github.com/bcnmy/go-ethereum@v1.10.27/les/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 // This is a temporary package whilst working on the eth/66 blocking refactors. 18 // After that work is done, les needs to be refactored to use the new package, 19 // or alternatively use a stripped down version of it. Either way, we need to 20 // keep the changes scoped so duplicating temporarily seems the sanest. 21 package downloader 22 23 import ( 24 "errors" 25 "fmt" 26 "math/big" 27 "sync" 28 "sync/atomic" 29 "time" 30 31 "github.com/ethereum/go-ethereum" 32 "github.com/ethereum/go-ethereum/common" 33 "github.com/ethereum/go-ethereum/core/rawdb" 34 "github.com/ethereum/go-ethereum/core/state/snapshot" 35 "github.com/ethereum/go-ethereum/core/types" 36 "github.com/ethereum/go-ethereum/eth/protocols/eth" 37 "github.com/ethereum/go-ethereum/eth/protocols/snap" 38 "github.com/ethereum/go-ethereum/ethdb" 39 "github.com/ethereum/go-ethereum/event" 40 "github.com/ethereum/go-ethereum/log" 41 "github.com/ethereum/go-ethereum/metrics" 42 "github.com/ethereum/go-ethereum/params" 43 ) 44 45 var ( 46 MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request 47 MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request 48 MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly 49 MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request 50 MaxStateFetch = 384 // Amount of node state values to allow fetching per request 51 52 maxQueuedHeaders = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection) 53 maxHeadersProcess = 2048 // Number of header download results to import at once into the chain 54 maxResultsProcess = 2048 // Number of content download results to import at once into the chain 55 fullMaxForkAncestry uint64 = params.FullImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 56 lightMaxForkAncestry uint64 = params.LightImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 57 58 reorgProtThreshold = 48 // Threshold number of recent blocks to disable mini reorg protection 59 reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs 60 61 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 62 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 63 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 64 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 65 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 66 ) 67 68 var ( 69 errBusy = errors.New("busy") 70 errUnknownPeer = errors.New("peer is unknown or unhealthy") 71 errBadPeer = errors.New("action from bad peer ignored") 72 errStallingPeer = errors.New("peer is stalling") 73 errUnsyncedPeer = errors.New("unsynced peer") 74 errNoPeers = errors.New("no peers to keep download active") 75 errTimeout = errors.New("timeout") 76 errEmptyHeaderSet = errors.New("empty header set by peer") 77 errPeersUnavailable = errors.New("no peers available or all tried for download") 78 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 79 errInvalidChain = errors.New("retrieved hash chain is invalid") 80 errInvalidBody = errors.New("retrieved block body is invalid") 81 errInvalidReceipt = errors.New("retrieved receipt is invalid") 82 errCancelStateFetch = errors.New("state data download canceled (requested)") 83 errCancelContentProcessing = errors.New("content processing canceled (requested)") 84 errCanceled = errors.New("syncing canceled (requested)") 85 errNoSyncActive = errors.New("no sync active") 86 errTooOld = errors.New("peer's protocol version too old") 87 errNoAncestorFound = errors.New("no common ancestor found") 88 ) 89 90 type Downloader struct { 91 mode uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode 92 mux *event.TypeMux // Event multiplexer to announce sync operation events 93 94 checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync) 95 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 96 queue *queue // Scheduler for selecting the hashes to download 97 peers *peerSet // Set of active peers from which download can proceed 98 99 stateDB ethdb.Database // Database to state sync into (and deduplicate via) 100 101 // Statistics 102 syncStatsChainOrigin uint64 // Origin block number where syncing started at 103 syncStatsChainHeight uint64 // Highest block number known when syncing started 104 syncStatsState stateSyncStats 105 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 106 107 lightchain LightChain 108 blockchain BlockChain 109 110 // Callbacks 111 dropPeer peerDropFn // Drops a peer for misbehaving 112 113 // Status 114 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 115 synchronising int32 116 notified int32 117 committed int32 118 ancientLimit uint64 // The maximum block number which can be regarded as ancient data. 119 120 // Channels 121 headerCh chan dataPack // Channel receiving inbound block headers 122 bodyCh chan dataPack // Channel receiving inbound block bodies 123 receiptCh chan dataPack // Channel receiving inbound receipts 124 bodyWakeCh chan bool // Channel to signal the block body fetcher of new tasks 125 receiptWakeCh chan bool // Channel to signal the receipt fetcher of new tasks 126 headerProcCh chan []*types.Header // Channel to feed the header processor new tasks 127 128 // State sync 129 pivotHeader *types.Header // Pivot block header to dynamically push the syncing state root 130 pivotLock sync.RWMutex // Lock protecting pivot header reads from updates 131 132 snapSync bool // Whether to run state sync over the snap protocol 133 SnapSyncer *snap.Syncer // TODO(karalabe): make private! hack for now 134 stateSyncStart chan *stateSync 135 trackStateReq chan *stateReq 136 stateCh chan dataPack // Channel receiving inbound node state data 137 138 // Cancellation and termination 139 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 140 cancelCh chan struct{} // Channel to cancel mid-flight syncs 141 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 142 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 143 144 quitCh chan struct{} // Quit channel to signal termination 145 quitLock sync.Mutex // Lock to prevent double closes 146 147 // Testing hooks 148 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 149 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 150 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 151 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 152 } 153 154 // LightChain encapsulates functions required to synchronise a light chain. 155 type LightChain interface { 156 // HasHeader verifies a header's presence in the local chain. 157 HasHeader(common.Hash, uint64) bool 158 159 // GetHeaderByHash retrieves a header from the local chain. 160 GetHeaderByHash(common.Hash) *types.Header 161 162 // CurrentHeader retrieves the head header from the local chain. 163 CurrentHeader() *types.Header 164 165 // GetTd returns the total difficulty of a local block. 166 GetTd(common.Hash, uint64) *big.Int 167 168 // InsertHeaderChain inserts a batch of headers into the local chain. 169 InsertHeaderChain([]*types.Header, int) (int, error) 170 171 // SetHead rewinds the local chain to a new head. 172 SetHead(uint64) error 173 } 174 175 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 176 type BlockChain interface { 177 LightChain 178 179 // HasBlock verifies a block's presence in the local chain. 180 HasBlock(common.Hash, uint64) bool 181 182 // HasFastBlock verifies a fast block's presence in the local chain. 183 HasFastBlock(common.Hash, uint64) bool 184 185 // GetBlockByHash retrieves a block from the local chain. 186 GetBlockByHash(common.Hash) *types.Block 187 188 // CurrentBlock retrieves the head block from the local chain. 189 CurrentBlock() *types.Block 190 191 // CurrentFastBlock retrieves the head fast block from the local chain. 192 CurrentFastBlock() *types.Block 193 194 // FastSyncCommitHead directly commits the head block to a certain entity. 195 FastSyncCommitHead(common.Hash) error 196 197 // InsertChain inserts a batch of blocks into the local chain. 198 InsertChain(types.Blocks) (int, error) 199 200 // InsertReceiptChain inserts a batch of receipts into the local chain. 201 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 202 203 // Snapshots returns the blockchain snapshot tree to paused it during sync. 204 Snapshots() *snapshot.Tree 205 } 206 207 // New creates a new downloader to fetch hashes and blocks from remote peers. 208 func New(checkpoint uint64, stateDb ethdb.Database, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 209 if lightchain == nil { 210 lightchain = chain 211 } 212 dl := &Downloader{ 213 stateDB: stateDb, 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 snap sync was requested, create the snap scheduler and switch to fast 368 // sync mode. Long term we could drop fast sync or merge the two together, 369 // but until snap becomes prevalent, we should support both. TODO(karalabe). 370 if mode == SnapSync { 371 if !d.snapSync { 372 // Snap sync uses the snapshot namespace to store potentially flakey data until 373 // sync completely heals and finishes. Pause snapshot maintenance in the mean 374 // time to prevent access. 375 if snapshots := d.blockchain.Snapshots(); snapshots != nil { // Only nil in tests 376 snapshots.Disable() 377 } 378 log.Warn("Enabling snapshot sync prototype") 379 d.snapSync = true 380 } 381 mode = FastSync 382 } 383 // Reset the queue, peer set and wake channels to clean any internal leftover state 384 d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems) 385 d.peers.Reset() 386 387 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 388 select { 389 case <-ch: 390 default: 391 } 392 } 393 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 394 for empty := false; !empty; { 395 select { 396 case <-ch: 397 default: 398 empty = true 399 } 400 } 401 } 402 for empty := false; !empty; { 403 select { 404 case <-d.headerProcCh: 405 default: 406 empty = true 407 } 408 } 409 // Create cancel channel for aborting mid-flight and mark the master peer 410 d.cancelLock.Lock() 411 d.cancelCh = make(chan struct{}) 412 d.cancelPeer = id 413 d.cancelLock.Unlock() 414 415 defer d.Cancel() // No matter what, we can't leave the cancel channel open 416 417 // Atomically set the requested sync mode 418 atomic.StoreUint32(&d.mode, uint32(mode)) 419 420 // Retrieve the origin peer and initiate the downloading process 421 p := d.peers.Peer(id) 422 if p == nil { 423 return errUnknownPeer 424 } 425 return d.syncWithPeer(p, hash, td) 426 } 427 428 func (d *Downloader) getMode() SyncMode { 429 return SyncMode(atomic.LoadUint32(&d.mode)) 430 } 431 432 // syncWithPeer starts a block synchronization based on the hash chain from the 433 // specified peer and head hash. 434 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 435 d.mux.Post(StartEvent{}) 436 defer func() { 437 // reset on error 438 if err != nil { 439 d.mux.Post(FailedEvent{err}) 440 } else { 441 latest := d.lightchain.CurrentHeader() 442 d.mux.Post(DoneEvent{latest}) 443 } 444 }() 445 if p.version < eth.ETH66 { 446 return fmt.Errorf("%w: advertized %d < required %d", errTooOld, p.version, eth.ETH66) 447 } 448 mode := d.getMode() 449 450 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", mode) 451 defer func(start time.Time) { 452 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 453 }(time.Now()) 454 455 // Look up the sync boundaries: the common ancestor and the target block 456 latest, pivot, err := d.fetchHead(p) 457 if err != nil { 458 return err 459 } 460 if mode == FastSync && pivot == nil { 461 // If no pivot block was returned, the head is below the min full block 462 // threshold (i.e. new chain). In that case we won't really fast sync 463 // anyway, but still need a valid pivot block to avoid some code hitting 464 // nil panics on an access. 465 pivot = d.blockchain.CurrentBlock().Header() 466 } 467 height := latest.Number.Uint64() 468 469 origin, err := d.findAncestor(p, latest) 470 if err != nil { 471 return err 472 } 473 d.syncStatsLock.Lock() 474 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 475 d.syncStatsChainOrigin = origin 476 } 477 d.syncStatsChainHeight = height 478 d.syncStatsLock.Unlock() 479 480 // Ensure our origin point is below any fast sync pivot point 481 if mode == FastSync { 482 if height <= uint64(fsMinFullBlocks) { 483 origin = 0 484 } else { 485 pivotNumber := pivot.Number.Uint64() 486 if pivotNumber <= origin { 487 origin = pivotNumber - 1 488 } 489 // Write out the pivot into the database so a rollback beyond it will 490 // reenable fast sync 491 rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber) 492 } 493 } 494 d.committed = 1 495 if mode == FastSync && pivot.Number.Uint64() != 0 { 496 d.committed = 0 497 } 498 if mode == FastSync { 499 // Set the ancient data limitation. 500 // If we are running fast sync, all block data older than ancientLimit will be 501 // written to the ancient store. More recent data will be written to the active 502 // database and will wait for the freezer to migrate. 503 // 504 // If there is a checkpoint available, then calculate the ancientLimit through 505 // that. Otherwise calculate the ancient limit through the advertised height 506 // of the remote peer. 507 // 508 // The reason for picking checkpoint first is that a malicious peer can give us 509 // a fake (very high) height, forcing the ancient limit to also be very high. 510 // The peer would start to feed us valid blocks until head, resulting in all of 511 // the blocks might be written into the ancient store. A following mini-reorg 512 // could cause issues. 513 if d.checkpoint != 0 && d.checkpoint > fullMaxForkAncestry+1 { 514 d.ancientLimit = d.checkpoint 515 } else if height > fullMaxForkAncestry+1 { 516 d.ancientLimit = height - fullMaxForkAncestry - 1 517 } else { 518 d.ancientLimit = 0 519 } 520 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 521 522 // If a part of blockchain data has already been written into active store, 523 // disable the ancient style insertion explicitly. 524 if origin >= frozen && frozen != 0 { 525 d.ancientLimit = 0 526 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 527 } else if d.ancientLimit > 0 { 528 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 529 } 530 // Rewind the ancient store and blockchain if reorg happens. 531 if origin+1 < frozen { 532 if err := d.lightchain.SetHead(origin + 1); err != nil { 533 return err 534 } 535 } 536 } 537 // Initiate the sync using a concurrent header and content retrieval algorithm 538 d.queue.Prepare(origin+1, mode) 539 if d.syncInitHook != nil { 540 d.syncInitHook(origin, height) 541 } 542 fetchers := []func() error{ 543 func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved 544 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 545 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 546 func() error { return d.processHeaders(origin+1, td) }, 547 } 548 if mode == FastSync { 549 d.pivotLock.Lock() 550 d.pivotHeader = pivot 551 d.pivotLock.Unlock() 552 553 fetchers = append(fetchers, func() error { return d.processFastSyncContent() }) 554 } else if mode == FullSync { 555 fetchers = append(fetchers, d.processFullSyncContent) 556 } 557 return d.spawnSync(fetchers) 558 } 559 560 // spawnSync runs d.process and all given fetcher functions to completion in 561 // separate goroutines, returning the first error that appears. 562 func (d *Downloader) spawnSync(fetchers []func() error) error { 563 errc := make(chan error, len(fetchers)) 564 d.cancelWg.Add(len(fetchers)) 565 for _, fn := range fetchers { 566 fn := fn 567 go func() { defer d.cancelWg.Done(); errc <- fn() }() 568 } 569 // Wait for the first error, then terminate the others. 570 var err error 571 for i := 0; i < len(fetchers); i++ { 572 if i == len(fetchers)-1 { 573 // Close the queue when all fetchers have exited. 574 // This will cause the block processor to end when 575 // it has processed the queue. 576 d.queue.Close() 577 } 578 if err = <-errc; err != nil && err != errCanceled { 579 break 580 } 581 } 582 d.queue.Close() 583 d.Cancel() 584 return err 585 } 586 587 // cancel aborts all of the operations and resets the queue. However, cancel does 588 // not wait for the running download goroutines to finish. This method should be 589 // used when cancelling the downloads from inside the downloader. 590 func (d *Downloader) cancel() { 591 // Close the current cancel channel 592 d.cancelLock.Lock() 593 defer d.cancelLock.Unlock() 594 595 if d.cancelCh != nil { 596 select { 597 case <-d.cancelCh: 598 // Channel was already closed 599 default: 600 close(d.cancelCh) 601 } 602 } 603 } 604 605 // Cancel aborts all of the operations and waits for all download goroutines to 606 // finish before returning. 607 func (d *Downloader) Cancel() { 608 d.cancel() 609 d.cancelWg.Wait() 610 } 611 612 // Terminate interrupts the downloader, canceling all pending operations. 613 // The downloader cannot be reused after calling Terminate. 614 func (d *Downloader) Terminate() { 615 // Close the termination channel (make sure double close is allowed) 616 d.quitLock.Lock() 617 select { 618 case <-d.quitCh: 619 default: 620 close(d.quitCh) 621 } 622 d.quitLock.Unlock() 623 624 // Cancel any pending download requests 625 d.Cancel() 626 } 627 628 // fetchHead retrieves the head header and prior pivot block (if available) from 629 // a remote peer. 630 func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *types.Header, err error) { 631 p.log.Debug("Retrieving remote chain head") 632 mode := d.getMode() 633 634 // Request the advertised remote head block and wait for the response 635 latest, _ := p.peer.Head() 636 fetch := 1 637 if mode == FastSync { 638 fetch = 2 // head + pivot headers 639 } 640 go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true) 641 642 ttl := d.peers.rates.TargetTimeout() 643 timeout := time.After(ttl) 644 for { 645 select { 646 case <-d.cancelCh: 647 return nil, nil, errCanceled 648 649 case packet := <-d.headerCh: 650 // Discard anything not from the origin peer 651 if packet.PeerId() != p.id { 652 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 653 break 654 } 655 // Make sure the peer gave us at least one and at most the requested headers 656 headers := packet.(*headerPack).headers 657 if len(headers) == 0 || len(headers) > fetch { 658 return nil, nil, fmt.Errorf("%w: returned headers %d != requested %d", errBadPeer, len(headers), fetch) 659 } 660 // The first header needs to be the head, validate against the checkpoint 661 // and request. If only 1 header was returned, make sure there's no pivot 662 // or there was not one requested. 663 head := headers[0] 664 if (mode == FastSync || mode == LightSync) && head.Number.Uint64() < d.checkpoint { 665 return nil, nil, fmt.Errorf("%w: remote head %d below checkpoint %d", errUnsyncedPeer, head.Number, d.checkpoint) 666 } 667 if len(headers) == 1 { 668 if mode == FastSync && head.Number.Uint64() > uint64(fsMinFullBlocks) { 669 return nil, nil, fmt.Errorf("%w: no pivot included along head header", errBadPeer) 670 } 671 p.log.Debug("Remote head identified, no pivot", "number", head.Number, "hash", head.Hash()) 672 return head, nil, nil 673 } 674 // At this point we have 2 headers in total and the first is the 675 // validated head of the chain. Check the pivot number and return, 676 pivot := headers[1] 677 if pivot.Number.Uint64() != head.Number.Uint64()-uint64(fsMinFullBlocks) { 678 return nil, nil, fmt.Errorf("%w: remote pivot %d != requested %d", errInvalidChain, pivot.Number, head.Number.Uint64()-uint64(fsMinFullBlocks)) 679 } 680 return head, pivot, nil 681 682 case <-timeout: 683 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 684 return nil, nil, errTimeout 685 686 case <-d.bodyCh: 687 case <-d.receiptCh: 688 // Out of bounds delivery, ignore 689 } 690 } 691 } 692 693 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 694 // common ancestor. 695 // It returns parameters to be used for peer.RequestHeadersByNumber: 696 // from - starting block number 697 // count - number of headers to request 698 // skip - number of headers to skip 699 // and also returns 'max', the last block which is expected to be returned by the remote peers, 700 // given the (from,count,skip) 701 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 702 var ( 703 from int 704 count int 705 MaxCount = MaxHeaderFetch / 16 706 ) 707 // requestHead is the highest block that we will ask for. If requestHead is not offset, 708 // the highest block that we will get is 16 blocks back from head, which means we 709 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 710 // between us and the peer is 1-2 blocks, which is most common 711 requestHead := int(remoteHeight) - 1 712 if requestHead < 0 { 713 requestHead = 0 714 } 715 // requestBottom is the lowest block we want included in the query 716 // Ideally, we want to include the one just below our own head 717 requestBottom := int(localHeight - 1) 718 if requestBottom < 0 { 719 requestBottom = 0 720 } 721 totalSpan := requestHead - requestBottom 722 span := 1 + totalSpan/MaxCount 723 if span < 2 { 724 span = 2 725 } 726 if span > 16 { 727 span = 16 728 } 729 730 count = 1 + totalSpan/span 731 if count > MaxCount { 732 count = MaxCount 733 } 734 if count < 2 { 735 count = 2 736 } 737 from = requestHead - (count-1)*span 738 if from < 0 { 739 from = 0 740 } 741 max := from + (count-1)*span 742 return int64(from), count, span - 1, uint64(max) 743 } 744 745 // findAncestor tries to locate the common ancestor link of the local chain and 746 // a remote peers blockchain. In the general case when our node was in sync and 747 // on the correct chain, checking the top N links should already get us a match. 748 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 749 // the head links match), we do a binary search to find the common ancestor. 750 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 751 // Figure out the valid ancestor range to prevent rewrite attacks 752 var ( 753 floor = int64(-1) 754 localHeight uint64 755 remoteHeight = remoteHeader.Number.Uint64() 756 ) 757 mode := d.getMode() 758 switch mode { 759 case FullSync: 760 localHeight = d.blockchain.CurrentBlock().NumberU64() 761 case FastSync: 762 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 763 default: 764 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 765 } 766 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 767 768 // Recap floor value for binary search 769 maxForkAncestry := fullMaxForkAncestry 770 if d.getMode() == LightSync { 771 maxForkAncestry = lightMaxForkAncestry 772 } 773 if localHeight >= maxForkAncestry { 774 // We're above the max reorg threshold, find the earliest fork point 775 floor = int64(localHeight - maxForkAncestry) 776 } 777 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 778 // all headers before that point will be missing. 779 if mode == LightSync { 780 // If we don't know the current CHT position, find it 781 if d.genesis == 0 { 782 header := d.lightchain.CurrentHeader() 783 for header != nil { 784 d.genesis = header.Number.Uint64() 785 if floor >= int64(d.genesis)-1 { 786 break 787 } 788 header = d.lightchain.GetHeaderByHash(header.ParentHash) 789 } 790 } 791 // We already know the "genesis" block number, cap floor to that 792 if floor < int64(d.genesis)-1 { 793 floor = int64(d.genesis) - 1 794 } 795 } 796 797 ancestor, err := d.findAncestorSpanSearch(p, mode, remoteHeight, localHeight, floor) 798 if err == nil { 799 return ancestor, nil 800 } 801 // The returned error was not nil. 802 // If the error returned does not reflect that a common ancestor was not found, return it. 803 // If the error reflects that a common ancestor was not found, continue to binary search, 804 // where the error value will be reassigned. 805 if !errors.Is(err, errNoAncestorFound) { 806 return 0, err 807 } 808 809 ancestor, err = d.findAncestorBinarySearch(p, mode, remoteHeight, floor) 810 if err != nil { 811 return 0, err 812 } 813 return ancestor, nil 814 } 815 816 func (d *Downloader) findAncestorSpanSearch(p *peerConnection, mode SyncMode, remoteHeight, localHeight uint64, floor int64) (commonAncestor uint64, err error) { 817 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 818 819 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 820 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 821 822 // Wait for the remote response to the head fetch 823 number, hash := uint64(0), common.Hash{} 824 825 ttl := d.peers.rates.TargetTimeout() 826 timeout := time.After(ttl) 827 828 for finished := false; !finished; { 829 select { 830 case <-d.cancelCh: 831 return 0, errCanceled 832 833 case packet := <-d.headerCh: 834 // Discard anything not from the origin peer 835 if packet.PeerId() != p.id { 836 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 837 break 838 } 839 // Make sure the peer actually gave something valid 840 headers := packet.(*headerPack).headers 841 if len(headers) == 0 { 842 p.log.Warn("Empty head header set") 843 return 0, errEmptyHeaderSet 844 } 845 // Make sure the peer's reply conforms to the request 846 for i, header := range headers { 847 expectNumber := from + int64(i)*int64(skip+1) 848 if number := header.Number.Int64(); number != expectNumber { 849 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 850 return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering")) 851 } 852 } 853 // Check if a common ancestor was found 854 finished = true 855 for i := len(headers) - 1; i >= 0; i-- { 856 // Skip any headers that underflow/overflow our requested set 857 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 858 continue 859 } 860 // Otherwise check if we already know the header or not 861 h := headers[i].Hash() 862 n := headers[i].Number.Uint64() 863 864 var known bool 865 switch mode { 866 case FullSync: 867 known = d.blockchain.HasBlock(h, n) 868 case FastSync: 869 known = d.blockchain.HasFastBlock(h, n) 870 default: 871 known = d.lightchain.HasHeader(h, n) 872 } 873 if known { 874 number, hash = n, h 875 break 876 } 877 } 878 879 case <-timeout: 880 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 881 return 0, errTimeout 882 883 case <-d.bodyCh: 884 case <-d.receiptCh: 885 // Out of bounds delivery, ignore 886 } 887 } 888 // If the head fetch already found an ancestor, return 889 if hash != (common.Hash{}) { 890 if int64(number) <= floor { 891 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 892 return 0, errInvalidAncestor 893 } 894 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 895 return number, nil 896 } 897 return 0, errNoAncestorFound 898 } 899 900 func (d *Downloader) findAncestorBinarySearch(p *peerConnection, mode SyncMode, remoteHeight uint64, floor int64) (commonAncestor uint64, err error) { 901 hash := common.Hash{} 902 903 // Ancestor not found, we need to binary search over our chain 904 start, end := uint64(0), remoteHeight 905 if floor > 0 { 906 start = uint64(floor) 907 } 908 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 909 910 for start+1 < end { 911 // Split our chain interval in two, and request the hash to cross check 912 check := (start + end) / 2 913 914 ttl := d.peers.rates.TargetTimeout() 915 timeout := time.After(ttl) 916 917 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 918 919 // Wait until a reply arrives to this request 920 for arrived := false; !arrived; { 921 select { 922 case <-d.cancelCh: 923 return 0, errCanceled 924 925 case packet := <-d.headerCh: 926 // Discard anything not from the origin peer 927 if packet.PeerId() != p.id { 928 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 929 break 930 } 931 // Make sure the peer actually gave something valid 932 headers := packet.(*headerPack).headers 933 if len(headers) != 1 { 934 p.log.Warn("Multiple headers for single request", "headers", len(headers)) 935 return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers)) 936 } 937 arrived = true 938 939 // Modify the search interval based on the response 940 h := headers[0].Hash() 941 n := headers[0].Number.Uint64() 942 943 var known bool 944 switch mode { 945 case FullSync: 946 known = d.blockchain.HasBlock(h, n) 947 case FastSync: 948 known = d.blockchain.HasFastBlock(h, n) 949 default: 950 known = d.lightchain.HasHeader(h, n) 951 } 952 if !known { 953 end = check 954 break 955 } 956 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 957 if header.Number.Uint64() != check { 958 p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 959 return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number) 960 } 961 start = check 962 hash = h 963 964 case <-timeout: 965 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 966 return 0, errTimeout 967 968 case <-d.bodyCh: 969 case <-d.receiptCh: 970 // Out of bounds delivery, ignore 971 } 972 } 973 } 974 // Ensure valid ancestry and return 975 if int64(start) <= floor { 976 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 977 return 0, errInvalidAncestor 978 } 979 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 980 return start, nil 981 } 982 983 // fetchHeaders keeps retrieving headers concurrently from the number 984 // requested, until no more are returned, potentially throttling on the way. To 985 // facilitate concurrency but still protect against malicious nodes sending bad 986 // headers, we construct a header chain skeleton using the "origin" peer we are 987 // syncing with, and fill in the missing headers using anyone else. Headers from 988 // other peers are only accepted if they map cleanly to the skeleton. If no one 989 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 990 // the origin is dropped. 991 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error { 992 p.log.Debug("Directing header downloads", "origin", from) 993 defer p.log.Debug("Header download terminated") 994 995 // Create a timeout timer, and the associated header fetcher 996 skeleton := true // Skeleton assembly phase or finishing up 997 pivoting := false // Whether the next request is pivot verification 998 request := time.Now() // time of the last skeleton fetch request 999 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 1000 <-timeout.C // timeout channel should be initially empty 1001 defer timeout.Stop() 1002 1003 var ttl time.Duration 1004 getHeaders := func(from uint64) { 1005 request = time.Now() 1006 1007 ttl = d.peers.rates.TargetTimeout() 1008 timeout.Reset(ttl) 1009 1010 if skeleton { 1011 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 1012 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 1013 } else { 1014 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 1015 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 1016 } 1017 } 1018 getNextPivot := func() { 1019 pivoting = true 1020 request = time.Now() 1021 1022 ttl = d.peers.rates.TargetTimeout() 1023 timeout.Reset(ttl) 1024 1025 d.pivotLock.RLock() 1026 pivot := d.pivotHeader.Number.Uint64() 1027 d.pivotLock.RUnlock() 1028 1029 p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks)) 1030 go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep 1031 } 1032 // Start pulling the header chain skeleton until all is done 1033 ancestor := from 1034 getHeaders(from) 1035 1036 mode := d.getMode() 1037 for { 1038 select { 1039 case <-d.cancelCh: 1040 return errCanceled 1041 1042 case packet := <-d.headerCh: 1043 // Make sure the active peer is giving us the skeleton headers 1044 if packet.PeerId() != p.id { 1045 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 1046 break 1047 } 1048 headerReqTimer.UpdateSince(request) 1049 timeout.Stop() 1050 1051 // If the pivot is being checked, move if it became stale and run the real retrieval 1052 var pivot uint64 1053 1054 d.pivotLock.RLock() 1055 if d.pivotHeader != nil { 1056 pivot = d.pivotHeader.Number.Uint64() 1057 } 1058 d.pivotLock.RUnlock() 1059 1060 if pivoting { 1061 if packet.Items() == 2 { 1062 // Retrieve the headers and do some sanity checks, just in case 1063 headers := packet.(*headerPack).headers 1064 1065 if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want { 1066 log.Warn("Peer sent invalid next pivot", "have", have, "want", want) 1067 return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want) 1068 } 1069 if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want { 1070 log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want) 1071 return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want) 1072 } 1073 log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number) 1074 pivot = headers[0].Number.Uint64() 1075 1076 d.pivotLock.Lock() 1077 d.pivotHeader = headers[0] 1078 d.pivotLock.Unlock() 1079 1080 // Write out the pivot into the database so a rollback beyond 1081 // it will reenable fast sync and update the state root that 1082 // the state syncer will be downloading. 1083 rawdb.WriteLastPivotNumber(d.stateDB, pivot) 1084 } 1085 pivoting = false 1086 getHeaders(from) 1087 continue 1088 } 1089 // If the skeleton's finished, pull any remaining head headers directly from the origin 1090 if skeleton && packet.Items() == 0 { 1091 skeleton = false 1092 getHeaders(from) 1093 continue 1094 } 1095 // If no more headers are inbound, notify the content fetchers and return 1096 if packet.Items() == 0 { 1097 // Don't abort header fetches while the pivot is downloading 1098 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 1099 p.log.Debug("No headers, waiting for pivot commit") 1100 select { 1101 case <-time.After(fsHeaderContCheck): 1102 getHeaders(from) 1103 continue 1104 case <-d.cancelCh: 1105 return errCanceled 1106 } 1107 } 1108 // Pivot done (or not in fast sync) and no more headers, terminate the process 1109 p.log.Debug("No more headers available") 1110 select { 1111 case d.headerProcCh <- nil: 1112 return nil 1113 case <-d.cancelCh: 1114 return errCanceled 1115 } 1116 } 1117 headers := packet.(*headerPack).headers 1118 1119 // If we received a skeleton batch, resolve internals concurrently 1120 if skeleton { 1121 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1122 if err != nil { 1123 p.log.Debug("Skeleton chain invalid", "err", err) 1124 return fmt.Errorf("%w: %v", errInvalidChain, err) 1125 } 1126 headers = filled[proced:] 1127 from += uint64(proced) 1128 } else { 1129 // If we're closing in on the chain head, but haven't yet reached it, delay 1130 // the last few headers so mini reorgs on the head don't cause invalid hash 1131 // chain errors. 1132 if n := len(headers); n > 0 { 1133 // Retrieve the current head we're at 1134 var head uint64 1135 if mode == LightSync { 1136 head = d.lightchain.CurrentHeader().Number.Uint64() 1137 } else { 1138 head = d.blockchain.CurrentFastBlock().NumberU64() 1139 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1140 head = full 1141 } 1142 } 1143 // If the head is below the common ancestor, we're actually deduplicating 1144 // already existing chain segments, so use the ancestor as the fake head. 1145 // Otherwise we might end up delaying header deliveries pointlessly. 1146 if head < ancestor { 1147 head = ancestor 1148 } 1149 // If the head is way older than this batch, delay the last few headers 1150 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1151 delay := reorgProtHeaderDelay 1152 if delay > n { 1153 delay = n 1154 } 1155 headers = headers[:n-delay] 1156 } 1157 } 1158 } 1159 // Insert all the new headers and fetch the next batch 1160 if len(headers) > 0 { 1161 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1162 select { 1163 case d.headerProcCh <- headers: 1164 case <-d.cancelCh: 1165 return errCanceled 1166 } 1167 from += uint64(len(headers)) 1168 1169 // If we're still skeleton filling fast sync, check pivot staleness 1170 // before continuing to the next skeleton filling 1171 if skeleton && pivot > 0 { 1172 getNextPivot() 1173 } else { 1174 getHeaders(from) 1175 } 1176 } else { 1177 // No headers delivered, or all of them being delayed, sleep a bit and retry 1178 p.log.Trace("All headers delayed, waiting") 1179 select { 1180 case <-time.After(fsHeaderContCheck): 1181 getHeaders(from) 1182 continue 1183 case <-d.cancelCh: 1184 return errCanceled 1185 } 1186 } 1187 1188 case <-timeout.C: 1189 if d.dropPeer == nil { 1190 // The dropPeer method is nil when `--copydb` is used for a local copy. 1191 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1192 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1193 break 1194 } 1195 // Header retrieval timed out, consider the peer bad and drop 1196 p.log.Debug("Header request timed out", "elapsed", ttl) 1197 headerTimeoutMeter.Mark(1) 1198 d.dropPeer(p.id) 1199 1200 // Finish the sync gracefully instead of dumping the gathered data though 1201 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1202 select { 1203 case ch <- false: 1204 case <-d.cancelCh: 1205 } 1206 } 1207 select { 1208 case d.headerProcCh <- nil: 1209 case <-d.cancelCh: 1210 } 1211 return fmt.Errorf("%w: header request timed out", errBadPeer) 1212 } 1213 } 1214 } 1215 1216 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1217 // and maps them to the provided skeleton header chain. 1218 // 1219 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1220 // immediately to the header processor to keep the rest of the pipeline full even 1221 // in the case of header stalls. 1222 // 1223 // The method returns the entire filled skeleton and also the number of headers 1224 // already forwarded for processing. 1225 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1226 log.Debug("Filling up skeleton", "from", from) 1227 d.queue.ScheduleSkeleton(from, skeleton) 1228 1229 var ( 1230 deliver = func(packet dataPack) (int, error) { 1231 pack := packet.(*headerPack) 1232 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1233 } 1234 expire = func() map[string]int { return d.queue.ExpireHeaders(d.peers.rates.TargetTimeout()) } 1235 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) { 1236 return d.queue.ReserveHeaders(p, count), false, false 1237 } 1238 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1239 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.peers.rates.TargetRoundTrip()) } 1240 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1241 p.SetHeadersIdle(accepted, deliveryTime) 1242 } 1243 ) 1244 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1245 d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve, 1246 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1247 1248 log.Debug("Skeleton fill terminated", "err", err) 1249 1250 filled, proced := d.queue.RetrieveHeaders() 1251 return filled, proced, err 1252 } 1253 1254 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1255 // available peers, reserving a chunk of blocks for each, waiting for delivery 1256 // and also periodically checking for timeouts. 1257 func (d *Downloader) fetchBodies(from uint64) error { 1258 log.Debug("Downloading block bodies", "origin", from) 1259 1260 var ( 1261 deliver = func(packet dataPack) (int, error) { 1262 pack := packet.(*bodyPack) 1263 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1264 } 1265 expire = func() map[string]int { return d.queue.ExpireBodies(d.peers.rates.TargetTimeout()) } 1266 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1267 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.peers.rates.TargetRoundTrip()) } 1268 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) } 1269 ) 1270 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1271 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies, 1272 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1273 1274 log.Debug("Block body download terminated", "err", err) 1275 return err 1276 } 1277 1278 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1279 // available peers, reserving a chunk of receipts for each, waiting for delivery 1280 // and also periodically checking for timeouts. 1281 func (d *Downloader) fetchReceipts(from uint64) error { 1282 log.Debug("Downloading transaction receipts", "origin", from) 1283 1284 var ( 1285 deliver = func(packet dataPack) (int, error) { 1286 pack := packet.(*receiptPack) 1287 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1288 } 1289 expire = func() map[string]int { return d.queue.ExpireReceipts(d.peers.rates.TargetTimeout()) } 1290 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1291 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.peers.rates.TargetRoundTrip()) } 1292 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1293 p.SetReceiptsIdle(accepted, deliveryTime) 1294 } 1295 ) 1296 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1297 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts, 1298 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1299 1300 log.Debug("Transaction receipt download terminated", "err", err) 1301 return err 1302 } 1303 1304 // fetchParts iteratively downloads scheduled block parts, taking any available 1305 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1306 // also periodically checking for timeouts. 1307 // 1308 // As the scheduling/timeout logic mostly is the same for all downloaded data 1309 // types, this method is used by each for data gathering and is instrumented with 1310 // various callbacks to handle the slight differences between processing them. 1311 // 1312 // The instrumentation parameters: 1313 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1314 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1315 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1316 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1317 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1318 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1319 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1320 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1321 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1322 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1323 // - fetch: network callback to actually send a particular download request to a physical remote peer 1324 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1325 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1326 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1327 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1328 // - kind: textual label of the type being downloaded to display in log messages 1329 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1330 expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool), 1331 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1332 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error { 1333 // Create a ticker to detect expired retrieval tasks 1334 ticker := time.NewTicker(100 * time.Millisecond) 1335 defer ticker.Stop() 1336 1337 update := make(chan struct{}, 1) 1338 1339 // Prepare the queue and fetch block parts until the block header fetcher's done 1340 finished := false 1341 for { 1342 select { 1343 case <-d.cancelCh: 1344 return errCanceled 1345 1346 case packet := <-deliveryCh: 1347 deliveryTime := time.Now() 1348 // If the peer was previously banned and failed to deliver its pack 1349 // in a reasonable time frame, ignore its message. 1350 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1351 // Deliver the received chunk of data and check chain validity 1352 accepted, err := deliver(packet) 1353 if errors.Is(err, errInvalidChain) { 1354 return err 1355 } 1356 // Unless a peer delivered something completely else than requested (usually 1357 // caused by a timed out request which came through in the end), set it to 1358 // idle. If the delivery's stale, the peer should have already been idled. 1359 if !errors.Is(err, errStaleDelivery) { 1360 setIdle(peer, accepted, deliveryTime) 1361 } 1362 // Issue a log to the user to see what's going on 1363 switch { 1364 case err == nil && packet.Items() == 0: 1365 peer.log.Trace("Requested data not delivered", "type", kind) 1366 case err == nil: 1367 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1368 default: 1369 peer.log.Debug("Failed to deliver retrieved data", "type", kind, "err", err) 1370 } 1371 } 1372 // Blocks assembled, try to update the progress 1373 select { 1374 case update <- struct{}{}: 1375 default: 1376 } 1377 1378 case cont := <-wakeCh: 1379 // The header fetcher sent a continuation flag, check if it's done 1380 if !cont { 1381 finished = true 1382 } 1383 // Headers arrive, try to update the progress 1384 select { 1385 case update <- struct{}{}: 1386 default: 1387 } 1388 1389 case <-ticker.C: 1390 // Sanity check update the progress 1391 select { 1392 case update <- struct{}{}: 1393 default: 1394 } 1395 1396 case <-update: 1397 // Short circuit if we lost all our peers 1398 if d.peers.Len() == 0 { 1399 return errNoPeers 1400 } 1401 // Check for fetch request timeouts and demote the responsible peers 1402 for pid, fails := range expire() { 1403 if peer := d.peers.Peer(pid); peer != nil { 1404 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1405 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1406 // out that sync wise we need to get rid of the peer. 1407 // 1408 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1409 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1410 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1411 if fails > 2 { 1412 peer.log.Trace("Data delivery timed out", "type", kind) 1413 setIdle(peer, 0, time.Now()) 1414 } else { 1415 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1416 1417 if d.dropPeer == nil { 1418 // The dropPeer method is nil when `--copydb` is used for a local copy. 1419 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1420 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1421 } else { 1422 d.dropPeer(pid) 1423 1424 // If this peer was the master peer, abort sync immediately 1425 d.cancelLock.RLock() 1426 master := pid == d.cancelPeer 1427 d.cancelLock.RUnlock() 1428 1429 if master { 1430 d.cancel() 1431 return errTimeout 1432 } 1433 } 1434 } 1435 } 1436 } 1437 // If there's nothing more to fetch, wait or terminate 1438 if pending() == 0 { 1439 if !inFlight() && finished { 1440 log.Debug("Data fetching completed", "type", kind) 1441 return nil 1442 } 1443 break 1444 } 1445 // Send a download request to all idle peers, until throttled 1446 progressed, throttled, running := false, false, inFlight() 1447 idles, total := idle() 1448 pendCount := pending() 1449 for _, peer := range idles { 1450 // Short circuit if throttling activated 1451 if throttled { 1452 break 1453 } 1454 // Short circuit if there is no more available task. 1455 if pendCount = pending(); pendCount == 0 { 1456 break 1457 } 1458 // Reserve a chunk of fetches for a peer. A nil can mean either that 1459 // no more headers are available, or that the peer is known not to 1460 // have them. 1461 request, progress, throttle := reserve(peer, capacity(peer)) 1462 if progress { 1463 progressed = true 1464 } 1465 if throttle { 1466 throttled = true 1467 throttleCounter.Inc(1) 1468 } 1469 if request == nil { 1470 continue 1471 } 1472 if request.From > 0 { 1473 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1474 } else { 1475 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1476 } 1477 // Fetch the chunk and make sure any errors return the hashes to the queue 1478 if fetchHook != nil { 1479 fetchHook(request.Headers) 1480 } 1481 if err := fetch(peer, request); err != nil { 1482 // Although we could try and make an attempt to fix this, this error really 1483 // means that we've double allocated a fetch task to a peer. If that is the 1484 // case, the internal state of the downloader and the queue is very wrong so 1485 // better hard crash and note the error instead of silently accumulating into 1486 // a much bigger issue. 1487 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1488 } 1489 running = true 1490 } 1491 // Make sure that we have peers available for fetching. If all peers have been tried 1492 // and all failed throw an error 1493 if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 { 1494 return errPeersUnavailable 1495 } 1496 } 1497 } 1498 } 1499 1500 // processHeaders takes batches of retrieved headers from an input channel and 1501 // keeps processing and scheduling them into the header chain and downloader's 1502 // queue until the stream ends or a failure occurs. 1503 func (d *Downloader) processHeaders(origin uint64, td *big.Int) error { 1504 // Keep a count of uncertain headers to roll back 1505 var ( 1506 rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis) 1507 rollbackErr error 1508 mode = d.getMode() 1509 ) 1510 defer func() { 1511 if rollback > 0 { 1512 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1513 if mode != LightSync { 1514 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1515 lastBlock = d.blockchain.CurrentBlock().Number() 1516 } 1517 if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block 1518 // We're already unwinding the stack, only print the error to make it more visible 1519 log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err) 1520 } 1521 curFastBlock, curBlock := common.Big0, common.Big0 1522 if mode != LightSync { 1523 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1524 curBlock = d.blockchain.CurrentBlock().Number() 1525 } 1526 log.Warn("Rolled back chain segment", 1527 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1528 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1529 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr) 1530 } 1531 }() 1532 // Wait for batches of headers to process 1533 gotHeaders := false 1534 1535 for { 1536 select { 1537 case <-d.cancelCh: 1538 rollbackErr = errCanceled 1539 return errCanceled 1540 1541 case headers := <-d.headerProcCh: 1542 // Terminate header processing if we synced up 1543 if len(headers) == 0 { 1544 // Notify everyone that headers are fully processed 1545 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1546 select { 1547 case ch <- false: 1548 case <-d.cancelCh: 1549 } 1550 } 1551 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1552 // better chain compared to ours. The only exception is if its promised blocks were 1553 // already imported by other means (e.g. fetcher): 1554 // 1555 // R <remote peer>, L <local node>: Both at block 10 1556 // R: Mine block 11, and propagate it to L 1557 // L: Queue block 11 for import 1558 // L: Notice that R's head and TD increased compared to ours, start sync 1559 // L: Import of block 11 finishes 1560 // L: Sync begins, and finds common ancestor at 11 1561 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1562 // R: Nothing to give 1563 if mode != LightSync { 1564 head := d.blockchain.CurrentBlock() 1565 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1566 return errStallingPeer 1567 } 1568 } 1569 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1570 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1571 // of delivering the post-pivot blocks that would flag the invalid content. 1572 // 1573 // This check cannot be executed "as is" for full imports, since blocks may still be 1574 // queued for processing when the header download completes. However, as long as the 1575 // peer gave us something useful, we're already happy/progressed (above check). 1576 if mode == FastSync || mode == LightSync { 1577 head := d.lightchain.CurrentHeader() 1578 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1579 return errStallingPeer 1580 } 1581 } 1582 // Disable any rollback and return 1583 rollback = 0 1584 return nil 1585 } 1586 // Otherwise split the chunk of headers into batches and process them 1587 gotHeaders = true 1588 for len(headers) > 0 { 1589 // Terminate if something failed in between processing chunks 1590 select { 1591 case <-d.cancelCh: 1592 rollbackErr = errCanceled 1593 return errCanceled 1594 default: 1595 } 1596 // Select the next chunk of headers to import 1597 limit := maxHeadersProcess 1598 if limit > len(headers) { 1599 limit = len(headers) 1600 } 1601 chunk := headers[:limit] 1602 1603 // In case of header only syncing, validate the chunk immediately 1604 if mode == FastSync || mode == LightSync { 1605 // If we're importing pure headers, verify based on their recentness 1606 var pivot uint64 1607 1608 d.pivotLock.RLock() 1609 if d.pivotHeader != nil { 1610 pivot = d.pivotHeader.Number.Uint64() 1611 } 1612 d.pivotLock.RUnlock() 1613 1614 frequency := fsHeaderCheckFrequency 1615 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1616 frequency = 1 1617 } 1618 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1619 rollbackErr = err 1620 1621 // If some headers were inserted, track them as uncertain 1622 if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 { 1623 rollback = chunk[0].Number.Uint64() 1624 } 1625 log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err) 1626 return fmt.Errorf("%w: %v", errInvalidChain, err) 1627 } 1628 // All verifications passed, track all headers within the allotted limits 1629 if mode == FastSync { 1630 head := chunk[len(chunk)-1].Number.Uint64() 1631 if head-rollback > uint64(fsHeaderSafetyNet) { 1632 rollback = head - uint64(fsHeaderSafetyNet) 1633 } else { 1634 rollback = 1 1635 } 1636 } 1637 } 1638 // Unless we're doing light chains, schedule the headers for associated content retrieval 1639 if mode == FullSync || mode == FastSync { 1640 // If we've reached the allowed number of pending headers, stall a bit 1641 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1642 select { 1643 case <-d.cancelCh: 1644 rollbackErr = errCanceled 1645 return errCanceled 1646 case <-time.After(time.Second): 1647 } 1648 } 1649 // Otherwise insert the headers for content retrieval 1650 inserts := d.queue.Schedule(chunk, origin) 1651 if len(inserts) != len(chunk) { 1652 rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk)) 1653 return fmt.Errorf("%w: stale headers", errBadPeer) 1654 } 1655 } 1656 headers = headers[limit:] 1657 origin += uint64(limit) 1658 } 1659 // Update the highest block number we know if a higher one is found. 1660 d.syncStatsLock.Lock() 1661 if d.syncStatsChainHeight < origin { 1662 d.syncStatsChainHeight = origin - 1 1663 } 1664 d.syncStatsLock.Unlock() 1665 1666 // Signal the content downloaders of the availability of new tasks 1667 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1668 select { 1669 case ch <- true: 1670 default: 1671 } 1672 } 1673 } 1674 } 1675 } 1676 1677 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1678 func (d *Downloader) processFullSyncContent() error { 1679 for { 1680 results := d.queue.Results(true) 1681 if len(results) == 0 { 1682 return nil 1683 } 1684 if d.chainInsertHook != nil { 1685 d.chainInsertHook(results) 1686 } 1687 if err := d.importBlockResults(results); err != nil { 1688 return err 1689 } 1690 } 1691 } 1692 1693 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1694 // Check for any early termination requests 1695 if len(results) == 0 { 1696 return nil 1697 } 1698 select { 1699 case <-d.quitCh: 1700 return errCancelContentProcessing 1701 default: 1702 } 1703 // Retrieve the a batch of results to import 1704 first, last := results[0].Header, results[len(results)-1].Header 1705 log.Debug("Inserting downloaded chain", "items", len(results), 1706 "firstnum", first.Number, "firsthash", first.Hash(), 1707 "lastnum", last.Number, "lasthash", last.Hash(), 1708 ) 1709 blocks := make([]*types.Block, len(results)) 1710 for i, result := range results { 1711 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1712 } 1713 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1714 if index < len(results) { 1715 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1716 } else { 1717 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1718 // when it needs to preprocess blocks to import a sidechain. 1719 // The importer will put together a new list of blocks to import, which is a superset 1720 // of the blocks delivered from the downloader, and the indexing will be off. 1721 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1722 } 1723 return fmt.Errorf("%w: %v", errInvalidChain, err) 1724 } 1725 return nil 1726 } 1727 1728 // processFastSyncContent takes fetch results from the queue and writes them to the 1729 // database. It also controls the synchronisation of state nodes of the pivot block. 1730 func (d *Downloader) processFastSyncContent() error { 1731 // Start syncing state of the reported head block. This should get us most of 1732 // the state of the pivot block. 1733 d.pivotLock.RLock() 1734 sync := d.syncState(d.pivotHeader.Root) 1735 d.pivotLock.RUnlock() 1736 1737 defer func() { 1738 // The `sync` object is replaced every time the pivot moves. We need to 1739 // defer close the very last active one, hence the lazy evaluation vs. 1740 // calling defer sync.Cancel() !!! 1741 sync.Cancel() 1742 }() 1743 1744 closeOnErr := func(s *stateSync) { 1745 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled && err != snap.ErrCancelled { 1746 d.queue.Close() // wake up Results 1747 } 1748 } 1749 go closeOnErr(sync) 1750 1751 // To cater for moving pivot points, track the pivot block and subsequently 1752 // accumulated download results separately. 1753 var ( 1754 oldPivot *fetchResult // Locked in pivot block, might change eventually 1755 oldTail []*fetchResult // Downloaded content after the pivot 1756 ) 1757 for { 1758 // Wait for the next batch of downloaded data to be available, and if the pivot 1759 // block became stale, move the goalpost 1760 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1761 if len(results) == 0 { 1762 // If pivot sync is done, stop 1763 if oldPivot == nil { 1764 return sync.Cancel() 1765 } 1766 // If sync failed, stop 1767 select { 1768 case <-d.cancelCh: 1769 sync.Cancel() 1770 return errCanceled 1771 default: 1772 } 1773 } 1774 if d.chainInsertHook != nil { 1775 d.chainInsertHook(results) 1776 } 1777 // If we haven't downloaded the pivot block yet, check pivot staleness 1778 // notifications from the header downloader 1779 d.pivotLock.RLock() 1780 pivot := d.pivotHeader 1781 d.pivotLock.RUnlock() 1782 1783 if oldPivot == nil { 1784 if pivot.Root != sync.root { 1785 sync.Cancel() 1786 sync = d.syncState(pivot.Root) 1787 1788 go closeOnErr(sync) 1789 } 1790 } else { 1791 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1792 } 1793 // Split around the pivot block and process the two sides via fast/full sync 1794 if atomic.LoadInt32(&d.committed) == 0 { 1795 latest := results[len(results)-1].Header 1796 // If the height is above the pivot block by 2 sets, it means the pivot 1797 // become stale in the network and it was garbage collected, move to a 1798 // new pivot. 1799 // 1800 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 1801 // need to be taken into account, otherwise we're detecting the pivot move 1802 // late and will drop peers due to unavailable state!!! 1803 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 1804 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 1805 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 1806 1807 d.pivotLock.Lock() 1808 d.pivotHeader = pivot 1809 d.pivotLock.Unlock() 1810 1811 // Write out the pivot into the database so a rollback beyond it will 1812 // reenable fast sync 1813 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 1814 } 1815 } 1816 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 1817 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1818 return err 1819 } 1820 if P != nil { 1821 // If new pivot block found, cancel old state retrieval and restart 1822 if oldPivot != P { 1823 sync.Cancel() 1824 sync = d.syncState(P.Header.Root) 1825 1826 go closeOnErr(sync) 1827 oldPivot = P 1828 } 1829 // Wait for completion, occasionally checking for pivot staleness 1830 select { 1831 case <-sync.done: 1832 if sync.err != nil { 1833 return sync.err 1834 } 1835 if err := d.commitPivotBlock(P); err != nil { 1836 return err 1837 } 1838 oldPivot = nil 1839 1840 case <-time.After(time.Second): 1841 oldTail = afterP 1842 continue 1843 } 1844 } 1845 // Fast sync done, pivot commit done, full import 1846 if err := d.importBlockResults(afterP); err != nil { 1847 return err 1848 } 1849 } 1850 } 1851 1852 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1853 if len(results) == 0 { 1854 return nil, nil, nil 1855 } 1856 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 1857 // the pivot is somewhere in the future 1858 return nil, results, nil 1859 } 1860 // This can also be optimized, but only happens very seldom 1861 for _, result := range results { 1862 num := result.Header.Number.Uint64() 1863 switch { 1864 case num < pivot: 1865 before = append(before, result) 1866 case num == pivot: 1867 p = result 1868 default: 1869 after = append(after, result) 1870 } 1871 } 1872 return p, before, after 1873 } 1874 1875 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1876 // Check for any early termination requests 1877 if len(results) == 0 { 1878 return nil 1879 } 1880 select { 1881 case <-d.quitCh: 1882 return errCancelContentProcessing 1883 case <-stateSync.done: 1884 if err := stateSync.Wait(); err != nil { 1885 return err 1886 } 1887 default: 1888 } 1889 // Retrieve the a batch of results to import 1890 first, last := results[0].Header, results[len(results)-1].Header 1891 log.Debug("Inserting fast-sync blocks", "items", len(results), 1892 "firstnum", first.Number, "firsthash", first.Hash(), 1893 "lastnumn", last.Number, "lasthash", last.Hash(), 1894 ) 1895 blocks := make([]*types.Block, len(results)) 1896 receipts := make([]types.Receipts, len(results)) 1897 for i, result := range results { 1898 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1899 receipts[i] = result.Receipts 1900 } 1901 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1902 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1903 return fmt.Errorf("%w: %v", errInvalidChain, err) 1904 } 1905 return nil 1906 } 1907 1908 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1909 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1910 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1911 1912 // Commit the pivot block as the new head, will require full sync from here on 1913 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1914 return err 1915 } 1916 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1917 return err 1918 } 1919 atomic.StoreInt32(&d.committed, 1) 1920 return nil 1921 } 1922 1923 // DeliverHeaders injects a new batch of block headers received from a remote 1924 // node into the download schedule. 1925 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) error { 1926 return d.deliver(d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1927 } 1928 1929 // DeliverBodies injects a new batch of block bodies received from a remote node. 1930 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) error { 1931 return d.deliver(d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1932 } 1933 1934 // DeliverReceipts injects a new batch of receipts received from a remote node. 1935 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) error { 1936 return d.deliver(d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1937 } 1938 1939 // DeliverNodeData injects a new batch of node state data received from a remote node. 1940 func (d *Downloader) DeliverNodeData(id string, data [][]byte) error { 1941 return d.deliver(d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1942 } 1943 1944 // DeliverSnapPacket is invoked from a peer's message handler when it transmits a 1945 // data packet for the local node to consume. 1946 func (d *Downloader) DeliverSnapPacket(peer *snap.Peer, packet snap.Packet) error { 1947 switch packet := packet.(type) { 1948 case *snap.AccountRangePacket: 1949 hashes, accounts, err := packet.Unpack() 1950 if err != nil { 1951 return err 1952 } 1953 return d.SnapSyncer.OnAccounts(peer, packet.ID, hashes, accounts, packet.Proof) 1954 1955 case *snap.StorageRangesPacket: 1956 hashset, slotset := packet.Unpack() 1957 return d.SnapSyncer.OnStorage(peer, packet.ID, hashset, slotset, packet.Proof) 1958 1959 case *snap.ByteCodesPacket: 1960 return d.SnapSyncer.OnByteCodes(peer, packet.ID, packet.Codes) 1961 1962 case *snap.TrieNodesPacket: 1963 return d.SnapSyncer.OnTrieNodes(peer, packet.ID, packet.Nodes) 1964 1965 default: 1966 return fmt.Errorf("unexpected snap packet type: %T", packet) 1967 } 1968 } 1969 1970 // deliver injects a new batch of data received from a remote node. 1971 func (d *Downloader) deliver(destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1972 // Update the delivery metrics for both good and failed deliveries 1973 inMeter.Mark(int64(packet.Items())) 1974 defer func() { 1975 if err != nil { 1976 dropMeter.Mark(int64(packet.Items())) 1977 } 1978 }() 1979 // Deliver or abort if the sync is canceled while queuing 1980 d.cancelLock.RLock() 1981 cancel := d.cancelCh 1982 d.cancelLock.RUnlock() 1983 if cancel == nil { 1984 return errNoSyncActive 1985 } 1986 select { 1987 case destCh <- packet: 1988 return nil 1989 case <-cancel: 1990 return errNoSyncActive 1991 } 1992 }