github.com/aquanetwork/aquachain@v1.7.8/aqua/downloader/downloader.go (about) 1 // Copyright 2015 The aquachain Authors 2 // This file is part of the aquachain library. 3 // 4 // The aquachain 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 aquachain 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 aquachain 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 aquachain "gitlab.com/aquachain/aquachain" 29 "gitlab.com/aquachain/aquachain/aqua/event" 30 "gitlab.com/aquachain/aquachain/aquadb" 31 "gitlab.com/aquachain/aquachain/common" 32 "gitlab.com/aquachain/aquachain/common/log" 33 "gitlab.com/aquachain/aquachain/common/metrics" 34 "gitlab.com/aquachain/aquachain/core" 35 "gitlab.com/aquachain/aquachain/core/types" 36 "gitlab.com/aquachain/aquachain/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 rount-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 // [aqua/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 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 64 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 65 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 66 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 67 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 68 ) 69 70 var ( 71 errBusy = errors.New("busy") 72 errUnknownPeer = errors.New("peer is unknown or unhealthy") 73 errBadPeer = errors.New("action from evil peer ignored") 74 errStallingPeer = errors.New("peer is stalling") 75 errNoPeers = errors.New("no peers to keep download active") 76 errTimeout = errors.New("timeout") 77 errEmptyHeaderSet = errors.New("empty header set by peer") 78 errPeersUnavailable = errors.New("no peers available or all tried for download") 79 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 80 errInvalidChain = errors.New("retrieved hash chain is invalid") 81 errInvalidBlock = errors.New("retrieved block is invalid") 82 errInvalidBody = errors.New("retrieved block body is invalid") 83 errInvalidReceipt = errors.New("retrieved receipt is invalid") 84 errCancelBlockFetch = errors.New("block download canceled (requested)") 85 errCancelHeaderFetch = errors.New("block header download canceled (requested)") 86 errCancelBodyFetch = errors.New("block body download canceled (requested)") 87 errCancelReceiptFetch = errors.New("receipt download canceled (requested)") 88 errCancelStateFetch = errors.New("state data download canceled (requested)") 89 errCancelHeaderProcessing = errors.New("header processing canceled (requested)") 90 errCancelContentProcessing = errors.New("content processing canceled (requested)") 91 errNoSyncActive = errors.New("no sync active") 92 errTooOld = errors.New("peer doesn't speak recent enough protocol version (need version >= 62)") 93 ) 94 95 type Downloader struct { 96 mode SyncMode // Synchronisation mode defining the strategy used (per sync cycle) 97 mux *event.TypeMux // Event multiplexer to announce sync operation events 98 99 queue *queue // Scheduler for selecting the hashes to download 100 peers *peerSet // Set of active peers from which download can proceed 101 stateDB aquadb.Database 102 103 rttEstimate uint64 // Round trip time to target for download requests 104 rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops) 105 106 // Statistics 107 syncStatsChainOrigin uint64 // Origin block number where syncing started at 108 syncStatsChainHeight uint64 // Highest block number known when syncing started 109 syncStatsState stateSyncStats 110 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 111 112 lightchain LightChain 113 blockchain BlockChain 114 115 // Callbacks 116 dropPeer peerDropFn // Drops a peer for misbehaving 117 118 // Status 119 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 120 synchronising int32 121 notified int32 122 committed int32 123 124 // Channels 125 headerCh chan dataPack // [aqua/62] Channel receiving inbound block headers 126 bodyCh chan dataPack // [aqua/62] Channel receiving inbound block bodies 127 receiptCh chan dataPack // [aqua/63] Channel receiving inbound receipts 128 bodyWakeCh chan bool // [aqua/62] Channel to signal the block body fetcher of new tasks 129 receiptWakeCh chan bool // [aqua/63] Channel to signal the receipt fetcher of new tasks 130 headerProcCh chan []*types.Header // [aqua/62] Channel to feed the header processor new tasks 131 132 // for stateFetcher 133 stateSyncStart chan *stateSync 134 trackStateReq chan *stateReq 135 stateCh chan dataPack // [aqua/63] Channel receiving inbound node state data 136 137 // Cancellation and termination 138 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 139 cancelCh chan struct{} // Channel to cancel mid-flight syncs 140 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 141 142 quitCh chan struct{} // Quit channel to signal termination 143 quitLock sync.RWMutex // Lock to prevent double closes 144 145 // Testing hooks 146 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 147 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 148 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 149 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 150 } 151 152 // LightChain encapsulates functions required to synchronise a light chain. 153 type LightChain interface { 154 // HasHeader verifies a header's presence in the local chain. 155 HasHeader(common.Hash, uint64) bool 156 157 // GetHeaderByHash retrieves a header from the local chain. 158 GetHeaderByHash(common.Hash) *types.Header 159 160 // CurrentHeader retrieves the head header from the local chain. 161 CurrentHeader() *types.Header 162 163 // GetTd returns the total difficulty of a local block. 164 GetTd(common.Hash, uint64) *big.Int 165 166 // InsertHeaderChain inserts a batch of headers into the local chain. 167 InsertHeaderChain([]*types.Header, int) (int, error) 168 169 // Rollback removes a few recently added elements from the local chain. 170 Rollback([]common.Hash) 171 172 GetBlockVersion(*big.Int) types.HeaderVersion 173 } 174 175 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 176 type BlockChain interface { 177 LightChain 178 179 // HasBlock verifies a block's presence in the local chain. 180 HasBlock(common.Hash, uint64) bool 181 182 // GetBlockByHash retrieves a block from the local chain. 183 GetBlockByHash(common.Hash) *types.Block 184 185 // CurrentBlock retrieves the head block from the local chain. 186 CurrentBlock() *types.Block 187 188 // CurrentFastBlock retrieves the head fast block from the local chain. 189 CurrentFastBlock() *types.Block 190 191 // FastSyncCommitHead directly commits the head block to a certain entity. 192 FastSyncCommitHead(common.Hash) error 193 194 // InsertChain inserts a batch of blocks into the local chain. 195 InsertChain(types.Blocks) (int, error) 196 197 // InsertReceiptChain inserts a batch of receipts into the local chain. 198 InsertReceiptChain(types.Blocks, []types.Receipts) (int, error) 199 } 200 201 // New creates a new downloader to fetch hashes and blocks from remote peers. 202 func New(mode SyncMode, stateDb aquadb.Database, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 203 if lightchain == nil { 204 lightchain = chain 205 } 206 dl := &Downloader{ 207 mode: mode, 208 stateDB: stateDb, 209 mux: mux, 210 queue: newQueue(lightchain.GetBlockVersion), 211 peers: newPeerSet(), 212 rttEstimate: uint64(rttMaxEstimate), 213 rttConfidence: uint64(1000000), 214 blockchain: chain, 215 lightchain: lightchain, 216 dropPeer: dropPeer, 217 headerCh: make(chan dataPack, 1), 218 bodyCh: make(chan dataPack, 1), 219 receiptCh: make(chan dataPack, 1), 220 bodyWakeCh: make(chan bool, 1), 221 receiptWakeCh: make(chan bool, 1), 222 headerProcCh: make(chan []*types.Header, 1), 223 quitCh: make(chan struct{}), 224 stateCh: make(chan dataPack), 225 stateSyncStart: make(chan *stateSync), 226 syncStatsState: stateSyncStats{ 227 processed: core.GetTrieSyncProgress(stateDb), 228 }, 229 trackStateReq: make(chan *stateReq), 230 } 231 go dl.qosTuner() 232 go dl.stateFetcher() 233 return dl 234 } 235 236 // Progress retrieves the synchronisation boundaries, specifically the origin 237 // block where synchronisation started at (may have failed/suspended); the block 238 // or header sync is currently at; and the latest known block which the sync targets. 239 // 240 // In addition, during the state download phase of fast synchronisation the number 241 // of processed and the total number of known states are also returned. Otherwise 242 // these are zero. 243 func (d *Downloader) Progress() aquachain.SyncProgress { 244 // Lock the current stats and return the progress 245 d.syncStatsLock.RLock() 246 defer d.syncStatsLock.RUnlock() 247 248 current := uint64(0) 249 switch d.mode { 250 case FullSync: 251 current = d.blockchain.CurrentBlock().NumberU64() 252 case FastSync: 253 current = d.blockchain.CurrentFastBlock().NumberU64() 254 case LightSync: 255 current = d.lightchain.CurrentHeader().Number.Uint64() 256 common.Report("light sync activated but should not be") 257 } 258 return aquachain.SyncProgress{ 259 StartingBlock: d.syncStatsChainOrigin, 260 CurrentBlock: current, 261 HighestBlock: d.syncStatsChainHeight, 262 PulledStates: d.syncStatsState.processed, 263 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 264 } 265 } 266 267 // Synchronising returns whether the downloader is currently retrieving blocks. 268 func (d *Downloader) Synchronising() bool { 269 return atomic.LoadInt32(&d.synchronising) > 0 270 } 271 272 // RegisterPeer injects a new download peer into the set of block source to be 273 // used for fetching hashes and blocks from. 274 func (d *Downloader) RegisterPeer(id string, version int, peer Peer) error { 275 logger := log.New("peer", id) 276 logger.Trace("Registering sync peer") 277 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 278 logger.Error("Failed to register sync peer", "err", err) 279 return err 280 } 281 d.qosReduceConfidence() 282 283 return nil 284 } 285 286 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 287 func (d *Downloader) RegisterLightPeer(id string, version int, peer LightPeer) error { 288 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 289 } 290 291 // UnregisterPeer remove a peer from the known list, preventing any action from 292 // the specified peer. An effort is also made to return any pending fetches into 293 // the queue. 294 func (d *Downloader) UnregisterPeer(id string) error { 295 // Unregister the peer from the active peer set and revoke any fetch tasks 296 logger := log.New("peer", id) 297 logger.Trace("Unregistering sync peer") 298 if err := d.peers.Unregister(id); err != nil { 299 logger.Error("Failed to unregister sync peer", "err", err) 300 return err 301 } 302 d.queue.Revoke(id) 303 304 // If this peer was the master peer, abort sync immediately 305 d.cancelLock.RLock() 306 master := id == d.cancelPeer 307 d.cancelLock.RUnlock() 308 309 if master { 310 d.Cancel() 311 } 312 return nil 313 } 314 315 // Synchronise tries to sync up our local block chain with a remote peer, both 316 // adding various sanity checks as well as wrapping it with various log entries. 317 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 318 err := d.synchronise(id, head, td, mode) 319 switch err { 320 case nil: 321 case errBusy: 322 case errBadPeer: 323 log.Debug("Synchronisation failed, dropping peer", "peer", id, "err", err) 324 if d.dropPeer == nil { 325 // The dropPeer method is nil when `--copydb` is used for a local copy. 326 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 327 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 328 } else { 329 d.dropPeer(id) 330 } 331 case errTimeout, errStallingPeer, 332 errEmptyHeaderSet, errPeersUnavailable, errTooOld, 333 errInvalidAncestor, errInvalidChain: 334 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 335 if d.dropPeer == nil { 336 // The dropPeer method is nil when `--copydb` is used for a local copy. 337 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 338 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 339 } else { 340 d.dropPeer(id) 341 } 342 default: 343 log.Warn("Synchronisation failed, retrying", "err", err) 344 } 345 return err 346 } 347 348 // synchronise will select the peer and use it for synchronising. If an empty string is given 349 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 350 // checks fail an error will be returned. This method is synchronous 351 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 352 // Mock out the synchronisation if testing 353 if d.synchroniseMock != nil { 354 return d.synchroniseMock(id, hash) 355 } 356 // Make sure only one goroutine is ever allowed past this point at once 357 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 358 return errBusy 359 } 360 defer atomic.StoreInt32(&d.synchronising, 0) 361 362 // Post a user notification of the sync (only once per session) 363 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 364 log.Info("Block synchronisation started") 365 } 366 // Reset the queue, peer set and wake channels to clean any internal leftover state 367 d.queue.Reset() 368 d.peers.Reset() 369 370 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 371 select { 372 case <-ch: 373 default: 374 } 375 } 376 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 377 for empty := false; !empty; { 378 select { 379 case <-ch: 380 default: 381 empty = true 382 } 383 } 384 } 385 for empty := false; !empty; { 386 select { 387 case <-d.headerProcCh: 388 default: 389 empty = true 390 } 391 } 392 // Create cancel channel for aborting mid-flight and mark the master peer 393 d.cancelLock.Lock() 394 d.cancelCh = make(chan struct{}) 395 d.cancelPeer = id 396 d.cancelLock.Unlock() 397 398 defer d.Cancel() // No matter what, we can't leave the cancel channel open 399 400 // Set the requested sync mode, unless it's forbidden 401 d.mode = mode 402 403 // Retrieve the origin peer and initiate the downloading process 404 p := d.peers.Peer(id) 405 if p == nil { 406 return errUnknownPeer 407 } 408 return d.syncWithPeer(p, hash, td) 409 } 410 411 // syncWithPeer starts a block synchronization based on the hash chain from the 412 // specified peer and head hash. 413 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 414 d.mux.Post(StartEvent{}) 415 defer func() { 416 // reset on error 417 if err != nil { 418 d.mux.Post(FailedEvent{err}) 419 } else { 420 d.mux.Post(DoneEvent{}) 421 } 422 }() 423 if p.version < 62 { 424 return errTooOld 425 } 426 427 log.Debug("Synchronising with the network", "peer", p.id, "aqua", p.version, "head", hash, "td", td, "mode", d.mode) 428 defer func(start time.Time) { 429 log.Debug("Synchronisation terminated", "elapsed", time.Since(start)) 430 }(time.Now()) 431 432 // Look up the sync boundaries: the common ancestor and the target block 433 latest, err := d.fetchHeight(p) 434 if err != nil { 435 return err 436 } 437 height := latest.Number.Uint64() 438 439 origin, err := d.findAncestor(p, height) 440 if err != nil { 441 return err 442 } 443 d.syncStatsLock.Lock() 444 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 445 d.syncStatsChainOrigin = origin 446 } 447 d.syncStatsChainHeight = height 448 d.syncStatsLock.Unlock() 449 450 // Ensure our origin point is below any fast sync pivot point 451 pivot := uint64(0) 452 if d.mode == FastSync { 453 if height <= uint64(fsMinFullBlocks) { 454 origin = 0 455 } else { 456 pivot = height - uint64(fsMinFullBlocks) 457 if pivot <= origin { 458 origin = pivot - 1 459 } 460 } 461 } 462 d.committed = 1 463 if d.mode == FastSync && pivot != 0 { 464 d.committed = 0 465 } 466 // Initiate the sync using a concurrent header and content retrieval algorithm 467 d.queue.Prepare(origin+1, d.mode) 468 if d.syncInitHook != nil { 469 d.syncInitHook(origin, height) 470 } 471 472 fetchers := []func() error{ 473 func() error { return d.fetchHeaders(p, origin+1, pivot) }, // Headers are always retrieved 474 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 475 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 476 func() error { return d.processHeaders(origin+1, pivot, td) }, 477 } 478 if d.mode == FastSync { 479 fetchers = append(fetchers, func() error { return d.processFastSyncContent(latest) }) 480 } else if d.mode == FullSync { 481 fetchers = append(fetchers, d.processFullSyncContent) 482 } 483 return d.spawnSync(fetchers) 484 } 485 486 // spawnSync runs d.process and all given fetcher functions to completion in 487 // separate goroutines, returning the first error that appears. 488 func (d *Downloader) spawnSync(fetchers []func() error) error { 489 var wg sync.WaitGroup 490 errc := make(chan error, len(fetchers)) 491 wg.Add(len(fetchers)) 492 for _, fn := range fetchers { 493 fn := fn 494 go func() { defer wg.Done(); errc <- fn() }() 495 } 496 // Wait for the first error, then terminate the others. 497 var err error 498 for i := 0; i < len(fetchers); i++ { 499 if i == len(fetchers)-1 { 500 // Close the queue when all fetchers have exited. 501 // This will cause the block processor to end when 502 // it has processed the queue. 503 d.queue.Close() 504 } 505 if err = <-errc; err != nil { 506 break 507 } 508 } 509 d.queue.Close() 510 d.Cancel() 511 wg.Wait() 512 return err 513 } 514 515 // Cancel cancels all of the operations and resets the queue. It returns true 516 // if the cancel operation was completed. 517 func (d *Downloader) Cancel() { 518 // Close the current cancel channel 519 d.cancelLock.Lock() 520 if d.cancelCh != nil { 521 select { 522 case <-d.cancelCh: 523 // Channel was already closed 524 default: 525 close(d.cancelCh) 526 } 527 } 528 d.cancelLock.Unlock() 529 } 530 531 // Terminate interrupts the downloader, canceling all pending operations. 532 // The downloader cannot be reused after calling Terminate. 533 func (d *Downloader) Terminate() { 534 // Close the termination channel (make sure double close is allowed) 535 d.quitLock.Lock() 536 select { 537 case <-d.quitCh: 538 default: 539 close(d.quitCh) 540 } 541 d.quitLock.Unlock() 542 543 // Cancel any pending download requests 544 d.Cancel() 545 } 546 547 // fetchHeight retrieves the head header of the remote peer to aid in estimating 548 // the total time a pending synchronisation would take. 549 func (d *Downloader) fetchHeight(p *peerConnection) (*types.Header, error) { 550 p.log.Debug("Retrieving remote chain height") 551 552 // Request the advertised remote head block and wait for the response 553 head, _ := p.peer.Head() 554 555 go p.peer.RequestHeadersByHash(head, 1, 0, false) 556 557 ttl := d.requestTTL() 558 timeout := time.After(ttl) 559 for { 560 select { 561 case <-d.cancelCh: 562 return nil, errCancelBlockFetch 563 564 case packet := <-d.headerCh: 565 // Discard anything not from the origin peer 566 if packet.PeerId() != p.id { 567 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 568 break 569 } 570 // Make sure the peer actually gave something valid 571 headers := packet.(*headerPack).headers 572 if len(headers) != 1 { 573 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 574 return nil, errBadPeer 575 } 576 head := headers[0] 577 p.log.Debug("Remote head header identified", "number", head.Number, "hash-no-nonce", head.HashNoNonce(), "nonce", head.Nonce) 578 return head, nil 579 580 case <-timeout: 581 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 582 return nil, errTimeout 583 584 case <-d.bodyCh: 585 case <-d.receiptCh: 586 // Out of bounds delivery, ignore 587 } 588 } 589 } 590 591 // findAncestor tries to locate the common ancestor link of the local chain and 592 // a remote peers blockchain. In the general case when our node was in sync and 593 // on the correct chain, checking the top N links should already get us a match. 594 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 595 // the head links match), we do a binary search to find the common ancestor. 596 func (d *Downloader) findAncestor(p *peerConnection, height uint64) (uint64, error) { 597 // Figure out the valid ancestor range to prevent rewrite attacks 598 floor, ceil := int64(-1), d.lightchain.CurrentHeader().Number.Uint64() 599 600 if d.mode == FullSync { 601 ceil = d.blockchain.CurrentBlock().NumberU64() 602 } else if d.mode == FastSync { 603 ceil = d.blockchain.CurrentFastBlock().NumberU64() 604 } 605 if ceil >= MaxForkAncestry { 606 floor = int64(ceil - MaxForkAncestry) 607 } 608 p.log.Debug("Looking for common ancestor", "local", ceil, "remote", height) 609 610 // Request the topmost blocks to short circuit binary ancestor lookup 611 head := ceil 612 if head > height { 613 head = height 614 } 615 from := int64(head) - int64(MaxHeaderFetch) 616 if from < 0 { 617 from = 0 618 } 619 // Span out with 15 block gaps into the future to catch bad head reports 620 limit := 2 * MaxHeaderFetch / 16 621 count := 1 + int((int64(ceil)-from)/16) 622 if count > limit { 623 count = limit 624 } 625 go p.peer.RequestHeadersByNumber(uint64(from), count, 15, false) 626 627 // Wait for the remote response to the head fetch 628 number, hash := uint64(0), common.Hash{} 629 630 ttl := d.requestTTL() 631 timeout := time.After(ttl) 632 633 for finished := false; !finished; { 634 select { 635 case <-d.cancelCh: 636 return 0, errCancelHeaderFetch 637 638 case packet := <-d.headerCh: 639 // Discard anything not from the origin peer 640 if packet.PeerId() != p.id { 641 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 642 break 643 } 644 // Make sure the peer actually gave something valid 645 headers := packet.(*headerPack).headers 646 if len(headers) == 0 { 647 p.log.Warn("Empty head header set") 648 return 0, errEmptyHeaderSet 649 } 650 // Make sure the peer's reply conforms to the request 651 for i := 0; i < len(headers); i++ { 652 if number := headers[i].Number.Int64(); number != from+int64(i)*16 { 653 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", from+int64(i)*16, "received", number) 654 return 0, errInvalidChain 655 } 656 } 657 // Check if a common ancestor was found 658 finished = true 659 for i := len(headers) - 1; i >= 0; i-- { 660 // Skip any headers that underflow/overflow our requested set 661 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > ceil { 662 continue 663 } 664 var version byte = 0 665 switch d.mode { 666 case FullSync: 667 // cache the header hash with the correct version (using the block height) 668 version = byte(d.blockchain.GetBlockVersion(headers[i].Number)) 669 hcache := headers[i].SetVersion(version) 670 // check if we already know the header or not 671 if d.blockchain.HasBlock(hcache, headers[i].Number.Uint64()) { 672 number, hash = headers[i].Number.Uint64(), hcache 673 } 674 default: 675 // cache the header hash with the correct version (using the block height) 676 version = byte(d.lightchain.GetBlockVersion(headers[i].Number)) 677 hcache := headers[i].SetVersion(version) 678 // check if we already know the header or not 679 if d.lightchain.HasHeader(hcache, headers[i].Number.Uint64()) { 680 number, hash = headers[i].Number.Uint64(), hcache 681 } 682 } 683 684 // If every header is known, even future ones, the peer straight out lied about its head 685 if number > height && i == limit-1 { 686 p.log.Warn("Lied about chain head", "reported", height, "found", number) 687 return 0, errStallingPeer 688 } 689 break 690 } 691 692 case <-timeout: 693 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 694 return 0, errTimeout 695 696 case <-d.bodyCh: 697 case <-d.receiptCh: 698 // Out of bounds delivery, ignore 699 } 700 } 701 // If the head fetch already found an ancestor, return 702 if !common.EmptyHash(hash) { 703 if int64(number) <= floor { 704 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 705 return 0, errInvalidAncestor 706 } 707 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 708 return number, nil 709 } 710 // Ancestor not found, we need to binary search over our chain 711 start, end := uint64(0), head 712 if floor > 0 { 713 start = uint64(floor) 714 } 715 for start+1 < end { 716 // Split our chain interval in two, and request the hash to cross check 717 check := (start + end) / 2 718 719 ttl := d.requestTTL() 720 timeout := time.After(ttl) 721 722 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 723 724 // Wait until a reply arrives to this request 725 for arrived := false; !arrived; { 726 select { 727 case <-d.cancelCh: 728 return 0, errCancelHeaderFetch 729 730 case packer := <-d.headerCh: 731 // Discard anything not from the origin peer 732 if packer.PeerId() != p.id { 733 log.Debug("Received headers from incorrect peer", "peer", packer.PeerId()) 734 break 735 } 736 // Make sure the peer actually gave something valid 737 headers := packer.(*headerPack).headers 738 if len(headers) != 1 { 739 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 740 return 0, errBadPeer 741 } 742 arrived = true 743 744 // Modify the search interval based on the response 745 firstversion := byte(d.blockchain.GetBlockVersion(headers[0].Number)) 746 if (d.mode == FullSync && !d.blockchain.HasBlock(headers[0].SetVersion(firstversion), headers[0].Number.Uint64())) || 747 (d.mode != FullSync && !d.lightchain.HasHeader(headers[0].SetVersion(firstversion), headers[0].Number.Uint64())) { 748 end = check 749 break 750 } 751 header := d.lightchain.GetHeaderByHash(headers[0].Hash()) // Independent of sync mode, header surely exists 752 if header.Number.Uint64() != check { 753 p.log.Debug("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 754 return 0, errBadPeer 755 } 756 start = check 757 758 case <-timeout: 759 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 760 return 0, errTimeout 761 762 case <-d.bodyCh: 763 case <-d.receiptCh: 764 // Out of bounds delivery, ignore 765 } 766 } 767 } 768 // Ensure valid ancestry and return 769 if int64(start) <= floor { 770 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 771 return 0, errInvalidAncestor 772 } 773 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 774 return start, nil 775 } 776 777 // fetchHeaders keeps retrieving headers concurrently from the number 778 // requested, until no more are returned, potentially throttling on the way. To 779 // facilitate concurrency but still protect against malicious nodes sending bad 780 // headers, we construct a header chain skeleton using the "origin" peer we are 781 // syncing with, and fill in the missing headers using anyone else. Headers from 782 // other peers are only accepted if they map cleanly to the skeleton. If no one 783 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 784 // the origin is dropped. 785 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64, pivot uint64) error { 786 p.log.Debug("Directing header downloads", "origin", from) 787 defer p.log.Debug("Header download terminated") 788 789 // Create a timeout timer, and the associated header fetcher 790 skeleton := true // Skeleton assembly phase or finishing up 791 request := time.Now() // time of the last skeleton fetch request 792 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 793 <-timeout.C // timeout channel should be initially empty 794 defer timeout.Stop() 795 796 var ttl time.Duration 797 getHeaders := func(from uint64) { 798 request = time.Now() 799 800 ttl = d.requestTTL() 801 timeout.Reset(ttl) 802 803 if skeleton { 804 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 805 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 806 } else { 807 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 808 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 809 } 810 } 811 // Start pulling the header chain skeleton until all is done 812 getHeaders(from) 813 814 for { 815 select { 816 case <-d.cancelCh: 817 return errCancelHeaderFetch 818 819 case packet := <-d.headerCh: 820 // Make sure the active peer is giving us the skeleton headers 821 if packet.PeerId() != p.id { 822 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 823 break 824 } 825 headerReqTimer.UpdateSince(request) 826 timeout.Stop() 827 828 // If the skeleton's finished, pull any remaining head headers directly from the origin 829 if packet.Items() == 0 && skeleton { 830 skeleton = false 831 getHeaders(from) 832 continue 833 } 834 // If no more headers are inbound, notify the content fetchers and return 835 if packet.Items() == 0 { 836 // Don't abort header fetches while the pivot is downloading 837 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 838 p.log.Debug("No headers, waiting for pivot commit") 839 select { 840 case <-time.After(fsHeaderContCheck): 841 getHeaders(from) 842 continue 843 case <-d.cancelCh: 844 return errCancelHeaderFetch 845 } 846 } 847 // Pivot done (or not in fast sync) and no more headers, terminate the process 848 p.log.Debug("No more headers available") 849 select { 850 case d.headerProcCh <- nil: 851 return nil 852 case <-d.cancelCh: 853 return errCancelHeaderFetch 854 } 855 } 856 headers := packet.(*headerPack).headers 857 858 // If we received a skeleton batch, resolve internals concurrently 859 if skeleton { 860 filled, proced, err := d.fillHeaderSkeleton(from, headers) 861 if err != nil { 862 p.log.Debug("Skeleton chain invalid", "err", err) 863 return errInvalidChain 864 } 865 headers = filled[proced:] 866 from += uint64(proced) 867 } 868 // Insert all the new headers and fetch the next batch 869 if len(headers) > 0 { 870 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 871 select { 872 case d.headerProcCh <- headers: 873 case <-d.cancelCh: 874 return errCancelHeaderFetch 875 } 876 from += uint64(len(headers)) 877 } 878 getHeaders(from) 879 880 case <-timeout.C: 881 if d.dropPeer == nil { 882 // The dropPeer method is nil when `--copydb` is used for a local copy. 883 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 884 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 885 break 886 } 887 // Header retrieval timed out, consider the peer bad and drop 888 p.log.Debug("Header request timed out", "elapsed", ttl) 889 headerTimeoutMeter.Mark(1) 890 d.dropPeer(p.id) 891 892 // Finish the sync gracefully instead of dumping the gathered data though 893 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 894 select { 895 case ch <- false: 896 case <-d.cancelCh: 897 } 898 } 899 select { 900 case d.headerProcCh <- nil: 901 case <-d.cancelCh: 902 } 903 return errBadPeer 904 } 905 } 906 } 907 908 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 909 // and maps them to the provided skeleton header chain. 910 // 911 // Any partial results from the beginning of the skeleton is (if possible) forwarded 912 // immediately to the header processor to keep the rest of the pipeline full even 913 // in the case of header stalls. 914 // 915 // The method returs the entire filled skeleton and also the number of headers 916 // already forwarded for processing. 917 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 918 log.Debug("Filling up skeleton", "from", from) 919 d.queue.ScheduleSkeleton(from, skeleton) 920 921 var ( 922 deliver = func(packet dataPack) (int, error) { 923 pack := packet.(*headerPack) 924 return d.queue.DeliverHeaders(pack.peerId, pack.headers, d.headerProcCh) 925 } 926 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 927 throttle = func() bool { return false } 928 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, error) { 929 return d.queue.ReserveHeaders(p, count), false, nil 930 } 931 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 932 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 933 setIdle = func(p *peerConnection, accepted int) { p.SetHeadersIdle(accepted) } 934 ) 935 err := d.fetchParts(errCancelHeaderFetch, d.headerCh, deliver, d.queue.headerContCh, expire, 936 d.queue.PendingHeaders, d.queue.InFlightHeaders, throttle, reserve, 937 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 938 939 log.Debug("Skeleton fill terminated", "err", err) 940 941 filled, proced := d.queue.RetrieveHeaders() 942 return filled, proced, err 943 } 944 945 // fetchBodies iteratively downloads the scheduled block bodies, taking any 946 // available peers, reserving a chunk of blocks for each, waiting for delivery 947 // and also periodically checking for timeouts. 948 func (d *Downloader) fetchBodies(from uint64) error { 949 log.Debug("Downloading block bodies", "origin", from) 950 951 var ( 952 deliver = func(packet dataPack) (int, error) { 953 pack := packet.(*bodyPack) 954 return d.queue.DeliverBodies(pack.peerId, pack.transactions, pack.uncles) 955 } 956 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 957 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 958 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 959 setIdle = func(p *peerConnection, accepted int) { p.SetBodiesIdle(accepted) } 960 ) 961 err := d.fetchParts(errCancelBodyFetch, d.bodyCh, deliver, d.bodyWakeCh, expire, 962 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ShouldThrottleBlocks, d.queue.ReserveBodies, 963 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 964 965 log.Debug("Block body download terminated", "err", err) 966 return err 967 } 968 969 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 970 // available peers, reserving a chunk of receipts for each, waiting for delivery 971 // and also periodically checking for timeouts. 972 func (d *Downloader) fetchReceipts(from uint64) error { 973 log.Debug("Downloading transaction receipts", "origin", from) 974 975 var ( 976 deliver = func(packet dataPack) (int, error) { 977 pack := packet.(*receiptPack) 978 return d.queue.DeliverReceipts(pack.peerId, pack.receipts) 979 } 980 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 981 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 982 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 983 setIdle = func(p *peerConnection, accepted int) { p.SetReceiptsIdle(accepted) } 984 ) 985 err := d.fetchParts(errCancelReceiptFetch, d.receiptCh, deliver, d.receiptWakeCh, expire, 986 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ShouldThrottleReceipts, d.queue.ReserveReceipts, 987 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 988 989 log.Debug("Transaction receipt download terminated", "err", err) 990 return err 991 } 992 993 // fetchParts iteratively downloads scheduled block parts, taking any available 994 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 995 // also periodically checking for timeouts. 996 // 997 // As the scheduling/timeout logic mostly is the same for all downloaded data 998 // types, this method is used by each for data gathering and is instrumented with 999 // various callbacks to handle the slight differences between processing them. 1000 // 1001 // The instrumentation parameters: 1002 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1003 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1004 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1005 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1006 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1007 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1008 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1009 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1010 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1011 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1012 // - fetch: network callback to actually send a particular download request to a physical remote peer 1013 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1014 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1015 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1016 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1017 // - kind: textual label of the type being downloaded to display in log mesages 1018 func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1019 expire func() map[string]int, pending func() int, inFlight func() bool, throttle func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, error), 1020 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1021 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int), kind string) error { 1022 1023 // Create a ticker to detect expired retrieval tasks 1024 ticker := time.NewTicker(100 * time.Millisecond) 1025 defer ticker.Stop() 1026 1027 update := make(chan struct{}, 1) 1028 1029 // Prepare the queue and fetch block parts until the block header fetcher's done 1030 finished := false 1031 for { 1032 select { 1033 case <-d.cancelCh: 1034 return errCancel 1035 1036 case packet := <-deliveryCh: 1037 // If the peer was previously banned and failed to deliver its pack 1038 // in a reasonable time frame, ignore its message. 1039 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1040 // Deliver the received chunk of data and check chain validity 1041 accepted, err := deliver(packet) 1042 if err == errInvalidChain { 1043 return err 1044 } 1045 // Unless a peer delivered something completely else than requested (usually 1046 // caused by a timed out request which came through in the end), set it to 1047 // idle. If the delivery's stale, the peer should have already been idled. 1048 if err != errStaleDelivery { 1049 setIdle(peer, accepted) 1050 } 1051 // Issue a log to the user to see what's going on 1052 switch { 1053 case err == nil && packet.Items() == 0: 1054 peer.log.Trace("Requested data not delivered", "type", kind) 1055 case err == nil: 1056 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1057 default: 1058 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1059 } 1060 } 1061 // Blocks assembled, try to update the progress 1062 select { 1063 case update <- struct{}{}: 1064 default: 1065 } 1066 1067 case cont := <-wakeCh: 1068 // The header fetcher sent a continuation flag, check if it's done 1069 if !cont { 1070 finished = true 1071 } 1072 // Headers arrive, try to update the progress 1073 select { 1074 case update <- struct{}{}: 1075 default: 1076 } 1077 1078 case <-ticker.C: 1079 // Sanity check update the progress 1080 select { 1081 case update <- struct{}{}: 1082 default: 1083 } 1084 1085 case <-update: 1086 // Short circuit if we lost all our peers 1087 if d.peers.Len() == 0 { 1088 return errNoPeers 1089 } 1090 // Check for fetch request timeouts and demote the responsible peers 1091 for pid, fails := range expire() { 1092 if peer := d.peers.Peer(pid); peer != nil { 1093 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1094 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1095 // out that sync wise we need to get rid of the peer. 1096 // 1097 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1098 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1099 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1100 if fails > 2 { 1101 peer.log.Trace("Data delivery timed out", "type", kind) 1102 setIdle(peer, 0) 1103 } else { 1104 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1105 if d.dropPeer == nil { 1106 // The dropPeer method is nil when `--copydb` is used for a local copy. 1107 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1108 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1109 } else { 1110 d.dropPeer(pid) 1111 } 1112 } 1113 } 1114 } 1115 // If there's nothing more to fetch, wait or terminate 1116 if pending() == 0 { 1117 if !inFlight() && finished { 1118 log.Debug("Data fetching completed", "type", kind) 1119 return nil 1120 } 1121 break 1122 } 1123 // Send a download request to all idle peers, until throttled 1124 progressed, throttled, running := false, false, inFlight() 1125 idles, total := idle() 1126 1127 for _, peer := range idles { 1128 // Short circuit if throttling activated 1129 if throttle() { 1130 throttled = true 1131 break 1132 } 1133 // Short circuit if there is no more available task. 1134 if pending() == 0 { 1135 break 1136 } 1137 // Reserve a chunk of fetches for a peer. A nil can mean either that 1138 // no more headers are available, or that the peer is known not to 1139 // have them. 1140 request, progress, err := reserve(peer, capacity(peer)) 1141 if err != nil { 1142 return err 1143 } 1144 if progress { 1145 progressed = true 1146 } 1147 if request == nil { 1148 continue 1149 } 1150 if request.From > 0 { 1151 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1152 } else { 1153 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1154 } 1155 // Fetch the chunk and make sure any errors return the hashes to the queue 1156 if fetchHook != nil { 1157 fetchHook(request.Headers) 1158 } 1159 if err := fetch(peer, request); err != nil { 1160 // Although we could try and make an attempt to fix this, this error really 1161 // means that we've double allocated a fetch task to a peer. If that is the 1162 // case, the internal state of the downloader and the queue is very wrong so 1163 // better hard crash and note the error instead of silently accumulating into 1164 // a much bigger issue. 1165 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1166 } 1167 running = true 1168 } 1169 // Make sure that we have peers available for fetching. If all peers have been tried 1170 // and all failed throw an error 1171 if !progressed && !throttled && !running && len(idles) == total && pending() > 0 { 1172 return errPeersUnavailable 1173 } 1174 } 1175 } 1176 } 1177 1178 // processHeaders takes batches of retrieved headers from an input channel and 1179 // keeps processing and scheduling them into the header chain and downloader's 1180 // queue until the stream ends or a failure occurs. 1181 func (d *Downloader) processHeaders(origin uint64, pivot uint64, td *big.Int) error { 1182 // Keep a count of uncertain headers to roll back 1183 rollback := []*types.Header{} 1184 defer func() { 1185 if len(rollback) > 0 { 1186 // Flatten the headers and roll them back 1187 hashes := make([]common.Hash, len(rollback)) 1188 for i, header := range rollback { 1189 hashes[i] = header.Hash() 1190 } 1191 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1192 //if d.mode != LightSync { 1193 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1194 lastBlock = d.blockchain.CurrentBlock().Number() 1195 //} 1196 d.lightchain.Rollback(hashes) 1197 curFastBlock, curBlock := common.Big0, common.Big0 1198 //if d.mode != LightSync { 1199 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1200 curBlock = d.blockchain.CurrentBlock().Number() 1201 //} 1202 log.Warn("Rolled back headers", "count", len(hashes), 1203 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1204 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1205 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock)) 1206 } 1207 }() 1208 1209 // Wait for batches of headers to process 1210 gotHeaders := false 1211 1212 for { 1213 select { 1214 case <-d.cancelCh: 1215 return errCancelHeaderProcessing 1216 1217 case headers := <-d.headerProcCh: 1218 // Terminate header processing if we synced up 1219 if len(headers) == 0 { 1220 // Notify everyone that headers are fully processed 1221 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1222 select { 1223 case ch <- false: 1224 case <-d.cancelCh: 1225 } 1226 } 1227 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1228 // better chain compared to ours. The only exception is if its promised blocks were 1229 // already imported by other means (e.g. fecher): 1230 // 1231 // R <remote peer>, L <local node>: Both at block 10 1232 // R: Mine block 11, and propagate it to L 1233 // L: Queue block 11 for import 1234 // L: Notice that R's head and TD increased compared to ours, start sync 1235 // L: Import of block 11 finishes 1236 // L: Sync begins, and finds common ancestor at 11 1237 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1238 // R: Nothing to give 1239 //if d.mode != LightSync { 1240 head := d.blockchain.CurrentBlock() 1241 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1242 return errStallingPeer 1243 } 1244 //} 1245 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1246 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1247 // of delivering the post-pivot blocks that would flag the invalid content. 1248 // 1249 // This check cannot be executed "as is" for full imports, since blocks may still be 1250 // queued for processing when the header download completes. However, as long as the 1251 // peer gave us something useful, we're already happy/progressed (above check). 1252 //if d.mode == FastSync || d.mode == LightSync { 1253 if d.mode == FastSync { 1254 head := d.lightchain.CurrentHeader() 1255 if td.Cmp(d.lightchain.GetTd(head.SetVersion(byte(d.lightchain.GetBlockVersion(head.Number))), head.Number.Uint64())) > 0 { 1256 return errStallingPeer 1257 } 1258 } 1259 // Disable any rollback and return 1260 rollback = nil 1261 return nil 1262 } 1263 // Otherwise split the chunk of headers into batches and process them 1264 gotHeaders = true 1265 1266 for len(headers) > 0 { 1267 // Terminate if something failed in between processing chunks 1268 select { 1269 case <-d.cancelCh: 1270 return errCancelHeaderProcessing 1271 default: 1272 } 1273 // Select the next chunk of headers to import 1274 limit := maxHeadersProcess 1275 if limit > len(headers) { 1276 limit = len(headers) 1277 } 1278 chunk := headers[:limit] 1279 1280 // In case of header only syncing, validate the chunk immediately 1281 //if d.mode == FastSync || d.mode == LightSync { 1282 if d.mode == FastSync { 1283 // Collect the yet unknown headers to mark them as uncertain 1284 unknown := make([]*types.Header, 0, len(headers)) 1285 for _, header := range chunk { // copies 1286 header.SetVersion(byte(d.lightchain.GetBlockVersion(header.Number))) 1287 if !d.lightchain.HasHeader(header.Hash(), header.Number.Uint64()) { 1288 unknown = append(unknown, header) 1289 } 1290 } 1291 // If we're importing pure headers, verify based on their recentness 1292 frequency := fsHeaderCheckFrequency 1293 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1294 frequency = 1 1295 } 1296 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1297 // If some headers were inserted, add them too to the rollback list 1298 if n > 0 { 1299 rollback = append(rollback, chunk[:n]...) 1300 } 1301 log.Debug("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "algo", chunk[n].Version, "err", err) 1302 return errInvalidChain 1303 } 1304 // All verifications passed, store newly found uncertain headers 1305 rollback = append(rollback, unknown...) 1306 if len(rollback) > fsHeaderSafetyNet { 1307 rollback = append(rollback[:0], rollback[len(rollback)-fsHeaderSafetyNet:]...) 1308 } 1309 } 1310 // Unless we're doing light chains, schedule the headers for associated content retrieval 1311 if d.mode == FullSync || d.mode == FastSync { 1312 // If we've reached the allowed number of pending headers, stall a bit 1313 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1314 select { 1315 case <-d.cancelCh: 1316 return errCancelHeaderProcessing 1317 case <-time.After(time.Second): 1318 } 1319 } 1320 // Otherwise insert the headers for content retrieval 1321 inserts := d.queue.Schedule(chunk, origin) 1322 if len(inserts) != len(chunk) { 1323 log.Debug("Stale headers") 1324 return errBadPeer 1325 } 1326 } 1327 headers = headers[limit:] 1328 origin += uint64(limit) 1329 } 1330 1331 // Update the highest block number we know if a higher one is found. 1332 d.syncStatsLock.Lock() 1333 if d.syncStatsChainHeight < origin { 1334 d.syncStatsChainHeight = origin - 1 1335 } 1336 d.syncStatsLock.Unlock() 1337 1338 // Signal the content downloaders of the availablility of new tasks 1339 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1340 select { 1341 case ch <- true: 1342 default: 1343 } 1344 } 1345 } 1346 } 1347 } 1348 1349 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1350 func (d *Downloader) processFullSyncContent() error { 1351 for { 1352 results := d.queue.Results(true) 1353 if len(results) == 0 { 1354 return nil 1355 } 1356 if d.chainInsertHook != nil { 1357 d.chainInsertHook(results) 1358 } 1359 if err := d.importBlockResults(results); err != nil { 1360 return err 1361 } 1362 } 1363 } 1364 1365 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1366 // Check for any early termination requests 1367 if len(results) == 0 { 1368 return nil 1369 } 1370 select { 1371 case <-d.quitCh: 1372 return errCancelContentProcessing 1373 default: 1374 } 1375 // Retrieve the a batch of results to import 1376 first, last := results[0].Header, results[len(results)-1].Header 1377 log.Debug("Inserting downloaded chain", "items", len(results), 1378 "firstnum", first.Number, "firsthash", first.Hash(), "algo", first.Version, 1379 "lastnum", last.Number, "lasthash", last.Hash(), "algo", last.Version, 1380 ) 1381 blocks := make([]*types.Block, len(results)) 1382 for i, result := range results { 1383 result.Header.Version = d.blockchain.GetBlockVersion(result.Header.Number) 1384 for i := range result.Uncles { 1385 result.Uncles[i].Version = d.blockchain.GetBlockVersion(result.Uncles[i].Number) 1386 } 1387 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1388 } 1389 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1390 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "err", err) 1391 return errInvalidChain 1392 } 1393 return nil 1394 } 1395 1396 // processFastSyncContent takes fetch results from the queue and writes them to the 1397 // database. It also controls the synchronisation of state nodes of the pivot block. 1398 func (d *Downloader) processFastSyncContent(latest *types.Header) error { 1399 // Start syncing state of the reported head block. This should get us most of 1400 // the state of the pivot block. 1401 stateSync := d.syncState(latest.Root) 1402 defer stateSync.Cancel() 1403 go func() { 1404 if err := stateSync.Wait(); err != nil && err != errCancelStateFetch { 1405 d.queue.Close() // wake up WaitResults 1406 } 1407 }() 1408 // Figure out the ideal pivot block. Note, that this goalpost may move if the 1409 // sync takes long enough for the chain head to move significantly. 1410 pivot := uint64(0) 1411 if height := latest.Number.Uint64(); height > uint64(fsMinFullBlocks) { 1412 pivot = height - uint64(fsMinFullBlocks) 1413 } 1414 // To cater for moving pivot points, track the pivot block and subsequently 1415 // accumulated download results separatey. 1416 var ( 1417 oldPivot *fetchResult // Locked in pivot block, might change eventually 1418 oldTail []*fetchResult // Downloaded content after the pivot 1419 ) 1420 for { 1421 // Wait for the next batch of downloaded data to be available, and if the pivot 1422 // block became stale, move the goalpost 1423 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1424 if len(results) == 0 { 1425 // If pivot sync is done, stop 1426 if oldPivot == nil { 1427 return stateSync.Cancel() 1428 } 1429 // If sync failed, stop 1430 select { 1431 case <-d.cancelCh: 1432 return stateSync.Cancel() 1433 default: 1434 } 1435 } 1436 if d.chainInsertHook != nil { 1437 d.chainInsertHook(results) 1438 } 1439 if oldPivot != nil { 1440 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1441 } 1442 // Split around the pivot block and process the two sides via fast/full sync 1443 if atomic.LoadInt32(&d.committed) == 0 { 1444 latest = results[len(results)-1].Header 1445 if height := latest.Number.Uint64(); height > pivot+2*uint64(fsMinFullBlocks) { 1446 log.Warn("Pivot became stale, moving", "old", pivot, "new", height-uint64(fsMinFullBlocks)) 1447 pivot = height - uint64(fsMinFullBlocks) 1448 } 1449 } 1450 P, beforeP, afterP := splitAroundPivot(pivot, results) 1451 if err := d.commitFastSyncData(beforeP, stateSync); err != nil { 1452 return err 1453 } 1454 if P != nil { 1455 // If new pivot block found, cancel old state retrieval and restart 1456 if oldPivot != P { 1457 stateSync.Cancel() 1458 1459 stateSync = d.syncState(P.Header.Root) 1460 defer stateSync.Cancel() 1461 go func() { 1462 if err := stateSync.Wait(); err != nil && err != errCancelStateFetch { 1463 d.queue.Close() // wake up WaitResults 1464 } 1465 }() 1466 oldPivot = P 1467 } 1468 // Wait for completion, occasionally checking for pivot staleness 1469 select { 1470 case <-stateSync.done: 1471 if stateSync.err != nil { 1472 return stateSync.err 1473 } 1474 if err := d.commitPivotBlock(P); err != nil { 1475 return err 1476 } 1477 oldPivot = nil 1478 1479 case <-time.After(time.Second): 1480 oldTail = afterP 1481 continue 1482 } 1483 } 1484 // Fast sync done, pivot commit done, full import 1485 if err := d.importBlockResults(afterP); err != nil { 1486 return err 1487 } 1488 } 1489 } 1490 1491 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1492 for _, result := range results { 1493 num := result.Header.Number.Uint64() 1494 switch { 1495 case num < pivot: 1496 before = append(before, result) 1497 case num == pivot: 1498 p = result 1499 default: 1500 after = append(after, result) 1501 } 1502 } 1503 return p, before, after 1504 } 1505 1506 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1507 // Check for any early termination requests 1508 if len(results) == 0 { 1509 return nil 1510 } 1511 select { 1512 case <-d.quitCh: 1513 return errCancelContentProcessing 1514 case <-stateSync.done: 1515 if err := stateSync.Wait(); err != nil { 1516 return err 1517 } 1518 default: 1519 } 1520 // Retrieve the a batch of results to import 1521 first, last := results[0].Header, results[len(results)-1].Header 1522 log.Debug("Inserting fast-sync blocks", "items", len(results), 1523 "firstnum", first.Number, "firsthash", first.Hash(), 1524 "lastnumn", last.Number, "lasthash", last.Hash(), 1525 ) 1526 blocks := make([]*types.Block, len(results)) 1527 receipts := make([]types.Receipts, len(results)) 1528 for i, result := range results { 1529 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1530 receipts[i] = result.Receipts 1531 } 1532 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts); err != nil { 1533 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1534 return errInvalidChain 1535 } 1536 return nil 1537 } 1538 1539 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1540 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1541 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1542 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}); err != nil { 1543 return err 1544 } 1545 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1546 return err 1547 } 1548 atomic.StoreInt32(&d.committed, 1) 1549 return nil 1550 } 1551 1552 // DeliverHeaders injects a new batch of block headers received from a remote 1553 // node into the download schedule. 1554 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) { 1555 return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1556 } 1557 1558 // DeliverBodies injects a new batch of block bodies received from a remote node. 1559 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) { 1560 return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1561 } 1562 1563 // DeliverReceipts injects a new batch of receipts received from a remote node. 1564 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) { 1565 return d.deliver(id, d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1566 } 1567 1568 // DeliverNodeData injects a new batch of node state data received from a remote node. 1569 func (d *Downloader) DeliverNodeData(id string, data [][]byte) (err error) { 1570 return d.deliver(id, d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1571 } 1572 1573 // deliver injects a new batch of data received from a remote node. 1574 func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1575 // Update the delivery metrics for both good and failed deliveries 1576 inMeter.Mark(int64(packet.Items())) 1577 defer func() { 1578 if err != nil { 1579 dropMeter.Mark(int64(packet.Items())) 1580 } 1581 }() 1582 // Deliver or abort if the sync is canceled while queuing 1583 d.cancelLock.RLock() 1584 cancel := d.cancelCh 1585 d.cancelLock.RUnlock() 1586 if cancel == nil { 1587 return errNoSyncActive 1588 } 1589 select { 1590 case destCh <- packet: 1591 return nil 1592 case <-cancel: 1593 return errNoSyncActive 1594 } 1595 } 1596 1597 // qosTuner is the quality of service tuning loop that occasionally gathers the 1598 // peer latency statistics and updates the estimated request round trip time. 1599 func (d *Downloader) qosTuner() { 1600 for { 1601 // Retrieve the current median RTT and integrate into the previoust target RTT 1602 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 1603 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 1604 1605 // A new RTT cycle passed, increase our confidence in the estimated RTT 1606 conf := atomic.LoadUint64(&d.rttConfidence) 1607 conf = conf + (1000000-conf)/2 1608 atomic.StoreUint64(&d.rttConfidence, conf) 1609 1610 // Log the new QoS values and sleep until the next RTT 1611 log.Trace("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1612 select { 1613 case <-d.quitCh: 1614 return 1615 case <-time.After(rtt): 1616 } 1617 } 1618 } 1619 1620 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 1621 // peer set, needing to reduce the confidence we have in out QoS estimates. 1622 func (d *Downloader) qosReduceConfidence() { 1623 // If we have a single peer, confidence is always 1 1624 peers := uint64(d.peers.Len()) 1625 if peers == 0 { 1626 // Ensure peer connectivity races don't catch us off guard 1627 return 1628 } 1629 if peers == 1 { 1630 atomic.StoreUint64(&d.rttConfidence, 1000000) 1631 return 1632 } 1633 // If we have a ton of peers, don't drop confidence) 1634 if peers >= uint64(qosConfidenceCap) { 1635 return 1636 } 1637 // Otherwise drop the confidence factor 1638 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 1639 if float64(conf)/1000000 < rttMinConfidence { 1640 conf = uint64(rttMinConfidence * 1000000) 1641 } 1642 atomic.StoreUint64(&d.rttConfidence, conf) 1643 1644 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1645 log.Trace("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1646 } 1647 1648 // requestRTT returns the current target round trip time for a download request 1649 // to complete in. 1650 // 1651 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 1652 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 1653 // be adapted to, but smaller ones are preffered (stabler download stream). 1654 func (d *Downloader) requestRTT() time.Duration { 1655 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 1656 } 1657 1658 // requestTTL returns the current timeout allowance for a single download request 1659 // to finish under. 1660 func (d *Downloader) requestTTL() time.Duration { 1661 var ( 1662 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1663 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 1664 ) 1665 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 1666 if ttl > ttlLimit { 1667 ttl = ttlLimit 1668 } 1669 return ttl 1670 }