github.com/shrimpyuk/bor@v0.2.15-0.20220224151350-fb4ec6020bae/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 "github.com/ethereum/go-ethereum/trie" 44 ) 45 46 var ( 47 MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request 48 MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request 49 MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly 50 MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request 51 MaxStateFetch = 384 // Amount of node state values to allow fetching per request 52 53 maxQueuedHeaders = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection) 54 maxHeadersProcess = 2048 // Number of header download results to import at once into the chain 55 maxResultsProcess = 2048 // Number of content download results to import at once into the chain 56 fullMaxForkAncestry uint64 = params.FullImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 57 lightMaxForkAncestry uint64 = params.LightImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 58 59 reorgProtThreshold = 48 // Threshold number of recent blocks to disable mini reorg protection 60 reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs 61 62 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 63 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 64 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 65 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 66 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 67 ) 68 69 var ( 70 errBusy = errors.New("busy") 71 errUnknownPeer = errors.New("peer is unknown or unhealthy") 72 errBadPeer = errors.New("action from bad peer ignored") 73 errStallingPeer = errors.New("peer is stalling") 74 errUnsyncedPeer = errors.New("unsynced peer") 75 errNoPeers = errors.New("no peers to keep download active") 76 errTimeout = errors.New("timeout") 77 errEmptyHeaderSet = errors.New("empty header set by peer") 78 errPeersUnavailable = errors.New("no peers available or all tried for download") 79 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 80 errInvalidChain = errors.New("retrieved hash chain is invalid") 81 errInvalidBody = errors.New("retrieved block body is invalid") 82 errInvalidReceipt = errors.New("retrieved receipt is invalid") 83 errCancelStateFetch = errors.New("state data download canceled (requested)") 84 errCancelContentProcessing = errors.New("content processing canceled (requested)") 85 errCanceled = errors.New("syncing canceled (requested)") 86 errNoSyncActive = errors.New("no sync active") 87 errTooOld = errors.New("peer's protocol version too old") 88 errNoAncestorFound = errors.New("no common ancestor found") 89 ) 90 91 type Downloader struct { 92 mode uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode 93 mux *event.TypeMux // Event multiplexer to announce sync operation events 94 95 checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync) 96 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 97 queue *queue // Scheduler for selecting the hashes to download 98 peers *peerSet // Set of active peers from which download can proceed 99 100 stateDB ethdb.Database // Database to state sync into (and deduplicate via) 101 stateBloom *trie.SyncBloom // Bloom filter for fast trie node and contract code existence checks 102 103 // Statistics 104 syncStatsChainOrigin uint64 // Origin block number where syncing started at 105 syncStatsChainHeight uint64 // Highest block number known when syncing started 106 syncStatsState stateSyncStats 107 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 108 109 lightchain LightChain 110 blockchain BlockChain 111 112 // Callbacks 113 dropPeer peerDropFn // Drops a peer for misbehaving 114 115 // Status 116 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 117 synchronising int32 118 notified int32 119 committed int32 120 ancientLimit uint64 // The maximum block number which can be regarded as ancient data. 121 122 // Channels 123 headerCh chan dataPack // Channel receiving inbound block headers 124 bodyCh chan dataPack // Channel receiving inbound block bodies 125 receiptCh chan dataPack // Channel receiving inbound receipts 126 bodyWakeCh chan bool // Channel to signal the block body fetcher of new tasks 127 receiptWakeCh chan bool // Channel to signal the receipt fetcher of new tasks 128 headerProcCh chan []*types.Header // Channel to feed the header processor new tasks 129 130 // State sync 131 pivotHeader *types.Header // Pivot block header to dynamically push the syncing state root 132 pivotLock sync.RWMutex // Lock protecting pivot header reads from updates 133 134 snapSync bool // Whether to run state sync over the snap protocol 135 SnapSyncer *snap.Syncer // TODO(karalabe): make private! hack for now 136 stateSyncStart chan *stateSync 137 trackStateReq chan *stateReq 138 stateCh chan dataPack // Channel receiving inbound node state data 139 140 // Cancellation and termination 141 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 142 cancelCh chan struct{} // Channel to cancel mid-flight syncs 143 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 144 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 145 146 quitCh chan struct{} // Quit channel to signal termination 147 quitLock sync.Mutex // Lock to prevent double closes 148 149 // Testing hooks 150 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 151 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 152 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 153 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 154 } 155 156 // LightChain encapsulates functions required to synchronise a light chain. 157 type LightChain interface { 158 // HasHeader verifies a header's presence in the local chain. 159 HasHeader(common.Hash, uint64) bool 160 161 // GetHeaderByHash retrieves a header from the local chain. 162 GetHeaderByHash(common.Hash) *types.Header 163 164 // CurrentHeader retrieves the head header from the local chain. 165 CurrentHeader() *types.Header 166 167 // GetTd returns the total difficulty of a local block. 168 GetTd(common.Hash, uint64) *big.Int 169 170 // InsertHeaderChain inserts a batch of headers into the local chain. 171 InsertHeaderChain([]*types.Header, int) (int, error) 172 173 // SetHead rewinds the local chain to a new head. 174 SetHead(uint64) error 175 } 176 177 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 178 type BlockChain interface { 179 LightChain 180 181 // HasBlock verifies a block's presence in the local chain. 182 HasBlock(common.Hash, uint64) bool 183 184 // HasFastBlock verifies a fast block's presence in the local chain. 185 HasFastBlock(common.Hash, uint64) bool 186 187 // GetBlockByHash retrieves a block from the local chain. 188 GetBlockByHash(common.Hash) *types.Block 189 190 // CurrentBlock retrieves the head block from the local chain. 191 CurrentBlock() *types.Block 192 193 // CurrentFastBlock retrieves the head fast block from the local chain. 194 CurrentFastBlock() *types.Block 195 196 // FastSyncCommitHead directly commits the head block to a certain entity. 197 FastSyncCommitHead(common.Hash) error 198 199 // InsertChain inserts a batch of blocks into the local chain. 200 InsertChain(types.Blocks) (int, error) 201 202 // InsertReceiptChain inserts a batch of receipts into the local chain. 203 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 204 205 // Snapshots returns the blockchain snapshot tree to paused it during sync. 206 Snapshots() *snapshot.Tree 207 } 208 209 // New creates a new downloader to fetch hashes and blocks from remote peers. 210 func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 211 if lightchain == nil { 212 lightchain = chain 213 } 214 dl := &Downloader{ 215 stateDB: stateDb, 216 stateBloom: stateBloom, 217 mux: mux, 218 checkpoint: checkpoint, 219 queue: newQueue(blockCacheMaxItems, blockCacheInitialItems), 220 peers: newPeerSet(), 221 blockchain: chain, 222 lightchain: lightchain, 223 dropPeer: dropPeer, 224 headerCh: make(chan dataPack, 1), 225 bodyCh: make(chan dataPack, 1), 226 receiptCh: make(chan dataPack, 1), 227 bodyWakeCh: make(chan bool, 1), 228 receiptWakeCh: make(chan bool, 1), 229 headerProcCh: make(chan []*types.Header, 1), 230 quitCh: make(chan struct{}), 231 stateCh: make(chan dataPack), 232 SnapSyncer: snap.NewSyncer(stateDb), 233 stateSyncStart: make(chan *stateSync), 234 syncStatsState: stateSyncStats{ 235 processed: rawdb.ReadFastTrieProgress(stateDb), 236 }, 237 trackStateReq: make(chan *stateReq), 238 } 239 go dl.stateFetcher() 240 return dl 241 } 242 243 // Progress retrieves the synchronisation boundaries, specifically the origin 244 // block where synchronisation started at (may have failed/suspended); the block 245 // or header sync is currently at; and the latest known block which the sync targets. 246 // 247 // In addition, during the state download phase of fast synchronisation the number 248 // of processed and the total number of known states are also returned. Otherwise 249 // these are zero. 250 func (d *Downloader) Progress() ethereum.SyncProgress { 251 // Lock the current stats and return the progress 252 d.syncStatsLock.RLock() 253 defer d.syncStatsLock.RUnlock() 254 255 current := uint64(0) 256 mode := d.getMode() 257 switch { 258 case d.blockchain != nil && mode == FullSync: 259 current = d.blockchain.CurrentBlock().NumberU64() 260 case d.blockchain != nil && mode == FastSync: 261 current = d.blockchain.CurrentFastBlock().NumberU64() 262 case d.lightchain != nil: 263 current = d.lightchain.CurrentHeader().Number.Uint64() 264 default: 265 log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", mode) 266 } 267 return ethereum.SyncProgress{ 268 StartingBlock: d.syncStatsChainOrigin, 269 CurrentBlock: current, 270 HighestBlock: d.syncStatsChainHeight, 271 PulledStates: d.syncStatsState.processed, 272 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 273 } 274 } 275 276 // Synchronising returns whether the downloader is currently retrieving blocks. 277 func (d *Downloader) Synchronising() bool { 278 return atomic.LoadInt32(&d.synchronising) > 0 279 } 280 281 // RegisterPeer injects a new download peer into the set of block source to be 282 // used for fetching hashes and blocks from. 283 func (d *Downloader) RegisterPeer(id string, version uint, peer Peer) error { 284 var logger log.Logger 285 if len(id) < 16 { 286 // Tests use short IDs, don't choke on them 287 logger = log.New("peer", id) 288 } else { 289 logger = log.New("peer", id[:8]) 290 } 291 logger.Trace("Registering sync peer") 292 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 293 logger.Error("Failed to register sync peer", "err", err) 294 return err 295 } 296 return nil 297 } 298 299 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 300 func (d *Downloader) RegisterLightPeer(id string, version uint, peer LightPeer) error { 301 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 302 } 303 304 // UnregisterPeer remove a peer from the known list, preventing any action from 305 // the specified peer. An effort is also made to return any pending fetches into 306 // the queue. 307 func (d *Downloader) UnregisterPeer(id string) error { 308 // Unregister the peer from the active peer set and revoke any fetch tasks 309 var logger log.Logger 310 if len(id) < 16 { 311 // Tests use short IDs, don't choke on them 312 logger = log.New("peer", id) 313 } else { 314 logger = log.New("peer", id[:8]) 315 } 316 logger.Trace("Unregistering sync peer") 317 if err := d.peers.Unregister(id); err != nil { 318 logger.Error("Failed to unregister sync peer", "err", err) 319 return err 320 } 321 d.queue.Revoke(id) 322 323 return nil 324 } 325 326 // Synchronise tries to sync up our local block chain with a remote peer, both 327 // adding various sanity checks as well as wrapping it with various log entries. 328 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 329 err := d.synchronise(id, head, td, mode) 330 331 switch err { 332 case nil, errBusy, errCanceled: 333 return err 334 } 335 if errors.Is(err, errInvalidChain) || errors.Is(err, errBadPeer) || errors.Is(err, errTimeout) || 336 errors.Is(err, errStallingPeer) || errors.Is(err, errUnsyncedPeer) || errors.Is(err, errEmptyHeaderSet) || 337 errors.Is(err, errPeersUnavailable) || errors.Is(err, errTooOld) || errors.Is(err, errInvalidAncestor) { 338 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 339 if d.dropPeer == nil { 340 // The dropPeer method is nil when `--copydb` is used for a local copy. 341 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 342 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 343 } else { 344 d.dropPeer(id) 345 } 346 return err 347 } 348 log.Warn("Synchronisation failed, retrying", "err", err) 349 return err 350 } 351 352 // synchronise will select the peer and use it for synchronising. If an empty string is given 353 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 354 // checks fail an error will be returned. This method is synchronous 355 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 356 // Mock out the synchronisation if testing 357 if d.synchroniseMock != nil { 358 return d.synchroniseMock(id, hash) 359 } 360 // Make sure only one goroutine is ever allowed past this point at once 361 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 362 return errBusy 363 } 364 defer atomic.StoreInt32(&d.synchronising, 0) 365 366 // Post a user notification of the sync (only once per session) 367 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 368 log.Info("Block synchronisation started") 369 } 370 // If we are already full syncing, but have a fast-sync bloom filter laying 371 // around, make sure it doesn't use memory any more. This is a special case 372 // when the user attempts to fast sync a new empty network. 373 if mode == FullSync && d.stateBloom != nil { 374 d.stateBloom.Close() 375 } 376 // If snap sync was requested, create the snap scheduler and switch to fast 377 // sync mode. Long term we could drop fast sync or merge the two together, 378 // but until snap becomes prevalent, we should support both. TODO(karalabe). 379 if mode == SnapSync { 380 if !d.snapSync { 381 // Snap sync uses the snapshot namespace to store potentially flakey data until 382 // sync completely heals and finishes. Pause snapshot maintenance in the mean 383 // time to prevent access. 384 if snapshots := d.blockchain.Snapshots(); snapshots != nil { // Only nil in tests 385 snapshots.Disable() 386 } 387 log.Warn("Enabling snapshot sync prototype") 388 d.snapSync = true 389 } 390 mode = FastSync 391 } 392 // Reset the queue, peer set and wake channels to clean any internal leftover state 393 d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems) 394 d.peers.Reset() 395 396 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 397 select { 398 case <-ch: 399 default: 400 } 401 } 402 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 403 for empty := false; !empty; { 404 select { 405 case <-ch: 406 default: 407 empty = true 408 } 409 } 410 } 411 for empty := false; !empty; { 412 select { 413 case <-d.headerProcCh: 414 default: 415 empty = true 416 } 417 } 418 // Create cancel channel for aborting mid-flight and mark the master peer 419 d.cancelLock.Lock() 420 d.cancelCh = make(chan struct{}) 421 d.cancelPeer = id 422 d.cancelLock.Unlock() 423 424 defer d.Cancel() // No matter what, we can't leave the cancel channel open 425 426 // Atomically set the requested sync mode 427 atomic.StoreUint32(&d.mode, uint32(mode)) 428 429 // Retrieve the origin peer and initiate the downloading process 430 p := d.peers.Peer(id) 431 if p == nil { 432 return errUnknownPeer 433 } 434 return d.syncWithPeer(p, hash, td) 435 } 436 437 func (d *Downloader) getMode() SyncMode { 438 return SyncMode(atomic.LoadUint32(&d.mode)) 439 } 440 441 // syncWithPeer starts a block synchronization based on the hash chain from the 442 // specified peer and head hash. 443 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 444 d.mux.Post(StartEvent{}) 445 defer func() { 446 // reset on error 447 if err != nil { 448 d.mux.Post(FailedEvent{err}) 449 } else { 450 latest := d.lightchain.CurrentHeader() 451 d.mux.Post(DoneEvent{latest}) 452 } 453 }() 454 if p.version < eth.ETH66 { 455 return fmt.Errorf("%w: advertized %d < required %d", errTooOld, p.version, eth.ETH66) 456 } 457 mode := d.getMode() 458 459 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", mode) 460 defer func(start time.Time) { 461 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 462 }(time.Now()) 463 464 // Look up the sync boundaries: the common ancestor and the target block 465 latest, pivot, err := d.fetchHead(p) 466 if err != nil { 467 return err 468 } 469 if mode == FastSync && pivot == nil { 470 // If no pivot block was returned, the head is below the min full block 471 // threshold (i.e. new chain). In that case we won't really fast sync 472 // anyway, but still need a valid pivot block to avoid some code hitting 473 // nil panics on an access. 474 pivot = d.blockchain.CurrentBlock().Header() 475 } 476 height := latest.Number.Uint64() 477 478 origin, err := d.findAncestor(p, latest) 479 if err != nil { 480 return err 481 } 482 d.syncStatsLock.Lock() 483 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 484 d.syncStatsChainOrigin = origin 485 } 486 d.syncStatsChainHeight = height 487 d.syncStatsLock.Unlock() 488 489 // Ensure our origin point is below any fast sync pivot point 490 if mode == FastSync { 491 if height <= uint64(fsMinFullBlocks) { 492 origin = 0 493 } else { 494 pivotNumber := pivot.Number.Uint64() 495 if pivotNumber <= origin { 496 origin = pivotNumber - 1 497 } 498 // Write out the pivot into the database so a rollback beyond it will 499 // reenable fast sync 500 rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber) 501 } 502 } 503 d.committed = 1 504 if mode == FastSync && pivot.Number.Uint64() != 0 { 505 d.committed = 0 506 } 507 if mode == FastSync { 508 // Set the ancient data limitation. 509 // If we are running fast sync, all block data older than ancientLimit will be 510 // written to the ancient store. More recent data will be written to the active 511 // database and will wait for the freezer to migrate. 512 // 513 // If there is a checkpoint available, then calculate the ancientLimit through 514 // that. Otherwise calculate the ancient limit through the advertised height 515 // of the remote peer. 516 // 517 // The reason for picking checkpoint first is that a malicious peer can give us 518 // a fake (very high) height, forcing the ancient limit to also be very high. 519 // The peer would start to feed us valid blocks until head, resulting in all of 520 // the blocks might be written into the ancient store. A following mini-reorg 521 // could cause issues. 522 if d.checkpoint != 0 && d.checkpoint > fullMaxForkAncestry+1 { 523 d.ancientLimit = d.checkpoint 524 } else if height > fullMaxForkAncestry+1 { 525 d.ancientLimit = height - fullMaxForkAncestry - 1 526 } else { 527 d.ancientLimit = 0 528 } 529 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 530 531 // If a part of blockchain data has already been written into active store, 532 // disable the ancient style insertion explicitly. 533 if origin >= frozen && frozen != 0 { 534 d.ancientLimit = 0 535 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 536 } else if d.ancientLimit > 0 { 537 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 538 } 539 // Rewind the ancient store and blockchain if reorg happens. 540 if origin+1 < frozen { 541 if err := d.lightchain.SetHead(origin + 1); err != nil { 542 return err 543 } 544 } 545 } 546 // Initiate the sync using a concurrent header and content retrieval algorithm 547 d.queue.Prepare(origin+1, mode) 548 if d.syncInitHook != nil { 549 d.syncInitHook(origin, height) 550 } 551 fetchers := []func() error{ 552 func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved 553 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 554 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 555 func() error { return d.processHeaders(origin+1, td) }, 556 } 557 if mode == FastSync { 558 d.pivotLock.Lock() 559 d.pivotHeader = pivot 560 d.pivotLock.Unlock() 561 562 fetchers = append(fetchers, func() error { return d.processFastSyncContent() }) 563 } else if mode == FullSync { 564 fetchers = append(fetchers, d.processFullSyncContent) 565 } 566 return d.spawnSync(fetchers) 567 } 568 569 // spawnSync runs d.process and all given fetcher functions to completion in 570 // separate goroutines, returning the first error that appears. 571 func (d *Downloader) spawnSync(fetchers []func() error) error { 572 errc := make(chan error, len(fetchers)) 573 d.cancelWg.Add(len(fetchers)) 574 for _, fn := range fetchers { 575 fn := fn 576 go func() { defer d.cancelWg.Done(); errc <- fn() }() 577 } 578 // Wait for the first error, then terminate the others. 579 var err error 580 for i := 0; i < len(fetchers); i++ { 581 if i == len(fetchers)-1 { 582 // Close the queue when all fetchers have exited. 583 // This will cause the block processor to end when 584 // it has processed the queue. 585 d.queue.Close() 586 } 587 if err = <-errc; err != nil && err != errCanceled { 588 break 589 } 590 } 591 d.queue.Close() 592 d.Cancel() 593 return err 594 } 595 596 // cancel aborts all of the operations and resets the queue. However, cancel does 597 // not wait for the running download goroutines to finish. This method should be 598 // used when cancelling the downloads from inside the downloader. 599 func (d *Downloader) cancel() { 600 // Close the current cancel channel 601 d.cancelLock.Lock() 602 defer d.cancelLock.Unlock() 603 604 if d.cancelCh != nil { 605 select { 606 case <-d.cancelCh: 607 // Channel was already closed 608 default: 609 close(d.cancelCh) 610 } 611 } 612 } 613 614 // Cancel aborts all of the operations and waits for all download goroutines to 615 // finish before returning. 616 func (d *Downloader) Cancel() { 617 d.cancel() 618 d.cancelWg.Wait() 619 } 620 621 // Terminate interrupts the downloader, canceling all pending operations. 622 // The downloader cannot be reused after calling Terminate. 623 func (d *Downloader) Terminate() { 624 // Close the termination channel (make sure double close is allowed) 625 d.quitLock.Lock() 626 select { 627 case <-d.quitCh: 628 default: 629 close(d.quitCh) 630 } 631 if d.stateBloom != nil { 632 d.stateBloom.Close() 633 } 634 d.quitLock.Unlock() 635 636 // Cancel any pending download requests 637 d.Cancel() 638 } 639 640 // fetchHead retrieves the head header and prior pivot block (if available) from 641 // a remote peer. 642 func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *types.Header, err error) { 643 p.log.Debug("Retrieving remote chain head") 644 mode := d.getMode() 645 646 // Request the advertised remote head block and wait for the response 647 latest, _ := p.peer.Head() 648 fetch := 1 649 if mode == FastSync { 650 fetch = 2 // head + pivot headers 651 } 652 go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true) 653 654 ttl := d.peers.rates.TargetTimeout() 655 timeout := time.After(ttl) 656 for { 657 select { 658 case <-d.cancelCh: 659 return nil, nil, errCanceled 660 661 case packet := <-d.headerCh: 662 // Discard anything not from the origin peer 663 if packet.PeerId() != p.id { 664 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 665 break 666 } 667 // Make sure the peer gave us at least one and at most the requested headers 668 headers := packet.(*headerPack).headers 669 if len(headers) == 0 || len(headers) > fetch { 670 return nil, nil, fmt.Errorf("%w: returned headers %d != requested %d", errBadPeer, len(headers), fetch) 671 } 672 // The first header needs to be the head, validate against the checkpoint 673 // and request. If only 1 header was returned, make sure there's no pivot 674 // or there was not one requested. 675 head := headers[0] 676 if (mode == FastSync || mode == LightSync) && head.Number.Uint64() < d.checkpoint { 677 return nil, nil, fmt.Errorf("%w: remote head %d below checkpoint %d", errUnsyncedPeer, head.Number, d.checkpoint) 678 } 679 if len(headers) == 1 { 680 if mode == FastSync && head.Number.Uint64() > uint64(fsMinFullBlocks) { 681 return nil, nil, fmt.Errorf("%w: no pivot included along head header", errBadPeer) 682 } 683 p.log.Debug("Remote head identified, no pivot", "number", head.Number, "hash", head.Hash()) 684 return head, nil, nil 685 } 686 // At this point we have 2 headers in total and the first is the 687 // validated head of the chain. Check the pivot number and return, 688 pivot := headers[1] 689 if pivot.Number.Uint64() != head.Number.Uint64()-uint64(fsMinFullBlocks) { 690 return nil, nil, fmt.Errorf("%w: remote pivot %d != requested %d", errInvalidChain, pivot.Number, head.Number.Uint64()-uint64(fsMinFullBlocks)) 691 } 692 return head, pivot, nil 693 694 case <-timeout: 695 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 696 return nil, nil, errTimeout 697 698 case <-d.bodyCh: 699 case <-d.receiptCh: 700 // Out of bounds delivery, ignore 701 } 702 } 703 } 704 705 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 706 // common ancestor. 707 // It returns parameters to be used for peer.RequestHeadersByNumber: 708 // from - starting block number 709 // count - number of headers to request 710 // skip - number of headers to skip 711 // and also returns 'max', the last block which is expected to be returned by the remote peers, 712 // given the (from,count,skip) 713 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 714 var ( 715 from int 716 count int 717 MaxCount = MaxHeaderFetch / 16 718 ) 719 // requestHead is the highest block that we will ask for. If requestHead is not offset, 720 // the highest block that we will get is 16 blocks back from head, which means we 721 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 722 // between us and the peer is 1-2 blocks, which is most common 723 requestHead := int(remoteHeight) - 1 724 if requestHead < 0 { 725 requestHead = 0 726 } 727 // requestBottom is the lowest block we want included in the query 728 // Ideally, we want to include the one just below our own head 729 requestBottom := int(localHeight - 1) 730 if requestBottom < 0 { 731 requestBottom = 0 732 } 733 totalSpan := requestHead - requestBottom 734 span := 1 + totalSpan/MaxCount 735 if span < 2 { 736 span = 2 737 } 738 if span > 16 { 739 span = 16 740 } 741 742 count = 1 + totalSpan/span 743 if count > MaxCount { 744 count = MaxCount 745 } 746 if count < 2 { 747 count = 2 748 } 749 from = requestHead - (count-1)*span 750 if from < 0 { 751 from = 0 752 } 753 max := from + (count-1)*span 754 return int64(from), count, span - 1, uint64(max) 755 } 756 757 // findAncestor tries to locate the common ancestor link of the local chain and 758 // a remote peers blockchain. In the general case when our node was in sync and 759 // on the correct chain, checking the top N links should already get us a match. 760 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 761 // the head links match), we do a binary search to find the common ancestor. 762 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 763 // Figure out the valid ancestor range to prevent rewrite attacks 764 var ( 765 floor = int64(-1) 766 localHeight uint64 767 remoteHeight = remoteHeader.Number.Uint64() 768 ) 769 mode := d.getMode() 770 switch mode { 771 case FullSync: 772 localHeight = d.blockchain.CurrentBlock().NumberU64() 773 case FastSync: 774 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 775 default: 776 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 777 } 778 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 779 780 // Recap floor value for binary search 781 maxForkAncestry := fullMaxForkAncestry 782 if d.getMode() == LightSync { 783 maxForkAncestry = lightMaxForkAncestry 784 } 785 if localHeight >= maxForkAncestry { 786 // We're above the max reorg threshold, find the earliest fork point 787 floor = int64(localHeight - maxForkAncestry) 788 } 789 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 790 // all headers before that point will be missing. 791 if mode == LightSync { 792 // If we don't know the current CHT position, find it 793 if d.genesis == 0 { 794 header := d.lightchain.CurrentHeader() 795 for header != nil { 796 d.genesis = header.Number.Uint64() 797 if floor >= int64(d.genesis)-1 { 798 break 799 } 800 header = d.lightchain.GetHeaderByHash(header.ParentHash) 801 } 802 } 803 // We already know the "genesis" block number, cap floor to that 804 if floor < int64(d.genesis)-1 { 805 floor = int64(d.genesis) - 1 806 } 807 } 808 809 ancestor, err := d.findAncestorSpanSearch(p, mode, remoteHeight, localHeight, floor) 810 if err == nil { 811 return ancestor, nil 812 } 813 // The returned error was not nil. 814 // If the error returned does not reflect that a common ancestor was not found, return it. 815 // If the error reflects that a common ancestor was not found, continue to binary search, 816 // where the error value will be reassigned. 817 if !errors.Is(err, errNoAncestorFound) { 818 return 0, err 819 } 820 821 ancestor, err = d.findAncestorBinarySearch(p, mode, remoteHeight, floor) 822 if err != nil { 823 return 0, err 824 } 825 return ancestor, nil 826 } 827 828 func (d *Downloader) findAncestorSpanSearch(p *peerConnection, mode SyncMode, remoteHeight, localHeight uint64, floor int64) (commonAncestor uint64, err error) { 829 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 830 831 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 832 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 833 834 // Wait for the remote response to the head fetch 835 number, hash := uint64(0), common.Hash{} 836 837 ttl := d.peers.rates.TargetTimeout() 838 timeout := time.After(ttl) 839 840 for finished := false; !finished; { 841 select { 842 case <-d.cancelCh: 843 return 0, errCanceled 844 845 case packet := <-d.headerCh: 846 // Discard anything not from the origin peer 847 if packet.PeerId() != p.id { 848 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 849 break 850 } 851 // Make sure the peer actually gave something valid 852 headers := packet.(*headerPack).headers 853 if len(headers) == 0 { 854 p.log.Warn("Empty head header set") 855 return 0, errEmptyHeaderSet 856 } 857 // Make sure the peer's reply conforms to the request 858 for i, header := range headers { 859 expectNumber := from + int64(i)*int64(skip+1) 860 if number := header.Number.Int64(); number != expectNumber { 861 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 862 return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering")) 863 } 864 } 865 // Check if a common ancestor was found 866 finished = true 867 for i := len(headers) - 1; i >= 0; i-- { 868 // Skip any headers that underflow/overflow our requested set 869 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 870 continue 871 } 872 // Otherwise check if we already know the header or not 873 h := headers[i].Hash() 874 n := headers[i].Number.Uint64() 875 876 var known bool 877 switch mode { 878 case FullSync: 879 known = d.blockchain.HasBlock(h, n) 880 case FastSync: 881 known = d.blockchain.HasFastBlock(h, n) 882 default: 883 known = d.lightchain.HasHeader(h, n) 884 } 885 if known { 886 number, hash = n, h 887 break 888 } 889 } 890 891 case <-timeout: 892 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 893 return 0, errTimeout 894 895 case <-d.bodyCh: 896 case <-d.receiptCh: 897 // Out of bounds delivery, ignore 898 } 899 } 900 // If the head fetch already found an ancestor, return 901 if hash != (common.Hash{}) { 902 if int64(number) <= floor { 903 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 904 return 0, errInvalidAncestor 905 } 906 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 907 return number, nil 908 } 909 return 0, errNoAncestorFound 910 } 911 912 func (d *Downloader) findAncestorBinarySearch(p *peerConnection, mode SyncMode, remoteHeight uint64, floor int64) (commonAncestor uint64, err error) { 913 hash := common.Hash{} 914 915 // Ancestor not found, we need to binary search over our chain 916 start, end := uint64(0), remoteHeight 917 if floor > 0 { 918 start = uint64(floor) 919 } 920 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 921 922 for start+1 < end { 923 // Split our chain interval in two, and request the hash to cross check 924 check := (start + end) / 2 925 926 ttl := d.peers.rates.TargetTimeout() 927 timeout := time.After(ttl) 928 929 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 930 931 // Wait until a reply arrives to this request 932 for arrived := false; !arrived; { 933 select { 934 case <-d.cancelCh: 935 return 0, errCanceled 936 937 case packet := <-d.headerCh: 938 // Discard anything not from the origin peer 939 if packet.PeerId() != p.id { 940 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 941 break 942 } 943 // Make sure the peer actually gave something valid 944 headers := packet.(*headerPack).headers 945 if len(headers) != 1 { 946 p.log.Warn("Multiple headers for single request", "headers", len(headers)) 947 return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers)) 948 } 949 arrived = true 950 951 // Modify the search interval based on the response 952 h := headers[0].Hash() 953 n := headers[0].Number.Uint64() 954 955 var known bool 956 switch mode { 957 case FullSync: 958 known = d.blockchain.HasBlock(h, n) 959 case FastSync: 960 known = d.blockchain.HasFastBlock(h, n) 961 default: 962 known = d.lightchain.HasHeader(h, n) 963 } 964 if !known { 965 end = check 966 break 967 } 968 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 969 if header.Number.Uint64() != check { 970 p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 971 return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number) 972 } 973 start = check 974 hash = h 975 976 case <-timeout: 977 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 978 return 0, errTimeout 979 980 case <-d.bodyCh: 981 case <-d.receiptCh: 982 // Out of bounds delivery, ignore 983 } 984 } 985 } 986 // Ensure valid ancestry and return 987 if int64(start) <= floor { 988 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 989 return 0, errInvalidAncestor 990 } 991 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 992 return start, nil 993 } 994 995 // fetchHeaders keeps retrieving headers concurrently from the number 996 // requested, until no more are returned, potentially throttling on the way. To 997 // facilitate concurrency but still protect against malicious nodes sending bad 998 // headers, we construct a header chain skeleton using the "origin" peer we are 999 // syncing with, and fill in the missing headers using anyone else. Headers from 1000 // other peers are only accepted if they map cleanly to the skeleton. If no one 1001 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 1002 // the origin is dropped. 1003 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error { 1004 p.log.Debug("Directing header downloads", "origin", from) 1005 defer p.log.Debug("Header download terminated") 1006 1007 // Create a timeout timer, and the associated header fetcher 1008 skeleton := true // Skeleton assembly phase or finishing up 1009 pivoting := false // Whether the next request is pivot verification 1010 request := time.Now() // time of the last skeleton fetch request 1011 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 1012 <-timeout.C // timeout channel should be initially empty 1013 defer timeout.Stop() 1014 1015 var ttl time.Duration 1016 getHeaders := func(from uint64) { 1017 request = time.Now() 1018 1019 ttl = d.peers.rates.TargetTimeout() 1020 timeout.Reset(ttl) 1021 1022 if skeleton { 1023 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 1024 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 1025 } else { 1026 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 1027 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 1028 } 1029 } 1030 getNextPivot := func() { 1031 pivoting = true 1032 request = time.Now() 1033 1034 ttl = d.peers.rates.TargetTimeout() 1035 timeout.Reset(ttl) 1036 1037 d.pivotLock.RLock() 1038 pivot := d.pivotHeader.Number.Uint64() 1039 d.pivotLock.RUnlock() 1040 1041 p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks)) 1042 go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep 1043 } 1044 // Start pulling the header chain skeleton until all is done 1045 ancestor := from 1046 getHeaders(from) 1047 1048 mode := d.getMode() 1049 for { 1050 select { 1051 case <-d.cancelCh: 1052 return errCanceled 1053 1054 case packet := <-d.headerCh: 1055 // Make sure the active peer is giving us the skeleton headers 1056 if packet.PeerId() != p.id { 1057 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 1058 break 1059 } 1060 headerReqTimer.UpdateSince(request) 1061 timeout.Stop() 1062 1063 // If the pivot is being checked, move if it became stale and run the real retrieval 1064 var pivot uint64 1065 1066 d.pivotLock.RLock() 1067 if d.pivotHeader != nil { 1068 pivot = d.pivotHeader.Number.Uint64() 1069 } 1070 d.pivotLock.RUnlock() 1071 1072 if pivoting { 1073 if packet.Items() == 2 { 1074 // Retrieve the headers and do some sanity checks, just in case 1075 headers := packet.(*headerPack).headers 1076 1077 if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want { 1078 log.Warn("Peer sent invalid next pivot", "have", have, "want", want) 1079 return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want) 1080 } 1081 if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want { 1082 log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want) 1083 return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want) 1084 } 1085 log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number) 1086 pivot = headers[0].Number.Uint64() 1087 1088 d.pivotLock.Lock() 1089 d.pivotHeader = headers[0] 1090 d.pivotLock.Unlock() 1091 1092 // Write out the pivot into the database so a rollback beyond 1093 // it will reenable fast sync and update the state root that 1094 // the state syncer will be downloading. 1095 rawdb.WriteLastPivotNumber(d.stateDB, pivot) 1096 } 1097 pivoting = false 1098 getHeaders(from) 1099 continue 1100 } 1101 // If the skeleton's finished, pull any remaining head headers directly from the origin 1102 if skeleton && packet.Items() == 0 { 1103 skeleton = false 1104 getHeaders(from) 1105 continue 1106 } 1107 // If no more headers are inbound, notify the content fetchers and return 1108 if packet.Items() == 0 { 1109 // Don't abort header fetches while the pivot is downloading 1110 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 1111 p.log.Debug("No headers, waiting for pivot commit") 1112 select { 1113 case <-time.After(fsHeaderContCheck): 1114 getHeaders(from) 1115 continue 1116 case <-d.cancelCh: 1117 return errCanceled 1118 } 1119 } 1120 // Pivot done (or not in fast sync) and no more headers, terminate the process 1121 p.log.Debug("No more headers available") 1122 select { 1123 case d.headerProcCh <- nil: 1124 return nil 1125 case <-d.cancelCh: 1126 return errCanceled 1127 } 1128 } 1129 headers := packet.(*headerPack).headers 1130 1131 // If we received a skeleton batch, resolve internals concurrently 1132 if skeleton { 1133 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1134 if err != nil { 1135 p.log.Debug("Skeleton chain invalid", "err", err) 1136 return fmt.Errorf("%w: %v", errInvalidChain, err) 1137 } 1138 headers = filled[proced:] 1139 from += uint64(proced) 1140 } else { 1141 // If we're closing in on the chain head, but haven't yet reached it, delay 1142 // the last few headers so mini reorgs on the head don't cause invalid hash 1143 // chain errors. 1144 if n := len(headers); n > 0 { 1145 // Retrieve the current head we're at 1146 var head uint64 1147 if mode == LightSync { 1148 head = d.lightchain.CurrentHeader().Number.Uint64() 1149 } else { 1150 head = d.blockchain.CurrentFastBlock().NumberU64() 1151 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1152 head = full 1153 } 1154 } 1155 // If the head is below the common ancestor, we're actually deduplicating 1156 // already existing chain segments, so use the ancestor as the fake head. 1157 // Otherwise we might end up delaying header deliveries pointlessly. 1158 if head < ancestor { 1159 head = ancestor 1160 } 1161 // If the head is way older than this batch, delay the last few headers 1162 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1163 delay := reorgProtHeaderDelay 1164 if delay > n { 1165 delay = n 1166 } 1167 headers = headers[:n-delay] 1168 } 1169 } 1170 } 1171 // Insert all the new headers and fetch the next batch 1172 if len(headers) > 0 { 1173 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1174 select { 1175 case d.headerProcCh <- headers: 1176 case <-d.cancelCh: 1177 return errCanceled 1178 } 1179 from += uint64(len(headers)) 1180 1181 // If we're still skeleton filling fast sync, check pivot staleness 1182 // before continuing to the next skeleton filling 1183 if skeleton && pivot > 0 { 1184 getNextPivot() 1185 } else { 1186 getHeaders(from) 1187 } 1188 } else { 1189 // No headers delivered, or all of them being delayed, sleep a bit and retry 1190 p.log.Trace("All headers delayed, waiting") 1191 select { 1192 case <-time.After(fsHeaderContCheck): 1193 getHeaders(from) 1194 continue 1195 case <-d.cancelCh: 1196 return errCanceled 1197 } 1198 } 1199 1200 case <-timeout.C: 1201 if d.dropPeer == nil { 1202 // The dropPeer method is nil when `--copydb` is used for a local copy. 1203 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1204 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1205 break 1206 } 1207 // Header retrieval timed out, consider the peer bad and drop 1208 p.log.Debug("Header request timed out", "elapsed", ttl) 1209 headerTimeoutMeter.Mark(1) 1210 d.dropPeer(p.id) 1211 1212 // Finish the sync gracefully instead of dumping the gathered data though 1213 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1214 select { 1215 case ch <- false: 1216 case <-d.cancelCh: 1217 } 1218 } 1219 select { 1220 case d.headerProcCh <- nil: 1221 case <-d.cancelCh: 1222 } 1223 return fmt.Errorf("%w: header request timed out", errBadPeer) 1224 } 1225 } 1226 } 1227 1228 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1229 // and maps them to the provided skeleton header chain. 1230 // 1231 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1232 // immediately to the header processor to keep the rest of the pipeline full even 1233 // in the case of header stalls. 1234 // 1235 // The method returns the entire filled skeleton and also the number of headers 1236 // already forwarded for processing. 1237 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1238 log.Debug("Filling up skeleton", "from", from) 1239 d.queue.ScheduleSkeleton(from, skeleton) 1240 1241 var ( 1242 deliver = func(packet dataPack) (int, error) { 1243 pack := packet.(*headerPack) 1244 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1245 } 1246 expire = func() map[string]int { return d.queue.ExpireHeaders(d.peers.rates.TargetTimeout()) } 1247 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) { 1248 return d.queue.ReserveHeaders(p, count), false, false 1249 } 1250 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1251 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.peers.rates.TargetRoundTrip()) } 1252 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1253 p.SetHeadersIdle(accepted, deliveryTime) 1254 } 1255 ) 1256 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1257 d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve, 1258 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1259 1260 log.Debug("Skeleton fill terminated", "err", err) 1261 1262 filled, proced := d.queue.RetrieveHeaders() 1263 return filled, proced, err 1264 } 1265 1266 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1267 // available peers, reserving a chunk of blocks for each, waiting for delivery 1268 // and also periodically checking for timeouts. 1269 func (d *Downloader) fetchBodies(from uint64) error { 1270 log.Debug("Downloading block bodies", "origin", from) 1271 1272 var ( 1273 deliver = func(packet dataPack) (int, error) { 1274 pack := packet.(*bodyPack) 1275 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1276 } 1277 expire = func() map[string]int { return d.queue.ExpireBodies(d.peers.rates.TargetTimeout()) } 1278 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1279 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.peers.rates.TargetRoundTrip()) } 1280 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) } 1281 ) 1282 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1283 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies, 1284 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1285 1286 log.Debug("Block body download terminated", "err", err) 1287 return err 1288 } 1289 1290 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1291 // available peers, reserving a chunk of receipts for each, waiting for delivery 1292 // and also periodically checking for timeouts. 1293 func (d *Downloader) fetchReceipts(from uint64) error { 1294 log.Debug("Downloading transaction receipts", "origin", from) 1295 1296 var ( 1297 deliver = func(packet dataPack) (int, error) { 1298 pack := packet.(*receiptPack) 1299 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1300 } 1301 expire = func() map[string]int { return d.queue.ExpireReceipts(d.peers.rates.TargetTimeout()) } 1302 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1303 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.peers.rates.TargetRoundTrip()) } 1304 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1305 p.SetReceiptsIdle(accepted, deliveryTime) 1306 } 1307 ) 1308 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1309 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts, 1310 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1311 1312 log.Debug("Transaction receipt download terminated", "err", err) 1313 return err 1314 } 1315 1316 // fetchParts iteratively downloads scheduled block parts, taking any available 1317 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1318 // also periodically checking for timeouts. 1319 // 1320 // As the scheduling/timeout logic mostly is the same for all downloaded data 1321 // types, this method is used by each for data gathering and is instrumented with 1322 // various callbacks to handle the slight differences between processing them. 1323 // 1324 // The instrumentation parameters: 1325 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1326 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1327 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1328 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1329 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1330 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1331 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1332 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1333 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1334 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1335 // - fetch: network callback to actually send a particular download request to a physical remote peer 1336 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1337 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1338 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1339 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1340 // - kind: textual label of the type being downloaded to display in log messages 1341 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1342 expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool), 1343 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1344 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error { 1345 1346 // Create a ticker to detect expired retrieval tasks 1347 ticker := time.NewTicker(100 * time.Millisecond) 1348 defer ticker.Stop() 1349 1350 update := make(chan struct{}, 1) 1351 1352 // Prepare the queue and fetch block parts until the block header fetcher's done 1353 finished := false 1354 for { 1355 select { 1356 case <-d.cancelCh: 1357 return errCanceled 1358 1359 case packet := <-deliveryCh: 1360 deliveryTime := time.Now() 1361 // If the peer was previously banned and failed to deliver its pack 1362 // in a reasonable time frame, ignore its message. 1363 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1364 // Deliver the received chunk of data and check chain validity 1365 accepted, err := deliver(packet) 1366 if errors.Is(err, errInvalidChain) { 1367 return err 1368 } 1369 // Unless a peer delivered something completely else than requested (usually 1370 // caused by a timed out request which came through in the end), set it to 1371 // idle. If the delivery's stale, the peer should have already been idled. 1372 if !errors.Is(err, errStaleDelivery) { 1373 setIdle(peer, accepted, deliveryTime) 1374 } 1375 // Issue a log to the user to see what's going on 1376 switch { 1377 case err == nil && packet.Items() == 0: 1378 peer.log.Trace("Requested data not delivered", "type", kind) 1379 case err == nil: 1380 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1381 default: 1382 peer.log.Debug("Failed to deliver retrieved data", "type", kind, "err", err) 1383 } 1384 } 1385 // Blocks assembled, try to update the progress 1386 select { 1387 case update <- struct{}{}: 1388 default: 1389 } 1390 1391 case cont := <-wakeCh: 1392 // The header fetcher sent a continuation flag, check if it's done 1393 if !cont { 1394 finished = true 1395 } 1396 // Headers arrive, try to update the progress 1397 select { 1398 case update <- struct{}{}: 1399 default: 1400 } 1401 1402 case <-ticker.C: 1403 // Sanity check update the progress 1404 select { 1405 case update <- struct{}{}: 1406 default: 1407 } 1408 1409 case <-update: 1410 // Short circuit if we lost all our peers 1411 if d.peers.Len() == 0 { 1412 return errNoPeers 1413 } 1414 // Check for fetch request timeouts and demote the responsible peers 1415 for pid, fails := range expire() { 1416 if peer := d.peers.Peer(pid); peer != nil { 1417 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1418 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1419 // out that sync wise we need to get rid of the peer. 1420 // 1421 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1422 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1423 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1424 if fails > 2 { 1425 peer.log.Trace("Data delivery timed out", "type", kind) 1426 setIdle(peer, 0, time.Now()) 1427 } else { 1428 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1429 1430 if d.dropPeer == nil { 1431 // The dropPeer method is nil when `--copydb` is used for a local copy. 1432 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1433 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1434 } else { 1435 d.dropPeer(pid) 1436 1437 // If this peer was the master peer, abort sync immediately 1438 d.cancelLock.RLock() 1439 master := pid == d.cancelPeer 1440 d.cancelLock.RUnlock() 1441 1442 if master { 1443 d.cancel() 1444 return errTimeout 1445 } 1446 } 1447 } 1448 } 1449 } 1450 // If there's nothing more to fetch, wait or terminate 1451 if pending() == 0 { 1452 if !inFlight() && finished { 1453 log.Debug("Data fetching completed", "type", kind) 1454 return nil 1455 } 1456 break 1457 } 1458 // Send a download request to all idle peers, until throttled 1459 progressed, throttled, running := false, false, inFlight() 1460 idles, total := idle() 1461 pendCount := pending() 1462 for _, peer := range idles { 1463 // Short circuit if throttling activated 1464 if throttled { 1465 break 1466 } 1467 // Short circuit if there is no more available task. 1468 if pendCount = pending(); pendCount == 0 { 1469 break 1470 } 1471 // Reserve a chunk of fetches for a peer. A nil can mean either that 1472 // no more headers are available, or that the peer is known not to 1473 // have them. 1474 request, progress, throttle := reserve(peer, capacity(peer)) 1475 if progress { 1476 progressed = true 1477 } 1478 if throttle { 1479 throttled = true 1480 throttleCounter.Inc(1) 1481 } 1482 if request == nil { 1483 continue 1484 } 1485 if request.From > 0 { 1486 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1487 } else { 1488 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1489 } 1490 // Fetch the chunk and make sure any errors return the hashes to the queue 1491 if fetchHook != nil { 1492 fetchHook(request.Headers) 1493 } 1494 if err := fetch(peer, request); err != nil { 1495 // Although we could try and make an attempt to fix this, this error really 1496 // means that we've double allocated a fetch task to a peer. If that is the 1497 // case, the internal state of the downloader and the queue is very wrong so 1498 // better hard crash and note the error instead of silently accumulating into 1499 // a much bigger issue. 1500 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1501 } 1502 running = true 1503 } 1504 // Make sure that we have peers available for fetching. If all peers have been tried 1505 // and all failed throw an error 1506 if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 { 1507 return errPeersUnavailable 1508 } 1509 } 1510 } 1511 } 1512 1513 // processHeaders takes batches of retrieved headers from an input channel and 1514 // keeps processing and scheduling them into the header chain and downloader's 1515 // queue until the stream ends or a failure occurs. 1516 func (d *Downloader) processHeaders(origin uint64, td *big.Int) error { 1517 // Keep a count of uncertain headers to roll back 1518 var ( 1519 rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis) 1520 rollbackErr error 1521 mode = d.getMode() 1522 ) 1523 defer func() { 1524 if rollback > 0 { 1525 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1526 if mode != LightSync { 1527 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1528 lastBlock = d.blockchain.CurrentBlock().Number() 1529 } 1530 if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block 1531 // We're already unwinding the stack, only print the error to make it more visible 1532 log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err) 1533 } 1534 curFastBlock, curBlock := common.Big0, common.Big0 1535 if mode != LightSync { 1536 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1537 curBlock = d.blockchain.CurrentBlock().Number() 1538 } 1539 log.Warn("Rolled back chain segment", 1540 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1541 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1542 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr) 1543 } 1544 }() 1545 // Wait for batches of headers to process 1546 gotHeaders := false 1547 1548 for { 1549 select { 1550 case <-d.cancelCh: 1551 rollbackErr = errCanceled 1552 return errCanceled 1553 1554 case headers := <-d.headerProcCh: 1555 // Terminate header processing if we synced up 1556 if len(headers) == 0 { 1557 // Notify everyone that headers are fully processed 1558 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1559 select { 1560 case ch <- false: 1561 case <-d.cancelCh: 1562 } 1563 } 1564 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1565 // better chain compared to ours. The only exception is if its promised blocks were 1566 // already imported by other means (e.g. fetcher): 1567 // 1568 // R <remote peer>, L <local node>: Both at block 10 1569 // R: Mine block 11, and propagate it to L 1570 // L: Queue block 11 for import 1571 // L: Notice that R's head and TD increased compared to ours, start sync 1572 // L: Import of block 11 finishes 1573 // L: Sync begins, and finds common ancestor at 11 1574 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1575 // R: Nothing to give 1576 if mode != LightSync { 1577 head := d.blockchain.CurrentBlock() 1578 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1579 return errStallingPeer 1580 } 1581 } 1582 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1583 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1584 // of delivering the post-pivot blocks that would flag the invalid content. 1585 // 1586 // This check cannot be executed "as is" for full imports, since blocks may still be 1587 // queued for processing when the header download completes. However, as long as the 1588 // peer gave us something useful, we're already happy/progressed (above check). 1589 if mode == FastSync || mode == LightSync { 1590 head := d.lightchain.CurrentHeader() 1591 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1592 return errStallingPeer 1593 } 1594 } 1595 // Disable any rollback and return 1596 rollback = 0 1597 return nil 1598 } 1599 // Otherwise split the chunk of headers into batches and process them 1600 gotHeaders = true 1601 for len(headers) > 0 { 1602 // Terminate if something failed in between processing chunks 1603 select { 1604 case <-d.cancelCh: 1605 rollbackErr = errCanceled 1606 return errCanceled 1607 default: 1608 } 1609 // Select the next chunk of headers to import 1610 limit := maxHeadersProcess 1611 if limit > len(headers) { 1612 limit = len(headers) 1613 } 1614 chunk := headers[:limit] 1615 1616 // In case of header only syncing, validate the chunk immediately 1617 if mode == FastSync || mode == LightSync { 1618 // If we're importing pure headers, verify based on their recentness 1619 var pivot uint64 1620 1621 d.pivotLock.RLock() 1622 if d.pivotHeader != nil { 1623 pivot = d.pivotHeader.Number.Uint64() 1624 } 1625 d.pivotLock.RUnlock() 1626 1627 frequency := fsHeaderCheckFrequency 1628 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1629 frequency = 1 1630 } 1631 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1632 rollbackErr = err 1633 1634 // If some headers were inserted, track them as uncertain 1635 if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 { 1636 rollback = chunk[0].Number.Uint64() 1637 } 1638 log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err) 1639 return fmt.Errorf("%w: %v", errInvalidChain, err) 1640 } 1641 // All verifications passed, track all headers within the alloted limits 1642 if mode == FastSync { 1643 head := chunk[len(chunk)-1].Number.Uint64() 1644 if head-rollback > uint64(fsHeaderSafetyNet) { 1645 rollback = head - uint64(fsHeaderSafetyNet) 1646 } else { 1647 rollback = 1 1648 } 1649 } 1650 } 1651 // Unless we're doing light chains, schedule the headers for associated content retrieval 1652 if mode == FullSync || mode == FastSync { 1653 // If we've reached the allowed number of pending headers, stall a bit 1654 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1655 select { 1656 case <-d.cancelCh: 1657 rollbackErr = errCanceled 1658 return errCanceled 1659 case <-time.After(time.Second): 1660 } 1661 } 1662 // Otherwise insert the headers for content retrieval 1663 inserts := d.queue.Schedule(chunk, origin) 1664 if len(inserts) != len(chunk) { 1665 rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk)) 1666 return fmt.Errorf("%w: stale headers", errBadPeer) 1667 } 1668 } 1669 headers = headers[limit:] 1670 origin += uint64(limit) 1671 } 1672 // Update the highest block number we know if a higher one is found. 1673 d.syncStatsLock.Lock() 1674 if d.syncStatsChainHeight < origin { 1675 d.syncStatsChainHeight = origin - 1 1676 } 1677 d.syncStatsLock.Unlock() 1678 1679 // Signal the content downloaders of the availablility of new tasks 1680 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1681 select { 1682 case ch <- true: 1683 default: 1684 } 1685 } 1686 } 1687 } 1688 } 1689 1690 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1691 func (d *Downloader) processFullSyncContent() error { 1692 for { 1693 results := d.queue.Results(true) 1694 if len(results) == 0 { 1695 return nil 1696 } 1697 if d.chainInsertHook != nil { 1698 d.chainInsertHook(results) 1699 } 1700 if err := d.importBlockResults(results); err != nil { 1701 return err 1702 } 1703 } 1704 } 1705 1706 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1707 // Check for any early termination requests 1708 if len(results) == 0 { 1709 return nil 1710 } 1711 select { 1712 case <-d.quitCh: 1713 return errCancelContentProcessing 1714 default: 1715 } 1716 // Retrieve the a batch of results to import 1717 first, last := results[0].Header, results[len(results)-1].Header 1718 log.Debug("Inserting downloaded chain", "items", len(results), 1719 "firstnum", first.Number, "firsthash", first.Hash(), 1720 "lastnum", last.Number, "lasthash", last.Hash(), 1721 ) 1722 blocks := make([]*types.Block, len(results)) 1723 for i, result := range results { 1724 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1725 } 1726 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1727 if index < len(results) { 1728 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1729 } else { 1730 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1731 // when it needs to preprocess blocks to import a sidechain. 1732 // The importer will put together a new list of blocks to import, which is a superset 1733 // of the blocks delivered from the downloader, and the indexing will be off. 1734 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1735 } 1736 return fmt.Errorf("%w: %v", errInvalidChain, err) 1737 } 1738 return nil 1739 } 1740 1741 // processFastSyncContent takes fetch results from the queue and writes them to the 1742 // database. It also controls the synchronisation of state nodes of the pivot block. 1743 func (d *Downloader) processFastSyncContent() error { 1744 // Start syncing state of the reported head block. This should get us most of 1745 // the state of the pivot block. 1746 d.pivotLock.RLock() 1747 sync := d.syncState(d.pivotHeader.Root) 1748 d.pivotLock.RUnlock() 1749 1750 defer func() { 1751 // The `sync` object is replaced every time the pivot moves. We need to 1752 // defer close the very last active one, hence the lazy evaluation vs. 1753 // calling defer sync.Cancel() !!! 1754 sync.Cancel() 1755 }() 1756 1757 closeOnErr := func(s *stateSync) { 1758 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled && err != snap.ErrCancelled { 1759 d.queue.Close() // wake up Results 1760 } 1761 } 1762 go closeOnErr(sync) 1763 1764 // To cater for moving pivot points, track the pivot block and subsequently 1765 // accumulated download results separately. 1766 var ( 1767 oldPivot *fetchResult // Locked in pivot block, might change eventually 1768 oldTail []*fetchResult // Downloaded content after the pivot 1769 ) 1770 for { 1771 // Wait for the next batch of downloaded data to be available, and if the pivot 1772 // block became stale, move the goalpost 1773 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1774 if len(results) == 0 { 1775 // If pivot sync is done, stop 1776 if oldPivot == nil { 1777 return sync.Cancel() 1778 } 1779 // If sync failed, stop 1780 select { 1781 case <-d.cancelCh: 1782 sync.Cancel() 1783 return errCanceled 1784 default: 1785 } 1786 } 1787 if d.chainInsertHook != nil { 1788 d.chainInsertHook(results) 1789 } 1790 // If we haven't downloaded the pivot block yet, check pivot staleness 1791 // notifications from the header downloader 1792 d.pivotLock.RLock() 1793 pivot := d.pivotHeader 1794 d.pivotLock.RUnlock() 1795 1796 if oldPivot == nil { 1797 if pivot.Root != sync.root { 1798 sync.Cancel() 1799 sync = d.syncState(pivot.Root) 1800 1801 go closeOnErr(sync) 1802 } 1803 } else { 1804 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1805 } 1806 // Split around the pivot block and process the two sides via fast/full sync 1807 if atomic.LoadInt32(&d.committed) == 0 { 1808 latest := results[len(results)-1].Header 1809 // If the height is above the pivot block by 2 sets, it means the pivot 1810 // become stale in the network and it was garbage collected, move to a 1811 // new pivot. 1812 // 1813 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 1814 // need to be taken into account, otherwise we're detecting the pivot move 1815 // late and will drop peers due to unavailable state!!! 1816 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 1817 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 1818 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 1819 1820 d.pivotLock.Lock() 1821 d.pivotHeader = pivot 1822 d.pivotLock.Unlock() 1823 1824 // Write out the pivot into the database so a rollback beyond it will 1825 // reenable fast sync 1826 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 1827 } 1828 } 1829 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 1830 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1831 return err 1832 } 1833 if P != nil { 1834 // If new pivot block found, cancel old state retrieval and restart 1835 if oldPivot != P { 1836 sync.Cancel() 1837 sync = d.syncState(P.Header.Root) 1838 1839 go closeOnErr(sync) 1840 oldPivot = P 1841 } 1842 // Wait for completion, occasionally checking for pivot staleness 1843 select { 1844 case <-sync.done: 1845 if sync.err != nil { 1846 return sync.err 1847 } 1848 if err := d.commitPivotBlock(P); err != nil { 1849 return err 1850 } 1851 oldPivot = nil 1852 1853 case <-time.After(time.Second): 1854 oldTail = afterP 1855 continue 1856 } 1857 } 1858 // Fast sync done, pivot commit done, full import 1859 if err := d.importBlockResults(afterP); err != nil { 1860 return err 1861 } 1862 } 1863 } 1864 1865 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1866 if len(results) == 0 { 1867 return nil, nil, nil 1868 } 1869 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 1870 // the pivot is somewhere in the future 1871 return nil, results, nil 1872 } 1873 // This can also be optimized, but only happens very seldom 1874 for _, result := range results { 1875 num := result.Header.Number.Uint64() 1876 switch { 1877 case num < pivot: 1878 before = append(before, result) 1879 case num == pivot: 1880 p = result 1881 default: 1882 after = append(after, result) 1883 } 1884 } 1885 return p, before, after 1886 } 1887 1888 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1889 // Check for any early termination requests 1890 if len(results) == 0 { 1891 return nil 1892 } 1893 select { 1894 case <-d.quitCh: 1895 return errCancelContentProcessing 1896 case <-stateSync.done: 1897 if err := stateSync.Wait(); err != nil { 1898 return err 1899 } 1900 default: 1901 } 1902 // Retrieve the a batch of results to import 1903 first, last := results[0].Header, results[len(results)-1].Header 1904 log.Debug("Inserting fast-sync blocks", "items", len(results), 1905 "firstnum", first.Number, "firsthash", first.Hash(), 1906 "lastnumn", last.Number, "lasthash", last.Hash(), 1907 ) 1908 blocks := make([]*types.Block, len(results)) 1909 receipts := make([]types.Receipts, len(results)) 1910 for i, result := range results { 1911 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1912 receipts[i] = result.Receipts 1913 } 1914 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1915 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1916 return fmt.Errorf("%w: %v", errInvalidChain, err) 1917 } 1918 return nil 1919 } 1920 1921 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1922 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1923 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1924 1925 // Commit the pivot block as the new head, will require full sync from here on 1926 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1927 return err 1928 } 1929 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1930 return err 1931 } 1932 atomic.StoreInt32(&d.committed, 1) 1933 1934 // If we had a bloom filter for the state sync, deallocate it now. Note, we only 1935 // deallocate internally, but keep the empty wrapper. This ensures that if we do 1936 // a rollback after committing the pivot and restarting fast sync, we don't end 1937 // up using a nil bloom. Empty bloom is fine, it just returns that it does not 1938 // have the info we need, so reach down to the database instead. 1939 if d.stateBloom != nil { 1940 d.stateBloom.Close() 1941 } 1942 return nil 1943 } 1944 1945 // DeliverHeaders injects a new batch of block headers received from a remote 1946 // node into the download schedule. 1947 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) error { 1948 return d.deliver(d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1949 } 1950 1951 // DeliverBodies injects a new batch of block bodies received from a remote node. 1952 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) error { 1953 return d.deliver(d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1954 } 1955 1956 // DeliverReceipts injects a new batch of receipts received from a remote node. 1957 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) error { 1958 return d.deliver(d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1959 } 1960 1961 // DeliverNodeData injects a new batch of node state data received from a remote node. 1962 func (d *Downloader) DeliverNodeData(id string, data [][]byte) error { 1963 return d.deliver(d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1964 } 1965 1966 // DeliverSnapPacket is invoked from a peer's message handler when it transmits a 1967 // data packet for the local node to consume. 1968 func (d *Downloader) DeliverSnapPacket(peer *snap.Peer, packet snap.Packet) error { 1969 switch packet := packet.(type) { 1970 case *snap.AccountRangePacket: 1971 hashes, accounts, err := packet.Unpack() 1972 if err != nil { 1973 return err 1974 } 1975 return d.SnapSyncer.OnAccounts(peer, packet.ID, hashes, accounts, packet.Proof) 1976 1977 case *snap.StorageRangesPacket: 1978 hashset, slotset := packet.Unpack() 1979 return d.SnapSyncer.OnStorage(peer, packet.ID, hashset, slotset, packet.Proof) 1980 1981 case *snap.ByteCodesPacket: 1982 return d.SnapSyncer.OnByteCodes(peer, packet.ID, packet.Codes) 1983 1984 case *snap.TrieNodesPacket: 1985 return d.SnapSyncer.OnTrieNodes(peer, packet.ID, packet.Nodes) 1986 1987 default: 1988 return fmt.Errorf("unexpected snap packet type: %T", packet) 1989 } 1990 } 1991 1992 // deliver injects a new batch of data received from a remote node. 1993 func (d *Downloader) deliver(destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1994 // Update the delivery metrics for both good and failed deliveries 1995 inMeter.Mark(int64(packet.Items())) 1996 defer func() { 1997 if err != nil { 1998 dropMeter.Mark(int64(packet.Items())) 1999 } 2000 }() 2001 // Deliver or abort if the sync is canceled while queuing 2002 d.cancelLock.RLock() 2003 cancel := d.cancelCh 2004 d.cancelLock.RUnlock() 2005 if cancel == nil { 2006 return errNoSyncActive 2007 } 2008 select { 2009 case destCh <- packet: 2010 return nil 2011 case <-cancel: 2012 return errNoSyncActive 2013 } 2014 }