github.com/samgwo/go-ethereum@v1.8.2-0.20180302101319-49bcb5fbd55e/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/types" 31 "github.com/ethereum/go-ethereum/ethdb" 32 "github.com/ethereum/go-ethereum/event" 33 "github.com/ethereum/go-ethereum/log" 34 "github.com/ethereum/go-ethereum/metrics" 35 "github.com/ethereum/go-ethereum/params" 36 ) 37 38 var ( 39 MaxHashFetch = 512 // Amount of hashes to be fetched per retrieval request 40 MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request 41 MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request 42 MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly 43 MaxBodyFetch = 128 // Amount of block bodies to be fetched per retrieval request 44 MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request 45 MaxStateFetch = 384 // Amount of node state values to allow fetching per request 46 47 MaxForkAncestry = 3 * params.EpochDuration // Maximum chain reorganisation 48 rttMinEstimate = 2 * time.Second // Minimum round-trip time to target for download requests 49 rttMaxEstimate = 20 * time.Second // Maximum rount-trip time to target for download requests 50 rttMinConfidence = 0.1 // Worse confidence factor in our estimated RTT value 51 ttlScaling = 3 // Constant scaling factor for RTT -> TTL conversion 52 ttlLimit = time.Minute // Maximum TTL allowance to prevent reaching crazy timeouts 53 54 qosTuningPeers = 5 // Number of peers to tune based on (best peers) 55 qosConfidenceCap = 10 // Number of peers above which not to modify RTT confidence 56 qosTuningImpact = 0.25 // Impact that a new tuning target has on the previous value 57 58 maxQueuedHeaders = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection) 59 maxHeadersProcess = 2048 // Number of header download results to import at once into the chain 60 maxResultsProcess = 2048 // Number of content download results to import at once into the chain 61 62 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 63 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 64 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 65 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 66 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 67 ) 68 69 var ( 70 errBusy = errors.New("busy") 71 errUnknownPeer = errors.New("peer is unknown or unhealthy") 72 errBadPeer = errors.New("action from bad peer ignored") 73 errStallingPeer = errors.New("peer is stalling") 74 errNoPeers = errors.New("no peers to keep download active") 75 errTimeout = errors.New("timeout") 76 errEmptyHeaderSet = errors.New("empty header set by peer") 77 errPeersUnavailable = errors.New("no peers available or all tried for download") 78 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 79 errInvalidChain = errors.New("retrieved hash chain is invalid") 80 errInvalidBlock = errors.New("retrieved block is invalid") 81 errInvalidBody = errors.New("retrieved block body is invalid") 82 errInvalidReceipt = errors.New("retrieved receipt is invalid") 83 errCancelBlockFetch = errors.New("block download canceled (requested)") 84 errCancelHeaderFetch = errors.New("block header download canceled (requested)") 85 errCancelBodyFetch = errors.New("block body download canceled (requested)") 86 errCancelReceiptFetch = errors.New("receipt download canceled (requested)") 87 errCancelStateFetch = errors.New("state data download canceled (requested)") 88 errCancelHeaderProcessing = errors.New("header processing canceled (requested)") 89 errCancelContentProcessing = errors.New("content processing canceled (requested)") 90 errNoSyncActive = errors.New("no sync active") 91 errTooOld = errors.New("peer doesn't speak recent enough protocol version (need version >= 62)") 92 ) 93 94 type Downloader struct { 95 mode SyncMode // Synchronisation mode defining the strategy used (per sync cycle) 96 mux *event.TypeMux // Event multiplexer to announce sync operation events 97 98 queue *queue // Scheduler for selecting the hashes to download 99 peers *peerSet // Set of active peers from which download can proceed 100 stateDB ethdb.Database 101 102 rttEstimate uint64 // Round trip time to target for download requests 103 rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops) 104 105 // Statistics 106 syncStatsChainOrigin uint64 // Origin block number where syncing started at 107 syncStatsChainHeight uint64 // Highest block number known when syncing started 108 syncStatsState stateSyncStats 109 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 110 111 lightchain LightChain 112 blockchain BlockChain 113 114 // Callbacks 115 dropPeer peerDropFn // Drops a peer for misbehaving 116 117 // Status 118 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 119 synchronising int32 120 notified int32 121 committed int32 122 123 // Channels 124 headerCh chan dataPack // [eth/62] Channel receiving inbound block headers 125 bodyCh chan dataPack // [eth/62] Channel receiving inbound block bodies 126 receiptCh chan dataPack // [eth/63] Channel receiving inbound receipts 127 bodyWakeCh chan bool // [eth/62] Channel to signal the block body fetcher of new tasks 128 receiptWakeCh chan bool // [eth/63] Channel to signal the receipt fetcher of new tasks 129 headerProcCh chan []*types.Header // [eth/62] Channel to feed the header processor new tasks 130 131 // for stateFetcher 132 stateSyncStart chan *stateSync 133 trackStateReq chan *stateReq 134 stateCh chan dataPack // [eth/63] Channel receiving inbound node state data 135 136 // Cancellation and termination 137 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 138 cancelCh chan struct{} // Channel to cancel mid-flight syncs 139 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 140 141 quitCh chan struct{} // Quit channel to signal termination 142 quitLock sync.RWMutex // Lock to prevent double closes 143 144 // Testing hooks 145 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 146 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 147 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 148 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 149 } 150 151 // LightChain encapsulates functions required to synchronise a light chain. 152 type LightChain interface { 153 // HasHeader verifies a header's presence in the local chain. 154 HasHeader(common.Hash, uint64) bool 155 156 // GetHeaderByHash retrieves a header from the local chain. 157 GetHeaderByHash(common.Hash) *types.Header 158 159 // CurrentHeader retrieves the head header from the local chain. 160 CurrentHeader() *types.Header 161 162 // GetTd returns the total difficulty of a local block. 163 GetTd(common.Hash, uint64) *big.Int 164 165 // InsertHeaderChain inserts a batch of headers into the local chain. 166 InsertHeaderChain([]*types.Header, int) (int, error) 167 168 // Rollback removes a few recently added elements from the local chain. 169 Rollback([]common.Hash) 170 } 171 172 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 173 type BlockChain interface { 174 LightChain 175 176 // HasBlock verifies a block's presence in the local chain. 177 HasBlock(common.Hash, uint64) bool 178 179 // GetBlockByHash retrieves a block from the local chain. 180 GetBlockByHash(common.Hash) *types.Block 181 182 // CurrentBlock retrieves the head block from the local chain. 183 CurrentBlock() *types.Block 184 185 // CurrentFastBlock retrieves the head fast block from the local chain. 186 CurrentFastBlock() *types.Block 187 188 // FastSyncCommitHead directly commits the head block to a certain entity. 189 FastSyncCommitHead(common.Hash) error 190 191 // InsertChain inserts a batch of blocks into the local chain. 192 InsertChain(types.Blocks) (int, error) 193 194 // InsertReceiptChain inserts a batch of receipts into the local chain. 195 InsertReceiptChain(types.Blocks, []types.Receipts) (int, error) 196 } 197 198 // New creates a new downloader to fetch hashes and blocks from remote peers. 199 func New(mode SyncMode, stateDb ethdb.Database, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 200 if lightchain == nil { 201 lightchain = chain 202 } 203 204 dl := &Downloader{ 205 mode: mode, 206 stateDB: stateDb, 207 mux: mux, 208 queue: newQueue(), 209 peers: newPeerSet(), 210 rttEstimate: uint64(rttMaxEstimate), 211 rttConfidence: uint64(1000000), 212 blockchain: chain, 213 lightchain: lightchain, 214 dropPeer: dropPeer, 215 headerCh: make(chan dataPack, 1), 216 bodyCh: make(chan dataPack, 1), 217 receiptCh: make(chan dataPack, 1), 218 bodyWakeCh: make(chan bool, 1), 219 receiptWakeCh: make(chan bool, 1), 220 headerProcCh: make(chan []*types.Header, 1), 221 quitCh: make(chan struct{}), 222 stateCh: make(chan dataPack), 223 stateSyncStart: make(chan *stateSync), 224 trackStateReq: make(chan *stateReq), 225 } 226 go dl.qosTuner() 227 go dl.stateFetcher() 228 return dl 229 } 230 231 // Progress retrieves the synchronisation boundaries, specifically the origin 232 // block where synchronisation started at (may have failed/suspended); the block 233 // or header sync is currently at; and the latest known block which the sync targets. 234 // 235 // In addition, during the state download phase of fast synchronisation the number 236 // of processed and the total number of known states are also returned. Otherwise 237 // these are zero. 238 func (d *Downloader) Progress() ethereum.SyncProgress { 239 // Lock the current stats and return the progress 240 d.syncStatsLock.RLock() 241 defer d.syncStatsLock.RUnlock() 242 243 current := uint64(0) 244 switch d.mode { 245 case FullSync: 246 current = d.blockchain.CurrentBlock().NumberU64() 247 case FastSync: 248 current = d.blockchain.CurrentFastBlock().NumberU64() 249 case LightSync: 250 current = d.lightchain.CurrentHeader().Number.Uint64() 251 } 252 return ethereum.SyncProgress{ 253 StartingBlock: d.syncStatsChainOrigin, 254 CurrentBlock: current, 255 HighestBlock: d.syncStatsChainHeight, 256 PulledStates: d.syncStatsState.processed, 257 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 258 } 259 } 260 261 // Synchronising returns whether the downloader is currently retrieving blocks. 262 func (d *Downloader) Synchronising() bool { 263 return atomic.LoadInt32(&d.synchronising) > 0 264 } 265 266 // RegisterPeer injects a new download peer into the set of block source to be 267 // used for fetching hashes and blocks from. 268 func (d *Downloader) RegisterPeer(id string, version int, peer Peer) error { 269 logger := log.New("peer", id) 270 logger.Trace("Registering sync peer") 271 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 272 logger.Error("Failed to register sync peer", "err", err) 273 return err 274 } 275 d.qosReduceConfidence() 276 277 return nil 278 } 279 280 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 281 func (d *Downloader) RegisterLightPeer(id string, version int, peer LightPeer) error { 282 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 283 } 284 285 // UnregisterPeer remove a peer from the known list, preventing any action from 286 // the specified peer. An effort is also made to return any pending fetches into 287 // the queue. 288 func (d *Downloader) UnregisterPeer(id string) error { 289 // Unregister the peer from the active peer set and revoke any fetch tasks 290 logger := log.New("peer", id) 291 logger.Trace("Unregistering sync peer") 292 if err := d.peers.Unregister(id); err != nil { 293 logger.Error("Failed to unregister sync peer", "err", err) 294 return err 295 } 296 d.queue.Revoke(id) 297 298 // If this peer was the master peer, abort sync immediately 299 d.cancelLock.RLock() 300 master := id == d.cancelPeer 301 d.cancelLock.RUnlock() 302 303 if master { 304 d.Cancel() 305 } 306 return nil 307 } 308 309 // Synchronise tries to sync up our local block chain with a remote peer, both 310 // adding various sanity checks as well as wrapping it with various log entries. 311 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 312 err := d.synchronise(id, head, td, mode) 313 switch err { 314 case nil: 315 case errBusy: 316 317 case errTimeout, errBadPeer, errStallingPeer, 318 errEmptyHeaderSet, errPeersUnavailable, errTooOld, 319 errInvalidAncestor, errInvalidChain: 320 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 321 if d.dropPeer == nil { 322 // The dropPeer method is nil when `--copydb` is used for a local copy. 323 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 324 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 325 } else { 326 d.dropPeer(id) 327 } 328 default: 329 log.Warn("Synchronisation failed, retrying", "err", err) 330 } 331 return err 332 } 333 334 // synchronise will select the peer and use it for synchronising. If an empty string is given 335 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 336 // checks fail an error will be returned. This method is synchronous 337 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 338 // Mock out the synchronisation if testing 339 if d.synchroniseMock != nil { 340 return d.synchroniseMock(id, hash) 341 } 342 // Make sure only one goroutine is ever allowed past this point at once 343 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 344 return errBusy 345 } 346 defer atomic.StoreInt32(&d.synchronising, 0) 347 348 // Post a user notification of the sync (only once per session) 349 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 350 log.Info("Block synchronisation started") 351 } 352 // Reset the queue, peer set and wake channels to clean any internal leftover state 353 d.queue.Reset() 354 d.peers.Reset() 355 356 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 357 select { 358 case <-ch: 359 default: 360 } 361 } 362 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 363 for empty := false; !empty; { 364 select { 365 case <-ch: 366 default: 367 empty = true 368 } 369 } 370 } 371 for empty := false; !empty; { 372 select { 373 case <-d.headerProcCh: 374 default: 375 empty = true 376 } 377 } 378 // Create cancel channel for aborting mid-flight and mark the master peer 379 d.cancelLock.Lock() 380 d.cancelCh = make(chan struct{}) 381 d.cancelPeer = id 382 d.cancelLock.Unlock() 383 384 defer d.Cancel() // No matter what, we can't leave the cancel channel open 385 386 // Set the requested sync mode, unless it's forbidden 387 d.mode = mode 388 389 // Retrieve the origin peer and initiate the downloading process 390 p := d.peers.Peer(id) 391 if p == nil { 392 return errUnknownPeer 393 } 394 return d.syncWithPeer(p, hash, td) 395 } 396 397 // syncWithPeer starts a block synchronization based on the hash chain from the 398 // specified peer and head hash. 399 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 400 d.mux.Post(StartEvent{}) 401 defer func() { 402 // reset on error 403 if err != nil { 404 d.mux.Post(FailedEvent{err}) 405 } else { 406 d.mux.Post(DoneEvent{}) 407 } 408 }() 409 if p.version < 62 { 410 return errTooOld 411 } 412 413 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", d.mode) 414 defer func(start time.Time) { 415 log.Debug("Synchronisation terminated", "elapsed", time.Since(start)) 416 }(time.Now()) 417 418 // Look up the sync boundaries: the common ancestor and the target block 419 latest, err := d.fetchHeight(p) 420 if err != nil { 421 return err 422 } 423 height := latest.Number.Uint64() 424 425 origin, err := d.findAncestor(p, height) 426 if err != nil { 427 return err 428 } 429 d.syncStatsLock.Lock() 430 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 431 d.syncStatsChainOrigin = origin 432 } 433 d.syncStatsChainHeight = height 434 d.syncStatsLock.Unlock() 435 436 // Ensure our origin point is below any fast sync pivot point 437 pivot := uint64(0) 438 if d.mode == FastSync { 439 if height <= uint64(fsMinFullBlocks) { 440 origin = 0 441 } else { 442 pivot = height - uint64(fsMinFullBlocks) 443 if pivot <= origin { 444 origin = pivot - 1 445 } 446 } 447 } 448 d.committed = 1 449 if d.mode == FastSync && pivot != 0 { 450 d.committed = 0 451 } 452 // Initiate the sync using a concurrent header and content retrieval algorithm 453 d.queue.Prepare(origin+1, d.mode) 454 if d.syncInitHook != nil { 455 d.syncInitHook(origin, height) 456 } 457 458 fetchers := []func() error{ 459 func() error { return d.fetchHeaders(p, origin+1, pivot) }, // Headers are always retrieved 460 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 461 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 462 func() error { return d.processHeaders(origin+1, pivot, td) }, 463 } 464 if d.mode == FastSync { 465 fetchers = append(fetchers, func() error { return d.processFastSyncContent(latest) }) 466 } else if d.mode == FullSync { 467 fetchers = append(fetchers, d.processFullSyncContent) 468 } 469 return d.spawnSync(fetchers) 470 } 471 472 // spawnSync runs d.process and all given fetcher functions to completion in 473 // separate goroutines, returning the first error that appears. 474 func (d *Downloader) spawnSync(fetchers []func() error) error { 475 var wg sync.WaitGroup 476 errc := make(chan error, len(fetchers)) 477 wg.Add(len(fetchers)) 478 for _, fn := range fetchers { 479 fn := fn 480 go func() { defer wg.Done(); errc <- fn() }() 481 } 482 // Wait for the first error, then terminate the others. 483 var err error 484 for i := 0; i < len(fetchers); i++ { 485 if i == len(fetchers)-1 { 486 // Close the queue when all fetchers have exited. 487 // This will cause the block processor to end when 488 // it has processed the queue. 489 d.queue.Close() 490 } 491 if err = <-errc; err != nil { 492 break 493 } 494 } 495 d.queue.Close() 496 d.Cancel() 497 wg.Wait() 498 return err 499 } 500 501 // Cancel cancels all of the operations and resets the queue. It returns true 502 // if the cancel operation was completed. 503 func (d *Downloader) Cancel() { 504 // Close the current cancel channel 505 d.cancelLock.Lock() 506 if d.cancelCh != nil { 507 select { 508 case <-d.cancelCh: 509 // Channel was already closed 510 default: 511 close(d.cancelCh) 512 } 513 } 514 d.cancelLock.Unlock() 515 } 516 517 // Terminate interrupts the downloader, canceling all pending operations. 518 // The downloader cannot be reused after calling Terminate. 519 func (d *Downloader) Terminate() { 520 // Close the termination channel (make sure double close is allowed) 521 d.quitLock.Lock() 522 select { 523 case <-d.quitCh: 524 default: 525 close(d.quitCh) 526 } 527 d.quitLock.Unlock() 528 529 // Cancel any pending download requests 530 d.Cancel() 531 } 532 533 // fetchHeight retrieves the head header of the remote peer to aid in estimating 534 // the total time a pending synchronisation would take. 535 func (d *Downloader) fetchHeight(p *peerConnection) (*types.Header, error) { 536 p.log.Debug("Retrieving remote chain height") 537 538 // Request the advertised remote head block and wait for the response 539 head, _ := p.peer.Head() 540 go p.peer.RequestHeadersByHash(head, 1, 0, false) 541 542 ttl := d.requestTTL() 543 timeout := time.After(ttl) 544 for { 545 select { 546 case <-d.cancelCh: 547 return nil, errCancelBlockFetch 548 549 case packet := <-d.headerCh: 550 // Discard anything not from the origin peer 551 if packet.PeerId() != p.id { 552 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 553 break 554 } 555 // Make sure the peer actually gave something valid 556 headers := packet.(*headerPack).headers 557 if len(headers) != 1 { 558 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 559 return nil, errBadPeer 560 } 561 head := headers[0] 562 p.log.Debug("Remote head header identified", "number", head.Number, "hash", head.Hash()) 563 return head, nil 564 565 case <-timeout: 566 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 567 return nil, errTimeout 568 569 case <-d.bodyCh: 570 case <-d.receiptCh: 571 // Out of bounds delivery, ignore 572 } 573 } 574 } 575 576 // findAncestor tries to locate the common ancestor link of the local chain and 577 // a remote peers blockchain. In the general case when our node was in sync and 578 // on the correct chain, checking the top N links should already get us a match. 579 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 580 // the head links match), we do a binary search to find the common ancestor. 581 func (d *Downloader) findAncestor(p *peerConnection, height uint64) (uint64, error) { 582 // Figure out the valid ancestor range to prevent rewrite attacks 583 floor, ceil := int64(-1), d.lightchain.CurrentHeader().Number.Uint64() 584 585 if d.mode == FullSync { 586 ceil = d.blockchain.CurrentBlock().NumberU64() 587 } else if d.mode == FastSync { 588 ceil = d.blockchain.CurrentFastBlock().NumberU64() 589 } 590 if ceil >= MaxForkAncestry { 591 floor = int64(ceil - MaxForkAncestry) 592 } 593 p.log.Debug("Looking for common ancestor", "local", ceil, "remote", height) 594 595 // Request the topmost blocks to short circuit binary ancestor lookup 596 head := ceil 597 if head > height { 598 head = height 599 } 600 from := int64(head) - int64(MaxHeaderFetch) 601 if from < 0 { 602 from = 0 603 } 604 // Span out with 15 block gaps into the future to catch bad head reports 605 limit := 2 * MaxHeaderFetch / 16 606 count := 1 + int((int64(ceil)-from)/16) 607 if count > limit { 608 count = limit 609 } 610 go p.peer.RequestHeadersByNumber(uint64(from), count, 15, false) 611 612 // Wait for the remote response to the head fetch 613 number, hash := uint64(0), common.Hash{} 614 615 ttl := d.requestTTL() 616 timeout := time.After(ttl) 617 618 for finished := false; !finished; { 619 select { 620 case <-d.cancelCh: 621 return 0, errCancelHeaderFetch 622 623 case packet := <-d.headerCh: 624 // Discard anything not from the origin peer 625 if packet.PeerId() != p.id { 626 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 627 break 628 } 629 // Make sure the peer actually gave something valid 630 headers := packet.(*headerPack).headers 631 if len(headers) == 0 { 632 p.log.Warn("Empty head header set") 633 return 0, errEmptyHeaderSet 634 } 635 // Make sure the peer's reply conforms to the request 636 for i := 0; i < len(headers); i++ { 637 if number := headers[i].Number.Int64(); number != from+int64(i)*16 { 638 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", from+int64(i)*16, "received", number) 639 return 0, errInvalidChain 640 } 641 } 642 // Check if a common ancestor was found 643 finished = true 644 for i := len(headers) - 1; i >= 0; i-- { 645 // Skip any headers that underflow/overflow our requested set 646 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > ceil { 647 continue 648 } 649 // Otherwise check if we already know the header or not 650 if (d.mode == FullSync && d.blockchain.HasBlock(headers[i].Hash(), headers[i].Number.Uint64())) || (d.mode != FullSync && d.lightchain.HasHeader(headers[i].Hash(), headers[i].Number.Uint64())) { 651 number, hash = headers[i].Number.Uint64(), headers[i].Hash() 652 653 // If every header is known, even future ones, the peer straight out lied about its head 654 if number > height && i == limit-1 { 655 p.log.Warn("Lied about chain head", "reported", height, "found", number) 656 return 0, errStallingPeer 657 } 658 break 659 } 660 } 661 662 case <-timeout: 663 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 664 return 0, errTimeout 665 666 case <-d.bodyCh: 667 case <-d.receiptCh: 668 // Out of bounds delivery, ignore 669 } 670 } 671 // If the head fetch already found an ancestor, return 672 if !common.EmptyHash(hash) { 673 if int64(number) <= floor { 674 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 675 return 0, errInvalidAncestor 676 } 677 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 678 return number, nil 679 } 680 // Ancestor not found, we need to binary search over our chain 681 start, end := uint64(0), head 682 if floor > 0 { 683 start = uint64(floor) 684 } 685 for start+1 < end { 686 // Split our chain interval in two, and request the hash to cross check 687 check := (start + end) / 2 688 689 ttl := d.requestTTL() 690 timeout := time.After(ttl) 691 692 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 693 694 // Wait until a reply arrives to this request 695 for arrived := false; !arrived; { 696 select { 697 case <-d.cancelCh: 698 return 0, errCancelHeaderFetch 699 700 case packer := <-d.headerCh: 701 // Discard anything not from the origin peer 702 if packer.PeerId() != p.id { 703 log.Debug("Received headers from incorrect peer", "peer", packer.PeerId()) 704 break 705 } 706 // Make sure the peer actually gave something valid 707 headers := packer.(*headerPack).headers 708 if len(headers) != 1 { 709 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 710 return 0, errBadPeer 711 } 712 arrived = true 713 714 // Modify the search interval based on the response 715 if (d.mode == FullSync && !d.blockchain.HasBlock(headers[0].Hash(), headers[0].Number.Uint64())) || (d.mode != FullSync && !d.lightchain.HasHeader(headers[0].Hash(), headers[0].Number.Uint64())) { 716 end = check 717 break 718 } 719 header := d.lightchain.GetHeaderByHash(headers[0].Hash()) // Independent of sync mode, header surely exists 720 if header.Number.Uint64() != check { 721 p.log.Debug("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 722 return 0, errBadPeer 723 } 724 start = check 725 726 case <-timeout: 727 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 728 return 0, errTimeout 729 730 case <-d.bodyCh: 731 case <-d.receiptCh: 732 // Out of bounds delivery, ignore 733 } 734 } 735 } 736 // Ensure valid ancestry and return 737 if int64(start) <= floor { 738 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 739 return 0, errInvalidAncestor 740 } 741 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 742 return start, nil 743 } 744 745 // fetchHeaders keeps retrieving headers concurrently from the number 746 // requested, until no more are returned, potentially throttling on the way. To 747 // facilitate concurrency but still protect against malicious nodes sending bad 748 // headers, we construct a header chain skeleton using the "origin" peer we are 749 // syncing with, and fill in the missing headers using anyone else. Headers from 750 // other peers are only accepted if they map cleanly to the skeleton. If no one 751 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 752 // the origin is dropped. 753 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64, pivot uint64) error { 754 p.log.Debug("Directing header downloads", "origin", from) 755 defer p.log.Debug("Header download terminated") 756 757 // Create a timeout timer, and the associated header fetcher 758 skeleton := true // Skeleton assembly phase or finishing up 759 request := time.Now() // time of the last skeleton fetch request 760 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 761 <-timeout.C // timeout channel should be initially empty 762 defer timeout.Stop() 763 764 var ttl time.Duration 765 getHeaders := func(from uint64) { 766 request = time.Now() 767 768 ttl = d.requestTTL() 769 timeout.Reset(ttl) 770 771 if skeleton { 772 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 773 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 774 } else { 775 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 776 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 777 } 778 } 779 // Start pulling the header chain skeleton until all is done 780 getHeaders(from) 781 782 for { 783 select { 784 case <-d.cancelCh: 785 return errCancelHeaderFetch 786 787 case packet := <-d.headerCh: 788 // Make sure the active peer is giving us the skeleton headers 789 if packet.PeerId() != p.id { 790 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 791 break 792 } 793 headerReqTimer.UpdateSince(request) 794 timeout.Stop() 795 796 // If the skeleton's finished, pull any remaining head headers directly from the origin 797 if packet.Items() == 0 && skeleton { 798 skeleton = false 799 getHeaders(from) 800 continue 801 } 802 // If no more headers are inbound, notify the content fetchers and return 803 if packet.Items() == 0 { 804 // Don't abort header fetches while the pivot is downloading 805 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 806 p.log.Debug("No headers, waiting for pivot commit") 807 select { 808 case <-time.After(fsHeaderContCheck): 809 getHeaders(from) 810 continue 811 case <-d.cancelCh: 812 return errCancelHeaderFetch 813 } 814 } 815 // Pivot done (or not in fast sync) and no more headers, terminate the process 816 p.log.Debug("No more headers available") 817 select { 818 case d.headerProcCh <- nil: 819 return nil 820 case <-d.cancelCh: 821 return errCancelHeaderFetch 822 } 823 } 824 headers := packet.(*headerPack).headers 825 826 // If we received a skeleton batch, resolve internals concurrently 827 if skeleton { 828 filled, proced, err := d.fillHeaderSkeleton(from, headers) 829 if err != nil { 830 p.log.Debug("Skeleton chain invalid", "err", err) 831 return errInvalidChain 832 } 833 headers = filled[proced:] 834 from += uint64(proced) 835 } 836 // Insert all the new headers and fetch the next batch 837 if len(headers) > 0 { 838 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 839 select { 840 case d.headerProcCh <- headers: 841 case <-d.cancelCh: 842 return errCancelHeaderFetch 843 } 844 from += uint64(len(headers)) 845 } 846 getHeaders(from) 847 848 case <-timeout.C: 849 if d.dropPeer == nil { 850 // The dropPeer method is nil when `--copydb` is used for a local copy. 851 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 852 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 853 break 854 } 855 // Header retrieval timed out, consider the peer bad and drop 856 p.log.Debug("Header request timed out", "elapsed", ttl) 857 headerTimeoutMeter.Mark(1) 858 d.dropPeer(p.id) 859 860 // Finish the sync gracefully instead of dumping the gathered data though 861 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 862 select { 863 case ch <- false: 864 case <-d.cancelCh: 865 } 866 } 867 select { 868 case d.headerProcCh <- nil: 869 case <-d.cancelCh: 870 } 871 return errBadPeer 872 } 873 } 874 } 875 876 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 877 // and maps them to the provided skeleton header chain. 878 // 879 // Any partial results from the beginning of the skeleton is (if possible) forwarded 880 // immediately to the header processor to keep the rest of the pipeline full even 881 // in the case of header stalls. 882 // 883 // The method returs the entire filled skeleton and also the number of headers 884 // already forwarded for processing. 885 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 886 log.Debug("Filling up skeleton", "from", from) 887 d.queue.ScheduleSkeleton(from, skeleton) 888 889 var ( 890 deliver = func(packet dataPack) (int, error) { 891 pack := packet.(*headerPack) 892 return d.queue.DeliverHeaders(pack.peerId, pack.headers, d.headerProcCh) 893 } 894 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 895 throttle = func() bool { return false } 896 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, error) { 897 return d.queue.ReserveHeaders(p, count), false, nil 898 } 899 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 900 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 901 setIdle = func(p *peerConnection, accepted int) { p.SetHeadersIdle(accepted) } 902 ) 903 err := d.fetchParts(errCancelHeaderFetch, d.headerCh, deliver, d.queue.headerContCh, expire, 904 d.queue.PendingHeaders, d.queue.InFlightHeaders, throttle, reserve, 905 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 906 907 log.Debug("Skeleton fill terminated", "err", err) 908 909 filled, proced := d.queue.RetrieveHeaders() 910 return filled, proced, err 911 } 912 913 // fetchBodies iteratively downloads the scheduled block bodies, taking any 914 // available peers, reserving a chunk of blocks for each, waiting for delivery 915 // and also periodically checking for timeouts. 916 func (d *Downloader) fetchBodies(from uint64) error { 917 log.Debug("Downloading block bodies", "origin", from) 918 919 var ( 920 deliver = func(packet dataPack) (int, error) { 921 pack := packet.(*bodyPack) 922 return d.queue.DeliverBodies(pack.peerId, pack.transactions, pack.uncles) 923 } 924 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 925 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 926 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 927 setIdle = func(p *peerConnection, accepted int) { p.SetBodiesIdle(accepted) } 928 ) 929 err := d.fetchParts(errCancelBodyFetch, d.bodyCh, deliver, d.bodyWakeCh, expire, 930 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ShouldThrottleBlocks, d.queue.ReserveBodies, 931 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 932 933 log.Debug("Block body download terminated", "err", err) 934 return err 935 } 936 937 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 938 // available peers, reserving a chunk of receipts for each, waiting for delivery 939 // and also periodically checking for timeouts. 940 func (d *Downloader) fetchReceipts(from uint64) error { 941 log.Debug("Downloading transaction receipts", "origin", from) 942 943 var ( 944 deliver = func(packet dataPack) (int, error) { 945 pack := packet.(*receiptPack) 946 return d.queue.DeliverReceipts(pack.peerId, pack.receipts) 947 } 948 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 949 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 950 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 951 setIdle = func(p *peerConnection, accepted int) { p.SetReceiptsIdle(accepted) } 952 ) 953 err := d.fetchParts(errCancelReceiptFetch, d.receiptCh, deliver, d.receiptWakeCh, expire, 954 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ShouldThrottleReceipts, d.queue.ReserveReceipts, 955 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 956 957 log.Debug("Transaction receipt download terminated", "err", err) 958 return err 959 } 960 961 // fetchParts iteratively downloads scheduled block parts, taking any available 962 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 963 // also periodically checking for timeouts. 964 // 965 // As the scheduling/timeout logic mostly is the same for all downloaded data 966 // types, this method is used by each for data gathering and is instrumented with 967 // various callbacks to handle the slight differences between processing them. 968 // 969 // The instrumentation parameters: 970 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 971 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 972 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 973 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 974 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 975 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 976 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 977 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 978 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 979 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 980 // - fetch: network callback to actually send a particular download request to a physical remote peer 981 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 982 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 983 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 984 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 985 // - kind: textual label of the type being downloaded to display in log mesages 986 func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 987 expire func() map[string]int, pending func() int, inFlight func() bool, throttle func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, error), 988 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 989 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int), kind string) error { 990 991 // Create a ticker to detect expired retrieval tasks 992 ticker := time.NewTicker(100 * time.Millisecond) 993 defer ticker.Stop() 994 995 update := make(chan struct{}, 1) 996 997 // Prepare the queue and fetch block parts until the block header fetcher's done 998 finished := false 999 for { 1000 select { 1001 case <-d.cancelCh: 1002 return errCancel 1003 1004 case packet := <-deliveryCh: 1005 // If the peer was previously banned and failed to deliver its pack 1006 // in a reasonable time frame, ignore its message. 1007 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1008 // Deliver the received chunk of data and check chain validity 1009 accepted, err := deliver(packet) 1010 if err == errInvalidChain { 1011 return err 1012 } 1013 // Unless a peer delivered something completely else than requested (usually 1014 // caused by a timed out request which came through in the end), set it to 1015 // idle. If the delivery's stale, the peer should have already been idled. 1016 if err != errStaleDelivery { 1017 setIdle(peer, accepted) 1018 } 1019 // Issue a log to the user to see what's going on 1020 switch { 1021 case err == nil && packet.Items() == 0: 1022 peer.log.Trace("Requested data not delivered", "type", kind) 1023 case err == nil: 1024 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1025 default: 1026 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1027 } 1028 } 1029 // Blocks assembled, try to update the progress 1030 select { 1031 case update <- struct{}{}: 1032 default: 1033 } 1034 1035 case cont := <-wakeCh: 1036 // The header fetcher sent a continuation flag, check if it's done 1037 if !cont { 1038 finished = true 1039 } 1040 // Headers arrive, try to update the progress 1041 select { 1042 case update <- struct{}{}: 1043 default: 1044 } 1045 1046 case <-ticker.C: 1047 // Sanity check update the progress 1048 select { 1049 case update <- struct{}{}: 1050 default: 1051 } 1052 1053 case <-update: 1054 // Short circuit if we lost all our peers 1055 if d.peers.Len() == 0 { 1056 return errNoPeers 1057 } 1058 // Check for fetch request timeouts and demote the responsible peers 1059 for pid, fails := range expire() { 1060 if peer := d.peers.Peer(pid); peer != nil { 1061 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1062 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1063 // out that sync wise we need to get rid of the peer. 1064 // 1065 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1066 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1067 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1068 if fails > 2 { 1069 peer.log.Trace("Data delivery timed out", "type", kind) 1070 setIdle(peer, 0) 1071 } else { 1072 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1073 if d.dropPeer == nil { 1074 // The dropPeer method is nil when `--copydb` is used for a local copy. 1075 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1076 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1077 } else { 1078 d.dropPeer(pid) 1079 } 1080 } 1081 } 1082 } 1083 // If there's nothing more to fetch, wait or terminate 1084 if pending() == 0 { 1085 if !inFlight() && finished { 1086 log.Debug("Data fetching completed", "type", kind) 1087 return nil 1088 } 1089 break 1090 } 1091 // Send a download request to all idle peers, until throttled 1092 progressed, throttled, running := false, false, inFlight() 1093 idles, total := idle() 1094 1095 for _, peer := range idles { 1096 // Short circuit if throttling activated 1097 if throttle() { 1098 throttled = true 1099 break 1100 } 1101 // Short circuit if there is no more available task. 1102 if pending() == 0 { 1103 break 1104 } 1105 // Reserve a chunk of fetches for a peer. A nil can mean either that 1106 // no more headers are available, or that the peer is known not to 1107 // have them. 1108 request, progress, err := reserve(peer, capacity(peer)) 1109 if err != nil { 1110 return err 1111 } 1112 if progress { 1113 progressed = true 1114 } 1115 if request == nil { 1116 continue 1117 } 1118 if request.From > 0 { 1119 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1120 } else { 1121 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1122 } 1123 // Fetch the chunk and make sure any errors return the hashes to the queue 1124 if fetchHook != nil { 1125 fetchHook(request.Headers) 1126 } 1127 if err := fetch(peer, request); err != nil { 1128 // Although we could try and make an attempt to fix this, this error really 1129 // means that we've double allocated a fetch task to a peer. If that is the 1130 // case, the internal state of the downloader and the queue is very wrong so 1131 // better hard crash and note the error instead of silently accumulating into 1132 // a much bigger issue. 1133 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1134 } 1135 running = true 1136 } 1137 // Make sure that we have peers available for fetching. If all peers have been tried 1138 // and all failed throw an error 1139 if !progressed && !throttled && !running && len(idles) == total && pending() > 0 { 1140 return errPeersUnavailable 1141 } 1142 } 1143 } 1144 } 1145 1146 // processHeaders takes batches of retrieved headers from an input channel and 1147 // keeps processing and scheduling them into the header chain and downloader's 1148 // queue until the stream ends or a failure occurs. 1149 func (d *Downloader) processHeaders(origin uint64, pivot uint64, td *big.Int) error { 1150 // Keep a count of uncertain headers to roll back 1151 rollback := []*types.Header{} 1152 defer func() { 1153 if len(rollback) > 0 { 1154 // Flatten the headers and roll them back 1155 hashes := make([]common.Hash, len(rollback)) 1156 for i, header := range rollback { 1157 hashes[i] = header.Hash() 1158 } 1159 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1160 if d.mode != LightSync { 1161 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1162 lastBlock = d.blockchain.CurrentBlock().Number() 1163 } 1164 d.lightchain.Rollback(hashes) 1165 curFastBlock, curBlock := common.Big0, common.Big0 1166 if d.mode != LightSync { 1167 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1168 curBlock = d.blockchain.CurrentBlock().Number() 1169 } 1170 log.Warn("Rolled back headers", "count", len(hashes), 1171 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1172 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1173 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock)) 1174 } 1175 }() 1176 1177 // Wait for batches of headers to process 1178 gotHeaders := false 1179 1180 for { 1181 select { 1182 case <-d.cancelCh: 1183 return errCancelHeaderProcessing 1184 1185 case headers := <-d.headerProcCh: 1186 // Terminate header processing if we synced up 1187 if len(headers) == 0 { 1188 // Notify everyone that headers are fully processed 1189 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1190 select { 1191 case ch <- false: 1192 case <-d.cancelCh: 1193 } 1194 } 1195 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1196 // better chain compared to ours. The only exception is if its promised blocks were 1197 // already imported by other means (e.g. fecher): 1198 // 1199 // R <remote peer>, L <local node>: Both at block 10 1200 // R: Mine block 11, and propagate it to L 1201 // L: Queue block 11 for import 1202 // L: Notice that R's head and TD increased compared to ours, start sync 1203 // L: Import of block 11 finishes 1204 // L: Sync begins, and finds common ancestor at 11 1205 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1206 // R: Nothing to give 1207 if d.mode != LightSync { 1208 head := d.blockchain.CurrentBlock() 1209 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1210 return errStallingPeer 1211 } 1212 } 1213 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1214 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1215 // of delivering the post-pivot blocks that would flag the invalid content. 1216 // 1217 // This check cannot be executed "as is" for full imports, since blocks may still be 1218 // queued for processing when the header download completes. However, as long as the 1219 // peer gave us something useful, we're already happy/progressed (above check). 1220 if d.mode == FastSync || d.mode == LightSync { 1221 head := d.lightchain.CurrentHeader() 1222 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1223 return errStallingPeer 1224 } 1225 } 1226 // Disable any rollback and return 1227 rollback = nil 1228 return nil 1229 } 1230 // Otherwise split the chunk of headers into batches and process them 1231 gotHeaders = true 1232 1233 for len(headers) > 0 { 1234 // Terminate if something failed in between processing chunks 1235 select { 1236 case <-d.cancelCh: 1237 return errCancelHeaderProcessing 1238 default: 1239 } 1240 // Select the next chunk of headers to import 1241 limit := maxHeadersProcess 1242 if limit > len(headers) { 1243 limit = len(headers) 1244 } 1245 chunk := headers[:limit] 1246 1247 // In case of header only syncing, validate the chunk immediately 1248 if d.mode == FastSync || d.mode == LightSync { 1249 // Collect the yet unknown headers to mark them as uncertain 1250 unknown := make([]*types.Header, 0, len(headers)) 1251 for _, header := range chunk { 1252 if !d.lightchain.HasHeader(header.Hash(), header.Number.Uint64()) { 1253 unknown = append(unknown, header) 1254 } 1255 } 1256 // If we're importing pure headers, verify based on their recentness 1257 frequency := fsHeaderCheckFrequency 1258 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1259 frequency = 1 1260 } 1261 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1262 // If some headers were inserted, add them too to the rollback list 1263 if n > 0 { 1264 rollback = append(rollback, chunk[:n]...) 1265 } 1266 log.Debug("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "err", err) 1267 return errInvalidChain 1268 } 1269 // All verifications passed, store newly found uncertain headers 1270 rollback = append(rollback, unknown...) 1271 if len(rollback) > fsHeaderSafetyNet { 1272 rollback = append(rollback[:0], rollback[len(rollback)-fsHeaderSafetyNet:]...) 1273 } 1274 } 1275 // Unless we're doing light chains, schedule the headers for associated content retrieval 1276 if d.mode == FullSync || d.mode == FastSync { 1277 // If we've reached the allowed number of pending headers, stall a bit 1278 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1279 select { 1280 case <-d.cancelCh: 1281 return errCancelHeaderProcessing 1282 case <-time.After(time.Second): 1283 } 1284 } 1285 // Otherwise insert the headers for content retrieval 1286 inserts := d.queue.Schedule(chunk, origin) 1287 if len(inserts) != len(chunk) { 1288 log.Debug("Stale headers") 1289 return errBadPeer 1290 } 1291 } 1292 headers = headers[limit:] 1293 origin += uint64(limit) 1294 } 1295 // Signal the content downloaders of the availablility of new tasks 1296 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1297 select { 1298 case ch <- true: 1299 default: 1300 } 1301 } 1302 } 1303 } 1304 } 1305 1306 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1307 func (d *Downloader) processFullSyncContent() error { 1308 for { 1309 results := d.queue.Results(true) 1310 if len(results) == 0 { 1311 return nil 1312 } 1313 if d.chainInsertHook != nil { 1314 d.chainInsertHook(results) 1315 } 1316 if err := d.importBlockResults(results); err != nil { 1317 return err 1318 } 1319 } 1320 } 1321 1322 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1323 // Check for any early termination requests 1324 if len(results) == 0 { 1325 return nil 1326 } 1327 select { 1328 case <-d.quitCh: 1329 return errCancelContentProcessing 1330 default: 1331 } 1332 // Retrieve the a batch of results to import 1333 first, last := results[0].Header, results[len(results)-1].Header 1334 log.Debug("Inserting downloaded chain", "items", len(results), 1335 "firstnum", first.Number, "firsthash", first.Hash(), 1336 "lastnum", last.Number, "lasthash", last.Hash(), 1337 ) 1338 blocks := make([]*types.Block, len(results)) 1339 for i, result := range results { 1340 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1341 } 1342 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1343 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1344 return errInvalidChain 1345 } 1346 return nil 1347 } 1348 1349 // processFastSyncContent takes fetch results from the queue and writes them to the 1350 // database. It also controls the synchronisation of state nodes of the pivot block. 1351 func (d *Downloader) processFastSyncContent(latest *types.Header) error { 1352 // Start syncing state of the reported head block. This should get us most of 1353 // the state of the pivot block. 1354 stateSync := d.syncState(latest.Root) 1355 defer stateSync.Cancel() 1356 go func() { 1357 if err := stateSync.Wait(); err != nil && err != errCancelStateFetch { 1358 d.queue.Close() // wake up WaitResults 1359 } 1360 }() 1361 // Figure out the ideal pivot block. Note, that this goalpost may move if the 1362 // sync takes long enough for the chain head to move significantly. 1363 pivot := uint64(0) 1364 if height := latest.Number.Uint64(); height > uint64(fsMinFullBlocks) { 1365 pivot = height - uint64(fsMinFullBlocks) 1366 } 1367 // To cater for moving pivot points, track the pivot block and subsequently 1368 // accumulated download results separatey. 1369 var ( 1370 oldPivot *fetchResult // Locked in pivot block, might change eventually 1371 oldTail []*fetchResult // Downloaded content after the pivot 1372 ) 1373 for { 1374 // Wait for the next batch of downloaded data to be available, and if the pivot 1375 // block became stale, move the goalpost 1376 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1377 if len(results) == 0 { 1378 // If pivot sync is done, stop 1379 if oldPivot == nil { 1380 return stateSync.Cancel() 1381 } 1382 // If sync failed, stop 1383 select { 1384 case <-d.cancelCh: 1385 return stateSync.Cancel() 1386 default: 1387 } 1388 } 1389 if d.chainInsertHook != nil { 1390 d.chainInsertHook(results) 1391 } 1392 if oldPivot != nil { 1393 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1394 } 1395 // Split around the pivot block and process the two sides via fast/full sync 1396 if atomic.LoadInt32(&d.committed) == 0 { 1397 latest = results[len(results)-1].Header 1398 if height := latest.Number.Uint64(); height > pivot+2*uint64(fsMinFullBlocks) { 1399 log.Warn("Pivot became stale, moving", "old", pivot, "new", height-uint64(fsMinFullBlocks)) 1400 pivot = height - uint64(fsMinFullBlocks) 1401 } 1402 } 1403 P, beforeP, afterP := splitAroundPivot(pivot, results) 1404 if err := d.commitFastSyncData(beforeP, stateSync); err != nil { 1405 return err 1406 } 1407 if P != nil { 1408 // If new pivot block found, cancel old state retrieval and restart 1409 if oldPivot != P { 1410 stateSync.Cancel() 1411 1412 stateSync = d.syncState(P.Header.Root) 1413 defer stateSync.Cancel() 1414 go func() { 1415 if err := stateSync.Wait(); err != nil && err != errCancelStateFetch { 1416 d.queue.Close() // wake up WaitResults 1417 } 1418 }() 1419 oldPivot = P 1420 } 1421 // Wait for completion, occasionally checking for pivot staleness 1422 select { 1423 case <-stateSync.done: 1424 if stateSync.err != nil { 1425 return stateSync.err 1426 } 1427 if err := d.commitPivotBlock(P); err != nil { 1428 return err 1429 } 1430 oldPivot = nil 1431 1432 case <-time.After(time.Second): 1433 oldTail = afterP 1434 continue 1435 } 1436 } 1437 // Fast sync done, pivot commit done, full import 1438 if err := d.importBlockResults(afterP); err != nil { 1439 return err 1440 } 1441 } 1442 } 1443 1444 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1445 for _, result := range results { 1446 num := result.Header.Number.Uint64() 1447 switch { 1448 case num < pivot: 1449 before = append(before, result) 1450 case num == pivot: 1451 p = result 1452 default: 1453 after = append(after, result) 1454 } 1455 } 1456 return p, before, after 1457 } 1458 1459 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1460 // Check for any early termination requests 1461 if len(results) == 0 { 1462 return nil 1463 } 1464 select { 1465 case <-d.quitCh: 1466 return errCancelContentProcessing 1467 case <-stateSync.done: 1468 if err := stateSync.Wait(); err != nil { 1469 return err 1470 } 1471 default: 1472 } 1473 // Retrieve the a batch of results to import 1474 first, last := results[0].Header, results[len(results)-1].Header 1475 log.Debug("Inserting fast-sync blocks", "items", len(results), 1476 "firstnum", first.Number, "firsthash", first.Hash(), 1477 "lastnumn", last.Number, "lasthash", last.Hash(), 1478 ) 1479 blocks := make([]*types.Block, len(results)) 1480 receipts := make([]types.Receipts, len(results)) 1481 for i, result := range results { 1482 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1483 receipts[i] = result.Receipts 1484 } 1485 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts); err != nil { 1486 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1487 return errInvalidChain 1488 } 1489 return nil 1490 } 1491 1492 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1493 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1494 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1495 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}); err != nil { 1496 return err 1497 } 1498 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1499 return err 1500 } 1501 atomic.StoreInt32(&d.committed, 1) 1502 return nil 1503 } 1504 1505 // DeliverHeaders injects a new batch of block headers received from a remote 1506 // node into the download schedule. 1507 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) { 1508 return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1509 } 1510 1511 // DeliverBodies injects a new batch of block bodies received from a remote node. 1512 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) { 1513 return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1514 } 1515 1516 // DeliverReceipts injects a new batch of receipts received from a remote node. 1517 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) { 1518 return d.deliver(id, d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1519 } 1520 1521 // DeliverNodeData injects a new batch of node state data received from a remote node. 1522 func (d *Downloader) DeliverNodeData(id string, data [][]byte) (err error) { 1523 return d.deliver(id, d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1524 } 1525 1526 // deliver injects a new batch of data received from a remote node. 1527 func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1528 // Update the delivery metrics for both good and failed deliveries 1529 inMeter.Mark(int64(packet.Items())) 1530 defer func() { 1531 if err != nil { 1532 dropMeter.Mark(int64(packet.Items())) 1533 } 1534 }() 1535 // Deliver or abort if the sync is canceled while queuing 1536 d.cancelLock.RLock() 1537 cancel := d.cancelCh 1538 d.cancelLock.RUnlock() 1539 if cancel == nil { 1540 return errNoSyncActive 1541 } 1542 select { 1543 case destCh <- packet: 1544 return nil 1545 case <-cancel: 1546 return errNoSyncActive 1547 } 1548 } 1549 1550 // qosTuner is the quality of service tuning loop that occasionally gathers the 1551 // peer latency statistics and updates the estimated request round trip time. 1552 func (d *Downloader) qosTuner() { 1553 for { 1554 // Retrieve the current median RTT and integrate into the previoust target RTT 1555 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 1556 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 1557 1558 // A new RTT cycle passed, increase our confidence in the estimated RTT 1559 conf := atomic.LoadUint64(&d.rttConfidence) 1560 conf = conf + (1000000-conf)/2 1561 atomic.StoreUint64(&d.rttConfidence, conf) 1562 1563 // Log the new QoS values and sleep until the next RTT 1564 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1565 select { 1566 case <-d.quitCh: 1567 return 1568 case <-time.After(rtt): 1569 } 1570 } 1571 } 1572 1573 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 1574 // peer set, needing to reduce the confidence we have in out QoS estimates. 1575 func (d *Downloader) qosReduceConfidence() { 1576 // If we have a single peer, confidence is always 1 1577 peers := uint64(d.peers.Len()) 1578 if peers == 0 { 1579 // Ensure peer connectivity races don't catch us off guard 1580 return 1581 } 1582 if peers == 1 { 1583 atomic.StoreUint64(&d.rttConfidence, 1000000) 1584 return 1585 } 1586 // If we have a ton of peers, don't drop confidence) 1587 if peers >= uint64(qosConfidenceCap) { 1588 return 1589 } 1590 // Otherwise drop the confidence factor 1591 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 1592 if float64(conf)/1000000 < rttMinConfidence { 1593 conf = uint64(rttMinConfidence * 1000000) 1594 } 1595 atomic.StoreUint64(&d.rttConfidence, conf) 1596 1597 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1598 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1599 } 1600 1601 // requestRTT returns the current target round trip time for a download request 1602 // to complete in. 1603 // 1604 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 1605 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 1606 // be adapted to, but smaller ones are preffered (stabler download stream). 1607 func (d *Downloader) requestRTT() time.Duration { 1608 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 1609 } 1610 1611 // requestTTL returns the current timeout allowance for a single download request 1612 // to finish under. 1613 func (d *Downloader) requestTTL() time.Duration { 1614 var ( 1615 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1616 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 1617 ) 1618 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 1619 if ttl > ttlLimit { 1620 ttl = ttlLimit 1621 } 1622 return ttl 1623 }