github.com/theQRL/go-zond@v0.2.1/zond/downloader/downloader.go (about) 1 // Copyright 2015 The go-ethereum Authors 2 // This file is part of the go-ethereum library. 3 // 4 // The go-ethereum library is free software: you can redistribute it and/or modify 5 // it under the terms of the GNU Lesser General Public License as published by 6 // the Free Software Foundation, either version 3 of the License, or 7 // (at your option) any later version. 8 // 9 // The go-ethereum library is distributed in the hope that it will be useful, 10 // but WITHOUT ANY WARRANTY; without even the implied warranty of 11 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 // GNU Lesser General Public License for more details. 13 // 14 // You should have received a copy of the GNU Lesser General Public License 15 // along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>. 16 17 // Package downloader contains the manual full chain synchronisation. 18 package downloader 19 20 import ( 21 "errors" 22 "fmt" 23 "sync" 24 "sync/atomic" 25 "time" 26 27 "github.com/theQRL/go-zond" 28 "github.com/theQRL/go-zond/common" 29 "github.com/theQRL/go-zond/core/rawdb" 30 "github.com/theQRL/go-zond/core/state/snapshot" 31 "github.com/theQRL/go-zond/core/types" 32 "github.com/theQRL/go-zond/event" 33 "github.com/theQRL/go-zond/log" 34 "github.com/theQRL/go-zond/params" 35 "github.com/theQRL/go-zond/trie" 36 "github.com/theQRL/go-zond/zond/protocols/snap" 37 "github.com/theQRL/go-zond/zonddb" 38 ) 39 40 var ( 41 MaxBlockFetch = 128 // Number of blocks to be fetched per retrieval request 42 MaxHeaderFetch = 192 // Number of block headers to be fetched per retrieval request 43 MaxReceiptFetch = 256 // Number of transaction receipts to allow fetching per request 44 45 maxQueuedHeaders = 32 * 1024 // [zond/62] Maximum number of headers to queue for import (DOS protection) 46 maxHeadersProcess = 2048 // Number of header download results to import at once into the chain 47 maxResultsProcess = 2048 // Number of content download results to import at once into the chain 48 fullMaxForkAncestry uint64 = params.FullImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 49 50 reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs 51 52 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 53 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 54 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in snap sync 55 ) 56 57 var ( 58 errBusy = errors.New("busy") 59 errBadPeer = errors.New("action from bad peer ignored") 60 errTimeout = errors.New("timeout") 61 errInvalidChain = errors.New("retrieved hash chain is invalid") 62 errInvalidBody = errors.New("retrieved block body is invalid") 63 errInvalidReceipt = errors.New("retrieved receipt is invalid") 64 errCancelStateFetch = errors.New("state data download canceled (requested)") 65 errCancelContentProcessing = errors.New("content processing canceled (requested)") 66 errCanceled = errors.New("syncing canceled (requested)") 67 errNoPivotHeader = errors.New("pivot header is not found") 68 ) 69 70 // peerDropFn is a callback type for dropping a peer detected as malicious. 71 type peerDropFn func(id string) 72 73 // badBlockFn is a callback for the async beacon sync to notify the caller that 74 // the origin header requested to sync to, produced a chain with a bad block. 75 type badBlockFn func(invalid *types.Header, origin *types.Header) 76 77 // headerTask is a set of downloaded headers to queue along with their precomputed 78 // hashes to avoid constant rehashing. 79 type headerTask struct { 80 headers []*types.Header 81 hashes []common.Hash 82 } 83 84 type Downloader struct { 85 mode atomic.Uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode 86 mux *event.TypeMux // Event multiplexer to announce sync operation events 87 88 queue *queue // Scheduler for selecting the hashes to download 89 peers *peerSet // Set of active peers from which download can proceed 90 91 stateDB zonddb.Database // Database to state sync into (and deduplicate via) 92 93 // Statistics 94 syncStatsChainOrigin uint64 // Origin block number where syncing started at 95 syncStatsChainHeight uint64 // Highest block number known when syncing started 96 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 97 98 blockchain BlockChain 99 100 // Callbacks 101 dropPeer peerDropFn // Drops a peer for misbehaving 102 badBlock badBlockFn // Reports a block as rejected by the chain 103 104 // Status 105 synchronising atomic.Bool 106 notified atomic.Bool 107 committed atomic.Bool 108 ancientLimit uint64 // The maximum block number which can be regarded as ancient data. 109 110 // Channels 111 headerProcCh chan *headerTask // Channel to feed the header processor new tasks 112 113 // Skeleton sync 114 skeleton *skeleton // Header skeleton to backfill the chain with (eth2 mode) 115 116 // State sync 117 pivotHeader *types.Header // Pivot block header to dynamically push the syncing state root 118 pivotLock sync.RWMutex // Lock protecting pivot header reads from updates 119 120 SnapSyncer *snap.Syncer // TODO(karalabe): make private! hack for now 121 stateSyncStart chan *stateSync 122 123 // Cancellation and termination 124 cancelCh chan struct{} // Channel to cancel mid-flight syncs 125 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 126 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 127 128 quitCh chan struct{} // Quit channel to signal termination 129 quitLock sync.Mutex // Lock to prevent double closes 130 131 // Testing hooks 132 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 133 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 134 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 135 136 // Progress reporting metrics 137 syncStartBlock uint64 // Head snap block when Gzond was started 138 syncStartTime time.Time // Time instance when chain sync started 139 syncLogTime time.Time // Time instance when status was last reported 140 } 141 142 // BlockChain encapsulates functions required to sync a (full or snap) blockchain. 143 type BlockChain interface { 144 // HasHeader verifies a header's presence in the local chain. 145 HasHeader(common.Hash, uint64) bool 146 147 // GetHeaderByHash retrieves a header from the local chain. 148 GetHeaderByHash(common.Hash) *types.Header 149 150 // CurrentHeader retrieves the head header from the local chain. 151 CurrentHeader() *types.Header 152 153 // InsertHeaderChain inserts a batch of headers into the local chain. 154 InsertHeaderChain([]*types.Header) (int, error) 155 156 // SetHead rewinds the local chain to a new head. 157 SetHead(uint64) error 158 159 // HasBlock verifies a block's presence in the local chain. 160 HasBlock(common.Hash, uint64) bool 161 162 // HasFastBlock verifies a snap block's presence in the local chain. 163 HasFastBlock(common.Hash, uint64) bool 164 165 // GetBlockByHash retrieves a block from the local chain. 166 GetBlockByHash(common.Hash) *types.Block 167 168 // CurrentBlock retrieves the head block from the local chain. 169 CurrentBlock() *types.Header 170 171 // CurrentSnapBlock retrieves the head snap block from the local chain. 172 CurrentSnapBlock() *types.Header 173 174 // SnapSyncCommitHead directly commits the head block to a certain entity. 175 SnapSyncCommitHead(common.Hash) error 176 177 // InsertChain inserts a batch of blocks into the local chain. 178 InsertChain(types.Blocks) (int, error) 179 180 // InsertReceiptChain inserts a batch of receipts into the local chain. 181 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 182 183 // Snapshots returns the blockchain snapshot tree to paused it during sync. 184 Snapshots() *snapshot.Tree 185 186 // TrieDB retrieves the low level trie database used for interacting 187 // with trie nodes. 188 TrieDB() *trie.Database 189 } 190 191 // New creates a new downloader to fetch hashes and blocks from remote peers. 192 func New(stateDb zonddb.Database, mux *event.TypeMux, chain BlockChain, dropPeer peerDropFn, success func()) *Downloader { 193 dl := &Downloader{ 194 stateDB: stateDb, 195 mux: mux, 196 queue: newQueue(blockCacheMaxItems, blockCacheInitialItems), 197 peers: newPeerSet(), 198 blockchain: chain, 199 dropPeer: dropPeer, 200 headerProcCh: make(chan *headerTask, 1), 201 quitCh: make(chan struct{}), 202 SnapSyncer: snap.NewSyncer(stateDb, chain.TrieDB().Scheme()), 203 stateSyncStart: make(chan *stateSync), 204 syncStartBlock: chain.CurrentSnapBlock().Number.Uint64(), 205 } 206 // Create the post-merge skeleton syncer and start the process 207 dl.skeleton = newSkeleton(stateDb, dl.peers, dropPeer, newBeaconBackfiller(dl, success)) 208 209 go dl.stateFetcher() 210 return dl 211 } 212 213 // Progress retrieves the synchronisation boundaries, specifically the origin 214 // block where synchronisation started at (may have failed/suspended); the block 215 // or header sync is currently at; and the latest known block which the sync targets. 216 // 217 // In addition, during the state download phase of snap synchronisation the number 218 // of processed and the total number of known states are also returned. Otherwise 219 // these are zero. 220 func (d *Downloader) Progress() zond.SyncProgress { 221 // Lock the current stats and return the progress 222 d.syncStatsLock.RLock() 223 defer d.syncStatsLock.RUnlock() 224 225 current := uint64(0) 226 mode := d.getMode() 227 switch mode { 228 case FullSync: 229 current = d.blockchain.CurrentBlock().Number.Uint64() 230 case SnapSync: 231 current = d.blockchain.CurrentSnapBlock().Number.Uint64() 232 default: 233 log.Error("Unknown downloader mode", "mode", mode) 234 } 235 progress, pending := d.SnapSyncer.Progress() 236 237 return zond.SyncProgress{ 238 StartingBlock: d.syncStatsChainOrigin, 239 CurrentBlock: current, 240 HighestBlock: d.syncStatsChainHeight, 241 SyncedAccounts: progress.AccountSynced, 242 SyncedAccountBytes: uint64(progress.AccountBytes), 243 SyncedBytecodes: progress.BytecodeSynced, 244 SyncedBytecodeBytes: uint64(progress.BytecodeBytes), 245 SyncedStorage: progress.StorageSynced, 246 SyncedStorageBytes: uint64(progress.StorageBytes), 247 HealedTrienodes: progress.TrienodeHealSynced, 248 HealedTrienodeBytes: uint64(progress.TrienodeHealBytes), 249 HealedBytecodes: progress.BytecodeHealSynced, 250 HealedBytecodeBytes: uint64(progress.BytecodeHealBytes), 251 HealingTrienodes: pending.TrienodeHeal, 252 HealingBytecode: pending.BytecodeHeal, 253 } 254 } 255 256 // RegisterPeer injects a new download peer into the set of block source to be 257 // used for fetching hashes and blocks from. 258 func (d *Downloader) RegisterPeer(id string, version uint, peer Peer) error { 259 var logger log.Logger 260 if len(id) < 16 { 261 // Tests use short IDs, don't choke on them 262 logger = log.New("peer", id) 263 } else { 264 logger = log.New("peer", id[:8]) 265 } 266 logger.Trace("Registering sync peer") 267 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 268 logger.Error("Failed to register sync peer", "err", err) 269 return err 270 } 271 return nil 272 } 273 274 // UnregisterPeer remove a peer from the known list, preventing any action from 275 // the specified peer. An effort is also made to return any pending fetches into 276 // the queue. 277 func (d *Downloader) UnregisterPeer(id string) error { 278 // Unregister the peer from the active peer set and revoke any fetch tasks 279 var logger log.Logger 280 if len(id) < 16 { 281 // Tests use short IDs, don't choke on them 282 logger = log.New("peer", id) 283 } else { 284 logger = log.New("peer", id[:8]) 285 } 286 logger.Trace("Unregistering sync peer") 287 if err := d.peers.Unregister(id); err != nil { 288 logger.Error("Failed to unregister sync peer", "err", err) 289 return err 290 } 291 d.queue.Revoke(id) 292 293 return nil 294 } 295 296 // synchronise will select the peer and use it for synchronising. If an empty string is given 297 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 298 // checks fail an error will be returned. This method is synchronous 299 func (d *Downloader) synchronise(mode SyncMode, beaconPing chan struct{}) error { 300 // The beacon header syncer is async. It will start this synchronization and 301 // will continue doing other tasks. However, if synchronization needs to be 302 // cancelled, the syncer needs to know if we reached the startup point (and 303 // inited the cancel channel) or not yet. Make sure that we'll signal even in 304 // case of a failure. 305 if beaconPing != nil { 306 defer func() { 307 select { 308 case <-beaconPing: // already notified 309 default: 310 close(beaconPing) // weird exit condition, notify that it's safe to cancel (the nothing) 311 } 312 }() 313 } 314 // Make sure only one goroutine is ever allowed past this point at once 315 if !d.synchronising.CompareAndSwap(false, true) { 316 return errBusy 317 } 318 defer d.synchronising.Store(false) 319 320 // Post a user notification of the sync (only once per session) 321 if d.notified.CompareAndSwap(false, true) { 322 log.Info("Block synchronisation started") 323 } 324 if mode == SnapSync { 325 // Snap sync will directly modify the persistent state, making the entire 326 // trie database unusable until the state is fully synced. To prevent any 327 // subsequent state reads, explicitly disable the trie database and state 328 // syncer is responsible to address and correct any state missing. 329 if d.blockchain.TrieDB().Scheme() == rawdb.PathScheme { 330 d.blockchain.TrieDB().Reset(types.EmptyRootHash) 331 } 332 // Snap sync uses the snapshot namespace to store potentially flaky data until 333 // sync completely heals and finishes. Pause snapshot maintenance in the mean- 334 // time to prevent access. 335 if snapshots := d.blockchain.Snapshots(); snapshots != nil { // Only nil in tests 336 snapshots.Disable() 337 } 338 } 339 // Reset the queue, peer set and wake channels to clean any internal leftover state 340 d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems) 341 d.peers.Reset() 342 343 for _, ch := range []chan bool{d.queue.blockWakeCh, d.queue.receiptWakeCh} { 344 select { 345 case <-ch: 346 default: 347 } 348 } 349 for empty := false; !empty; { 350 select { 351 case <-d.headerProcCh: 352 default: 353 empty = true 354 } 355 } 356 // Create cancel channel for aborting mid-flight and mark the master peer 357 d.cancelLock.Lock() 358 d.cancelCh = make(chan struct{}) 359 d.cancelLock.Unlock() 360 361 defer d.Cancel() // No matter what, we can't leave the cancel channel open 362 363 // Atomically set the requested sync mode 364 d.mode.Store(uint32(mode)) 365 366 if beaconPing != nil { 367 close(beaconPing) 368 } 369 return d.syncToHead() 370 } 371 372 func (d *Downloader) getMode() SyncMode { 373 return SyncMode(d.mode.Load()) 374 } 375 376 // syncToHead starts a block synchronization based on the hash chain from 377 // the specified head hash. 378 func (d *Downloader) syncToHead() (err error) { 379 d.mux.Post(StartEvent{}) 380 defer func() { 381 // reset on error 382 if err != nil { 383 d.mux.Post(FailedEvent{err}) 384 } else { 385 latest := d.blockchain.CurrentHeader() 386 d.mux.Post(DoneEvent{latest}) 387 } 388 }() 389 mode := d.getMode() 390 391 log.Debug("Backfilling with the network", "mode", mode) 392 defer func(start time.Time) { 393 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 394 }(time.Now()) 395 396 // Look up the sync boundaries: the common ancestor and the target block 397 var latest, pivot, final *types.Header 398 399 latest, _, final, err = d.skeleton.Bounds() 400 if err != nil { 401 return err 402 } 403 if latest.Number.Uint64() > uint64(fsMinFullBlocks) { 404 number := latest.Number.Uint64() - uint64(fsMinFullBlocks) 405 406 // Retrieve the pivot header from the skeleton chain segment but 407 // fallback to local chain if it's not found in skeleton space. 408 if pivot = d.skeleton.Header(number); pivot == nil { 409 _, oldest, _, _ := d.skeleton.Bounds() // error is already checked 410 if number < oldest.Number.Uint64() { 411 count := int(oldest.Number.Uint64() - number) // it's capped by fsMinFullBlocks 412 headers := d.readHeaderRange(oldest, count) 413 if len(headers) == count { 414 pivot = headers[len(headers)-1] 415 log.Warn("Retrieved pivot header from local", "number", pivot.Number, "hash", pivot.Hash(), "latest", latest.Number, "oldest", oldest.Number) 416 } 417 } 418 } 419 // Print an error log and return directly in case the pivot header 420 // is still not found. It means the skeleton chain is not linked 421 // correctly with local chain. 422 if pivot == nil { 423 log.Error("Pivot header is not found", "number", number) 424 return errNoPivotHeader 425 } 426 } 427 // If no pivot block was returned, the head is below the min full block 428 // threshold (i.e. new chain). In that case we won't really snap sync 429 // anyway, but still need a valid pivot block to avoid some code hitting 430 // nil panics on access. 431 if mode == SnapSync && pivot == nil { 432 pivot = d.blockchain.CurrentBlock() 433 } 434 height := latest.Number.Uint64() 435 436 // In beacon mode, use the skeleton chain for the ancestor lookup 437 origin, err := d.findBeaconAncestor() 438 if err != nil { 439 return err 440 } 441 d.syncStatsLock.Lock() 442 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 443 d.syncStatsChainOrigin = origin 444 } 445 d.syncStatsChainHeight = height 446 d.syncStatsLock.Unlock() 447 448 // Ensure our origin point is below any snap sync pivot point 449 if mode == SnapSync { 450 if height <= uint64(fsMinFullBlocks) { 451 origin = 0 452 } else { 453 pivotNumber := pivot.Number.Uint64() 454 if pivotNumber <= origin { 455 origin = pivotNumber - 1 456 } 457 // Write out the pivot into the database so a rollback beyond it will 458 // reenable snap sync 459 rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber) 460 } 461 } 462 d.committed.Store(true) 463 if mode == SnapSync && pivot.Number.Uint64() != 0 { 464 d.committed.Store(false) 465 } 466 if mode == SnapSync { 467 // Set the ancient data limitation. If we are running snap sync, all block 468 // data older than ancientLimit will be written to the ancient store. More 469 // recent data will be written to the active database and will wait for the 470 // freezer to migrate. 471 // 472 // If the network is post-merge, use either the last announced finalized 473 // block as the ancient limit, or if we haven't yet received one, the head- 474 // a max fork ancestry limit. One quirky case if we've already passed the 475 // finalized block, in which case the skeleton.Bounds will return nil and 476 // we'll revert to head - 90K. That's fine, we're finishing sync anyway. 477 // 478 // For non-merged networks, if there is a checkpoint available, then calculate 479 // the ancientLimit through that. Otherwise calculate the ancient limit through 480 // the advertised height of the remote peer. This most is mostly a fallback for 481 // legacy networks, but should eventually be dropped. TODO(karalabe). 482 // 483 // Beacon sync, use the latest finalized block as the ancient limit 484 // or a reasonable height if no finalized block is yet announced. 485 if final != nil { 486 d.ancientLimit = final.Number.Uint64() 487 } else if height > fullMaxForkAncestry+1 { 488 d.ancientLimit = height - fullMaxForkAncestry - 1 489 } else { 490 d.ancientLimit = 0 491 } 492 493 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 494 495 // If a part of blockchain data has already been written into active store, 496 // disable the ancient style insertion explicitly. 497 if origin >= frozen && frozen != 0 { 498 d.ancientLimit = 0 499 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 500 } else if d.ancientLimit > 0 { 501 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 502 } 503 // Rewind the ancient store and blockchain if reorg happens. 504 if origin+1 < frozen { 505 if err := d.blockchain.SetHead(origin); err != nil { 506 return err 507 } 508 log.Info("Truncated excess ancient chain segment", "oldhead", frozen-1, "newhead", origin) 509 } 510 } 511 // Initiate the sync using a concurrent header and content retrieval algorithm 512 d.queue.Prepare(origin+1, mode) 513 // In beacon mode, headers are served by the skeleton syncer 514 fetchers := []func() error{ 515 func() error { return d.fetchHeaders(origin + 1) }, // Headers are always retrieved 516 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and snap sync 517 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during snap sync 518 func() error { return d.processHeaders(origin + 1) }, 519 } 520 if mode == SnapSync { 521 d.pivotLock.Lock() 522 d.pivotHeader = pivot 523 d.pivotLock.Unlock() 524 525 fetchers = append(fetchers, func() error { return d.processSnapSyncContent() }) 526 } else if mode == FullSync { 527 fetchers = append(fetchers, func() error { return d.processFullSyncContent() }) 528 } 529 return d.spawnSync(fetchers) 530 } 531 532 // spawnSync runs d.process and all given fetcher functions to completion in 533 // separate goroutines, returning the first error that appears. 534 func (d *Downloader) spawnSync(fetchers []func() error) error { 535 errc := make(chan error, len(fetchers)) 536 d.cancelWg.Add(len(fetchers)) 537 for _, fn := range fetchers { 538 fn := fn 539 go func() { defer d.cancelWg.Done(); errc <- fn() }() 540 } 541 // Wait for the first error, then terminate the others. 542 var err error 543 for i := 0; i < len(fetchers); i++ { 544 if i == len(fetchers)-1 { 545 // Close the queue when all fetchers have exited. 546 // This will cause the block processor to end when 547 // it has processed the queue. 548 d.queue.Close() 549 } 550 if got := <-errc; got != nil { 551 err = got 552 if got != errCanceled { 553 break // receive a meaningful error, bubble it up 554 } 555 } 556 } 557 d.queue.Close() 558 d.Cancel() 559 return err 560 } 561 562 // cancel aborts all of the operations and resets the queue. However, cancel does 563 // not wait for the running download goroutines to finish. This method should be 564 // used when cancelling the downloads from inside the downloader. 565 func (d *Downloader) cancel() { 566 // Close the current cancel channel 567 d.cancelLock.Lock() 568 defer d.cancelLock.Unlock() 569 570 if d.cancelCh != nil { 571 select { 572 case <-d.cancelCh: 573 // Channel was already closed 574 default: 575 close(d.cancelCh) 576 } 577 } 578 } 579 580 // Cancel aborts all of the operations and waits for all download goroutines to 581 // finish before returning. 582 func (d *Downloader) Cancel() { 583 d.cancel() 584 d.cancelWg.Wait() 585 } 586 587 // Terminate interrupts the downloader, canceling all pending operations. 588 // The downloader cannot be reused after calling Terminate. 589 func (d *Downloader) Terminate() { 590 // Close the termination channel (make sure double close is allowed) 591 d.quitLock.Lock() 592 select { 593 case <-d.quitCh: 594 default: 595 close(d.quitCh) 596 597 // Terminate the internal beacon syncer 598 d.skeleton.Terminate() 599 } 600 d.quitLock.Unlock() 601 602 // Cancel any pending download requests 603 d.Cancel() 604 } 605 606 // fetchBodies iteratively downloads the scheduled block bodies, taking any 607 // available peers, reserving a chunk of blocks for each, waiting for delivery 608 // and also periodically checking for timeouts. 609 func (d *Downloader) fetchBodies(from uint64) error { 610 log.Debug("Downloading block bodies", "origin", from) 611 err := d.concurrentFetch((*bodyQueue)(d)) 612 613 log.Debug("Block body download terminated", "err", err) 614 return err 615 } 616 617 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 618 // available peers, reserving a chunk of receipts for each, waiting for delivery 619 // and also periodically checking for timeouts. 620 func (d *Downloader) fetchReceipts(from uint64) error { 621 log.Debug("Downloading receipts", "origin", from) 622 err := d.concurrentFetch((*receiptQueue)(d)) 623 624 log.Debug("Receipt download terminated", "err", err) 625 return err 626 } 627 628 // processHeaders takes batches of retrieved headers from an input channel and 629 // keeps processing and scheduling them into the header chain and downloader's 630 // queue until the stream ends or a failure occurs. 631 func (d *Downloader) processHeaders(origin uint64) error { 632 var ( 633 mode = d.getMode() 634 timer = time.NewTimer(time.Second) 635 ) 636 defer timer.Stop() 637 638 for { 639 select { 640 case <-d.cancelCh: 641 return errCanceled 642 643 case task := <-d.headerProcCh: 644 // Terminate header processing if we synced up 645 if task == nil || len(task.headers) == 0 { 646 // Notify everyone that headers are fully processed 647 for _, ch := range []chan bool{d.queue.blockWakeCh, d.queue.receiptWakeCh} { 648 select { 649 case ch <- false: 650 case <-d.cancelCh: 651 } 652 } 653 return nil 654 } 655 // Otherwise split the chunk of headers into batches and process them 656 headers, hashes := task.headers, task.hashes 657 658 for len(headers) > 0 { 659 // Terminate if something failed in between processing chunks 660 select { 661 case <-d.cancelCh: 662 return errCanceled 663 default: 664 } 665 // Select the next chunk of headers to import 666 limit := maxHeadersProcess 667 if limit > len(headers) { 668 limit = len(headers) 669 } 670 chunkHeaders := headers[:limit] 671 chunkHashes := hashes[:limit] 672 673 // In case of header only syncing, validate the chunk immediately 674 if mode == SnapSync { 675 // Although the received headers might be all valid, a legacy 676 // PoW/PoA sync must not accept post-merge headers. Make sure 677 // that any transition is rejected at this point. 678 if len(chunkHeaders) > 0 { 679 if n, err := d.blockchain.InsertHeaderChain(chunkHeaders); err != nil { 680 log.Warn("Invalid header encountered", "number", chunkHeaders[n].Number, "hash", chunkHashes[n], "parent", chunkHeaders[n].ParentHash, "err", err) 681 return fmt.Errorf("%w: %v", errInvalidChain, err) 682 } 683 } 684 } 685 // If we've reached the allowed number of pending headers, stall a bit 686 for d.queue.PendingBodies() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 687 timer.Reset(time.Second) 688 select { 689 case <-d.cancelCh: 690 return errCanceled 691 case <-timer.C: 692 } 693 } 694 // Otherwise insert the headers for content retrieval 695 inserts := d.queue.Schedule(chunkHeaders, chunkHashes, origin) 696 if len(inserts) != len(chunkHeaders) { 697 return fmt.Errorf("%w: stale headers", errBadPeer) 698 } 699 700 headers = headers[limit:] 701 hashes = hashes[limit:] 702 origin += uint64(limit) 703 } 704 // Update the highest block number we know if a higher one is found. 705 d.syncStatsLock.Lock() 706 if d.syncStatsChainHeight < origin { 707 d.syncStatsChainHeight = origin - 1 708 } 709 d.syncStatsLock.Unlock() 710 711 // Signal the content downloaders of the availability of new tasks 712 for _, ch := range []chan bool{d.queue.blockWakeCh, d.queue.receiptWakeCh} { 713 select { 714 case ch <- true: 715 default: 716 } 717 } 718 } 719 } 720 } 721 722 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 723 func (d *Downloader) processFullSyncContent() error { 724 for { 725 results := d.queue.Results(true) 726 if len(results) == 0 { 727 return nil 728 } 729 if d.chainInsertHook != nil { 730 d.chainInsertHook(results) 731 } 732 if err := d.importBlockResults(results); err != nil { 733 return err 734 } 735 } 736 } 737 738 func (d *Downloader) importBlockResults(results []*fetchResult) error { 739 // Check for any early termination requests 740 if len(results) == 0 { 741 return nil 742 } 743 select { 744 case <-d.quitCh: 745 return errCancelContentProcessing 746 default: 747 } 748 // Retrieve a batch of results to import 749 first, last := results[0].Header, results[len(results)-1].Header 750 log.Debug("Inserting downloaded chain", "items", len(results), 751 "firstnum", first.Number, "firsthash", first.Hash(), 752 "lastnum", last.Number, "lasthash", last.Hash(), 753 ) 754 blocks := make([]*types.Block, len(results)) 755 for i, result := range results { 756 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.body()) 757 } 758 // Downloaded blocks are always regarded as trusted after the 759 // transition. Because the downloaded chain is guided by the 760 // consensus-layer. 761 if index, err := d.blockchain.InsertChain(blocks); err != nil { 762 if index < len(results) { 763 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 764 765 // In post-merge, notify the engine API of encountered bad chains 766 if d.badBlock != nil { 767 head, _, _, err := d.skeleton.Bounds() 768 if err != nil { 769 log.Error("Failed to retrieve beacon bounds for bad block reporting", "err", err) 770 } else { 771 d.badBlock(blocks[index].Header(), head) 772 } 773 } 774 } else { 775 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 776 // when it needs to preprocess blocks to import a sidechain. 777 // The importer will put together a new list of blocks to import, which is a superset 778 // of the blocks delivered from the downloader, and the indexing will be off. 779 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 780 } 781 return fmt.Errorf("%w: %v", errInvalidChain, err) 782 } 783 return nil 784 } 785 786 // processSnapSyncContent takes fetch results from the queue and writes them to the 787 // database. It also controls the synchronisation of state nodes of the pivot block. 788 func (d *Downloader) processSnapSyncContent() error { 789 // Start syncing state of the reported head block. This should get us most of 790 // the state of the pivot block. 791 d.pivotLock.RLock() 792 sync := d.syncState(d.pivotHeader.Root) 793 d.pivotLock.RUnlock() 794 795 defer func() { 796 // The `sync` object is replaced every time the pivot moves. We need to 797 // defer close the very last active one, hence the lazy evaluation vs. 798 // calling defer sync.Cancel() !!! 799 sync.Cancel() 800 }() 801 802 closeOnErr := func(s *stateSync) { 803 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled && err != snap.ErrCancelled { 804 d.queue.Close() // wake up Results 805 } 806 } 807 go closeOnErr(sync) 808 809 // To cater for moving pivot points, track the pivot block and subsequently 810 // accumulated download results separately. 811 // 812 // These will be nil up to the point where we reach the pivot, and will only 813 // be set temporarily if the synced blocks are piling up, but the pivot is 814 // still busy downloading. In that case, we need to occasionally check for 815 // pivot moves, so need to unblock the loop. These fields will accumulate 816 // the results in the meantime. 817 // 818 // Note, there's no issue with memory piling up since after 64 blocks the 819 // pivot will forcefully move so these accumulators will be dropped. 820 var ( 821 oldPivot *fetchResult // Locked in pivot block, might change eventually 822 oldTail []*fetchResult // Downloaded content after the pivot 823 timer = time.NewTimer(time.Second) 824 ) 825 defer timer.Stop() 826 827 for { 828 // Wait for the next batch of downloaded data to be available. If we have 829 // not yet reached the pivot point, wait blockingly as there's no need to 830 // spin-loop check for pivot moves. If we reached the pivot but have not 831 // yet processed it, check for results async, so we might notice pivot 832 // moves while state syncing. If the pivot was passed fully, block again 833 // as there's no more reason to check for pivot moves at all. 834 results := d.queue.Results(oldPivot == nil) 835 if len(results) == 0 { 836 // If pivot sync is done, stop 837 if d.committed.Load() { 838 d.reportSnapSyncProgress(true) 839 return sync.Cancel() 840 } 841 // If sync failed, stop 842 select { 843 case <-d.cancelCh: 844 sync.Cancel() 845 return errCanceled 846 default: 847 } 848 } 849 if d.chainInsertHook != nil { 850 d.chainInsertHook(results) 851 } 852 d.reportSnapSyncProgress(false) 853 854 // If we haven't downloaded the pivot block yet, check pivot staleness 855 // notifications from the header downloader 856 d.pivotLock.RLock() 857 pivot := d.pivotHeader 858 d.pivotLock.RUnlock() 859 860 if oldPivot == nil { // no results piling up, we can move the pivot 861 if !d.committed.Load() { // not yet passed the pivot, we can move the pivot 862 if pivot.Root != sync.root { // pivot position changed, we can move the pivot 863 sync.Cancel() 864 sync = d.syncState(pivot.Root) 865 866 go closeOnErr(sync) 867 } 868 } 869 } else { // results already piled up, consume before handling pivot move 870 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 871 } 872 // Split around the pivot block and process the two sides via snap/full sync 873 if !d.committed.Load() { 874 latest := results[len(results)-1].Header 875 // If the height is above the pivot block by 2 sets, it means the pivot 876 // become stale in the network, and it was garbage collected, move to a 877 // new pivot. 878 // 879 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 880 // need to be taken into account, otherwise we're detecting the pivot move 881 // late and will drop peers due to unavailable state!!! 882 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 883 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 884 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 885 886 d.pivotLock.Lock() 887 d.pivotHeader = pivot 888 d.pivotLock.Unlock() 889 890 // Write out the pivot into the database so a rollback beyond it will 891 // reenable snap sync 892 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 893 } 894 } 895 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 896 if err := d.commitSnapSyncData(beforeP, sync); err != nil { 897 return err 898 } 899 if P != nil { 900 // If new pivot block found, cancel old state retrieval and restart 901 if oldPivot != P { 902 sync.Cancel() 903 sync = d.syncState(P.Header.Root) 904 905 go closeOnErr(sync) 906 oldPivot = P 907 } 908 // Wait for completion, occasionally checking for pivot staleness 909 timer.Reset(time.Second) 910 select { 911 case <-sync.done: 912 if sync.err != nil { 913 return sync.err 914 } 915 if err := d.commitPivotBlock(P); err != nil { 916 return err 917 } 918 oldPivot = nil 919 920 case <-timer.C: 921 oldTail = afterP 922 continue 923 } 924 } 925 // Fast sync done, pivot commit done, full import 926 if err := d.importBlockResults(afterP); err != nil { 927 return err 928 } 929 } 930 } 931 932 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 933 if len(results) == 0 { 934 return nil, nil, nil 935 } 936 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 937 // the pivot is somewhere in the future 938 return nil, results, nil 939 } 940 // This can also be optimized, but only happens very seldom 941 for _, result := range results { 942 num := result.Header.Number.Uint64() 943 switch { 944 case num < pivot: 945 before = append(before, result) 946 case num == pivot: 947 p = result 948 default: 949 after = append(after, result) 950 } 951 } 952 return p, before, after 953 } 954 955 func (d *Downloader) commitSnapSyncData(results []*fetchResult, stateSync *stateSync) error { 956 // Check for any early termination requests 957 if len(results) == 0 { 958 return nil 959 } 960 select { 961 case <-d.quitCh: 962 return errCancelContentProcessing 963 case <-stateSync.done: 964 if err := stateSync.Wait(); err != nil { 965 return err 966 } 967 default: 968 } 969 // Retrieve the batch of results to import 970 first, last := results[0].Header, results[len(results)-1].Header 971 log.Debug("Inserting snap-sync blocks", "items", len(results), 972 "firstnum", first.Number, "firsthash", first.Hash(), 973 "lastnumn", last.Number, "lasthash", last.Hash(), 974 ) 975 blocks := make([]*types.Block, len(results)) 976 receipts := make([]types.Receipts, len(results)) 977 for i, result := range results { 978 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.body()) 979 receipts[i] = result.Receipts 980 } 981 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 982 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 983 return fmt.Errorf("%w: %v", errInvalidChain, err) 984 } 985 return nil 986 } 987 988 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 989 block := types.NewBlockWithHeader(result.Header).WithBody(result.body()) 990 log.Debug("Committing snap sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 991 992 // Commit the pivot block as the new head, will require full sync from here on 993 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 994 return err 995 } 996 if err := d.blockchain.SnapSyncCommitHead(block.Hash()); err != nil { 997 return err 998 } 999 d.committed.Store(true) 1000 return nil 1001 } 1002 1003 // DeliverSnapPacket is invoked from a peer's message handler when it transmits a 1004 // data packet for the local node to consume. 1005 func (d *Downloader) DeliverSnapPacket(peer *snap.Peer, packet snap.Packet) error { 1006 switch packet := packet.(type) { 1007 case *snap.AccountRangePacket: 1008 hashes, accounts, err := packet.Unpack() 1009 if err != nil { 1010 return err 1011 } 1012 return d.SnapSyncer.OnAccounts(peer, packet.ID, hashes, accounts, packet.Proof) 1013 1014 case *snap.StorageRangesPacket: 1015 hashset, slotset := packet.Unpack() 1016 return d.SnapSyncer.OnStorage(peer, packet.ID, hashset, slotset, packet.Proof) 1017 1018 case *snap.ByteCodesPacket: 1019 return d.SnapSyncer.OnByteCodes(peer, packet.ID, packet.Codes) 1020 1021 case *snap.TrieNodesPacket: 1022 return d.SnapSyncer.OnTrieNodes(peer, packet.ID, packet.Nodes) 1023 1024 default: 1025 return fmt.Errorf("unexpected snap packet type: %T", packet) 1026 } 1027 } 1028 1029 // readHeaderRange returns a list of headers, using the given last header as the base, 1030 // and going backwards towards genesis. This method assumes that the caller already has 1031 // placed a reasonable cap on count. 1032 func (d *Downloader) readHeaderRange(last *types.Header, count int) []*types.Header { 1033 var ( 1034 current = last 1035 headers []*types.Header 1036 ) 1037 for { 1038 parent := d.blockchain.GetHeaderByHash(current.ParentHash) 1039 if parent == nil { 1040 break // The chain is not continuous, or the chain is exhausted 1041 } 1042 headers = append(headers, parent) 1043 if len(headers) >= count { 1044 break 1045 } 1046 current = parent 1047 } 1048 return headers 1049 } 1050 1051 // reportSnapSyncProgress calculates various status reports and provides it to the user. 1052 func (d *Downloader) reportSnapSyncProgress(force bool) { 1053 // Initialize the sync start time if it's the first time we're reporting 1054 if d.syncStartTime.IsZero() { 1055 d.syncStartTime = time.Now().Add(-time.Millisecond) // -1ms offset to avoid division by zero 1056 } 1057 // Don't report all the events, just occasionally 1058 if !force && time.Since(d.syncLogTime) < 8*time.Second { 1059 return 1060 } 1061 // Don't report anything until we have a meaningful progress 1062 var ( 1063 headerBytes, _ = d.stateDB.AncientSize(rawdb.ChainFreezerHeaderTable) 1064 bodyBytes, _ = d.stateDB.AncientSize(rawdb.ChainFreezerBodiesTable) 1065 receiptBytes, _ = d.stateDB.AncientSize(rawdb.ChainFreezerReceiptTable) 1066 ) 1067 syncedBytes := common.StorageSize(headerBytes + bodyBytes + receiptBytes) 1068 if syncedBytes == 0 { 1069 return 1070 } 1071 var ( 1072 header = d.blockchain.CurrentHeader() 1073 block = d.blockchain.CurrentSnapBlock() 1074 ) 1075 syncedBlocks := block.Number.Uint64() - d.syncStartBlock 1076 if syncedBlocks == 0 { 1077 return 1078 } 1079 // Retrieve the current chain head and calculate the ETA 1080 latest, _, _, err := d.skeleton.Bounds() 1081 if err != nil { 1082 // We're going to cheat for non-merged networks, but that's fine 1083 latest = d.pivotHeader 1084 } 1085 if latest == nil { 1086 // This should really never happen, but add some defensive code for now. 1087 // TODO(karalabe): Remove it eventually if we don't see it blow. 1088 log.Error("Nil latest block in sync progress report") 1089 return 1090 } 1091 var ( 1092 left = latest.Number.Uint64() - block.Number.Uint64() 1093 eta = time.Since(d.syncStartTime) / time.Duration(syncedBlocks) * time.Duration(left) 1094 1095 progress = fmt.Sprintf("%.2f%%", float64(block.Number.Uint64())*100/float64(latest.Number.Uint64())) 1096 headers = fmt.Sprintf("%v@%v", log.FormatLogfmtUint64(header.Number.Uint64()), common.StorageSize(headerBytes).TerminalString()) 1097 bodies = fmt.Sprintf("%v@%v", log.FormatLogfmtUint64(block.Number.Uint64()), common.StorageSize(bodyBytes).TerminalString()) 1098 receipts = fmt.Sprintf("%v@%v", log.FormatLogfmtUint64(block.Number.Uint64()), common.StorageSize(receiptBytes).TerminalString()) 1099 ) 1100 log.Info("Syncing: chain download in progress", "synced", progress, "chain", syncedBytes, "headers", headers, "bodies", bodies, "receipts", receipts, "eta", common.PrettyDuration(eta)) 1101 d.syncLogTime = time.Now() 1102 }