github.com/CommerciumBlockchain/go-commercium@v0.0.0-20220709212705-b46438a77516/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/CommerciumBlockchain/go-commercium" 29 "github.com/CommerciumBlockchain/go-commercium/common" 30 "github.com/CommerciumBlockchain/go-commercium/core/rawdb" 31 "github.com/CommerciumBlockchain/go-commercium/core/types" 32 "github.com/CommerciumBlockchain/go-commercium/eth/protocols/snap" 33 "github.com/CommerciumBlockchain/go-commercium/ethdb" 34 "github.com/CommerciumBlockchain/go-commercium/event" 35 "github.com/CommerciumBlockchain/go-commercium/log" 36 "github.com/CommerciumBlockchain/go-commercium/metrics" 37 "github.com/CommerciumBlockchain/go-commercium/params" 38 "github.com/CommerciumBlockchain/go-commercium/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 doesn't speak recent enough protocol version (need version >= 64)") 93 ) 94 95 type Downloader struct { 96 // WARNING: The `rttEstimate` and `rttConfidence` fields are accessed atomically. 97 // On 32 bit platforms, only 64-bit aligned fields can be atomic. The struct is 98 // guaranteed to be so aligned, so take advantage of that. For more information, 99 // see https://golang.org/pkg/sync/atomic/#pkg-note-BUG. 100 rttEstimate uint64 // Round trip time to target for download requests 101 rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops) 102 103 mode uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode 104 mux *event.TypeMux // Event multiplexer to announce sync operation events 105 106 checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync) 107 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 108 queue *queue // Scheduler for selecting the hashes to download 109 peers *peerSet // Set of active peers from which download can proceed 110 111 stateDB ethdb.Database // Database to state sync into (and deduplicate via) 112 stateBloom *trie.SyncBloom // Bloom filter for fast trie node and contract code existence checks 113 114 // Statistics 115 syncStatsChainOrigin uint64 // Origin block number where syncing started at 116 syncStatsChainHeight uint64 // Highest block number known when syncing started 117 syncStatsState stateSyncStats 118 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 119 120 lightchain LightChain 121 blockchain BlockChain 122 123 // Callbacks 124 dropPeer peerDropFn // Drops a peer for misbehaving 125 126 // Status 127 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 128 synchronising int32 129 notified int32 130 committed int32 131 ancientLimit uint64 // The maximum block number which can be regarded as ancient data. 132 133 // Channels 134 headerCh chan dataPack // Channel receiving inbound block headers 135 bodyCh chan dataPack // Channel receiving inbound block bodies 136 receiptCh chan dataPack // Channel receiving inbound receipts 137 bodyWakeCh chan bool // Channel to signal the block body fetcher of new tasks 138 receiptWakeCh chan bool // Channel to signal the receipt fetcher of new tasks 139 headerProcCh chan []*types.Header // Channel to feed the header processor new tasks 140 141 // State sync 142 pivotHeader *types.Header // Pivot block header to dynamically push the syncing state root 143 pivotLock sync.RWMutex // Lock protecting pivot header reads from updates 144 145 snapSync bool // Whether to run state sync over the snap protocol 146 SnapSyncer *snap.Syncer // TODO(karalabe): make private! hack for now 147 stateSyncStart chan *stateSync 148 trackStateReq chan *stateReq 149 stateCh chan dataPack // Channel receiving inbound node state data 150 151 // Cancellation and termination 152 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 153 cancelCh chan struct{} // Channel to cancel mid-flight syncs 154 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 155 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 156 157 quitCh chan struct{} // Quit channel to signal termination 158 quitLock sync.Mutex // Lock to prevent double closes 159 160 // Testing hooks 161 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 162 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 163 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 164 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 165 } 166 167 // LightChain encapsulates functions required to synchronise a light chain. 168 type LightChain interface { 169 // HasHeader verifies a header's presence in the local chain. 170 HasHeader(common.Hash, uint64) bool 171 172 // GetHeaderByHash retrieves a header from the local chain. 173 GetHeaderByHash(common.Hash) *types.Header 174 175 // CurrentHeader retrieves the head header from the local chain. 176 CurrentHeader() *types.Header 177 178 // GetTd returns the total difficulty of a local block. 179 GetTd(common.Hash, uint64) *big.Int 180 181 // InsertHeaderChain inserts a batch of headers into the local chain. 182 InsertHeaderChain([]*types.Header, int) (int, error) 183 184 // SetHead rewinds the local chain to a new head. 185 SetHead(uint64) error 186 } 187 188 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 189 type BlockChain interface { 190 LightChain 191 192 // HasBlock verifies a block's presence in the local chain. 193 HasBlock(common.Hash, uint64) bool 194 195 // HasFastBlock verifies a fast block's presence in the local chain. 196 HasFastBlock(common.Hash, uint64) bool 197 198 // GetBlockByHash retrieves a block from the local chain. 199 GetBlockByHash(common.Hash) *types.Block 200 201 // CurrentBlock retrieves the head block from the local chain. 202 CurrentBlock() *types.Block 203 204 // CurrentFastBlock retrieves the head fast block from the local chain. 205 CurrentFastBlock() *types.Block 206 207 // FastSyncCommitHead directly commits the head block to a certain entity. 208 FastSyncCommitHead(common.Hash) error 209 210 // InsertChain inserts a batch of blocks into the local chain. 211 InsertChain(types.Blocks) (int, error) 212 213 // InsertReceiptChain inserts a batch of receipts into the local chain. 214 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 215 } 216 217 // New creates a new downloader to fetch hashes and blocks from remote peers. 218 func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 219 if lightchain == nil { 220 lightchain = chain 221 } 222 dl := &Downloader{ 223 stateDB: stateDb, 224 stateBloom: stateBloom, 225 mux: mux, 226 checkpoint: checkpoint, 227 queue: newQueue(blockCacheMaxItems, blockCacheInitialItems), 228 peers: newPeerSet(), 229 rttEstimate: uint64(rttMaxEstimate), 230 rttConfidence: uint64(1000000), 231 blockchain: chain, 232 lightchain: lightchain, 233 dropPeer: dropPeer, 234 headerCh: make(chan dataPack, 1), 235 bodyCh: make(chan dataPack, 1), 236 receiptCh: make(chan dataPack, 1), 237 bodyWakeCh: make(chan bool, 1), 238 receiptWakeCh: make(chan bool, 1), 239 headerProcCh: make(chan []*types.Header, 1), 240 quitCh: make(chan struct{}), 241 stateCh: make(chan dataPack), 242 SnapSyncer: snap.NewSyncer(stateDb, stateBloom), 243 stateSyncStart: make(chan *stateSync), 244 syncStatsState: stateSyncStats{ 245 processed: rawdb.ReadFastTrieProgress(stateDb), 246 }, 247 trackStateReq: make(chan *stateReq), 248 } 249 go dl.qosTuner() 250 go dl.stateFetcher() 251 return dl 252 } 253 254 // Progress retrieves the synchronisation boundaries, specifically the origin 255 // block where synchronisation started at (may have failed/suspended); the block 256 // or header sync is currently at; and the latest known block which the sync targets. 257 // 258 // In addition, during the state download phase of fast synchronisation the number 259 // of processed and the total number of known states are also returned. Otherwise 260 // these are zero. 261 func (d *Downloader) Progress() ethereum.SyncProgress { 262 // Lock the current stats and return the progress 263 d.syncStatsLock.RLock() 264 defer d.syncStatsLock.RUnlock() 265 266 current := uint64(0) 267 mode := d.getMode() 268 switch { 269 case d.blockchain != nil && mode == FullSync: 270 current = d.blockchain.CurrentBlock().NumberU64() 271 case d.blockchain != nil && mode == FastSync: 272 current = d.blockchain.CurrentFastBlock().NumberU64() 273 case d.lightchain != nil: 274 current = d.lightchain.CurrentHeader().Number.Uint64() 275 default: 276 log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", mode) 277 } 278 return ethereum.SyncProgress{ 279 StartingBlock: d.syncStatsChainOrigin, 280 CurrentBlock: current, 281 HighestBlock: d.syncStatsChainHeight, 282 PulledStates: d.syncStatsState.processed, 283 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 284 } 285 } 286 287 // Synchronising returns whether the downloader is currently retrieving blocks. 288 func (d *Downloader) Synchronising() bool { 289 return atomic.LoadInt32(&d.synchronising) > 0 290 } 291 292 // RegisterPeer injects a new download peer into the set of block source to be 293 // used for fetching hashes and blocks from. 294 func (d *Downloader) RegisterPeer(id string, version uint, peer Peer) error { 295 var logger log.Logger 296 if len(id) < 16 { 297 // Tests use short IDs, don't choke on them 298 logger = log.New("peer", id) 299 } else { 300 logger = log.New("peer", id[:16]) 301 } 302 logger.Trace("Registering sync peer") 303 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 304 logger.Error("Failed to register sync peer", "err", err) 305 return err 306 } 307 d.qosReduceConfidence() 308 309 return nil 310 } 311 312 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 313 func (d *Downloader) RegisterLightPeer(id string, version uint, peer LightPeer) error { 314 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 315 } 316 317 // UnregisterPeer remove a peer from the known list, preventing any action from 318 // the specified peer. An effort is also made to return any pending fetches into 319 // the queue. 320 func (d *Downloader) UnregisterPeer(id string) error { 321 // Unregister the peer from the active peer set and revoke any fetch tasks 322 var logger log.Logger 323 if len(id) < 16 { 324 // Tests use short IDs, don't choke on them 325 logger = log.New("peer", id) 326 } else { 327 logger = log.New("peer", id[:16]) 328 } 329 logger.Trace("Unregistering sync peer") 330 if err := d.peers.Unregister(id); err != nil { 331 logger.Error("Failed to unregister sync peer", "err", err) 332 return err 333 } 334 d.queue.Revoke(id) 335 336 return nil 337 } 338 339 // Synchronise tries to sync up our local block chain with a remote peer, both 340 // adding various sanity checks as well as wrapping it with various log entries. 341 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 342 err := d.synchronise(id, head, td, mode) 343 344 switch err { 345 case nil, errBusy, errCanceled: 346 return err 347 } 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, peer version: %d", errTooOld, p.version) 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 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 818 819 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 820 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 821 822 // Wait for the remote response to the head fetch 823 number, hash := uint64(0), common.Hash{} 824 825 ttl := d.requestTTL() 826 timeout := time.After(ttl) 827 828 for finished := false; !finished; { 829 select { 830 case <-d.cancelCh: 831 return 0, errCanceled 832 833 case packet := <-d.headerCh: 834 // Discard anything not from the origin peer 835 if packet.PeerId() != p.id { 836 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 837 break 838 } 839 // Make sure the peer actually gave something valid 840 headers := packet.(*headerPack).headers 841 if len(headers) == 0 { 842 p.log.Warn("Empty head header set") 843 return 0, errEmptyHeaderSet 844 } 845 // Make sure the peer's reply conforms to the request 846 for i, header := range headers { 847 expectNumber := from + int64(i)*int64(skip+1) 848 if number := header.Number.Int64(); number != expectNumber { 849 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 850 return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering")) 851 } 852 } 853 // Check if a common ancestor was found 854 finished = true 855 for i := len(headers) - 1; i >= 0; i-- { 856 // Skip any headers that underflow/overflow our requested set 857 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 858 continue 859 } 860 // Otherwise check if we already know the header or not 861 h := headers[i].Hash() 862 n := headers[i].Number.Uint64() 863 864 var known bool 865 switch mode { 866 case FullSync: 867 known = d.blockchain.HasBlock(h, n) 868 case FastSync: 869 known = d.blockchain.HasFastBlock(h, n) 870 default: 871 known = d.lightchain.HasHeader(h, n) 872 } 873 if known { 874 number, hash = n, h 875 break 876 } 877 } 878 879 case <-timeout: 880 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 881 return 0, errTimeout 882 883 case <-d.bodyCh: 884 case <-d.receiptCh: 885 // Out of bounds delivery, ignore 886 } 887 } 888 // If the head fetch already found an ancestor, return 889 if hash != (common.Hash{}) { 890 if int64(number) <= floor { 891 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 892 return 0, errInvalidAncestor 893 } 894 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 895 return number, nil 896 } 897 // Ancestor not found, we need to binary search over our chain 898 start, end := uint64(0), remoteHeight 899 if floor > 0 { 900 start = uint64(floor) 901 } 902 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 903 904 for start+1 < end { 905 // Split our chain interval in two, and request the hash to cross check 906 check := (start + end) / 2 907 908 ttl := d.requestTTL() 909 timeout := time.After(ttl) 910 911 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 912 913 // Wait until a reply arrives to this request 914 for arrived := false; !arrived; { 915 select { 916 case <-d.cancelCh: 917 return 0, errCanceled 918 919 case packet := <-d.headerCh: 920 // Discard anything not from the origin peer 921 if packet.PeerId() != p.id { 922 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 923 break 924 } 925 // Make sure the peer actually gave something valid 926 headers := packet.(*headerPack).headers 927 if len(headers) != 1 { 928 p.log.Warn("Multiple headers for single request", "headers", len(headers)) 929 return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers)) 930 } 931 arrived = true 932 933 // Modify the search interval based on the response 934 h := headers[0].Hash() 935 n := headers[0].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 end = check 948 break 949 } 950 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 951 if header.Number.Uint64() != check { 952 p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 953 return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number) 954 } 955 start = check 956 hash = h 957 958 case <-timeout: 959 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 960 return 0, errTimeout 961 962 case <-d.bodyCh: 963 case <-d.receiptCh: 964 // Out of bounds delivery, ignore 965 } 966 } 967 } 968 // Ensure valid ancestry and return 969 if int64(start) <= floor { 970 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 971 return 0, errInvalidAncestor 972 } 973 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 974 return start, nil 975 } 976 977 // fetchHeaders keeps retrieving headers concurrently from the number 978 // requested, until no more are returned, potentially throttling on the way. To 979 // facilitate concurrency but still protect against malicious nodes sending bad 980 // headers, we construct a header chain skeleton using the "origin" peer we are 981 // syncing with, and fill in the missing headers using anyone else. Headers from 982 // other peers are only accepted if they map cleanly to the skeleton. If no one 983 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 984 // the origin is dropped. 985 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error { 986 p.log.Debug("Directing header downloads", "origin", from) 987 defer p.log.Debug("Header download terminated") 988 989 // Create a timeout timer, and the associated header fetcher 990 skeleton := true // Skeleton assembly phase or finishing up 991 pivoting := false // Whether the next request is pivot verification 992 request := time.Now() // time of the last skeleton fetch request 993 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 994 <-timeout.C // timeout channel should be initially empty 995 defer timeout.Stop() 996 997 var ttl time.Duration 998 getHeaders := func(from uint64) { 999 request = time.Now() 1000 1001 ttl = d.requestTTL() 1002 timeout.Reset(ttl) 1003 1004 if skeleton { 1005 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 1006 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 1007 } else { 1008 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 1009 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 1010 } 1011 } 1012 getNextPivot := func() { 1013 pivoting = true 1014 request = time.Now() 1015 1016 ttl = d.requestTTL() 1017 timeout.Reset(ttl) 1018 1019 d.pivotLock.RLock() 1020 pivot := d.pivotHeader.Number.Uint64() 1021 d.pivotLock.RUnlock() 1022 1023 p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks)) 1024 go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep 1025 } 1026 // Start pulling the header chain skeleton until all is done 1027 ancestor := from 1028 getHeaders(from) 1029 1030 mode := d.getMode() 1031 for { 1032 select { 1033 case <-d.cancelCh: 1034 return errCanceled 1035 1036 case packet := <-d.headerCh: 1037 // Make sure the active peer is giving us the skeleton headers 1038 if packet.PeerId() != p.id { 1039 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 1040 break 1041 } 1042 headerReqTimer.UpdateSince(request) 1043 timeout.Stop() 1044 1045 // If the pivot is being checked, move if it became stale and run the real retrieval 1046 var pivot uint64 1047 1048 d.pivotLock.RLock() 1049 if d.pivotHeader != nil { 1050 pivot = d.pivotHeader.Number.Uint64() 1051 } 1052 d.pivotLock.RUnlock() 1053 1054 if pivoting { 1055 if packet.Items() == 2 { 1056 // Retrieve the headers and do some sanity checks, just in case 1057 headers := packet.(*headerPack).headers 1058 1059 if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want { 1060 log.Warn("Peer sent invalid next pivot", "have", have, "want", want) 1061 return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want) 1062 } 1063 if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want { 1064 log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want) 1065 return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want) 1066 } 1067 log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number) 1068 pivot = headers[0].Number.Uint64() 1069 1070 d.pivotLock.Lock() 1071 d.pivotHeader = headers[0] 1072 d.pivotLock.Unlock() 1073 1074 // Write out the pivot into the database so a rollback beyond 1075 // it will reenable fast sync and update the state root that 1076 // the state syncer will be downloading. 1077 rawdb.WriteLastPivotNumber(d.stateDB, pivot) 1078 } 1079 pivoting = false 1080 getHeaders(from) 1081 continue 1082 } 1083 // If the skeleton's finished, pull any remaining head headers directly from the origin 1084 if skeleton && packet.Items() == 0 { 1085 skeleton = false 1086 getHeaders(from) 1087 continue 1088 } 1089 // If no more headers are inbound, notify the content fetchers and return 1090 if packet.Items() == 0 { 1091 // Don't abort header fetches while the pivot is downloading 1092 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 1093 p.log.Debug("No headers, waiting for pivot commit") 1094 select { 1095 case <-time.After(fsHeaderContCheck): 1096 getHeaders(from) 1097 continue 1098 case <-d.cancelCh: 1099 return errCanceled 1100 } 1101 } 1102 // Pivot done (or not in fast sync) and no more headers, terminate the process 1103 p.log.Debug("No more headers available") 1104 select { 1105 case d.headerProcCh <- nil: 1106 return nil 1107 case <-d.cancelCh: 1108 return errCanceled 1109 } 1110 } 1111 headers := packet.(*headerPack).headers 1112 1113 // If we received a skeleton batch, resolve internals concurrently 1114 if skeleton { 1115 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1116 if err != nil { 1117 p.log.Debug("Skeleton chain invalid", "err", err) 1118 return fmt.Errorf("%w: %v", errInvalidChain, err) 1119 } 1120 headers = filled[proced:] 1121 from += uint64(proced) 1122 } else { 1123 // If we're closing in on the chain head, but haven't yet reached it, delay 1124 // the last few headers so mini reorgs on the head don't cause invalid hash 1125 // chain errors. 1126 if n := len(headers); n > 0 { 1127 // Retrieve the current head we're at 1128 var head uint64 1129 if mode == LightSync { 1130 head = d.lightchain.CurrentHeader().Number.Uint64() 1131 } else { 1132 head = d.blockchain.CurrentFastBlock().NumberU64() 1133 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1134 head = full 1135 } 1136 } 1137 // If the head is below the common ancestor, we're actually deduplicating 1138 // already existing chain segments, so use the ancestor as the fake head. 1139 // Otherwise we might end up delaying header deliveries pointlessly. 1140 if head < ancestor { 1141 head = ancestor 1142 } 1143 // If the head is way older than this batch, delay the last few headers 1144 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1145 delay := reorgProtHeaderDelay 1146 if delay > n { 1147 delay = n 1148 } 1149 headers = headers[:n-delay] 1150 } 1151 } 1152 } 1153 // Insert all the new headers and fetch the next batch 1154 if len(headers) > 0 { 1155 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1156 select { 1157 case d.headerProcCh <- headers: 1158 case <-d.cancelCh: 1159 return errCanceled 1160 } 1161 from += uint64(len(headers)) 1162 1163 // If we're still skeleton filling fast sync, check pivot staleness 1164 // before continuing to the next skeleton filling 1165 if skeleton && pivot > 0 { 1166 getNextPivot() 1167 } else { 1168 getHeaders(from) 1169 } 1170 } else { 1171 // No headers delivered, or all of them being delayed, sleep a bit and retry 1172 p.log.Trace("All headers delayed, waiting") 1173 select { 1174 case <-time.After(fsHeaderContCheck): 1175 getHeaders(from) 1176 continue 1177 case <-d.cancelCh: 1178 return errCanceled 1179 } 1180 } 1181 1182 case <-timeout.C: 1183 if d.dropPeer == nil { 1184 // The dropPeer method is nil when `--copydb` is used for a local copy. 1185 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1186 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1187 break 1188 } 1189 // Header retrieval timed out, consider the peer bad and drop 1190 p.log.Debug("Header request timed out", "elapsed", ttl) 1191 headerTimeoutMeter.Mark(1) 1192 d.dropPeer(p.id) 1193 1194 // Finish the sync gracefully instead of dumping the gathered data though 1195 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1196 select { 1197 case ch <- false: 1198 case <-d.cancelCh: 1199 } 1200 } 1201 select { 1202 case d.headerProcCh <- nil: 1203 case <-d.cancelCh: 1204 } 1205 return fmt.Errorf("%w: header request timed out", errBadPeer) 1206 } 1207 } 1208 } 1209 1210 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1211 // and maps them to the provided skeleton header chain. 1212 // 1213 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1214 // immediately to the header processor to keep the rest of the pipeline full even 1215 // in the case of header stalls. 1216 // 1217 // The method returns the entire filled skeleton and also the number of headers 1218 // already forwarded for processing. 1219 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1220 log.Debug("Filling up skeleton", "from", from) 1221 d.queue.ScheduleSkeleton(from, skeleton) 1222 1223 var ( 1224 deliver = func(packet dataPack) (int, error) { 1225 pack := packet.(*headerPack) 1226 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1227 } 1228 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 1229 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) { 1230 return d.queue.ReserveHeaders(p, count), false, false 1231 } 1232 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1233 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 1234 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1235 p.SetHeadersIdle(accepted, deliveryTime) 1236 } 1237 ) 1238 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1239 d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve, 1240 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1241 1242 log.Debug("Skeleton fill terminated", "err", err) 1243 1244 filled, proced := d.queue.RetrieveHeaders() 1245 return filled, proced, err 1246 } 1247 1248 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1249 // available peers, reserving a chunk of blocks for each, waiting for delivery 1250 // and also periodically checking for timeouts. 1251 func (d *Downloader) fetchBodies(from uint64) error { 1252 log.Debug("Downloading block bodies", "origin", from) 1253 1254 var ( 1255 deliver = func(packet dataPack) (int, error) { 1256 pack := packet.(*bodyPack) 1257 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1258 } 1259 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 1260 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1261 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 1262 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) } 1263 ) 1264 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1265 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies, 1266 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1267 1268 log.Debug("Block body download terminated", "err", err) 1269 return err 1270 } 1271 1272 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1273 // available peers, reserving a chunk of receipts for each, waiting for delivery 1274 // and also periodically checking for timeouts. 1275 func (d *Downloader) fetchReceipts(from uint64) error { 1276 log.Debug("Downloading transaction receipts", "origin", from) 1277 1278 var ( 1279 deliver = func(packet dataPack) (int, error) { 1280 pack := packet.(*receiptPack) 1281 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1282 } 1283 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1284 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1285 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1286 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1287 p.SetReceiptsIdle(accepted, deliveryTime) 1288 } 1289 ) 1290 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1291 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts, 1292 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1293 1294 log.Debug("Transaction receipt download terminated", "err", err) 1295 return err 1296 } 1297 1298 // fetchParts iteratively downloads scheduled block parts, taking any available 1299 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1300 // also periodically checking for timeouts. 1301 // 1302 // As the scheduling/timeout logic mostly is the same for all downloaded data 1303 // types, this method is used by each for data gathering and is instrumented with 1304 // various callbacks to handle the slight differences between processing them. 1305 // 1306 // The instrumentation parameters: 1307 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1308 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1309 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1310 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1311 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1312 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1313 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1314 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1315 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1316 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1317 // - fetch: network callback to actually send a particular download request to a physical remote peer 1318 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1319 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1320 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1321 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1322 // - kind: textual label of the type being downloaded to display in log messages 1323 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1324 expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool), 1325 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1326 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error { 1327 1328 // Create a ticker to detect expired retrieval tasks 1329 ticker := time.NewTicker(100 * time.Millisecond) 1330 defer ticker.Stop() 1331 1332 update := make(chan struct{}, 1) 1333 1334 // Prepare the queue and fetch block parts until the block header fetcher's done 1335 finished := false 1336 for { 1337 select { 1338 case <-d.cancelCh: 1339 return errCanceled 1340 1341 case packet := <-deliveryCh: 1342 deliveryTime := time.Now() 1343 // If the peer was previously banned and failed to deliver its pack 1344 // in a reasonable time frame, ignore its message. 1345 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1346 // Deliver the received chunk of data and check chain validity 1347 accepted, err := deliver(packet) 1348 if errors.Is(err, errInvalidChain) { 1349 return err 1350 } 1351 // Unless a peer delivered something completely else than requested (usually 1352 // caused by a timed out request which came through in the end), set it to 1353 // idle. If the delivery's stale, the peer should have already been idled. 1354 if !errors.Is(err, errStaleDelivery) { 1355 setIdle(peer, accepted, deliveryTime) 1356 } 1357 // Issue a log to the user to see what's going on 1358 switch { 1359 case err == nil && packet.Items() == 0: 1360 peer.log.Trace("Requested data not delivered", "type", kind) 1361 case err == nil: 1362 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1363 default: 1364 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1365 } 1366 } 1367 // Blocks assembled, try to update the progress 1368 select { 1369 case update <- struct{}{}: 1370 default: 1371 } 1372 1373 case cont := <-wakeCh: 1374 // The header fetcher sent a continuation flag, check if it's done 1375 if !cont { 1376 finished = true 1377 } 1378 // Headers arrive, try to update the progress 1379 select { 1380 case update <- struct{}{}: 1381 default: 1382 } 1383 1384 case <-ticker.C: 1385 // Sanity check update the progress 1386 select { 1387 case update <- struct{}{}: 1388 default: 1389 } 1390 1391 case <-update: 1392 // Short circuit if we lost all our peers 1393 if d.peers.Len() == 0 { 1394 return errNoPeers 1395 } 1396 // Check for fetch request timeouts and demote the responsible peers 1397 for pid, fails := range expire() { 1398 if peer := d.peers.Peer(pid); peer != nil { 1399 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1400 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1401 // out that sync wise we need to get rid of the peer. 1402 // 1403 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1404 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1405 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1406 if fails > 2 { 1407 peer.log.Trace("Data delivery timed out", "type", kind) 1408 setIdle(peer, 0, time.Now()) 1409 } else { 1410 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1411 1412 if d.dropPeer == nil { 1413 // The dropPeer method is nil when `--copydb` is used for a local copy. 1414 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1415 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1416 } else { 1417 d.dropPeer(pid) 1418 1419 // If this peer was the master peer, abort sync immediately 1420 d.cancelLock.RLock() 1421 master := pid == d.cancelPeer 1422 d.cancelLock.RUnlock() 1423 1424 if master { 1425 d.cancel() 1426 return errTimeout 1427 } 1428 } 1429 } 1430 } 1431 } 1432 // If there's nothing more to fetch, wait or terminate 1433 if pending() == 0 { 1434 if !inFlight() && finished { 1435 log.Debug("Data fetching completed", "type", kind) 1436 return nil 1437 } 1438 break 1439 } 1440 // Send a download request to all idle peers, until throttled 1441 progressed, throttled, running := false, false, inFlight() 1442 idles, total := idle() 1443 pendCount := pending() 1444 for _, peer := range idles { 1445 // Short circuit if throttling activated 1446 if throttled { 1447 break 1448 } 1449 // Short circuit if there is no more available task. 1450 if pendCount = pending(); pendCount == 0 { 1451 break 1452 } 1453 // Reserve a chunk of fetches for a peer. A nil can mean either that 1454 // no more headers are available, or that the peer is known not to 1455 // have them. 1456 request, progress, throttle := reserve(peer, capacity(peer)) 1457 if progress { 1458 progressed = true 1459 } 1460 if throttle { 1461 throttled = true 1462 throttleCounter.Inc(1) 1463 } 1464 if request == nil { 1465 continue 1466 } 1467 if request.From > 0 { 1468 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1469 } else { 1470 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1471 } 1472 // Fetch the chunk and make sure any errors return the hashes to the queue 1473 if fetchHook != nil { 1474 fetchHook(request.Headers) 1475 } 1476 if err := fetch(peer, request); err != nil { 1477 // Although we could try and make an attempt to fix this, this error really 1478 // means that we've double allocated a fetch task to a peer. If that is the 1479 // case, the internal state of the downloader and the queue is very wrong so 1480 // better hard crash and note the error instead of silently accumulating into 1481 // a much bigger issue. 1482 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1483 } 1484 running = true 1485 } 1486 // Make sure that we have peers available for fetching. If all peers have been tried 1487 // and all failed throw an error 1488 if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 { 1489 return errPeersUnavailable 1490 } 1491 } 1492 } 1493 } 1494 1495 // processHeaders takes batches of retrieved headers from an input channel and 1496 // keeps processing and scheduling them into the header chain and downloader's 1497 // queue until the stream ends or a failure occurs. 1498 func (d *Downloader) processHeaders(origin uint64, td *big.Int) error { 1499 // Keep a count of uncertain headers to roll back 1500 var ( 1501 rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis) 1502 rollbackErr error 1503 mode = d.getMode() 1504 ) 1505 defer func() { 1506 if rollback > 0 { 1507 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1508 if mode != LightSync { 1509 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1510 lastBlock = d.blockchain.CurrentBlock().Number() 1511 } 1512 if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block 1513 // We're already unwinding the stack, only print the error to make it more visible 1514 log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err) 1515 } 1516 curFastBlock, curBlock := common.Big0, common.Big0 1517 if mode != LightSync { 1518 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1519 curBlock = d.blockchain.CurrentBlock().Number() 1520 } 1521 log.Warn("Rolled back chain segment", 1522 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1523 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1524 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr) 1525 } 1526 }() 1527 // Wait for batches of headers to process 1528 gotHeaders := false 1529 1530 for { 1531 select { 1532 case <-d.cancelCh: 1533 rollbackErr = errCanceled 1534 return errCanceled 1535 1536 case headers := <-d.headerProcCh: 1537 // Terminate header processing if we synced up 1538 if len(headers) == 0 { 1539 // Notify everyone that headers are fully processed 1540 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1541 select { 1542 case ch <- false: 1543 case <-d.cancelCh: 1544 } 1545 } 1546 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1547 // better chain compared to ours. The only exception is if its promised blocks were 1548 // already imported by other means (e.g. fetcher): 1549 // 1550 // R <remote peer>, L <local node>: Both at block 10 1551 // R: Mine block 11, and propagate it to L 1552 // L: Queue block 11 for import 1553 // L: Notice that R's head and TD increased compared to ours, start sync 1554 // L: Import of block 11 finishes 1555 // L: Sync begins, and finds common ancestor at 11 1556 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1557 // R: Nothing to give 1558 if mode != LightSync { 1559 head := d.blockchain.CurrentBlock() 1560 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1561 return errStallingPeer 1562 } 1563 } 1564 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1565 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1566 // of delivering the post-pivot blocks that would flag the invalid content. 1567 // 1568 // This check cannot be executed "as is" for full imports, since blocks may still be 1569 // queued for processing when the header download completes. However, as long as the 1570 // peer gave us something useful, we're already happy/progressed (above check). 1571 if mode == FastSync || mode == LightSync { 1572 head := d.lightchain.CurrentHeader() 1573 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1574 return errStallingPeer 1575 } 1576 } 1577 // Disable any rollback and return 1578 rollback = 0 1579 return nil 1580 } 1581 // Otherwise split the chunk of headers into batches and process them 1582 gotHeaders = true 1583 for len(headers) > 0 { 1584 // Terminate if something failed in between processing chunks 1585 select { 1586 case <-d.cancelCh: 1587 rollbackErr = errCanceled 1588 return errCanceled 1589 default: 1590 } 1591 // Select the next chunk of headers to import 1592 limit := maxHeadersProcess 1593 if limit > len(headers) { 1594 limit = len(headers) 1595 } 1596 chunk := headers[:limit] 1597 1598 // In case of header only syncing, validate the chunk immediately 1599 if mode == FastSync || mode == LightSync { 1600 // If we're importing pure headers, verify based on their recentness 1601 var pivot uint64 1602 1603 d.pivotLock.RLock() 1604 if d.pivotHeader != nil { 1605 pivot = d.pivotHeader.Number.Uint64() 1606 } 1607 d.pivotLock.RUnlock() 1608 1609 frequency := fsHeaderCheckFrequency 1610 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1611 frequency = 1 1612 } 1613 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1614 rollbackErr = err 1615 1616 // If some headers were inserted, track them as uncertain 1617 if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 { 1618 rollback = chunk[0].Number.Uint64() 1619 } 1620 log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err) 1621 return fmt.Errorf("%w: %v", errInvalidChain, err) 1622 } 1623 // All verifications passed, track all headers within the alloted limits 1624 if mode == FastSync { 1625 head := chunk[len(chunk)-1].Number.Uint64() 1626 if head-rollback > uint64(fsHeaderSafetyNet) { 1627 rollback = head - uint64(fsHeaderSafetyNet) 1628 } else { 1629 rollback = 1 1630 } 1631 } 1632 } 1633 // Unless we're doing light chains, schedule the headers for associated content retrieval 1634 if mode == FullSync || mode == FastSync { 1635 // If we've reached the allowed number of pending headers, stall a bit 1636 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1637 select { 1638 case <-d.cancelCh: 1639 rollbackErr = errCanceled 1640 return errCanceled 1641 case <-time.After(time.Second): 1642 } 1643 } 1644 // Otherwise insert the headers for content retrieval 1645 inserts := d.queue.Schedule(chunk, origin) 1646 if len(inserts) != len(chunk) { 1647 rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk)) 1648 return fmt.Errorf("%w: stale headers", errBadPeer) 1649 } 1650 } 1651 headers = headers[limit:] 1652 origin += uint64(limit) 1653 } 1654 // Update the highest block number we know if a higher one is found. 1655 d.syncStatsLock.Lock() 1656 if d.syncStatsChainHeight < origin { 1657 d.syncStatsChainHeight = origin - 1 1658 } 1659 d.syncStatsLock.Unlock() 1660 1661 // Signal the content downloaders of the availablility of new tasks 1662 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1663 select { 1664 case ch <- true: 1665 default: 1666 } 1667 } 1668 } 1669 } 1670 } 1671 1672 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1673 func (d *Downloader) processFullSyncContent() error { 1674 for { 1675 results := d.queue.Results(true) 1676 if len(results) == 0 { 1677 return nil 1678 } 1679 if d.chainInsertHook != nil { 1680 d.chainInsertHook(results) 1681 } 1682 if err := d.importBlockResults(results); err != nil { 1683 return err 1684 } 1685 } 1686 } 1687 1688 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1689 // Check for any early termination requests 1690 if len(results) == 0 { 1691 return nil 1692 } 1693 select { 1694 case <-d.quitCh: 1695 return errCancelContentProcessing 1696 default: 1697 } 1698 // Retrieve the a batch of results to import 1699 first, last := results[0].Header, results[len(results)-1].Header 1700 log.Debug("Inserting downloaded chain", "items", len(results), 1701 "firstnum", first.Number, "firsthash", first.Hash(), 1702 "lastnum", last.Number, "lasthash", last.Hash(), 1703 ) 1704 blocks := make([]*types.Block, len(results)) 1705 for i, result := range results { 1706 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1707 } 1708 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1709 if index < len(results) { 1710 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1711 } else { 1712 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1713 // when it needs to preprocess blocks to import a sidechain. 1714 // The importer will put together a new list of blocks to import, which is a superset 1715 // of the blocks delivered from the downloader, and the indexing will be off. 1716 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1717 } 1718 return fmt.Errorf("%w: %v", errInvalidChain, err) 1719 } 1720 return nil 1721 } 1722 1723 // processFastSyncContent takes fetch results from the queue and writes them to the 1724 // database. It also controls the synchronisation of state nodes of the pivot block. 1725 func (d *Downloader) processFastSyncContent() error { 1726 // Start syncing state of the reported head block. This should get us most of 1727 // the state of the pivot block. 1728 d.pivotLock.RLock() 1729 sync := d.syncState(d.pivotHeader.Root) 1730 d.pivotLock.RUnlock() 1731 1732 defer func() { 1733 // The `sync` object is replaced every time the pivot moves. We need to 1734 // defer close the very last active one, hence the lazy evaluation vs. 1735 // calling defer sync.Cancel() !!! 1736 sync.Cancel() 1737 }() 1738 1739 closeOnErr := func(s *stateSync) { 1740 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled { 1741 d.queue.Close() // wake up Results 1742 } 1743 } 1744 go closeOnErr(sync) 1745 1746 // To cater for moving pivot points, track the pivot block and subsequently 1747 // accumulated download results separately. 1748 var ( 1749 oldPivot *fetchResult // Locked in pivot block, might change eventually 1750 oldTail []*fetchResult // Downloaded content after the pivot 1751 ) 1752 for { 1753 // Wait for the next batch of downloaded data to be available, and if the pivot 1754 // block became stale, move the goalpost 1755 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1756 if len(results) == 0 { 1757 // If pivot sync is done, stop 1758 if oldPivot == nil { 1759 return sync.Cancel() 1760 } 1761 // If sync failed, stop 1762 select { 1763 case <-d.cancelCh: 1764 sync.Cancel() 1765 return errCanceled 1766 default: 1767 } 1768 } 1769 if d.chainInsertHook != nil { 1770 d.chainInsertHook(results) 1771 } 1772 // If we haven't downloaded the pivot block yet, check pivot staleness 1773 // notifications from the header downloader 1774 d.pivotLock.RLock() 1775 pivot := d.pivotHeader 1776 d.pivotLock.RUnlock() 1777 1778 if oldPivot == nil { 1779 if pivot.Root != sync.root { 1780 sync.Cancel() 1781 sync = d.syncState(pivot.Root) 1782 1783 go closeOnErr(sync) 1784 } 1785 } else { 1786 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1787 } 1788 // Split around the pivot block and process the two sides via fast/full sync 1789 if atomic.LoadInt32(&d.committed) == 0 { 1790 latest := results[len(results)-1].Header 1791 // If the height is above the pivot block by 2 sets, it means the pivot 1792 // become stale in the network and it was garbage collected, move to a 1793 // new pivot. 1794 // 1795 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 1796 // need to be taken into account, otherwise we're detecting the pivot move 1797 // late and will drop peers due to unavailable state!!! 1798 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 1799 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 1800 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 1801 1802 d.pivotLock.Lock() 1803 d.pivotHeader = pivot 1804 d.pivotLock.Unlock() 1805 1806 // Write out the pivot into the database so a rollback beyond it will 1807 // reenable fast sync 1808 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 1809 } 1810 } 1811 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 1812 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1813 return err 1814 } 1815 if P != nil { 1816 // If new pivot block found, cancel old state retrieval and restart 1817 if oldPivot != P { 1818 sync.Cancel() 1819 sync = d.syncState(P.Header.Root) 1820 1821 go closeOnErr(sync) 1822 oldPivot = P 1823 } 1824 // Wait for completion, occasionally checking for pivot staleness 1825 select { 1826 case <-sync.done: 1827 if sync.err != nil { 1828 return sync.err 1829 } 1830 if err := d.commitPivotBlock(P); err != nil { 1831 return err 1832 } 1833 oldPivot = nil 1834 1835 case <-time.After(time.Second): 1836 oldTail = afterP 1837 continue 1838 } 1839 } 1840 // Fast sync done, pivot commit done, full import 1841 if err := d.importBlockResults(afterP); err != nil { 1842 return err 1843 } 1844 } 1845 } 1846 1847 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1848 if len(results) == 0 { 1849 return nil, nil, nil 1850 } 1851 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 1852 // the pivot is somewhere in the future 1853 return nil, results, nil 1854 } 1855 // This can also be optimized, but only happens very seldom 1856 for _, result := range results { 1857 num := result.Header.Number.Uint64() 1858 switch { 1859 case num < pivot: 1860 before = append(before, result) 1861 case num == pivot: 1862 p = result 1863 default: 1864 after = append(after, result) 1865 } 1866 } 1867 return p, before, after 1868 } 1869 1870 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1871 // Check for any early termination requests 1872 if len(results) == 0 { 1873 return nil 1874 } 1875 select { 1876 case <-d.quitCh: 1877 return errCancelContentProcessing 1878 case <-stateSync.done: 1879 if err := stateSync.Wait(); err != nil { 1880 return err 1881 } 1882 default: 1883 } 1884 // Retrieve the a batch of results to import 1885 first, last := results[0].Header, results[len(results)-1].Header 1886 log.Debug("Inserting fast-sync blocks", "items", len(results), 1887 "firstnum", first.Number, "firsthash", first.Hash(), 1888 "lastnumn", last.Number, "lasthash", last.Hash(), 1889 ) 1890 blocks := make([]*types.Block, len(results)) 1891 receipts := make([]types.Receipts, len(results)) 1892 for i, result := range results { 1893 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1894 receipts[i] = result.Receipts 1895 } 1896 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1897 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1898 return fmt.Errorf("%w: %v", errInvalidChain, err) 1899 } 1900 return nil 1901 } 1902 1903 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1904 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1905 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1906 1907 // Commit the pivot block as the new head, will require full sync from here on 1908 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1909 return err 1910 } 1911 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1912 return err 1913 } 1914 atomic.StoreInt32(&d.committed, 1) 1915 1916 // If we had a bloom filter for the state sync, deallocate it now. Note, we only 1917 // deallocate internally, but keep the empty wrapper. This ensures that if we do 1918 // a rollback after committing the pivot and restarting fast sync, we don't end 1919 // up using a nil bloom. Empty bloom is fine, it just returns that it does not 1920 // have the info we need, so reach down to the database instead. 1921 if d.stateBloom != nil { 1922 d.stateBloom.Close() 1923 } 1924 return nil 1925 } 1926 1927 // DeliverHeaders injects a new batch of block headers received from a remote 1928 // node into the download schedule. 1929 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) error { 1930 return d.deliver(d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1931 } 1932 1933 // DeliverBodies injects a new batch of block bodies received from a remote node. 1934 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) error { 1935 return d.deliver(d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1936 } 1937 1938 // DeliverReceipts injects a new batch of receipts received from a remote node. 1939 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) error { 1940 return d.deliver(d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1941 } 1942 1943 // DeliverNodeData injects a new batch of node state data received from a remote node. 1944 func (d *Downloader) DeliverNodeData(id string, data [][]byte) error { 1945 return d.deliver(d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1946 } 1947 1948 // DeliverSnapPacket is invoked from a peer's message handler when it transmits a 1949 // data packet for the local node to consume. 1950 func (d *Downloader) DeliverSnapPacket(peer *snap.Peer, packet snap.Packet) error { 1951 switch packet := packet.(type) { 1952 case *snap.AccountRangePacket: 1953 hashes, accounts, err := packet.Unpack() 1954 if err != nil { 1955 return err 1956 } 1957 return d.SnapSyncer.OnAccounts(peer, packet.ID, hashes, accounts, packet.Proof) 1958 1959 case *snap.StorageRangesPacket: 1960 hashset, slotset := packet.Unpack() 1961 return d.SnapSyncer.OnStorage(peer, packet.ID, hashset, slotset, packet.Proof) 1962 1963 case *snap.ByteCodesPacket: 1964 return d.SnapSyncer.OnByteCodes(peer, packet.ID, packet.Codes) 1965 1966 case *snap.TrieNodesPacket: 1967 return d.SnapSyncer.OnTrieNodes(peer, packet.ID, packet.Nodes) 1968 1969 default: 1970 return fmt.Errorf("unexpected snap packet type: %T", packet) 1971 } 1972 } 1973 1974 // deliver injects a new batch of data received from a remote node. 1975 func (d *Downloader) deliver(destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1976 // Update the delivery metrics for both good and failed deliveries 1977 inMeter.Mark(int64(packet.Items())) 1978 defer func() { 1979 if err != nil { 1980 dropMeter.Mark(int64(packet.Items())) 1981 } 1982 }() 1983 // Deliver or abort if the sync is canceled while queuing 1984 d.cancelLock.RLock() 1985 cancel := d.cancelCh 1986 d.cancelLock.RUnlock() 1987 if cancel == nil { 1988 return errNoSyncActive 1989 } 1990 select { 1991 case destCh <- packet: 1992 return nil 1993 case <-cancel: 1994 return errNoSyncActive 1995 } 1996 } 1997 1998 // qosTuner is the quality of service tuning loop that occasionally gathers the 1999 // peer latency statistics and updates the estimated request round trip time. 2000 func (d *Downloader) qosTuner() { 2001 for { 2002 // Retrieve the current median RTT and integrate into the previoust target RTT 2003 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 2004 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 2005 2006 // A new RTT cycle passed, increase our confidence in the estimated RTT 2007 conf := atomic.LoadUint64(&d.rttConfidence) 2008 conf = conf + (1000000-conf)/2 2009 atomic.StoreUint64(&d.rttConfidence, conf) 2010 2011 // Log the new QoS values and sleep until the next RTT 2012 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 2013 select { 2014 case <-d.quitCh: 2015 return 2016 case <-time.After(rtt): 2017 } 2018 } 2019 } 2020 2021 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 2022 // peer set, needing to reduce the confidence we have in out QoS estimates. 2023 func (d *Downloader) qosReduceConfidence() { 2024 // If we have a single peer, confidence is always 1 2025 peers := uint64(d.peers.Len()) 2026 if peers == 0 { 2027 // Ensure peer connectivity races don't catch us off guard 2028 return 2029 } 2030 if peers == 1 { 2031 atomic.StoreUint64(&d.rttConfidence, 1000000) 2032 return 2033 } 2034 // If we have a ton of peers, don't drop confidence) 2035 if peers >= uint64(qosConfidenceCap) { 2036 return 2037 } 2038 // Otherwise drop the confidence factor 2039 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 2040 if float64(conf)/1000000 < rttMinConfidence { 2041 conf = uint64(rttMinConfidence * 1000000) 2042 } 2043 atomic.StoreUint64(&d.rttConfidence, conf) 2044 2045 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2046 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 2047 } 2048 2049 // requestRTT returns the current target round trip time for a download request 2050 // to complete in. 2051 // 2052 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 2053 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 2054 // be adapted to, but smaller ones are preferred (stabler download stream). 2055 func (d *Downloader) requestRTT() time.Duration { 2056 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 2057 } 2058 2059 // requestTTL returns the current timeout allowance for a single download request 2060 // to finish under. 2061 func (d *Downloader) requestTTL() time.Duration { 2062 var ( 2063 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2064 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 2065 ) 2066 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 2067 if ttl > ttlLimit { 2068 ttl = ttlLimit 2069 } 2070 return ttl 2071 }