gitee.com/liu-zhao234568/cntest@v1.0.0/eth/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 "math/big" 24 "sync" 25 "sync/atomic" 26 "time" 27 28 ethereum "gitee.com/liu-zhao234568/cntest" 29 "gitee.com/liu-zhao234568/cntest/common" 30 "gitee.com/liu-zhao234568/cntest/core/rawdb" 31 "gitee.com/liu-zhao234568/cntest/core/state/snapshot" 32 "gitee.com/liu-zhao234568/cntest/core/types" 33 "gitee.com/liu-zhao234568/cntest/eth/protocols/eth" 34 "gitee.com/liu-zhao234568/cntest/eth/protocols/snap" 35 "gitee.com/liu-zhao234568/cntest/ethdb" 36 "gitee.com/liu-zhao234568/cntest/event" 37 "gitee.com/liu-zhao234568/cntest/log" 38 "gitee.com/liu-zhao234568/cntest/metrics" 39 "gitee.com/liu-zhao234568/cntest/params" 40 "gitee.com/liu-zhao234568/cntest/trie" 41 ) 42 43 var ( 44 MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request 45 MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request 46 MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly 47 MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request 48 MaxStateFetch = 384 // Amount of node state values to allow fetching per request 49 50 maxQueuedHeaders = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection) 51 maxHeadersProcess = 2048 // Number of header download results to import at once into the chain 52 maxResultsProcess = 2048 // Number of content download results to import at once into the chain 53 fullMaxForkAncestry uint64 = params.FullImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 54 lightMaxForkAncestry uint64 = params.LightImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 55 56 reorgProtThreshold = 48 // Threshold number of recent blocks to disable mini reorg protection 57 reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs 58 59 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 60 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 61 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 62 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 63 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 64 ) 65 66 var ( 67 errBusy = errors.New("busy") 68 errUnknownPeer = errors.New("peer is unknown or unhealthy") 69 errBadPeer = errors.New("action from bad peer ignored") 70 errStallingPeer = errors.New("peer is stalling") 71 errUnsyncedPeer = errors.New("unsynced peer") 72 errNoPeers = errors.New("no peers to keep download active") 73 errTimeout = errors.New("timeout") 74 errEmptyHeaderSet = errors.New("empty header set by peer") 75 errPeersUnavailable = errors.New("no peers available or all tried for download") 76 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 77 errInvalidChain = errors.New("retrieved hash chain is invalid") 78 errInvalidBody = errors.New("retrieved block body is invalid") 79 errInvalidReceipt = errors.New("retrieved receipt is invalid") 80 errCancelStateFetch = errors.New("state data download canceled (requested)") 81 errCancelContentProcessing = errors.New("content processing canceled (requested)") 82 errCanceled = errors.New("syncing canceled (requested)") 83 errNoSyncActive = errors.New("no sync active") 84 errTooOld = errors.New("peer's protocol version too old") 85 errNoAncestorFound = errors.New("no common ancestor found") 86 ) 87 88 type Downloader struct { 89 mode uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode 90 mux *event.TypeMux // Event multiplexer to announce sync operation events 91 92 checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync) 93 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 94 queue *queue // Scheduler for selecting the hashes to download 95 peers *peerSet // Set of active peers from which download can proceed 96 97 stateDB ethdb.Database // Database to state sync into (and deduplicate via) 98 stateBloom *trie.SyncBloom // Bloom filter for fast trie node and contract code existence checks 99 100 // Statistics 101 syncStatsChainOrigin uint64 // Origin block number where syncing started at 102 syncStatsChainHeight uint64 // Highest block number known when syncing started 103 syncStatsState stateSyncStats 104 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 105 106 lightchain LightChain 107 blockchain BlockChain 108 109 // Callbacks 110 dropPeer peerDropFn // Drops a peer for misbehaving 111 112 // Status 113 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 114 synchronising int32 115 notified int32 116 committed int32 117 ancientLimit uint64 // The maximum block number which can be regarded as ancient data. 118 119 // Channels 120 headerCh chan dataPack // Channel receiving inbound block headers 121 bodyCh chan dataPack // Channel receiving inbound block bodies 122 receiptCh chan dataPack // Channel receiving inbound receipts 123 bodyWakeCh chan bool // Channel to signal the block body fetcher of new tasks 124 receiptWakeCh chan bool // Channel to signal the receipt fetcher of new tasks 125 headerProcCh chan []*types.Header // Channel to feed the header processor new tasks 126 127 // State sync 128 pivotHeader *types.Header // Pivot block header to dynamically push the syncing state root 129 pivotLock sync.RWMutex // Lock protecting pivot header reads from updates 130 131 snapSync bool // Whether to run state sync over the snap protocol 132 SnapSyncer *snap.Syncer // TODO(karalabe): make private! hack for now 133 stateSyncStart chan *stateSync 134 trackStateReq chan *stateReq 135 stateCh chan dataPack // Channel receiving inbound node state data 136 137 // Cancellation and termination 138 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 139 cancelCh chan struct{} // Channel to cancel mid-flight syncs 140 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 141 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 142 143 quitCh chan struct{} // Quit channel to signal termination 144 quitLock sync.Mutex // Lock to prevent double closes 145 146 // Testing hooks 147 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 148 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 149 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 150 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 151 } 152 153 // LightChain encapsulates functions required to synchronise a light chain. 154 type LightChain interface { 155 // HasHeader verifies a header's presence in the local chain. 156 HasHeader(common.Hash, uint64) bool 157 158 // GetHeaderByHash retrieves a header from the local chain. 159 GetHeaderByHash(common.Hash) *types.Header 160 161 // CurrentHeader retrieves the head header from the local chain. 162 CurrentHeader() *types.Header 163 164 // GetTd returns the total difficulty of a local block. 165 GetTd(common.Hash, uint64) *big.Int 166 167 // InsertHeaderChain inserts a batch of headers into the local chain. 168 InsertHeaderChain([]*types.Header, int) (int, error) 169 170 // SetHead rewinds the local chain to a new head. 171 SetHead(uint64) error 172 } 173 174 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 175 type BlockChain interface { 176 LightChain 177 178 // HasBlock verifies a block's presence in the local chain. 179 HasBlock(common.Hash, uint64) bool 180 181 // HasFastBlock verifies a fast block's presence in the local chain. 182 HasFastBlock(common.Hash, uint64) bool 183 184 // GetBlockByHash retrieves a block from the local chain. 185 GetBlockByHash(common.Hash) *types.Block 186 187 // CurrentBlock retrieves the head block from the local chain. 188 CurrentBlock() *types.Block 189 190 // CurrentFastBlock retrieves the head fast block from the local chain. 191 CurrentFastBlock() *types.Block 192 193 // FastSyncCommitHead directly commits the head block to a certain entity. 194 FastSyncCommitHead(common.Hash) error 195 196 // InsertChain inserts a batch of blocks into the local chain. 197 InsertChain(types.Blocks) (int, error) 198 199 // InsertReceiptChain inserts a batch of receipts into the local chain. 200 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 201 202 // Snapshots returns the blockchain snapshot tree to paused it during sync. 203 Snapshots() *snapshot.Tree 204 } 205 206 // New creates a new downloader to fetch hashes and blocks from remote peers. 207 func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 208 if lightchain == nil { 209 lightchain = chain 210 } 211 dl := &Downloader{ 212 stateDB: stateDb, 213 stateBloom: stateBloom, 214 mux: mux, 215 checkpoint: checkpoint, 216 queue: newQueue(blockCacheMaxItems, blockCacheInitialItems), 217 peers: newPeerSet(), 218 blockchain: chain, 219 lightchain: lightchain, 220 dropPeer: dropPeer, 221 headerCh: make(chan dataPack, 1), 222 bodyCh: make(chan dataPack, 1), 223 receiptCh: make(chan dataPack, 1), 224 bodyWakeCh: make(chan bool, 1), 225 receiptWakeCh: make(chan bool, 1), 226 headerProcCh: make(chan []*types.Header, 1), 227 quitCh: make(chan struct{}), 228 stateCh: make(chan dataPack), 229 SnapSyncer: snap.NewSyncer(stateDb), 230 stateSyncStart: make(chan *stateSync), 231 syncStatsState: stateSyncStats{ 232 processed: rawdb.ReadFastTrieProgress(stateDb), 233 }, 234 trackStateReq: make(chan *stateReq), 235 } 236 go dl.stateFetcher() 237 return dl 238 } 239 240 // Progress retrieves the synchronisation boundaries, specifically the origin 241 // block where synchronisation started at (may have failed/suspended); the block 242 // or header sync is currently at; and the latest known block which the sync targets. 243 // 244 // In addition, during the state download phase of fast synchronisation the number 245 // of processed and the total number of known states are also returned. Otherwise 246 // these are zero. 247 func (d *Downloader) Progress() ethereum.SyncProgress { 248 // Lock the current stats and return the progress 249 d.syncStatsLock.RLock() 250 defer d.syncStatsLock.RUnlock() 251 252 current := uint64(0) 253 mode := d.getMode() 254 switch { 255 case d.blockchain != nil && mode == FullSync: 256 current = d.blockchain.CurrentBlock().NumberU64() 257 case d.blockchain != nil && mode == FastSync: 258 current = d.blockchain.CurrentFastBlock().NumberU64() 259 case d.lightchain != nil: 260 current = d.lightchain.CurrentHeader().Number.Uint64() 261 default: 262 log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", mode) 263 } 264 return ethereum.SyncProgress{ 265 StartingBlock: d.syncStatsChainOrigin, 266 CurrentBlock: current, 267 HighestBlock: d.syncStatsChainHeight, 268 PulledStates: d.syncStatsState.processed, 269 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 270 } 271 } 272 273 // Synchronising returns whether the downloader is currently retrieving blocks. 274 func (d *Downloader) Synchronising() bool { 275 return atomic.LoadInt32(&d.synchronising) > 0 276 } 277 278 // RegisterPeer injects a new download peer into the set of block source to be 279 // used for fetching hashes and blocks from. 280 func (d *Downloader) RegisterPeer(id string, version uint, peer Peer) error { 281 var logger log.Logger 282 if len(id) < 16 { 283 // Tests use short IDs, don't choke on them 284 logger = log.New("peer", id) 285 } else { 286 logger = log.New("peer", id[:8]) 287 } 288 logger.Trace("Registering sync peer") 289 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 290 logger.Error("Failed to register sync peer", "err", err) 291 return err 292 } 293 return nil 294 } 295 296 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 297 func (d *Downloader) RegisterLightPeer(id string, version uint, peer LightPeer) error { 298 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 299 } 300 301 // UnregisterPeer remove a peer from the known list, preventing any action from 302 // the specified peer. An effort is also made to return any pending fetches into 303 // the queue. 304 func (d *Downloader) UnregisterPeer(id string) error { 305 // Unregister the peer from the active peer set and revoke any fetch tasks 306 var logger log.Logger 307 if len(id) < 16 { 308 // Tests use short IDs, don't choke on them 309 logger = log.New("peer", id) 310 } else { 311 logger = log.New("peer", id[:8]) 312 } 313 logger.Trace("Unregistering sync peer") 314 if err := d.peers.Unregister(id); err != nil { 315 logger.Error("Failed to unregister sync peer", "err", err) 316 return err 317 } 318 d.queue.Revoke(id) 319 320 return nil 321 } 322 323 // Synchronise tries to sync up our local block chain with a remote peer, both 324 // adding various sanity checks as well as wrapping it with various log entries. 325 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 326 err := d.synchronise(id, head, td, mode) 327 328 switch err { 329 case nil, errBusy, errCanceled: 330 return err 331 } 332 if errors.Is(err, errInvalidChain) || errors.Is(err, errBadPeer) || errors.Is(err, errTimeout) || 333 errors.Is(err, errStallingPeer) || errors.Is(err, errUnsyncedPeer) || errors.Is(err, errEmptyHeaderSet) || 334 errors.Is(err, errPeersUnavailable) || errors.Is(err, errTooOld) || errors.Is(err, errInvalidAncestor) { 335 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 336 if d.dropPeer == nil { 337 // The dropPeer method is nil when `--copydb` is used for a local copy. 338 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 339 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 340 } else { 341 d.dropPeer(id) 342 } 343 return err 344 } 345 log.Warn("Synchronisation failed, retrying", "err", err) 346 return err 347 } 348 349 // synchronise will select the peer and use it for synchronising. If an empty string is given 350 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 351 // checks fail an error will be returned. This method is synchronous 352 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 353 // Mock out the synchronisation if testing 354 if d.synchroniseMock != nil { 355 return d.synchroniseMock(id, hash) 356 } 357 // Make sure only one goroutine is ever allowed past this point at once 358 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 359 return errBusy 360 } 361 defer atomic.StoreInt32(&d.synchronising, 0) 362 363 // Post a user notification of the sync (only once per session) 364 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 365 log.Info("Block synchronisation started") 366 } 367 // If we are already full syncing, but have a fast-sync bloom filter laying 368 // around, make sure it doesn't use memory any more. This is a special case 369 // when the user attempts to fast sync a new empty network. 370 if mode == FullSync && d.stateBloom != nil { 371 d.stateBloom.Close() 372 } 373 // If snap sync was requested, create the snap scheduler and switch to fast 374 // sync mode. Long term we could drop fast sync or merge the two together, 375 // but until snap becomes prevalent, we should support both. TODO(karalabe). 376 if mode == SnapSync { 377 if !d.snapSync { 378 // Snap sync uses the snapshot namespace to store potentially flakey data until 379 // sync completely heals and finishes. Pause snapshot maintenance in the mean 380 // time to prevent access. 381 if snapshots := d.blockchain.Snapshots(); snapshots != nil { // Only nil in tests 382 snapshots.Disable() 383 } 384 log.Warn("Enabling snapshot sync prototype") 385 d.snapSync = true 386 } 387 mode = FastSync 388 } 389 // Reset the queue, peer set and wake channels to clean any internal leftover state 390 d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems) 391 d.peers.Reset() 392 393 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 394 select { 395 case <-ch: 396 default: 397 } 398 } 399 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 400 for empty := false; !empty; { 401 select { 402 case <-ch: 403 default: 404 empty = true 405 } 406 } 407 } 408 for empty := false; !empty; { 409 select { 410 case <-d.headerProcCh: 411 default: 412 empty = true 413 } 414 } 415 // Create cancel channel for aborting mid-flight and mark the master peer 416 d.cancelLock.Lock() 417 d.cancelCh = make(chan struct{}) 418 d.cancelPeer = id 419 d.cancelLock.Unlock() 420 421 defer d.Cancel() // No matter what, we can't leave the cancel channel open 422 423 // Atomically set the requested sync mode 424 atomic.StoreUint32(&d.mode, uint32(mode)) 425 426 // Retrieve the origin peer and initiate the downloading process 427 p := d.peers.Peer(id) 428 if p == nil { 429 return errUnknownPeer 430 } 431 return d.syncWithPeer(p, hash, td) 432 } 433 434 func (d *Downloader) getMode() SyncMode { 435 return SyncMode(atomic.LoadUint32(&d.mode)) 436 } 437 438 // syncWithPeer starts a block synchronization based on the hash chain from the 439 // specified peer and head hash. 440 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 441 d.mux.Post(StartEvent{}) 442 defer func() { 443 // reset on error 444 if err != nil { 445 d.mux.Post(FailedEvent{err}) 446 } else { 447 latest := d.lightchain.CurrentHeader() 448 d.mux.Post(DoneEvent{latest}) 449 } 450 }() 451 if p.version < eth.ETH65 { 452 return fmt.Errorf("%w: advertized %d < required %d", errTooOld, p.version, eth.ETH65) 453 } 454 mode := d.getMode() 455 456 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", mode) 457 defer func(start time.Time) { 458 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 459 }(time.Now()) 460 461 // Look up the sync boundaries: the common ancestor and the target block 462 latest, pivot, err := d.fetchHead(p) 463 if err != nil { 464 return err 465 } 466 if mode == FastSync && pivot == nil { 467 // If no pivot block was returned, the head is below the min full block 468 // threshold (i.e. new chian). In that case we won't really fast sync 469 // anyway, but still need a valid pivot block to avoid some code hitting 470 // nil panics on an access. 471 pivot = d.blockchain.CurrentBlock().Header() 472 } 473 height := latest.Number.Uint64() 474 475 origin, err := d.findAncestor(p, latest) 476 if err != nil { 477 return err 478 } 479 d.syncStatsLock.Lock() 480 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 481 d.syncStatsChainOrigin = origin 482 } 483 d.syncStatsChainHeight = height 484 d.syncStatsLock.Unlock() 485 486 // Ensure our origin point is below any fast sync pivot point 487 if mode == FastSync { 488 if height <= uint64(fsMinFullBlocks) { 489 origin = 0 490 } else { 491 pivotNumber := pivot.Number.Uint64() 492 if pivotNumber <= origin { 493 origin = pivotNumber - 1 494 } 495 // Write out the pivot into the database so a rollback beyond it will 496 // reenable fast sync 497 rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber) 498 } 499 } 500 d.committed = 1 501 if mode == FastSync && pivot.Number.Uint64() != 0 { 502 d.committed = 0 503 } 504 if mode == FastSync { 505 // Set the ancient data limitation. 506 // If we are running fast sync, all block data older than ancientLimit will be 507 // written to the ancient store. More recent data will be written to the active 508 // database and will wait for the freezer to migrate. 509 // 510 // If there is a checkpoint available, then calculate the ancientLimit through 511 // that. Otherwise calculate the ancient limit through the advertised height 512 // of the remote peer. 513 // 514 // The reason for picking checkpoint first is that a malicious peer can give us 515 // a fake (very high) height, forcing the ancient limit to also be very high. 516 // The peer would start to feed us valid blocks until head, resulting in all of 517 // the blocks might be written into the ancient store. A following mini-reorg 518 // could cause issues. 519 if d.checkpoint != 0 && d.checkpoint > fullMaxForkAncestry+1 { 520 d.ancientLimit = d.checkpoint 521 } else if height > fullMaxForkAncestry+1 { 522 d.ancientLimit = height - fullMaxForkAncestry - 1 523 } else { 524 d.ancientLimit = 0 525 } 526 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 527 528 // If a part of blockchain data has already been written into active store, 529 // disable the ancient style insertion explicitly. 530 if origin >= frozen && frozen != 0 { 531 d.ancientLimit = 0 532 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 533 } else if d.ancientLimit > 0 { 534 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 535 } 536 // Rewind the ancient store and blockchain if reorg happens. 537 if origin+1 < frozen { 538 if err := d.lightchain.SetHead(origin + 1); err != nil { 539 return err 540 } 541 } 542 } 543 // Initiate the sync using a concurrent header and content retrieval algorithm 544 d.queue.Prepare(origin+1, mode) 545 if d.syncInitHook != nil { 546 d.syncInitHook(origin, height) 547 } 548 fetchers := []func() error{ 549 func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved 550 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 551 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 552 func() error { return d.processHeaders(origin+1, td) }, 553 } 554 if mode == FastSync { 555 d.pivotLock.Lock() 556 d.pivotHeader = pivot 557 d.pivotLock.Unlock() 558 559 fetchers = append(fetchers, func() error { return d.processFastSyncContent() }) 560 } else if mode == FullSync { 561 fetchers = append(fetchers, d.processFullSyncContent) 562 } 563 return d.spawnSync(fetchers) 564 } 565 566 // spawnSync runs d.process and all given fetcher functions to completion in 567 // separate goroutines, returning the first error that appears. 568 func (d *Downloader) spawnSync(fetchers []func() error) error { 569 errc := make(chan error, len(fetchers)) 570 d.cancelWg.Add(len(fetchers)) 571 for _, fn := range fetchers { 572 fn := fn 573 go func() { defer d.cancelWg.Done(); errc <- fn() }() 574 } 575 // Wait for the first error, then terminate the others. 576 var err error 577 for i := 0; i < len(fetchers); i++ { 578 if i == len(fetchers)-1 { 579 // Close the queue when all fetchers have exited. 580 // This will cause the block processor to end when 581 // it has processed the queue. 582 d.queue.Close() 583 } 584 if err = <-errc; err != nil && err != errCanceled { 585 break 586 } 587 } 588 d.queue.Close() 589 d.Cancel() 590 return err 591 } 592 593 // cancel aborts all of the operations and resets the queue. However, cancel does 594 // not wait for the running download goroutines to finish. This method should be 595 // used when cancelling the downloads from inside the downloader. 596 func (d *Downloader) cancel() { 597 // Close the current cancel channel 598 d.cancelLock.Lock() 599 defer d.cancelLock.Unlock() 600 601 if d.cancelCh != nil { 602 select { 603 case <-d.cancelCh: 604 // Channel was already closed 605 default: 606 close(d.cancelCh) 607 } 608 } 609 } 610 611 // Cancel aborts all of the operations and waits for all download goroutines to 612 // finish before returning. 613 func (d *Downloader) Cancel() { 614 d.cancel() 615 d.cancelWg.Wait() 616 } 617 618 // Terminate interrupts the downloader, canceling all pending operations. 619 // The downloader cannot be reused after calling Terminate. 620 func (d *Downloader) Terminate() { 621 // Close the termination channel (make sure double close is allowed) 622 d.quitLock.Lock() 623 select { 624 case <-d.quitCh: 625 default: 626 close(d.quitCh) 627 } 628 if d.stateBloom != nil { 629 d.stateBloom.Close() 630 } 631 d.quitLock.Unlock() 632 633 // Cancel any pending download requests 634 d.Cancel() 635 } 636 637 // fetchHead retrieves the head header and prior pivot block (if available) from 638 // a remote peer. 639 func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *types.Header, err error) { 640 p.log.Debug("Retrieving remote chain head") 641 mode := d.getMode() 642 643 // Request the advertised remote head block and wait for the response 644 latest, _ := p.peer.Head() 645 fetch := 1 646 if mode == FastSync { 647 fetch = 2 // head + pivot headers 648 } 649 go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true) 650 651 ttl := d.peers.rates.TargetTimeout() 652 timeout := time.After(ttl) 653 for { 654 select { 655 case <-d.cancelCh: 656 return nil, nil, errCanceled 657 658 case packet := <-d.headerCh: 659 // Discard anything not from the origin peer 660 if packet.PeerId() != p.id { 661 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 662 break 663 } 664 // Make sure the peer gave us at least one and at most the requested headers 665 headers := packet.(*headerPack).headers 666 if len(headers) == 0 || len(headers) > fetch { 667 return nil, nil, fmt.Errorf("%w: returned headers %d != requested %d", errBadPeer, len(headers), fetch) 668 } 669 // The first header needs to be the head, validate against the checkpoint 670 // and request. If only 1 header was returned, make sure there's no pivot 671 // or there was not one requested. 672 head := headers[0] 673 if (mode == FastSync || mode == LightSync) && head.Number.Uint64() < d.checkpoint { 674 return nil, nil, fmt.Errorf("%w: remote head %d below checkpoint %d", errUnsyncedPeer, head.Number, d.checkpoint) 675 } 676 if len(headers) == 1 { 677 if mode == FastSync && head.Number.Uint64() > uint64(fsMinFullBlocks) { 678 return nil, nil, fmt.Errorf("%w: no pivot included along head header", errBadPeer) 679 } 680 p.log.Debug("Remote head identified, no pivot", "number", head.Number, "hash", head.Hash()) 681 return head, nil, nil 682 } 683 // At this point we have 2 headers in total and the first is the 684 // validated head of the chian. Check the pivot number and return, 685 pivot := headers[1] 686 if pivot.Number.Uint64() != head.Number.Uint64()-uint64(fsMinFullBlocks) { 687 return nil, nil, fmt.Errorf("%w: remote pivot %d != requested %d", errInvalidChain, pivot.Number, head.Number.Uint64()-uint64(fsMinFullBlocks)) 688 } 689 return head, pivot, nil 690 691 case <-timeout: 692 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 693 return nil, nil, errTimeout 694 695 case <-d.bodyCh: 696 case <-d.receiptCh: 697 // Out of bounds delivery, ignore 698 } 699 } 700 } 701 702 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 703 // common ancestor. 704 // It returns parameters to be used for peer.RequestHeadersByNumber: 705 // from - starting block number 706 // count - number of headers to request 707 // skip - number of headers to skip 708 // and also returns 'max', the last block which is expected to be returned by the remote peers, 709 // given the (from,count,skip) 710 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 711 var ( 712 from int 713 count int 714 MaxCount = MaxHeaderFetch / 16 715 ) 716 // requestHead is the highest block that we will ask for. If requestHead is not offset, 717 // the highest block that we will get is 16 blocks back from head, which means we 718 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 719 // between us and the peer is 1-2 blocks, which is most common 720 requestHead := int(remoteHeight) - 1 721 if requestHead < 0 { 722 requestHead = 0 723 } 724 // requestBottom is the lowest block we want included in the query 725 // Ideally, we want to include the one just below our own head 726 requestBottom := int(localHeight - 1) 727 if requestBottom < 0 { 728 requestBottom = 0 729 } 730 totalSpan := requestHead - requestBottom 731 span := 1 + totalSpan/MaxCount 732 if span < 2 { 733 span = 2 734 } 735 if span > 16 { 736 span = 16 737 } 738 739 count = 1 + totalSpan/span 740 if count > MaxCount { 741 count = MaxCount 742 } 743 if count < 2 { 744 count = 2 745 } 746 from = requestHead - (count-1)*span 747 if from < 0 { 748 from = 0 749 } 750 max := from + (count-1)*span 751 return int64(from), count, span - 1, uint64(max) 752 } 753 754 // findAncestor tries to locate the common ancestor link of the local chain and 755 // a remote peers blockchain. In the general case when our node was in sync and 756 // on the correct chain, checking the top N links should already get us a match. 757 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 758 // the head links match), we do a binary search to find the common ancestor. 759 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 760 // Figure out the valid ancestor range to prevent rewrite attacks 761 var ( 762 floor = int64(-1) 763 localHeight uint64 764 remoteHeight = remoteHeader.Number.Uint64() 765 ) 766 mode := d.getMode() 767 switch mode { 768 case FullSync: 769 localHeight = d.blockchain.CurrentBlock().NumberU64() 770 case FastSync: 771 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 772 default: 773 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 774 } 775 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 776 777 // Recap floor value for binary search 778 maxForkAncestry := fullMaxForkAncestry 779 if d.getMode() == LightSync { 780 maxForkAncestry = lightMaxForkAncestry 781 } 782 if localHeight >= maxForkAncestry { 783 // We're above the max reorg threshold, find the earliest fork point 784 floor = int64(localHeight - maxForkAncestry) 785 } 786 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 787 // all headers before that point will be missing. 788 if mode == LightSync { 789 // If we don't know the current CHT position, find it 790 if d.genesis == 0 { 791 header := d.lightchain.CurrentHeader() 792 for header != nil { 793 d.genesis = header.Number.Uint64() 794 if floor >= int64(d.genesis)-1 { 795 break 796 } 797 header = d.lightchain.GetHeaderByHash(header.ParentHash) 798 } 799 } 800 // We already know the "genesis" block number, cap floor to that 801 if floor < int64(d.genesis)-1 { 802 floor = int64(d.genesis) - 1 803 } 804 } 805 806 ancestor, err := d.findAncestorSpanSearch(p, mode, remoteHeight, localHeight, floor) 807 if err == nil { 808 return ancestor, nil 809 } 810 // The returned error was not nil. 811 // If the error returned does not reflect that a common ancestor was not found, return it. 812 // If the error reflects that a common ancestor was not found, continue to binary search, 813 // where the error value will be reassigned. 814 if !errors.Is(err, errNoAncestorFound) { 815 return 0, err 816 } 817 818 ancestor, err = d.findAncestorBinarySearch(p, mode, remoteHeight, floor) 819 if err != nil { 820 return 0, err 821 } 822 return ancestor, nil 823 } 824 825 func (d *Downloader) findAncestorSpanSearch(p *peerConnection, mode SyncMode, remoteHeight, localHeight uint64, floor int64) (commonAncestor uint64, err error) { 826 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 827 828 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 829 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 830 831 // Wait for the remote response to the head fetch 832 number, hash := uint64(0), common.Hash{} 833 834 ttl := d.peers.rates.TargetTimeout() 835 timeout := time.After(ttl) 836 837 for finished := false; !finished; { 838 select { 839 case <-d.cancelCh: 840 return 0, errCanceled 841 842 case packet := <-d.headerCh: 843 // Discard anything not from the origin peer 844 if packet.PeerId() != p.id { 845 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 846 break 847 } 848 // Make sure the peer actually gave something valid 849 headers := packet.(*headerPack).headers 850 if len(headers) == 0 { 851 p.log.Warn("Empty head header set") 852 return 0, errEmptyHeaderSet 853 } 854 // Make sure the peer's reply conforms to the request 855 for i, header := range headers { 856 expectNumber := from + int64(i)*int64(skip+1) 857 if number := header.Number.Int64(); number != expectNumber { 858 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 859 return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering")) 860 } 861 } 862 // Check if a common ancestor was found 863 finished = true 864 for i := len(headers) - 1; i >= 0; i-- { 865 // Skip any headers that underflow/overflow our requested set 866 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 867 continue 868 } 869 // Otherwise check if we already know the header or not 870 h := headers[i].Hash() 871 n := headers[i].Number.Uint64() 872 873 var known bool 874 switch mode { 875 case FullSync: 876 known = d.blockchain.HasBlock(h, n) 877 case FastSync: 878 known = d.blockchain.HasFastBlock(h, n) 879 default: 880 known = d.lightchain.HasHeader(h, n) 881 } 882 if known { 883 number, hash = n, h 884 break 885 } 886 } 887 888 case <-timeout: 889 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 890 return 0, errTimeout 891 892 case <-d.bodyCh: 893 case <-d.receiptCh: 894 // Out of bounds delivery, ignore 895 } 896 } 897 // If the head fetch already found an ancestor, return 898 if hash != (common.Hash{}) { 899 if int64(number) <= floor { 900 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 901 return 0, errInvalidAncestor 902 } 903 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 904 return number, nil 905 } 906 return 0, errNoAncestorFound 907 } 908 909 func (d *Downloader) findAncestorBinarySearch(p *peerConnection, mode SyncMode, remoteHeight uint64, floor int64) (commonAncestor uint64, err error) { 910 hash := common.Hash{} 911 912 // Ancestor not found, we need to binary search over our chain 913 start, end := uint64(0), remoteHeight 914 if floor > 0 { 915 start = uint64(floor) 916 } 917 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 918 919 for start+1 < end { 920 // Split our chain interval in two, and request the hash to cross check 921 check := (start + end) / 2 922 923 ttl := d.peers.rates.TargetTimeout() 924 timeout := time.After(ttl) 925 926 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 927 928 // Wait until a reply arrives to this request 929 for arrived := false; !arrived; { 930 select { 931 case <-d.cancelCh: 932 return 0, errCanceled 933 934 case packet := <-d.headerCh: 935 // Discard anything not from the origin peer 936 if packet.PeerId() != p.id { 937 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 938 break 939 } 940 // Make sure the peer actually gave something valid 941 headers := packet.(*headerPack).headers 942 if len(headers) != 1 { 943 p.log.Warn("Multiple headers for single request", "headers", len(headers)) 944 return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers)) 945 } 946 arrived = true 947 948 // Modify the search interval based on the response 949 h := headers[0].Hash() 950 n := headers[0].Number.Uint64() 951 952 var known bool 953 switch mode { 954 case FullSync: 955 known = d.blockchain.HasBlock(h, n) 956 case FastSync: 957 known = d.blockchain.HasFastBlock(h, n) 958 default: 959 known = d.lightchain.HasHeader(h, n) 960 } 961 if !known { 962 end = check 963 break 964 } 965 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 966 if header.Number.Uint64() != check { 967 p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 968 return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number) 969 } 970 start = check 971 hash = h 972 973 case <-timeout: 974 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 975 return 0, errTimeout 976 977 case <-d.bodyCh: 978 case <-d.receiptCh: 979 // Out of bounds delivery, ignore 980 } 981 } 982 } 983 // Ensure valid ancestry and return 984 if int64(start) <= floor { 985 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 986 return 0, errInvalidAncestor 987 } 988 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 989 return start, nil 990 } 991 992 // fetchHeaders keeps retrieving headers concurrently from the number 993 // requested, until no more are returned, potentially throttling on the way. To 994 // facilitate concurrency but still protect against malicious nodes sending bad 995 // headers, we construct a header chain skeleton using the "origin" peer we are 996 // syncing with, and fill in the missing headers using anyone else. Headers from 997 // other peers are only accepted if they map cleanly to the skeleton. If no one 998 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 999 // the origin is dropped. 1000 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error { 1001 p.log.Debug("Directing header downloads", "origin", from) 1002 defer p.log.Debug("Header download terminated") 1003 1004 // Create a timeout timer, and the associated header fetcher 1005 skeleton := true // Skeleton assembly phase or finishing up 1006 pivoting := false // Whether the next request is pivot verification 1007 request := time.Now() // time of the last skeleton fetch request 1008 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 1009 <-timeout.C // timeout channel should be initially empty 1010 defer timeout.Stop() 1011 1012 var ttl time.Duration 1013 getHeaders := func(from uint64) { 1014 request = time.Now() 1015 1016 ttl = d.peers.rates.TargetTimeout() 1017 timeout.Reset(ttl) 1018 1019 if skeleton { 1020 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 1021 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 1022 } else { 1023 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 1024 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 1025 } 1026 } 1027 getNextPivot := func() { 1028 pivoting = true 1029 request = time.Now() 1030 1031 ttl = d.peers.rates.TargetTimeout() 1032 timeout.Reset(ttl) 1033 1034 d.pivotLock.RLock() 1035 pivot := d.pivotHeader.Number.Uint64() 1036 d.pivotLock.RUnlock() 1037 1038 p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks)) 1039 go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep 1040 } 1041 // Start pulling the header chain skeleton until all is done 1042 ancestor := from 1043 getHeaders(from) 1044 1045 mode := d.getMode() 1046 for { 1047 select { 1048 case <-d.cancelCh: 1049 return errCanceled 1050 1051 case packet := <-d.headerCh: 1052 // Make sure the active peer is giving us the skeleton headers 1053 if packet.PeerId() != p.id { 1054 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 1055 break 1056 } 1057 headerReqTimer.UpdateSince(request) 1058 timeout.Stop() 1059 1060 // If the pivot is being checked, move if it became stale and run the real retrieval 1061 var pivot uint64 1062 1063 d.pivotLock.RLock() 1064 if d.pivotHeader != nil { 1065 pivot = d.pivotHeader.Number.Uint64() 1066 } 1067 d.pivotLock.RUnlock() 1068 1069 if pivoting { 1070 if packet.Items() == 2 { 1071 // Retrieve the headers and do some sanity checks, just in case 1072 headers := packet.(*headerPack).headers 1073 1074 if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want { 1075 log.Warn("Peer sent invalid next pivot", "have", have, "want", want) 1076 return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want) 1077 } 1078 if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want { 1079 log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want) 1080 return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want) 1081 } 1082 log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number) 1083 pivot = headers[0].Number.Uint64() 1084 1085 d.pivotLock.Lock() 1086 d.pivotHeader = headers[0] 1087 d.pivotLock.Unlock() 1088 1089 // Write out the pivot into the database so a rollback beyond 1090 // it will reenable fast sync and update the state root that 1091 // the state syncer will be downloading. 1092 rawdb.WriteLastPivotNumber(d.stateDB, pivot) 1093 } 1094 pivoting = false 1095 getHeaders(from) 1096 continue 1097 } 1098 // If the skeleton's finished, pull any remaining head headers directly from the origin 1099 if skeleton && packet.Items() == 0 { 1100 skeleton = false 1101 getHeaders(from) 1102 continue 1103 } 1104 // If no more headers are inbound, notify the content fetchers and return 1105 if packet.Items() == 0 { 1106 // Don't abort header fetches while the pivot is downloading 1107 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 1108 p.log.Debug("No headers, waiting for pivot commit") 1109 select { 1110 case <-time.After(fsHeaderContCheck): 1111 getHeaders(from) 1112 continue 1113 case <-d.cancelCh: 1114 return errCanceled 1115 } 1116 } 1117 // Pivot done (or not in fast sync) and no more headers, terminate the process 1118 p.log.Debug("No more headers available") 1119 select { 1120 case d.headerProcCh <- nil: 1121 return nil 1122 case <-d.cancelCh: 1123 return errCanceled 1124 } 1125 } 1126 headers := packet.(*headerPack).headers 1127 1128 // If we received a skeleton batch, resolve internals concurrently 1129 if skeleton { 1130 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1131 if err != nil { 1132 p.log.Debug("Skeleton chain invalid", "err", err) 1133 return fmt.Errorf("%w: %v", errInvalidChain, err) 1134 } 1135 headers = filled[proced:] 1136 from += uint64(proced) 1137 } else { 1138 // If we're closing in on the chain head, but haven't yet reached it, delay 1139 // the last few headers so mini reorgs on the head don't cause invalid hash 1140 // chain errors. 1141 if n := len(headers); n > 0 { 1142 // Retrieve the current head we're at 1143 var head uint64 1144 if mode == LightSync { 1145 head = d.lightchain.CurrentHeader().Number.Uint64() 1146 } else { 1147 head = d.blockchain.CurrentFastBlock().NumberU64() 1148 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1149 head = full 1150 } 1151 } 1152 // If the head is below the common ancestor, we're actually deduplicating 1153 // already existing chain segments, so use the ancestor as the fake head. 1154 // Otherwise we might end up delaying header deliveries pointlessly. 1155 if head < ancestor { 1156 head = ancestor 1157 } 1158 // If the head is way older than this batch, delay the last few headers 1159 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1160 delay := reorgProtHeaderDelay 1161 if delay > n { 1162 delay = n 1163 } 1164 headers = headers[:n-delay] 1165 } 1166 } 1167 } 1168 // Insert all the new headers and fetch the next batch 1169 if len(headers) > 0 { 1170 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1171 select { 1172 case d.headerProcCh <- headers: 1173 case <-d.cancelCh: 1174 return errCanceled 1175 } 1176 from += uint64(len(headers)) 1177 1178 // If we're still skeleton filling fast sync, check pivot staleness 1179 // before continuing to the next skeleton filling 1180 if skeleton && pivot > 0 { 1181 getNextPivot() 1182 } else { 1183 getHeaders(from) 1184 } 1185 } else { 1186 // No headers delivered, or all of them being delayed, sleep a bit and retry 1187 p.log.Trace("All headers delayed, waiting") 1188 select { 1189 case <-time.After(fsHeaderContCheck): 1190 getHeaders(from) 1191 continue 1192 case <-d.cancelCh: 1193 return errCanceled 1194 } 1195 } 1196 1197 case <-timeout.C: 1198 if d.dropPeer == nil { 1199 // The dropPeer method is nil when `--copydb` is used for a local copy. 1200 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1201 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1202 break 1203 } 1204 // Header retrieval timed out, consider the peer bad and drop 1205 p.log.Debug("Header request timed out", "elapsed", ttl) 1206 headerTimeoutMeter.Mark(1) 1207 d.dropPeer(p.id) 1208 1209 // Finish the sync gracefully instead of dumping the gathered data though 1210 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1211 select { 1212 case ch <- false: 1213 case <-d.cancelCh: 1214 } 1215 } 1216 select { 1217 case d.headerProcCh <- nil: 1218 case <-d.cancelCh: 1219 } 1220 return fmt.Errorf("%w: header request timed out", errBadPeer) 1221 } 1222 } 1223 } 1224 1225 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1226 // and maps them to the provided skeleton header chain. 1227 // 1228 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1229 // immediately to the header processor to keep the rest of the pipeline full even 1230 // in the case of header stalls. 1231 // 1232 // The method returns the entire filled skeleton and also the number of headers 1233 // already forwarded for processing. 1234 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1235 log.Debug("Filling up skeleton", "from", from) 1236 d.queue.ScheduleSkeleton(from, skeleton) 1237 1238 var ( 1239 deliver = func(packet dataPack) (int, error) { 1240 pack := packet.(*headerPack) 1241 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1242 } 1243 expire = func() map[string]int { return d.queue.ExpireHeaders(d.peers.rates.TargetTimeout()) } 1244 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) { 1245 return d.queue.ReserveHeaders(p, count), false, false 1246 } 1247 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1248 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.peers.rates.TargetRoundTrip()) } 1249 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1250 p.SetHeadersIdle(accepted, deliveryTime) 1251 } 1252 ) 1253 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1254 d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve, 1255 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1256 1257 log.Debug("Skeleton fill terminated", "err", err) 1258 1259 filled, proced := d.queue.RetrieveHeaders() 1260 return filled, proced, err 1261 } 1262 1263 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1264 // available peers, reserving a chunk of blocks for each, waiting for delivery 1265 // and also periodically checking for timeouts. 1266 func (d *Downloader) fetchBodies(from uint64) error { 1267 log.Debug("Downloading block bodies", "origin", from) 1268 1269 var ( 1270 deliver = func(packet dataPack) (int, error) { 1271 pack := packet.(*bodyPack) 1272 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1273 } 1274 expire = func() map[string]int { return d.queue.ExpireBodies(d.peers.rates.TargetTimeout()) } 1275 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1276 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.peers.rates.TargetRoundTrip()) } 1277 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) } 1278 ) 1279 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1280 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies, 1281 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1282 1283 log.Debug("Block body download terminated", "err", err) 1284 return err 1285 } 1286 1287 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1288 // available peers, reserving a chunk of receipts for each, waiting for delivery 1289 // and also periodically checking for timeouts. 1290 func (d *Downloader) fetchReceipts(from uint64) error { 1291 log.Debug("Downloading transaction receipts", "origin", from) 1292 1293 var ( 1294 deliver = func(packet dataPack) (int, error) { 1295 pack := packet.(*receiptPack) 1296 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1297 } 1298 expire = func() map[string]int { return d.queue.ExpireReceipts(d.peers.rates.TargetTimeout()) } 1299 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1300 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.peers.rates.TargetRoundTrip()) } 1301 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1302 p.SetReceiptsIdle(accepted, deliveryTime) 1303 } 1304 ) 1305 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1306 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts, 1307 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1308 1309 log.Debug("Transaction receipt download terminated", "err", err) 1310 return err 1311 } 1312 1313 // fetchParts iteratively downloads scheduled block parts, taking any available 1314 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1315 // also periodically checking for timeouts. 1316 // 1317 // As the scheduling/timeout logic mostly is the same for all downloaded data 1318 // types, this method is used by each for data gathering and is instrumented with 1319 // various callbacks to handle the slight differences between processing them. 1320 // 1321 // The instrumentation parameters: 1322 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1323 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1324 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1325 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1326 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1327 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1328 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1329 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1330 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1331 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1332 // - fetch: network callback to actually send a particular download request to a physical remote peer 1333 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1334 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1335 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1336 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1337 // - kind: textual label of the type being downloaded to display in log messages 1338 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1339 expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool), 1340 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1341 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error { 1342 1343 // Create a ticker to detect expired retrieval tasks 1344 ticker := time.NewTicker(100 * time.Millisecond) 1345 defer ticker.Stop() 1346 1347 update := make(chan struct{}, 1) 1348 1349 // Prepare the queue and fetch block parts until the block header fetcher's done 1350 finished := false 1351 for { 1352 select { 1353 case <-d.cancelCh: 1354 return errCanceled 1355 1356 case packet := <-deliveryCh: 1357 deliveryTime := time.Now() 1358 // If the peer was previously banned and failed to deliver its pack 1359 // in a reasonable time frame, ignore its message. 1360 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1361 // Deliver the received chunk of data and check chain validity 1362 accepted, err := deliver(packet) 1363 if errors.Is(err, errInvalidChain) { 1364 return err 1365 } 1366 // Unless a peer delivered something completely else than requested (usually 1367 // caused by a timed out request which came through in the end), set it to 1368 // idle. If the delivery's stale, the peer should have already been idled. 1369 if !errors.Is(err, errStaleDelivery) { 1370 setIdle(peer, accepted, deliveryTime) 1371 } 1372 // Issue a log to the user to see what's going on 1373 switch { 1374 case err == nil && packet.Items() == 0: 1375 peer.log.Trace("Requested data not delivered", "type", kind) 1376 case err == nil: 1377 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1378 default: 1379 peer.log.Debug("Failed to deliver retrieved data", "type", kind, "err", err) 1380 } 1381 } 1382 // Blocks assembled, try to update the progress 1383 select { 1384 case update <- struct{}{}: 1385 default: 1386 } 1387 1388 case cont := <-wakeCh: 1389 // The header fetcher sent a continuation flag, check if it's done 1390 if !cont { 1391 finished = true 1392 } 1393 // Headers arrive, try to update the progress 1394 select { 1395 case update <- struct{}{}: 1396 default: 1397 } 1398 1399 case <-ticker.C: 1400 // Sanity check update the progress 1401 select { 1402 case update <- struct{}{}: 1403 default: 1404 } 1405 1406 case <-update: 1407 // Short circuit if we lost all our peers 1408 if d.peers.Len() == 0 { 1409 return errNoPeers 1410 } 1411 // Check for fetch request timeouts and demote the responsible peers 1412 for pid, fails := range expire() { 1413 if peer := d.peers.Peer(pid); peer != nil { 1414 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1415 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1416 // out that sync wise we need to get rid of the peer. 1417 // 1418 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1419 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1420 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1421 if fails > 2 { 1422 peer.log.Trace("Data delivery timed out", "type", kind) 1423 setIdle(peer, 0, time.Now()) 1424 } else { 1425 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1426 1427 if d.dropPeer == nil { 1428 // The dropPeer method is nil when `--copydb` is used for a local copy. 1429 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1430 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1431 } else { 1432 d.dropPeer(pid) 1433 1434 // If this peer was the master peer, abort sync immediately 1435 d.cancelLock.RLock() 1436 master := pid == d.cancelPeer 1437 d.cancelLock.RUnlock() 1438 1439 if master { 1440 d.cancel() 1441 return errTimeout 1442 } 1443 } 1444 } 1445 } 1446 } 1447 // If there's nothing more to fetch, wait or terminate 1448 if pending() == 0 { 1449 if !inFlight() && finished { 1450 log.Debug("Data fetching completed", "type", kind) 1451 return nil 1452 } 1453 break 1454 } 1455 // Send a download request to all idle peers, until throttled 1456 progressed, throttled, running := false, false, inFlight() 1457 idles, total := idle() 1458 pendCount := pending() 1459 for _, peer := range idles { 1460 // Short circuit if throttling activated 1461 if throttled { 1462 break 1463 } 1464 // Short circuit if there is no more available task. 1465 if pendCount = pending(); pendCount == 0 { 1466 break 1467 } 1468 // Reserve a chunk of fetches for a peer. A nil can mean either that 1469 // no more headers are available, or that the peer is known not to 1470 // have them. 1471 request, progress, throttle := reserve(peer, capacity(peer)) 1472 if progress { 1473 progressed = true 1474 } 1475 if throttle { 1476 throttled = true 1477 throttleCounter.Inc(1) 1478 } 1479 if request == nil { 1480 continue 1481 } 1482 if request.From > 0 { 1483 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1484 } else { 1485 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1486 } 1487 // Fetch the chunk and make sure any errors return the hashes to the queue 1488 if fetchHook != nil { 1489 fetchHook(request.Headers) 1490 } 1491 if err := fetch(peer, request); err != nil { 1492 // Although we could try and make an attempt to fix this, this error really 1493 // means that we've double allocated a fetch task to a peer. If that is the 1494 // case, the internal state of the downloader and the queue is very wrong so 1495 // better hard crash and note the error instead of silently accumulating into 1496 // a much bigger issue. 1497 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1498 } 1499 running = true 1500 } 1501 // Make sure that we have peers available for fetching. If all peers have been tried 1502 // and all failed throw an error 1503 if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 { 1504 return errPeersUnavailable 1505 } 1506 } 1507 } 1508 } 1509 1510 // processHeaders takes batches of retrieved headers from an input channel and 1511 // keeps processing and scheduling them into the header chain and downloader's 1512 // queue until the stream ends or a failure occurs. 1513 func (d *Downloader) processHeaders(origin uint64, td *big.Int) error { 1514 // Keep a count of uncertain headers to roll back 1515 var ( 1516 rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis) 1517 rollbackErr error 1518 mode = d.getMode() 1519 ) 1520 defer func() { 1521 if rollback > 0 { 1522 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1523 if mode != LightSync { 1524 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1525 lastBlock = d.blockchain.CurrentBlock().Number() 1526 } 1527 if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block 1528 // We're already unwinding the stack, only print the error to make it more visible 1529 log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err) 1530 } 1531 curFastBlock, curBlock := common.Big0, common.Big0 1532 if mode != LightSync { 1533 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1534 curBlock = d.blockchain.CurrentBlock().Number() 1535 } 1536 log.Warn("Rolled back chain segment", 1537 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1538 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1539 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr) 1540 } 1541 }() 1542 // Wait for batches of headers to process 1543 gotHeaders := false 1544 1545 for { 1546 select { 1547 case <-d.cancelCh: 1548 rollbackErr = errCanceled 1549 return errCanceled 1550 1551 case headers := <-d.headerProcCh: 1552 // Terminate header processing if we synced up 1553 if len(headers) == 0 { 1554 // Notify everyone that headers are fully processed 1555 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1556 select { 1557 case ch <- false: 1558 case <-d.cancelCh: 1559 } 1560 } 1561 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1562 // better chain compared to ours. The only exception is if its promised blocks were 1563 // already imported by other means (e.g. fetcher): 1564 // 1565 // R <remote peer>, L <local node>: Both at block 10 1566 // R: Mine block 11, and propagate it to L 1567 // L: Queue block 11 for import 1568 // L: Notice that R's head and TD increased compared to ours, start sync 1569 // L: Import of block 11 finishes 1570 // L: Sync begins, and finds common ancestor at 11 1571 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1572 // R: Nothing to give 1573 if mode != LightSync { 1574 head := d.blockchain.CurrentBlock() 1575 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1576 return errStallingPeer 1577 } 1578 } 1579 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1580 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1581 // of delivering the post-pivot blocks that would flag the invalid content. 1582 // 1583 // This check cannot be executed "as is" for full imports, since blocks may still be 1584 // queued for processing when the header download completes. However, as long as the 1585 // peer gave us something useful, we're already happy/progressed (above check). 1586 if mode == FastSync || mode == LightSync { 1587 head := d.lightchain.CurrentHeader() 1588 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1589 return errStallingPeer 1590 } 1591 } 1592 // Disable any rollback and return 1593 rollback = 0 1594 return nil 1595 } 1596 // Otherwise split the chunk of headers into batches and process them 1597 gotHeaders = true 1598 for len(headers) > 0 { 1599 // Terminate if something failed in between processing chunks 1600 select { 1601 case <-d.cancelCh: 1602 rollbackErr = errCanceled 1603 return errCanceled 1604 default: 1605 } 1606 // Select the next chunk of headers to import 1607 limit := maxHeadersProcess 1608 if limit > len(headers) { 1609 limit = len(headers) 1610 } 1611 chunk := headers[:limit] 1612 1613 // In case of header only syncing, validate the chunk immediately 1614 if mode == FastSync || mode == LightSync { 1615 // If we're importing pure headers, verify based on their recentness 1616 var pivot uint64 1617 1618 d.pivotLock.RLock() 1619 if d.pivotHeader != nil { 1620 pivot = d.pivotHeader.Number.Uint64() 1621 } 1622 d.pivotLock.RUnlock() 1623 1624 frequency := fsHeaderCheckFrequency 1625 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1626 frequency = 1 1627 } 1628 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1629 rollbackErr = err 1630 1631 // If some headers were inserted, track them as uncertain 1632 if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 { 1633 rollback = chunk[0].Number.Uint64() 1634 } 1635 log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err) 1636 return fmt.Errorf("%w: %v", errInvalidChain, err) 1637 } 1638 // All verifications passed, track all headers within the alloted limits 1639 if mode == FastSync { 1640 head := chunk[len(chunk)-1].Number.Uint64() 1641 if head-rollback > uint64(fsHeaderSafetyNet) { 1642 rollback = head - uint64(fsHeaderSafetyNet) 1643 } else { 1644 rollback = 1 1645 } 1646 } 1647 } 1648 // Unless we're doing light chains, schedule the headers for associated content retrieval 1649 if mode == FullSync || mode == FastSync { 1650 // If we've reached the allowed number of pending headers, stall a bit 1651 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1652 select { 1653 case <-d.cancelCh: 1654 rollbackErr = errCanceled 1655 return errCanceled 1656 case <-time.After(time.Second): 1657 } 1658 } 1659 // Otherwise insert the headers for content retrieval 1660 inserts := d.queue.Schedule(chunk, origin) 1661 if len(inserts) != len(chunk) { 1662 rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk)) 1663 return fmt.Errorf("%w: stale headers", errBadPeer) 1664 } 1665 } 1666 headers = headers[limit:] 1667 origin += uint64(limit) 1668 } 1669 // Update the highest block number we know if a higher one is found. 1670 d.syncStatsLock.Lock() 1671 if d.syncStatsChainHeight < origin { 1672 d.syncStatsChainHeight = origin - 1 1673 } 1674 d.syncStatsLock.Unlock() 1675 1676 // Signal the content downloaders of the availablility of new tasks 1677 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1678 select { 1679 case ch <- true: 1680 default: 1681 } 1682 } 1683 } 1684 } 1685 } 1686 1687 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1688 func (d *Downloader) processFullSyncContent() error { 1689 for { 1690 results := d.queue.Results(true) 1691 if len(results) == 0 { 1692 return nil 1693 } 1694 if d.chainInsertHook != nil { 1695 d.chainInsertHook(results) 1696 } 1697 if err := d.importBlockResults(results); err != nil { 1698 return err 1699 } 1700 } 1701 } 1702 1703 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1704 // Check for any early termination requests 1705 if len(results) == 0 { 1706 return nil 1707 } 1708 select { 1709 case <-d.quitCh: 1710 return errCancelContentProcessing 1711 default: 1712 } 1713 // Retrieve the a batch of results to import 1714 first, last := results[0].Header, results[len(results)-1].Header 1715 log.Debug("Inserting downloaded chain", "items", len(results), 1716 "firstnum", first.Number, "firsthash", first.Hash(), 1717 "lastnum", last.Number, "lasthash", last.Hash(), 1718 ) 1719 blocks := make([]*types.Block, len(results)) 1720 for i, result := range results { 1721 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1722 } 1723 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1724 if index < len(results) { 1725 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1726 } else { 1727 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1728 // when it needs to preprocess blocks to import a sidechain. 1729 // The importer will put together a new list of blocks to import, which is a superset 1730 // of the blocks delivered from the downloader, and the indexing will be off. 1731 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1732 } 1733 return fmt.Errorf("%w: %v", errInvalidChain, err) 1734 } 1735 return nil 1736 } 1737 1738 // processFastSyncContent takes fetch results from the queue and writes them to the 1739 // database. It also controls the synchronisation of state nodes of the pivot block. 1740 func (d *Downloader) processFastSyncContent() error { 1741 // Start syncing state of the reported head block. This should get us most of 1742 // the state of the pivot block. 1743 d.pivotLock.RLock() 1744 sync := d.syncState(d.pivotHeader.Root) 1745 d.pivotLock.RUnlock() 1746 1747 defer func() { 1748 // The `sync` object is replaced every time the pivot moves. We need to 1749 // defer close the very last active one, hence the lazy evaluation vs. 1750 // calling defer sync.Cancel() !!! 1751 sync.Cancel() 1752 }() 1753 1754 closeOnErr := func(s *stateSync) { 1755 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled && err != snap.ErrCancelled { 1756 d.queue.Close() // wake up Results 1757 } 1758 } 1759 go closeOnErr(sync) 1760 1761 // To cater for moving pivot points, track the pivot block and subsequently 1762 // accumulated download results separately. 1763 var ( 1764 oldPivot *fetchResult // Locked in pivot block, might change eventually 1765 oldTail []*fetchResult // Downloaded content after the pivot 1766 ) 1767 for { 1768 // Wait for the next batch of downloaded data to be available, and if the pivot 1769 // block became stale, move the goalpost 1770 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1771 if len(results) == 0 { 1772 // If pivot sync is done, stop 1773 if oldPivot == nil { 1774 return sync.Cancel() 1775 } 1776 // If sync failed, stop 1777 select { 1778 case <-d.cancelCh: 1779 sync.Cancel() 1780 return errCanceled 1781 default: 1782 } 1783 } 1784 if d.chainInsertHook != nil { 1785 d.chainInsertHook(results) 1786 } 1787 // If we haven't downloaded the pivot block yet, check pivot staleness 1788 // notifications from the header downloader 1789 d.pivotLock.RLock() 1790 pivot := d.pivotHeader 1791 d.pivotLock.RUnlock() 1792 1793 if oldPivot == nil { 1794 if pivot.Root != sync.root { 1795 sync.Cancel() 1796 sync = d.syncState(pivot.Root) 1797 1798 go closeOnErr(sync) 1799 } 1800 } else { 1801 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1802 } 1803 // Split around the pivot block and process the two sides via fast/full sync 1804 if atomic.LoadInt32(&d.committed) == 0 { 1805 latest := results[len(results)-1].Header 1806 // If the height is above the pivot block by 2 sets, it means the pivot 1807 // become stale in the network and it was garbage collected, move to a 1808 // new pivot. 1809 // 1810 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 1811 // need to be taken into account, otherwise we're detecting the pivot move 1812 // late and will drop peers due to unavailable state!!! 1813 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 1814 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 1815 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 1816 1817 d.pivotLock.Lock() 1818 d.pivotHeader = pivot 1819 d.pivotLock.Unlock() 1820 1821 // Write out the pivot into the database so a rollback beyond it will 1822 // reenable fast sync 1823 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 1824 } 1825 } 1826 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 1827 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1828 return err 1829 } 1830 if P != nil { 1831 // If new pivot block found, cancel old state retrieval and restart 1832 if oldPivot != P { 1833 sync.Cancel() 1834 sync = d.syncState(P.Header.Root) 1835 1836 go closeOnErr(sync) 1837 oldPivot = P 1838 } 1839 // Wait for completion, occasionally checking for pivot staleness 1840 select { 1841 case <-sync.done: 1842 if sync.err != nil { 1843 return sync.err 1844 } 1845 if err := d.commitPivotBlock(P); err != nil { 1846 return err 1847 } 1848 oldPivot = nil 1849 1850 case <-time.After(time.Second): 1851 oldTail = afterP 1852 continue 1853 } 1854 } 1855 // Fast sync done, pivot commit done, full import 1856 if err := d.importBlockResults(afterP); err != nil { 1857 return err 1858 } 1859 } 1860 } 1861 1862 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1863 if len(results) == 0 { 1864 return nil, nil, nil 1865 } 1866 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 1867 // the pivot is somewhere in the future 1868 return nil, results, nil 1869 } 1870 // This can also be optimized, but only happens very seldom 1871 for _, result := range results { 1872 num := result.Header.Number.Uint64() 1873 switch { 1874 case num < pivot: 1875 before = append(before, result) 1876 case num == pivot: 1877 p = result 1878 default: 1879 after = append(after, result) 1880 } 1881 } 1882 return p, before, after 1883 } 1884 1885 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1886 // Check for any early termination requests 1887 if len(results) == 0 { 1888 return nil 1889 } 1890 select { 1891 case <-d.quitCh: 1892 return errCancelContentProcessing 1893 case <-stateSync.done: 1894 if err := stateSync.Wait(); err != nil { 1895 return err 1896 } 1897 default: 1898 } 1899 // Retrieve the a batch of results to import 1900 first, last := results[0].Header, results[len(results)-1].Header 1901 log.Debug("Inserting fast-sync blocks", "items", len(results), 1902 "firstnum", first.Number, "firsthash", first.Hash(), 1903 "lastnumn", last.Number, "lasthash", last.Hash(), 1904 ) 1905 blocks := make([]*types.Block, len(results)) 1906 receipts := make([]types.Receipts, len(results)) 1907 for i, result := range results { 1908 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1909 receipts[i] = result.Receipts 1910 } 1911 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1912 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1913 return fmt.Errorf("%w: %v", errInvalidChain, err) 1914 } 1915 return nil 1916 } 1917 1918 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1919 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1920 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1921 1922 // Commit the pivot block as the new head, will require full sync from here on 1923 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1924 return err 1925 } 1926 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1927 return err 1928 } 1929 atomic.StoreInt32(&d.committed, 1) 1930 1931 // If we had a bloom filter for the state sync, deallocate it now. Note, we only 1932 // deallocate internally, but keep the empty wrapper. This ensures that if we do 1933 // a rollback after committing the pivot and restarting fast sync, we don't end 1934 // up using a nil bloom. Empty bloom is fine, it just returns that it does not 1935 // have the info we need, so reach down to the database instead. 1936 if d.stateBloom != nil { 1937 d.stateBloom.Close() 1938 } 1939 return nil 1940 } 1941 1942 // DeliverHeaders injects a new batch of block headers received from a remote 1943 // node into the download schedule. 1944 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) error { 1945 return d.deliver(d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1946 } 1947 1948 // DeliverBodies injects a new batch of block bodies received from a remote node. 1949 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) error { 1950 return d.deliver(d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1951 } 1952 1953 // DeliverReceipts injects a new batch of receipts received from a remote node. 1954 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) error { 1955 return d.deliver(d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1956 } 1957 1958 // DeliverNodeData injects a new batch of node state data received from a remote node. 1959 func (d *Downloader) DeliverNodeData(id string, data [][]byte) error { 1960 return d.deliver(d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1961 } 1962 1963 // DeliverSnapPacket is invoked from a peer's message handler when it transmits a 1964 // data packet for the local node to consume. 1965 func (d *Downloader) DeliverSnapPacket(peer *snap.Peer, packet snap.Packet) error { 1966 switch packet := packet.(type) { 1967 case *snap.AccountRangePacket: 1968 hashes, accounts, err := packet.Unpack() 1969 if err != nil { 1970 return err 1971 } 1972 return d.SnapSyncer.OnAccounts(peer, packet.ID, hashes, accounts, packet.Proof) 1973 1974 case *snap.StorageRangesPacket: 1975 hashset, slotset := packet.Unpack() 1976 return d.SnapSyncer.OnStorage(peer, packet.ID, hashset, slotset, packet.Proof) 1977 1978 case *snap.ByteCodesPacket: 1979 return d.SnapSyncer.OnByteCodes(peer, packet.ID, packet.Codes) 1980 1981 case *snap.TrieNodesPacket: 1982 return d.SnapSyncer.OnTrieNodes(peer, packet.ID, packet.Nodes) 1983 1984 default: 1985 return fmt.Errorf("unexpected snap packet type: %T", packet) 1986 } 1987 } 1988 1989 // deliver injects a new batch of data received from a remote node. 1990 func (d *Downloader) deliver(destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1991 // Update the delivery metrics for both good and failed deliveries 1992 inMeter.Mark(int64(packet.Items())) 1993 defer func() { 1994 if err != nil { 1995 dropMeter.Mark(int64(packet.Items())) 1996 } 1997 }() 1998 // Deliver or abort if the sync is canceled while queuing 1999 d.cancelLock.RLock() 2000 cancel := d.cancelCh 2001 d.cancelLock.RUnlock() 2002 if cancel == nil { 2003 return errNoSyncActive 2004 } 2005 select { 2006 case destCh <- packet: 2007 return nil 2008 case <-cancel: 2009 return errNoSyncActive 2010 } 2011 }