github.com/hyperion-hyn/go-ethereum@v2.4.0+incompatible/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 ethereum "github.com/ethereum/go-ethereum" 29 "github.com/ethereum/go-ethereum/common" 30 "github.com/ethereum/go-ethereum/core/rawdb" 31 "github.com/ethereum/go-ethereum/core/types" 32 "github.com/ethereum/go-ethereum/ethdb" 33 "github.com/ethereum/go-ethereum/event" 34 "github.com/ethereum/go-ethereum/log" 35 "github.com/ethereum/go-ethereum/metrics" 36 "github.com/ethereum/go-ethereum/params" 37 ) 38 39 var ( 40 MaxHashFetch = 512 // Amount of hashes to be fetched per retrieval request 41 MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request 42 MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request 43 MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly 44 MaxBodyFetch = 128 // Amount of block bodies to be fetched per retrieval request 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 MaxForkAncestry = 3 * params.EpochDuration // Maximum chain reorganisation 49 rttMinEstimate = 2 * time.Second // Minimum round-trip time to target for download requests 50 rttMaxEstimate = 20 * time.Second // Maximum round-trip time to target for download requests 51 rttMinConfidence = 0.1 // Worse confidence factor in our estimated RTT value 52 ttlScaling = 3 // Constant scaling factor for RTT -> TTL conversion 53 ttlLimit = time.Minute // Maximum TTL allowance to prevent reaching crazy timeouts 54 55 qosTuningPeers = 5 // Number of peers to tune based on (best peers) 56 qosConfidenceCap = 10 // Number of peers above which not to modify RTT confidence 57 qosTuningImpact = 0.25 // Impact that a new tuning target has on the previous value 58 59 maxQueuedHeaders = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection) 60 maxHeadersProcess = 2048 // Number of header download results to import at once into the chain 61 maxResultsProcess = 2048 // Number of content download results to import at once into the chain 62 63 reorgProtThreshold = 48 // Threshold number of recent blocks to disable mini reorg protection 64 reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs 65 66 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 67 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 68 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 69 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 70 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 71 ) 72 73 var ( 74 errBusy = errors.New("busy") 75 errUnknownPeer = errors.New("peer is unknown or unhealthy") 76 errBadPeer = errors.New("action from bad peer ignored") 77 errStallingPeer = errors.New("peer is stalling") 78 errNoPeers = errors.New("no peers to keep download active") 79 errTimeout = errors.New("timeout") 80 errEmptyHeaderSet = errors.New("empty header set by peer") 81 errPeersUnavailable = errors.New("no peers available or all tried for download") 82 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 83 errInvalidChain = errors.New("retrieved hash chain is invalid") 84 errInvalidBlock = errors.New("retrieved block is invalid") 85 errInvalidBody = errors.New("retrieved block body is invalid") 86 errInvalidReceipt = errors.New("retrieved receipt is invalid") 87 errCancelBlockFetch = errors.New("block download canceled (requested)") 88 errCancelHeaderFetch = errors.New("block header download canceled (requested)") 89 errCancelBodyFetch = errors.New("block body download canceled (requested)") 90 errCancelReceiptFetch = errors.New("receipt download canceled (requested)") 91 errCancelStateFetch = errors.New("state data download canceled (requested)") 92 errCancelHeaderProcessing = errors.New("header processing canceled (requested)") 93 errCancelContentProcessing = errors.New("content processing canceled (requested)") 94 errNoSyncActive = errors.New("no sync active") 95 errTooOld = errors.New("peer doesn't speak recent enough protocol version (need version >= 62)") 96 ) 97 98 type Downloader struct { 99 mode SyncMode // Synchronisation mode defining the strategy used (per sync cycle) 100 mux *event.TypeMux // Event multiplexer to announce sync operation events 101 102 queue *queue // Scheduler for selecting the hashes to download 103 peers *peerSet // Set of active peers from which download can proceed 104 stateDB ethdb.Database 105 106 rttEstimate uint64 // Round trip time to target for download requests 107 rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops) 108 109 // Statistics 110 syncStatsChainOrigin uint64 // Origin block number where syncing started at 111 syncStatsChainHeight uint64 // Highest block number known when syncing started 112 syncStatsState stateSyncStats 113 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 114 115 lightchain LightChain 116 blockchain BlockChain 117 118 // Callbacks 119 dropPeer peerDropFn // Drops a peer for misbehaving 120 121 // Status 122 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 123 synchronising int32 124 notified int32 125 committed int32 126 127 // Channels 128 headerCh chan dataPack // [eth/62] Channel receiving inbound block headers 129 bodyCh chan dataPack // [eth/62] Channel receiving inbound block bodies 130 receiptCh chan dataPack // [eth/63] Channel receiving inbound receipts 131 bodyWakeCh chan bool // [eth/62] Channel to signal the block body fetcher of new tasks 132 receiptWakeCh chan bool // [eth/63] Channel to signal the receipt fetcher of new tasks 133 headerProcCh chan []*types.Header // [eth/62] Channel to feed the header processor new tasks 134 135 // for stateFetcher 136 stateSyncStart chan *stateSync 137 trackStateReq chan *stateReq 138 stateCh chan dataPack // [eth/63] Channel receiving inbound node state data 139 140 // Cancellation and termination 141 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 142 cancelCh chan struct{} // Channel to cancel mid-flight syncs 143 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 144 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 145 146 quitCh chan struct{} // Quit channel to signal termination 147 quitLock sync.RWMutex // Lock to prevent double closes 148 149 // Testing hooks 150 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 151 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 152 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 153 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 154 } 155 156 // LightChain encapsulates functions required to synchronise a light chain. 157 type LightChain interface { 158 // HasHeader verifies a header's presence in the local chain. 159 HasHeader(common.Hash, uint64) bool 160 161 // GetHeaderByHash retrieves a header from the local chain. 162 GetHeaderByHash(common.Hash) *types.Header 163 164 // CurrentHeader retrieves the head header from the local chain. 165 CurrentHeader() *types.Header 166 167 // GetTd returns the total difficulty of a local block. 168 GetTd(common.Hash, uint64) *big.Int 169 170 // InsertHeaderChain inserts a batch of headers into the local chain. 171 InsertHeaderChain([]*types.Header, int) (int, error) 172 173 // Rollback removes a few recently added elements from the local chain. 174 Rollback([]common.Hash) 175 } 176 177 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 178 type BlockChain interface { 179 LightChain 180 181 // HasBlock verifies a block's presence in the local chain. 182 HasBlock(common.Hash, uint64) bool 183 184 // GetBlockByHash retrieves a block from the local chain. 185 GetBlockByHash(common.Hash) *types.Block 186 187 // CurrentBlock retrieves the head block from the local chain. 188 CurrentBlock() *types.Block 189 190 // CurrentFastBlock retrieves the head fast block from the local chain. 191 CurrentFastBlock() *types.Block 192 193 // FastSyncCommitHead directly commits the head block to a certain entity. 194 FastSyncCommitHead(common.Hash) error 195 196 // InsertChain inserts a batch of blocks into the local chain. 197 InsertChain(types.Blocks) (int, error) 198 199 // InsertReceiptChain inserts a batch of receipts into the local chain. 200 InsertReceiptChain(types.Blocks, []types.Receipts) (int, error) 201 } 202 203 // New creates a new downloader to fetch hashes and blocks from remote peers. 204 func New(mode SyncMode, stateDb ethdb.Database, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 205 if lightchain == nil { 206 lightchain = chain 207 } 208 209 dl := &Downloader{ 210 mode: mode, 211 stateDB: stateDb, 212 mux: mux, 213 queue: newQueue(), 214 peers: newPeerSet(), 215 rttEstimate: uint64(rttMaxEstimate), 216 rttConfidence: uint64(1000000), 217 blockchain: chain, 218 lightchain: lightchain, 219 dropPeer: dropPeer, 220 headerCh: make(chan dataPack, 1), 221 bodyCh: make(chan dataPack, 1), 222 receiptCh: make(chan dataPack, 1), 223 bodyWakeCh: make(chan bool, 1), 224 receiptWakeCh: make(chan bool, 1), 225 headerProcCh: make(chan []*types.Header, 1), 226 quitCh: make(chan struct{}), 227 stateCh: make(chan dataPack), 228 stateSyncStart: make(chan *stateSync), 229 syncStatsState: stateSyncStats{ 230 processed: rawdb.ReadFastTrieProgress(stateDb), 231 }, 232 trackStateReq: make(chan *stateReq), 233 } 234 go dl.qosTuner() 235 go dl.stateFetcher() 236 return dl 237 } 238 239 // Progress retrieves the synchronisation boundaries, specifically the origin 240 // block where synchronisation started at (may have failed/suspended); the block 241 // or header sync is currently at; and the latest known block which the sync targets. 242 // 243 // In addition, during the state download phase of fast synchronisation the number 244 // of processed and the total number of known states are also returned. Otherwise 245 // these are zero. 246 func (d *Downloader) Progress() ethereum.SyncProgress { 247 // Lock the current stats and return the progress 248 d.syncStatsLock.RLock() 249 defer d.syncStatsLock.RUnlock() 250 251 current := uint64(0) 252 switch d.mode { 253 case FullSync: 254 current = d.blockchain.CurrentBlock().NumberU64() 255 case FastSync: 256 current = d.blockchain.CurrentFastBlock().NumberU64() 257 case LightSync: 258 current = d.lightchain.CurrentHeader().Number.Uint64() 259 } 260 return ethereum.SyncProgress{ 261 StartingBlock: d.syncStatsChainOrigin, 262 CurrentBlock: current, 263 HighestBlock: d.syncStatsChainHeight, 264 PulledStates: d.syncStatsState.processed, 265 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 266 } 267 } 268 269 // Synchronising returns whether the downloader is currently retrieving blocks. 270 func (d *Downloader) Synchronising() bool { 271 return atomic.LoadInt32(&d.synchronising) > 0 272 } 273 274 // RegisterPeer injects a new download peer into the set of block source to be 275 // used for fetching hashes and blocks from. 276 func (d *Downloader) RegisterPeer(id string, version int, peer Peer) error { 277 logger := log.New("peer", id) 278 logger.Trace("Registering sync peer") 279 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 280 logger.Error("Failed to register sync peer", "err", err) 281 return err 282 } 283 d.qosReduceConfidence() 284 285 return nil 286 } 287 288 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 289 func (d *Downloader) RegisterLightPeer(id string, version int, peer LightPeer) error { 290 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 291 } 292 293 // UnregisterPeer remove a peer from the known list, preventing any action from 294 // the specified peer. An effort is also made to return any pending fetches into 295 // the queue. 296 func (d *Downloader) UnregisterPeer(id string) error { 297 // Unregister the peer from the active peer set and revoke any fetch tasks 298 logger := log.New("peer", id) 299 logger.Trace("Unregistering sync peer") 300 if err := d.peers.Unregister(id); err != nil { 301 logger.Error("Failed to unregister sync peer", "err", err) 302 return err 303 } 304 d.queue.Revoke(id) 305 306 // If this peer was the master peer, abort sync immediately 307 d.cancelLock.RLock() 308 master := id == d.cancelPeer 309 d.cancelLock.RUnlock() 310 311 if master { 312 d.cancel() 313 } 314 return nil 315 } 316 317 // Synchronise tries to sync up our local block chain with a remote peer, both 318 // adding various sanity checks as well as wrapping it with various log entries. 319 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 320 err := d.synchronise(id, head, td, mode) 321 switch err { 322 case nil: 323 case errBusy: 324 325 case errTimeout, errBadPeer, errStallingPeer, 326 errEmptyHeaderSet, errPeersUnavailable, errTooOld, 327 errInvalidAncestor, errInvalidChain: 328 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 329 if d.dropPeer == nil { 330 // The dropPeer method is nil when `--copydb` is used for a local copy. 331 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 332 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 333 } else { 334 d.dropPeer(id) 335 } 336 default: 337 log.Warn("Synchronisation failed, retrying", "err", err) 338 } 339 return err 340 } 341 342 // synchronise will select the peer and use it for synchronising. If an empty string is given 343 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 344 // checks fail an error will be returned. This method is synchronous 345 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 346 // Mock out the synchronisation if testing 347 if d.synchroniseMock != nil { 348 return d.synchroniseMock(id, hash) 349 } 350 // Make sure only one goroutine is ever allowed past this point at once 351 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 352 return errBusy 353 } 354 // changes for permissions. added set sync status to indicate permisssions that node sync has started 355 types.SetSyncStatus() 356 defer atomic.StoreInt32(&d.synchronising, 0) 357 358 // Post a user notification of the sync (only once per session) 359 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 360 log.Info("Block synchronisation started") 361 } 362 // Reset the queue, peer set and wake channels to clean any internal leftover state 363 d.queue.Reset() 364 d.peers.Reset() 365 366 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 367 select { 368 case <-ch: 369 default: 370 } 371 } 372 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 373 for empty := false; !empty; { 374 select { 375 case <-ch: 376 default: 377 empty = true 378 } 379 } 380 } 381 for empty := false; !empty; { 382 select { 383 case <-d.headerProcCh: 384 default: 385 empty = true 386 } 387 } 388 // Create cancel channel for aborting mid-flight and mark the master peer 389 d.cancelLock.Lock() 390 d.cancelCh = make(chan struct{}) 391 d.cancelPeer = id 392 d.cancelLock.Unlock() 393 394 defer d.Cancel() // No matter what, we can't leave the cancel channel open 395 396 // Set the requested sync mode, unless it's forbidden 397 d.mode = mode 398 399 // Retrieve the origin peer and initiate the downloading process 400 p := d.peers.Peer(id) 401 if p == nil { 402 return errUnknownPeer 403 } 404 if d.mode == BoundedFullSync { 405 return d.syncWithPeerUntil(p, hash, td) 406 } 407 return d.syncWithPeer(p, hash, td) 408 } 409 410 // syncWithPeer starts a block synchronization based on the hash chain from the 411 // specified peer and head hash. 412 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 413 d.mux.Post(StartEvent{}) 414 defer func() { 415 // reset on error 416 if err != nil { 417 d.mux.Post(FailedEvent{err}) 418 } else { 419 d.mux.Post(DoneEvent{}) 420 } 421 }() 422 if p.version < 62 { 423 return errTooOld 424 } 425 426 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", d.mode) 427 defer func(start time.Time) { 428 log.Debug("Synchronisation terminated", "elapsed", time.Since(start)) 429 }(time.Now()) 430 431 // Look up the sync boundaries: the common ancestor and the target block 432 latest, err := d.fetchHeight(p) 433 if err != nil { 434 return err 435 } 436 height := latest.Number.Uint64() 437 438 origin, err := d.findAncestor(p, height) 439 if err != nil { 440 return err 441 } 442 d.syncStatsLock.Lock() 443 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 444 d.syncStatsChainOrigin = origin 445 } 446 d.syncStatsChainHeight = height 447 d.syncStatsLock.Unlock() 448 449 // Ensure our origin point is below any fast sync pivot point 450 pivot := uint64(0) 451 if d.mode == FastSync { 452 if height <= uint64(fsMinFullBlocks) { 453 origin = 0 454 } else { 455 pivot = height - uint64(fsMinFullBlocks) 456 if pivot <= origin { 457 origin = pivot - 1 458 } 459 } 460 } 461 d.committed = 1 462 if d.mode == FastSync && pivot != 0 { 463 d.committed = 0 464 } 465 // Initiate the sync using a concurrent header and content retrieval algorithm 466 d.queue.Prepare(origin+1, d.mode) 467 if d.syncInitHook != nil { 468 d.syncInitHook(origin, height) 469 } 470 471 fetchers := []func() error{ 472 func() error { return d.fetchHeaders(p, origin+1, pivot) }, // Headers are always retrieved 473 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 474 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 475 func() error { return d.processHeaders(origin+1, pivot, td) }, 476 } 477 if d.mode == FastSync { 478 fetchers = append(fetchers, func() error { return d.processFastSyncContent(latest) }) 479 } else if d.mode == FullSync { 480 fetchers = append(fetchers, d.processFullSyncContent) 481 } 482 return d.spawnSync(fetchers) 483 } 484 485 // spawnSync runs d.process and all given fetcher functions to completion in 486 // separate goroutines, returning the first error that appears. 487 func (d *Downloader) spawnSync(fetchers []func() error) error { 488 errc := make(chan error, len(fetchers)) 489 d.cancelWg.Add(len(fetchers)) 490 for _, fn := range fetchers { 491 fn := fn 492 go func() { defer d.cancelWg.Done(); errc <- fn() }() 493 } 494 // Wait for the first error, then terminate the others. 495 var err error 496 for i := 0; i < len(fetchers); i++ { 497 if i == len(fetchers)-1 { 498 // Close the queue when all fetchers have exited. 499 // This will cause the block processor to end when 500 // it has processed the queue. 501 d.queue.Close() 502 } 503 if err = <-errc; err != nil { 504 break 505 } 506 } 507 d.queue.Close() 508 d.Cancel() 509 return err 510 } 511 512 // cancel aborts all of the operations and resets the queue. However, cancel does 513 // not wait for the running download goroutines to finish. This method should be 514 // used when cancelling the downloads from inside the downloader. 515 func (d *Downloader) cancel() { 516 // Close the current cancel channel 517 d.cancelLock.Lock() 518 if d.cancelCh != nil { 519 select { 520 case <-d.cancelCh: 521 // Channel was already closed 522 default: 523 close(d.cancelCh) 524 } 525 } 526 d.cancelLock.Unlock() 527 } 528 529 // Cancel aborts all of the operations and waits for all download goroutines to 530 // finish before returning. 531 func (d *Downloader) Cancel() { 532 d.cancel() 533 d.cancelWg.Wait() 534 } 535 536 // Terminate interrupts the downloader, canceling all pending operations. 537 // The downloader cannot be reused after calling Terminate. 538 func (d *Downloader) Terminate() { 539 // Close the termination channel (make sure double close is allowed) 540 d.quitLock.Lock() 541 select { 542 case <-d.quitCh: 543 default: 544 close(d.quitCh) 545 } 546 d.quitLock.Unlock() 547 548 // Cancel any pending download requests 549 d.Cancel() 550 } 551 552 // fetchHeight retrieves the head header of the remote peer to aid in estimating 553 // the total time a pending synchronisation would take. 554 func (d *Downloader) fetchHeight(p *peerConnection) (*types.Header, error) { 555 p.log.Debug("Retrieving remote chain height") 556 557 // Request the advertised remote head block and wait for the response 558 head, _ := p.peer.Head() 559 go p.peer.RequestHeadersByHash(head, 1, 0, false) 560 561 ttl := d.requestTTL() 562 timeout := time.After(ttl) 563 for { 564 select { 565 case <-d.cancelCh: 566 return nil, errCancelBlockFetch 567 568 case packet := <-d.headerCh: 569 // Discard anything not from the origin peer 570 if packet.PeerId() != p.id { 571 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 572 break 573 } 574 // Make sure the peer actually gave something valid 575 headers := packet.(*headerPack).headers 576 if len(headers) != 1 { 577 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 578 return nil, errBadPeer 579 } 580 head := headers[0] 581 p.log.Debug("Remote head header identified", "number", head.Number, "hash", head.Hash()) 582 return head, nil 583 584 case <-timeout: 585 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 586 return nil, errTimeout 587 588 case <-d.bodyCh: 589 case <-d.receiptCh: 590 // Out of bounds delivery, ignore 591 } 592 } 593 } 594 595 // findAncestor tries to locate the common ancestor link of the local chain and 596 // a remote peers blockchain. In the general case when our node was in sync and 597 // on the correct chain, checking the top N links should already get us a match. 598 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 599 // the head links match), we do a binary search to find the common ancestor. 600 func (d *Downloader) findAncestor(p *peerConnection, height uint64) (uint64, error) { 601 // Figure out the valid ancestor range to prevent rewrite attacks 602 floor, ceil := int64(-1), d.lightchain.CurrentHeader().Number.Uint64() 603 604 if d.mode == FullSync { 605 ceil = d.blockchain.CurrentBlock().NumberU64() 606 } else if d.mode == FastSync { 607 ceil = d.blockchain.CurrentFastBlock().NumberU64() 608 } 609 if ceil >= MaxForkAncestry { 610 floor = int64(ceil - MaxForkAncestry) 611 } 612 p.log.Debug("Looking for common ancestor", "local", ceil, "remote", height) 613 614 // Request the topmost blocks to short circuit binary ancestor lookup 615 head := ceil 616 if head > height { 617 head = height 618 } 619 from := int64(head) - int64(MaxHeaderFetch) 620 if from < 0 { 621 from = 0 622 } 623 // Span out with 15 block gaps into the future to catch bad head reports 624 limit := 2 * MaxHeaderFetch / 16 625 count := 1 + int((int64(ceil)-from)/16) 626 if count > limit { 627 count = limit 628 } 629 go p.peer.RequestHeadersByNumber(uint64(from), count, 15, false) 630 631 // Wait for the remote response to the head fetch 632 number, hash := uint64(0), common.Hash{} 633 634 ttl := d.requestTTL() 635 timeout := time.After(ttl) 636 637 for finished := false; !finished; { 638 select { 639 case <-d.cancelCh: 640 return 0, errCancelHeaderFetch 641 642 case packet := <-d.headerCh: 643 // Discard anything not from the origin peer 644 if packet.PeerId() != p.id { 645 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 646 break 647 } 648 // Make sure the peer actually gave something valid 649 headers := packet.(*headerPack).headers 650 if len(headers) == 0 { 651 p.log.Warn("Empty head header set") 652 return 0, errEmptyHeaderSet 653 } 654 // Make sure the peer's reply conforms to the request 655 for i := 0; i < len(headers); i++ { 656 if number := headers[i].Number.Int64(); number != from+int64(i)*16 { 657 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", from+int64(i)*16, "received", number) 658 return 0, errInvalidChain 659 } 660 } 661 // Check if a common ancestor was found 662 finished = true 663 for i := len(headers) - 1; i >= 0; i-- { 664 // Skip any headers that underflow/overflow our requested set 665 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > ceil { 666 continue 667 } 668 // Otherwise check if we already know the header or not 669 h := headers[i].Hash() 670 n := headers[i].Number.Uint64() 671 if (d.mode == FullSync && d.blockchain.HasBlock(h, n)) || (d.mode != FullSync && d.lightchain.HasHeader(h, n)) { 672 number, hash = n, h 673 674 // If every header is known, even future ones, the peer straight out lied about its head 675 if number > height && i == limit-1 { 676 p.log.Warn("Lied about chain head", "reported", height, "found", number) 677 return 0, errStallingPeer 678 } 679 break 680 } 681 } 682 683 case <-timeout: 684 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 685 return 0, errTimeout 686 687 case <-d.bodyCh: 688 case <-d.receiptCh: 689 // Out of bounds delivery, ignore 690 } 691 } 692 // If the head fetch already found an ancestor, return 693 if hash != (common.Hash{}) { 694 if int64(number) <= floor { 695 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 696 return 0, errInvalidAncestor 697 } 698 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 699 return number, nil 700 } 701 // Ancestor not found, we need to binary search over our chain 702 start, end := uint64(0), head 703 if floor > 0 { 704 start = uint64(floor) 705 } 706 for start+1 < end { 707 // Split our chain interval in two, and request the hash to cross check 708 check := (start + end) / 2 709 710 ttl := d.requestTTL() 711 timeout := time.After(ttl) 712 713 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 714 715 // Wait until a reply arrives to this request 716 for arrived := false; !arrived; { 717 select { 718 case <-d.cancelCh: 719 return 0, errCancelHeaderFetch 720 721 case packer := <-d.headerCh: 722 // Discard anything not from the origin peer 723 if packer.PeerId() != p.id { 724 log.Debug("Received headers from incorrect peer", "peer", packer.PeerId()) 725 break 726 } 727 // Make sure the peer actually gave something valid 728 headers := packer.(*headerPack).headers 729 if len(headers) != 1 { 730 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 731 return 0, errBadPeer 732 } 733 arrived = true 734 735 // Modify the search interval based on the response 736 h := headers[0].Hash() 737 n := headers[0].Number.Uint64() 738 if (d.mode == FullSync && !d.blockchain.HasBlock(h, n)) || (d.mode != FullSync && !d.lightchain.HasHeader(h, n)) { 739 end = check 740 break 741 } 742 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 743 if header.Number.Uint64() != check { 744 p.log.Debug("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 745 return 0, errBadPeer 746 } 747 start = check 748 hash = h 749 750 case <-timeout: 751 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 752 return 0, errTimeout 753 754 case <-d.bodyCh: 755 case <-d.receiptCh: 756 // Out of bounds delivery, ignore 757 } 758 } 759 } 760 // Ensure valid ancestry and return 761 if int64(start) <= floor { 762 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 763 return 0, errInvalidAncestor 764 } 765 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 766 return start, nil 767 } 768 769 // fetchHeaders keeps retrieving headers concurrently from the number 770 // requested, until no more are returned, potentially throttling on the way. To 771 // facilitate concurrency but still protect against malicious nodes sending bad 772 // headers, we construct a header chain skeleton using the "origin" peer we are 773 // syncing with, and fill in the missing headers using anyone else. Headers from 774 // other peers are only accepted if they map cleanly to the skeleton. If no one 775 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 776 // the origin is dropped. 777 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64, pivot uint64) error { 778 p.log.Debug("Directing header downloads", "origin", from) 779 defer p.log.Debug("Header download terminated") 780 781 // Create a timeout timer, and the associated header fetcher 782 skeleton := true // Skeleton assembly phase or finishing up 783 request := time.Now() // time of the last skeleton fetch request 784 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 785 <-timeout.C // timeout channel should be initially empty 786 defer timeout.Stop() 787 788 var ttl time.Duration 789 getHeaders := func(from uint64) { 790 request = time.Now() 791 792 ttl = d.requestTTL() 793 timeout.Reset(ttl) 794 795 if skeleton { 796 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 797 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 798 } else { 799 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 800 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 801 } 802 } 803 // Start pulling the header chain skeleton until all is done 804 getHeaders(from) 805 806 for { 807 select { 808 case <-d.cancelCh: 809 return errCancelHeaderFetch 810 811 case packet := <-d.headerCh: 812 // Make sure the active peer is giving us the skeleton headers 813 if packet.PeerId() != p.id { 814 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 815 break 816 } 817 headerReqTimer.UpdateSince(request) 818 timeout.Stop() 819 820 // If the skeleton's finished, pull any remaining head headers directly from the origin 821 if packet.Items() == 0 && skeleton { 822 skeleton = false 823 getHeaders(from) 824 continue 825 } 826 // If no more headers are inbound, notify the content fetchers and return 827 if packet.Items() == 0 { 828 // Don't abort header fetches while the pivot is downloading 829 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 830 p.log.Debug("No headers, waiting for pivot commit") 831 select { 832 case <-time.After(fsHeaderContCheck): 833 getHeaders(from) 834 continue 835 case <-d.cancelCh: 836 return errCancelHeaderFetch 837 } 838 } 839 // Pivot done (or not in fast sync) and no more headers, terminate the process 840 p.log.Debug("No more headers available") 841 select { 842 case d.headerProcCh <- nil: 843 return nil 844 case <-d.cancelCh: 845 return errCancelHeaderFetch 846 } 847 } 848 headers := packet.(*headerPack).headers 849 850 // If we received a skeleton batch, resolve internals concurrently 851 if skeleton { 852 filled, proced, err := d.fillHeaderSkeleton(from, headers) 853 if err != nil { 854 p.log.Debug("Skeleton chain invalid", "err", err) 855 return errInvalidChain 856 } 857 headers = filled[proced:] 858 from += uint64(proced) 859 } else { 860 // If we're closing in on the chain head, but haven't yet reached it, delay 861 // the last few headers so mini reorgs on the head don't cause invalid hash 862 // chain errors. 863 if n := len(headers); n > 0 { 864 // Retrieve the current head we're at 865 head := uint64(0) 866 if d.mode == LightSync { 867 head = d.lightchain.CurrentHeader().Number.Uint64() 868 } else { 869 head = d.blockchain.CurrentFastBlock().NumberU64() 870 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 871 head = full 872 } 873 } 874 // If the head is way older than this batch, delay the last few headers 875 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 876 delay := reorgProtHeaderDelay 877 if delay > n { 878 delay = n 879 } 880 headers = headers[:n-delay] 881 } 882 } 883 } 884 // Insert all the new headers and fetch the next batch 885 if len(headers) > 0 { 886 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 887 select { 888 case d.headerProcCh <- headers: 889 case <-d.cancelCh: 890 return errCancelHeaderFetch 891 } 892 from += uint64(len(headers)) 893 getHeaders(from) 894 } else { 895 // No headers delivered, or all of them being delayed, sleep a bit and retry 896 p.log.Trace("All headers delayed, waiting") 897 select { 898 case <-time.After(fsHeaderContCheck): 899 getHeaders(from) 900 continue 901 case <-d.cancelCh: 902 return errCancelHeaderFetch 903 } 904 } 905 906 case <-timeout.C: 907 if d.dropPeer == nil { 908 // The dropPeer method is nil when `--copydb` is used for a local copy. 909 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 910 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 911 break 912 } 913 // Header retrieval timed out, consider the peer bad and drop 914 p.log.Debug("Header request timed out", "elapsed", ttl) 915 headerTimeoutMeter.Mark(1) 916 d.dropPeer(p.id) 917 918 // Finish the sync gracefully instead of dumping the gathered data though 919 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 920 select { 921 case ch <- false: 922 case <-d.cancelCh: 923 } 924 } 925 select { 926 case d.headerProcCh <- nil: 927 case <-d.cancelCh: 928 } 929 return errBadPeer 930 } 931 } 932 } 933 934 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 935 // and maps them to the provided skeleton header chain. 936 // 937 // Any partial results from the beginning of the skeleton is (if possible) forwarded 938 // immediately to the header processor to keep the rest of the pipeline full even 939 // in the case of header stalls. 940 // 941 // The method returns the entire filled skeleton and also the number of headers 942 // already forwarded for processing. 943 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 944 log.Debug("Filling up skeleton", "from", from) 945 d.queue.ScheduleSkeleton(from, skeleton) 946 947 var ( 948 deliver = func(packet dataPack) (int, error) { 949 pack := packet.(*headerPack) 950 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 951 } 952 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 953 throttle = func() bool { return false } 954 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, error) { 955 return d.queue.ReserveHeaders(p, count), false, nil 956 } 957 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 958 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 959 setIdle = func(p *peerConnection, accepted int) { p.SetHeadersIdle(accepted) } 960 ) 961 err := d.fetchParts(errCancelHeaderFetch, d.headerCh, deliver, d.queue.headerContCh, expire, 962 d.queue.PendingHeaders, d.queue.InFlightHeaders, throttle, reserve, 963 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 964 965 log.Debug("Skeleton fill terminated", "err", err) 966 967 filled, proced := d.queue.RetrieveHeaders() 968 return filled, proced, err 969 } 970 971 // fetchBodies iteratively downloads the scheduled block bodies, taking any 972 // available peers, reserving a chunk of blocks for each, waiting for delivery 973 // and also periodically checking for timeouts. 974 func (d *Downloader) fetchBodies(from uint64) error { 975 log.Debug("Downloading block bodies", "origin", from) 976 977 var ( 978 deliver = func(packet dataPack) (int, error) { 979 pack := packet.(*bodyPack) 980 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 981 } 982 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 983 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 984 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 985 setIdle = func(p *peerConnection, accepted int) { p.SetBodiesIdle(accepted) } 986 ) 987 err := d.fetchParts(errCancelBodyFetch, d.bodyCh, deliver, d.bodyWakeCh, expire, 988 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ShouldThrottleBlocks, d.queue.ReserveBodies, 989 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 990 991 log.Debug("Block body download terminated", "err", err) 992 return err 993 } 994 995 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 996 // available peers, reserving a chunk of receipts for each, waiting for delivery 997 // and also periodically checking for timeouts. 998 func (d *Downloader) fetchReceipts(from uint64) error { 999 log.Debug("Downloading transaction receipts", "origin", from) 1000 1001 var ( 1002 deliver = func(packet dataPack) (int, error) { 1003 pack := packet.(*receiptPack) 1004 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1005 } 1006 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1007 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1008 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1009 setIdle = func(p *peerConnection, accepted int) { p.SetReceiptsIdle(accepted) } 1010 ) 1011 err := d.fetchParts(errCancelReceiptFetch, d.receiptCh, deliver, d.receiptWakeCh, expire, 1012 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ShouldThrottleReceipts, d.queue.ReserveReceipts, 1013 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1014 1015 log.Debug("Transaction receipt download terminated", "err", err) 1016 return err 1017 } 1018 1019 // fetchParts iteratively downloads scheduled block parts, taking any available 1020 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1021 // also periodically checking for timeouts. 1022 // 1023 // As the scheduling/timeout logic mostly is the same for all downloaded data 1024 // types, this method is used by each for data gathering and is instrumented with 1025 // various callbacks to handle the slight differences between processing them. 1026 // 1027 // The instrumentation parameters: 1028 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1029 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1030 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1031 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1032 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1033 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1034 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1035 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1036 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1037 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1038 // - fetch: network callback to actually send a particular download request to a physical remote peer 1039 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1040 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1041 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1042 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1043 // - kind: textual label of the type being downloaded to display in log mesages 1044 func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1045 expire func() map[string]int, pending func() int, inFlight func() bool, throttle func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, error), 1046 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1047 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int), kind string) error { 1048 1049 // Create a ticker to detect expired retrieval tasks 1050 ticker := time.NewTicker(100 * time.Millisecond) 1051 defer ticker.Stop() 1052 1053 update := make(chan struct{}, 1) 1054 1055 // Prepare the queue and fetch block parts until the block header fetcher's done 1056 finished := false 1057 for { 1058 select { 1059 case <-d.cancelCh: 1060 return errCancel 1061 1062 case packet := <-deliveryCh: 1063 // If the peer was previously banned and failed to deliver its pack 1064 // in a reasonable time frame, ignore its message. 1065 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1066 // Deliver the received chunk of data and check chain validity 1067 accepted, err := deliver(packet) 1068 if err == errInvalidChain { 1069 return err 1070 } 1071 // Unless a peer delivered something completely else than requested (usually 1072 // caused by a timed out request which came through in the end), set it to 1073 // idle. If the delivery's stale, the peer should have already been idled. 1074 if err != errStaleDelivery { 1075 setIdle(peer, accepted) 1076 } 1077 // Issue a log to the user to see what's going on 1078 switch { 1079 case err == nil && packet.Items() == 0: 1080 peer.log.Trace("Requested data not delivered", "type", kind) 1081 case err == nil: 1082 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1083 default: 1084 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1085 } 1086 } 1087 // Blocks assembled, try to update the progress 1088 select { 1089 case update <- struct{}{}: 1090 default: 1091 } 1092 1093 case cont := <-wakeCh: 1094 // The header fetcher sent a continuation flag, check if it's done 1095 if !cont { 1096 finished = true 1097 } 1098 // Headers arrive, try to update the progress 1099 select { 1100 case update <- struct{}{}: 1101 default: 1102 } 1103 1104 case <-ticker.C: 1105 // Sanity check update the progress 1106 select { 1107 case update <- struct{}{}: 1108 default: 1109 } 1110 1111 case <-update: 1112 // Short circuit if we lost all our peers 1113 if d.peers.Len() == 0 { 1114 return errNoPeers 1115 } 1116 // Check for fetch request timeouts and demote the responsible peers 1117 for pid, fails := range expire() { 1118 if peer := d.peers.Peer(pid); peer != nil { 1119 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1120 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1121 // out that sync wise we need to get rid of the peer. 1122 // 1123 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1124 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1125 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1126 if fails > 2 { 1127 peer.log.Trace("Data delivery timed out", "type", kind) 1128 setIdle(peer, 0) 1129 } else { 1130 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1131 if d.dropPeer == nil { 1132 // The dropPeer method is nil when `--copydb` is used for a local copy. 1133 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1134 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1135 } else { 1136 d.dropPeer(pid) 1137 } 1138 } 1139 } 1140 } 1141 // If there's nothing more to fetch, wait or terminate 1142 if pending() == 0 { 1143 if !inFlight() && finished { 1144 log.Debug("Data fetching completed", "type", kind) 1145 return nil 1146 } 1147 break 1148 } 1149 // Send a download request to all idle peers, until throttled 1150 progressed, throttled, running := false, false, inFlight() 1151 idles, total := idle() 1152 1153 for _, peer := range idles { 1154 // Short circuit if throttling activated 1155 if throttle() { 1156 throttled = true 1157 break 1158 } 1159 // Short circuit if there is no more available task. 1160 if pending() == 0 { 1161 break 1162 } 1163 // Reserve a chunk of fetches for a peer. A nil can mean either that 1164 // no more headers are available, or that the peer is known not to 1165 // have them. 1166 request, progress, err := reserve(peer, capacity(peer)) 1167 if err != nil { 1168 return err 1169 } 1170 if progress { 1171 progressed = true 1172 } 1173 if request == nil { 1174 continue 1175 } 1176 if request.From > 0 { 1177 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1178 } else { 1179 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1180 } 1181 // Fetch the chunk and make sure any errors return the hashes to the queue 1182 if fetchHook != nil { 1183 fetchHook(request.Headers) 1184 } 1185 if err := fetch(peer, request); err != nil { 1186 // Although we could try and make an attempt to fix this, this error really 1187 // means that we've double allocated a fetch task to a peer. If that is the 1188 // case, the internal state of the downloader and the queue is very wrong so 1189 // better hard crash and note the error instead of silently accumulating into 1190 // a much bigger issue. 1191 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1192 } 1193 running = true 1194 } 1195 // Make sure that we have peers available for fetching. If all peers have been tried 1196 // and all failed throw an error 1197 if !progressed && !throttled && !running && len(idles) == total && pending() > 0 { 1198 return errPeersUnavailable 1199 } 1200 } 1201 } 1202 } 1203 1204 // processHeaders takes batches of retrieved headers from an input channel and 1205 // keeps processing and scheduling them into the header chain and downloader's 1206 // queue until the stream ends or a failure occurs. 1207 func (d *Downloader) processHeaders(origin uint64, pivot uint64, td *big.Int) error { 1208 // Keep a count of uncertain headers to roll back 1209 rollback := []*types.Header{} 1210 defer func() { 1211 if len(rollback) > 0 { 1212 // Flatten the headers and roll them back 1213 hashes := make([]common.Hash, len(rollback)) 1214 for i, header := range rollback { 1215 hashes[i] = header.Hash() 1216 } 1217 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1218 if d.mode != LightSync { 1219 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1220 lastBlock = d.blockchain.CurrentBlock().Number() 1221 } 1222 d.lightchain.Rollback(hashes) 1223 curFastBlock, curBlock := common.Big0, common.Big0 1224 if d.mode != LightSync { 1225 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1226 curBlock = d.blockchain.CurrentBlock().Number() 1227 } 1228 log.Warn("Rolled back headers", "count", len(hashes), 1229 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1230 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1231 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock)) 1232 } 1233 }() 1234 1235 // Wait for batches of headers to process 1236 gotHeaders := false 1237 1238 for { 1239 select { 1240 case <-d.cancelCh: 1241 return errCancelHeaderProcessing 1242 1243 case headers := <-d.headerProcCh: 1244 // Terminate header processing if we synced up 1245 if len(headers) == 0 { 1246 // Notify everyone that headers are fully processed 1247 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1248 select { 1249 case ch <- false: 1250 case <-d.cancelCh: 1251 } 1252 } 1253 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1254 // better chain compared to ours. The only exception is if its promised blocks were 1255 // already imported by other means (e.g. fetcher): 1256 // 1257 // R <remote peer>, L <local node>: Both at block 10 1258 // R: Mine block 11, and propagate it to L 1259 // L: Queue block 11 for import 1260 // L: Notice that R's head and TD increased compared to ours, start sync 1261 // L: Import of block 11 finishes 1262 // L: Sync begins, and finds common ancestor at 11 1263 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1264 // R: Nothing to give 1265 if d.mode != LightSync { 1266 head := d.blockchain.CurrentBlock() 1267 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1268 return errStallingPeer 1269 } 1270 } 1271 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1272 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1273 // of delivering the post-pivot blocks that would flag the invalid content. 1274 // 1275 // This check cannot be executed "as is" for full imports, since blocks may still be 1276 // queued for processing when the header download completes. However, as long as the 1277 // peer gave us something useful, we're already happy/progressed (above check). 1278 if d.mode == FastSync || d.mode == LightSync { 1279 head := d.lightchain.CurrentHeader() 1280 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1281 return errStallingPeer 1282 } 1283 } 1284 // Disable any rollback and return 1285 rollback = nil 1286 return nil 1287 } 1288 // Otherwise split the chunk of headers into batches and process them 1289 gotHeaders = true 1290 1291 for len(headers) > 0 { 1292 // Terminate if something failed in between processing chunks 1293 select { 1294 case <-d.cancelCh: 1295 return errCancelHeaderProcessing 1296 default: 1297 } 1298 // Select the next chunk of headers to import 1299 limit := maxHeadersProcess 1300 if limit > len(headers) { 1301 limit = len(headers) 1302 } 1303 chunk := headers[:limit] 1304 1305 // In case of header only syncing, validate the chunk immediately 1306 if d.mode == FastSync || d.mode == LightSync { 1307 // Collect the yet unknown headers to mark them as uncertain 1308 unknown := make([]*types.Header, 0, len(headers)) 1309 for _, header := range chunk { 1310 if !d.lightchain.HasHeader(header.Hash(), header.Number.Uint64()) { 1311 unknown = append(unknown, header) 1312 } 1313 } 1314 // If we're importing pure headers, verify based on their recentness 1315 frequency := fsHeaderCheckFrequency 1316 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1317 frequency = 1 1318 } 1319 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1320 // If some headers were inserted, add them too to the rollback list 1321 if n > 0 { 1322 rollback = append(rollback, chunk[:n]...) 1323 } 1324 log.Debug("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "err", err) 1325 return errInvalidChain 1326 } 1327 // All verifications passed, store newly found uncertain headers 1328 rollback = append(rollback, unknown...) 1329 if len(rollback) > fsHeaderSafetyNet { 1330 rollback = append(rollback[:0], rollback[len(rollback)-fsHeaderSafetyNet:]...) 1331 } 1332 } 1333 // Unless we're doing light chains, schedule the headers for associated content retrieval 1334 if d.mode == FullSync || d.mode == FastSync || d.mode == BoundedFullSync { 1335 // If we've reached the allowed number of pending headers, stall a bit 1336 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1337 select { 1338 case <-d.cancelCh: 1339 return errCancelHeaderProcessing 1340 case <-time.After(time.Second): 1341 } 1342 } 1343 // Otherwise insert the headers for content retrieval 1344 inserts := d.queue.Schedule(chunk, origin) 1345 if len(inserts) != len(chunk) { 1346 log.Debug("Stale headers") 1347 return errBadPeer 1348 } 1349 } 1350 headers = headers[limit:] 1351 origin += uint64(limit) 1352 } 1353 1354 // Update the highest block number we know if a higher one is found. 1355 d.syncStatsLock.Lock() 1356 if d.syncStatsChainHeight < origin { 1357 d.syncStatsChainHeight = origin - 1 1358 } 1359 d.syncStatsLock.Unlock() 1360 1361 // Signal the content downloaders of the availablility of new tasks 1362 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1363 select { 1364 case ch <- true: 1365 default: 1366 } 1367 } 1368 } 1369 } 1370 } 1371 1372 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1373 func (d *Downloader) processFullSyncContent() error { 1374 for { 1375 results := d.queue.Results(true) 1376 if len(results) == 0 { 1377 return nil 1378 } 1379 if d.chainInsertHook != nil { 1380 d.chainInsertHook(results) 1381 } 1382 if err := d.importBlockResults(results); err != nil { 1383 return err 1384 } 1385 } 1386 } 1387 1388 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1389 // Check for any early termination requests 1390 if len(results) == 0 { 1391 return nil 1392 } 1393 select { 1394 case <-d.quitCh: 1395 return errCancelContentProcessing 1396 default: 1397 } 1398 // Retrieve the a batch of results to import 1399 first, last := results[0].Header, results[len(results)-1].Header 1400 log.Debug("Inserting downloaded chain", "items", len(results), 1401 "firstnum", first.Number, "firsthash", first.Hash(), 1402 "lastnum", last.Number, "lasthash", last.Hash(), 1403 ) 1404 blocks := make([]*types.Block, len(results)) 1405 for i, result := range results { 1406 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1407 } 1408 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1409 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1410 return errInvalidChain 1411 } 1412 return nil 1413 } 1414 1415 // processFastSyncContent takes fetch results from the queue and writes them to the 1416 // database. It also controls the synchronisation of state nodes of the pivot block. 1417 func (d *Downloader) processFastSyncContent(latest *types.Header) error { 1418 // Start syncing state of the reported head block. This should get us most of 1419 // the state of the pivot block. 1420 stateSync := d.syncState(latest.Root) 1421 defer stateSync.Cancel() 1422 go func() { 1423 if err := stateSync.Wait(); err != nil && err != errCancelStateFetch { 1424 d.queue.Close() // wake up Results 1425 } 1426 }() 1427 // Figure out the ideal pivot block. Note, that this goalpost may move if the 1428 // sync takes long enough for the chain head to move significantly. 1429 pivot := uint64(0) 1430 if height := latest.Number.Uint64(); height > uint64(fsMinFullBlocks) { 1431 pivot = height - uint64(fsMinFullBlocks) 1432 } 1433 // To cater for moving pivot points, track the pivot block and subsequently 1434 // accumulated download results separately. 1435 var ( 1436 oldPivot *fetchResult // Locked in pivot block, might change eventually 1437 oldTail []*fetchResult // Downloaded content after the pivot 1438 ) 1439 for { 1440 // Wait for the next batch of downloaded data to be available, and if the pivot 1441 // block became stale, move the goalpost 1442 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1443 if len(results) == 0 { 1444 // If pivot sync is done, stop 1445 if oldPivot == nil { 1446 return stateSync.Cancel() 1447 } 1448 // If sync failed, stop 1449 select { 1450 case <-d.cancelCh: 1451 return stateSync.Cancel() 1452 default: 1453 } 1454 } 1455 if d.chainInsertHook != nil { 1456 d.chainInsertHook(results) 1457 } 1458 if oldPivot != nil { 1459 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1460 } 1461 // Split around the pivot block and process the two sides via fast/full sync 1462 if atomic.LoadInt32(&d.committed) == 0 { 1463 latest = results[len(results)-1].Header 1464 if height := latest.Number.Uint64(); height > pivot+2*uint64(fsMinFullBlocks) { 1465 log.Warn("Pivot became stale, moving", "old", pivot, "new", height-uint64(fsMinFullBlocks)) 1466 pivot = height - uint64(fsMinFullBlocks) 1467 } 1468 } 1469 P, beforeP, afterP := splitAroundPivot(pivot, results) 1470 if err := d.commitFastSyncData(beforeP, stateSync); err != nil { 1471 return err 1472 } 1473 if P != nil { 1474 // If new pivot block found, cancel old state retrieval and restart 1475 if oldPivot != P { 1476 stateSync.Cancel() 1477 1478 stateSync = d.syncState(P.Header.Root) 1479 defer stateSync.Cancel() 1480 go func() { 1481 if err := stateSync.Wait(); err != nil && err != errCancelStateFetch { 1482 d.queue.Close() // wake up Results 1483 } 1484 }() 1485 oldPivot = P 1486 } 1487 // Wait for completion, occasionally checking for pivot staleness 1488 select { 1489 case <-stateSync.done: 1490 if stateSync.err != nil { 1491 return stateSync.err 1492 } 1493 if err := d.commitPivotBlock(P); err != nil { 1494 return err 1495 } 1496 oldPivot = nil 1497 1498 case <-time.After(time.Second): 1499 oldTail = afterP 1500 continue 1501 } 1502 } 1503 // Fast sync done, pivot commit done, full import 1504 if err := d.importBlockResults(afterP); err != nil { 1505 return err 1506 } 1507 } 1508 } 1509 1510 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1511 for _, result := range results { 1512 num := result.Header.Number.Uint64() 1513 switch { 1514 case num < pivot: 1515 before = append(before, result) 1516 case num == pivot: 1517 p = result 1518 default: 1519 after = append(after, result) 1520 } 1521 } 1522 return p, before, after 1523 } 1524 1525 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1526 // Check for any early termination requests 1527 if len(results) == 0 { 1528 return nil 1529 } 1530 select { 1531 case <-d.quitCh: 1532 return errCancelContentProcessing 1533 case <-stateSync.done: 1534 if err := stateSync.Wait(); err != nil { 1535 return err 1536 } 1537 default: 1538 } 1539 // Retrieve the a batch of results to import 1540 first, last := results[0].Header, results[len(results)-1].Header 1541 log.Debug("Inserting fast-sync blocks", "items", len(results), 1542 "firstnum", first.Number, "firsthash", first.Hash(), 1543 "lastnumn", last.Number, "lasthash", last.Hash(), 1544 ) 1545 blocks := make([]*types.Block, len(results)) 1546 receipts := make([]types.Receipts, len(results)) 1547 for i, result := range results { 1548 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1549 receipts[i] = result.Receipts 1550 } 1551 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts); err != nil { 1552 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1553 return errInvalidChain 1554 } 1555 return nil 1556 } 1557 1558 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1559 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1560 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1561 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}); err != nil { 1562 return err 1563 } 1564 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1565 return err 1566 } 1567 atomic.StoreInt32(&d.committed, 1) 1568 return nil 1569 } 1570 1571 // DeliverHeaders injects a new batch of block headers received from a remote 1572 // node into the download schedule. 1573 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) { 1574 return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1575 } 1576 1577 // DeliverBodies injects a new batch of block bodies received from a remote node. 1578 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) { 1579 return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1580 } 1581 1582 // DeliverReceipts injects a new batch of receipts received from a remote node. 1583 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) { 1584 return d.deliver(id, d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1585 } 1586 1587 // DeliverNodeData injects a new batch of node state data received from a remote node. 1588 func (d *Downloader) DeliverNodeData(id string, data [][]byte) (err error) { 1589 return d.deliver(id, d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1590 } 1591 1592 // deliver injects a new batch of data received from a remote node. 1593 func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1594 // Update the delivery metrics for both good and failed deliveries 1595 inMeter.Mark(int64(packet.Items())) 1596 defer func() { 1597 if err != nil { 1598 dropMeter.Mark(int64(packet.Items())) 1599 } 1600 }() 1601 // Deliver or abort if the sync is canceled while queuing 1602 d.cancelLock.RLock() 1603 cancel := d.cancelCh 1604 d.cancelLock.RUnlock() 1605 if cancel == nil { 1606 return errNoSyncActive 1607 } 1608 select { 1609 case destCh <- packet: 1610 return nil 1611 case <-cancel: 1612 return errNoSyncActive 1613 } 1614 } 1615 1616 // qosTuner is the quality of service tuning loop that occasionally gathers the 1617 // peer latency statistics and updates the estimated request round trip time. 1618 func (d *Downloader) qosTuner() { 1619 for { 1620 // Retrieve the current median RTT and integrate into the previoust target RTT 1621 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 1622 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 1623 1624 // A new RTT cycle passed, increase our confidence in the estimated RTT 1625 conf := atomic.LoadUint64(&d.rttConfidence) 1626 conf = conf + (1000000-conf)/2 1627 atomic.StoreUint64(&d.rttConfidence, conf) 1628 1629 // Log the new QoS values and sleep until the next RTT 1630 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1631 select { 1632 case <-d.quitCh: 1633 return 1634 case <-time.After(rtt): 1635 } 1636 } 1637 } 1638 1639 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 1640 // peer set, needing to reduce the confidence we have in out QoS estimates. 1641 func (d *Downloader) qosReduceConfidence() { 1642 // If we have a single peer, confidence is always 1 1643 peers := uint64(d.peers.Len()) 1644 if peers == 0 { 1645 // Ensure peer connectivity races don't catch us off guard 1646 return 1647 } 1648 if peers == 1 { 1649 atomic.StoreUint64(&d.rttConfidence, 1000000) 1650 return 1651 } 1652 // If we have a ton of peers, don't drop confidence) 1653 if peers >= uint64(qosConfidenceCap) { 1654 return 1655 } 1656 // Otherwise drop the confidence factor 1657 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 1658 if float64(conf)/1000000 < rttMinConfidence { 1659 conf = uint64(rttMinConfidence * 1000000) 1660 } 1661 atomic.StoreUint64(&d.rttConfidence, conf) 1662 1663 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1664 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1665 } 1666 1667 // requestRTT returns the current target round trip time for a download request 1668 // to complete in. 1669 // 1670 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 1671 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 1672 // be adapted to, but smaller ones are preferred (stabler download stream). 1673 func (d *Downloader) requestRTT() time.Duration { 1674 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 1675 } 1676 1677 // requestTTL returns the current timeout allowance for a single download request 1678 // to finish under. 1679 func (d *Downloader) requestTTL() time.Duration { 1680 var ( 1681 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1682 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 1683 ) 1684 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 1685 if ttl > ttlLimit { 1686 ttl = ttlLimit 1687 } 1688 return ttl 1689 } 1690 1691 // Extra downloader functionality for non-proof-of-work consensus 1692 1693 // Synchronizes with a peer, but only up to the provided Hash 1694 func (d *Downloader) syncWithPeerUntil(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 1695 d.mux.Post(StartEvent{}) 1696 defer func() { 1697 // reset on error 1698 if err != nil { 1699 d.mux.Post(FailedEvent{err}) 1700 } else { 1701 d.mux.Post(DoneEvent{}) 1702 } 1703 }() 1704 if p.version < 62 { 1705 return errTooOld 1706 } 1707 1708 log.Info("Synchronising with the network", "id", p.id, "version", p.version) 1709 defer func(start time.Time) { 1710 log.Info("Synchronisation terminated", "duration", time.Since(start)) 1711 }(time.Now()) 1712 1713 remoteHeader, err := d.fetchHeader(p, hash) 1714 if err != nil { 1715 return err 1716 } 1717 1718 remoteHeight := remoteHeader.Number.Uint64() 1719 localHeight := d.blockchain.CurrentBlock().NumberU64() 1720 1721 d.syncStatsLock.Lock() 1722 if d.syncStatsChainHeight <= localHeight || d.syncStatsChainOrigin > localHeight { 1723 d.syncStatsChainOrigin = localHeight 1724 } 1725 d.syncStatsChainHeight = remoteHeight 1726 d.syncStatsLock.Unlock() 1727 1728 d.queue.Prepare(localHeight+1, d.mode) 1729 if d.syncInitHook != nil { 1730 d.syncInitHook(localHeight, remoteHeight) 1731 } 1732 1733 pivot := uint64(0) 1734 1735 fetchers := []func() error{ 1736 func() error { return d.fetchBoundedHeaders(p, localHeight+1, remoteHeight) }, 1737 func() error { return d.fetchBodies(localHeight + 1) }, 1738 func() error { return d.fetchReceipts(localHeight + 1) }, // Receipts are only retrieved during fast sync 1739 func() error { return d.processHeaders(localHeight+1, pivot, td) }, 1740 d.processFullSyncContent, //This must be added to clear the buffer of downloaded content as it's being filled 1741 } 1742 return d.spawnSync(fetchers) 1743 } 1744 1745 // Fetches a single header from a peer 1746 func (d *Downloader) fetchHeader(p *peerConnection, hash common.Hash) (*types.Header, error) { 1747 log.Info("retrieving remote chain height", "peer", p) 1748 1749 go p.peer.RequestHeadersByHash(hash, 1, 0, false) 1750 1751 timeout := time.After(d.requestTTL()) 1752 for { 1753 select { 1754 case <-d.cancelCh: 1755 return nil, errCancelBlockFetch 1756 1757 case packet := <-d.headerCh: 1758 // Discard anything not from the origin peer 1759 if packet.PeerId() != p.id { 1760 log.Info("Received headers from incorrect peer", "peer id", packet.PeerId()) 1761 break 1762 } 1763 // Make sure the peer actually gave something valid 1764 headers := packet.(*headerPack).headers 1765 if len(headers) != 1 { 1766 log.Info("invalid number of head headers (!= 1)", "peer", p, "len(headers)", len(headers)) 1767 return nil, errBadPeer 1768 } 1769 return headers[0], nil 1770 1771 case <-timeout: 1772 log.Info("head header timeout", "peer", p) 1773 return nil, errTimeout 1774 1775 case <-d.bodyCh: 1776 case <-d.stateCh: 1777 case <-d.receiptCh: 1778 // Out of bounds delivery, ignore 1779 } 1780 } 1781 } 1782 1783 // Not defined in go's stdlib: 1784 func minInt(a, b int) int { 1785 if a < b { 1786 return a 1787 } 1788 return b 1789 } 1790 1791 // Fetches headers between `from` and `to`, inclusive. 1792 // Assumes invariant: from <= to. 1793 func (d *Downloader) fetchBoundedHeaders(p *peerConnection, from uint64, to uint64) error { 1794 log.Info("directing header downloads", "peer", p, "from", from, "to", to) 1795 defer log.Info("header download terminated", "peer", p) 1796 1797 // Create a timeout timer, and the associated header fetcher 1798 skeleton := true // Skeleton assembly phase or finishing up 1799 request := time.Now() // time of the last skeleton fetch request 1800 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 1801 <-timeout.C // timeout channel should be initially empty 1802 defer timeout.Stop() 1803 1804 getHeaders := func(from uint64) { 1805 request = time.Now() 1806 timeout.Reset(d.requestTTL()) 1807 1808 skeletonStart := from + uint64(MaxHeaderFetch) - 1 1809 1810 if skeleton { 1811 if skeletonStart > to { 1812 skeleton = false 1813 } 1814 } 1815 1816 if skeleton { 1817 numSkeletonHeaders := minInt(MaxSkeletonSize, (int(to-from)+1)/MaxHeaderFetch) 1818 log.Trace("fetching skeleton headers", "peer", p, "num skeleton headers", numSkeletonHeaders, "from", from) 1819 go p.peer.RequestHeadersByNumber(skeletonStart, numSkeletonHeaders, MaxHeaderFetch-1, false) 1820 } else { 1821 // There are not enough headers remaining to warrant a skeleton fetch. 1822 // Grab all of the remaining headers. 1823 1824 numHeaders := int(to-from) + 1 1825 log.Trace("fetching full headers", "peer", p, "num headers", numHeaders, "from", from) 1826 go p.peer.RequestHeadersByNumber(from, numHeaders, 0, false) 1827 } 1828 } 1829 // Start pulling the header chain skeleton until all is done 1830 getHeaders(from) 1831 1832 for { 1833 select { 1834 case <-d.cancelCh: 1835 return errCancelHeaderFetch 1836 1837 case packet := <-d.headerCh: 1838 // Make sure the active peer is giving us the skeleton headers 1839 if packet.PeerId() != p.id { 1840 log.Info("Received headers from incorrect peer", "peer id", packet.PeerId()) 1841 break 1842 } 1843 headerReqTimer.UpdateSince(request) 1844 timeout.Stop() 1845 1846 headers := packet.(*headerPack).headers 1847 1848 // If we received a skeleton batch, resolve internals concurrently 1849 if skeleton { 1850 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1851 if err != nil { 1852 log.Debug("skeleton chain invalid", "peer", p, "err", err) 1853 return errInvalidChain 1854 } 1855 headers = filled[proced:] 1856 from += uint64(proced) 1857 } 1858 // Insert all the new headers and fetch the next batch 1859 if len(headers) > 0 { 1860 log.Trace("schedule headers", "peer", p, "num headers", len(headers), "from", from) 1861 select { 1862 case d.headerProcCh <- headers: 1863 case <-d.cancelCh: 1864 return errCancelHeaderFetch 1865 } 1866 from += uint64(len(headers)) 1867 } 1868 1869 if from <= to { 1870 getHeaders(from) 1871 } else { 1872 // Notify the content fetchers that no more headers are inbound and return. 1873 select { 1874 case d.headerProcCh <- nil: 1875 return nil 1876 case <-d.cancelCh: 1877 return errCancelHeaderFetch 1878 } 1879 } 1880 1881 case <-timeout.C: 1882 // Header retrieval timed out, consider the peer bad and drop 1883 log.Info("header request timed out", "peer", p) 1884 headerTimeoutMeter.Mark(1) 1885 d.dropPeer(p.id) 1886 1887 // Finish the sync gracefully instead of dumping the gathered data though 1888 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1889 select { 1890 case ch <- false: 1891 case <-d.cancelCh: 1892 } 1893 } 1894 select { 1895 case d.headerProcCh <- nil: 1896 case <-d.cancelCh: 1897 } 1898 return errBadPeer 1899 } 1900 } 1901 }