github.com/bencicandrej/quorum@v2.2.6-0.20190909091323-878cab86f711+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 defer atomic.StoreInt32(&d.synchronising, 0) 355 356 // Post a user notification of the sync (only once per session) 357 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 358 log.Info("Block synchronisation started") 359 } 360 // Reset the queue, peer set and wake channels to clean any internal leftover state 361 d.queue.Reset() 362 d.peers.Reset() 363 364 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 365 select { 366 case <-ch: 367 default: 368 } 369 } 370 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 371 for empty := false; !empty; { 372 select { 373 case <-ch: 374 default: 375 empty = true 376 } 377 } 378 } 379 for empty := false; !empty; { 380 select { 381 case <-d.headerProcCh: 382 default: 383 empty = true 384 } 385 } 386 // Create cancel channel for aborting mid-flight and mark the master peer 387 d.cancelLock.Lock() 388 d.cancelCh = make(chan struct{}) 389 d.cancelPeer = id 390 d.cancelLock.Unlock() 391 392 defer d.Cancel() // No matter what, we can't leave the cancel channel open 393 394 // Set the requested sync mode, unless it's forbidden 395 d.mode = mode 396 397 // Retrieve the origin peer and initiate the downloading process 398 p := d.peers.Peer(id) 399 if p == nil { 400 return errUnknownPeer 401 } 402 if d.mode == BoundedFullSync { 403 return d.syncWithPeerUntil(p, hash, td) 404 } 405 return d.syncWithPeer(p, hash, td) 406 } 407 408 // syncWithPeer starts a block synchronization based on the hash chain from the 409 // specified peer and head hash. 410 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 411 d.mux.Post(StartEvent{}) 412 defer func() { 413 // reset on error 414 if err != nil { 415 d.mux.Post(FailedEvent{err}) 416 } else { 417 d.mux.Post(DoneEvent{}) 418 } 419 }() 420 if p.version < 62 { 421 return errTooOld 422 } 423 424 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", d.mode) 425 defer func(start time.Time) { 426 log.Debug("Synchronisation terminated", "elapsed", time.Since(start)) 427 }(time.Now()) 428 429 // Look up the sync boundaries: the common ancestor and the target block 430 latest, err := d.fetchHeight(p) 431 if err != nil { 432 return err 433 } 434 height := latest.Number.Uint64() 435 436 origin, err := d.findAncestor(p, height) 437 if err != nil { 438 return err 439 } 440 d.syncStatsLock.Lock() 441 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 442 d.syncStatsChainOrigin = origin 443 } 444 d.syncStatsChainHeight = height 445 d.syncStatsLock.Unlock() 446 447 // Ensure our origin point is below any fast sync pivot point 448 pivot := uint64(0) 449 if d.mode == FastSync { 450 if height <= uint64(fsMinFullBlocks) { 451 origin = 0 452 } else { 453 pivot = height - uint64(fsMinFullBlocks) 454 if pivot <= origin { 455 origin = pivot - 1 456 } 457 } 458 } 459 d.committed = 1 460 if d.mode == FastSync && pivot != 0 { 461 d.committed = 0 462 } 463 // Initiate the sync using a concurrent header and content retrieval algorithm 464 d.queue.Prepare(origin+1, d.mode) 465 if d.syncInitHook != nil { 466 d.syncInitHook(origin, height) 467 } 468 469 fetchers := []func() error{ 470 func() error { return d.fetchHeaders(p, origin+1, pivot) }, // Headers are always retrieved 471 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 472 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 473 func() error { return d.processHeaders(origin+1, pivot, td) }, 474 } 475 if d.mode == FastSync { 476 fetchers = append(fetchers, func() error { return d.processFastSyncContent(latest) }) 477 } else if d.mode == FullSync { 478 fetchers = append(fetchers, d.processFullSyncContent) 479 } 480 return d.spawnSync(fetchers) 481 } 482 483 // spawnSync runs d.process and all given fetcher functions to completion in 484 // separate goroutines, returning the first error that appears. 485 func (d *Downloader) spawnSync(fetchers []func() error) error { 486 errc := make(chan error, len(fetchers)) 487 d.cancelWg.Add(len(fetchers)) 488 for _, fn := range fetchers { 489 fn := fn 490 go func() { defer d.cancelWg.Done(); errc <- fn() }() 491 } 492 // Wait for the first error, then terminate the others. 493 var err error 494 for i := 0; i < len(fetchers); i++ { 495 if i == len(fetchers)-1 { 496 // Close the queue when all fetchers have exited. 497 // This will cause the block processor to end when 498 // it has processed the queue. 499 d.queue.Close() 500 } 501 if err = <-errc; err != nil { 502 break 503 } 504 } 505 d.queue.Close() 506 d.Cancel() 507 return err 508 } 509 510 // cancel aborts all of the operations and resets the queue. However, cancel does 511 // not wait for the running download goroutines to finish. This method should be 512 // used when cancelling the downloads from inside the downloader. 513 func (d *Downloader) cancel() { 514 // Close the current cancel channel 515 d.cancelLock.Lock() 516 if d.cancelCh != nil { 517 select { 518 case <-d.cancelCh: 519 // Channel was already closed 520 default: 521 close(d.cancelCh) 522 } 523 } 524 d.cancelLock.Unlock() 525 } 526 527 // Cancel aborts all of the operations and waits for all download goroutines to 528 // finish before returning. 529 func (d *Downloader) Cancel() { 530 d.cancel() 531 d.cancelWg.Wait() 532 } 533 534 // Terminate interrupts the downloader, canceling all pending operations. 535 // The downloader cannot be reused after calling Terminate. 536 func (d *Downloader) Terminate() { 537 // Close the termination channel (make sure double close is allowed) 538 d.quitLock.Lock() 539 select { 540 case <-d.quitCh: 541 default: 542 close(d.quitCh) 543 } 544 d.quitLock.Unlock() 545 546 // Cancel any pending download requests 547 d.Cancel() 548 } 549 550 // fetchHeight retrieves the head header of the remote peer to aid in estimating 551 // the total time a pending synchronisation would take. 552 func (d *Downloader) fetchHeight(p *peerConnection) (*types.Header, error) { 553 p.log.Debug("Retrieving remote chain height") 554 555 // Request the advertised remote head block and wait for the response 556 head, _ := p.peer.Head() 557 go p.peer.RequestHeadersByHash(head, 1, 0, false) 558 559 ttl := d.requestTTL() 560 timeout := time.After(ttl) 561 for { 562 select { 563 case <-d.cancelCh: 564 return nil, errCancelBlockFetch 565 566 case packet := <-d.headerCh: 567 // Discard anything not from the origin peer 568 if packet.PeerId() != p.id { 569 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 570 break 571 } 572 // Make sure the peer actually gave something valid 573 headers := packet.(*headerPack).headers 574 if len(headers) != 1 { 575 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 576 return nil, errBadPeer 577 } 578 head := headers[0] 579 p.log.Debug("Remote head header identified", "number", head.Number, "hash", head.Hash()) 580 return head, nil 581 582 case <-timeout: 583 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 584 return nil, errTimeout 585 586 case <-d.bodyCh: 587 case <-d.receiptCh: 588 // Out of bounds delivery, ignore 589 } 590 } 591 } 592 593 // findAncestor tries to locate the common ancestor link of the local chain and 594 // a remote peers blockchain. In the general case when our node was in sync and 595 // on the correct chain, checking the top N links should already get us a match. 596 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 597 // the head links match), we do a binary search to find the common ancestor. 598 func (d *Downloader) findAncestor(p *peerConnection, height uint64) (uint64, error) { 599 // Figure out the valid ancestor range to prevent rewrite attacks 600 floor, ceil := int64(-1), d.lightchain.CurrentHeader().Number.Uint64() 601 602 if d.mode == FullSync { 603 ceil = d.blockchain.CurrentBlock().NumberU64() 604 } else if d.mode == FastSync { 605 ceil = d.blockchain.CurrentFastBlock().NumberU64() 606 } 607 if ceil >= MaxForkAncestry { 608 floor = int64(ceil - MaxForkAncestry) 609 } 610 p.log.Debug("Looking for common ancestor", "local", ceil, "remote", height) 611 612 // Request the topmost blocks to short circuit binary ancestor lookup 613 head := ceil 614 if head > height { 615 head = height 616 } 617 from := int64(head) - int64(MaxHeaderFetch) 618 if from < 0 { 619 from = 0 620 } 621 // Span out with 15 block gaps into the future to catch bad head reports 622 limit := 2 * MaxHeaderFetch / 16 623 count := 1 + int((int64(ceil)-from)/16) 624 if count > limit { 625 count = limit 626 } 627 go p.peer.RequestHeadersByNumber(uint64(from), count, 15, false) 628 629 // Wait for the remote response to the head fetch 630 number, hash := uint64(0), common.Hash{} 631 632 ttl := d.requestTTL() 633 timeout := time.After(ttl) 634 635 for finished := false; !finished; { 636 select { 637 case <-d.cancelCh: 638 return 0, errCancelHeaderFetch 639 640 case packet := <-d.headerCh: 641 // Discard anything not from the origin peer 642 if packet.PeerId() != p.id { 643 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 644 break 645 } 646 // Make sure the peer actually gave something valid 647 headers := packet.(*headerPack).headers 648 if len(headers) == 0 { 649 p.log.Warn("Empty head header set") 650 return 0, errEmptyHeaderSet 651 } 652 // Make sure the peer's reply conforms to the request 653 for i := 0; i < len(headers); i++ { 654 if number := headers[i].Number.Int64(); number != from+int64(i)*16 { 655 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", from+int64(i)*16, "received", number) 656 return 0, errInvalidChain 657 } 658 } 659 // Check if a common ancestor was found 660 finished = true 661 for i := len(headers) - 1; i >= 0; i-- { 662 // Skip any headers that underflow/overflow our requested set 663 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > ceil { 664 continue 665 } 666 // Otherwise check if we already know the header or not 667 h := headers[i].Hash() 668 n := headers[i].Number.Uint64() 669 if (d.mode == FullSync && d.blockchain.HasBlock(h, n)) || (d.mode != FullSync && d.lightchain.HasHeader(h, n)) { 670 number, hash = n, h 671 672 // If every header is known, even future ones, the peer straight out lied about its head 673 if number > height && i == limit-1 { 674 p.log.Warn("Lied about chain head", "reported", height, "found", number) 675 return 0, errStallingPeer 676 } 677 break 678 } 679 } 680 681 case <-timeout: 682 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 683 return 0, errTimeout 684 685 case <-d.bodyCh: 686 case <-d.receiptCh: 687 // Out of bounds delivery, ignore 688 } 689 } 690 // If the head fetch already found an ancestor, return 691 if hash != (common.Hash{}) { 692 if int64(number) <= floor { 693 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 694 return 0, errInvalidAncestor 695 } 696 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 697 return number, nil 698 } 699 // Ancestor not found, we need to binary search over our chain 700 start, end := uint64(0), head 701 if floor > 0 { 702 start = uint64(floor) 703 } 704 for start+1 < end { 705 // Split our chain interval in two, and request the hash to cross check 706 check := (start + end) / 2 707 708 ttl := d.requestTTL() 709 timeout := time.After(ttl) 710 711 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 712 713 // Wait until a reply arrives to this request 714 for arrived := false; !arrived; { 715 select { 716 case <-d.cancelCh: 717 return 0, errCancelHeaderFetch 718 719 case packer := <-d.headerCh: 720 // Discard anything not from the origin peer 721 if packer.PeerId() != p.id { 722 log.Debug("Received headers from incorrect peer", "peer", packer.PeerId()) 723 break 724 } 725 // Make sure the peer actually gave something valid 726 headers := packer.(*headerPack).headers 727 if len(headers) != 1 { 728 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 729 return 0, errBadPeer 730 } 731 arrived = true 732 733 // Modify the search interval based on the response 734 h := headers[0].Hash() 735 n := headers[0].Number.Uint64() 736 if (d.mode == FullSync && !d.blockchain.HasBlock(h, n)) || (d.mode != FullSync && !d.lightchain.HasHeader(h, n)) { 737 end = check 738 break 739 } 740 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 741 if header.Number.Uint64() != check { 742 p.log.Debug("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 743 return 0, errBadPeer 744 } 745 start = check 746 hash = h 747 748 case <-timeout: 749 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 750 return 0, errTimeout 751 752 case <-d.bodyCh: 753 case <-d.receiptCh: 754 // Out of bounds delivery, ignore 755 } 756 } 757 } 758 // Ensure valid ancestry and return 759 if int64(start) <= floor { 760 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 761 return 0, errInvalidAncestor 762 } 763 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 764 return start, nil 765 } 766 767 // fetchHeaders keeps retrieving headers concurrently from the number 768 // requested, until no more are returned, potentially throttling on the way. To 769 // facilitate concurrency but still protect against malicious nodes sending bad 770 // headers, we construct a header chain skeleton using the "origin" peer we are 771 // syncing with, and fill in the missing headers using anyone else. Headers from 772 // other peers are only accepted if they map cleanly to the skeleton. If no one 773 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 774 // the origin is dropped. 775 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64, pivot uint64) error { 776 p.log.Debug("Directing header downloads", "origin", from) 777 defer p.log.Debug("Header download terminated") 778 779 // Create a timeout timer, and the associated header fetcher 780 skeleton := true // Skeleton assembly phase or finishing up 781 request := time.Now() // time of the last skeleton fetch request 782 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 783 <-timeout.C // timeout channel should be initially empty 784 defer timeout.Stop() 785 786 var ttl time.Duration 787 getHeaders := func(from uint64) { 788 request = time.Now() 789 790 ttl = d.requestTTL() 791 timeout.Reset(ttl) 792 793 if skeleton { 794 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 795 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 796 } else { 797 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 798 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 799 } 800 } 801 // Start pulling the header chain skeleton until all is done 802 getHeaders(from) 803 804 for { 805 select { 806 case <-d.cancelCh: 807 return errCancelHeaderFetch 808 809 case packet := <-d.headerCh: 810 // Make sure the active peer is giving us the skeleton headers 811 if packet.PeerId() != p.id { 812 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 813 break 814 } 815 headerReqTimer.UpdateSince(request) 816 timeout.Stop() 817 818 // If the skeleton's finished, pull any remaining head headers directly from the origin 819 if packet.Items() == 0 && skeleton { 820 skeleton = false 821 getHeaders(from) 822 continue 823 } 824 // If no more headers are inbound, notify the content fetchers and return 825 if packet.Items() == 0 { 826 // Don't abort header fetches while the pivot is downloading 827 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 828 p.log.Debug("No headers, waiting for pivot commit") 829 select { 830 case <-time.After(fsHeaderContCheck): 831 getHeaders(from) 832 continue 833 case <-d.cancelCh: 834 return errCancelHeaderFetch 835 } 836 } 837 // Pivot done (or not in fast sync) and no more headers, terminate the process 838 p.log.Debug("No more headers available") 839 select { 840 case d.headerProcCh <- nil: 841 return nil 842 case <-d.cancelCh: 843 return errCancelHeaderFetch 844 } 845 } 846 headers := packet.(*headerPack).headers 847 848 // If we received a skeleton batch, resolve internals concurrently 849 if skeleton { 850 filled, proced, err := d.fillHeaderSkeleton(from, headers) 851 if err != nil { 852 p.log.Debug("Skeleton chain invalid", "err", err) 853 return errInvalidChain 854 } 855 headers = filled[proced:] 856 from += uint64(proced) 857 } else { 858 // If we're closing in on the chain head, but haven't yet reached it, delay 859 // the last few headers so mini reorgs on the head don't cause invalid hash 860 // chain errors. 861 if n := len(headers); n > 0 { 862 // Retrieve the current head we're at 863 head := uint64(0) 864 if d.mode == LightSync { 865 head = d.lightchain.CurrentHeader().Number.Uint64() 866 } else { 867 head = d.blockchain.CurrentFastBlock().NumberU64() 868 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 869 head = full 870 } 871 } 872 // If the head is way older than this batch, delay the last few headers 873 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 874 delay := reorgProtHeaderDelay 875 if delay > n { 876 delay = n 877 } 878 headers = headers[:n-delay] 879 } 880 } 881 } 882 // Insert all the new headers and fetch the next batch 883 if len(headers) > 0 { 884 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 885 select { 886 case d.headerProcCh <- headers: 887 case <-d.cancelCh: 888 return errCancelHeaderFetch 889 } 890 from += uint64(len(headers)) 891 getHeaders(from) 892 } else { 893 // No headers delivered, or all of them being delayed, sleep a bit and retry 894 p.log.Trace("All headers delayed, waiting") 895 select { 896 case <-time.After(fsHeaderContCheck): 897 getHeaders(from) 898 continue 899 case <-d.cancelCh: 900 return errCancelHeaderFetch 901 } 902 } 903 904 case <-timeout.C: 905 if d.dropPeer == nil { 906 // The dropPeer method is nil when `--copydb` is used for a local copy. 907 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 908 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 909 break 910 } 911 // Header retrieval timed out, consider the peer bad and drop 912 p.log.Debug("Header request timed out", "elapsed", ttl) 913 headerTimeoutMeter.Mark(1) 914 d.dropPeer(p.id) 915 916 // Finish the sync gracefully instead of dumping the gathered data though 917 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 918 select { 919 case ch <- false: 920 case <-d.cancelCh: 921 } 922 } 923 select { 924 case d.headerProcCh <- nil: 925 case <-d.cancelCh: 926 } 927 return errBadPeer 928 } 929 } 930 } 931 932 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 933 // and maps them to the provided skeleton header chain. 934 // 935 // Any partial results from the beginning of the skeleton is (if possible) forwarded 936 // immediately to the header processor to keep the rest of the pipeline full even 937 // in the case of header stalls. 938 // 939 // The method returns the entire filled skeleton and also the number of headers 940 // already forwarded for processing. 941 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 942 log.Debug("Filling up skeleton", "from", from) 943 d.queue.ScheduleSkeleton(from, skeleton) 944 945 var ( 946 deliver = func(packet dataPack) (int, error) { 947 pack := packet.(*headerPack) 948 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 949 } 950 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 951 throttle = func() bool { return false } 952 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, error) { 953 return d.queue.ReserveHeaders(p, count), false, nil 954 } 955 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 956 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 957 setIdle = func(p *peerConnection, accepted int) { p.SetHeadersIdle(accepted) } 958 ) 959 err := d.fetchParts(errCancelHeaderFetch, d.headerCh, deliver, d.queue.headerContCh, expire, 960 d.queue.PendingHeaders, d.queue.InFlightHeaders, throttle, reserve, 961 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 962 963 log.Debug("Skeleton fill terminated", "err", err) 964 965 filled, proced := d.queue.RetrieveHeaders() 966 return filled, proced, err 967 } 968 969 // fetchBodies iteratively downloads the scheduled block bodies, taking any 970 // available peers, reserving a chunk of blocks for each, waiting for delivery 971 // and also periodically checking for timeouts. 972 func (d *Downloader) fetchBodies(from uint64) error { 973 log.Debug("Downloading block bodies", "origin", from) 974 975 var ( 976 deliver = func(packet dataPack) (int, error) { 977 pack := packet.(*bodyPack) 978 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 979 } 980 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 981 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 982 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 983 setIdle = func(p *peerConnection, accepted int) { p.SetBodiesIdle(accepted) } 984 ) 985 err := d.fetchParts(errCancelBodyFetch, d.bodyCh, deliver, d.bodyWakeCh, expire, 986 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ShouldThrottleBlocks, d.queue.ReserveBodies, 987 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 988 989 log.Debug("Block body download terminated", "err", err) 990 return err 991 } 992 993 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 994 // available peers, reserving a chunk of receipts for each, waiting for delivery 995 // and also periodically checking for timeouts. 996 func (d *Downloader) fetchReceipts(from uint64) error { 997 log.Debug("Downloading transaction receipts", "origin", from) 998 999 var ( 1000 deliver = func(packet dataPack) (int, error) { 1001 pack := packet.(*receiptPack) 1002 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1003 } 1004 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1005 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1006 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1007 setIdle = func(p *peerConnection, accepted int) { p.SetReceiptsIdle(accepted) } 1008 ) 1009 err := d.fetchParts(errCancelReceiptFetch, d.receiptCh, deliver, d.receiptWakeCh, expire, 1010 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ShouldThrottleReceipts, d.queue.ReserveReceipts, 1011 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1012 1013 log.Debug("Transaction receipt download terminated", "err", err) 1014 return err 1015 } 1016 1017 // fetchParts iteratively downloads scheduled block parts, taking any available 1018 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1019 // also periodically checking for timeouts. 1020 // 1021 // As the scheduling/timeout logic mostly is the same for all downloaded data 1022 // types, this method is used by each for data gathering and is instrumented with 1023 // various callbacks to handle the slight differences between processing them. 1024 // 1025 // The instrumentation parameters: 1026 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1027 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1028 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1029 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1030 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1031 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1032 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1033 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1034 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1035 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1036 // - fetch: network callback to actually send a particular download request to a physical remote peer 1037 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1038 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1039 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1040 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1041 // - kind: textual label of the type being downloaded to display in log mesages 1042 func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1043 expire func() map[string]int, pending func() int, inFlight func() bool, throttle func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, error), 1044 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1045 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int), kind string) error { 1046 1047 // Create a ticker to detect expired retrieval tasks 1048 ticker := time.NewTicker(100 * time.Millisecond) 1049 defer ticker.Stop() 1050 1051 update := make(chan struct{}, 1) 1052 1053 // Prepare the queue and fetch block parts until the block header fetcher's done 1054 finished := false 1055 for { 1056 select { 1057 case <-d.cancelCh: 1058 return errCancel 1059 1060 case packet := <-deliveryCh: 1061 // If the peer was previously banned and failed to deliver its pack 1062 // in a reasonable time frame, ignore its message. 1063 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1064 // Deliver the received chunk of data and check chain validity 1065 accepted, err := deliver(packet) 1066 if err == errInvalidChain { 1067 return err 1068 } 1069 // Unless a peer delivered something completely else than requested (usually 1070 // caused by a timed out request which came through in the end), set it to 1071 // idle. If the delivery's stale, the peer should have already been idled. 1072 if err != errStaleDelivery { 1073 setIdle(peer, accepted) 1074 } 1075 // Issue a log to the user to see what's going on 1076 switch { 1077 case err == nil && packet.Items() == 0: 1078 peer.log.Trace("Requested data not delivered", "type", kind) 1079 case err == nil: 1080 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1081 default: 1082 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1083 } 1084 } 1085 // Blocks assembled, try to update the progress 1086 select { 1087 case update <- struct{}{}: 1088 default: 1089 } 1090 1091 case cont := <-wakeCh: 1092 // The header fetcher sent a continuation flag, check if it's done 1093 if !cont { 1094 finished = true 1095 } 1096 // Headers arrive, try to update the progress 1097 select { 1098 case update <- struct{}{}: 1099 default: 1100 } 1101 1102 case <-ticker.C: 1103 // Sanity check update the progress 1104 select { 1105 case update <- struct{}{}: 1106 default: 1107 } 1108 1109 case <-update: 1110 // Short circuit if we lost all our peers 1111 if d.peers.Len() == 0 { 1112 return errNoPeers 1113 } 1114 // Check for fetch request timeouts and demote the responsible peers 1115 for pid, fails := range expire() { 1116 if peer := d.peers.Peer(pid); peer != nil { 1117 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1118 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1119 // out that sync wise we need to get rid of the peer. 1120 // 1121 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1122 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1123 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1124 if fails > 2 { 1125 peer.log.Trace("Data delivery timed out", "type", kind) 1126 setIdle(peer, 0) 1127 } else { 1128 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1129 if d.dropPeer == nil { 1130 // The dropPeer method is nil when `--copydb` is used for a local copy. 1131 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1132 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1133 } else { 1134 d.dropPeer(pid) 1135 } 1136 } 1137 } 1138 } 1139 // If there's nothing more to fetch, wait or terminate 1140 if pending() == 0 { 1141 if !inFlight() && finished { 1142 log.Debug("Data fetching completed", "type", kind) 1143 return nil 1144 } 1145 break 1146 } 1147 // Send a download request to all idle peers, until throttled 1148 progressed, throttled, running := false, false, inFlight() 1149 idles, total := idle() 1150 1151 for _, peer := range idles { 1152 // Short circuit if throttling activated 1153 if throttle() { 1154 throttled = true 1155 break 1156 } 1157 // Short circuit if there is no more available task. 1158 if pending() == 0 { 1159 break 1160 } 1161 // Reserve a chunk of fetches for a peer. A nil can mean either that 1162 // no more headers are available, or that the peer is known not to 1163 // have them. 1164 request, progress, err := reserve(peer, capacity(peer)) 1165 if err != nil { 1166 return err 1167 } 1168 if progress { 1169 progressed = true 1170 } 1171 if request == nil { 1172 continue 1173 } 1174 if request.From > 0 { 1175 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1176 } else { 1177 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1178 } 1179 // Fetch the chunk and make sure any errors return the hashes to the queue 1180 if fetchHook != nil { 1181 fetchHook(request.Headers) 1182 } 1183 if err := fetch(peer, request); err != nil { 1184 // Although we could try and make an attempt to fix this, this error really 1185 // means that we've double allocated a fetch task to a peer. If that is the 1186 // case, the internal state of the downloader and the queue is very wrong so 1187 // better hard crash and note the error instead of silently accumulating into 1188 // a much bigger issue. 1189 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1190 } 1191 running = true 1192 } 1193 // Make sure that we have peers available for fetching. If all peers have been tried 1194 // and all failed throw an error 1195 if !progressed && !throttled && !running && len(idles) == total && pending() > 0 { 1196 return errPeersUnavailable 1197 } 1198 } 1199 } 1200 } 1201 1202 // processHeaders takes batches of retrieved headers from an input channel and 1203 // keeps processing and scheduling them into the header chain and downloader's 1204 // queue until the stream ends or a failure occurs. 1205 func (d *Downloader) processHeaders(origin uint64, pivot uint64, td *big.Int) error { 1206 // Keep a count of uncertain headers to roll back 1207 rollback := []*types.Header{} 1208 defer func() { 1209 if len(rollback) > 0 { 1210 // Flatten the headers and roll them back 1211 hashes := make([]common.Hash, len(rollback)) 1212 for i, header := range rollback { 1213 hashes[i] = header.Hash() 1214 } 1215 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1216 if d.mode != LightSync { 1217 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1218 lastBlock = d.blockchain.CurrentBlock().Number() 1219 } 1220 d.lightchain.Rollback(hashes) 1221 curFastBlock, curBlock := common.Big0, common.Big0 1222 if d.mode != LightSync { 1223 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1224 curBlock = d.blockchain.CurrentBlock().Number() 1225 } 1226 log.Warn("Rolled back headers", "count", len(hashes), 1227 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1228 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1229 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock)) 1230 } 1231 }() 1232 1233 // Wait for batches of headers to process 1234 gotHeaders := false 1235 1236 for { 1237 select { 1238 case <-d.cancelCh: 1239 return errCancelHeaderProcessing 1240 1241 case headers := <-d.headerProcCh: 1242 // Terminate header processing if we synced up 1243 if len(headers) == 0 { 1244 // Notify everyone that headers are fully processed 1245 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1246 select { 1247 case ch <- false: 1248 case <-d.cancelCh: 1249 } 1250 } 1251 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1252 // better chain compared to ours. The only exception is if its promised blocks were 1253 // already imported by other means (e.g. fetcher): 1254 // 1255 // R <remote peer>, L <local node>: Both at block 10 1256 // R: Mine block 11, and propagate it to L 1257 // L: Queue block 11 for import 1258 // L: Notice that R's head and TD increased compared to ours, start sync 1259 // L: Import of block 11 finishes 1260 // L: Sync begins, and finds common ancestor at 11 1261 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1262 // R: Nothing to give 1263 if d.mode != LightSync { 1264 head := d.blockchain.CurrentBlock() 1265 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1266 return errStallingPeer 1267 } 1268 } 1269 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1270 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1271 // of delivering the post-pivot blocks that would flag the invalid content. 1272 // 1273 // This check cannot be executed "as is" for full imports, since blocks may still be 1274 // queued for processing when the header download completes. However, as long as the 1275 // peer gave us something useful, we're already happy/progressed (above check). 1276 if d.mode == FastSync || d.mode == LightSync { 1277 head := d.lightchain.CurrentHeader() 1278 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1279 return errStallingPeer 1280 } 1281 } 1282 // Disable any rollback and return 1283 rollback = nil 1284 return nil 1285 } 1286 // Otherwise split the chunk of headers into batches and process them 1287 gotHeaders = true 1288 1289 for len(headers) > 0 { 1290 // Terminate if something failed in between processing chunks 1291 select { 1292 case <-d.cancelCh: 1293 return errCancelHeaderProcessing 1294 default: 1295 } 1296 // Select the next chunk of headers to import 1297 limit := maxHeadersProcess 1298 if limit > len(headers) { 1299 limit = len(headers) 1300 } 1301 chunk := headers[:limit] 1302 1303 // In case of header only syncing, validate the chunk immediately 1304 if d.mode == FastSync || d.mode == LightSync { 1305 // Collect the yet unknown headers to mark them as uncertain 1306 unknown := make([]*types.Header, 0, len(headers)) 1307 for _, header := range chunk { 1308 if !d.lightchain.HasHeader(header.Hash(), header.Number.Uint64()) { 1309 unknown = append(unknown, header) 1310 } 1311 } 1312 // If we're importing pure headers, verify based on their recentness 1313 frequency := fsHeaderCheckFrequency 1314 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1315 frequency = 1 1316 } 1317 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1318 // If some headers were inserted, add them too to the rollback list 1319 if n > 0 { 1320 rollback = append(rollback, chunk[:n]...) 1321 } 1322 log.Debug("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "err", err) 1323 return errInvalidChain 1324 } 1325 // All verifications passed, store newly found uncertain headers 1326 rollback = append(rollback, unknown...) 1327 if len(rollback) > fsHeaderSafetyNet { 1328 rollback = append(rollback[:0], rollback[len(rollback)-fsHeaderSafetyNet:]...) 1329 } 1330 } 1331 // Unless we're doing light chains, schedule the headers for associated content retrieval 1332 if d.mode == FullSync || d.mode == FastSync || d.mode == BoundedFullSync { 1333 // If we've reached the allowed number of pending headers, stall a bit 1334 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1335 select { 1336 case <-d.cancelCh: 1337 return errCancelHeaderProcessing 1338 case <-time.After(time.Second): 1339 } 1340 } 1341 // Otherwise insert the headers for content retrieval 1342 inserts := d.queue.Schedule(chunk, origin) 1343 if len(inserts) != len(chunk) { 1344 log.Debug("Stale headers") 1345 return errBadPeer 1346 } 1347 } 1348 headers = headers[limit:] 1349 origin += uint64(limit) 1350 } 1351 1352 // Update the highest block number we know if a higher one is found. 1353 d.syncStatsLock.Lock() 1354 if d.syncStatsChainHeight < origin { 1355 d.syncStatsChainHeight = origin - 1 1356 } 1357 d.syncStatsLock.Unlock() 1358 1359 // Signal the content downloaders of the availablility of new tasks 1360 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1361 select { 1362 case ch <- true: 1363 default: 1364 } 1365 } 1366 } 1367 } 1368 } 1369 1370 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1371 func (d *Downloader) processFullSyncContent() error { 1372 for { 1373 results := d.queue.Results(true) 1374 if len(results) == 0 { 1375 return nil 1376 } 1377 if d.chainInsertHook != nil { 1378 d.chainInsertHook(results) 1379 } 1380 if err := d.importBlockResults(results); err != nil { 1381 return err 1382 } 1383 } 1384 } 1385 1386 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1387 // Check for any early termination requests 1388 if len(results) == 0 { 1389 return nil 1390 } 1391 select { 1392 case <-d.quitCh: 1393 return errCancelContentProcessing 1394 default: 1395 } 1396 // Retrieve the a batch of results to import 1397 first, last := results[0].Header, results[len(results)-1].Header 1398 log.Debug("Inserting downloaded chain", "items", len(results), 1399 "firstnum", first.Number, "firsthash", first.Hash(), 1400 "lastnum", last.Number, "lasthash", last.Hash(), 1401 ) 1402 blocks := make([]*types.Block, len(results)) 1403 for i, result := range results { 1404 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1405 } 1406 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1407 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1408 return errInvalidChain 1409 } 1410 return nil 1411 } 1412 1413 // processFastSyncContent takes fetch results from the queue and writes them to the 1414 // database. It also controls the synchronisation of state nodes of the pivot block. 1415 func (d *Downloader) processFastSyncContent(latest *types.Header) error { 1416 // Start syncing state of the reported head block. This should get us most of 1417 // the state of the pivot block. 1418 stateSync := d.syncState(latest.Root) 1419 defer stateSync.Cancel() 1420 go func() { 1421 if err := stateSync.Wait(); err != nil && err != errCancelStateFetch { 1422 d.queue.Close() // wake up Results 1423 } 1424 }() 1425 // Figure out the ideal pivot block. Note, that this goalpost may move if the 1426 // sync takes long enough for the chain head to move significantly. 1427 pivot := uint64(0) 1428 if height := latest.Number.Uint64(); height > uint64(fsMinFullBlocks) { 1429 pivot = height - uint64(fsMinFullBlocks) 1430 } 1431 // To cater for moving pivot points, track the pivot block and subsequently 1432 // accumulated download results separately. 1433 var ( 1434 oldPivot *fetchResult // Locked in pivot block, might change eventually 1435 oldTail []*fetchResult // Downloaded content after the pivot 1436 ) 1437 for { 1438 // Wait for the next batch of downloaded data to be available, and if the pivot 1439 // block became stale, move the goalpost 1440 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1441 if len(results) == 0 { 1442 // If pivot sync is done, stop 1443 if oldPivot == nil { 1444 return stateSync.Cancel() 1445 } 1446 // If sync failed, stop 1447 select { 1448 case <-d.cancelCh: 1449 return stateSync.Cancel() 1450 default: 1451 } 1452 } 1453 if d.chainInsertHook != nil { 1454 d.chainInsertHook(results) 1455 } 1456 if oldPivot != nil { 1457 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1458 } 1459 // Split around the pivot block and process the two sides via fast/full sync 1460 if atomic.LoadInt32(&d.committed) == 0 { 1461 latest = results[len(results)-1].Header 1462 if height := latest.Number.Uint64(); height > pivot+2*uint64(fsMinFullBlocks) { 1463 log.Warn("Pivot became stale, moving", "old", pivot, "new", height-uint64(fsMinFullBlocks)) 1464 pivot = height - uint64(fsMinFullBlocks) 1465 } 1466 } 1467 P, beforeP, afterP := splitAroundPivot(pivot, results) 1468 if err := d.commitFastSyncData(beforeP, stateSync); err != nil { 1469 return err 1470 } 1471 if P != nil { 1472 // If new pivot block found, cancel old state retrieval and restart 1473 if oldPivot != P { 1474 stateSync.Cancel() 1475 1476 stateSync = d.syncState(P.Header.Root) 1477 defer stateSync.Cancel() 1478 go func() { 1479 if err := stateSync.Wait(); err != nil && err != errCancelStateFetch { 1480 d.queue.Close() // wake up Results 1481 } 1482 }() 1483 oldPivot = P 1484 } 1485 // Wait for completion, occasionally checking for pivot staleness 1486 select { 1487 case <-stateSync.done: 1488 if stateSync.err != nil { 1489 return stateSync.err 1490 } 1491 if err := d.commitPivotBlock(P); err != nil { 1492 return err 1493 } 1494 oldPivot = nil 1495 1496 case <-time.After(time.Second): 1497 oldTail = afterP 1498 continue 1499 } 1500 } 1501 // Fast sync done, pivot commit done, full import 1502 if err := d.importBlockResults(afterP); err != nil { 1503 return err 1504 } 1505 } 1506 } 1507 1508 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1509 for _, result := range results { 1510 num := result.Header.Number.Uint64() 1511 switch { 1512 case num < pivot: 1513 before = append(before, result) 1514 case num == pivot: 1515 p = result 1516 default: 1517 after = append(after, result) 1518 } 1519 } 1520 return p, before, after 1521 } 1522 1523 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1524 // Check for any early termination requests 1525 if len(results) == 0 { 1526 return nil 1527 } 1528 select { 1529 case <-d.quitCh: 1530 return errCancelContentProcessing 1531 case <-stateSync.done: 1532 if err := stateSync.Wait(); err != nil { 1533 return err 1534 } 1535 default: 1536 } 1537 // Retrieve the a batch of results to import 1538 first, last := results[0].Header, results[len(results)-1].Header 1539 log.Debug("Inserting fast-sync blocks", "items", len(results), 1540 "firstnum", first.Number, "firsthash", first.Hash(), 1541 "lastnumn", last.Number, "lasthash", last.Hash(), 1542 ) 1543 blocks := make([]*types.Block, len(results)) 1544 receipts := make([]types.Receipts, len(results)) 1545 for i, result := range results { 1546 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1547 receipts[i] = result.Receipts 1548 } 1549 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts); err != nil { 1550 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1551 return errInvalidChain 1552 } 1553 return nil 1554 } 1555 1556 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1557 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1558 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1559 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}); err != nil { 1560 return err 1561 } 1562 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1563 return err 1564 } 1565 atomic.StoreInt32(&d.committed, 1) 1566 return nil 1567 } 1568 1569 // DeliverHeaders injects a new batch of block headers received from a remote 1570 // node into the download schedule. 1571 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) { 1572 return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1573 } 1574 1575 // DeliverBodies injects a new batch of block bodies received from a remote node. 1576 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) { 1577 return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1578 } 1579 1580 // DeliverReceipts injects a new batch of receipts received from a remote node. 1581 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) { 1582 return d.deliver(id, d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1583 } 1584 1585 // DeliverNodeData injects a new batch of node state data received from a remote node. 1586 func (d *Downloader) DeliverNodeData(id string, data [][]byte) (err error) { 1587 return d.deliver(id, d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1588 } 1589 1590 // deliver injects a new batch of data received from a remote node. 1591 func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1592 // Update the delivery metrics for both good and failed deliveries 1593 inMeter.Mark(int64(packet.Items())) 1594 defer func() { 1595 if err != nil { 1596 dropMeter.Mark(int64(packet.Items())) 1597 } 1598 }() 1599 // Deliver or abort if the sync is canceled while queuing 1600 d.cancelLock.RLock() 1601 cancel := d.cancelCh 1602 d.cancelLock.RUnlock() 1603 if cancel == nil { 1604 return errNoSyncActive 1605 } 1606 select { 1607 case destCh <- packet: 1608 return nil 1609 case <-cancel: 1610 return errNoSyncActive 1611 } 1612 } 1613 1614 // qosTuner is the quality of service tuning loop that occasionally gathers the 1615 // peer latency statistics and updates the estimated request round trip time. 1616 func (d *Downloader) qosTuner() { 1617 for { 1618 // Retrieve the current median RTT and integrate into the previoust target RTT 1619 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 1620 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 1621 1622 // A new RTT cycle passed, increase our confidence in the estimated RTT 1623 conf := atomic.LoadUint64(&d.rttConfidence) 1624 conf = conf + (1000000-conf)/2 1625 atomic.StoreUint64(&d.rttConfidence, conf) 1626 1627 // Log the new QoS values and sleep until the next RTT 1628 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1629 select { 1630 case <-d.quitCh: 1631 return 1632 case <-time.After(rtt): 1633 } 1634 } 1635 } 1636 1637 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 1638 // peer set, needing to reduce the confidence we have in out QoS estimates. 1639 func (d *Downloader) qosReduceConfidence() { 1640 // If we have a single peer, confidence is always 1 1641 peers := uint64(d.peers.Len()) 1642 if peers == 0 { 1643 // Ensure peer connectivity races don't catch us off guard 1644 return 1645 } 1646 if peers == 1 { 1647 atomic.StoreUint64(&d.rttConfidence, 1000000) 1648 return 1649 } 1650 // If we have a ton of peers, don't drop confidence) 1651 if peers >= uint64(qosConfidenceCap) { 1652 return 1653 } 1654 // Otherwise drop the confidence factor 1655 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 1656 if float64(conf)/1000000 < rttMinConfidence { 1657 conf = uint64(rttMinConfidence * 1000000) 1658 } 1659 atomic.StoreUint64(&d.rttConfidence, conf) 1660 1661 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1662 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1663 } 1664 1665 // requestRTT returns the current target round trip time for a download request 1666 // to complete in. 1667 // 1668 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 1669 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 1670 // be adapted to, but smaller ones are preferred (stabler download stream). 1671 func (d *Downloader) requestRTT() time.Duration { 1672 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 1673 } 1674 1675 // requestTTL returns the current timeout allowance for a single download request 1676 // to finish under. 1677 func (d *Downloader) requestTTL() time.Duration { 1678 var ( 1679 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1680 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 1681 ) 1682 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 1683 if ttl > ttlLimit { 1684 ttl = ttlLimit 1685 } 1686 return ttl 1687 } 1688 1689 // Extra downloader functionality for non-proof-of-work consensus 1690 1691 // Synchronizes with a peer, but only up to the provided Hash 1692 func (d *Downloader) syncWithPeerUntil(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 1693 d.mux.Post(StartEvent{}) 1694 defer func() { 1695 // reset on error 1696 if err != nil { 1697 d.mux.Post(FailedEvent{err}) 1698 } else { 1699 d.mux.Post(DoneEvent{}) 1700 } 1701 }() 1702 if p.version < 62 { 1703 return errTooOld 1704 } 1705 1706 log.Info("Synchronising with the network", "id", p.id, "version", p.version) 1707 defer func(start time.Time) { 1708 log.Info("Synchronisation terminated", "duration", time.Since(start)) 1709 }(time.Now()) 1710 1711 remoteHeader, err := d.fetchHeader(p, hash) 1712 if err != nil { 1713 return err 1714 } 1715 1716 remoteHeight := remoteHeader.Number.Uint64() 1717 localHeight := d.blockchain.CurrentBlock().NumberU64() 1718 1719 d.syncStatsLock.Lock() 1720 if d.syncStatsChainHeight <= localHeight || d.syncStatsChainOrigin > localHeight { 1721 d.syncStatsChainOrigin = localHeight 1722 } 1723 d.syncStatsChainHeight = remoteHeight 1724 d.syncStatsLock.Unlock() 1725 1726 d.queue.Prepare(localHeight+1, d.mode) 1727 if d.syncInitHook != nil { 1728 d.syncInitHook(localHeight, remoteHeight) 1729 } 1730 1731 pivot := uint64(0) 1732 1733 fetchers := []func() error{ 1734 func() error { return d.fetchBoundedHeaders(p, localHeight+1, remoteHeight) }, 1735 func() error { return d.fetchBodies(localHeight + 1) }, 1736 func() error { return d.fetchReceipts(localHeight + 1) }, // Receipts are only retrieved during fast sync 1737 func() error { return d.processHeaders(localHeight+1, pivot, td) }, 1738 d.processFullSyncContent, //This must be added to clear the buffer of downloaded content as it's being filled 1739 } 1740 return d.spawnSync(fetchers) 1741 } 1742 1743 // Fetches a single header from a peer 1744 func (d *Downloader) fetchHeader(p *peerConnection, hash common.Hash) (*types.Header, error) { 1745 log.Info("retrieving remote chain height", "peer", p) 1746 1747 go p.peer.RequestHeadersByHash(hash, 1, 0, false) 1748 1749 timeout := time.After(d.requestTTL()) 1750 for { 1751 select { 1752 case <-d.cancelCh: 1753 return nil, errCancelBlockFetch 1754 1755 case packet := <-d.headerCh: 1756 // Discard anything not from the origin peer 1757 if packet.PeerId() != p.id { 1758 log.Info("Received headers from incorrect peer", "peer id", packet.PeerId()) 1759 break 1760 } 1761 // Make sure the peer actually gave something valid 1762 headers := packet.(*headerPack).headers 1763 if len(headers) != 1 { 1764 log.Info("invalid number of head headers (!= 1)", "peer", p, "len(headers)", len(headers)) 1765 return nil, errBadPeer 1766 } 1767 return headers[0], nil 1768 1769 case <-timeout: 1770 log.Info("head header timeout", "peer", p) 1771 return nil, errTimeout 1772 1773 case <-d.bodyCh: 1774 case <-d.stateCh: 1775 case <-d.receiptCh: 1776 // Out of bounds delivery, ignore 1777 } 1778 } 1779 } 1780 1781 // Not defined in go's stdlib: 1782 func minInt(a, b int) int { 1783 if a < b { 1784 return a 1785 } 1786 return b 1787 } 1788 1789 // Fetches headers between `from` and `to`, inclusive. 1790 // Assumes invariant: from <= to. 1791 func (d *Downloader) fetchBoundedHeaders(p *peerConnection, from uint64, to uint64) error { 1792 log.Info("directing header downloads", "peer", p, "from", from, "to", to) 1793 defer log.Info("header download terminated", "peer", p) 1794 1795 // Create a timeout timer, and the associated header fetcher 1796 skeleton := true // Skeleton assembly phase or finishing up 1797 request := time.Now() // time of the last skeleton fetch request 1798 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 1799 <-timeout.C // timeout channel should be initially empty 1800 defer timeout.Stop() 1801 1802 getHeaders := func(from uint64) { 1803 request = time.Now() 1804 timeout.Reset(d.requestTTL()) 1805 1806 skeletonStart := from + uint64(MaxHeaderFetch) - 1 1807 1808 if skeleton { 1809 if skeletonStart > to { 1810 skeleton = false 1811 } 1812 } 1813 1814 if skeleton { 1815 numSkeletonHeaders := minInt(MaxSkeletonSize, (int(to-from)+1)/MaxHeaderFetch) 1816 log.Trace("fetching skeleton headers", "peer", p, "num skeleton headers", numSkeletonHeaders, "from", from) 1817 go p.peer.RequestHeadersByNumber(skeletonStart, numSkeletonHeaders, MaxHeaderFetch-1, false) 1818 } else { 1819 // There are not enough headers remaining to warrant a skeleton fetch. 1820 // Grab all of the remaining headers. 1821 1822 numHeaders := int(to-from) + 1 1823 log.Trace("fetching full headers", "peer", p, "num headers", numHeaders, "from", from) 1824 go p.peer.RequestHeadersByNumber(from, numHeaders, 0, false) 1825 } 1826 } 1827 // Start pulling the header chain skeleton until all is done 1828 getHeaders(from) 1829 1830 for { 1831 select { 1832 case <-d.cancelCh: 1833 return errCancelHeaderFetch 1834 1835 case packet := <-d.headerCh: 1836 // Make sure the active peer is giving us the skeleton headers 1837 if packet.PeerId() != p.id { 1838 log.Info("Received headers from incorrect peer", "peer id", packet.PeerId()) 1839 break 1840 } 1841 headerReqTimer.UpdateSince(request) 1842 timeout.Stop() 1843 1844 headers := packet.(*headerPack).headers 1845 1846 // If we received a skeleton batch, resolve internals concurrently 1847 if skeleton { 1848 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1849 if err != nil { 1850 log.Debug("skeleton chain invalid", "peer", p, "err", err) 1851 return errInvalidChain 1852 } 1853 headers = filled[proced:] 1854 from += uint64(proced) 1855 } 1856 // Insert all the new headers and fetch the next batch 1857 if len(headers) > 0 { 1858 log.Trace("schedule headers", "peer", p, "num headers", len(headers), "from", from) 1859 select { 1860 case d.headerProcCh <- headers: 1861 case <-d.cancelCh: 1862 return errCancelHeaderFetch 1863 } 1864 from += uint64(len(headers)) 1865 } 1866 1867 if from <= to { 1868 getHeaders(from) 1869 } else { 1870 // Notify the content fetchers that no more headers are inbound and return. 1871 select { 1872 case d.headerProcCh <- nil: 1873 return nil 1874 case <-d.cancelCh: 1875 return errCancelHeaderFetch 1876 } 1877 } 1878 1879 case <-timeout.C: 1880 // Header retrieval timed out, consider the peer bad and drop 1881 log.Info("header request timed out", "peer", p) 1882 headerTimeoutMeter.Mark(1) 1883 d.dropPeer(p.id) 1884 1885 // Finish the sync gracefully instead of dumping the gathered data though 1886 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1887 select { 1888 case ch <- false: 1889 case <-d.cancelCh: 1890 } 1891 } 1892 select { 1893 case d.headerProcCh <- nil: 1894 case <-d.cancelCh: 1895 } 1896 return errBadPeer 1897 } 1898 } 1899 }