github.com/daeglee/go-ethereum@v0.0.0-20190504220456-cad3e8d18e9b/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 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 103 queue *queue // Scheduler for selecting the hashes to download 104 peers *peerSet // Set of active peers from which download can proceed 105 stateDB ethdb.Database 106 107 rttEstimate uint64 // Round trip time to target for download requests 108 rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops) 109 110 // Statistics 111 syncStatsChainOrigin uint64 // Origin block number where syncing started at 112 syncStatsChainHeight uint64 // Highest block number known when syncing started 113 syncStatsState stateSyncStats 114 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 115 116 lightchain LightChain 117 blockchain BlockChain 118 119 // Callbacks 120 dropPeer peerDropFn // Drops a peer for misbehaving 121 122 // Status 123 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 124 synchronising int32 125 notified int32 126 committed int32 127 128 // Channels 129 headerCh chan dataPack // [eth/62] Channel receiving inbound block headers 130 bodyCh chan dataPack // [eth/62] Channel receiving inbound block bodies 131 receiptCh chan dataPack // [eth/63] Channel receiving inbound receipts 132 bodyWakeCh chan bool // [eth/62] Channel to signal the block body fetcher of new tasks 133 receiptWakeCh chan bool // [eth/63] Channel to signal the receipt fetcher of new tasks 134 headerProcCh chan []*types.Header // [eth/62] Channel to feed the header processor new tasks 135 136 // for stateFetcher 137 stateSyncStart chan *stateSync 138 trackStateReq chan *stateReq 139 stateCh chan dataPack // [eth/63] Channel receiving inbound node state data 140 141 // Cancellation and termination 142 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 143 cancelCh chan struct{} // Channel to cancel mid-flight syncs 144 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 145 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 146 147 quitCh chan struct{} // Quit channel to signal termination 148 quitLock sync.RWMutex // Lock to prevent double closes 149 150 // Testing hooks 151 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 152 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 153 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 154 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 155 } 156 157 // LightChain encapsulates functions required to synchronise a light chain. 158 type LightChain interface { 159 // HasHeader verifies a header's presence in the local chain. 160 HasHeader(common.Hash, uint64) bool 161 162 // GetHeaderByHash retrieves a header from the local chain. 163 GetHeaderByHash(common.Hash) *types.Header 164 165 // CurrentHeader retrieves the head header from the local chain. 166 CurrentHeader() *types.Header 167 168 // GetTd returns the total difficulty of a local block. 169 GetTd(common.Hash, uint64) *big.Int 170 171 // InsertHeaderChain inserts a batch of headers into the local chain. 172 InsertHeaderChain([]*types.Header, int) (int, error) 173 174 // Rollback removes a few recently added elements from the local chain. 175 Rollback([]common.Hash) 176 } 177 178 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 179 type BlockChain interface { 180 LightChain 181 182 // HasBlock verifies a block's presence in the local chain. 183 HasBlock(common.Hash, uint64) bool 184 185 // HasFastBlock verifies a fast block's presence in the local chain. 186 HasFastBlock(common.Hash, uint64) bool 187 188 // GetBlockByHash retrieves a block from the local chain. 189 GetBlockByHash(common.Hash) *types.Block 190 191 // CurrentBlock retrieves the head block from the local chain. 192 CurrentBlock() *types.Block 193 194 // CurrentFastBlock retrieves the head fast block from the local chain. 195 CurrentFastBlock() *types.Block 196 197 // FastSyncCommitHead directly commits the head block to a certain entity. 198 FastSyncCommitHead(common.Hash) error 199 200 // InsertChain inserts a batch of blocks into the local chain. 201 InsertChain(types.Blocks) (int, error) 202 203 // InsertReceiptChain inserts a batch of receipts into the local chain. 204 InsertReceiptChain(types.Blocks, []types.Receipts) (int, error) 205 } 206 207 // New creates a new downloader to fetch hashes and blocks from remote peers. 208 func New(mode SyncMode, stateDb ethdb.Database, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 209 if lightchain == nil { 210 lightchain = chain 211 } 212 213 dl := &Downloader{ 214 mode: mode, 215 stateDB: stateDb, 216 mux: mux, 217 queue: newQueue(), 218 peers: newPeerSet(), 219 rttEstimate: uint64(rttMaxEstimate), 220 rttConfidence: uint64(1000000), 221 blockchain: chain, 222 lightchain: lightchain, 223 dropPeer: dropPeer, 224 headerCh: make(chan dataPack, 1), 225 bodyCh: make(chan dataPack, 1), 226 receiptCh: make(chan dataPack, 1), 227 bodyWakeCh: make(chan bool, 1), 228 receiptWakeCh: make(chan bool, 1), 229 headerProcCh: make(chan []*types.Header, 1), 230 quitCh: make(chan struct{}), 231 stateCh: make(chan dataPack), 232 stateSyncStart: make(chan *stateSync), 233 syncStatsState: stateSyncStats{ 234 processed: rawdb.ReadFastTrieProgress(stateDb), 235 }, 236 trackStateReq: make(chan *stateReq), 237 } 238 go dl.qosTuner() 239 go dl.stateFetcher() 240 return dl 241 } 242 243 // Progress retrieves the synchronisation boundaries, specifically the origin 244 // block where synchronisation started at (may have failed/suspended); the block 245 // or header sync is currently at; and the latest known block which the sync targets. 246 // 247 // In addition, during the state download phase of fast synchronisation the number 248 // of processed and the total number of known states are also returned. Otherwise 249 // these are zero. 250 func (d *Downloader) Progress() ethereum.SyncProgress { 251 // Lock the current stats and return the progress 252 d.syncStatsLock.RLock() 253 defer d.syncStatsLock.RUnlock() 254 255 current := uint64(0) 256 switch d.mode { 257 case FullSync: 258 current = d.blockchain.CurrentBlock().NumberU64() 259 case FastSync: 260 current = d.blockchain.CurrentFastBlock().NumberU64() 261 case LightSync: 262 current = d.lightchain.CurrentHeader().Number.Uint64() 263 } 264 return ethereum.SyncProgress{ 265 StartingBlock: d.syncStatsChainOrigin, 266 CurrentBlock: current, 267 HighestBlock: d.syncStatsChainHeight, 268 PulledStates: d.syncStatsState.processed, 269 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 270 } 271 } 272 273 // Synchronising returns whether the downloader is currently retrieving blocks. 274 func (d *Downloader) Synchronising() bool { 275 return atomic.LoadInt32(&d.synchronising) > 0 276 } 277 278 // RegisterPeer injects a new download peer into the set of block source to be 279 // used for fetching hashes and blocks from. 280 func (d *Downloader) RegisterPeer(id string, version int, peer Peer) error { 281 logger := log.New("peer", id) 282 logger.Trace("Registering sync peer") 283 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 284 logger.Error("Failed to register sync peer", "err", err) 285 return err 286 } 287 d.qosReduceConfidence() 288 289 return nil 290 } 291 292 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 293 func (d *Downloader) RegisterLightPeer(id string, version int, peer LightPeer) error { 294 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 295 } 296 297 // UnregisterPeer remove a peer from the known list, preventing any action from 298 // the specified peer. An effort is also made to return any pending fetches into 299 // the queue. 300 func (d *Downloader) UnregisterPeer(id string) error { 301 // Unregister the peer from the active peer set and revoke any fetch tasks 302 logger := log.New("peer", id) 303 logger.Trace("Unregistering sync peer") 304 if err := d.peers.Unregister(id); err != nil { 305 logger.Error("Failed to unregister sync peer", "err", err) 306 return err 307 } 308 d.queue.Revoke(id) 309 310 // If this peer was the master peer, abort sync immediately 311 d.cancelLock.RLock() 312 master := id == d.cancelPeer 313 d.cancelLock.RUnlock() 314 315 if master { 316 d.cancel() 317 } 318 return nil 319 } 320 321 // Synchronise tries to sync up our local block chain with a remote peer, both 322 // adding various sanity checks as well as wrapping it with various log entries. 323 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 324 err := d.synchronise(id, head, td, mode) 325 switch err { 326 case nil: 327 case errBusy: 328 329 case errTimeout, errBadPeer, errStallingPeer, 330 errEmptyHeaderSet, errPeersUnavailable, errTooOld, 331 errInvalidAncestor, errInvalidChain: 332 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 333 if d.dropPeer == nil { 334 // The dropPeer method is nil when `--copydb` is used for a local copy. 335 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 336 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 337 } else { 338 d.dropPeer(id) 339 } 340 default: 341 log.Warn("Synchronisation failed, retrying", "err", err) 342 } 343 return err 344 } 345 346 // synchronise will select the peer and use it for synchronising. If an empty string is given 347 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 348 // checks fail an error will be returned. This method is synchronous 349 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 350 // Mock out the synchronisation if testing 351 if d.synchroniseMock != nil { 352 return d.synchroniseMock(id, hash) 353 } 354 // Make sure only one goroutine is ever allowed past this point at once 355 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 356 return errBusy 357 } 358 defer atomic.StoreInt32(&d.synchronising, 0) 359 360 // Post a user notification of the sync (only once per session) 361 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 362 log.Info("Block synchronisation started") 363 } 364 // Reset the queue, peer set and wake channels to clean any internal leftover state 365 d.queue.Reset() 366 d.peers.Reset() 367 368 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 369 select { 370 case <-ch: 371 default: 372 } 373 } 374 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 375 for empty := false; !empty; { 376 select { 377 case <-ch: 378 default: 379 empty = true 380 } 381 } 382 } 383 for empty := false; !empty; { 384 select { 385 case <-d.headerProcCh: 386 default: 387 empty = true 388 } 389 } 390 // Create cancel channel for aborting mid-flight and mark the master peer 391 d.cancelLock.Lock() 392 d.cancelCh = make(chan struct{}) 393 d.cancelPeer = id 394 d.cancelLock.Unlock() 395 396 defer d.Cancel() // No matter what, we can't leave the cancel channel open 397 398 // Set the requested sync mode, unless it's forbidden 399 d.mode = mode 400 401 // Retrieve the origin peer and initiate the downloading process 402 p := d.peers.Peer(id) 403 if p == nil { 404 return errUnknownPeer 405 } 406 return d.syncWithPeer(p, hash, td) 407 } 408 409 // syncWithPeer starts a block synchronization based on the hash chain from the 410 // specified peer and head hash. 411 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 412 d.mux.Post(StartEvent{}) 413 defer func() { 414 // reset on error 415 if err != nil { 416 d.mux.Post(FailedEvent{err}) 417 } else { 418 latest := d.lightchain.CurrentHeader() 419 d.mux.Post(DoneEvent{latest}) 420 } 421 }() 422 if p.version < 62 { 423 return errTooOld 424 } 425 426 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", d.mode) 427 defer func(start time.Time) { 428 log.Debug("Synchronisation terminated", "elapsed", time.Since(start)) 429 }(time.Now()) 430 431 // Look up the sync boundaries: the common ancestor and the target block 432 latest, err := d.fetchHeight(p) 433 if err != nil { 434 return err 435 } 436 height := latest.Number.Uint64() 437 438 origin, err := d.findAncestor(p, latest) 439 if err != nil { 440 return err 441 } 442 d.syncStatsLock.Lock() 443 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 444 d.syncStatsChainOrigin = origin 445 } 446 d.syncStatsChainHeight = height 447 d.syncStatsLock.Unlock() 448 449 // Ensure our origin point is below any fast sync pivot point 450 pivot := uint64(0) 451 if d.mode == FastSync { 452 if height <= uint64(fsMinFullBlocks) { 453 origin = 0 454 } else { 455 pivot = height - uint64(fsMinFullBlocks) 456 if pivot <= origin { 457 origin = pivot - 1 458 } 459 } 460 } 461 d.committed = 1 462 if d.mode == FastSync && pivot != 0 { 463 d.committed = 0 464 } 465 // Initiate the sync using a concurrent header and content retrieval algorithm 466 d.queue.Prepare(origin+1, d.mode) 467 if d.syncInitHook != nil { 468 d.syncInitHook(origin, height) 469 } 470 471 fetchers := []func() error{ 472 func() error { return d.fetchHeaders(p, origin+1, pivot) }, // Headers are always retrieved 473 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 474 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 475 func() error { return d.processHeaders(origin+1, pivot, td) }, 476 } 477 if d.mode == FastSync { 478 fetchers = append(fetchers, func() error { return d.processFastSyncContent(latest) }) 479 } else if d.mode == FullSync { 480 fetchers = append(fetchers, d.processFullSyncContent) 481 } 482 return d.spawnSync(fetchers) 483 } 484 485 // spawnSync runs d.process and all given fetcher functions to completion in 486 // separate goroutines, returning the first error that appears. 487 func (d *Downloader) spawnSync(fetchers []func() error) error { 488 errc := make(chan error, len(fetchers)) 489 d.cancelWg.Add(len(fetchers)) 490 for _, fn := range fetchers { 491 fn := fn 492 go func() { defer d.cancelWg.Done(); errc <- fn() }() 493 } 494 // Wait for the first error, then terminate the others. 495 var err error 496 for i := 0; i < len(fetchers); i++ { 497 if i == len(fetchers)-1 { 498 // Close the queue when all fetchers have exited. 499 // This will cause the block processor to end when 500 // it has processed the queue. 501 d.queue.Close() 502 } 503 if err = <-errc; err != nil { 504 break 505 } 506 } 507 d.queue.Close() 508 d.Cancel() 509 return err 510 } 511 512 // cancel aborts all of the operations and resets the queue. However, cancel does 513 // not wait for the running download goroutines to finish. This method should be 514 // used when cancelling the downloads from inside the downloader. 515 func (d *Downloader) cancel() { 516 // Close the current cancel channel 517 d.cancelLock.Lock() 518 if d.cancelCh != nil { 519 select { 520 case <-d.cancelCh: 521 // Channel was already closed 522 default: 523 close(d.cancelCh) 524 } 525 } 526 d.cancelLock.Unlock() 527 } 528 529 // Cancel aborts all of the operations and waits for all download goroutines to 530 // finish before returning. 531 func (d *Downloader) Cancel() { 532 d.cancel() 533 d.cancelWg.Wait() 534 } 535 536 // Terminate interrupts the downloader, canceling all pending operations. 537 // The downloader cannot be reused after calling Terminate. 538 func (d *Downloader) Terminate() { 539 // Close the termination channel (make sure double close is allowed) 540 d.quitLock.Lock() 541 select { 542 case <-d.quitCh: 543 default: 544 close(d.quitCh) 545 } 546 d.quitLock.Unlock() 547 548 // Cancel any pending download requests 549 d.Cancel() 550 } 551 552 // fetchHeight retrieves the head header of the remote peer to aid in estimating 553 // the total time a pending synchronisation would take. 554 func (d *Downloader) fetchHeight(p *peerConnection) (*types.Header, error) { 555 p.log.Debug("Retrieving remote chain height") 556 557 // Request the advertised remote head block and wait for the response 558 head, _ := p.peer.Head() 559 go p.peer.RequestHeadersByHash(head, 1, 0, false) 560 561 ttl := d.requestTTL() 562 timeout := time.After(ttl) 563 for { 564 select { 565 case <-d.cancelCh: 566 return nil, errCancelBlockFetch 567 568 case packet := <-d.headerCh: 569 // Discard anything not from the origin peer 570 if packet.PeerId() != p.id { 571 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 572 break 573 } 574 // Make sure the peer actually gave something valid 575 headers := packet.(*headerPack).headers 576 if len(headers) != 1 { 577 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 578 return nil, errBadPeer 579 } 580 head := headers[0] 581 p.log.Debug("Remote head header identified", "number", head.Number, "hash", head.Hash()) 582 return head, nil 583 584 case <-timeout: 585 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 586 return nil, errTimeout 587 588 case <-d.bodyCh: 589 case <-d.receiptCh: 590 // Out of bounds delivery, ignore 591 } 592 } 593 } 594 595 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 596 // common ancestor. 597 // It returns parameters to be used for peer.RequestHeadersByNumber: 598 // from - starting block number 599 // count - number of headers to request 600 // skip - number of headers to skip 601 // and also returns 'max', the last block which is expected to be returned by the remote peers, 602 // given the (from,count,skip) 603 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 604 var ( 605 from int 606 count int 607 MaxCount = MaxHeaderFetch / 16 608 ) 609 // requestHead is the highest block that we will ask for. If requestHead is not offset, 610 // the highest block that we will get is 16 blocks back from head, which means we 611 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 612 // between us and the peer is 1-2 blocks, which is most common 613 requestHead := int(remoteHeight) - 1 614 if requestHead < 0 { 615 requestHead = 0 616 } 617 // requestBottom is the lowest block we want included in the query 618 // Ideally, we want to include just below own head 619 requestBottom := int(localHeight - 1) 620 if requestBottom < 0 { 621 requestBottom = 0 622 } 623 totalSpan := requestHead - requestBottom 624 span := 1 + totalSpan/MaxCount 625 if span < 2 { 626 span = 2 627 } 628 if span > 16 { 629 span = 16 630 } 631 632 count = 1 + totalSpan/span 633 if count > MaxCount { 634 count = MaxCount 635 } 636 if count < 2 { 637 count = 2 638 } 639 from = requestHead - (count-1)*span 640 if from < 0 { 641 from = 0 642 } 643 max := from + (count-1)*span 644 return int64(from), count, span - 1, uint64(max) 645 } 646 647 // findAncestor tries to locate the common ancestor link of the local chain and 648 // a remote peers blockchain. In the general case when our node was in sync and 649 // on the correct chain, checking the top N links should already get us a match. 650 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 651 // the head links match), we do a binary search to find the common ancestor. 652 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 653 // Figure out the valid ancestor range to prevent rewrite attacks 654 var ( 655 floor = int64(-1) 656 localHeight uint64 657 remoteHeight = remoteHeader.Number.Uint64() 658 ) 659 switch d.mode { 660 case FullSync: 661 localHeight = d.blockchain.CurrentBlock().NumberU64() 662 case FastSync: 663 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 664 default: 665 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 666 } 667 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 668 669 // Recap floor value for binary search 670 if localHeight >= MaxForkAncestry { 671 // We're above the max reorg threshold, find the earliest fork point 672 floor = int64(localHeight - MaxForkAncestry) 673 } 674 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 675 // all headers before that point will be missing. 676 if d.mode == LightSync { 677 // If we dont know the current CHT position, find it 678 if d.genesis == 0 { 679 header := d.lightchain.CurrentHeader() 680 for header != nil { 681 d.genesis = header.Number.Uint64() 682 if floor >= int64(d.genesis)-1 { 683 break 684 } 685 header = d.lightchain.GetHeaderByHash(header.ParentHash) 686 } 687 } 688 // We already know the "genesis" block number, cap floor to that 689 if floor < int64(d.genesis)-1 { 690 floor = int64(d.genesis) - 1 691 } 692 } 693 694 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 695 696 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 697 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 698 699 // Wait for the remote response to the head fetch 700 number, hash := uint64(0), common.Hash{} 701 702 ttl := d.requestTTL() 703 timeout := time.After(ttl) 704 705 for finished := false; !finished; { 706 select { 707 case <-d.cancelCh: 708 return 0, errCancelHeaderFetch 709 710 case packet := <-d.headerCh: 711 // Discard anything not from the origin peer 712 if packet.PeerId() != p.id { 713 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 714 break 715 } 716 // Make sure the peer actually gave something valid 717 headers := packet.(*headerPack).headers 718 if len(headers) == 0 { 719 p.log.Warn("Empty head header set") 720 return 0, errEmptyHeaderSet 721 } 722 // Make sure the peer's reply conforms to the request 723 for i, header := range headers { 724 expectNumber := from + int64(i)*int64(skip+1) 725 if number := header.Number.Int64(); number != expectNumber { 726 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 727 return 0, errInvalidChain 728 } 729 } 730 // Check if a common ancestor was found 731 finished = true 732 for i := len(headers) - 1; i >= 0; i-- { 733 // Skip any headers that underflow/overflow our requested set 734 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 735 continue 736 } 737 // Otherwise check if we already know the header or not 738 h := headers[i].Hash() 739 n := headers[i].Number.Uint64() 740 741 var known bool 742 switch d.mode { 743 case FullSync: 744 known = d.blockchain.HasBlock(h, n) 745 case FastSync: 746 known = d.blockchain.HasFastBlock(h, n) 747 default: 748 known = d.lightchain.HasHeader(h, n) 749 } 750 if known { 751 number, hash = n, h 752 break 753 } 754 } 755 756 case <-timeout: 757 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 758 return 0, errTimeout 759 760 case <-d.bodyCh: 761 case <-d.receiptCh: 762 // Out of bounds delivery, ignore 763 } 764 } 765 // If the head fetch already found an ancestor, return 766 if hash != (common.Hash{}) { 767 if int64(number) <= floor { 768 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 769 return 0, errInvalidAncestor 770 } 771 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 772 return number, nil 773 } 774 // Ancestor not found, we need to binary search over our chain 775 start, end := uint64(0), remoteHeight 776 if floor > 0 { 777 start = uint64(floor) 778 } 779 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 780 781 for start+1 < end { 782 // Split our chain interval in two, and request the hash to cross check 783 check := (start + end) / 2 784 785 ttl := d.requestTTL() 786 timeout := time.After(ttl) 787 788 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 789 790 // Wait until a reply arrives to this request 791 for arrived := false; !arrived; { 792 select { 793 case <-d.cancelCh: 794 return 0, errCancelHeaderFetch 795 796 case packer := <-d.headerCh: 797 // Discard anything not from the origin peer 798 if packer.PeerId() != p.id { 799 log.Debug("Received headers from incorrect peer", "peer", packer.PeerId()) 800 break 801 } 802 // Make sure the peer actually gave something valid 803 headers := packer.(*headerPack).headers 804 if len(headers) != 1 { 805 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 806 return 0, errBadPeer 807 } 808 arrived = true 809 810 // Modify the search interval based on the response 811 h := headers[0].Hash() 812 n := headers[0].Number.Uint64() 813 814 var known bool 815 switch d.mode { 816 case FullSync: 817 known = d.blockchain.HasBlock(h, n) 818 case FastSync: 819 known = d.blockchain.HasFastBlock(h, n) 820 default: 821 known = d.lightchain.HasHeader(h, n) 822 } 823 if !known { 824 end = check 825 break 826 } 827 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 828 if header.Number.Uint64() != check { 829 p.log.Debug("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 830 return 0, errBadPeer 831 } 832 start = check 833 hash = h 834 835 case <-timeout: 836 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 837 return 0, errTimeout 838 839 case <-d.bodyCh: 840 case <-d.receiptCh: 841 // Out of bounds delivery, ignore 842 } 843 } 844 } 845 // Ensure valid ancestry and return 846 if int64(start) <= floor { 847 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 848 return 0, errInvalidAncestor 849 } 850 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 851 return start, nil 852 } 853 854 // fetchHeaders keeps retrieving headers concurrently from the number 855 // requested, until no more are returned, potentially throttling on the way. To 856 // facilitate concurrency but still protect against malicious nodes sending bad 857 // headers, we construct a header chain skeleton using the "origin" peer we are 858 // syncing with, and fill in the missing headers using anyone else. Headers from 859 // other peers are only accepted if they map cleanly to the skeleton. If no one 860 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 861 // the origin is dropped. 862 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64, pivot uint64) error { 863 p.log.Debug("Directing header downloads", "origin", from) 864 defer p.log.Debug("Header download terminated") 865 866 // Create a timeout timer, and the associated header fetcher 867 skeleton := true // Skeleton assembly phase or finishing up 868 request := time.Now() // time of the last skeleton fetch request 869 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 870 <-timeout.C // timeout channel should be initially empty 871 defer timeout.Stop() 872 873 var ttl time.Duration 874 getHeaders := func(from uint64) { 875 request = time.Now() 876 877 ttl = d.requestTTL() 878 timeout.Reset(ttl) 879 880 if skeleton { 881 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 882 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 883 } else { 884 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 885 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 886 } 887 } 888 // Start pulling the header chain skeleton until all is done 889 getHeaders(from) 890 891 for { 892 select { 893 case <-d.cancelCh: 894 return errCancelHeaderFetch 895 896 case packet := <-d.headerCh: 897 // Make sure the active peer is giving us the skeleton headers 898 if packet.PeerId() != p.id { 899 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 900 break 901 } 902 headerReqTimer.UpdateSince(request) 903 timeout.Stop() 904 905 // If the skeleton's finished, pull any remaining head headers directly from the origin 906 if packet.Items() == 0 && skeleton { 907 skeleton = false 908 getHeaders(from) 909 continue 910 } 911 // If no more headers are inbound, notify the content fetchers and return 912 if packet.Items() == 0 { 913 // Don't abort header fetches while the pivot is downloading 914 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 915 p.log.Debug("No headers, waiting for pivot commit") 916 select { 917 case <-time.After(fsHeaderContCheck): 918 getHeaders(from) 919 continue 920 case <-d.cancelCh: 921 return errCancelHeaderFetch 922 } 923 } 924 // Pivot done (or not in fast sync) and no more headers, terminate the process 925 p.log.Debug("No more headers available") 926 select { 927 case d.headerProcCh <- nil: 928 return nil 929 case <-d.cancelCh: 930 return errCancelHeaderFetch 931 } 932 } 933 headers := packet.(*headerPack).headers 934 935 // If we received a skeleton batch, resolve internals concurrently 936 if skeleton { 937 filled, proced, err := d.fillHeaderSkeleton(from, headers) 938 if err != nil { 939 p.log.Debug("Skeleton chain invalid", "err", err) 940 return errInvalidChain 941 } 942 headers = filled[proced:] 943 from += uint64(proced) 944 } else { 945 // If we're closing in on the chain head, but haven't yet reached it, delay 946 // the last few headers so mini reorgs on the head don't cause invalid hash 947 // chain errors. 948 if n := len(headers); n > 0 { 949 // Retrieve the current head we're at 950 head := uint64(0) 951 if d.mode == LightSync { 952 head = d.lightchain.CurrentHeader().Number.Uint64() 953 } else { 954 head = d.blockchain.CurrentFastBlock().NumberU64() 955 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 956 head = full 957 } 958 } 959 // If the head is way older than this batch, delay the last few headers 960 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 961 delay := reorgProtHeaderDelay 962 if delay > n { 963 delay = n 964 } 965 headers = headers[:n-delay] 966 } 967 } 968 } 969 // Insert all the new headers and fetch the next batch 970 if len(headers) > 0 { 971 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 972 select { 973 case d.headerProcCh <- headers: 974 case <-d.cancelCh: 975 return errCancelHeaderFetch 976 } 977 from += uint64(len(headers)) 978 getHeaders(from) 979 } else { 980 // No headers delivered, or all of them being delayed, sleep a bit and retry 981 p.log.Trace("All headers delayed, waiting") 982 select { 983 case <-time.After(fsHeaderContCheck): 984 getHeaders(from) 985 continue 986 case <-d.cancelCh: 987 return errCancelHeaderFetch 988 } 989 } 990 991 case <-timeout.C: 992 if d.dropPeer == nil { 993 // The dropPeer method is nil when `--copydb` is used for a local copy. 994 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 995 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 996 break 997 } 998 // Header retrieval timed out, consider the peer bad and drop 999 p.log.Debug("Header request timed out", "elapsed", ttl) 1000 headerTimeoutMeter.Mark(1) 1001 d.dropPeer(p.id) 1002 1003 // Finish the sync gracefully instead of dumping the gathered data though 1004 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1005 select { 1006 case ch <- false: 1007 case <-d.cancelCh: 1008 } 1009 } 1010 select { 1011 case d.headerProcCh <- nil: 1012 case <-d.cancelCh: 1013 } 1014 return errBadPeer 1015 } 1016 } 1017 } 1018 1019 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1020 // and maps them to the provided skeleton header chain. 1021 // 1022 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1023 // immediately to the header processor to keep the rest of the pipeline full even 1024 // in the case of header stalls. 1025 // 1026 // The method returns the entire filled skeleton and also the number of headers 1027 // already forwarded for processing. 1028 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1029 log.Debug("Filling up skeleton", "from", from) 1030 d.queue.ScheduleSkeleton(from, skeleton) 1031 1032 var ( 1033 deliver = func(packet dataPack) (int, error) { 1034 pack := packet.(*headerPack) 1035 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1036 } 1037 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 1038 throttle = func() bool { return false } 1039 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, error) { 1040 return d.queue.ReserveHeaders(p, count), false, nil 1041 } 1042 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1043 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 1044 setIdle = func(p *peerConnection, accepted int) { p.SetHeadersIdle(accepted) } 1045 ) 1046 err := d.fetchParts(errCancelHeaderFetch, d.headerCh, deliver, d.queue.headerContCh, expire, 1047 d.queue.PendingHeaders, d.queue.InFlightHeaders, throttle, reserve, 1048 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1049 1050 log.Debug("Skeleton fill terminated", "err", err) 1051 1052 filled, proced := d.queue.RetrieveHeaders() 1053 return filled, proced, err 1054 } 1055 1056 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1057 // available peers, reserving a chunk of blocks for each, waiting for delivery 1058 // and also periodically checking for timeouts. 1059 func (d *Downloader) fetchBodies(from uint64) error { 1060 log.Debug("Downloading block bodies", "origin", from) 1061 1062 var ( 1063 deliver = func(packet dataPack) (int, error) { 1064 pack := packet.(*bodyPack) 1065 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1066 } 1067 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 1068 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1069 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 1070 setIdle = func(p *peerConnection, accepted int) { p.SetBodiesIdle(accepted) } 1071 ) 1072 err := d.fetchParts(errCancelBodyFetch, d.bodyCh, deliver, d.bodyWakeCh, expire, 1073 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ShouldThrottleBlocks, d.queue.ReserveBodies, 1074 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1075 1076 log.Debug("Block body download terminated", "err", err) 1077 return err 1078 } 1079 1080 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1081 // available peers, reserving a chunk of receipts for each, waiting for delivery 1082 // and also periodically checking for timeouts. 1083 func (d *Downloader) fetchReceipts(from uint64) error { 1084 log.Debug("Downloading transaction receipts", "origin", from) 1085 1086 var ( 1087 deliver = func(packet dataPack) (int, error) { 1088 pack := packet.(*receiptPack) 1089 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1090 } 1091 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1092 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1093 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1094 setIdle = func(p *peerConnection, accepted int) { p.SetReceiptsIdle(accepted) } 1095 ) 1096 err := d.fetchParts(errCancelReceiptFetch, d.receiptCh, deliver, d.receiptWakeCh, expire, 1097 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ShouldThrottleReceipts, d.queue.ReserveReceipts, 1098 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1099 1100 log.Debug("Transaction receipt download terminated", "err", err) 1101 return err 1102 } 1103 1104 // fetchParts iteratively downloads scheduled block parts, taking any available 1105 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1106 // also periodically checking for timeouts. 1107 // 1108 // As the scheduling/timeout logic mostly is the same for all downloaded data 1109 // types, this method is used by each for data gathering and is instrumented with 1110 // various callbacks to handle the slight differences between processing them. 1111 // 1112 // The instrumentation parameters: 1113 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1114 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1115 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1116 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1117 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1118 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1119 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1120 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1121 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1122 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1123 // - fetch: network callback to actually send a particular download request to a physical remote peer 1124 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1125 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1126 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1127 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1128 // - kind: textual label of the type being downloaded to display in log mesages 1129 func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1130 expire func() map[string]int, pending func() int, inFlight func() bool, throttle func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, error), 1131 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1132 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int), kind string) error { 1133 1134 // Create a ticker to detect expired retrieval tasks 1135 ticker := time.NewTicker(100 * time.Millisecond) 1136 defer ticker.Stop() 1137 1138 update := make(chan struct{}, 1) 1139 1140 // Prepare the queue and fetch block parts until the block header fetcher's done 1141 finished := false 1142 for { 1143 select { 1144 case <-d.cancelCh: 1145 return errCancel 1146 1147 case packet := <-deliveryCh: 1148 // If the peer was previously banned and failed to deliver its pack 1149 // in a reasonable time frame, ignore its message. 1150 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1151 // Deliver the received chunk of data and check chain validity 1152 accepted, err := deliver(packet) 1153 if err == errInvalidChain { 1154 return err 1155 } 1156 // Unless a peer delivered something completely else than requested (usually 1157 // caused by a timed out request which came through in the end), set it to 1158 // idle. If the delivery's stale, the peer should have already been idled. 1159 if err != errStaleDelivery { 1160 setIdle(peer, accepted) 1161 } 1162 // Issue a log to the user to see what's going on 1163 switch { 1164 case err == nil && packet.Items() == 0: 1165 peer.log.Trace("Requested data not delivered", "type", kind) 1166 case err == nil: 1167 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1168 default: 1169 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1170 } 1171 } 1172 // Blocks assembled, try to update the progress 1173 select { 1174 case update <- struct{}{}: 1175 default: 1176 } 1177 1178 case cont := <-wakeCh: 1179 // The header fetcher sent a continuation flag, check if it's done 1180 if !cont { 1181 finished = true 1182 } 1183 // Headers arrive, try to update the progress 1184 select { 1185 case update <- struct{}{}: 1186 default: 1187 } 1188 1189 case <-ticker.C: 1190 // Sanity check update the progress 1191 select { 1192 case update <- struct{}{}: 1193 default: 1194 } 1195 1196 case <-update: 1197 // Short circuit if we lost all our peers 1198 if d.peers.Len() == 0 { 1199 return errNoPeers 1200 } 1201 // Check for fetch request timeouts and demote the responsible peers 1202 for pid, fails := range expire() { 1203 if peer := d.peers.Peer(pid); peer != nil { 1204 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1205 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1206 // out that sync wise we need to get rid of the peer. 1207 // 1208 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1209 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1210 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1211 if fails > 2 { 1212 peer.log.Trace("Data delivery timed out", "type", kind) 1213 setIdle(peer, 0) 1214 } else { 1215 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1216 if d.dropPeer == nil { 1217 // The dropPeer method is nil when `--copydb` is used for a local copy. 1218 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1219 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1220 } else { 1221 d.dropPeer(pid) 1222 } 1223 } 1224 } 1225 } 1226 // If there's nothing more to fetch, wait or terminate 1227 if pending() == 0 { 1228 if !inFlight() && finished { 1229 log.Debug("Data fetching completed", "type", kind) 1230 return nil 1231 } 1232 break 1233 } 1234 // Send a download request to all idle peers, until throttled 1235 progressed, throttled, running := false, false, inFlight() 1236 idles, total := idle() 1237 1238 for _, peer := range idles { 1239 // Short circuit if throttling activated 1240 if throttle() { 1241 throttled = true 1242 break 1243 } 1244 // Short circuit if there is no more available task. 1245 if pending() == 0 { 1246 break 1247 } 1248 // Reserve a chunk of fetches for a peer. A nil can mean either that 1249 // no more headers are available, or that the peer is known not to 1250 // have them. 1251 request, progress, err := reserve(peer, capacity(peer)) 1252 if err != nil { 1253 return err 1254 } 1255 if progress { 1256 progressed = true 1257 } 1258 if request == nil { 1259 continue 1260 } 1261 if request.From > 0 { 1262 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1263 } else { 1264 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1265 } 1266 // Fetch the chunk and make sure any errors return the hashes to the queue 1267 if fetchHook != nil { 1268 fetchHook(request.Headers) 1269 } 1270 if err := fetch(peer, request); err != nil { 1271 // Although we could try and make an attempt to fix this, this error really 1272 // means that we've double allocated a fetch task to a peer. If that is the 1273 // case, the internal state of the downloader and the queue is very wrong so 1274 // better hard crash and note the error instead of silently accumulating into 1275 // a much bigger issue. 1276 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1277 } 1278 running = true 1279 } 1280 // Make sure that we have peers available for fetching. If all peers have been tried 1281 // and all failed throw an error 1282 if !progressed && !throttled && !running && len(idles) == total && pending() > 0 { 1283 return errPeersUnavailable 1284 } 1285 } 1286 } 1287 } 1288 1289 // processHeaders takes batches of retrieved headers from an input channel and 1290 // keeps processing and scheduling them into the header chain and downloader's 1291 // queue until the stream ends or a failure occurs. 1292 func (d *Downloader) processHeaders(origin uint64, pivot uint64, td *big.Int) error { 1293 // Keep a count of uncertain headers to roll back 1294 rollback := []*types.Header{} 1295 defer func() { 1296 if len(rollback) > 0 { 1297 // Flatten the headers and roll them back 1298 hashes := make([]common.Hash, len(rollback)) 1299 for i, header := range rollback { 1300 hashes[i] = header.Hash() 1301 } 1302 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1303 if d.mode != LightSync { 1304 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1305 lastBlock = d.blockchain.CurrentBlock().Number() 1306 } 1307 d.lightchain.Rollback(hashes) 1308 curFastBlock, curBlock := common.Big0, common.Big0 1309 if d.mode != LightSync { 1310 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1311 curBlock = d.blockchain.CurrentBlock().Number() 1312 } 1313 log.Warn("Rolled back headers", "count", len(hashes), 1314 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1315 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1316 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock)) 1317 } 1318 }() 1319 1320 // Wait for batches of headers to process 1321 gotHeaders := false 1322 1323 for { 1324 select { 1325 case <-d.cancelCh: 1326 return errCancelHeaderProcessing 1327 1328 case headers := <-d.headerProcCh: 1329 // Terminate header processing if we synced up 1330 if len(headers) == 0 { 1331 // Notify everyone that headers are fully processed 1332 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1333 select { 1334 case ch <- false: 1335 case <-d.cancelCh: 1336 } 1337 } 1338 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1339 // better chain compared to ours. The only exception is if its promised blocks were 1340 // already imported by other means (e.g. fetcher): 1341 // 1342 // R <remote peer>, L <local node>: Both at block 10 1343 // R: Mine block 11, and propagate it to L 1344 // L: Queue block 11 for import 1345 // L: Notice that R's head and TD increased compared to ours, start sync 1346 // L: Import of block 11 finishes 1347 // L: Sync begins, and finds common ancestor at 11 1348 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1349 // R: Nothing to give 1350 if d.mode != LightSync { 1351 head := d.blockchain.CurrentBlock() 1352 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1353 return errStallingPeer 1354 } 1355 } 1356 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1357 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1358 // of delivering the post-pivot blocks that would flag the invalid content. 1359 // 1360 // This check cannot be executed "as is" for full imports, since blocks may still be 1361 // queued for processing when the header download completes. However, as long as the 1362 // peer gave us something useful, we're already happy/progressed (above check). 1363 if d.mode == FastSync || d.mode == LightSync { 1364 head := d.lightchain.CurrentHeader() 1365 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1366 return errStallingPeer 1367 } 1368 } 1369 // Disable any rollback and return 1370 rollback = nil 1371 return nil 1372 } 1373 // Otherwise split the chunk of headers into batches and process them 1374 gotHeaders = true 1375 for len(headers) > 0 { 1376 // Terminate if something failed in between processing chunks 1377 select { 1378 case <-d.cancelCh: 1379 return errCancelHeaderProcessing 1380 default: 1381 } 1382 // Select the next chunk of headers to import 1383 limit := maxHeadersProcess 1384 if limit > len(headers) { 1385 limit = len(headers) 1386 } 1387 chunk := headers[:limit] 1388 1389 // In case of header only syncing, validate the chunk immediately 1390 if d.mode == FastSync || d.mode == LightSync { 1391 // Collect the yet unknown headers to mark them as uncertain 1392 unknown := make([]*types.Header, 0, len(headers)) 1393 for _, header := range chunk { 1394 if !d.lightchain.HasHeader(header.Hash(), header.Number.Uint64()) { 1395 unknown = append(unknown, header) 1396 } 1397 } 1398 // If we're importing pure headers, verify based on their recentness 1399 frequency := fsHeaderCheckFrequency 1400 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1401 frequency = 1 1402 } 1403 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1404 // If some headers were inserted, add them too to the rollback list 1405 if n > 0 { 1406 rollback = append(rollback, chunk[:n]...) 1407 } 1408 log.Debug("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "err", err) 1409 return errInvalidChain 1410 } 1411 // All verifications passed, store newly found uncertain headers 1412 rollback = append(rollback, unknown...) 1413 if len(rollback) > fsHeaderSafetyNet { 1414 rollback = append(rollback[:0], rollback[len(rollback)-fsHeaderSafetyNet:]...) 1415 } 1416 } 1417 // Unless we're doing light chains, schedule the headers for associated content retrieval 1418 if d.mode == FullSync || d.mode == FastSync { 1419 // If we've reached the allowed number of pending headers, stall a bit 1420 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1421 select { 1422 case <-d.cancelCh: 1423 return errCancelHeaderProcessing 1424 case <-time.After(time.Second): 1425 } 1426 } 1427 // Otherwise insert the headers for content retrieval 1428 inserts := d.queue.Schedule(chunk, origin) 1429 if len(inserts) != len(chunk) { 1430 log.Debug("Stale headers") 1431 return errBadPeer 1432 } 1433 } 1434 headers = headers[limit:] 1435 origin += uint64(limit) 1436 } 1437 // Update the highest block number we know if a higher one is found. 1438 d.syncStatsLock.Lock() 1439 if d.syncStatsChainHeight < origin { 1440 d.syncStatsChainHeight = origin - 1 1441 } 1442 d.syncStatsLock.Unlock() 1443 1444 // Signal the content downloaders of the availablility of new tasks 1445 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1446 select { 1447 case ch <- true: 1448 default: 1449 } 1450 } 1451 } 1452 } 1453 } 1454 1455 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1456 func (d *Downloader) processFullSyncContent() error { 1457 for { 1458 results := d.queue.Results(true) 1459 if len(results) == 0 { 1460 return nil 1461 } 1462 if d.chainInsertHook != nil { 1463 d.chainInsertHook(results) 1464 } 1465 if err := d.importBlockResults(results); err != nil { 1466 return err 1467 } 1468 } 1469 } 1470 1471 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1472 // Check for any early termination requests 1473 if len(results) == 0 { 1474 return nil 1475 } 1476 select { 1477 case <-d.quitCh: 1478 return errCancelContentProcessing 1479 default: 1480 } 1481 // Retrieve the a batch of results to import 1482 first, last := results[0].Header, results[len(results)-1].Header 1483 log.Debug("Inserting downloaded chain", "items", len(results), 1484 "firstnum", first.Number, "firsthash", first.Hash(), 1485 "lastnum", last.Number, "lasthash", last.Hash(), 1486 ) 1487 blocks := make([]*types.Block, len(results)) 1488 for i, result := range results { 1489 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1490 } 1491 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1492 if index < len(results) { 1493 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1494 } else { 1495 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1496 // when it needs to preprocess blocks to import a sidechain. 1497 // The importer will put together a new list of blocks to import, which is a superset 1498 // of the blocks delivered from the downloader, and the indexing will be off. 1499 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1500 } 1501 return errInvalidChain 1502 } 1503 return nil 1504 } 1505 1506 // processFastSyncContent takes fetch results from the queue and writes them to the 1507 // database. It also controls the synchronisation of state nodes of the pivot block. 1508 func (d *Downloader) processFastSyncContent(latest *types.Header) error { 1509 // Start syncing state of the reported head block. This should get us most of 1510 // the state of the pivot block. 1511 stateSync := d.syncState(latest.Root) 1512 defer stateSync.Cancel() 1513 go func() { 1514 if err := stateSync.Wait(); err != nil && err != errCancelStateFetch { 1515 d.queue.Close() // wake up Results 1516 } 1517 }() 1518 // Figure out the ideal pivot block. Note, that this goalpost may move if the 1519 // sync takes long enough for the chain head to move significantly. 1520 pivot := uint64(0) 1521 if height := latest.Number.Uint64(); height > uint64(fsMinFullBlocks) { 1522 pivot = height - uint64(fsMinFullBlocks) 1523 } 1524 // To cater for moving pivot points, track the pivot block and subsequently 1525 // accumulated download results separately. 1526 var ( 1527 oldPivot *fetchResult // Locked in pivot block, might change eventually 1528 oldTail []*fetchResult // Downloaded content after the pivot 1529 ) 1530 for { 1531 // Wait for the next batch of downloaded data to be available, and if the pivot 1532 // block became stale, move the goalpost 1533 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1534 if len(results) == 0 { 1535 // If pivot sync is done, stop 1536 if oldPivot == nil { 1537 return stateSync.Cancel() 1538 } 1539 // If sync failed, stop 1540 select { 1541 case <-d.cancelCh: 1542 return stateSync.Cancel() 1543 default: 1544 } 1545 } 1546 if d.chainInsertHook != nil { 1547 d.chainInsertHook(results) 1548 } 1549 if oldPivot != nil { 1550 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1551 } 1552 // Split around the pivot block and process the two sides via fast/full sync 1553 if atomic.LoadInt32(&d.committed) == 0 { 1554 latest = results[len(results)-1].Header 1555 if height := latest.Number.Uint64(); height > pivot+2*uint64(fsMinFullBlocks) { 1556 log.Warn("Pivot became stale, moving", "old", pivot, "new", height-uint64(fsMinFullBlocks)) 1557 pivot = height - uint64(fsMinFullBlocks) 1558 } 1559 } 1560 P, beforeP, afterP := splitAroundPivot(pivot, results) 1561 if err := d.commitFastSyncData(beforeP, stateSync); err != nil { 1562 return err 1563 } 1564 if P != nil { 1565 // If new pivot block found, cancel old state retrieval and restart 1566 if oldPivot != P { 1567 stateSync.Cancel() 1568 1569 stateSync = d.syncState(P.Header.Root) 1570 defer stateSync.Cancel() 1571 go func() { 1572 if err := stateSync.Wait(); err != nil && err != errCancelStateFetch { 1573 d.queue.Close() // wake up Results 1574 } 1575 }() 1576 oldPivot = P 1577 } 1578 // Wait for completion, occasionally checking for pivot staleness 1579 select { 1580 case <-stateSync.done: 1581 if stateSync.err != nil { 1582 return stateSync.err 1583 } 1584 if err := d.commitPivotBlock(P); err != nil { 1585 return err 1586 } 1587 oldPivot = nil 1588 1589 case <-time.After(time.Second): 1590 oldTail = afterP 1591 continue 1592 } 1593 } 1594 // Fast sync done, pivot commit done, full import 1595 if err := d.importBlockResults(afterP); err != nil { 1596 return err 1597 } 1598 } 1599 } 1600 1601 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1602 for _, result := range results { 1603 num := result.Header.Number.Uint64() 1604 switch { 1605 case num < pivot: 1606 before = append(before, result) 1607 case num == pivot: 1608 p = result 1609 default: 1610 after = append(after, result) 1611 } 1612 } 1613 return p, before, after 1614 } 1615 1616 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1617 // Check for any early termination requests 1618 if len(results) == 0 { 1619 return nil 1620 } 1621 select { 1622 case <-d.quitCh: 1623 return errCancelContentProcessing 1624 case <-stateSync.done: 1625 if err := stateSync.Wait(); err != nil { 1626 return err 1627 } 1628 default: 1629 } 1630 // Retrieve the a batch of results to import 1631 first, last := results[0].Header, results[len(results)-1].Header 1632 log.Debug("Inserting fast-sync blocks", "items", len(results), 1633 "firstnum", first.Number, "firsthash", first.Hash(), 1634 "lastnumn", last.Number, "lasthash", last.Hash(), 1635 ) 1636 blocks := make([]*types.Block, len(results)) 1637 receipts := make([]types.Receipts, len(results)) 1638 for i, result := range results { 1639 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1640 receipts[i] = result.Receipts 1641 } 1642 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts); err != nil { 1643 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1644 return errInvalidChain 1645 } 1646 return nil 1647 } 1648 1649 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1650 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1651 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1652 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}); err != nil { 1653 return err 1654 } 1655 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1656 return err 1657 } 1658 atomic.StoreInt32(&d.committed, 1) 1659 return nil 1660 } 1661 1662 // DeliverHeaders injects a new batch of block headers received from a remote 1663 // node into the download schedule. 1664 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) { 1665 return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1666 } 1667 1668 // DeliverBodies injects a new batch of block bodies received from a remote node. 1669 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) { 1670 return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1671 } 1672 1673 // DeliverReceipts injects a new batch of receipts received from a remote node. 1674 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) { 1675 return d.deliver(id, d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1676 } 1677 1678 // DeliverNodeData injects a new batch of node state data received from a remote node. 1679 func (d *Downloader) DeliverNodeData(id string, data [][]byte) (err error) { 1680 return d.deliver(id, d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1681 } 1682 1683 // deliver injects a new batch of data received from a remote node. 1684 func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1685 // Update the delivery metrics for both good and failed deliveries 1686 inMeter.Mark(int64(packet.Items())) 1687 defer func() { 1688 if err != nil { 1689 dropMeter.Mark(int64(packet.Items())) 1690 } 1691 }() 1692 // Deliver or abort if the sync is canceled while queuing 1693 d.cancelLock.RLock() 1694 cancel := d.cancelCh 1695 d.cancelLock.RUnlock() 1696 if cancel == nil { 1697 return errNoSyncActive 1698 } 1699 select { 1700 case destCh <- packet: 1701 return nil 1702 case <-cancel: 1703 return errNoSyncActive 1704 } 1705 } 1706 1707 // qosTuner is the quality of service tuning loop that occasionally gathers the 1708 // peer latency statistics and updates the estimated request round trip time. 1709 func (d *Downloader) qosTuner() { 1710 for { 1711 // Retrieve the current median RTT and integrate into the previoust target RTT 1712 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 1713 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 1714 1715 // A new RTT cycle passed, increase our confidence in the estimated RTT 1716 conf := atomic.LoadUint64(&d.rttConfidence) 1717 conf = conf + (1000000-conf)/2 1718 atomic.StoreUint64(&d.rttConfidence, conf) 1719 1720 // Log the new QoS values and sleep until the next RTT 1721 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1722 select { 1723 case <-d.quitCh: 1724 return 1725 case <-time.After(rtt): 1726 } 1727 } 1728 } 1729 1730 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 1731 // peer set, needing to reduce the confidence we have in out QoS estimates. 1732 func (d *Downloader) qosReduceConfidence() { 1733 // If we have a single peer, confidence is always 1 1734 peers := uint64(d.peers.Len()) 1735 if peers == 0 { 1736 // Ensure peer connectivity races don't catch us off guard 1737 return 1738 } 1739 if peers == 1 { 1740 atomic.StoreUint64(&d.rttConfidence, 1000000) 1741 return 1742 } 1743 // If we have a ton of peers, don't drop confidence) 1744 if peers >= uint64(qosConfidenceCap) { 1745 return 1746 } 1747 // Otherwise drop the confidence factor 1748 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 1749 if float64(conf)/1000000 < rttMinConfidence { 1750 conf = uint64(rttMinConfidence * 1000000) 1751 } 1752 atomic.StoreUint64(&d.rttConfidence, conf) 1753 1754 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1755 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1756 } 1757 1758 // requestRTT returns the current target round trip time for a download request 1759 // to complete in. 1760 // 1761 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 1762 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 1763 // be adapted to, but smaller ones are preferred (stabler download stream). 1764 func (d *Downloader) requestRTT() time.Duration { 1765 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 1766 } 1767 1768 // requestTTL returns the current timeout allowance for a single download request 1769 // to finish under. 1770 func (d *Downloader) requestTTL() time.Duration { 1771 var ( 1772 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1773 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 1774 ) 1775 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 1776 if ttl > ttlLimit { 1777 ttl = ttlLimit 1778 } 1779 return ttl 1780 }