/ 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/types.h>
 13  #include <node/abort.h>
 14  #include <node/blockstorage.h>
 15  #include <node/context.h>
 16  #include <node/database_args.h>
 17  #include <node/interface_ui.h>
 18  #include <primitives/block.h>
 19  #include <sync.h>
 20  #include <tinyformat.h>
 21  #include <uint256.h>
 22  #include <undo.h>
 23  #include <util/check.h>
 24  #include <util/fs.h>
 25  #include <util/log.h>
 26  #include <util/string.h>
 27  #include <util/thread.h>
 28  #include <util/threadinterrupt.h>
 29  #include <util/time.h>
 30  #include <util/translation.h>
 31  #include <validation.h>
 32  #include <validationinterface.h>
 33  
 34  #include <compare>
 35  #include <cstdint>
 36  #include <functional>
 37  #include <memory>
 38  #include <optional>
 39  #include <stdexcept>
 40  #include <string>
 41  #include <thread>
 42  #include <utility>
 43  #include <vector>
 44  
 45  using kernel::ChainstateRole;
 46  
 47  constexpr uint8_t DB_BEST_BLOCK{'B'};
 48  
 49  constexpr auto SYNC_LOG_INTERVAL{30s};
 50  constexpr auto SYNC_LOCATOR_WRITE_INTERVAL{30s};
 51  
 52  template <typename... Args>
 53  void BaseIndex::FatalErrorf(util::ConstevalFormatString<sizeof...(Args)> fmt, const Args&... args)
 54  {
 55      auto message = tfm::format(fmt, args...);
 56      node::AbortNode(m_chain->context()->shutdown_request, m_chain->context()->exit_status, Untranslated(message), m_chain->context()->warnings.get());
 57  }
 58  
 59  CBlockLocator GetLocator(interfaces::Chain& chain, const uint256& block_hash)
 60  {
 61      CBlockLocator locator;
 62      bool found = chain.findBlock(block_hash, interfaces::FoundBlock().locator(locator));
 63      assert(found);
 64      assert(!locator.IsNull());
 65      return locator;
 66  }
 67  
 68  BaseIndex::DB::DB(const fs::path& path, size_t n_cache_size, bool f_memory, bool f_wipe, bool f_obfuscate) :
 69      CDBWrapper{DBParams{
 70          .path = path,
 71          .cache_bytes = n_cache_size,
 72          .memory_only = f_memory,
 73          .wipe_data = f_wipe,
 74          .obfuscate = f_obfuscate,
 75          .options = [] { DBOptions options; node::ReadDatabaseArgs(gArgs, options); return options; }()}}
 76  {}
 77  
 78  CBlockLocator BaseIndex::DB::ReadBestBlock() const
 79  {
 80      CBlockLocator locator;
 81  
 82      bool success = Read(DB_BEST_BLOCK, locator);
 83      if (!success) {
 84          locator.SetNull();
 85      }
 86  
 87      return locator;
 88  }
 89  
 90  void BaseIndex::DB::WriteBestBlock(CDBBatch& batch, const CBlockLocator& locator)
 91  {
 92      batch.Write(DB_BEST_BLOCK, locator);
 93  }
 94  
 95  BaseIndex::BaseIndex(std::unique_ptr<interfaces::Chain> chain, std::string name)
 96      : m_chain{std::move(chain)}, m_name{std::move(name)} {}
 97  
 98  BaseIndex::~BaseIndex()
 99  {
100      Interrupt();
101      Stop();
102  }
103  
104  bool BaseIndex::Init()
105  {
106      AssertLockNotHeld(cs_main);
107  
108      // May need reset if index is being restarted.
109      m_interrupt.reset();
110  
111      // m_chainstate member gives indexing code access to node internals. It is
112      // removed in followup https://github.com/bitcoin/bitcoin/pull/24230
113      m_chainstate = WITH_LOCK(::cs_main,
114                               return &m_chain->context()->chainman->ValidatedChainstate());
115      // Register to validation interface before setting the 'm_synced' flag, so that
116      // callbacks are not missed once m_synced is true.
117      m_chain->context()->validation_signals->RegisterValidationInterface(this);
118  
119      const auto locator{GetDB().ReadBestBlock()};
120  
121      LOCK(cs_main);
122      CChain& index_chain = m_chainstate->m_chain;
123  
124      if (locator.IsNull()) {
125          SetBestBlockIndex(nullptr);
126      } else {
127          // Setting the best block to the locator's top block. If it is not part of the
128          // best chain, we will rewind to the fork point during index sync
129          const CBlockIndex* locator_index{m_chainstate->m_blockman.LookupBlockIndex(locator.vHave.at(0))};
130          if (!locator_index) {
131              return InitError(Untranslated(strprintf("best block of %s not found. Please rebuild the index.", GetName())));
132          }
133          SetBestBlockIndex(locator_index);
134      }
135  
136      // Child init
137      const CBlockIndex* start_block = m_best_block_index.load();
138      if (!CustomInit(start_block ? std::make_optional(interfaces::BlockRef{start_block->GetBlockHash(), start_block->nHeight}) : std::nullopt)) {
139          return false;
140      }
141  
142      // Note: this will latch to true immediately if the user starts up with an empty
143      // datadir and an index enabled. If this is the case, indexation will happen solely
144      // via `BlockConnected` signals until, possibly, the next restart.
145      m_synced = start_block == index_chain.Tip();
146      m_init = true;
147      return true;
148  }
149  
150  static const CBlockIndex* NextSyncBlock(const CBlockIndex* const pindex_prev, CChain& chain) EXCLUSIVE_LOCKS_REQUIRED(cs_main)
151  {
152      AssertLockHeld(cs_main);
153  
154      if (!pindex_prev) {
155          return chain.Genesis();
156      }
157  
158      if (const auto* pindex{chain.Next(*pindex_prev)}) {
159          return pindex;
160      }
161  
162      // If there is no next block, we might be synced
163      if (pindex_prev == chain.Tip()) {
164          return nullptr;
165      }
166  
167      // Since block is not in the chain, return the next block in the chain AFTER the last common ancestor.
168      // Caller will be responsible for rewinding back to the common ancestor.
169      const auto* fork{chain.FindFork(*pindex_prev)};
170      // Common ancestor must exist (genesis).
171      return chain.Next(*Assert(fork));
172  }
173  
174  bool BaseIndex::ProcessBlock(const CBlockIndex* pindex, const CBlock* block_data)
175  {
176      interfaces::BlockInfo block_info = kernel::MakeBlockInfo(pindex, block_data);
177  
178      CBlock block;
179      if (!block_data) { // disk lookup if block data wasn't provided
180          if (!m_chainstate->m_blockman.ReadBlock(block, *pindex)) {
181              FatalErrorf("Failed to read block %s from disk",
182                          pindex->GetBlockHash().ToString());
183              return false;
184          }
185          block_info.data = &block;
186      }
187  
188      CBlockUndo block_undo;
189      if (CustomOptions().connect_undo_data) {
190          if (pindex->nHeight > 0 && !m_chainstate->m_blockman.ReadBlockUndo(block_undo, *pindex)) {
191              FatalErrorf("Failed to read undo block data %s from disk",
192                          pindex->GetBlockHash().ToString());
193              return false;
194          }
195          block_info.undo_data = &block_undo;
196      }
197  
198      if (!CustomAppend(block_info)) {
199          FatalErrorf("Failed to write block %s to index database",
200                      pindex->GetBlockHash().ToString());
201          return false;
202      }
203  
204      return true;
205  }
206  
207  void BaseIndex::Sync()
208  {
209      const CBlockIndex* pindex = m_best_block_index.load();
210      if (!m_synced) {
211          auto last_log_time{NodeClock::now()};
212          auto last_locator_write_time{last_log_time};
213          while (true) {
214              if (m_interrupt) {
215                  LogInfo("%s: m_interrupt set; exiting ThreadSync", GetName());
216  
217                  SetBestBlockIndex(pindex);
218                  // No need to handle errors in Commit. If it fails, the error will be already be
219                  // logged. The best way to recover is to continue, as index cannot be corrupted by
220                  // a missed commit to disk for an advanced index state.
221                  Commit();
222                  return;
223              }
224  
225              const CBlockIndex* pindex_next = WITH_LOCK(cs_main, return NextSyncBlock(pindex, m_chainstate->m_chain));
226              // If pindex_next is null, it means pindex is the chain tip, so
227              // commit data indexed so far.
228              if (!pindex_next) {
229                  SetBestBlockIndex(pindex);
230                  // No need to handle errors in Commit. See rationale above.
231                  Commit();
232  
233                  // If pindex is still the chain tip after committing, exit the
234                  // sync loop. It is important for cs_main to be locked while
235                  // setting m_synced = true, otherwise a new block could be
236                  // attached while m_synced is still false, and it would not be
237                  // indexed.
238                  LOCK(::cs_main);
239                  pindex_next = NextSyncBlock(pindex, m_chainstate->m_chain);
240                  if (!pindex_next) {
241                      m_synced = true;
242                      break;
243                  }
244              }
245              if (pindex_next->pprev != pindex && !Rewind(pindex, pindex_next->pprev)) {
246                  FatalErrorf("Failed to rewind %s to a previous chain tip", GetName());
247                  return;
248              }
249              pindex = pindex_next;
250  
251  
252              if (!ProcessBlock(pindex)) return; // error logged internally
253  
254              auto current_time{NodeClock::now()};
255              if (current_time - last_log_time >= SYNC_LOG_INTERVAL) {
256                  LogInfo("Syncing %s with block chain from height %d", GetName(), pindex->nHeight);
257                  last_log_time = current_time;
258              }
259  
260              if (current_time - last_locator_write_time >= SYNC_LOCATOR_WRITE_INTERVAL) {
261                  SetBestBlockIndex(pindex);
262                  last_locator_write_time = current_time;
263                  // No need to handle errors in Commit. See rationale above.
264                  Commit();
265              }
266          }
267      }
268  
269      if (pindex) {
270          LogInfo("%s is enabled at height %d", GetName(), pindex->nHeight);
271      } else {
272          LogInfo("%s is enabled", GetName());
273      }
274  }
275  
276  bool BaseIndex::Commit()
277  {
278      // Don't commit anything if we haven't indexed any block yet
279      // (this could happen if init is interrupted).
280      bool ok = m_best_block_index != nullptr;
281      if (ok) {
282          CDBBatch batch(GetDB());
283          ok = CustomCommit(batch);
284          if (ok) {
285              GetDB().WriteBestBlock(batch, GetLocator(*m_chain, m_best_block_index.load()->GetBlockHash()));
286              GetDB().WriteBatch(batch);
287          }
288      }
289      if (!ok) {
290          LogError("Failed to commit latest %s state", GetName());
291          return false;
292      }
293      return true;
294  }
295  
296  bool BaseIndex::Rewind(const CBlockIndex* current_tip, const CBlockIndex* new_tip)
297  {
298      assert(current_tip->GetAncestor(new_tip->nHeight) == new_tip);
299  
300      CBlock block;
301      CBlockUndo block_undo;
302  
303      for (const CBlockIndex* iter_tip = current_tip; iter_tip != new_tip; iter_tip = iter_tip->pprev) {
304          interfaces::BlockInfo block_info = kernel::MakeBlockInfo(iter_tip);
305          if (CustomOptions().disconnect_data) {
306              if (!m_chainstate->m_blockman.ReadBlock(block, *iter_tip)) {
307                  LogError("Failed to read block %s from disk",
308                           iter_tip->GetBlockHash().ToString());
309                  return false;
310              }
311              block_info.data = &block;
312          }
313          if (CustomOptions().disconnect_undo_data && iter_tip->nHeight > 0) {
314              if (!m_chainstate->m_blockman.ReadBlockUndo(block_undo, *iter_tip)) {
315                  return false;
316              }
317              block_info.undo_data = &block_undo;
318          }
319          if (!CustomRemove(block_info)) {
320              return false;
321          }
322      }
323  
324      // Don't commit here - the committed index state must never be ahead of the
325      // flushed chainstate, otherwise unclean restarts would lead to index corruption.
326      // Pruning has a minimum of 288 blocks-to-keep and getting the index
327      // out of sync may be possible but a users fault.
328      // In case we reorg beyond the pruned depth, ReadBlock would
329      // throw and lead to a graceful shutdown
330      SetBestBlockIndex(new_tip);
331      return true;
332  }
333  
334  void BaseIndex::BlockConnected(const ChainstateRole& role, const std::shared_ptr<const CBlock>& block, const CBlockIndex* pindex)
335  {
336      // Ignore events from not fully validated chains to avoid out-of-order indexing.
337      //
338      // TODO at some point we could parameterize whether a particular index can be
339      // built out of order, but for now just do the conservative simple thing.
340      if (!role.validated) {
341          return;
342      }
343  
344      // Ignore BlockConnected signals until we have fully indexed the chain.
345      if (!m_synced) {
346          return;
347      }
348  
349      const CBlockIndex* best_block_index = m_best_block_index.load();
350      if (!best_block_index) {
351          if (pindex->nHeight != 0) {
352              FatalErrorf("First block connected is not the genesis block (height=%d)",
353                         pindex->nHeight);
354              return;
355          }
356      } else {
357          // Ensure block connects to an ancestor of the current best block. This should be the case
358          // most of the time, but may not be immediately after the sync thread catches up and sets
359          // m_synced. Consider the case where there is a reorg and the blocks on the stale branch are
360          // in the ValidationInterface queue backlog even after the sync thread has caught up to the
361          // new chain tip. In this unlikely event, log a warning and let the queue clear.
362          if (best_block_index->GetAncestor(pindex->nHeight - 1) != pindex->pprev) {
363              LogWarning("Block %s does not connect to an ancestor of "
364                        "known best chain (tip=%s); not updating index",
365                        pindex->GetBlockHash().ToString(),
366                        best_block_index->GetBlockHash().ToString());
367              return;
368          }
369          if (best_block_index != pindex->pprev && !Rewind(best_block_index, pindex->pprev)) {
370              FatalErrorf("Failed to rewind %s to a previous chain tip",
371                         GetName());
372              return;
373          }
374      }
375  
376      // Dispatch block to child class; errors are logged internally and abort the node.
377      if (ProcessBlock(pindex, block.get())) {
378          // Setting the best block index is intentionally the last step of this
379          // function, so BlockUntilSyncedToCurrentChain callers waiting for the
380          // best block index to be updated can rely on the block being fully
381          // processed, and the index object being safe to delete.
382          SetBestBlockIndex(pindex);
383      }
384  }
385  
386  void BaseIndex::ChainStateFlushed(const ChainstateRole& role, const CBlockLocator& locator)
387  {
388      // Ignore events from not fully validated chains to avoid out-of-order indexing.
389      if (!role.validated) {
390          return;
391      }
392  
393      if (!m_synced) {
394          return;
395      }
396  
397      const uint256& locator_tip_hash = locator.vHave.front();
398      const CBlockIndex* locator_tip_index;
399      {
400          LOCK(cs_main);
401          locator_tip_index = m_chainstate->m_blockman.LookupBlockIndex(locator_tip_hash);
402      }
403  
404      if (!locator_tip_index) {
405          FatalErrorf("First block (hash=%s) in locator was not found",
406                     locator_tip_hash.ToString());
407          return;
408      }
409  
410      // This checks that ChainStateFlushed callbacks are received after BlockConnected. The check may fail
411      // immediately after the sync thread catches up and sets m_synced. Consider the case where
412      // there is a reorg and the blocks on the stale branch are in the ValidationInterface queue
413      // backlog even after the sync thread has caught up to the new chain tip. In this unlikely
414      // event, log a warning and let the queue clear.
415      const CBlockIndex* best_block_index = m_best_block_index.load();
416      if (best_block_index->GetAncestor(locator_tip_index->nHeight) != locator_tip_index) {
417          LogWarning("Locator contains block (hash=%s) not on known best "
418                    "chain (tip=%s); not writing index locator",
419                    locator_tip_hash.ToString(),
420                    best_block_index->GetBlockHash().ToString());
421          return;
422      }
423  
424      // No need to handle errors in Commit. If it fails, the error will be already be logged. The
425      // best way to recover is to continue, as index cannot be corrupted by a missed commit to disk
426      // for an advanced index state.
427      Commit();
428  }
429  
430  bool BaseIndex::BlockUntilSyncedToCurrentChain() const
431  {
432      AssertLockNotHeld(cs_main);
433  
434      if (!m_synced) {
435          return false;
436      }
437  
438      {
439          // Skip the queue-draining stuff if we know we're caught up with
440          // m_chain.Tip().
441          LOCK(cs_main);
442          const CBlockIndex* chain_tip = m_chainstate->m_chain.Tip();
443          const CBlockIndex* best_block_index = m_best_block_index.load();
444          if (best_block_index->GetAncestor(chain_tip->nHeight) == chain_tip) {
445              return true;
446          }
447      }
448  
449      LogInfo("%s is catching up on block notifications", GetName());
450      m_chain->context()->validation_signals->SyncWithValidationInterfaceQueue();
451      return true;
452  }
453  
454  void BaseIndex::Interrupt()
455  {
456      m_interrupt();
457  }
458  
459  bool BaseIndex::StartBackgroundSync()
460  {
461      if (!m_init) throw std::logic_error("Error: Cannot start a non-initialized index");
462  
463      m_thread_sync = std::thread(&util::TraceThread, GetName(), [this] { Sync(); });
464      return true;
465  }
466  
467  void BaseIndex::Stop()
468  {
469      if (m_chain->context()->validation_signals) {
470          m_chain->context()->validation_signals->UnregisterValidationInterface(this);
471      }
472  
473      if (m_thread_sync.joinable()) {
474          m_thread_sync.join();
475      }
476  }
477  
478  IndexSummary BaseIndex::GetSummary() const
479  {
480      IndexSummary summary{};
481      summary.name = GetName();
482      summary.synced = m_synced;
483      if (const auto& pindex = m_best_block_index.load()) {
484          summary.best_block_height = pindex->nHeight;
485          summary.best_block_hash = pindex->GetBlockHash();
486      } else {
487          summary.best_block_height = 0;
488          summary.best_block_hash = m_chain->getBlockHash(0);
489      }
490      return summary;
491  }
492  
493  void BaseIndex::SetBestBlockIndex(const CBlockIndex* block)
494  {
495      assert(!m_chainstate->m_blockman.IsPruneMode() || AllowPrune());
496  
497      if (AllowPrune() && block) {
498          node::PruneLockInfo prune_lock;
499          prune_lock.height_first = block->nHeight;
500          WITH_LOCK(::cs_main, m_chainstate->m_blockman.UpdatePruneLock(GetName(), prune_lock));
501      }
502  
503      // Intentionally set m_best_block_index as the last step in this function,
504      // after updating prune locks above, and after making any other references
505      // to *this, so the BlockUntilSyncedToCurrentChain function (which checks
506      // m_best_block_index as an optimization) can be used to wait for the last
507      // BlockConnected notification and safely assume that prune locks are
508      // updated and that the index object is safe to delete.
509      m_best_block_index = block;
510  }