332      // TODO at some point we could parameterize whether a particular index can be
333      // built out of order, but for now just do the conservative simple thing.
334      if (role == ChainstateRole::ASSUMEDVALID) {
335          return;
336      }
337  
338      // Ignore BlockConnected signals until we have fully indexed the chain.
339      if (!m_synced) {
340          return;
341      }
342  
343      const CBlockIndex* best_block_index = m_best_block_index.load();
344      if (!best_block_index) {
345          if (pindex->nHeight != 0) {
346              FatalErrorf("First block connected is not the genesis block (height=%d)",
347                         pindex->nHeight);
348              return;
349          }
350      } else {
351          // Ensure block connects to an ancestor of the current best block. This should be the case
352          // most of the time, but may not be immediately after the sync thread catches up and sets
353          // m_synced. Consider the case where there is a reorg and the blocks on the stale branch are
354          // in the ValidationInterface queue backlog even after the sync thread has caught up to the
355          // new chain tip. In this unlikely event, log a warning and let the queue clear.
356          if (best_block_index->GetAncestor(pindex->nHeight - 1) != pindex->pprev) {
357              LogWarning("Block %s does not connect to an ancestor of "
358                        "known best chain (tip=%s); not updating index",
359                        pindex->GetBlockHash().ToString(),
360                        best_block_index->GetBlockHash().ToString());
361              return;
362          }
363          if (best_block_index != pindex->pprev && !Rewind(best_block_index, pindex->pprev)) {
364              FatalErrorf("Failed to rewind %s to a previous chain tip",
365                         GetName());
366              return;
367          }
368      }
369  
370      // Dispatch block to child class; errors are logged internally and abort the node.
371      if (ProcessBlock(pindex, block.get())) {
372          // Setting the best block index is intentionally the last step of this
373          // function, so BlockUntilSyncedToCurrentChain callers waiting for the
374          // best block index to be updated can rely on the block being fully
375          // processed, and the index object being safe to delete.
376          SetBestBlockIndex(pindex);
377      }
378  }
379  
380  void BaseIndex::ChainStateFlushed(ChainstateRole role, const CBlockLocator& locator)
381  {
382      // Ignore events from the assumed-valid chain; we will process its blocks
383      // (sequentially) after it is fully verified by the background chainstate.
384      if (role == ChainstateRole::ASSUMEDVALID) {
385          return;
386      }
387  
388      if (!m_synced) {
389          return;
390      }
391  
392      const uint256& locator_tip_hash = locator.vHave.front();
393      const CBlockIndex* locator_tip_index;
394      {
395          LOCK(cs_main);
396          locator_tip_index = m_chainstate->m_blockman.LookupBlockIndex(locator_tip_hash);
397      }
398  
399      if (!locator_tip_index) {
400          FatalErrorf("First block (hash=%s) in locator was not found",
401                     locator_tip_hash.ToString());
402          return;
403      }
404  
405      // This checks that ChainStateFlushed callbacks are received after BlockConnected. The check may fail
406      // immediately after the sync thread catches up and sets m_synced. Consider the case where
407      // there is a reorg and the blocks on the stale branch are in the ValidationInterface queue
408      // backlog even after the sync thread has caught up to the new chain tip. In this unlikely
409      // event, log a warning and let the queue clear.
410      const CBlockIndex* best_block_index = m_best_block_index.load();
411      if (best_block_index->GetAncestor(locator_tip_index->nHeight) != locator_tip_index) {
412          LogWarning("Locator contains block (hash=%s) not on known best "
413                    "chain (tip=%s); not writing index locator",
414                    locator_tip_hash.ToString(),
415                    best_block_index->GetBlockHash().ToString());
416          return;
417      }
418  
419      // No need to handle errors in Commit. If it fails, the error will be already be logged. The
420      // best way to recover is to continue, as index cannot be corrupted by a missed commit to disk
421      // for an advanced index state.
422      Commit();
423  }
424  
425  bool BaseIndex::BlockUntilSyncedToCurrentChain() const
426  {
427      AssertLockNotHeld(cs_main);
428  
429      if (!m_synced) {
430          return false;
431      }
432  
433      {
434          // Skip the queue-draining stuff if we know we're caught up with
435          // m_chain.Tip().
436          LOCK(cs_main);
437          const CBlockIndex* chain_tip = m_chainstate->m_chain.Tip();
438          const CBlockIndex* best_block_index = m_best_block_index.load();
439          if (best_block_index->GetAncestor(chain_tip->nHeight) == chain_tip) {
440              return true;
441          }
442      }
443  
444      LogInfo("%s is catching up on block notifications", GetName());
445      m_chain->context()->validation_signals->SyncWithValidationInterfaceQueue();
446      return true;
447  }
448  
449  void BaseIndex::Interrupt()
450  {
451      m_interrupt();
452  }
453  
454  bool BaseIndex::StartBackgroundSync()
455  {
456      if (!m_init) throw std::logic_error("Error: Cannot start a non-initialized index");
457  
458      m_thread_sync = std::thread(&util::TraceThread, GetName(), [this] { Sync(); });
459      return true;
460  }
461  
462  void BaseIndex::Stop()
463  {
464      if (m_chain->context()->validation_signals) {
465          m_chain->context()->validation_signals->UnregisterValidationInterface(this);
466      }
467  
468      if (m_thread_sync.joinable()) {
469          m_thread_sync.join();
470      }
471  }
472  
473  IndexSummary BaseIndex::GetSummary() const
474  {
475      IndexSummary summary{};
476      summary.name = GetName();
477      summary.synced = m_synced;
478      if (const auto& pindex = m_best_block_index.load()) {
479          summary.best_block_height = pindex->nHeight;
480          summary.best_block_hash = pindex->GetBlockHash();
481      } else {
482          summary.best_block_height = 0;
483          summary.best_block_hash = m_chain->getBlockHash(0);
484      }
485      return summary;
486  }
487  
488  void BaseIndex::SetBestBlockIndex(const CBlockIndex* block)
489  {
490      assert(!m_chainstate->m_blockman.IsPruneMode() || AllowPrune());
491  
492      if (AllowPrune() && block) {
493          node::PruneLockInfo prune_lock;
494          prune_lock.height_first = block->nHeight;
495          WITH_LOCK(::cs_main, m_chainstate->m_blockman.UpdatePruneLock(GetName(), prune_lock));
496      }
497  
498      // Intentionally set m_best_block_index as the last step in this function,
499      // after updating prune locks above, and after making any other references
500      // to *this, so the BlockUntilSyncedToCurrentChain function (which checks
501      // m_best_block_index as an optimization) can be used to wait for the last
502      // BlockConnected notification and safely assume that prune locks are
503      // updated and that the index object is safe to delete.
504      m_best_block_index = block;
505  }