github.com/juliankolbe/go-ethereum@v1.9.992/eth/downloader/downloader.go (about) 1 // Copyright 2015 The go-ethereum Authors 2 // This file is part of the go-ethereum library. 3 // 4 // The go-ethereum library is free software: you can redistribute it and/or modify 5 // it under the terms of the GNU Lesser General Public License as published by 6 // the Free Software Foundation, either version 3 of the License, or 7 // (at your option) any later version. 8 // 9 // The go-ethereum library is distributed in the hope that it will be useful, 10 // but WITHOUT ANY WARRANTY; without even the implied warranty of 11 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 // GNU Lesser General Public License for more details. 13 // 14 // You should have received a copy of the GNU Lesser General Public License 15 // along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>. 16 17 // Package downloader contains the manual full chain synchronisation. 18 package downloader 19 20 import ( 21 "errors" 22 "fmt" 23 "math/big" 24 "sync" 25 "sync/atomic" 26 "time" 27 28 "github.com/juliankolbe/go-ethereum" 29 "github.com/juliankolbe/go-ethereum/common" 30 "github.com/juliankolbe/go-ethereum/core/rawdb" 31 "github.com/juliankolbe/go-ethereum/core/types" 32 "github.com/juliankolbe/go-ethereum/eth/protocols/snap" 33 "github.com/juliankolbe/go-ethereum/ethdb" 34 "github.com/juliankolbe/go-ethereum/event" 35 "github.com/juliankolbe/go-ethereum/log" 36 "github.com/juliankolbe/go-ethereum/metrics" 37 "github.com/juliankolbe/go-ethereum/params" 38 "github.com/juliankolbe/go-ethereum/trie" 39 ) 40 41 var ( 42 MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request 43 MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request 44 MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly 45 MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request 46 MaxStateFetch = 384 // Amount of node state values to allow fetching per request 47 48 rttMinEstimate = 2 * time.Second // Minimum round-trip time to target for download requests 49 rttMaxEstimate = 20 * time.Second // Maximum round-trip time to target for download requests 50 rttMinConfidence = 0.1 // Worse confidence factor in our estimated RTT value 51 ttlScaling = 3 // Constant scaling factor for RTT -> TTL conversion 52 ttlLimit = time.Minute // Maximum TTL allowance to prevent reaching crazy timeouts 53 54 qosTuningPeers = 5 // Number of peers to tune based on (best peers) 55 qosConfidenceCap = 10 // Number of peers above which not to modify RTT confidence 56 qosTuningImpact = 0.25 // Impact that a new tuning target has on the previous value 57 58 maxQueuedHeaders = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection) 59 maxHeadersProcess = 2048 // Number of header download results to import at once into the chain 60 maxResultsProcess = 2048 // Number of content download results to import at once into the chain 61 fullMaxForkAncestry uint64 = params.FullImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 62 lightMaxForkAncestry uint64 = params.LightImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 63 64 reorgProtThreshold = 48 // Threshold number of recent blocks to disable mini reorg protection 65 reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs 66 67 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 68 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 69 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 70 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 71 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 72 ) 73 74 var ( 75 errBusy = errors.New("busy") 76 errUnknownPeer = errors.New("peer is unknown or unhealthy") 77 errBadPeer = errors.New("action from bad peer ignored") 78 errStallingPeer = errors.New("peer is stalling") 79 errUnsyncedPeer = errors.New("unsynced peer") 80 errNoPeers = errors.New("no peers to keep download active") 81 errTimeout = errors.New("timeout") 82 errEmptyHeaderSet = errors.New("empty header set by peer") 83 errPeersUnavailable = errors.New("no peers available or all tried for download") 84 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 85 errInvalidChain = errors.New("retrieved hash chain is invalid") 86 errInvalidBody = errors.New("retrieved block body is invalid") 87 errInvalidReceipt = errors.New("retrieved receipt is invalid") 88 errCancelStateFetch = errors.New("state data download canceled (requested)") 89 errCancelContentProcessing = errors.New("content processing canceled (requested)") 90 errCanceled = errors.New("syncing canceled (requested)") 91 errNoSyncActive = errors.New("no sync active") 92 errTooOld = errors.New("peer's protocol version too old") 93 errNoAncestorFound = errors.New("no common ancestor found") 94 ) 95 96 type Downloader struct { 97 // WARNING: The `rttEstimate` and `rttConfidence` fields are accessed atomically. 98 // On 32 bit platforms, only 64-bit aligned fields can be atomic. The struct is 99 // guaranteed to be so aligned, so take advantage of that. For more information, 100 // see https://golang.org/pkg/sync/atomic/#pkg-note-BUG. 101 rttEstimate uint64 // Round trip time to target for download requests 102 rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops) 103 104 mode uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode 105 mux *event.TypeMux // Event multiplexer to announce sync operation events 106 107 checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync) 108 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 109 queue *queue // Scheduler for selecting the hashes to download 110 peers *peerSet // Set of active peers from which download can proceed 111 112 stateDB ethdb.Database // Database to state sync into (and deduplicate via) 113 stateBloom *trie.SyncBloom // Bloom filter for fast trie node and contract code existence checks 114 115 // Statistics 116 syncStatsChainOrigin uint64 // Origin block number where syncing started at 117 syncStatsChainHeight uint64 // Highest block number known when syncing started 118 syncStatsState stateSyncStats 119 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 120 121 lightchain LightChain 122 blockchain BlockChain 123 124 // Callbacks 125 dropPeer peerDropFn // Drops a peer for misbehaving 126 127 // Status 128 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 129 synchronising int32 130 notified int32 131 committed int32 132 ancientLimit uint64 // The maximum block number which can be regarded as ancient data. 133 134 // Channels 135 headerCh chan dataPack // Channel receiving inbound block headers 136 bodyCh chan dataPack // Channel receiving inbound block bodies 137 receiptCh chan dataPack // Channel receiving inbound receipts 138 bodyWakeCh chan bool // Channel to signal the block body fetcher of new tasks 139 receiptWakeCh chan bool // Channel to signal the receipt fetcher of new tasks 140 headerProcCh chan []*types.Header // Channel to feed the header processor new tasks 141 142 // State sync 143 pivotHeader *types.Header // Pivot block header to dynamically push the syncing state root 144 pivotLock sync.RWMutex // Lock protecting pivot header reads from updates 145 146 snapSync bool // Whether to run state sync over the snap protocol 147 SnapSyncer *snap.Syncer // TODO(karalabe): make private! hack for now 148 stateSyncStart chan *stateSync 149 trackStateReq chan *stateReq 150 stateCh chan dataPack // Channel receiving inbound node state data 151 152 // Cancellation and termination 153 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 154 cancelCh chan struct{} // Channel to cancel mid-flight syncs 155 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 156 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 157 158 quitCh chan struct{} // Quit channel to signal termination 159 quitLock sync.Mutex // Lock to prevent double closes 160 161 // Testing hooks 162 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 163 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 164 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 165 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 166 } 167 168 // LightChain encapsulates functions required to synchronise a light chain. 169 type LightChain interface { 170 // HasHeader verifies a header's presence in the local chain. 171 HasHeader(common.Hash, uint64) bool 172 173 // GetHeaderByHash retrieves a header from the local chain. 174 GetHeaderByHash(common.Hash) *types.Header 175 176 // CurrentHeader retrieves the head header from the local chain. 177 CurrentHeader() *types.Header 178 179 // GetTd returns the total difficulty of a local block. 180 GetTd(common.Hash, uint64) *big.Int 181 182 // InsertHeaderChain inserts a batch of headers into the local chain. 183 InsertHeaderChain([]*types.Header, int) (int, error) 184 185 // SetHead rewinds the local chain to a new head. 186 SetHead(uint64) error 187 } 188 189 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 190 type BlockChain interface { 191 LightChain 192 193 // HasBlock verifies a block's presence in the local chain. 194 HasBlock(common.Hash, uint64) bool 195 196 // HasFastBlock verifies a fast block's presence in the local chain. 197 HasFastBlock(common.Hash, uint64) bool 198 199 // GetBlockByHash retrieves a block from the local chain. 200 GetBlockByHash(common.Hash) *types.Block 201 202 // CurrentBlock retrieves the head block from the local chain. 203 CurrentBlock() *types.Block 204 205 // CurrentFastBlock retrieves the head fast block from the local chain. 206 CurrentFastBlock() *types.Block 207 208 // FastSyncCommitHead directly commits the head block to a certain entity. 209 FastSyncCommitHead(common.Hash) error 210 211 // InsertChain inserts a batch of blocks into the local chain. 212 InsertChain(types.Blocks) (int, error) 213 214 // InsertReceiptChain inserts a batch of receipts into the local chain. 215 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 216 } 217 218 // New creates a new downloader to fetch hashes and blocks from remote peers. 219 func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 220 if lightchain == nil { 221 lightchain = chain 222 } 223 dl := &Downloader{ 224 stateDB: stateDb, 225 stateBloom: stateBloom, 226 mux: mux, 227 checkpoint: checkpoint, 228 queue: newQueue(blockCacheMaxItems, blockCacheInitialItems), 229 peers: newPeerSet(), 230 rttEstimate: uint64(rttMaxEstimate), 231 rttConfidence: uint64(1000000), 232 blockchain: chain, 233 lightchain: lightchain, 234 dropPeer: dropPeer, 235 headerCh: make(chan dataPack, 1), 236 bodyCh: make(chan dataPack, 1), 237 receiptCh: make(chan dataPack, 1), 238 bodyWakeCh: make(chan bool, 1), 239 receiptWakeCh: make(chan bool, 1), 240 headerProcCh: make(chan []*types.Header, 1), 241 quitCh: make(chan struct{}), 242 stateCh: make(chan dataPack), 243 SnapSyncer: snap.NewSyncer(stateDb, stateBloom), 244 stateSyncStart: make(chan *stateSync), 245 syncStatsState: stateSyncStats{ 246 processed: rawdb.ReadFastTrieProgress(stateDb), 247 }, 248 trackStateReq: make(chan *stateReq), 249 } 250 go dl.qosTuner() 251 go dl.stateFetcher() 252 return dl 253 } 254 255 // Progress retrieves the synchronisation boundaries, specifically the origin 256 // block where synchronisation started at (may have failed/suspended); the block 257 // or header sync is currently at; and the latest known block which the sync targets. 258 // 259 // In addition, during the state download phase of fast synchronisation the number 260 // of processed and the total number of known states are also returned. Otherwise 261 // these are zero. 262 func (d *Downloader) Progress() ethereum.SyncProgress { 263 // Lock the current stats and return the progress 264 d.syncStatsLock.RLock() 265 defer d.syncStatsLock.RUnlock() 266 267 current := uint64(0) 268 mode := d.getMode() 269 switch { 270 case d.blockchain != nil && mode == FullSync: 271 current = d.blockchain.CurrentBlock().NumberU64() 272 case d.blockchain != nil && mode == FastSync: 273 current = d.blockchain.CurrentFastBlock().NumberU64() 274 case d.lightchain != nil: 275 current = d.lightchain.CurrentHeader().Number.Uint64() 276 default: 277 log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", mode) 278 } 279 return ethereum.SyncProgress{ 280 StartingBlock: d.syncStatsChainOrigin, 281 CurrentBlock: current, 282 HighestBlock: d.syncStatsChainHeight, 283 PulledStates: d.syncStatsState.processed, 284 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 285 } 286 } 287 288 // Synchronising returns whether the downloader is currently retrieving blocks. 289 func (d *Downloader) Synchronising() bool { 290 return atomic.LoadInt32(&d.synchronising) > 0 291 } 292 293 // RegisterPeer injects a new download peer into the set of block source to be 294 // used for fetching hashes and blocks from. 295 func (d *Downloader) RegisterPeer(id string, version uint, peer Peer) error { 296 var logger log.Logger 297 if len(id) < 16 { 298 // Tests use short IDs, don't choke on them 299 logger = log.New("peer", id) 300 } else { 301 logger = log.New("peer", id[:8]) 302 } 303 logger.Trace("Registering sync peer") 304 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 305 logger.Error("Failed to register sync peer", "err", err) 306 return err 307 } 308 d.qosReduceConfidence() 309 310 return nil 311 } 312 313 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 314 func (d *Downloader) RegisterLightPeer(id string, version uint, peer LightPeer) error { 315 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 316 } 317 318 // UnregisterPeer remove a peer from the known list, preventing any action from 319 // the specified peer. An effort is also made to return any pending fetches into 320 // the queue. 321 func (d *Downloader) UnregisterPeer(id string) error { 322 // Unregister the peer from the active peer set and revoke any fetch tasks 323 var logger log.Logger 324 if len(id) < 16 { 325 // Tests use short IDs, don't choke on them 326 logger = log.New("peer", id) 327 } else { 328 logger = log.New("peer", id[:8]) 329 } 330 logger.Trace("Unregistering sync peer") 331 if err := d.peers.Unregister(id); err != nil { 332 logger.Error("Failed to unregister sync peer", "err", err) 333 return err 334 } 335 d.queue.Revoke(id) 336 337 return nil 338 } 339 340 // Synchronise tries to sync up our local block chain with a remote peer, both 341 // adding various sanity checks as well as wrapping it with various log entries. 342 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 343 err := d.synchronise(id, head, td, mode) 344 345 switch err { 346 case nil, errBusy, errCanceled: 347 return err 348 } 349 if errors.Is(err, errInvalidChain) || errors.Is(err, errBadPeer) || errors.Is(err, errTimeout) || 350 errors.Is(err, errStallingPeer) || errors.Is(err, errUnsyncedPeer) || errors.Is(err, errEmptyHeaderSet) || 351 errors.Is(err, errPeersUnavailable) || errors.Is(err, errTooOld) || errors.Is(err, errInvalidAncestor) { 352 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 353 if d.dropPeer == nil { 354 // The dropPeer method is nil when `--copydb` is used for a local copy. 355 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 356 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 357 } else { 358 d.dropPeer(id) 359 } 360 return err 361 } 362 log.Warn("Synchronisation failed, retrying", "err", err) 363 return err 364 } 365 366 // synchronise will select the peer and use it for synchronising. If an empty string is given 367 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 368 // checks fail an error will be returned. This method is synchronous 369 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 370 // Mock out the synchronisation if testing 371 if d.synchroniseMock != nil { 372 return d.synchroniseMock(id, hash) 373 } 374 // Make sure only one goroutine is ever allowed past this point at once 375 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 376 return errBusy 377 } 378 defer atomic.StoreInt32(&d.synchronising, 0) 379 380 // Post a user notification of the sync (only once per session) 381 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 382 log.Info("Block synchronisation started") 383 } 384 // If we are already full syncing, but have a fast-sync bloom filter laying 385 // around, make sure it doesn't use memory any more. This is a special case 386 // when the user attempts to fast sync a new empty network. 387 if mode == FullSync && d.stateBloom != nil { 388 d.stateBloom.Close() 389 } 390 // If snap sync was requested, create the snap scheduler and switch to fast 391 // sync mode. Long term we could drop fast sync or merge the two together, 392 // but until snap becomes prevalent, we should support both. TODO(karalabe). 393 if mode == SnapSync { 394 if !d.snapSync { 395 log.Warn("Enabling snapshot sync prototype") 396 d.snapSync = true 397 } 398 mode = FastSync 399 } 400 // Reset the queue, peer set and wake channels to clean any internal leftover state 401 d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems) 402 d.peers.Reset() 403 404 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 405 select { 406 case <-ch: 407 default: 408 } 409 } 410 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 411 for empty := false; !empty; { 412 select { 413 case <-ch: 414 default: 415 empty = true 416 } 417 } 418 } 419 for empty := false; !empty; { 420 select { 421 case <-d.headerProcCh: 422 default: 423 empty = true 424 } 425 } 426 // Create cancel channel for aborting mid-flight and mark the master peer 427 d.cancelLock.Lock() 428 d.cancelCh = make(chan struct{}) 429 d.cancelPeer = id 430 d.cancelLock.Unlock() 431 432 defer d.Cancel() // No matter what, we can't leave the cancel channel open 433 434 // Atomically set the requested sync mode 435 atomic.StoreUint32(&d.mode, uint32(mode)) 436 437 // Retrieve the origin peer and initiate the downloading process 438 p := d.peers.Peer(id) 439 if p == nil { 440 return errUnknownPeer 441 } 442 return d.syncWithPeer(p, hash, td) 443 } 444 445 func (d *Downloader) getMode() SyncMode { 446 return SyncMode(atomic.LoadUint32(&d.mode)) 447 } 448 449 // syncWithPeer starts a block synchronization based on the hash chain from the 450 // specified peer and head hash. 451 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 452 d.mux.Post(StartEvent{}) 453 defer func() { 454 // reset on error 455 if err != nil { 456 d.mux.Post(FailedEvent{err}) 457 } else { 458 latest := d.lightchain.CurrentHeader() 459 d.mux.Post(DoneEvent{latest}) 460 } 461 }() 462 if p.version < 64 { 463 return fmt.Errorf("%w: advertized %d < required %d", errTooOld, p.version, 64) 464 } 465 mode := d.getMode() 466 467 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", mode) 468 defer func(start time.Time) { 469 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 470 }(time.Now()) 471 472 // Look up the sync boundaries: the common ancestor and the target block 473 latest, pivot, err := d.fetchHead(p) 474 if err != nil { 475 return err 476 } 477 if mode == FastSync && pivot == nil { 478 // If no pivot block was returned, the head is below the min full block 479 // threshold (i.e. new chian). In that case we won't really fast sync 480 // anyway, but still need a valid pivot block to avoid some code hitting 481 // nil panics on an access. 482 pivot = d.blockchain.CurrentBlock().Header() 483 } 484 height := latest.Number.Uint64() 485 486 origin, err := d.findAncestor(p, latest) 487 if err != nil { 488 return err 489 } 490 d.syncStatsLock.Lock() 491 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 492 d.syncStatsChainOrigin = origin 493 } 494 d.syncStatsChainHeight = height 495 d.syncStatsLock.Unlock() 496 497 // Ensure our origin point is below any fast sync pivot point 498 if mode == FastSync { 499 if height <= uint64(fsMinFullBlocks) { 500 origin = 0 501 } else { 502 pivotNumber := pivot.Number.Uint64() 503 if pivotNumber <= origin { 504 origin = pivotNumber - 1 505 } 506 // Write out the pivot into the database so a rollback beyond it will 507 // reenable fast sync 508 rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber) 509 } 510 } 511 d.committed = 1 512 if mode == FastSync && pivot.Number.Uint64() != 0 { 513 d.committed = 0 514 } 515 if mode == FastSync { 516 // Set the ancient data limitation. 517 // If we are running fast sync, all block data older than ancientLimit will be 518 // written to the ancient store. More recent data will be written to the active 519 // database and will wait for the freezer to migrate. 520 // 521 // If there is a checkpoint available, then calculate the ancientLimit through 522 // that. Otherwise calculate the ancient limit through the advertised height 523 // of the remote peer. 524 // 525 // The reason for picking checkpoint first is that a malicious peer can give us 526 // a fake (very high) height, forcing the ancient limit to also be very high. 527 // The peer would start to feed us valid blocks until head, resulting in all of 528 // the blocks might be written into the ancient store. A following mini-reorg 529 // could cause issues. 530 if d.checkpoint != 0 && d.checkpoint > fullMaxForkAncestry+1 { 531 d.ancientLimit = d.checkpoint 532 } else if height > fullMaxForkAncestry+1 { 533 d.ancientLimit = height - fullMaxForkAncestry - 1 534 } else { 535 d.ancientLimit = 0 536 } 537 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 538 539 // If a part of blockchain data has already been written into active store, 540 // disable the ancient style insertion explicitly. 541 if origin >= frozen && frozen != 0 { 542 d.ancientLimit = 0 543 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 544 } else if d.ancientLimit > 0 { 545 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 546 } 547 // Rewind the ancient store and blockchain if reorg happens. 548 if origin+1 < frozen { 549 if err := d.lightchain.SetHead(origin + 1); err != nil { 550 return err 551 } 552 } 553 } 554 // Initiate the sync using a concurrent header and content retrieval algorithm 555 d.queue.Prepare(origin+1, mode) 556 if d.syncInitHook != nil { 557 d.syncInitHook(origin, height) 558 } 559 fetchers := []func() error{ 560 func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved 561 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 562 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 563 func() error { return d.processHeaders(origin+1, td) }, 564 } 565 if mode == FastSync { 566 d.pivotLock.Lock() 567 d.pivotHeader = pivot 568 d.pivotLock.Unlock() 569 570 fetchers = append(fetchers, func() error { return d.processFastSyncContent() }) 571 } else if mode == FullSync { 572 fetchers = append(fetchers, d.processFullSyncContent) 573 } 574 return d.spawnSync(fetchers) 575 } 576 577 // spawnSync runs d.process and all given fetcher functions to completion in 578 // separate goroutines, returning the first error that appears. 579 func (d *Downloader) spawnSync(fetchers []func() error) error { 580 errc := make(chan error, len(fetchers)) 581 d.cancelWg.Add(len(fetchers)) 582 for _, fn := range fetchers { 583 fn := fn 584 go func() { defer d.cancelWg.Done(); errc <- fn() }() 585 } 586 // Wait for the first error, then terminate the others. 587 var err error 588 for i := 0; i < len(fetchers); i++ { 589 if i == len(fetchers)-1 { 590 // Close the queue when all fetchers have exited. 591 // This will cause the block processor to end when 592 // it has processed the queue. 593 d.queue.Close() 594 } 595 if err = <-errc; err != nil && err != errCanceled { 596 break 597 } 598 } 599 d.queue.Close() 600 d.Cancel() 601 return err 602 } 603 604 // cancel aborts all of the operations and resets the queue. However, cancel does 605 // not wait for the running download goroutines to finish. This method should be 606 // used when cancelling the downloads from inside the downloader. 607 func (d *Downloader) cancel() { 608 // Close the current cancel channel 609 d.cancelLock.Lock() 610 defer d.cancelLock.Unlock() 611 612 if d.cancelCh != nil { 613 select { 614 case <-d.cancelCh: 615 // Channel was already closed 616 default: 617 close(d.cancelCh) 618 } 619 } 620 } 621 622 // Cancel aborts all of the operations and waits for all download goroutines to 623 // finish before returning. 624 func (d *Downloader) Cancel() { 625 d.cancel() 626 d.cancelWg.Wait() 627 } 628 629 // Terminate interrupts the downloader, canceling all pending operations. 630 // The downloader cannot be reused after calling Terminate. 631 func (d *Downloader) Terminate() { 632 // Close the termination channel (make sure double close is allowed) 633 d.quitLock.Lock() 634 select { 635 case <-d.quitCh: 636 default: 637 close(d.quitCh) 638 } 639 if d.stateBloom != nil { 640 d.stateBloom.Close() 641 } 642 d.quitLock.Unlock() 643 644 // Cancel any pending download requests 645 d.Cancel() 646 } 647 648 // fetchHead retrieves the head header and prior pivot block (if available) from 649 // a remote peer. 650 func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *types.Header, err error) { 651 p.log.Debug("Retrieving remote chain head") 652 mode := d.getMode() 653 654 // Request the advertised remote head block and wait for the response 655 latest, _ := p.peer.Head() 656 fetch := 1 657 if mode == FastSync { 658 fetch = 2 // head + pivot headers 659 } 660 go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true) 661 662 ttl := d.requestTTL() 663 timeout := time.After(ttl) 664 for { 665 select { 666 case <-d.cancelCh: 667 return nil, nil, errCanceled 668 669 case packet := <-d.headerCh: 670 // Discard anything not from the origin peer 671 if packet.PeerId() != p.id { 672 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 673 break 674 } 675 // Make sure the peer gave us at least one and at most the requested headers 676 headers := packet.(*headerPack).headers 677 if len(headers) == 0 || len(headers) > fetch { 678 return nil, nil, fmt.Errorf("%w: returned headers %d != requested %d", errBadPeer, len(headers), fetch) 679 } 680 // The first header needs to be the head, validate against the checkpoint 681 // and request. If only 1 header was returned, make sure there's no pivot 682 // or there was not one requested. 683 head := headers[0] 684 if (mode == FastSync || mode == LightSync) && head.Number.Uint64() < d.checkpoint { 685 return nil, nil, fmt.Errorf("%w: remote head %d below checkpoint %d", errUnsyncedPeer, head.Number, d.checkpoint) 686 } 687 if len(headers) == 1 { 688 if mode == FastSync && head.Number.Uint64() > uint64(fsMinFullBlocks) { 689 return nil, nil, fmt.Errorf("%w: no pivot included along head header", errBadPeer) 690 } 691 p.log.Debug("Remote head identified, no pivot", "number", head.Number, "hash", head.Hash()) 692 return head, nil, nil 693 } 694 // At this point we have 2 headers in total and the first is the 695 // validated head of the chian. Check the pivot number and return, 696 pivot := headers[1] 697 if pivot.Number.Uint64() != head.Number.Uint64()-uint64(fsMinFullBlocks) { 698 return nil, nil, fmt.Errorf("%w: remote pivot %d != requested %d", errInvalidChain, pivot.Number, head.Number.Uint64()-uint64(fsMinFullBlocks)) 699 } 700 return head, pivot, nil 701 702 case <-timeout: 703 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 704 return nil, nil, errTimeout 705 706 case <-d.bodyCh: 707 case <-d.receiptCh: 708 // Out of bounds delivery, ignore 709 } 710 } 711 } 712 713 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 714 // common ancestor. 715 // It returns parameters to be used for peer.RequestHeadersByNumber: 716 // from - starting block number 717 // count - number of headers to request 718 // skip - number of headers to skip 719 // and also returns 'max', the last block which is expected to be returned by the remote peers, 720 // given the (from,count,skip) 721 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 722 var ( 723 from int 724 count int 725 MaxCount = MaxHeaderFetch / 16 726 ) 727 // requestHead is the highest block that we will ask for. If requestHead is not offset, 728 // the highest block that we will get is 16 blocks back from head, which means we 729 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 730 // between us and the peer is 1-2 blocks, which is most common 731 requestHead := int(remoteHeight) - 1 732 if requestHead < 0 { 733 requestHead = 0 734 } 735 // requestBottom is the lowest block we want included in the query 736 // Ideally, we want to include the one just below our own head 737 requestBottom := int(localHeight - 1) 738 if requestBottom < 0 { 739 requestBottom = 0 740 } 741 totalSpan := requestHead - requestBottom 742 span := 1 + totalSpan/MaxCount 743 if span < 2 { 744 span = 2 745 } 746 if span > 16 { 747 span = 16 748 } 749 750 count = 1 + totalSpan/span 751 if count > MaxCount { 752 count = MaxCount 753 } 754 if count < 2 { 755 count = 2 756 } 757 from = requestHead - (count-1)*span 758 if from < 0 { 759 from = 0 760 } 761 max := from + (count-1)*span 762 return int64(from), count, span - 1, uint64(max) 763 } 764 765 // findAncestor tries to locate the common ancestor link of the local chain and 766 // a remote peers blockchain. In the general case when our node was in sync and 767 // on the correct chain, checking the top N links should already get us a match. 768 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 769 // the head links match), we do a binary search to find the common ancestor. 770 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 771 // Figure out the valid ancestor range to prevent rewrite attacks 772 var ( 773 floor = int64(-1) 774 localHeight uint64 775 remoteHeight = remoteHeader.Number.Uint64() 776 ) 777 mode := d.getMode() 778 switch mode { 779 case FullSync: 780 localHeight = d.blockchain.CurrentBlock().NumberU64() 781 case FastSync: 782 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 783 default: 784 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 785 } 786 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 787 788 // Recap floor value for binary search 789 maxForkAncestry := fullMaxForkAncestry 790 if d.getMode() == LightSync { 791 maxForkAncestry = lightMaxForkAncestry 792 } 793 if localHeight >= maxForkAncestry { 794 // We're above the max reorg threshold, find the earliest fork point 795 floor = int64(localHeight - maxForkAncestry) 796 } 797 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 798 // all headers before that point will be missing. 799 if mode == LightSync { 800 // If we don't know the current CHT position, find it 801 if d.genesis == 0 { 802 header := d.lightchain.CurrentHeader() 803 for header != nil { 804 d.genesis = header.Number.Uint64() 805 if floor >= int64(d.genesis)-1 { 806 break 807 } 808 header = d.lightchain.GetHeaderByHash(header.ParentHash) 809 } 810 } 811 // We already know the "genesis" block number, cap floor to that 812 if floor < int64(d.genesis)-1 { 813 floor = int64(d.genesis) - 1 814 } 815 } 816 817 ancestor, err := d.findAncestorSpanSearch(p, mode, remoteHeight, localHeight, floor) 818 if err == nil { 819 return ancestor, nil 820 } 821 // The returned error was not nil. 822 // If the error returned does not reflect that a common ancestor was not found, return it. 823 // If the error reflects that a common ancestor was not found, continue to binary search, 824 // where the error value will be reassigned. 825 if !errors.Is(err, errNoAncestorFound) { 826 return 0, err 827 } 828 829 ancestor, err = d.findAncestorBinarySearch(p, mode, remoteHeight, floor) 830 if err != nil { 831 return 0, err 832 } 833 return ancestor, nil 834 } 835 836 func (d *Downloader) findAncestorSpanSearch(p *peerConnection, mode SyncMode, remoteHeight, localHeight uint64, floor int64) (commonAncestor uint64, err error) { 837 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 838 839 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 840 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 841 842 // Wait for the remote response to the head fetch 843 number, hash := uint64(0), common.Hash{} 844 845 ttl := d.requestTTL() 846 timeout := time.After(ttl) 847 848 for finished := false; !finished; { 849 select { 850 case <-d.cancelCh: 851 return 0, errCanceled 852 853 case packet := <-d.headerCh: 854 // Discard anything not from the origin peer 855 if packet.PeerId() != p.id { 856 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 857 break 858 } 859 // Make sure the peer actually gave something valid 860 headers := packet.(*headerPack).headers 861 if len(headers) == 0 { 862 p.log.Warn("Empty head header set") 863 return 0, errEmptyHeaderSet 864 } 865 // Make sure the peer's reply conforms to the request 866 for i, header := range headers { 867 expectNumber := from + int64(i)*int64(skip+1) 868 if number := header.Number.Int64(); number != expectNumber { 869 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 870 return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering")) 871 } 872 } 873 // Check if a common ancestor was found 874 finished = true 875 for i := len(headers) - 1; i >= 0; i-- { 876 // Skip any headers that underflow/overflow our requested set 877 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 878 continue 879 } 880 // Otherwise check if we already know the header or not 881 h := headers[i].Hash() 882 n := headers[i].Number.Uint64() 883 884 var known bool 885 switch mode { 886 case FullSync: 887 known = d.blockchain.HasBlock(h, n) 888 case FastSync: 889 known = d.blockchain.HasFastBlock(h, n) 890 default: 891 known = d.lightchain.HasHeader(h, n) 892 } 893 if known { 894 number, hash = n, h 895 break 896 } 897 } 898 899 case <-timeout: 900 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 901 return 0, errTimeout 902 903 case <-d.bodyCh: 904 case <-d.receiptCh: 905 // Out of bounds delivery, ignore 906 } 907 } 908 // If the head fetch already found an ancestor, return 909 if hash != (common.Hash{}) { 910 if int64(number) <= floor { 911 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 912 return 0, errInvalidAncestor 913 } 914 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 915 return number, nil 916 } 917 return 0, errNoAncestorFound 918 } 919 920 func (d *Downloader) findAncestorBinarySearch(p *peerConnection, mode SyncMode, remoteHeight uint64, floor int64) (commonAncestor uint64, err error) { 921 hash := common.Hash{} 922 923 // Ancestor not found, we need to binary search over our chain 924 start, end := uint64(0), remoteHeight 925 if floor > 0 { 926 start = uint64(floor) 927 } 928 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 929 930 for start+1 < end { 931 // Split our chain interval in two, and request the hash to cross check 932 check := (start + end) / 2 933 934 ttl := d.requestTTL() 935 timeout := time.After(ttl) 936 937 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 938 939 // Wait until a reply arrives to this request 940 for arrived := false; !arrived; { 941 select { 942 case <-d.cancelCh: 943 return 0, errCanceled 944 945 case packet := <-d.headerCh: 946 // Discard anything not from the origin peer 947 if packet.PeerId() != p.id { 948 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 949 break 950 } 951 // Make sure the peer actually gave something valid 952 headers := packet.(*headerPack).headers 953 if len(headers) != 1 { 954 p.log.Warn("Multiple headers for single request", "headers", len(headers)) 955 return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers)) 956 } 957 arrived = true 958 959 // Modify the search interval based on the response 960 h := headers[0].Hash() 961 n := headers[0].Number.Uint64() 962 963 var known bool 964 switch mode { 965 case FullSync: 966 known = d.blockchain.HasBlock(h, n) 967 case FastSync: 968 known = d.blockchain.HasFastBlock(h, n) 969 default: 970 known = d.lightchain.HasHeader(h, n) 971 } 972 if !known { 973 end = check 974 break 975 } 976 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 977 if header.Number.Uint64() != check { 978 p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 979 return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number) 980 } 981 start = check 982 hash = h 983 984 case <-timeout: 985 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 986 return 0, errTimeout 987 988 case <-d.bodyCh: 989 case <-d.receiptCh: 990 // Out of bounds delivery, ignore 991 } 992 } 993 } 994 // Ensure valid ancestry and return 995 if int64(start) <= floor { 996 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 997 return 0, errInvalidAncestor 998 } 999 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 1000 return start, nil 1001 } 1002 1003 // fetchHeaders keeps retrieving headers concurrently from the number 1004 // requested, until no more are returned, potentially throttling on the way. To 1005 // facilitate concurrency but still protect against malicious nodes sending bad 1006 // headers, we construct a header chain skeleton using the "origin" peer we are 1007 // syncing with, and fill in the missing headers using anyone else. Headers from 1008 // other peers are only accepted if they map cleanly to the skeleton. If no one 1009 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 1010 // the origin is dropped. 1011 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error { 1012 p.log.Debug("Directing header downloads", "origin", from) 1013 defer p.log.Debug("Header download terminated") 1014 1015 // Create a timeout timer, and the associated header fetcher 1016 skeleton := true // Skeleton assembly phase or finishing up 1017 pivoting := false // Whether the next request is pivot verification 1018 request := time.Now() // time of the last skeleton fetch request 1019 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 1020 <-timeout.C // timeout channel should be initially empty 1021 defer timeout.Stop() 1022 1023 var ttl time.Duration 1024 getHeaders := func(from uint64) { 1025 request = time.Now() 1026 1027 ttl = d.requestTTL() 1028 timeout.Reset(ttl) 1029 1030 if skeleton { 1031 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 1032 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 1033 } else { 1034 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 1035 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 1036 } 1037 } 1038 getNextPivot := func() { 1039 pivoting = true 1040 request = time.Now() 1041 1042 ttl = d.requestTTL() 1043 timeout.Reset(ttl) 1044 1045 d.pivotLock.RLock() 1046 pivot := d.pivotHeader.Number.Uint64() 1047 d.pivotLock.RUnlock() 1048 1049 p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks)) 1050 go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep 1051 } 1052 // Start pulling the header chain skeleton until all is done 1053 ancestor := from 1054 getHeaders(from) 1055 1056 mode := d.getMode() 1057 for { 1058 select { 1059 case <-d.cancelCh: 1060 return errCanceled 1061 1062 case packet := <-d.headerCh: 1063 // Make sure the active peer is giving us the skeleton headers 1064 if packet.PeerId() != p.id { 1065 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 1066 break 1067 } 1068 headerReqTimer.UpdateSince(request) 1069 timeout.Stop() 1070 1071 // If the pivot is being checked, move if it became stale and run the real retrieval 1072 var pivot uint64 1073 1074 d.pivotLock.RLock() 1075 if d.pivotHeader != nil { 1076 pivot = d.pivotHeader.Number.Uint64() 1077 } 1078 d.pivotLock.RUnlock() 1079 1080 if pivoting { 1081 if packet.Items() == 2 { 1082 // Retrieve the headers and do some sanity checks, just in case 1083 headers := packet.(*headerPack).headers 1084 1085 if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want { 1086 log.Warn("Peer sent invalid next pivot", "have", have, "want", want) 1087 return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want) 1088 } 1089 if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want { 1090 log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want) 1091 return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want) 1092 } 1093 log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number) 1094 pivot = headers[0].Number.Uint64() 1095 1096 d.pivotLock.Lock() 1097 d.pivotHeader = headers[0] 1098 d.pivotLock.Unlock() 1099 1100 // Write out the pivot into the database so a rollback beyond 1101 // it will reenable fast sync and update the state root that 1102 // the state syncer will be downloading. 1103 rawdb.WriteLastPivotNumber(d.stateDB, pivot) 1104 } 1105 pivoting = false 1106 getHeaders(from) 1107 continue 1108 } 1109 // If the skeleton's finished, pull any remaining head headers directly from the origin 1110 if skeleton && packet.Items() == 0 { 1111 skeleton = false 1112 getHeaders(from) 1113 continue 1114 } 1115 // If no more headers are inbound, notify the content fetchers and return 1116 if packet.Items() == 0 { 1117 // Don't abort header fetches while the pivot is downloading 1118 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 1119 p.log.Debug("No headers, waiting for pivot commit") 1120 select { 1121 case <-time.After(fsHeaderContCheck): 1122 getHeaders(from) 1123 continue 1124 case <-d.cancelCh: 1125 return errCanceled 1126 } 1127 } 1128 // Pivot done (or not in fast sync) and no more headers, terminate the process 1129 p.log.Debug("No more headers available") 1130 select { 1131 case d.headerProcCh <- nil: 1132 return nil 1133 case <-d.cancelCh: 1134 return errCanceled 1135 } 1136 } 1137 headers := packet.(*headerPack).headers 1138 1139 // If we received a skeleton batch, resolve internals concurrently 1140 if skeleton { 1141 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1142 if err != nil { 1143 p.log.Debug("Skeleton chain invalid", "err", err) 1144 return fmt.Errorf("%w: %v", errInvalidChain, err) 1145 } 1146 headers = filled[proced:] 1147 from += uint64(proced) 1148 } else { 1149 // If we're closing in on the chain head, but haven't yet reached it, delay 1150 // the last few headers so mini reorgs on the head don't cause invalid hash 1151 // chain errors. 1152 if n := len(headers); n > 0 { 1153 // Retrieve the current head we're at 1154 var head uint64 1155 if mode == LightSync { 1156 head = d.lightchain.CurrentHeader().Number.Uint64() 1157 } else { 1158 head = d.blockchain.CurrentFastBlock().NumberU64() 1159 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1160 head = full 1161 } 1162 } 1163 // If the head is below the common ancestor, we're actually deduplicating 1164 // already existing chain segments, so use the ancestor as the fake head. 1165 // Otherwise we might end up delaying header deliveries pointlessly. 1166 if head < ancestor { 1167 head = ancestor 1168 } 1169 // If the head is way older than this batch, delay the last few headers 1170 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1171 delay := reorgProtHeaderDelay 1172 if delay > n { 1173 delay = n 1174 } 1175 headers = headers[:n-delay] 1176 } 1177 } 1178 } 1179 // Insert all the new headers and fetch the next batch 1180 if len(headers) > 0 { 1181 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1182 select { 1183 case d.headerProcCh <- headers: 1184 case <-d.cancelCh: 1185 return errCanceled 1186 } 1187 from += uint64(len(headers)) 1188 1189 // If we're still skeleton filling fast sync, check pivot staleness 1190 // before continuing to the next skeleton filling 1191 if skeleton && pivot > 0 { 1192 getNextPivot() 1193 } else { 1194 getHeaders(from) 1195 } 1196 } else { 1197 // No headers delivered, or all of them being delayed, sleep a bit and retry 1198 p.log.Trace("All headers delayed, waiting") 1199 select { 1200 case <-time.After(fsHeaderContCheck): 1201 getHeaders(from) 1202 continue 1203 case <-d.cancelCh: 1204 return errCanceled 1205 } 1206 } 1207 1208 case <-timeout.C: 1209 if d.dropPeer == nil { 1210 // The dropPeer method is nil when `--copydb` is used for a local copy. 1211 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1212 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1213 break 1214 } 1215 // Header retrieval timed out, consider the peer bad and drop 1216 p.log.Debug("Header request timed out", "elapsed", ttl) 1217 headerTimeoutMeter.Mark(1) 1218 d.dropPeer(p.id) 1219 1220 // Finish the sync gracefully instead of dumping the gathered data though 1221 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1222 select { 1223 case ch <- false: 1224 case <-d.cancelCh: 1225 } 1226 } 1227 select { 1228 case d.headerProcCh <- nil: 1229 case <-d.cancelCh: 1230 } 1231 return fmt.Errorf("%w: header request timed out", errBadPeer) 1232 } 1233 } 1234 } 1235 1236 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1237 // and maps them to the provided skeleton header chain. 1238 // 1239 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1240 // immediately to the header processor to keep the rest of the pipeline full even 1241 // in the case of header stalls. 1242 // 1243 // The method returns the entire filled skeleton and also the number of headers 1244 // already forwarded for processing. 1245 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1246 log.Debug("Filling up skeleton", "from", from) 1247 d.queue.ScheduleSkeleton(from, skeleton) 1248 1249 var ( 1250 deliver = func(packet dataPack) (int, error) { 1251 pack := packet.(*headerPack) 1252 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1253 } 1254 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 1255 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) { 1256 return d.queue.ReserveHeaders(p, count), false, false 1257 } 1258 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1259 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 1260 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1261 p.SetHeadersIdle(accepted, deliveryTime) 1262 } 1263 ) 1264 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1265 d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve, 1266 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1267 1268 log.Debug("Skeleton fill terminated", "err", err) 1269 1270 filled, proced := d.queue.RetrieveHeaders() 1271 return filled, proced, err 1272 } 1273 1274 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1275 // available peers, reserving a chunk of blocks for each, waiting for delivery 1276 // and also periodically checking for timeouts. 1277 func (d *Downloader) fetchBodies(from uint64) error { 1278 log.Debug("Downloading block bodies", "origin", from) 1279 1280 var ( 1281 deliver = func(packet dataPack) (int, error) { 1282 pack := packet.(*bodyPack) 1283 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1284 } 1285 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 1286 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1287 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 1288 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) } 1289 ) 1290 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1291 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies, 1292 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1293 1294 log.Debug("Block body download terminated", "err", err) 1295 return err 1296 } 1297 1298 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1299 // available peers, reserving a chunk of receipts for each, waiting for delivery 1300 // and also periodically checking for timeouts. 1301 func (d *Downloader) fetchReceipts(from uint64) error { 1302 log.Debug("Downloading transaction receipts", "origin", from) 1303 1304 var ( 1305 deliver = func(packet dataPack) (int, error) { 1306 pack := packet.(*receiptPack) 1307 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1308 } 1309 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1310 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1311 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1312 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1313 p.SetReceiptsIdle(accepted, deliveryTime) 1314 } 1315 ) 1316 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1317 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts, 1318 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1319 1320 log.Debug("Transaction receipt download terminated", "err", err) 1321 return err 1322 } 1323 1324 // fetchParts iteratively downloads scheduled block parts, taking any available 1325 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1326 // also periodically checking for timeouts. 1327 // 1328 // As the scheduling/timeout logic mostly is the same for all downloaded data 1329 // types, this method is used by each for data gathering and is instrumented with 1330 // various callbacks to handle the slight differences between processing them. 1331 // 1332 // The instrumentation parameters: 1333 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1334 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1335 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1336 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1337 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1338 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1339 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1340 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1341 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1342 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1343 // - fetch: network callback to actually send a particular download request to a physical remote peer 1344 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1345 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1346 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1347 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1348 // - kind: textual label of the type being downloaded to display in log messages 1349 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1350 expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool), 1351 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1352 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error { 1353 1354 // Create a ticker to detect expired retrieval tasks 1355 ticker := time.NewTicker(100 * time.Millisecond) 1356 defer ticker.Stop() 1357 1358 update := make(chan struct{}, 1) 1359 1360 // Prepare the queue and fetch block parts until the block header fetcher's done 1361 finished := false 1362 for { 1363 select { 1364 case <-d.cancelCh: 1365 return errCanceled 1366 1367 case packet := <-deliveryCh: 1368 deliveryTime := time.Now() 1369 // If the peer was previously banned and failed to deliver its pack 1370 // in a reasonable time frame, ignore its message. 1371 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1372 // Deliver the received chunk of data and check chain validity 1373 accepted, err := deliver(packet) 1374 if errors.Is(err, errInvalidChain) { 1375 return err 1376 } 1377 // Unless a peer delivered something completely else than requested (usually 1378 // caused by a timed out request which came through in the end), set it to 1379 // idle. If the delivery's stale, the peer should have already been idled. 1380 if !errors.Is(err, errStaleDelivery) { 1381 setIdle(peer, accepted, deliveryTime) 1382 } 1383 // Issue a log to the user to see what's going on 1384 switch { 1385 case err == nil && packet.Items() == 0: 1386 peer.log.Trace("Requested data not delivered", "type", kind) 1387 case err == nil: 1388 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1389 default: 1390 peer.log.Debug("Failed to deliver retrieved data", "type", kind, "err", err) 1391 } 1392 } 1393 // Blocks assembled, try to update the progress 1394 select { 1395 case update <- struct{}{}: 1396 default: 1397 } 1398 1399 case cont := <-wakeCh: 1400 // The header fetcher sent a continuation flag, check if it's done 1401 if !cont { 1402 finished = true 1403 } 1404 // Headers arrive, try to update the progress 1405 select { 1406 case update <- struct{}{}: 1407 default: 1408 } 1409 1410 case <-ticker.C: 1411 // Sanity check update the progress 1412 select { 1413 case update <- struct{}{}: 1414 default: 1415 } 1416 1417 case <-update: 1418 // Short circuit if we lost all our peers 1419 if d.peers.Len() == 0 { 1420 return errNoPeers 1421 } 1422 // Check for fetch request timeouts and demote the responsible peers 1423 for pid, fails := range expire() { 1424 if peer := d.peers.Peer(pid); peer != nil { 1425 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1426 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1427 // out that sync wise we need to get rid of the peer. 1428 // 1429 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1430 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1431 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1432 if fails > 2 { 1433 peer.log.Trace("Data delivery timed out", "type", kind) 1434 setIdle(peer, 0, time.Now()) 1435 } else { 1436 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1437 1438 if d.dropPeer == nil { 1439 // The dropPeer method is nil when `--copydb` is used for a local copy. 1440 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1441 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1442 } else { 1443 d.dropPeer(pid) 1444 1445 // If this peer was the master peer, abort sync immediately 1446 d.cancelLock.RLock() 1447 master := pid == d.cancelPeer 1448 d.cancelLock.RUnlock() 1449 1450 if master { 1451 d.cancel() 1452 return errTimeout 1453 } 1454 } 1455 } 1456 } 1457 } 1458 // If there's nothing more to fetch, wait or terminate 1459 if pending() == 0 { 1460 if !inFlight() && finished { 1461 log.Debug("Data fetching completed", "type", kind) 1462 return nil 1463 } 1464 break 1465 } 1466 // Send a download request to all idle peers, until throttled 1467 progressed, throttled, running := false, false, inFlight() 1468 idles, total := idle() 1469 pendCount := pending() 1470 for _, peer := range idles { 1471 // Short circuit if throttling activated 1472 if throttled { 1473 break 1474 } 1475 // Short circuit if there is no more available task. 1476 if pendCount = pending(); pendCount == 0 { 1477 break 1478 } 1479 // Reserve a chunk of fetches for a peer. A nil can mean either that 1480 // no more headers are available, or that the peer is known not to 1481 // have them. 1482 request, progress, throttle := reserve(peer, capacity(peer)) 1483 if progress { 1484 progressed = true 1485 } 1486 if throttle { 1487 throttled = true 1488 throttleCounter.Inc(1) 1489 } 1490 if request == nil { 1491 continue 1492 } 1493 if request.From > 0 { 1494 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1495 } else { 1496 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1497 } 1498 // Fetch the chunk and make sure any errors return the hashes to the queue 1499 if fetchHook != nil { 1500 fetchHook(request.Headers) 1501 } 1502 if err := fetch(peer, request); err != nil { 1503 // Although we could try and make an attempt to fix this, this error really 1504 // means that we've double allocated a fetch task to a peer. If that is the 1505 // case, the internal state of the downloader and the queue is very wrong so 1506 // better hard crash and note the error instead of silently accumulating into 1507 // a much bigger issue. 1508 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1509 } 1510 running = true 1511 } 1512 // Make sure that we have peers available for fetching. If all peers have been tried 1513 // and all failed throw an error 1514 if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 { 1515 return errPeersUnavailable 1516 } 1517 } 1518 } 1519 } 1520 1521 // processHeaders takes batches of retrieved headers from an input channel and 1522 // keeps processing and scheduling them into the header chain and downloader's 1523 // queue until the stream ends or a failure occurs. 1524 func (d *Downloader) processHeaders(origin uint64, td *big.Int) error { 1525 // Keep a count of uncertain headers to roll back 1526 var ( 1527 rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis) 1528 rollbackErr error 1529 mode = d.getMode() 1530 ) 1531 defer func() { 1532 if rollback > 0 { 1533 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1534 if mode != LightSync { 1535 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1536 lastBlock = d.blockchain.CurrentBlock().Number() 1537 } 1538 if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block 1539 // We're already unwinding the stack, only print the error to make it more visible 1540 log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err) 1541 } 1542 curFastBlock, curBlock := common.Big0, common.Big0 1543 if mode != LightSync { 1544 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1545 curBlock = d.blockchain.CurrentBlock().Number() 1546 } 1547 log.Warn("Rolled back chain segment", 1548 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1549 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1550 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr) 1551 } 1552 }() 1553 // Wait for batches of headers to process 1554 gotHeaders := false 1555 1556 for { 1557 select { 1558 case <-d.cancelCh: 1559 rollbackErr = errCanceled 1560 return errCanceled 1561 1562 case headers := <-d.headerProcCh: 1563 // Terminate header processing if we synced up 1564 if len(headers) == 0 { 1565 // Notify everyone that headers are fully processed 1566 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1567 select { 1568 case ch <- false: 1569 case <-d.cancelCh: 1570 } 1571 } 1572 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1573 // better chain compared to ours. The only exception is if its promised blocks were 1574 // already imported by other means (e.g. fetcher): 1575 // 1576 // R <remote peer>, L <local node>: Both at block 10 1577 // R: Mine block 11, and propagate it to L 1578 // L: Queue block 11 for import 1579 // L: Notice that R's head and TD increased compared to ours, start sync 1580 // L: Import of block 11 finishes 1581 // L: Sync begins, and finds common ancestor at 11 1582 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1583 // R: Nothing to give 1584 if mode != LightSync { 1585 head := d.blockchain.CurrentBlock() 1586 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1587 return errStallingPeer 1588 } 1589 } 1590 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1591 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1592 // of delivering the post-pivot blocks that would flag the invalid content. 1593 // 1594 // This check cannot be executed "as is" for full imports, since blocks may still be 1595 // queued for processing when the header download completes. However, as long as the 1596 // peer gave us something useful, we're already happy/progressed (above check). 1597 if mode == FastSync || mode == LightSync { 1598 head := d.lightchain.CurrentHeader() 1599 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1600 return errStallingPeer 1601 } 1602 } 1603 // Disable any rollback and return 1604 rollback = 0 1605 return nil 1606 } 1607 // Otherwise split the chunk of headers into batches and process them 1608 gotHeaders = true 1609 for len(headers) > 0 { 1610 // Terminate if something failed in between processing chunks 1611 select { 1612 case <-d.cancelCh: 1613 rollbackErr = errCanceled 1614 return errCanceled 1615 default: 1616 } 1617 // Select the next chunk of headers to import 1618 limit := maxHeadersProcess 1619 if limit > len(headers) { 1620 limit = len(headers) 1621 } 1622 chunk := headers[:limit] 1623 1624 // In case of header only syncing, validate the chunk immediately 1625 if mode == FastSync || mode == LightSync { 1626 // If we're importing pure headers, verify based on their recentness 1627 var pivot uint64 1628 1629 d.pivotLock.RLock() 1630 if d.pivotHeader != nil { 1631 pivot = d.pivotHeader.Number.Uint64() 1632 } 1633 d.pivotLock.RUnlock() 1634 1635 frequency := fsHeaderCheckFrequency 1636 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1637 frequency = 1 1638 } 1639 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1640 rollbackErr = err 1641 1642 // If some headers were inserted, track them as uncertain 1643 if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 { 1644 rollback = chunk[0].Number.Uint64() 1645 } 1646 log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err) 1647 return fmt.Errorf("%w: %v", errInvalidChain, err) 1648 } 1649 // All verifications passed, track all headers within the alloted limits 1650 if mode == FastSync { 1651 head := chunk[len(chunk)-1].Number.Uint64() 1652 if head-rollback > uint64(fsHeaderSafetyNet) { 1653 rollback = head - uint64(fsHeaderSafetyNet) 1654 } else { 1655 rollback = 1 1656 } 1657 } 1658 } 1659 // Unless we're doing light chains, schedule the headers for associated content retrieval 1660 if mode == FullSync || mode == FastSync { 1661 // If we've reached the allowed number of pending headers, stall a bit 1662 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1663 select { 1664 case <-d.cancelCh: 1665 rollbackErr = errCanceled 1666 return errCanceled 1667 case <-time.After(time.Second): 1668 } 1669 } 1670 // Otherwise insert the headers for content retrieval 1671 inserts := d.queue.Schedule(chunk, origin) 1672 if len(inserts) != len(chunk) { 1673 rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk)) 1674 return fmt.Errorf("%w: stale headers", errBadPeer) 1675 } 1676 } 1677 headers = headers[limit:] 1678 origin += uint64(limit) 1679 } 1680 // Update the highest block number we know if a higher one is found. 1681 d.syncStatsLock.Lock() 1682 if d.syncStatsChainHeight < origin { 1683 d.syncStatsChainHeight = origin - 1 1684 } 1685 d.syncStatsLock.Unlock() 1686 1687 // Signal the content downloaders of the availablility of new tasks 1688 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1689 select { 1690 case ch <- true: 1691 default: 1692 } 1693 } 1694 } 1695 } 1696 } 1697 1698 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1699 func (d *Downloader) processFullSyncContent() error { 1700 for { 1701 results := d.queue.Results(true) 1702 if len(results) == 0 { 1703 return nil 1704 } 1705 if d.chainInsertHook != nil { 1706 d.chainInsertHook(results) 1707 } 1708 if err := d.importBlockResults(results); err != nil { 1709 return err 1710 } 1711 } 1712 } 1713 1714 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1715 // Check for any early termination requests 1716 if len(results) == 0 { 1717 return nil 1718 } 1719 select { 1720 case <-d.quitCh: 1721 return errCancelContentProcessing 1722 default: 1723 } 1724 // Retrieve the a batch of results to import 1725 first, last := results[0].Header, results[len(results)-1].Header 1726 log.Debug("Inserting downloaded chain", "items", len(results), 1727 "firstnum", first.Number, "firsthash", first.Hash(), 1728 "lastnum", last.Number, "lasthash", last.Hash(), 1729 ) 1730 blocks := make([]*types.Block, len(results)) 1731 for i, result := range results { 1732 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1733 } 1734 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1735 if index < len(results) { 1736 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1737 } else { 1738 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1739 // when it needs to preprocess blocks to import a sidechain. 1740 // The importer will put together a new list of blocks to import, which is a superset 1741 // of the blocks delivered from the downloader, and the indexing will be off. 1742 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1743 } 1744 return fmt.Errorf("%w: %v", errInvalidChain, err) 1745 } 1746 return nil 1747 } 1748 1749 // processFastSyncContent takes fetch results from the queue and writes them to the 1750 // database. It also controls the synchronisation of state nodes of the pivot block. 1751 func (d *Downloader) processFastSyncContent() error { 1752 // Start syncing state of the reported head block. This should get us most of 1753 // the state of the pivot block. 1754 d.pivotLock.RLock() 1755 sync := d.syncState(d.pivotHeader.Root) 1756 d.pivotLock.RUnlock() 1757 1758 defer func() { 1759 // The `sync` object is replaced every time the pivot moves. We need to 1760 // defer close the very last active one, hence the lazy evaluation vs. 1761 // calling defer sync.Cancel() !!! 1762 sync.Cancel() 1763 }() 1764 1765 closeOnErr := func(s *stateSync) { 1766 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled && err != snap.ErrCancelled { 1767 d.queue.Close() // wake up Results 1768 } 1769 } 1770 go closeOnErr(sync) 1771 1772 // To cater for moving pivot points, track the pivot block and subsequently 1773 // accumulated download results separately. 1774 var ( 1775 oldPivot *fetchResult // Locked in pivot block, might change eventually 1776 oldTail []*fetchResult // Downloaded content after the pivot 1777 ) 1778 for { 1779 // Wait for the next batch of downloaded data to be available, and if the pivot 1780 // block became stale, move the goalpost 1781 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1782 if len(results) == 0 { 1783 // If pivot sync is done, stop 1784 if oldPivot == nil { 1785 return sync.Cancel() 1786 } 1787 // If sync failed, stop 1788 select { 1789 case <-d.cancelCh: 1790 sync.Cancel() 1791 return errCanceled 1792 default: 1793 } 1794 } 1795 if d.chainInsertHook != nil { 1796 d.chainInsertHook(results) 1797 } 1798 // If we haven't downloaded the pivot block yet, check pivot staleness 1799 // notifications from the header downloader 1800 d.pivotLock.RLock() 1801 pivot := d.pivotHeader 1802 d.pivotLock.RUnlock() 1803 1804 if oldPivot == nil { 1805 if pivot.Root != sync.root { 1806 sync.Cancel() 1807 sync = d.syncState(pivot.Root) 1808 1809 go closeOnErr(sync) 1810 } 1811 } else { 1812 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1813 } 1814 // Split around the pivot block and process the two sides via fast/full sync 1815 if atomic.LoadInt32(&d.committed) == 0 { 1816 latest := results[len(results)-1].Header 1817 // If the height is above the pivot block by 2 sets, it means the pivot 1818 // become stale in the network and it was garbage collected, move to a 1819 // new pivot. 1820 // 1821 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 1822 // need to be taken into account, otherwise we're detecting the pivot move 1823 // late and will drop peers due to unavailable state!!! 1824 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 1825 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 1826 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 1827 1828 d.pivotLock.Lock() 1829 d.pivotHeader = pivot 1830 d.pivotLock.Unlock() 1831 1832 // Write out the pivot into the database so a rollback beyond it will 1833 // reenable fast sync 1834 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 1835 } 1836 } 1837 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 1838 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1839 return err 1840 } 1841 if P != nil { 1842 // If new pivot block found, cancel old state retrieval and restart 1843 if oldPivot != P { 1844 sync.Cancel() 1845 sync = d.syncState(P.Header.Root) 1846 1847 go closeOnErr(sync) 1848 oldPivot = P 1849 } 1850 // Wait for completion, occasionally checking for pivot staleness 1851 select { 1852 case <-sync.done: 1853 if sync.err != nil { 1854 return sync.err 1855 } 1856 if err := d.commitPivotBlock(P); err != nil { 1857 return err 1858 } 1859 oldPivot = nil 1860 1861 case <-time.After(time.Second): 1862 oldTail = afterP 1863 continue 1864 } 1865 } 1866 // Fast sync done, pivot commit done, full import 1867 if err := d.importBlockResults(afterP); err != nil { 1868 return err 1869 } 1870 } 1871 } 1872 1873 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1874 if len(results) == 0 { 1875 return nil, nil, nil 1876 } 1877 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 1878 // the pivot is somewhere in the future 1879 return nil, results, nil 1880 } 1881 // This can also be optimized, but only happens very seldom 1882 for _, result := range results { 1883 num := result.Header.Number.Uint64() 1884 switch { 1885 case num < pivot: 1886 before = append(before, result) 1887 case num == pivot: 1888 p = result 1889 default: 1890 after = append(after, result) 1891 } 1892 } 1893 return p, before, after 1894 } 1895 1896 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1897 // Check for any early termination requests 1898 if len(results) == 0 { 1899 return nil 1900 } 1901 select { 1902 case <-d.quitCh: 1903 return errCancelContentProcessing 1904 case <-stateSync.done: 1905 if err := stateSync.Wait(); err != nil { 1906 return err 1907 } 1908 default: 1909 } 1910 // Retrieve the a batch of results to import 1911 first, last := results[0].Header, results[len(results)-1].Header 1912 log.Debug("Inserting fast-sync blocks", "items", len(results), 1913 "firstnum", first.Number, "firsthash", first.Hash(), 1914 "lastnumn", last.Number, "lasthash", last.Hash(), 1915 ) 1916 blocks := make([]*types.Block, len(results)) 1917 receipts := make([]types.Receipts, len(results)) 1918 for i, result := range results { 1919 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1920 receipts[i] = result.Receipts 1921 } 1922 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1923 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1924 return fmt.Errorf("%w: %v", errInvalidChain, err) 1925 } 1926 return nil 1927 } 1928 1929 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1930 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1931 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1932 1933 // Commit the pivot block as the new head, will require full sync from here on 1934 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1935 return err 1936 } 1937 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1938 return err 1939 } 1940 atomic.StoreInt32(&d.committed, 1) 1941 1942 // If we had a bloom filter for the state sync, deallocate it now. Note, we only 1943 // deallocate internally, but keep the empty wrapper. This ensures that if we do 1944 // a rollback after committing the pivot and restarting fast sync, we don't end 1945 // up using a nil bloom. Empty bloom is fine, it just returns that it does not 1946 // have the info we need, so reach down to the database instead. 1947 if d.stateBloom != nil { 1948 d.stateBloom.Close() 1949 } 1950 return nil 1951 } 1952 1953 // DeliverHeaders injects a new batch of block headers received from a remote 1954 // node into the download schedule. 1955 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) error { 1956 return d.deliver(d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1957 } 1958 1959 // DeliverBodies injects a new batch of block bodies received from a remote node. 1960 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) error { 1961 return d.deliver(d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1962 } 1963 1964 // DeliverReceipts injects a new batch of receipts received from a remote node. 1965 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) error { 1966 return d.deliver(d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1967 } 1968 1969 // DeliverNodeData injects a new batch of node state data received from a remote node. 1970 func (d *Downloader) DeliverNodeData(id string, data [][]byte) error { 1971 return d.deliver(d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1972 } 1973 1974 // DeliverSnapPacket is invoked from a peer's message handler when it transmits a 1975 // data packet for the local node to consume. 1976 func (d *Downloader) DeliverSnapPacket(peer *snap.Peer, packet snap.Packet) error { 1977 switch packet := packet.(type) { 1978 case *snap.AccountRangePacket: 1979 hashes, accounts, err := packet.Unpack() 1980 if err != nil { 1981 return err 1982 } 1983 return d.SnapSyncer.OnAccounts(peer, packet.ID, hashes, accounts, packet.Proof) 1984 1985 case *snap.StorageRangesPacket: 1986 hashset, slotset := packet.Unpack() 1987 return d.SnapSyncer.OnStorage(peer, packet.ID, hashset, slotset, packet.Proof) 1988 1989 case *snap.ByteCodesPacket: 1990 return d.SnapSyncer.OnByteCodes(peer, packet.ID, packet.Codes) 1991 1992 case *snap.TrieNodesPacket: 1993 return d.SnapSyncer.OnTrieNodes(peer, packet.ID, packet.Nodes) 1994 1995 default: 1996 return fmt.Errorf("unexpected snap packet type: %T", packet) 1997 } 1998 } 1999 2000 // deliver injects a new batch of data received from a remote node. 2001 func (d *Downloader) deliver(destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 2002 // Update the delivery metrics for both good and failed deliveries 2003 inMeter.Mark(int64(packet.Items())) 2004 defer func() { 2005 if err != nil { 2006 dropMeter.Mark(int64(packet.Items())) 2007 } 2008 }() 2009 // Deliver or abort if the sync is canceled while queuing 2010 d.cancelLock.RLock() 2011 cancel := d.cancelCh 2012 d.cancelLock.RUnlock() 2013 if cancel == nil { 2014 return errNoSyncActive 2015 } 2016 select { 2017 case destCh <- packet: 2018 return nil 2019 case <-cancel: 2020 return errNoSyncActive 2021 } 2022 } 2023 2024 // qosTuner is the quality of service tuning loop that occasionally gathers the 2025 // peer latency statistics and updates the estimated request round trip time. 2026 func (d *Downloader) qosTuner() { 2027 for { 2028 // Retrieve the current median RTT and integrate into the previoust target RTT 2029 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 2030 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 2031 2032 // A new RTT cycle passed, increase our confidence in the estimated RTT 2033 conf := atomic.LoadUint64(&d.rttConfidence) 2034 conf = conf + (1000000-conf)/2 2035 atomic.StoreUint64(&d.rttConfidence, conf) 2036 2037 // Log the new QoS values and sleep until the next RTT 2038 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 2039 select { 2040 case <-d.quitCh: 2041 return 2042 case <-time.After(rtt): 2043 } 2044 } 2045 } 2046 2047 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 2048 // peer set, needing to reduce the confidence we have in out QoS estimates. 2049 func (d *Downloader) qosReduceConfidence() { 2050 // If we have a single peer, confidence is always 1 2051 peers := uint64(d.peers.Len()) 2052 if peers == 0 { 2053 // Ensure peer connectivity races don't catch us off guard 2054 return 2055 } 2056 if peers == 1 { 2057 atomic.StoreUint64(&d.rttConfidence, 1000000) 2058 return 2059 } 2060 // If we have a ton of peers, don't drop confidence) 2061 if peers >= uint64(qosConfidenceCap) { 2062 return 2063 } 2064 // Otherwise drop the confidence factor 2065 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 2066 if float64(conf)/1000000 < rttMinConfidence { 2067 conf = uint64(rttMinConfidence * 1000000) 2068 } 2069 atomic.StoreUint64(&d.rttConfidence, conf) 2070 2071 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2072 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 2073 } 2074 2075 // requestRTT returns the current target round trip time for a download request 2076 // to complete in. 2077 // 2078 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 2079 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 2080 // be adapted to, but smaller ones are preferred (stabler download stream). 2081 func (d *Downloader) requestRTT() time.Duration { 2082 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 2083 } 2084 2085 // requestTTL returns the current timeout allowance for a single download request 2086 // to finish under. 2087 func (d *Downloader) requestTTL() time.Duration { 2088 var ( 2089 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2090 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 2091 ) 2092 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 2093 if ttl > ttlLimit { 2094 ttl = ttlLimit 2095 } 2096 return ttl 2097 }