github.com/ethxdao/go-ethereum@v0.0.0-20221218102228-5ae34a9cc189/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/ethxdao/go-ethereum/common" 32 "github.com/ethxdao/go-ethereum/core/rawdb" 33 "github.com/ethxdao/go-ethereum/core/state/snapshot" 34 "github.com/ethxdao/go-ethereum/core/types" 35 "github.com/ethxdao/go-ethereum/eth/protocols/eth" 36 "github.com/ethxdao/go-ethereum/eth/protocols/snap" 37 "github.com/ethxdao/go-ethereum/ethdb" 38 "github.com/ethxdao/go-ethereum/event" 39 "github.com/ethxdao/go-ethereum/log" 40 "github.com/ethxdao/go-ethereum/metrics" 41 "github.com/ethxdao/go-ethereum/params" 42 ) 43 44 var ( 45 MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request 46 MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request 47 MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly 48 MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request 49 MaxStateFetch = 384 // Amount of node state values to allow fetching per request 50 51 maxQueuedHeaders = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection) 52 maxHeadersProcess = 2048 // Number of header download results to import at once into the chain 53 maxResultsProcess = 2048 // Number of content download results to import at once into the chain 54 fullMaxForkAncestry uint64 = params.FullImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 55 lightMaxForkAncestry uint64 = params.LightImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 56 57 reorgProtThreshold = 48 // Threshold number of recent blocks to disable mini reorg protection 58 reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs 59 60 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 61 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 62 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 63 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 64 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 65 ) 66 67 var ( 68 errBusy = errors.New("busy") 69 errUnknownPeer = errors.New("peer is unknown or unhealthy") 70 errBadPeer = errors.New("action from bad peer ignored") 71 errStallingPeer = errors.New("peer is stalling") 72 errUnsyncedPeer = errors.New("unsynced peer") 73 errNoPeers = errors.New("no peers to keep download active") 74 errTimeout = errors.New("timeout") 75 errEmptyHeaderSet = errors.New("empty header set by peer") 76 errPeersUnavailable = errors.New("no peers available or all tried for download") 77 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 78 errInvalidChain = errors.New("retrieved hash chain is invalid") 79 errInvalidBody = errors.New("retrieved block body is invalid") 80 errInvalidReceipt = errors.New("retrieved receipt is invalid") 81 errCancelStateFetch = errors.New("state data download canceled (requested)") 82 errCancelContentProcessing = errors.New("content processing canceled (requested)") 83 errCanceled = errors.New("syncing canceled (requested)") 84 errNoSyncActive = errors.New("no sync active") 85 errTooOld = errors.New("peer's protocol version too old") 86 errNoAncestorFound = errors.New("no common ancestor found") 87 ) 88 89 type Downloader struct { 90 mode uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode 91 mux *event.TypeMux // Event multiplexer to announce sync operation events 92 93 checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync) 94 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 95 queue *queue // Scheduler for selecting the hashes to download 96 peers *peerSet // Set of active peers from which download can proceed 97 98 stateDB ethdb.Database // Database to state sync into (and deduplicate via) 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, 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 mux: mux, 214 checkpoint: checkpoint, 215 queue: newQueue(blockCacheMaxItems, blockCacheInitialItems), 216 peers: newPeerSet(), 217 blockchain: chain, 218 lightchain: lightchain, 219 dropPeer: dropPeer, 220 headerCh: make(chan dataPack, 1), 221 bodyCh: make(chan dataPack, 1), 222 receiptCh: make(chan dataPack, 1), 223 bodyWakeCh: make(chan bool, 1), 224 receiptWakeCh: make(chan bool, 1), 225 headerProcCh: make(chan []*types.Header, 1), 226 quitCh: make(chan struct{}), 227 stateCh: make(chan dataPack), 228 SnapSyncer: snap.NewSyncer(stateDb), 229 stateSyncStart: make(chan *stateSync), 230 //syncStatsState: stateSyncStats{ 231 // processed: rawdb.ReadFastTrieProgress(stateDb), 232 //}, 233 trackStateReq: make(chan *stateReq), 234 } 235 go dl.stateFetcher() 236 return dl 237 } 238 239 // Progress retrieves the synchronisation boundaries, specifically the origin 240 // block where synchronisation started at (may have failed/suspended); the block 241 // or header sync is currently at; and the latest known block which the sync targets. 242 // 243 // In addition, during the state download phase of fast synchronisation the number 244 // of processed and the total number of known states are also returned. Otherwise 245 // these are zero. 246 func (d *Downloader) Progress() ethereum.SyncProgress { 247 // Lock the current stats and return the progress 248 d.syncStatsLock.RLock() 249 defer d.syncStatsLock.RUnlock() 250 251 current := uint64(0) 252 mode := d.getMode() 253 switch { 254 case d.blockchain != nil && mode == FullSync: 255 current = d.blockchain.CurrentBlock().NumberU64() 256 case d.blockchain != nil && mode == FastSync: 257 current = d.blockchain.CurrentFastBlock().NumberU64() 258 case d.lightchain != nil: 259 current = d.lightchain.CurrentHeader().Number.Uint64() 260 default: 261 log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", mode) 262 } 263 return ethereum.SyncProgress{ 264 StartingBlock: d.syncStatsChainOrigin, 265 CurrentBlock: current, 266 HighestBlock: d.syncStatsChainHeight, 267 //PulledStates: d.syncStatsState.processed, 268 //KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 269 } 270 } 271 272 // Synchronising returns whether the downloader is currently retrieving blocks. 273 func (d *Downloader) Synchronising() bool { 274 return atomic.LoadInt32(&d.synchronising) > 0 275 } 276 277 // RegisterPeer injects a new download peer into the set of block source to be 278 // used for fetching hashes and blocks from. 279 func (d *Downloader) RegisterPeer(id string, version uint, peer Peer) error { 280 var logger log.Logger 281 if len(id) < 16 { 282 // Tests use short IDs, don't choke on them 283 logger = log.New("peer", id) 284 } else { 285 logger = log.New("peer", id[:8]) 286 } 287 logger.Trace("Registering sync peer") 288 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 289 logger.Error("Failed to register sync peer", "err", err) 290 return err 291 } 292 return nil 293 } 294 295 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 296 func (d *Downloader) RegisterLightPeer(id string, version uint, peer LightPeer) error { 297 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 298 } 299 300 // UnregisterPeer remove a peer from the known list, preventing any action from 301 // the specified peer. An effort is also made to return any pending fetches into 302 // the queue. 303 func (d *Downloader) UnregisterPeer(id string) error { 304 // Unregister the peer from the active peer set and revoke any fetch tasks 305 var logger log.Logger 306 if len(id) < 16 { 307 // Tests use short IDs, don't choke on them 308 logger = log.New("peer", id) 309 } else { 310 logger = log.New("peer", id[:8]) 311 } 312 logger.Trace("Unregistering sync peer") 313 if err := d.peers.Unregister(id); err != nil { 314 logger.Error("Failed to unregister sync peer", "err", err) 315 return err 316 } 317 d.queue.Revoke(id) 318 319 return nil 320 } 321 322 // Synchronise tries to sync up our local block chain with a remote peer, both 323 // adding various sanity checks as well as wrapping it with various log entries. 324 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 325 err := d.synchronise(id, head, td, mode) 326 327 switch err { 328 case nil, errBusy, errCanceled: 329 return err 330 } 331 if errors.Is(err, errInvalidChain) || errors.Is(err, errBadPeer) || errors.Is(err, errTimeout) || 332 errors.Is(err, errStallingPeer) || errors.Is(err, errUnsyncedPeer) || errors.Is(err, errEmptyHeaderSet) || 333 errors.Is(err, errPeersUnavailable) || errors.Is(err, errTooOld) || errors.Is(err, errInvalidAncestor) { 334 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 335 if d.dropPeer == nil { 336 // The dropPeer method is nil when `--copydb` is used for a local copy. 337 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 338 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 339 } else { 340 d.dropPeer(id) 341 } 342 return err 343 } 344 log.Warn("Synchronisation failed, retrying", "err", err) 345 return err 346 } 347 348 // synchronise will select the peer and use it for synchronising. If an empty string is given 349 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 350 // checks fail an error will be returned. This method is synchronous 351 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 352 // Mock out the synchronisation if testing 353 if d.synchroniseMock != nil { 354 return d.synchroniseMock(id, hash) 355 } 356 // Make sure only one goroutine is ever allowed past this point at once 357 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 358 return errBusy 359 } 360 defer atomic.StoreInt32(&d.synchronising, 0) 361 362 // Post a user notification of the sync (only once per session) 363 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 364 log.Info("Block synchronisation started") 365 } 366 // If snap sync was requested, create the snap scheduler and switch to fast 367 // sync mode. Long term we could drop fast sync or merge the two together, 368 // but until snap becomes prevalent, we should support both. TODO(karalabe). 369 if mode == SnapSync { 370 if !d.snapSync { 371 // Snap sync uses the snapshot namespace to store potentially flakey data until 372 // sync completely heals and finishes. Pause snapshot maintenance in the mean 373 // time to prevent access. 374 if snapshots := d.blockchain.Snapshots(); snapshots != nil { // Only nil in tests 375 snapshots.Disable() 376 } 377 log.Warn("Enabling snapshot sync prototype") 378 d.snapSync = true 379 } 380 mode = FastSync 381 } 382 // Reset the queue, peer set and wake channels to clean any internal leftover state 383 d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems) 384 d.peers.Reset() 385 386 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 387 select { 388 case <-ch: 389 default: 390 } 391 } 392 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 393 for empty := false; !empty; { 394 select { 395 case <-ch: 396 default: 397 empty = true 398 } 399 } 400 } 401 for empty := false; !empty; { 402 select { 403 case <-d.headerProcCh: 404 default: 405 empty = true 406 } 407 } 408 // Create cancel channel for aborting mid-flight and mark the master peer 409 d.cancelLock.Lock() 410 d.cancelCh = make(chan struct{}) 411 d.cancelPeer = id 412 d.cancelLock.Unlock() 413 414 defer d.Cancel() // No matter what, we can't leave the cancel channel open 415 416 // Atomically set the requested sync mode 417 atomic.StoreUint32(&d.mode, uint32(mode)) 418 419 // Retrieve the origin peer and initiate the downloading process 420 p := d.peers.Peer(id) 421 if p == nil { 422 return errUnknownPeer 423 } 424 return d.syncWithPeer(p, hash, td) 425 } 426 427 func (d *Downloader) getMode() SyncMode { 428 return SyncMode(atomic.LoadUint32(&d.mode)) 429 } 430 431 // syncWithPeer starts a block synchronization based on the hash chain from the 432 // specified peer and head hash. 433 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 434 d.mux.Post(StartEvent{}) 435 defer func() { 436 // reset on error 437 if err != nil { 438 d.mux.Post(FailedEvent{err}) 439 } else { 440 latest := d.lightchain.CurrentHeader() 441 d.mux.Post(DoneEvent{latest}) 442 } 443 }() 444 if p.version < eth.ETH66 { 445 return fmt.Errorf("%w: advertized %d < required %d", errTooOld, p.version, eth.ETH66) 446 } 447 mode := d.getMode() 448 449 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", mode) 450 defer func(start time.Time) { 451 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 452 }(time.Now()) 453 454 // Look up the sync boundaries: the common ancestor and the target block 455 latest, pivot, err := d.fetchHead(p) 456 if err != nil { 457 return err 458 } 459 if mode == FastSync && pivot == nil { 460 // If no pivot block was returned, the head is below the min full block 461 // threshold (i.e. new chain). In that case we won't really fast sync 462 // anyway, but still need a valid pivot block to avoid some code hitting 463 // nil panics on an access. 464 pivot = d.blockchain.CurrentBlock().Header() 465 } 466 height := latest.Number.Uint64() 467 468 origin, err := d.findAncestor(p, latest) 469 if err != nil { 470 return err 471 } 472 d.syncStatsLock.Lock() 473 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 474 d.syncStatsChainOrigin = origin 475 } 476 d.syncStatsChainHeight = height 477 d.syncStatsLock.Unlock() 478 479 // Ensure our origin point is below any fast sync pivot point 480 if mode == FastSync { 481 if height <= uint64(fsMinFullBlocks) { 482 origin = 0 483 } else { 484 pivotNumber := pivot.Number.Uint64() 485 if pivotNumber <= origin { 486 origin = pivotNumber - 1 487 } 488 // Write out the pivot into the database so a rollback beyond it will 489 // reenable fast sync 490 rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber) 491 } 492 } 493 d.committed = 1 494 if mode == FastSync && pivot.Number.Uint64() != 0 { 495 d.committed = 0 496 } 497 if mode == FastSync { 498 // Set the ancient data limitation. 499 // If we are running fast sync, all block data older than ancientLimit will be 500 // written to the ancient store. More recent data will be written to the active 501 // database and will wait for the freezer to migrate. 502 // 503 // If there is a checkpoint available, then calculate the ancientLimit through 504 // that. Otherwise calculate the ancient limit through the advertised height 505 // of the remote peer. 506 // 507 // The reason for picking checkpoint first is that a malicious peer can give us 508 // a fake (very high) height, forcing the ancient limit to also be very high. 509 // The peer would start to feed us valid blocks until head, resulting in all of 510 // the blocks might be written into the ancient store. A following mini-reorg 511 // could cause issues. 512 if d.checkpoint != 0 && d.checkpoint > fullMaxForkAncestry+1 { 513 d.ancientLimit = d.checkpoint 514 } else if height > fullMaxForkAncestry+1 { 515 d.ancientLimit = height - fullMaxForkAncestry - 1 516 } else { 517 d.ancientLimit = 0 518 } 519 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 520 521 // If a part of blockchain data has already been written into active store, 522 // disable the ancient style insertion explicitly. 523 if origin >= frozen && frozen != 0 { 524 d.ancientLimit = 0 525 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 526 } else if d.ancientLimit > 0 { 527 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 528 } 529 // Rewind the ancient store and blockchain if reorg happens. 530 if origin+1 < frozen { 531 if err := d.lightchain.SetHead(origin + 1); err != nil { 532 return err 533 } 534 } 535 } 536 // Initiate the sync using a concurrent header and content retrieval algorithm 537 d.queue.Prepare(origin+1, mode) 538 if d.syncInitHook != nil { 539 d.syncInitHook(origin, height) 540 } 541 fetchers := []func() error{ 542 func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved 543 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 544 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 545 func() error { return d.processHeaders(origin+1, td) }, 546 } 547 if mode == FastSync { 548 d.pivotLock.Lock() 549 d.pivotHeader = pivot 550 d.pivotLock.Unlock() 551 552 fetchers = append(fetchers, func() error { return d.processFastSyncContent() }) 553 } else if mode == FullSync { 554 fetchers = append(fetchers, d.processFullSyncContent) 555 } 556 return d.spawnSync(fetchers) 557 } 558 559 // spawnSync runs d.process and all given fetcher functions to completion in 560 // separate goroutines, returning the first error that appears. 561 func (d *Downloader) spawnSync(fetchers []func() error) error { 562 errc := make(chan error, len(fetchers)) 563 d.cancelWg.Add(len(fetchers)) 564 for _, fn := range fetchers { 565 fn := fn 566 go func() { defer d.cancelWg.Done(); errc <- fn() }() 567 } 568 // Wait for the first error, then terminate the others. 569 var err error 570 for i := 0; i < len(fetchers); i++ { 571 if i == len(fetchers)-1 { 572 // Close the queue when all fetchers have exited. 573 // This will cause the block processor to end when 574 // it has processed the queue. 575 d.queue.Close() 576 } 577 if err = <-errc; err != nil && err != errCanceled { 578 break 579 } 580 } 581 d.queue.Close() 582 d.Cancel() 583 return err 584 } 585 586 // cancel aborts all of the operations and resets the queue. However, cancel does 587 // not wait for the running download goroutines to finish. This method should be 588 // used when cancelling the downloads from inside the downloader. 589 func (d *Downloader) cancel() { 590 // Close the current cancel channel 591 d.cancelLock.Lock() 592 defer d.cancelLock.Unlock() 593 594 if d.cancelCh != nil { 595 select { 596 case <-d.cancelCh: 597 // Channel was already closed 598 default: 599 close(d.cancelCh) 600 } 601 } 602 } 603 604 // Cancel aborts all of the operations and waits for all download goroutines to 605 // finish before returning. 606 func (d *Downloader) Cancel() { 607 d.cancel() 608 d.cancelWg.Wait() 609 } 610 611 // Terminate interrupts the downloader, canceling all pending operations. 612 // The downloader cannot be reused after calling Terminate. 613 func (d *Downloader) Terminate() { 614 // Close the termination channel (make sure double close is allowed) 615 d.quitLock.Lock() 616 select { 617 case <-d.quitCh: 618 default: 619 close(d.quitCh) 620 } 621 d.quitLock.Unlock() 622 623 // Cancel any pending download requests 624 d.Cancel() 625 } 626 627 // fetchHead retrieves the head header and prior pivot block (if available) from 628 // a remote peer. 629 func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *types.Header, err error) { 630 p.log.Debug("Retrieving remote chain head") 631 mode := d.getMode() 632 633 // Request the advertised remote head block and wait for the response 634 latest, _ := p.peer.Head() 635 fetch := 1 636 if mode == FastSync { 637 fetch = 2 // head + pivot headers 638 } 639 go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true) 640 641 ttl := d.peers.rates.TargetTimeout() 642 timeout := time.After(ttl) 643 for { 644 select { 645 case <-d.cancelCh: 646 return nil, nil, errCanceled 647 648 case packet := <-d.headerCh: 649 // Discard anything not from the origin peer 650 if packet.PeerId() != p.id { 651 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 652 break 653 } 654 // Make sure the peer gave us at least one and at most the requested headers 655 headers := packet.(*headerPack).headers 656 if len(headers) == 0 || len(headers) > fetch { 657 return nil, nil, fmt.Errorf("%w: returned headers %d != requested %d", errBadPeer, len(headers), fetch) 658 } 659 // The first header needs to be the head, validate against the checkpoint 660 // and request. If only 1 header was returned, make sure there's no pivot 661 // or there was not one requested. 662 head := headers[0] 663 if (mode == FastSync || mode == LightSync) && head.Number.Uint64() < d.checkpoint { 664 return nil, nil, fmt.Errorf("%w: remote head %d below checkpoint %d", errUnsyncedPeer, head.Number, d.checkpoint) 665 } 666 if len(headers) == 1 { 667 if mode == FastSync && head.Number.Uint64() > uint64(fsMinFullBlocks) { 668 return nil, nil, fmt.Errorf("%w: no pivot included along head header", errBadPeer) 669 } 670 p.log.Debug("Remote head identified, no pivot", "number", head.Number, "hash", head.Hash()) 671 return head, nil, nil 672 } 673 // At this point we have 2 headers in total and the first is the 674 // validated head of the chain. Check the pivot number and return, 675 pivot := headers[1] 676 if pivot.Number.Uint64() != head.Number.Uint64()-uint64(fsMinFullBlocks) { 677 return nil, nil, fmt.Errorf("%w: remote pivot %d != requested %d", errInvalidChain, pivot.Number, head.Number.Uint64()-uint64(fsMinFullBlocks)) 678 } 679 return head, pivot, nil 680 681 case <-timeout: 682 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 683 return nil, nil, errTimeout 684 685 case <-d.bodyCh: 686 case <-d.receiptCh: 687 // Out of bounds delivery, ignore 688 } 689 } 690 } 691 692 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 693 // common ancestor. 694 // It returns parameters to be used for peer.RequestHeadersByNumber: 695 // from - starting block number 696 // count - number of headers to request 697 // skip - number of headers to skip 698 // and also returns 'max', the last block which is expected to be returned by the remote peers, 699 // given the (from,count,skip) 700 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 701 var ( 702 from int 703 count int 704 MaxCount = MaxHeaderFetch / 16 705 ) 706 // requestHead is the highest block that we will ask for. If requestHead is not offset, 707 // the highest block that we will get is 16 blocks back from head, which means we 708 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 709 // between us and the peer is 1-2 blocks, which is most common 710 requestHead := int(remoteHeight) - 1 711 if requestHead < 0 { 712 requestHead = 0 713 } 714 // requestBottom is the lowest block we want included in the query 715 // Ideally, we want to include the one just below our own head 716 requestBottom := int(localHeight - 1) 717 if requestBottom < 0 { 718 requestBottom = 0 719 } 720 totalSpan := requestHead - requestBottom 721 span := 1 + totalSpan/MaxCount 722 if span < 2 { 723 span = 2 724 } 725 if span > 16 { 726 span = 16 727 } 728 729 count = 1 + totalSpan/span 730 if count > MaxCount { 731 count = MaxCount 732 } 733 if count < 2 { 734 count = 2 735 } 736 from = requestHead - (count-1)*span 737 if from < 0 { 738 from = 0 739 } 740 max := from + (count-1)*span 741 return int64(from), count, span - 1, uint64(max) 742 } 743 744 // findAncestor tries to locate the common ancestor link of the local chain and 745 // a remote peers blockchain. In the general case when our node was in sync and 746 // on the correct chain, checking the top N links should already get us a match. 747 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 748 // the head links match), we do a binary search to find the common ancestor. 749 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 750 // Figure out the valid ancestor range to prevent rewrite attacks 751 var ( 752 floor = int64(-1) 753 localHeight uint64 754 remoteHeight = remoteHeader.Number.Uint64() 755 ) 756 mode := d.getMode() 757 switch mode { 758 case FullSync: 759 localHeight = d.blockchain.CurrentBlock().NumberU64() 760 case FastSync: 761 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 762 default: 763 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 764 } 765 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 766 767 // Recap floor value for binary search 768 maxForkAncestry := fullMaxForkAncestry 769 if d.getMode() == LightSync { 770 maxForkAncestry = lightMaxForkAncestry 771 } 772 if localHeight >= maxForkAncestry { 773 // We're above the max reorg threshold, find the earliest fork point 774 floor = int64(localHeight - maxForkAncestry) 775 } 776 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 777 // all headers before that point will be missing. 778 if mode == LightSync { 779 // If we don't know the current CHT position, find it 780 if d.genesis == 0 { 781 header := d.lightchain.CurrentHeader() 782 for header != nil { 783 d.genesis = header.Number.Uint64() 784 if floor >= int64(d.genesis)-1 { 785 break 786 } 787 header = d.lightchain.GetHeaderByHash(header.ParentHash) 788 } 789 } 790 // We already know the "genesis" block number, cap floor to that 791 if floor < int64(d.genesis)-1 { 792 floor = int64(d.genesis) - 1 793 } 794 } 795 796 ancestor, err := d.findAncestorSpanSearch(p, mode, remoteHeight, localHeight, floor) 797 if err == nil { 798 return ancestor, nil 799 } 800 // The returned error was not nil. 801 // If the error returned does not reflect that a common ancestor was not found, return it. 802 // If the error reflects that a common ancestor was not found, continue to binary search, 803 // where the error value will be reassigned. 804 if !errors.Is(err, errNoAncestorFound) { 805 return 0, err 806 } 807 808 ancestor, err = d.findAncestorBinarySearch(p, mode, remoteHeight, floor) 809 if err != nil { 810 return 0, err 811 } 812 return ancestor, nil 813 } 814 815 func (d *Downloader) findAncestorSpanSearch(p *peerConnection, mode SyncMode, remoteHeight, localHeight uint64, floor int64) (commonAncestor uint64, err error) { 816 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 817 818 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 819 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 820 821 // Wait for the remote response to the head fetch 822 number, hash := uint64(0), common.Hash{} 823 824 ttl := d.peers.rates.TargetTimeout() 825 timeout := time.After(ttl) 826 827 for finished := false; !finished; { 828 select { 829 case <-d.cancelCh: 830 return 0, errCanceled 831 832 case packet := <-d.headerCh: 833 // Discard anything not from the origin peer 834 if packet.PeerId() != p.id { 835 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 836 break 837 } 838 // Make sure the peer actually gave something valid 839 headers := packet.(*headerPack).headers 840 if len(headers) == 0 { 841 p.log.Warn("Empty head header set") 842 return 0, errEmptyHeaderSet 843 } 844 // Make sure the peer's reply conforms to the request 845 for i, header := range headers { 846 expectNumber := from + int64(i)*int64(skip+1) 847 if number := header.Number.Int64(); number != expectNumber { 848 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 849 return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering")) 850 } 851 } 852 // Check if a common ancestor was found 853 finished = true 854 for i := len(headers) - 1; i >= 0; i-- { 855 // Skip any headers that underflow/overflow our requested set 856 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 857 continue 858 } 859 // Otherwise check if we already know the header or not 860 h := headers[i].Hash() 861 n := headers[i].Number.Uint64() 862 863 var known bool 864 switch mode { 865 case FullSync: 866 known = d.blockchain.HasBlock(h, n) 867 case FastSync: 868 known = d.blockchain.HasFastBlock(h, n) 869 default: 870 known = d.lightchain.HasHeader(h, n) 871 } 872 if known { 873 number, hash = n, h 874 break 875 } 876 } 877 878 case <-timeout: 879 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 880 return 0, errTimeout 881 882 case <-d.bodyCh: 883 case <-d.receiptCh: 884 // Out of bounds delivery, ignore 885 } 886 } 887 // If the head fetch already found an ancestor, return 888 if hash != (common.Hash{}) { 889 if int64(number) <= floor { 890 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 891 return 0, errInvalidAncestor 892 } 893 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 894 return number, nil 895 } 896 return 0, errNoAncestorFound 897 } 898 899 func (d *Downloader) findAncestorBinarySearch(p *peerConnection, mode SyncMode, remoteHeight uint64, floor int64) (commonAncestor uint64, err error) { 900 hash := common.Hash{} 901 902 // Ancestor not found, we need to binary search over our chain 903 start, end := uint64(0), remoteHeight 904 if floor > 0 { 905 start = uint64(floor) 906 } 907 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 908 909 for start+1 < end { 910 // Split our chain interval in two, and request the hash to cross check 911 check := (start + end) / 2 912 913 ttl := d.peers.rates.TargetTimeout() 914 timeout := time.After(ttl) 915 916 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 917 918 // Wait until a reply arrives to this request 919 for arrived := false; !arrived; { 920 select { 921 case <-d.cancelCh: 922 return 0, errCanceled 923 924 case packet := <-d.headerCh: 925 // Discard anything not from the origin peer 926 if packet.PeerId() != p.id { 927 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 928 break 929 } 930 // Make sure the peer actually gave something valid 931 headers := packet.(*headerPack).headers 932 if len(headers) != 1 { 933 p.log.Warn("Multiple headers for single request", "headers", len(headers)) 934 return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers)) 935 } 936 arrived = true 937 938 // Modify the search interval based on the response 939 h := headers[0].Hash() 940 n := headers[0].Number.Uint64() 941 942 var known bool 943 switch mode { 944 case FullSync: 945 known = d.blockchain.HasBlock(h, n) 946 case FastSync: 947 known = d.blockchain.HasFastBlock(h, n) 948 default: 949 known = d.lightchain.HasHeader(h, n) 950 } 951 if !known { 952 end = check 953 break 954 } 955 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 956 if header.Number.Uint64() != check { 957 p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 958 return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number) 959 } 960 start = check 961 hash = h 962 963 case <-timeout: 964 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 965 return 0, errTimeout 966 967 case <-d.bodyCh: 968 case <-d.receiptCh: 969 // Out of bounds delivery, ignore 970 } 971 } 972 } 973 // Ensure valid ancestry and return 974 if int64(start) <= floor { 975 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 976 return 0, errInvalidAncestor 977 } 978 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 979 return start, nil 980 } 981 982 // fetchHeaders keeps retrieving headers concurrently from the number 983 // requested, until no more are returned, potentially throttling on the way. To 984 // facilitate concurrency but still protect against malicious nodes sending bad 985 // headers, we construct a header chain skeleton using the "origin" peer we are 986 // syncing with, and fill in the missing headers using anyone else. Headers from 987 // other peers are only accepted if they map cleanly to the skeleton. If no one 988 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 989 // the origin is dropped. 990 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error { 991 p.log.Debug("Directing header downloads", "origin", from) 992 defer p.log.Debug("Header download terminated") 993 994 // Create a timeout timer, and the associated header fetcher 995 skeleton := true // Skeleton assembly phase or finishing up 996 pivoting := false // Whether the next request is pivot verification 997 request := time.Now() // time of the last skeleton fetch request 998 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 999 <-timeout.C // timeout channel should be initially empty 1000 defer timeout.Stop() 1001 1002 var ttl time.Duration 1003 getHeaders := func(from uint64) { 1004 request = time.Now() 1005 1006 ttl = d.peers.rates.TargetTimeout() 1007 timeout.Reset(ttl) 1008 1009 if skeleton { 1010 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 1011 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 1012 } else { 1013 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 1014 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 1015 } 1016 } 1017 getNextPivot := func() { 1018 pivoting = true 1019 request = time.Now() 1020 1021 ttl = d.peers.rates.TargetTimeout() 1022 timeout.Reset(ttl) 1023 1024 d.pivotLock.RLock() 1025 pivot := d.pivotHeader.Number.Uint64() 1026 d.pivotLock.RUnlock() 1027 1028 p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks)) 1029 go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep 1030 } 1031 // Start pulling the header chain skeleton until all is done 1032 ancestor := from 1033 getHeaders(from) 1034 1035 mode := d.getMode() 1036 for { 1037 select { 1038 case <-d.cancelCh: 1039 return errCanceled 1040 1041 case packet := <-d.headerCh: 1042 // Make sure the active peer is giving us the skeleton headers 1043 if packet.PeerId() != p.id { 1044 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 1045 break 1046 } 1047 headerReqTimer.UpdateSince(request) 1048 timeout.Stop() 1049 1050 // If the pivot is being checked, move if it became stale and run the real retrieval 1051 var pivot uint64 1052 1053 d.pivotLock.RLock() 1054 if d.pivotHeader != nil { 1055 pivot = d.pivotHeader.Number.Uint64() 1056 } 1057 d.pivotLock.RUnlock() 1058 1059 if pivoting { 1060 if packet.Items() == 2 { 1061 // Retrieve the headers and do some sanity checks, just in case 1062 headers := packet.(*headerPack).headers 1063 1064 if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want { 1065 log.Warn("Peer sent invalid next pivot", "have", have, "want", want) 1066 return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want) 1067 } 1068 if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want { 1069 log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want) 1070 return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want) 1071 } 1072 log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number) 1073 pivot = headers[0].Number.Uint64() 1074 1075 d.pivotLock.Lock() 1076 d.pivotHeader = headers[0] 1077 d.pivotLock.Unlock() 1078 1079 // Write out the pivot into the database so a rollback beyond 1080 // it will reenable fast sync and update the state root that 1081 // the state syncer will be downloading. 1082 rawdb.WriteLastPivotNumber(d.stateDB, pivot) 1083 } 1084 pivoting = false 1085 getHeaders(from) 1086 continue 1087 } 1088 // If the skeleton's finished, pull any remaining head headers directly from the origin 1089 if skeleton && packet.Items() == 0 { 1090 skeleton = false 1091 getHeaders(from) 1092 continue 1093 } 1094 // If no more headers are inbound, notify the content fetchers and return 1095 if packet.Items() == 0 { 1096 // Don't abort header fetches while the pivot is downloading 1097 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 1098 p.log.Debug("No headers, waiting for pivot commit") 1099 select { 1100 case <-time.After(fsHeaderContCheck): 1101 getHeaders(from) 1102 continue 1103 case <-d.cancelCh: 1104 return errCanceled 1105 } 1106 } 1107 // Pivot done (or not in fast sync) and no more headers, terminate the process 1108 p.log.Debug("No more headers available") 1109 select { 1110 case d.headerProcCh <- nil: 1111 return nil 1112 case <-d.cancelCh: 1113 return errCanceled 1114 } 1115 } 1116 headers := packet.(*headerPack).headers 1117 1118 // If we received a skeleton batch, resolve internals concurrently 1119 if skeleton { 1120 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1121 if err != nil { 1122 p.log.Debug("Skeleton chain invalid", "err", err) 1123 return fmt.Errorf("%w: %v", errInvalidChain, err) 1124 } 1125 headers = filled[proced:] 1126 from += uint64(proced) 1127 } else { 1128 // If we're closing in on the chain head, but haven't yet reached it, delay 1129 // the last few headers so mini reorgs on the head don't cause invalid hash 1130 // chain errors. 1131 if n := len(headers); n > 0 { 1132 // Retrieve the current head we're at 1133 var head uint64 1134 if mode == LightSync { 1135 head = d.lightchain.CurrentHeader().Number.Uint64() 1136 } else { 1137 head = d.blockchain.CurrentFastBlock().NumberU64() 1138 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1139 head = full 1140 } 1141 } 1142 // If the head is below the common ancestor, we're actually deduplicating 1143 // already existing chain segments, so use the ancestor as the fake head. 1144 // Otherwise we might end up delaying header deliveries pointlessly. 1145 if head < ancestor { 1146 head = ancestor 1147 } 1148 // If the head is way older than this batch, delay the last few headers 1149 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1150 delay := reorgProtHeaderDelay 1151 if delay > n { 1152 delay = n 1153 } 1154 headers = headers[:n-delay] 1155 } 1156 } 1157 } 1158 // Insert all the new headers and fetch the next batch 1159 if len(headers) > 0 { 1160 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1161 select { 1162 case d.headerProcCh <- headers: 1163 case <-d.cancelCh: 1164 return errCanceled 1165 } 1166 from += uint64(len(headers)) 1167 1168 // If we're still skeleton filling fast sync, check pivot staleness 1169 // before continuing to the next skeleton filling 1170 if skeleton && pivot > 0 { 1171 getNextPivot() 1172 } else { 1173 getHeaders(from) 1174 } 1175 } else { 1176 // No headers delivered, or all of them being delayed, sleep a bit and retry 1177 p.log.Trace("All headers delayed, waiting") 1178 select { 1179 case <-time.After(fsHeaderContCheck): 1180 getHeaders(from) 1181 continue 1182 case <-d.cancelCh: 1183 return errCanceled 1184 } 1185 } 1186 1187 case <-timeout.C: 1188 if d.dropPeer == nil { 1189 // The dropPeer method is nil when `--copydb` is used for a local copy. 1190 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1191 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1192 break 1193 } 1194 // Header retrieval timed out, consider the peer bad and drop 1195 p.log.Debug("Header request timed out", "elapsed", ttl) 1196 headerTimeoutMeter.Mark(1) 1197 d.dropPeer(p.id) 1198 1199 // Finish the sync gracefully instead of dumping the gathered data though 1200 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1201 select { 1202 case ch <- false: 1203 case <-d.cancelCh: 1204 } 1205 } 1206 select { 1207 case d.headerProcCh <- nil: 1208 case <-d.cancelCh: 1209 } 1210 return fmt.Errorf("%w: header request timed out", errBadPeer) 1211 } 1212 } 1213 } 1214 1215 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1216 // and maps them to the provided skeleton header chain. 1217 // 1218 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1219 // immediately to the header processor to keep the rest of the pipeline full even 1220 // in the case of header stalls. 1221 // 1222 // The method returns the entire filled skeleton and also the number of headers 1223 // already forwarded for processing. 1224 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1225 log.Debug("Filling up skeleton", "from", from) 1226 d.queue.ScheduleSkeleton(from, skeleton) 1227 1228 var ( 1229 deliver = func(packet dataPack) (int, error) { 1230 pack := packet.(*headerPack) 1231 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1232 } 1233 expire = func() map[string]int { return d.queue.ExpireHeaders(d.peers.rates.TargetTimeout()) } 1234 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) { 1235 return d.queue.ReserveHeaders(p, count), false, false 1236 } 1237 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1238 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.peers.rates.TargetRoundTrip()) } 1239 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1240 p.SetHeadersIdle(accepted, deliveryTime) 1241 } 1242 ) 1243 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1244 d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve, 1245 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1246 1247 log.Debug("Skeleton fill terminated", "err", err) 1248 1249 filled, proced := d.queue.RetrieveHeaders() 1250 return filled, proced, err 1251 } 1252 1253 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1254 // available peers, reserving a chunk of blocks for each, waiting for delivery 1255 // and also periodically checking for timeouts. 1256 func (d *Downloader) fetchBodies(from uint64) error { 1257 log.Debug("Downloading block bodies", "origin", from) 1258 1259 var ( 1260 deliver = func(packet dataPack) (int, error) { 1261 pack := packet.(*bodyPack) 1262 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1263 } 1264 expire = func() map[string]int { return d.queue.ExpireBodies(d.peers.rates.TargetTimeout()) } 1265 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1266 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.peers.rates.TargetRoundTrip()) } 1267 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) } 1268 ) 1269 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1270 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies, 1271 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1272 1273 log.Debug("Block body download terminated", "err", err) 1274 return err 1275 } 1276 1277 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1278 // available peers, reserving a chunk of receipts for each, waiting for delivery 1279 // and also periodically checking for timeouts. 1280 func (d *Downloader) fetchReceipts(from uint64) error { 1281 log.Debug("Downloading transaction receipts", "origin", from) 1282 1283 var ( 1284 deliver = func(packet dataPack) (int, error) { 1285 pack := packet.(*receiptPack) 1286 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1287 } 1288 expire = func() map[string]int { return d.queue.ExpireReceipts(d.peers.rates.TargetTimeout()) } 1289 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1290 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.peers.rates.TargetRoundTrip()) } 1291 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1292 p.SetReceiptsIdle(accepted, deliveryTime) 1293 } 1294 ) 1295 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1296 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts, 1297 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1298 1299 log.Debug("Transaction receipt download terminated", "err", err) 1300 return err 1301 } 1302 1303 // fetchParts iteratively downloads scheduled block parts, taking any available 1304 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1305 // also periodically checking for timeouts. 1306 // 1307 // As the scheduling/timeout logic mostly is the same for all downloaded data 1308 // types, this method is used by each for data gathering and is instrumented with 1309 // various callbacks to handle the slight differences between processing them. 1310 // 1311 // The instrumentation parameters: 1312 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1313 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1314 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1315 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1316 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1317 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1318 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1319 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1320 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1321 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1322 // - fetch: network callback to actually send a particular download request to a physical remote peer 1323 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1324 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1325 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1326 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1327 // - kind: textual label of the type being downloaded to display in log messages 1328 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1329 expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool), 1330 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1331 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error { 1332 // Create a ticker to detect expired retrieval tasks 1333 ticker := time.NewTicker(100 * time.Millisecond) 1334 defer ticker.Stop() 1335 1336 update := make(chan struct{}, 1) 1337 1338 // Prepare the queue and fetch block parts until the block header fetcher's done 1339 finished := false 1340 for { 1341 select { 1342 case <-d.cancelCh: 1343 return errCanceled 1344 1345 case packet := <-deliveryCh: 1346 deliveryTime := time.Now() 1347 // If the peer was previously banned and failed to deliver its pack 1348 // in a reasonable time frame, ignore its message. 1349 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1350 // Deliver the received chunk of data and check chain validity 1351 accepted, err := deliver(packet) 1352 if errors.Is(err, errInvalidChain) { 1353 return err 1354 } 1355 // Unless a peer delivered something completely else than requested (usually 1356 // caused by a timed out request which came through in the end), set it to 1357 // idle. If the delivery's stale, the peer should have already been idled. 1358 if !errors.Is(err, errStaleDelivery) { 1359 setIdle(peer, accepted, deliveryTime) 1360 } 1361 // Issue a log to the user to see what's going on 1362 switch { 1363 case err == nil && packet.Items() == 0: 1364 peer.log.Trace("Requested data not delivered", "type", kind) 1365 case err == nil: 1366 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1367 default: 1368 peer.log.Debug("Failed to deliver retrieved data", "type", kind, "err", err) 1369 } 1370 } 1371 // Blocks assembled, try to update the progress 1372 select { 1373 case update <- struct{}{}: 1374 default: 1375 } 1376 1377 case cont := <-wakeCh: 1378 // The header fetcher sent a continuation flag, check if it's done 1379 if !cont { 1380 finished = true 1381 } 1382 // Headers arrive, try to update the progress 1383 select { 1384 case update <- struct{}{}: 1385 default: 1386 } 1387 1388 case <-ticker.C: 1389 // Sanity check update the progress 1390 select { 1391 case update <- struct{}{}: 1392 default: 1393 } 1394 1395 case <-update: 1396 // Short circuit if we lost all our peers 1397 if d.peers.Len() == 0 { 1398 return errNoPeers 1399 } 1400 // Check for fetch request timeouts and demote the responsible peers 1401 for pid, fails := range expire() { 1402 if peer := d.peers.Peer(pid); peer != nil { 1403 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1404 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1405 // out that sync wise we need to get rid of the peer. 1406 // 1407 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1408 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1409 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1410 if fails > 2 { 1411 peer.log.Trace("Data delivery timed out", "type", kind) 1412 setIdle(peer, 0, time.Now()) 1413 } else { 1414 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1415 1416 if d.dropPeer == nil { 1417 // The dropPeer method is nil when `--copydb` is used for a local copy. 1418 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1419 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1420 } else { 1421 d.dropPeer(pid) 1422 1423 // If this peer was the master peer, abort sync immediately 1424 d.cancelLock.RLock() 1425 master := pid == d.cancelPeer 1426 d.cancelLock.RUnlock() 1427 1428 if master { 1429 d.cancel() 1430 return errTimeout 1431 } 1432 } 1433 } 1434 } 1435 } 1436 // If there's nothing more to fetch, wait or terminate 1437 if pending() == 0 { 1438 if !inFlight() && finished { 1439 log.Debug("Data fetching completed", "type", kind) 1440 return nil 1441 } 1442 break 1443 } 1444 // Send a download request to all idle peers, until throttled 1445 progressed, throttled, running := false, false, inFlight() 1446 idles, total := idle() 1447 pendCount := pending() 1448 for _, peer := range idles { 1449 // Short circuit if throttling activated 1450 if throttled { 1451 break 1452 } 1453 // Short circuit if there is no more available task. 1454 if pendCount = pending(); pendCount == 0 { 1455 break 1456 } 1457 // Reserve a chunk of fetches for a peer. A nil can mean either that 1458 // no more headers are available, or that the peer is known not to 1459 // have them. 1460 request, progress, throttle := reserve(peer, capacity(peer)) 1461 if progress { 1462 progressed = true 1463 } 1464 if throttle { 1465 throttled = true 1466 throttleCounter.Inc(1) 1467 } 1468 if request == nil { 1469 continue 1470 } 1471 if request.From > 0 { 1472 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1473 } else { 1474 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1475 } 1476 // Fetch the chunk and make sure any errors return the hashes to the queue 1477 if fetchHook != nil { 1478 fetchHook(request.Headers) 1479 } 1480 if err := fetch(peer, request); err != nil { 1481 // Although we could try and make an attempt to fix this, this error really 1482 // means that we've double allocated a fetch task to a peer. If that is the 1483 // case, the internal state of the downloader and the queue is very wrong so 1484 // better hard crash and note the error instead of silently accumulating into 1485 // a much bigger issue. 1486 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1487 } 1488 running = true 1489 } 1490 // Make sure that we have peers available for fetching. If all peers have been tried 1491 // and all failed throw an error 1492 if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 { 1493 return errPeersUnavailable 1494 } 1495 } 1496 } 1497 } 1498 1499 // processHeaders takes batches of retrieved headers from an input channel and 1500 // keeps processing and scheduling them into the header chain and downloader's 1501 // queue until the stream ends or a failure occurs. 1502 func (d *Downloader) processHeaders(origin uint64, td *big.Int) error { 1503 // Keep a count of uncertain headers to roll back 1504 var ( 1505 rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis) 1506 rollbackErr error 1507 mode = d.getMode() 1508 ) 1509 defer func() { 1510 if rollback > 0 { 1511 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1512 if mode != LightSync { 1513 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1514 lastBlock = d.blockchain.CurrentBlock().Number() 1515 } 1516 if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block 1517 // We're already unwinding the stack, only print the error to make it more visible 1518 log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err) 1519 } 1520 curFastBlock, curBlock := common.Big0, common.Big0 1521 if mode != LightSync { 1522 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1523 curBlock = d.blockchain.CurrentBlock().Number() 1524 } 1525 log.Warn("Rolled back chain segment", 1526 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1527 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1528 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr) 1529 } 1530 }() 1531 // Wait for batches of headers to process 1532 gotHeaders := false 1533 1534 for { 1535 select { 1536 case <-d.cancelCh: 1537 rollbackErr = errCanceled 1538 return errCanceled 1539 1540 case headers := <-d.headerProcCh: 1541 // Terminate header processing if we synced up 1542 if len(headers) == 0 { 1543 // Notify everyone that headers are fully processed 1544 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1545 select { 1546 case ch <- false: 1547 case <-d.cancelCh: 1548 } 1549 } 1550 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1551 // better chain compared to ours. The only exception is if its promised blocks were 1552 // already imported by other means (e.g. fetcher): 1553 // 1554 // R <remote peer>, L <local node>: Both at block 10 1555 // R: Mine block 11, and propagate it to L 1556 // L: Queue block 11 for import 1557 // L: Notice that R's head and TD increased compared to ours, start sync 1558 // L: Import of block 11 finishes 1559 // L: Sync begins, and finds common ancestor at 11 1560 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1561 // R: Nothing to give 1562 if mode != LightSync { 1563 head := d.blockchain.CurrentBlock() 1564 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1565 return errStallingPeer 1566 } 1567 } 1568 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1569 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1570 // of delivering the post-pivot blocks that would flag the invalid content. 1571 // 1572 // This check cannot be executed "as is" for full imports, since blocks may still be 1573 // queued for processing when the header download completes. However, as long as the 1574 // peer gave us something useful, we're already happy/progressed (above check). 1575 if mode == FastSync || mode == LightSync { 1576 head := d.lightchain.CurrentHeader() 1577 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1578 return errStallingPeer 1579 } 1580 } 1581 // Disable any rollback and return 1582 rollback = 0 1583 return nil 1584 } 1585 // Otherwise split the chunk of headers into batches and process them 1586 gotHeaders = true 1587 for len(headers) > 0 { 1588 // Terminate if something failed in between processing chunks 1589 select { 1590 case <-d.cancelCh: 1591 rollbackErr = errCanceled 1592 return errCanceled 1593 default: 1594 } 1595 // Select the next chunk of headers to import 1596 limit := maxHeadersProcess 1597 if limit > len(headers) { 1598 limit = len(headers) 1599 } 1600 chunk := headers[:limit] 1601 1602 // In case of header only syncing, validate the chunk immediately 1603 if mode == FastSync || mode == LightSync { 1604 // If we're importing pure headers, verify based on their recentness 1605 var pivot uint64 1606 1607 d.pivotLock.RLock() 1608 if d.pivotHeader != nil { 1609 pivot = d.pivotHeader.Number.Uint64() 1610 } 1611 d.pivotLock.RUnlock() 1612 1613 frequency := fsHeaderCheckFrequency 1614 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1615 frequency = 1 1616 } 1617 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1618 rollbackErr = err 1619 1620 // If some headers were inserted, track them as uncertain 1621 if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 { 1622 rollback = chunk[0].Number.Uint64() 1623 } 1624 log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err) 1625 return fmt.Errorf("%w: %v", errInvalidChain, err) 1626 } 1627 // All verifications passed, track all headers within the allotted limits 1628 if mode == FastSync { 1629 head := chunk[len(chunk)-1].Number.Uint64() 1630 if head-rollback > uint64(fsHeaderSafetyNet) { 1631 rollback = head - uint64(fsHeaderSafetyNet) 1632 } else { 1633 rollback = 1 1634 } 1635 } 1636 } 1637 // Unless we're doing light chains, schedule the headers for associated content retrieval 1638 if mode == FullSync || mode == FastSync { 1639 // If we've reached the allowed number of pending headers, stall a bit 1640 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1641 select { 1642 case <-d.cancelCh: 1643 rollbackErr = errCanceled 1644 return errCanceled 1645 case <-time.After(time.Second): 1646 } 1647 } 1648 // Otherwise insert the headers for content retrieval 1649 inserts := d.queue.Schedule(chunk, origin) 1650 if len(inserts) != len(chunk) { 1651 rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk)) 1652 return fmt.Errorf("%w: stale headers", errBadPeer) 1653 } 1654 } 1655 headers = headers[limit:] 1656 origin += uint64(limit) 1657 } 1658 // Update the highest block number we know if a higher one is found. 1659 d.syncStatsLock.Lock() 1660 if d.syncStatsChainHeight < origin { 1661 d.syncStatsChainHeight = origin - 1 1662 } 1663 d.syncStatsLock.Unlock() 1664 1665 // Signal the content downloaders of the availability of new tasks 1666 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1667 select { 1668 case ch <- true: 1669 default: 1670 } 1671 } 1672 } 1673 } 1674 } 1675 1676 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1677 func (d *Downloader) processFullSyncContent() error { 1678 for { 1679 results := d.queue.Results(true) 1680 if len(results) == 0 { 1681 return nil 1682 } 1683 if d.chainInsertHook != nil { 1684 d.chainInsertHook(results) 1685 } 1686 if err := d.importBlockResults(results); err != nil { 1687 return err 1688 } 1689 } 1690 } 1691 1692 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1693 // Check for any early termination requests 1694 if len(results) == 0 { 1695 return nil 1696 } 1697 select { 1698 case <-d.quitCh: 1699 return errCancelContentProcessing 1700 default: 1701 } 1702 // Retrieve the a batch of results to import 1703 first, last := results[0].Header, results[len(results)-1].Header 1704 log.Debug("Inserting downloaded chain", "items", len(results), 1705 "firstnum", first.Number, "firsthash", first.Hash(), 1706 "lastnum", last.Number, "lasthash", last.Hash(), 1707 ) 1708 blocks := make([]*types.Block, len(results)) 1709 for i, result := range results { 1710 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1711 } 1712 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1713 if index < len(results) { 1714 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1715 } else { 1716 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1717 // when it needs to preprocess blocks to import a sidechain. 1718 // The importer will put together a new list of blocks to import, which is a superset 1719 // of the blocks delivered from the downloader, and the indexing will be off. 1720 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1721 } 1722 return fmt.Errorf("%w: %v", errInvalidChain, err) 1723 } 1724 return nil 1725 } 1726 1727 // processFastSyncContent takes fetch results from the queue and writes them to the 1728 // database. It also controls the synchronisation of state nodes of the pivot block. 1729 func (d *Downloader) processFastSyncContent() error { 1730 // Start syncing state of the reported head block. This should get us most of 1731 // the state of the pivot block. 1732 d.pivotLock.RLock() 1733 sync := d.syncState(d.pivotHeader.Root) 1734 d.pivotLock.RUnlock() 1735 1736 defer func() { 1737 // The `sync` object is replaced every time the pivot moves. We need to 1738 // defer close the very last active one, hence the lazy evaluation vs. 1739 // calling defer sync.Cancel() !!! 1740 sync.Cancel() 1741 }() 1742 1743 closeOnErr := func(s *stateSync) { 1744 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled && err != snap.ErrCancelled { 1745 d.queue.Close() // wake up Results 1746 } 1747 } 1748 go closeOnErr(sync) 1749 1750 // To cater for moving pivot points, track the pivot block and subsequently 1751 // accumulated download results separately. 1752 var ( 1753 oldPivot *fetchResult // Locked in pivot block, might change eventually 1754 oldTail []*fetchResult // Downloaded content after the pivot 1755 ) 1756 for { 1757 // Wait for the next batch of downloaded data to be available, and if the pivot 1758 // block became stale, move the goalpost 1759 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1760 if len(results) == 0 { 1761 // If pivot sync is done, stop 1762 if oldPivot == nil { 1763 return sync.Cancel() 1764 } 1765 // If sync failed, stop 1766 select { 1767 case <-d.cancelCh: 1768 sync.Cancel() 1769 return errCanceled 1770 default: 1771 } 1772 } 1773 if d.chainInsertHook != nil { 1774 d.chainInsertHook(results) 1775 } 1776 // If we haven't downloaded the pivot block yet, check pivot staleness 1777 // notifications from the header downloader 1778 d.pivotLock.RLock() 1779 pivot := d.pivotHeader 1780 d.pivotLock.RUnlock() 1781 1782 if oldPivot == nil { 1783 if pivot.Root != sync.root { 1784 sync.Cancel() 1785 sync = d.syncState(pivot.Root) 1786 1787 go closeOnErr(sync) 1788 } 1789 } else { 1790 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1791 } 1792 // Split around the pivot block and process the two sides via fast/full sync 1793 if atomic.LoadInt32(&d.committed) == 0 { 1794 latest := results[len(results)-1].Header 1795 // If the height is above the pivot block by 2 sets, it means the pivot 1796 // become stale in the network and it was garbage collected, move to a 1797 // new pivot. 1798 // 1799 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 1800 // need to be taken into account, otherwise we're detecting the pivot move 1801 // late and will drop peers due to unavailable state!!! 1802 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 1803 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 1804 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 1805 1806 d.pivotLock.Lock() 1807 d.pivotHeader = pivot 1808 d.pivotLock.Unlock() 1809 1810 // Write out the pivot into the database so a rollback beyond it will 1811 // reenable fast sync 1812 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 1813 } 1814 } 1815 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 1816 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1817 return err 1818 } 1819 if P != nil { 1820 // If new pivot block found, cancel old state retrieval and restart 1821 if oldPivot != P { 1822 sync.Cancel() 1823 sync = d.syncState(P.Header.Root) 1824 1825 go closeOnErr(sync) 1826 oldPivot = P 1827 } 1828 // Wait for completion, occasionally checking for pivot staleness 1829 select { 1830 case <-sync.done: 1831 if sync.err != nil { 1832 return sync.err 1833 } 1834 if err := d.commitPivotBlock(P); err != nil { 1835 return err 1836 } 1837 oldPivot = nil 1838 1839 case <-time.After(time.Second): 1840 oldTail = afterP 1841 continue 1842 } 1843 } 1844 // Fast sync done, pivot commit done, full import 1845 if err := d.importBlockResults(afterP); err != nil { 1846 return err 1847 } 1848 } 1849 } 1850 1851 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1852 if len(results) == 0 { 1853 return nil, nil, nil 1854 } 1855 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 1856 // the pivot is somewhere in the future 1857 return nil, results, nil 1858 } 1859 // This can also be optimized, but only happens very seldom 1860 for _, result := range results { 1861 num := result.Header.Number.Uint64() 1862 switch { 1863 case num < pivot: 1864 before = append(before, result) 1865 case num == pivot: 1866 p = result 1867 default: 1868 after = append(after, result) 1869 } 1870 } 1871 return p, before, after 1872 } 1873 1874 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1875 // Check for any early termination requests 1876 if len(results) == 0 { 1877 return nil 1878 } 1879 select { 1880 case <-d.quitCh: 1881 return errCancelContentProcessing 1882 case <-stateSync.done: 1883 if err := stateSync.Wait(); err != nil { 1884 return err 1885 } 1886 default: 1887 } 1888 // Retrieve the a batch of results to import 1889 first, last := results[0].Header, results[len(results)-1].Header 1890 log.Debug("Inserting fast-sync blocks", "items", len(results), 1891 "firstnum", first.Number, "firsthash", first.Hash(), 1892 "lastnumn", last.Number, "lasthash", last.Hash(), 1893 ) 1894 blocks := make([]*types.Block, len(results)) 1895 receipts := make([]types.Receipts, len(results)) 1896 for i, result := range results { 1897 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1898 receipts[i] = result.Receipts 1899 } 1900 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1901 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1902 return fmt.Errorf("%w: %v", errInvalidChain, err) 1903 } 1904 return nil 1905 } 1906 1907 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1908 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1909 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1910 1911 // Commit the pivot block as the new head, will require full sync from here on 1912 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1913 return err 1914 } 1915 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1916 return err 1917 } 1918 atomic.StoreInt32(&d.committed, 1) 1919 return nil 1920 } 1921 1922 // DeliverHeaders injects a new batch of block headers received from a remote 1923 // node into the download schedule. 1924 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) error { 1925 return d.deliver(d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1926 } 1927 1928 // DeliverBodies injects a new batch of block bodies received from a remote node. 1929 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) error { 1930 return d.deliver(d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1931 } 1932 1933 // DeliverReceipts injects a new batch of receipts received from a remote node. 1934 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) error { 1935 return d.deliver(d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1936 } 1937 1938 // DeliverNodeData injects a new batch of node state data received from a remote node. 1939 func (d *Downloader) DeliverNodeData(id string, data [][]byte) error { 1940 return d.deliver(d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1941 } 1942 1943 // DeliverSnapPacket is invoked from a peer's message handler when it transmits a 1944 // data packet for the local node to consume. 1945 func (d *Downloader) DeliverSnapPacket(peer *snap.Peer, packet snap.Packet) error { 1946 switch packet := packet.(type) { 1947 case *snap.AccountRangePacket: 1948 hashes, accounts, err := packet.Unpack() 1949 if err != nil { 1950 return err 1951 } 1952 return d.SnapSyncer.OnAccounts(peer, packet.ID, hashes, accounts, packet.Proof) 1953 1954 case *snap.StorageRangesPacket: 1955 hashset, slotset := packet.Unpack() 1956 return d.SnapSyncer.OnStorage(peer, packet.ID, hashset, slotset, packet.Proof) 1957 1958 case *snap.ByteCodesPacket: 1959 return d.SnapSyncer.OnByteCodes(peer, packet.ID, packet.Codes) 1960 1961 case *snap.TrieNodesPacket: 1962 return d.SnapSyncer.OnTrieNodes(peer, packet.ID, packet.Nodes) 1963 1964 default: 1965 return fmt.Errorf("unexpected snap packet type: %T", packet) 1966 } 1967 } 1968 1969 // deliver injects a new batch of data received from a remote node. 1970 func (d *Downloader) deliver(destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1971 // Update the delivery metrics for both good and failed deliveries 1972 inMeter.Mark(int64(packet.Items())) 1973 defer func() { 1974 if err != nil { 1975 dropMeter.Mark(int64(packet.Items())) 1976 } 1977 }() 1978 // Deliver or abort if the sync is canceled while queuing 1979 d.cancelLock.RLock() 1980 cancel := d.cancelCh 1981 d.cancelLock.RUnlock() 1982 if cancel == nil { 1983 return errNoSyncActive 1984 } 1985 select { 1986 case destCh <- packet: 1987 return nil 1988 case <-cancel: 1989 return errNoSyncActive 1990 } 1991 }