github.com/phillinzzz/newBsc@v1.1.6/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/phillinzzz/newBsc" 29 "github.com/phillinzzz/newBsc/common" 30 "github.com/phillinzzz/newBsc/core/rawdb" 31 "github.com/phillinzzz/newBsc/core/state/snapshot" 32 "github.com/phillinzzz/newBsc/core/types" 33 "github.com/phillinzzz/newBsc/eth/protocols/eth" 34 "github.com/phillinzzz/newBsc/eth/protocols/snap" 35 "github.com/phillinzzz/newBsc/ethdb" 36 "github.com/phillinzzz/newBsc/event" 37 "github.com/phillinzzz/newBsc/log" 38 "github.com/phillinzzz/newBsc/metrics" 39 "github.com/phillinzzz/newBsc/params" 40 "github.com/phillinzzz/newBsc/trie" 41 ) 42 43 var ( 44 MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request 45 MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request 46 MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly 47 MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request 48 MaxStateFetch = 384 // Amount of node state values to allow fetching per request 49 50 rttMinEstimate = 2 * time.Second // Minimum round-trip time to target for download requests 51 rttMaxEstimate = 20 * time.Second // Maximum round-trip time to target for download requests 52 rttMinConfidence = 0.1 // Worse confidence factor in our estimated RTT value 53 ttlScaling = 3 // Constant scaling factor for RTT -> TTL conversion 54 ttlLimit = time.Minute // Maximum TTL allowance to prevent reaching crazy timeouts 55 56 qosTuningPeers = 5 // Number of peers to tune based on (best peers) 57 qosConfidenceCap = 10 // Number of peers above which not to modify RTT confidence 58 qosTuningImpact = 0.25 // Impact that a new tuning target has on the previous value 59 60 maxQueuedHeaders = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection) 61 maxHeadersProcess = 2048 // Number of header download results to import at once into the chain 62 maxResultsProcess = 2048 // Number of content download results to import at once into the chain 63 fullMaxForkAncestry uint64 = params.FullImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 64 lightMaxForkAncestry uint64 = params.LightImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 65 66 reorgProtThreshold = 48 // Threshold number of recent blocks to disable mini reorg protection 67 reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs 68 69 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 70 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 71 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 72 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 73 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 74 ) 75 76 var ( 77 errBusy = errors.New("busy") 78 errUnknownPeer = errors.New("peer is unknown or unhealthy") 79 errBadPeer = errors.New("action from bad peer ignored") 80 errStallingPeer = errors.New("peer is stalling") 81 errUnsyncedPeer = errors.New("unsynced peer") 82 errNoPeers = errors.New("no peers to keep download active") 83 errTimeout = errors.New("timeout") 84 errEmptyHeaderSet = errors.New("empty header set by peer") 85 errPeersUnavailable = errors.New("no peers available or all tried for download") 86 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 87 errInvalidChain = errors.New("retrieved hash chain is invalid") 88 errInvalidBody = errors.New("retrieved block body is invalid") 89 errInvalidReceipt = errors.New("retrieved receipt is invalid") 90 errCancelStateFetch = errors.New("state data download canceled (requested)") 91 errCancelContentProcessing = errors.New("content processing canceled (requested)") 92 errCanceled = errors.New("syncing canceled (requested)") 93 errNoSyncActive = errors.New("no sync active") 94 errTooOld = errors.New("peer's protocol version too old") 95 errNoAncestorFound = errors.New("no common ancestor found") 96 ) 97 98 type Downloader struct { 99 // WARNING: The `rttEstimate` and `rttConfidence` fields are accessed atomically. 100 // On 32 bit platforms, only 64-bit aligned fields can be atomic. The struct is 101 // guaranteed to be so aligned, so take advantage of that. For more information, 102 // see https://golang.org/pkg/sync/atomic/#pkg-note-BUG. 103 rttEstimate uint64 // Round trip time to target for download requests 104 rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops) 105 106 mode uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode 107 mux *event.TypeMux // Event multiplexer to announce sync operation events 108 109 checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync) 110 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 111 queue *queue // Scheduler for selecting the hashes to download 112 peers *peerSet // Set of active peers from which download can proceed 113 114 stateDB ethdb.Database // Database to state sync into (and deduplicate via) 115 stateBloom *trie.SyncBloom // Bloom filter for fast trie node and contract code existence checks 116 117 // Statistics 118 syncStatsChainOrigin uint64 // Origin block number where syncing started at 119 syncStatsChainHeight uint64 // Highest block number known when syncing started 120 syncStatsState stateSyncStats 121 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 122 123 lightchain LightChain 124 blockchain BlockChain 125 126 // Callbacks 127 dropPeer peerDropFn // Drops a peer for misbehaving 128 129 // Status 130 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 131 synchronising int32 132 notified int32 133 committed int32 134 ancientLimit uint64 // The maximum block number which can be regarded as ancient data. 135 136 // Channels 137 headerCh chan dataPack // Channel receiving inbound block headers 138 bodyCh chan dataPack // Channel receiving inbound block bodies 139 receiptCh chan dataPack // Channel receiving inbound receipts 140 bodyWakeCh chan bool // Channel to signal the block body fetcher of new tasks 141 receiptWakeCh chan bool // Channel to signal the receipt fetcher of new tasks 142 headerProcCh chan []*types.Header // Channel to feed the header processor new tasks 143 144 // State sync 145 pivotHeader *types.Header // Pivot block header to dynamically push the syncing state root 146 pivotLock sync.RWMutex // Lock protecting pivot header reads from updates 147 148 snapSync bool // Whether to run state sync over the snap protocol 149 SnapSyncer *snap.Syncer // TODO(karalabe): make private! hack for now 150 stateSyncStart chan *stateSync 151 trackStateReq chan *stateReq 152 stateCh chan dataPack // Channel receiving inbound node state data 153 154 // Cancellation and termination 155 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 156 cancelCh chan struct{} // Channel to cancel mid-flight syncs 157 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 158 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 159 160 quitCh chan struct{} // Quit channel to signal termination 161 quitLock sync.Mutex // Lock to prevent double closes 162 163 // Testing hooks 164 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 165 bodyFetchHook func([]*types.Header, ...interface{}) // Method to call upon starting a block body fetch 166 receiptFetchHook func([]*types.Header, ...interface{}) // Method to call upon starting a receipt fetch 167 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 168 } 169 170 // LightChain encapsulates functions required to synchronise a light chain. 171 type LightChain interface { 172 // HasHeader verifies a header's presence in the local chain. 173 HasHeader(common.Hash, uint64) bool 174 175 // GetHeaderByHash retrieves a header from the local chain. 176 GetHeaderByHash(common.Hash) *types.Header 177 178 // CurrentHeader retrieves the head header from the local chain. 179 CurrentHeader() *types.Header 180 181 // GetTd returns the total difficulty of a local block. 182 GetTd(common.Hash, uint64) *big.Int 183 184 // InsertHeaderChain inserts a batch of headers into the local chain. 185 InsertHeaderChain([]*types.Header, int) (int, error) 186 187 // SetHead rewinds the local chain to a new head. 188 SetHead(uint64) error 189 } 190 191 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 192 type BlockChain interface { 193 LightChain 194 195 // HasBlock verifies a block's presence in the local chain. 196 HasBlock(common.Hash, uint64) bool 197 198 // HasFastBlock verifies a fast block's presence in the local chain. 199 HasFastBlock(common.Hash, uint64) bool 200 201 // GetBlockByHash retrieves a block from the local chain. 202 GetBlockByHash(common.Hash) *types.Block 203 204 // CurrentBlock retrieves the head block from the local chain. 205 CurrentBlock() *types.Block 206 207 // CurrentFastBlock retrieves the head fast block from the local chain. 208 CurrentFastBlock() *types.Block 209 210 // FastSyncCommitHead directly commits the head block to a certain entity. 211 FastSyncCommitHead(common.Hash) error 212 213 // InsertChain inserts a batch of blocks into the local chain. 214 InsertChain(types.Blocks) (int, error) 215 216 // InsertReceiptChain inserts a batch of receipts into the local chain. 217 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 218 219 // Snapshots returns the blockchain snapshot tree to paused it during sync. 220 Snapshots() *snapshot.Tree 221 } 222 223 type DownloadOption func(downloader *Downloader) *Downloader 224 225 type IDiffPeer interface { 226 RequestDiffLayers([]common.Hash) error 227 } 228 229 type IPeerSet interface { 230 GetDiffPeer(string) IDiffPeer 231 } 232 233 func EnableDiffFetchOp(peers IPeerSet) DownloadOption { 234 return func(dl *Downloader) *Downloader { 235 var hook = func(headers []*types.Header, args ...interface{}) { 236 if len(args) < 2 { 237 return 238 } 239 peerID, ok := args[1].(string) 240 if !ok { 241 return 242 } 243 mode, ok := args[0].(SyncMode) 244 if !ok { 245 return 246 } 247 if ep := peers.GetDiffPeer(peerID); mode == FullSync && ep != nil { 248 hashes := make([]common.Hash, 0, len(headers)) 249 for _, header := range headers { 250 hashes = append(hashes, header.Hash()) 251 } 252 ep.RequestDiffLayers(hashes) 253 } 254 } 255 dl.bodyFetchHook = hook 256 return dl 257 } 258 } 259 260 // New creates a new downloader to fetch hashes and blocks from remote peers. 261 func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn, options ...DownloadOption) *Downloader { 262 if lightchain == nil { 263 lightchain = chain 264 } 265 dl := &Downloader{ 266 stateDB: stateDb, 267 stateBloom: stateBloom, 268 mux: mux, 269 checkpoint: checkpoint, 270 queue: newQueue(blockCacheMaxItems, blockCacheInitialItems), 271 peers: newPeerSet(), 272 rttEstimate: uint64(rttMaxEstimate), 273 rttConfidence: uint64(1000000), 274 blockchain: chain, 275 lightchain: lightchain, 276 dropPeer: dropPeer, 277 headerCh: make(chan dataPack, 1), 278 bodyCh: make(chan dataPack, 1), 279 receiptCh: make(chan dataPack, 1), 280 bodyWakeCh: make(chan bool, 1), 281 receiptWakeCh: make(chan bool, 1), 282 headerProcCh: make(chan []*types.Header, 1), 283 quitCh: make(chan struct{}), 284 stateCh: make(chan dataPack), 285 SnapSyncer: snap.NewSyncer(stateDb), 286 stateSyncStart: make(chan *stateSync), 287 syncStatsState: stateSyncStats{ 288 processed: rawdb.ReadFastTrieProgress(stateDb), 289 }, 290 trackStateReq: make(chan *stateReq), 291 } 292 for _, option := range options { 293 if dl != nil { 294 dl = option(dl) 295 } 296 } 297 go dl.qosTuner() 298 go dl.stateFetcher() 299 return dl 300 } 301 302 // Progress retrieves the synchronisation boundaries, specifically the origin 303 // block where synchronisation started at (may have failed/suspended); the block 304 // or header sync is currently at; and the latest known block which the sync targets. 305 // 306 // In addition, during the state download phase of fast synchronisation the number 307 // of processed and the total number of known states are also returned. Otherwise 308 // these are zero. 309 func (d *Downloader) Progress() ethereum.SyncProgress { 310 // Lock the current stats and return the progress 311 d.syncStatsLock.RLock() 312 defer d.syncStatsLock.RUnlock() 313 314 current := uint64(0) 315 mode := d.getMode() 316 switch { 317 case d.blockchain != nil && mode == FullSync: 318 current = d.blockchain.CurrentBlock().NumberU64() 319 case d.blockchain != nil && mode == FastSync: 320 current = d.blockchain.CurrentFastBlock().NumberU64() 321 case d.lightchain != nil: 322 current = d.lightchain.CurrentHeader().Number.Uint64() 323 default: 324 log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", mode) 325 } 326 return ethereum.SyncProgress{ 327 StartingBlock: d.syncStatsChainOrigin, 328 CurrentBlock: current, 329 HighestBlock: d.syncStatsChainHeight, 330 PulledStates: d.syncStatsState.processed, 331 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 332 } 333 } 334 335 // Synchronising returns whether the downloader is currently retrieving blocks. 336 func (d *Downloader) Synchronising() bool { 337 return atomic.LoadInt32(&d.synchronising) > 0 338 } 339 340 // RegisterPeer injects a new download peer into the set of block source to be 341 // used for fetching hashes and blocks from. 342 func (d *Downloader) RegisterPeer(id string, version uint, peer Peer) error { 343 var logger log.Logger 344 if len(id) < 16 { 345 // Tests use short IDs, don't choke on them 346 logger = log.New("peer", id) 347 } else { 348 logger = log.New("peer", id[:8]) 349 } 350 logger.Trace("Registering sync peer") 351 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 352 logger.Error("Failed to register sync peer", "err", err) 353 return err 354 } 355 d.qosReduceConfidence() 356 357 return nil 358 } 359 360 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 361 func (d *Downloader) RegisterLightPeer(id string, version uint, peer LightPeer) error { 362 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 363 } 364 365 // UnregisterPeer remove a peer from the known list, preventing any action from 366 // the specified peer. An effort is also made to return any pending fetches into 367 // the queue. 368 func (d *Downloader) UnregisterPeer(id string) error { 369 // Unregister the peer from the active peer set and revoke any fetch tasks 370 var logger log.Logger 371 if len(id) < 16 { 372 // Tests use short IDs, don't choke on them 373 logger = log.New("peer", id) 374 } else { 375 logger = log.New("peer", id[:8]) 376 } 377 logger.Trace("Unregistering sync peer") 378 if err := d.peers.Unregister(id); err != nil { 379 logger.Error("Failed to unregister sync peer", "err", err) 380 return err 381 } 382 d.queue.Revoke(id) 383 384 return nil 385 } 386 387 // Synchronise tries to sync up our local block chain with a remote peer, both 388 // adding various sanity checks as well as wrapping it with various log entries. 389 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 390 err := d.synchronise(id, head, td, mode) 391 392 switch err { 393 case nil, errBusy, errCanceled: 394 return err 395 } 396 if errors.Is(err, errInvalidChain) || errors.Is(err, errBadPeer) || errors.Is(err, errTimeout) || 397 errors.Is(err, errStallingPeer) || errors.Is(err, errUnsyncedPeer) || errors.Is(err, errEmptyHeaderSet) || 398 errors.Is(err, errPeersUnavailable) || errors.Is(err, errTooOld) || errors.Is(err, errInvalidAncestor) { 399 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 400 if d.dropPeer == nil { 401 // The dropPeer method is nil when `--copydb` is used for a local copy. 402 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 403 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 404 } else { 405 d.dropPeer(id) 406 } 407 return err 408 } 409 log.Warn("Synchronisation failed, retrying", "err", err) 410 return err 411 } 412 413 // synchronise will select the peer and use it for synchronising. If an empty string is given 414 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 415 // checks fail an error will be returned. This method is synchronous 416 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 417 // Mock out the synchronisation if testing 418 if d.synchroniseMock != nil { 419 return d.synchroniseMock(id, hash) 420 } 421 // Make sure only one goroutine is ever allowed past this point at once 422 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 423 return errBusy 424 } 425 defer atomic.StoreInt32(&d.synchronising, 0) 426 427 // Post a user notification of the sync (only once per session) 428 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 429 log.Info("Block synchronisation started") 430 } 431 // If we are already full syncing, but have a fast-sync bloom filter laying 432 // around, make sure it doesn't use memory any more. This is a special case 433 // when the user attempts to fast sync a new empty network. 434 if mode == FullSync && d.stateBloom != nil { 435 d.stateBloom.Close() 436 } 437 // If snap sync was requested, create the snap scheduler and switch to fast 438 // sync mode. Long term we could drop fast sync or merge the two together, 439 // but until snap becomes prevalent, we should support both. TODO(karalabe). 440 if mode == SnapSync { 441 if !d.snapSync { 442 // Snap sync uses the snapshot namespace to store potentially flakey data until 443 // sync completely heals and finishes. Pause snapshot maintenance in the mean 444 // time to prevent access. 445 if snapshots := d.blockchain.Snapshots(); snapshots != nil { // Only nil in tests 446 snapshots.Disable() 447 } 448 log.Warn("Enabling snapshot sync prototype") 449 d.snapSync = true 450 } 451 mode = FastSync 452 } 453 // Reset the queue, peer set and wake channels to clean any internal leftover state 454 d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems) 455 d.peers.Reset() 456 457 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 458 select { 459 case <-ch: 460 default: 461 } 462 } 463 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 464 for empty := false; !empty; { 465 select { 466 case <-ch: 467 default: 468 empty = true 469 } 470 } 471 } 472 for empty := false; !empty; { 473 select { 474 case <-d.headerProcCh: 475 default: 476 empty = true 477 } 478 } 479 // Create cancel channel for aborting mid-flight and mark the master peer 480 d.cancelLock.Lock() 481 d.cancelCh = make(chan struct{}) 482 d.cancelPeer = id 483 d.cancelLock.Unlock() 484 485 defer d.Cancel() // No matter what, we can't leave the cancel channel open 486 487 // Atomically set the requested sync mode 488 atomic.StoreUint32(&d.mode, uint32(mode)) 489 490 // Retrieve the origin peer and initiate the downloading process 491 p := d.peers.Peer(id) 492 if p == nil { 493 return errUnknownPeer 494 } 495 return d.syncWithPeer(p, hash, td) 496 } 497 498 func (d *Downloader) getMode() SyncMode { 499 return SyncMode(atomic.LoadUint32(&d.mode)) 500 } 501 502 // syncWithPeer starts a block synchronization based on the hash chain from the 503 // specified peer and head hash. 504 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 505 d.mux.Post(StartEvent{}) 506 defer func() { 507 // reset on error 508 if err != nil { 509 d.mux.Post(FailedEvent{err}) 510 } else { 511 latest := d.lightchain.CurrentHeader() 512 d.mux.Post(DoneEvent{latest}) 513 } 514 }() 515 if p.version < eth.ETH65 { 516 return fmt.Errorf("%w: advertized %d < required %d", errTooOld, p.version, eth.ETH65) 517 } 518 mode := d.getMode() 519 520 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", mode) 521 defer func(start time.Time) { 522 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 523 }(time.Now()) 524 525 // Look up the sync boundaries: the common ancestor and the target block 526 latest, pivot, err := d.fetchHead(p) 527 if err != nil { 528 return err 529 } 530 if mode == FastSync && pivot == nil { 531 // If no pivot block was returned, the head is below the min full block 532 // threshold (i.e. new chian). In that case we won't really fast sync 533 // anyway, but still need a valid pivot block to avoid some code hitting 534 // nil panics on an access. 535 pivot = d.blockchain.CurrentBlock().Header() 536 } 537 height := latest.Number.Uint64() 538 539 origin, err := d.findAncestor(p, latest) 540 if err != nil { 541 return err 542 } 543 d.syncStatsLock.Lock() 544 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 545 d.syncStatsChainOrigin = origin 546 } 547 d.syncStatsChainHeight = height 548 d.syncStatsLock.Unlock() 549 550 // Ensure our origin point is below any fast sync pivot point 551 if mode == FastSync { 552 if height <= uint64(fsMinFullBlocks) { 553 origin = 0 554 } else { 555 pivotNumber := pivot.Number.Uint64() 556 if pivotNumber <= origin { 557 origin = pivotNumber - 1 558 } 559 // Write out the pivot into the database so a rollback beyond it will 560 // reenable fast sync 561 rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber) 562 } 563 } 564 d.committed = 1 565 if mode == FastSync && pivot.Number.Uint64() != 0 { 566 d.committed = 0 567 } 568 if mode == FastSync { 569 // Set the ancient data limitation. 570 // If we are running fast sync, all block data older than ancientLimit will be 571 // written to the ancient store. More recent data will be written to the active 572 // database and will wait for the freezer to migrate. 573 // 574 // If there is a checkpoint available, then calculate the ancientLimit through 575 // that. Otherwise calculate the ancient limit through the advertised height 576 // of the remote peer. 577 // 578 // The reason for picking checkpoint first is that a malicious peer can give us 579 // a fake (very high) height, forcing the ancient limit to also be very high. 580 // The peer would start to feed us valid blocks until head, resulting in all of 581 // the blocks might be written into the ancient store. A following mini-reorg 582 // could cause issues. 583 if d.checkpoint != 0 && d.checkpoint > fullMaxForkAncestry+1 { 584 d.ancientLimit = d.checkpoint 585 } else if height > fullMaxForkAncestry+1 { 586 d.ancientLimit = height - fullMaxForkAncestry - 1 587 } else { 588 d.ancientLimit = 0 589 } 590 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 591 592 // If a part of blockchain data has already been written into active store, 593 // disable the ancient style insertion explicitly. 594 if origin >= frozen && frozen != 0 { 595 d.ancientLimit = 0 596 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 597 } else if d.ancientLimit > 0 { 598 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 599 } 600 // Rewind the ancient store and blockchain if reorg happens. 601 if origin+1 < frozen { 602 if err := d.lightchain.SetHead(origin + 1); err != nil { 603 return err 604 } 605 } 606 } 607 // Initiate the sync using a concurrent header and content retrieval algorithm 608 d.queue.Prepare(origin+1, mode) 609 if d.syncInitHook != nil { 610 d.syncInitHook(origin, height) 611 } 612 fetchers := []func() error{ 613 func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved 614 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 615 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 616 func() error { return d.processHeaders(origin+1, td) }, 617 } 618 if mode == FastSync { 619 d.pivotLock.Lock() 620 d.pivotHeader = pivot 621 d.pivotLock.Unlock() 622 623 fetchers = append(fetchers, func() error { return d.processFastSyncContent() }) 624 } else if mode == FullSync { 625 fetchers = append(fetchers, d.processFullSyncContent) 626 } 627 return d.spawnSync(fetchers) 628 } 629 630 // spawnSync runs d.process and all given fetcher functions to completion in 631 // separate goroutines, returning the first error that appears. 632 func (d *Downloader) spawnSync(fetchers []func() error) error { 633 errc := make(chan error, len(fetchers)) 634 d.cancelWg.Add(len(fetchers)) 635 for _, fn := range fetchers { 636 fn := fn 637 go func() { defer d.cancelWg.Done(); errc <- fn() }() 638 } 639 // Wait for the first error, then terminate the others. 640 var err error 641 for i := 0; i < len(fetchers); i++ { 642 if i == len(fetchers)-1 { 643 // Close the queue when all fetchers have exited. 644 // This will cause the block processor to end when 645 // it has processed the queue. 646 d.queue.Close() 647 } 648 if err = <-errc; err != nil && err != errCanceled { 649 break 650 } 651 } 652 d.queue.Close() 653 d.Cancel() 654 return err 655 } 656 657 // cancel aborts all of the operations and resets the queue. However, cancel does 658 // not wait for the running download goroutines to finish. This method should be 659 // used when cancelling the downloads from inside the downloader. 660 func (d *Downloader) cancel() { 661 // Close the current cancel channel 662 d.cancelLock.Lock() 663 defer d.cancelLock.Unlock() 664 665 if d.cancelCh != nil { 666 select { 667 case <-d.cancelCh: 668 // Channel was already closed 669 default: 670 close(d.cancelCh) 671 } 672 } 673 } 674 675 // Cancel aborts all of the operations and waits for all download goroutines to 676 // finish before returning. 677 func (d *Downloader) Cancel() { 678 d.cancel() 679 d.cancelWg.Wait() 680 } 681 682 // Terminate interrupts the downloader, canceling all pending operations. 683 // The downloader cannot be reused after calling Terminate. 684 func (d *Downloader) Terminate() { 685 // Close the termination channel (make sure double close is allowed) 686 d.quitLock.Lock() 687 select { 688 case <-d.quitCh: 689 default: 690 close(d.quitCh) 691 } 692 if d.stateBloom != nil { 693 d.stateBloom.Close() 694 } 695 d.quitLock.Unlock() 696 697 // Cancel any pending download requests 698 d.Cancel() 699 } 700 701 // fetchHead retrieves the head header and prior pivot block (if available) from 702 // a remote peer. 703 func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *types.Header, err error) { 704 p.log.Debug("Retrieving remote chain head") 705 mode := d.getMode() 706 707 // Request the advertised remote head block and wait for the response 708 latest, _ := p.peer.Head() 709 fetch := 1 710 if mode == FastSync { 711 fetch = 2 // head + pivot headers 712 } 713 go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true) 714 715 ttl := d.requestTTL() 716 timeout := time.After(ttl) 717 for { 718 select { 719 case <-d.cancelCh: 720 return nil, nil, errCanceled 721 722 case packet := <-d.headerCh: 723 // Discard anything not from the origin peer 724 if packet.PeerId() != p.id { 725 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 726 break 727 } 728 // Make sure the peer gave us at least one and at most the requested headers 729 headers := packet.(*headerPack).headers 730 if len(headers) == 0 || len(headers) > fetch { 731 return nil, nil, fmt.Errorf("%w: returned headers %d != requested %d", errBadPeer, len(headers), fetch) 732 } 733 // The first header needs to be the head, validate against the checkpoint 734 // and request. If only 1 header was returned, make sure there's no pivot 735 // or there was not one requested. 736 head := headers[0] 737 if (mode == FastSync || mode == LightSync) && head.Number.Uint64() < d.checkpoint { 738 return nil, nil, fmt.Errorf("%w: remote head %d below checkpoint %d", errUnsyncedPeer, head.Number, d.checkpoint) 739 } 740 if len(headers) == 1 { 741 if mode == FastSync && head.Number.Uint64() > uint64(fsMinFullBlocks) { 742 return nil, nil, fmt.Errorf("%w: no pivot included along head header", errBadPeer) 743 } 744 p.log.Debug("Remote head identified, no pivot", "number", head.Number, "hash", head.Hash()) 745 return head, nil, nil 746 } 747 // At this point we have 2 headers in total and the first is the 748 // validated head of the chian. Check the pivot number and return, 749 pivot := headers[1] 750 if pivot.Number.Uint64() != head.Number.Uint64()-uint64(fsMinFullBlocks) { 751 return nil, nil, fmt.Errorf("%w: remote pivot %d != requested %d", errInvalidChain, pivot.Number, head.Number.Uint64()-uint64(fsMinFullBlocks)) 752 } 753 return head, pivot, nil 754 755 case <-timeout: 756 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 757 return nil, nil, errTimeout 758 759 case <-d.bodyCh: 760 case <-d.receiptCh: 761 // Out of bounds delivery, ignore 762 } 763 } 764 } 765 766 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 767 // common ancestor. 768 // It returns parameters to be used for peer.RequestHeadersByNumber: 769 // from - starting block number 770 // count - number of headers to request 771 // skip - number of headers to skip 772 // and also returns 'max', the last block which is expected to be returned by the remote peers, 773 // given the (from,count,skip) 774 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 775 var ( 776 from int 777 count int 778 MaxCount = MaxHeaderFetch / 16 779 ) 780 // requestHead is the highest block that we will ask for. If requestHead is not offset, 781 // the highest block that we will get is 16 blocks back from head, which means we 782 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 783 // between us and the peer is 1-2 blocks, which is most common 784 requestHead := int(remoteHeight) - 1 785 if requestHead < 0 { 786 requestHead = 0 787 } 788 // requestBottom is the lowest block we want included in the query 789 // Ideally, we want to include the one just below our own head 790 requestBottom := int(localHeight - 1) 791 if requestBottom < 0 { 792 requestBottom = 0 793 } 794 totalSpan := requestHead - requestBottom 795 span := 1 + totalSpan/MaxCount 796 if span < 2 { 797 span = 2 798 } 799 if span > 16 { 800 span = 16 801 } 802 803 count = 1 + totalSpan/span 804 if count > MaxCount { 805 count = MaxCount 806 } 807 if count < 2 { 808 count = 2 809 } 810 from = requestHead - (count-1)*span 811 if from < 0 { 812 from = 0 813 } 814 max := from + (count-1)*span 815 return int64(from), count, span - 1, uint64(max) 816 } 817 818 // findAncestor tries to locate the common ancestor link of the local chain and 819 // a remote peers blockchain. In the general case when our node was in sync and 820 // on the correct chain, checking the top N links should already get us a match. 821 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 822 // the head links match), we do a binary search to find the common ancestor. 823 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 824 // Figure out the valid ancestor range to prevent rewrite attacks 825 var ( 826 floor = int64(-1) 827 localHeight uint64 828 remoteHeight = remoteHeader.Number.Uint64() 829 ) 830 mode := d.getMode() 831 switch mode { 832 case FullSync: 833 localHeight = d.blockchain.CurrentBlock().NumberU64() 834 case FastSync: 835 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 836 default: 837 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 838 } 839 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 840 841 // Recap floor value for binary search 842 maxForkAncestry := fullMaxForkAncestry 843 if d.getMode() == LightSync { 844 maxForkAncestry = lightMaxForkAncestry 845 } 846 if localHeight >= maxForkAncestry { 847 // We're above the max reorg threshold, find the earliest fork point 848 floor = int64(localHeight - maxForkAncestry) 849 } 850 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 851 // all headers before that point will be missing. 852 if mode == LightSync { 853 // If we don't know the current CHT position, find it 854 if d.genesis == 0 { 855 header := d.lightchain.CurrentHeader() 856 for header != nil { 857 d.genesis = header.Number.Uint64() 858 if floor >= int64(d.genesis)-1 { 859 break 860 } 861 header = d.lightchain.GetHeaderByHash(header.ParentHash) 862 } 863 } 864 // We already know the "genesis" block number, cap floor to that 865 if floor < int64(d.genesis)-1 { 866 floor = int64(d.genesis) - 1 867 } 868 } 869 870 ancestor, err := d.findAncestorSpanSearch(p, mode, remoteHeight, localHeight, floor) 871 if err == nil { 872 return ancestor, nil 873 } 874 // The returned error was not nil. 875 // If the error returned does not reflect that a common ancestor was not found, return it. 876 // If the error reflects that a common ancestor was not found, continue to binary search, 877 // where the error value will be reassigned. 878 if !errors.Is(err, errNoAncestorFound) { 879 return 0, err 880 } 881 882 ancestor, err = d.findAncestorBinarySearch(p, mode, remoteHeight, floor) 883 if err != nil { 884 return 0, err 885 } 886 return ancestor, nil 887 } 888 889 func (d *Downloader) findAncestorSpanSearch(p *peerConnection, mode SyncMode, remoteHeight, localHeight uint64, floor int64) (commonAncestor uint64, err error) { 890 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 891 892 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 893 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 894 895 // Wait for the remote response to the head fetch 896 number, hash := uint64(0), common.Hash{} 897 898 ttl := d.requestTTL() 899 timeout := time.After(ttl) 900 901 for finished := false; !finished; { 902 select { 903 case <-d.cancelCh: 904 return 0, errCanceled 905 906 case packet := <-d.headerCh: 907 // Discard anything not from the origin peer 908 if packet.PeerId() != p.id { 909 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 910 break 911 } 912 // Make sure the peer actually gave something valid 913 headers := packet.(*headerPack).headers 914 if len(headers) == 0 { 915 p.log.Warn("Empty head header set") 916 return 0, errEmptyHeaderSet 917 } 918 // Make sure the peer's reply conforms to the request 919 for i, header := range headers { 920 expectNumber := from + int64(i)*int64(skip+1) 921 if number := header.Number.Int64(); number != expectNumber { 922 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 923 return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering")) 924 } 925 } 926 // Check if a common ancestor was found 927 finished = true 928 for i := len(headers) - 1; i >= 0; i-- { 929 // Skip any headers that underflow/overflow our requested set 930 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 931 continue 932 } 933 // Otherwise check if we already know the header or not 934 h := headers[i].Hash() 935 n := headers[i].Number.Uint64() 936 937 var known bool 938 switch mode { 939 case FullSync: 940 known = d.blockchain.HasBlock(h, n) 941 case FastSync: 942 known = d.blockchain.HasFastBlock(h, n) 943 default: 944 known = d.lightchain.HasHeader(h, n) 945 } 946 if known { 947 number, hash = n, h 948 break 949 } 950 } 951 952 case <-timeout: 953 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 954 return 0, errTimeout 955 956 case <-d.bodyCh: 957 case <-d.receiptCh: 958 // Out of bounds delivery, ignore 959 } 960 } 961 // If the head fetch already found an ancestor, return 962 if hash != (common.Hash{}) { 963 if int64(number) <= floor { 964 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 965 return 0, errInvalidAncestor 966 } 967 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 968 return number, nil 969 } 970 return 0, errNoAncestorFound 971 } 972 973 func (d *Downloader) findAncestorBinarySearch(p *peerConnection, mode SyncMode, remoteHeight uint64, floor int64) (commonAncestor uint64, err error) { 974 hash := common.Hash{} 975 976 // Ancestor not found, we need to binary search over our chain 977 start, end := uint64(0), remoteHeight 978 if floor > 0 { 979 start = uint64(floor) 980 } 981 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 982 983 for start+1 < end { 984 // Split our chain interval in two, and request the hash to cross check 985 check := (start + end) / 2 986 987 ttl := d.requestTTL() 988 timeout := time.After(ttl) 989 990 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 991 992 // Wait until a reply arrives to this request 993 for arrived := false; !arrived; { 994 select { 995 case <-d.cancelCh: 996 return 0, errCanceled 997 998 case packet := <-d.headerCh: 999 // Discard anything not from the origin peer 1000 if packet.PeerId() != p.id { 1001 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 1002 break 1003 } 1004 // Make sure the peer actually gave something valid 1005 headers := packet.(*headerPack).headers 1006 if len(headers) != 1 { 1007 p.log.Warn("Multiple headers for single request", "headers", len(headers)) 1008 return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers)) 1009 } 1010 arrived = true 1011 1012 // Modify the search interval based on the response 1013 h := headers[0].Hash() 1014 n := headers[0].Number.Uint64() 1015 1016 var known bool 1017 switch mode { 1018 case FullSync: 1019 known = d.blockchain.HasBlock(h, n) 1020 case FastSync: 1021 known = d.blockchain.HasFastBlock(h, n) 1022 default: 1023 known = d.lightchain.HasHeader(h, n) 1024 } 1025 if !known { 1026 end = check 1027 break 1028 } 1029 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 1030 if header == nil { 1031 p.log.Error("header not found", "hash", h, "request", check) 1032 return 0, fmt.Errorf("%w: header no found (%s)", errBadPeer, h) 1033 } 1034 if header.Number.Uint64() != check { 1035 p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 1036 return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number) 1037 } 1038 start = check 1039 hash = h 1040 1041 case <-timeout: 1042 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 1043 return 0, errTimeout 1044 1045 case <-d.bodyCh: 1046 case <-d.receiptCh: 1047 // Out of bounds delivery, ignore 1048 } 1049 } 1050 } 1051 // Ensure valid ancestry and return 1052 if int64(start) <= floor { 1053 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 1054 return 0, errInvalidAncestor 1055 } 1056 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 1057 return start, nil 1058 } 1059 1060 // fetchHeaders keeps retrieving headers concurrently from the number 1061 // requested, until no more are returned, potentially throttling on the way. To 1062 // facilitate concurrency but still protect against malicious nodes sending bad 1063 // headers, we construct a header chain skeleton using the "origin" peer we are 1064 // syncing with, and fill in the missing headers using anyone else. Headers from 1065 // other peers are only accepted if they map cleanly to the skeleton. If no one 1066 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 1067 // the origin is dropped. 1068 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error { 1069 p.log.Debug("Directing header downloads", "origin", from) 1070 defer p.log.Debug("Header download terminated") 1071 1072 // Create a timeout timer, and the associated header fetcher 1073 skeleton := true // Skeleton assembly phase or finishing up 1074 pivoting := false // Whether the next request is pivot verification 1075 request := time.Now() // time of the last skeleton fetch request 1076 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 1077 <-timeout.C // timeout channel should be initially empty 1078 defer timeout.Stop() 1079 1080 var ttl time.Duration 1081 getHeaders := func(from uint64) { 1082 request = time.Now() 1083 1084 ttl = d.requestTTL() 1085 timeout.Reset(ttl) 1086 1087 if skeleton { 1088 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 1089 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 1090 } else { 1091 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 1092 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 1093 } 1094 } 1095 getNextPivot := func() { 1096 pivoting = true 1097 request = time.Now() 1098 1099 ttl = d.requestTTL() 1100 timeout.Reset(ttl) 1101 1102 d.pivotLock.RLock() 1103 pivot := d.pivotHeader.Number.Uint64() 1104 d.pivotLock.RUnlock() 1105 1106 p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks)) 1107 go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep 1108 } 1109 // Start pulling the header chain skeleton until all is done 1110 ancestor := from 1111 getHeaders(from) 1112 1113 mode := d.getMode() 1114 for { 1115 select { 1116 case <-d.cancelCh: 1117 return errCanceled 1118 1119 case packet := <-d.headerCh: 1120 // Make sure the active peer is giving us the skeleton headers 1121 if packet.PeerId() != p.id { 1122 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 1123 break 1124 } 1125 headerReqTimer.UpdateSince(request) 1126 timeout.Stop() 1127 1128 // If the pivot is being checked, move if it became stale and run the real retrieval 1129 var pivot uint64 1130 1131 d.pivotLock.RLock() 1132 if d.pivotHeader != nil { 1133 pivot = d.pivotHeader.Number.Uint64() 1134 } 1135 d.pivotLock.RUnlock() 1136 1137 if pivoting { 1138 if packet.Items() == 2 { 1139 // Retrieve the headers and do some sanity checks, just in case 1140 headers := packet.(*headerPack).headers 1141 1142 if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want { 1143 log.Warn("Peer sent invalid next pivot", "have", have, "want", want) 1144 return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want) 1145 } 1146 if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want { 1147 log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want) 1148 return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want) 1149 } 1150 log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number) 1151 pivot = headers[0].Number.Uint64() 1152 1153 d.pivotLock.Lock() 1154 d.pivotHeader = headers[0] 1155 d.pivotLock.Unlock() 1156 1157 // Write out the pivot into the database so a rollback beyond 1158 // it will reenable fast sync and update the state root that 1159 // the state syncer will be downloading. 1160 rawdb.WriteLastPivotNumber(d.stateDB, pivot) 1161 } 1162 pivoting = false 1163 getHeaders(from) 1164 continue 1165 } 1166 // If the skeleton's finished, pull any remaining head headers directly from the origin 1167 if skeleton && packet.Items() == 0 { 1168 skeleton = false 1169 getHeaders(from) 1170 continue 1171 } 1172 // If no more headers are inbound, notify the content fetchers and return 1173 if packet.Items() == 0 { 1174 // Don't abort header fetches while the pivot is downloading 1175 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 1176 p.log.Debug("No headers, waiting for pivot commit") 1177 select { 1178 case <-time.After(fsHeaderContCheck): 1179 getHeaders(from) 1180 continue 1181 case <-d.cancelCh: 1182 return errCanceled 1183 } 1184 } 1185 // Pivot done (or not in fast sync) and no more headers, terminate the process 1186 p.log.Debug("No more headers available") 1187 select { 1188 case d.headerProcCh <- nil: 1189 return nil 1190 case <-d.cancelCh: 1191 return errCanceled 1192 } 1193 } 1194 headers := packet.(*headerPack).headers 1195 1196 // If we received a skeleton batch, resolve internals concurrently 1197 if skeleton { 1198 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1199 if err != nil { 1200 p.log.Debug("Skeleton chain invalid", "err", err) 1201 return fmt.Errorf("%w: %v", errInvalidChain, err) 1202 } 1203 headers = filled[proced:] 1204 from += uint64(proced) 1205 } else { 1206 // If we're closing in on the chain head, but haven't yet reached it, delay 1207 // the last few headers so mini reorgs on the head don't cause invalid hash 1208 // chain errors. 1209 if n := len(headers); n > 0 { 1210 // Retrieve the current head we're at 1211 var head uint64 1212 if mode == LightSync { 1213 head = d.lightchain.CurrentHeader().Number.Uint64() 1214 } else { 1215 head = d.blockchain.CurrentFastBlock().NumberU64() 1216 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1217 head = full 1218 } 1219 } 1220 // If the head is below the common ancestor, we're actually deduplicating 1221 // already existing chain segments, so use the ancestor as the fake head. 1222 // Otherwise we might end up delaying header deliveries pointlessly. 1223 if head < ancestor { 1224 head = ancestor 1225 } 1226 // If the head is way older than this batch, delay the last few headers 1227 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1228 delay := reorgProtHeaderDelay 1229 if delay > n { 1230 delay = n 1231 } 1232 headers = headers[:n-delay] 1233 } 1234 } 1235 } 1236 // Insert all the new headers and fetch the next batch 1237 if len(headers) > 0 { 1238 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1239 select { 1240 case d.headerProcCh <- headers: 1241 case <-d.cancelCh: 1242 return errCanceled 1243 } 1244 from += uint64(len(headers)) 1245 1246 // If we're still skeleton filling fast sync, check pivot staleness 1247 // before continuing to the next skeleton filling 1248 if skeleton && pivot > 0 { 1249 getNextPivot() 1250 } else { 1251 getHeaders(from) 1252 } 1253 } else { 1254 // No headers delivered, or all of them being delayed, sleep a bit and retry 1255 p.log.Trace("All headers delayed, waiting") 1256 select { 1257 case <-time.After(fsHeaderContCheck): 1258 getHeaders(from) 1259 continue 1260 case <-d.cancelCh: 1261 return errCanceled 1262 } 1263 } 1264 1265 case <-timeout.C: 1266 if d.dropPeer == nil { 1267 // The dropPeer method is nil when `--copydb` is used for a local copy. 1268 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1269 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1270 break 1271 } 1272 // Header retrieval timed out, consider the peer bad and drop 1273 p.log.Debug("Header request timed out", "elapsed", ttl) 1274 headerTimeoutMeter.Mark(1) 1275 d.dropPeer(p.id) 1276 1277 // Finish the sync gracefully instead of dumping the gathered data though 1278 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1279 select { 1280 case ch <- false: 1281 case <-d.cancelCh: 1282 } 1283 } 1284 select { 1285 case d.headerProcCh <- nil: 1286 case <-d.cancelCh: 1287 } 1288 return fmt.Errorf("%w: header request timed out", errBadPeer) 1289 } 1290 } 1291 } 1292 1293 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1294 // and maps them to the provided skeleton header chain. 1295 // 1296 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1297 // immediately to the header processor to keep the rest of the pipeline full even 1298 // in the case of header stalls. 1299 // 1300 // The method returns the entire filled skeleton and also the number of headers 1301 // already forwarded for processing. 1302 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1303 log.Debug("Filling up skeleton", "from", from) 1304 d.queue.ScheduleSkeleton(from, skeleton) 1305 1306 var ( 1307 deliver = func(packet dataPack) (int, error) { 1308 pack := packet.(*headerPack) 1309 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1310 } 1311 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 1312 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) { 1313 return d.queue.ReserveHeaders(p, count), false, false 1314 } 1315 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1316 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 1317 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1318 p.SetHeadersIdle(accepted, deliveryTime) 1319 } 1320 ) 1321 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1322 d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve, 1323 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1324 1325 log.Debug("Skeleton fill terminated", "err", err) 1326 1327 filled, proced := d.queue.RetrieveHeaders() 1328 return filled, proced, err 1329 } 1330 1331 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1332 // available peers, reserving a chunk of blocks for each, waiting for delivery 1333 // and also periodically checking for timeouts. 1334 func (d *Downloader) fetchBodies(from uint64) error { 1335 log.Debug("Downloading block bodies", "origin", from) 1336 1337 var ( 1338 deliver = func(packet dataPack) (int, error) { 1339 pack := packet.(*bodyPack) 1340 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1341 } 1342 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 1343 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1344 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 1345 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) } 1346 ) 1347 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1348 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies, 1349 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1350 1351 log.Debug("Block body download terminated", "err", err) 1352 return err 1353 } 1354 1355 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1356 // available peers, reserving a chunk of receipts for each, waiting for delivery 1357 // and also periodically checking for timeouts. 1358 func (d *Downloader) fetchReceipts(from uint64) error { 1359 log.Debug("Downloading transaction receipts", "origin", from) 1360 1361 var ( 1362 deliver = func(packet dataPack) (int, error) { 1363 pack := packet.(*receiptPack) 1364 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1365 } 1366 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1367 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1368 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1369 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1370 p.SetReceiptsIdle(accepted, deliveryTime) 1371 } 1372 ) 1373 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1374 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts, 1375 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1376 1377 log.Debug("Transaction receipt download terminated", "err", err) 1378 return err 1379 } 1380 1381 // fetchParts iteratively downloads scheduled block parts, taking any available 1382 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1383 // also periodically checking for timeouts. 1384 // 1385 // As the scheduling/timeout logic mostly is the same for all downloaded data 1386 // types, this method is used by each for data gathering and is instrumented with 1387 // various callbacks to handle the slight differences between processing them. 1388 // 1389 // The instrumentation parameters: 1390 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1391 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1392 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1393 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1394 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1395 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1396 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1397 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1398 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1399 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1400 // - fetch: network callback to actually send a particular download request to a physical remote peer 1401 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1402 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1403 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1404 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1405 // - kind: textual label of the type being downloaded to display in log messages 1406 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1407 expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool), 1408 fetchHook func([]*types.Header, ...interface{}), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1409 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error { 1410 1411 // Create a ticker to detect expired retrieval tasks 1412 ticker := time.NewTicker(100 * time.Millisecond) 1413 defer ticker.Stop() 1414 1415 update := make(chan struct{}, 1) 1416 1417 // Prepare the queue and fetch block parts until the block header fetcher's done 1418 finished := false 1419 for { 1420 select { 1421 case <-d.cancelCh: 1422 return errCanceled 1423 1424 case packet := <-deliveryCh: 1425 deliveryTime := time.Now() 1426 // If the peer was previously banned and failed to deliver its pack 1427 // in a reasonable time frame, ignore its message. 1428 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1429 // Deliver the received chunk of data and check chain validity 1430 accepted, err := deliver(packet) 1431 if errors.Is(err, errInvalidChain) { 1432 return err 1433 } 1434 // Unless a peer delivered something completely else than requested (usually 1435 // caused by a timed out request which came through in the end), set it to 1436 // idle. If the delivery's stale, the peer should have already been idled. 1437 if !errors.Is(err, errStaleDelivery) { 1438 setIdle(peer, accepted, deliveryTime) 1439 } 1440 // Issue a log to the user to see what's going on 1441 switch { 1442 case err == nil && packet.Items() == 0: 1443 peer.log.Trace("Requested data not delivered", "type", kind) 1444 case err == nil: 1445 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1446 default: 1447 peer.log.Debug("Failed to deliver retrieved data", "type", kind, "err", err) 1448 } 1449 } 1450 // Blocks assembled, try to update the progress 1451 select { 1452 case update <- struct{}{}: 1453 default: 1454 } 1455 1456 case cont := <-wakeCh: 1457 // The header fetcher sent a continuation flag, check if it's done 1458 if !cont { 1459 finished = true 1460 } 1461 // Headers arrive, try to update the progress 1462 select { 1463 case update <- struct{}{}: 1464 default: 1465 } 1466 1467 case <-ticker.C: 1468 // Sanity check update the progress 1469 select { 1470 case update <- struct{}{}: 1471 default: 1472 } 1473 1474 case <-update: 1475 // Short circuit if we lost all our peers 1476 if d.peers.Len() == 0 { 1477 return errNoPeers 1478 } 1479 // Check for fetch request timeouts and demote the responsible peers 1480 for pid, fails := range expire() { 1481 if peer := d.peers.Peer(pid); peer != nil { 1482 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1483 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1484 // out that sync wise we need to get rid of the peer. 1485 // 1486 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1487 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1488 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1489 if fails > 2 { 1490 peer.log.Trace("Data delivery timed out", "type", kind) 1491 setIdle(peer, 0, time.Now()) 1492 } else { 1493 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1494 1495 if d.dropPeer == nil { 1496 // The dropPeer method is nil when `--copydb` is used for a local copy. 1497 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1498 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1499 } else { 1500 d.dropPeer(pid) 1501 1502 // If this peer was the master peer, abort sync immediately 1503 d.cancelLock.RLock() 1504 master := pid == d.cancelPeer 1505 d.cancelLock.RUnlock() 1506 1507 if master { 1508 d.cancel() 1509 return errTimeout 1510 } 1511 } 1512 } 1513 } 1514 } 1515 // If there's nothing more to fetch, wait or terminate 1516 if pending() == 0 { 1517 if !inFlight() && finished { 1518 log.Debug("Data fetching completed", "type", kind) 1519 return nil 1520 } 1521 break 1522 } 1523 // Send a download request to all idle peers, until throttled 1524 progressed, throttled, running := false, false, inFlight() 1525 idles, total := idle() 1526 pendCount := pending() 1527 for _, peer := range idles { 1528 // Short circuit if throttling activated 1529 if throttled { 1530 break 1531 } 1532 // Short circuit if there is no more available task. 1533 if pendCount = pending(); pendCount == 0 { 1534 break 1535 } 1536 // Reserve a chunk of fetches for a peer. A nil can mean either that 1537 // no more headers are available, or that the peer is known not to 1538 // have them. 1539 request, progress, throttle := reserve(peer, capacity(peer)) 1540 if progress { 1541 progressed = true 1542 } 1543 if throttle { 1544 throttled = true 1545 throttleCounter.Inc(1) 1546 } 1547 if request == nil { 1548 continue 1549 } 1550 if request.From > 0 { 1551 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1552 } else { 1553 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1554 } 1555 // Fetch the chunk and make sure any errors return the hashes to the queue 1556 if fetchHook != nil { 1557 fetchHook(request.Headers, d.getMode(), peer.id) 1558 } 1559 if err := fetch(peer, request); err != nil { 1560 // Although we could try and make an attempt to fix this, this error really 1561 // means that we've double allocated a fetch task to a peer. If that is the 1562 // case, the internal state of the downloader and the queue is very wrong so 1563 // better hard crash and note the error instead of silently accumulating into 1564 // a much bigger issue. 1565 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1566 } 1567 running = true 1568 } 1569 // Make sure that we have peers available for fetching. If all peers have been tried 1570 // and all failed throw an error 1571 if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 { 1572 return errPeersUnavailable 1573 } 1574 } 1575 } 1576 } 1577 1578 // processHeaders takes batches of retrieved headers from an input channel and 1579 // keeps processing and scheduling them into the header chain and downloader's 1580 // queue until the stream ends or a failure occurs. 1581 func (d *Downloader) processHeaders(origin uint64, td *big.Int) error { 1582 // Keep a count of uncertain headers to roll back 1583 var ( 1584 rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis) 1585 rollbackErr error 1586 mode = d.getMode() 1587 ) 1588 defer func() { 1589 if rollback > 0 { 1590 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1591 if mode != LightSync { 1592 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1593 lastBlock = d.blockchain.CurrentBlock().Number() 1594 } 1595 if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block 1596 // We're already unwinding the stack, only print the error to make it more visible 1597 log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err) 1598 } 1599 curFastBlock, curBlock := common.Big0, common.Big0 1600 if mode != LightSync { 1601 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1602 curBlock = d.blockchain.CurrentBlock().Number() 1603 } 1604 log.Warn("Rolled back chain segment", 1605 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1606 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1607 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr) 1608 } 1609 }() 1610 // Wait for batches of headers to process 1611 gotHeaders := false 1612 1613 for { 1614 select { 1615 case <-d.cancelCh: 1616 rollbackErr = errCanceled 1617 return errCanceled 1618 1619 case headers := <-d.headerProcCh: 1620 // Terminate header processing if we synced up 1621 if len(headers) == 0 { 1622 // Notify everyone that headers are fully processed 1623 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1624 select { 1625 case ch <- false: 1626 case <-d.cancelCh: 1627 } 1628 } 1629 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1630 // better chain compared to ours. The only exception is if its promised blocks were 1631 // already imported by other means (e.g. fetcher): 1632 // 1633 // R <remote peer>, L <local node>: Both at block 10 1634 // R: Mine block 11, and propagate it to L 1635 // L: Queue block 11 for import 1636 // L: Notice that R's head and TD increased compared to ours, start sync 1637 // L: Import of block 11 finishes 1638 // L: Sync begins, and finds common ancestor at 11 1639 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1640 // R: Nothing to give 1641 if mode != LightSync { 1642 head := d.blockchain.CurrentBlock() 1643 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1644 return errStallingPeer 1645 } 1646 } 1647 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1648 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1649 // of delivering the post-pivot blocks that would flag the invalid content. 1650 // 1651 // This check cannot be executed "as is" for full imports, since blocks may still be 1652 // queued for processing when the header download completes. However, as long as the 1653 // peer gave us something useful, we're already happy/progressed (above check). 1654 if mode == FastSync || mode == LightSync { 1655 head := d.lightchain.CurrentHeader() 1656 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1657 return errStallingPeer 1658 } 1659 } 1660 // Disable any rollback and return 1661 rollback = 0 1662 return nil 1663 } 1664 // Otherwise split the chunk of headers into batches and process them 1665 gotHeaders = true 1666 for len(headers) > 0 { 1667 // Terminate if something failed in between processing chunks 1668 select { 1669 case <-d.cancelCh: 1670 rollbackErr = errCanceled 1671 return errCanceled 1672 default: 1673 } 1674 // Select the next chunk of headers to import 1675 limit := maxHeadersProcess 1676 if limit > len(headers) { 1677 limit = len(headers) 1678 } 1679 chunk := headers[:limit] 1680 1681 // In case of header only syncing, validate the chunk immediately 1682 if mode == FastSync || mode == LightSync { 1683 // If we're importing pure headers, verify based on their recentness 1684 var pivot uint64 1685 1686 d.pivotLock.RLock() 1687 if d.pivotHeader != nil { 1688 pivot = d.pivotHeader.Number.Uint64() 1689 } 1690 d.pivotLock.RUnlock() 1691 1692 frequency := fsHeaderCheckFrequency 1693 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1694 frequency = 1 1695 } 1696 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1697 rollbackErr = err 1698 1699 // If some headers were inserted, track them as uncertain 1700 if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 { 1701 rollback = chunk[0].Number.Uint64() 1702 } 1703 log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err) 1704 return fmt.Errorf("%w: %v", errInvalidChain, err) 1705 } 1706 // All verifications passed, track all headers within the alloted limits 1707 if mode == FastSync { 1708 head := chunk[len(chunk)-1].Number.Uint64() 1709 if head-rollback > uint64(fsHeaderSafetyNet) { 1710 rollback = head - uint64(fsHeaderSafetyNet) 1711 } else { 1712 rollback = 1 1713 } 1714 } 1715 } 1716 // Unless we're doing light chains, schedule the headers for associated content retrieval 1717 if mode == FullSync || mode == FastSync { 1718 // If we've reached the allowed number of pending headers, stall a bit 1719 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1720 select { 1721 case <-d.cancelCh: 1722 rollbackErr = errCanceled 1723 return errCanceled 1724 case <-time.After(time.Second): 1725 } 1726 } 1727 // Otherwise insert the headers for content retrieval 1728 inserts := d.queue.Schedule(chunk, origin) 1729 if len(inserts) != len(chunk) { 1730 rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk)) 1731 return fmt.Errorf("%w: stale headers", errBadPeer) 1732 } 1733 } 1734 headers = headers[limit:] 1735 origin += uint64(limit) 1736 } 1737 // Update the highest block number we know if a higher one is found. 1738 d.syncStatsLock.Lock() 1739 if d.syncStatsChainHeight < origin { 1740 d.syncStatsChainHeight = origin - 1 1741 } 1742 d.syncStatsLock.Unlock() 1743 1744 // Signal the content downloaders of the availablility of new tasks 1745 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1746 select { 1747 case ch <- true: 1748 default: 1749 } 1750 } 1751 } 1752 } 1753 } 1754 1755 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1756 func (d *Downloader) processFullSyncContent() error { 1757 for { 1758 results := d.queue.Results(true) 1759 if len(results) == 0 { 1760 return nil 1761 } 1762 if d.chainInsertHook != nil { 1763 d.chainInsertHook(results) 1764 } 1765 if err := d.importBlockResults(results); err != nil { 1766 return err 1767 } 1768 } 1769 } 1770 1771 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1772 // Check for any early termination requests 1773 if len(results) == 0 { 1774 return nil 1775 } 1776 select { 1777 case <-d.quitCh: 1778 return errCancelContentProcessing 1779 default: 1780 } 1781 // Retrieve the a batch of results to import 1782 first, last := results[0].Header, results[len(results)-1].Header 1783 log.Debug("Inserting downloaded chain", "items", len(results), 1784 "firstnum", first.Number, "firsthash", first.Hash(), 1785 "lastnum", last.Number, "lasthash", last.Hash(), 1786 ) 1787 blocks := make([]*types.Block, len(results)) 1788 for i, result := range results { 1789 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1790 } 1791 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1792 if index < len(results) { 1793 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1794 } else { 1795 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1796 // when it needs to preprocess blocks to import a sidechain. 1797 // The importer will put together a new list of blocks to import, which is a superset 1798 // of the blocks delivered from the downloader, and the indexing will be off. 1799 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1800 } 1801 return fmt.Errorf("%w: %v", errInvalidChain, err) 1802 } 1803 return nil 1804 } 1805 1806 // processFastSyncContent takes fetch results from the queue and writes them to the 1807 // database. It also controls the synchronisation of state nodes of the pivot block. 1808 func (d *Downloader) processFastSyncContent() error { 1809 // Start syncing state of the reported head block. This should get us most of 1810 // the state of the pivot block. 1811 d.pivotLock.RLock() 1812 sync := d.syncState(d.pivotHeader.Root) 1813 d.pivotLock.RUnlock() 1814 1815 defer func() { 1816 // The `sync` object is replaced every time the pivot moves. We need to 1817 // defer close the very last active one, hence the lazy evaluation vs. 1818 // calling defer sync.Cancel() !!! 1819 sync.Cancel() 1820 }() 1821 1822 closeOnErr := func(s *stateSync) { 1823 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled && err != snap.ErrCancelled { 1824 d.queue.Close() // wake up Results 1825 } 1826 } 1827 go closeOnErr(sync) 1828 1829 // To cater for moving pivot points, track the pivot block and subsequently 1830 // accumulated download results separately. 1831 var ( 1832 oldPivot *fetchResult // Locked in pivot block, might change eventually 1833 oldTail []*fetchResult // Downloaded content after the pivot 1834 ) 1835 for { 1836 // Wait for the next batch of downloaded data to be available, and if the pivot 1837 // block became stale, move the goalpost 1838 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1839 if len(results) == 0 { 1840 // If pivot sync is done, stop 1841 if oldPivot == nil { 1842 return sync.Cancel() 1843 } 1844 // If sync failed, stop 1845 select { 1846 case <-d.cancelCh: 1847 sync.Cancel() 1848 return errCanceled 1849 default: 1850 } 1851 } 1852 if d.chainInsertHook != nil { 1853 d.chainInsertHook(results) 1854 } 1855 // If we haven't downloaded the pivot block yet, check pivot staleness 1856 // notifications from the header downloader 1857 d.pivotLock.RLock() 1858 pivot := d.pivotHeader 1859 d.pivotLock.RUnlock() 1860 1861 if oldPivot == nil { 1862 if pivot.Root != sync.root { 1863 sync.Cancel() 1864 sync = d.syncState(pivot.Root) 1865 1866 go closeOnErr(sync) 1867 } 1868 } else { 1869 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1870 } 1871 // Split around the pivot block and process the two sides via fast/full sync 1872 if atomic.LoadInt32(&d.committed) == 0 { 1873 latest := results[len(results)-1].Header 1874 // If the height is above the pivot block by 2 sets, it means the pivot 1875 // become stale in the network and it was garbage collected, move to a 1876 // new pivot. 1877 // 1878 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 1879 // need to be taken into account, otherwise we're detecting the pivot move 1880 // late and will drop peers due to unavailable state!!! 1881 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 1882 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 1883 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 1884 1885 d.pivotLock.Lock() 1886 d.pivotHeader = pivot 1887 d.pivotLock.Unlock() 1888 1889 // Write out the pivot into the database so a rollback beyond it will 1890 // reenable fast sync 1891 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 1892 } 1893 } 1894 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 1895 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1896 return err 1897 } 1898 if P != nil { 1899 // If new pivot block found, cancel old state retrieval and restart 1900 if oldPivot != P { 1901 sync.Cancel() 1902 sync = d.syncState(P.Header.Root) 1903 1904 go closeOnErr(sync) 1905 oldPivot = P 1906 } 1907 // Wait for completion, occasionally checking for pivot staleness 1908 select { 1909 case <-sync.done: 1910 if sync.err != nil { 1911 return sync.err 1912 } 1913 if err := d.commitPivotBlock(P); err != nil { 1914 return err 1915 } 1916 oldPivot = nil 1917 1918 case <-time.After(time.Second): 1919 oldTail = afterP 1920 continue 1921 } 1922 } 1923 // Fast sync done, pivot commit done, full import 1924 if err := d.importBlockResults(afterP); err != nil { 1925 return err 1926 } 1927 } 1928 } 1929 1930 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1931 if len(results) == 0 { 1932 return nil, nil, nil 1933 } 1934 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 1935 // the pivot is somewhere in the future 1936 return nil, results, nil 1937 } 1938 // This can also be optimized, but only happens very seldom 1939 for _, result := range results { 1940 num := result.Header.Number.Uint64() 1941 switch { 1942 case num < pivot: 1943 before = append(before, result) 1944 case num == pivot: 1945 p = result 1946 default: 1947 after = append(after, result) 1948 } 1949 } 1950 return p, before, after 1951 } 1952 1953 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1954 // Check for any early termination requests 1955 if len(results) == 0 { 1956 return nil 1957 } 1958 select { 1959 case <-d.quitCh: 1960 return errCancelContentProcessing 1961 case <-stateSync.done: 1962 if err := stateSync.Wait(); err != nil { 1963 return err 1964 } 1965 default: 1966 } 1967 // Retrieve the a batch of results to import 1968 first, last := results[0].Header, results[len(results)-1].Header 1969 log.Debug("Inserting fast-sync blocks", "items", len(results), 1970 "firstnum", first.Number, "firsthash", first.Hash(), 1971 "lastnumn", last.Number, "lasthash", last.Hash(), 1972 ) 1973 blocks := make([]*types.Block, len(results)) 1974 receipts := make([]types.Receipts, len(results)) 1975 for i, result := range results { 1976 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1977 receipts[i] = result.Receipts 1978 } 1979 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1980 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1981 return fmt.Errorf("%w: %v", errInvalidChain, err) 1982 } 1983 return nil 1984 } 1985 1986 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1987 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1988 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1989 1990 // Commit the pivot block as the new head, will require full sync from here on 1991 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1992 return err 1993 } 1994 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1995 return err 1996 } 1997 atomic.StoreInt32(&d.committed, 1) 1998 1999 // If we had a bloom filter for the state sync, deallocate it now. Note, we only 2000 // deallocate internally, but keep the empty wrapper. This ensures that if we do 2001 // a rollback after committing the pivot and restarting fast sync, we don't end 2002 // up using a nil bloom. Empty bloom is fine, it just returns that it does not 2003 // have the info we need, so reach down to the database instead. 2004 if d.stateBloom != nil { 2005 d.stateBloom.Close() 2006 } 2007 return nil 2008 } 2009 2010 // DeliverHeaders injects a new batch of block headers received from a remote 2011 // node into the download schedule. 2012 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) error { 2013 return d.deliver(d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 2014 } 2015 2016 // DeliverBodies injects a new batch of block bodies received from a remote node. 2017 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) error { 2018 return d.deliver(d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 2019 } 2020 2021 // DeliverReceipts injects a new batch of receipts received from a remote node. 2022 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) error { 2023 return d.deliver(d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 2024 } 2025 2026 // DeliverNodeData injects a new batch of node state data received from a remote node. 2027 func (d *Downloader) DeliverNodeData(id string, data [][]byte) error { 2028 return d.deliver(d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 2029 } 2030 2031 // DeliverSnapPacket is invoked from a peer's message handler when it transmits a 2032 // data packet for the local node to consume. 2033 func (d *Downloader) DeliverSnapPacket(peer *snap.Peer, packet snap.Packet) error { 2034 switch packet := packet.(type) { 2035 case *snap.AccountRangePacket: 2036 hashes, accounts, err := packet.Unpack() 2037 if err != nil { 2038 return err 2039 } 2040 return d.SnapSyncer.OnAccounts(peer, packet.ID, hashes, accounts, packet.Proof) 2041 2042 case *snap.StorageRangesPacket: 2043 hashset, slotset := packet.Unpack() 2044 return d.SnapSyncer.OnStorage(peer, packet.ID, hashset, slotset, packet.Proof) 2045 2046 case *snap.ByteCodesPacket: 2047 return d.SnapSyncer.OnByteCodes(peer, packet.ID, packet.Codes) 2048 2049 case *snap.TrieNodesPacket: 2050 return d.SnapSyncer.OnTrieNodes(peer, packet.ID, packet.Nodes) 2051 2052 default: 2053 return fmt.Errorf("unexpected snap packet type: %T", packet) 2054 } 2055 } 2056 2057 // deliver injects a new batch of data received from a remote node. 2058 func (d *Downloader) deliver(destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 2059 // Update the delivery metrics for both good and failed deliveries 2060 inMeter.Mark(int64(packet.Items())) 2061 defer func() { 2062 if err != nil { 2063 dropMeter.Mark(int64(packet.Items())) 2064 } 2065 }() 2066 // Deliver or abort if the sync is canceled while queuing 2067 d.cancelLock.RLock() 2068 cancel := d.cancelCh 2069 d.cancelLock.RUnlock() 2070 if cancel == nil { 2071 return errNoSyncActive 2072 } 2073 select { 2074 case destCh <- packet: 2075 return nil 2076 case <-cancel: 2077 return errNoSyncActive 2078 } 2079 } 2080 2081 // qosTuner is the quality of service tuning loop that occasionally gathers the 2082 // peer latency statistics and updates the estimated request round trip time. 2083 func (d *Downloader) qosTuner() { 2084 for { 2085 // Retrieve the current median RTT and integrate into the previoust target RTT 2086 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 2087 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 2088 2089 // A new RTT cycle passed, increase our confidence in the estimated RTT 2090 conf := atomic.LoadUint64(&d.rttConfidence) 2091 conf = conf + (1000000-conf)/2 2092 atomic.StoreUint64(&d.rttConfidence, conf) 2093 2094 // Log the new QoS values and sleep until the next RTT 2095 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 2096 select { 2097 case <-d.quitCh: 2098 return 2099 case <-time.After(rtt): 2100 } 2101 } 2102 } 2103 2104 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 2105 // peer set, needing to reduce the confidence we have in out QoS estimates. 2106 func (d *Downloader) qosReduceConfidence() { 2107 // If we have a single peer, confidence is always 1 2108 peers := uint64(d.peers.Len()) 2109 if peers == 0 { 2110 // Ensure peer connectivity races don't catch us off guard 2111 return 2112 } 2113 if peers == 1 { 2114 atomic.StoreUint64(&d.rttConfidence, 1000000) 2115 return 2116 } 2117 // If we have a ton of peers, don't drop confidence) 2118 if peers >= uint64(qosConfidenceCap) { 2119 return 2120 } 2121 // Otherwise drop the confidence factor 2122 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 2123 if float64(conf)/1000000 < rttMinConfidence { 2124 conf = uint64(rttMinConfidence * 1000000) 2125 } 2126 atomic.StoreUint64(&d.rttConfidence, conf) 2127 2128 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2129 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 2130 } 2131 2132 // requestRTT returns the current target round trip time for a download request 2133 // to complete in. 2134 // 2135 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 2136 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 2137 // be adapted to, but smaller ones are preferred (stabler download stream). 2138 func (d *Downloader) requestRTT() time.Duration { 2139 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 2140 } 2141 2142 // requestTTL returns the current timeout allowance for a single download request 2143 // to finish under. 2144 func (d *Downloader) requestTTL() time.Duration { 2145 var ( 2146 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2147 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 2148 ) 2149 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 2150 if ttl > ttlLimit { 2151 ttl = ttlLimit 2152 } 2153 return ttl 2154 }