github.com/Openstars/go-ethereum@v1.9.7/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 "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 "github.com/ethereum/go-ethereum/trie" 38 ) 39 40 var ( 41 MaxHashFetch = 512 // Amount of hashes to be fetched per retrieval request 42 MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request 43 MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request 44 MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly 45 MaxBodyFetch = 128 // Amount of block bodies to be fetched per retrieval request 46 MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request 47 MaxStateFetch = 384 // Amount of node state values to allow fetching per request 48 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 maxForkAncestry uint64 = params.ImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 63 64 reorgProtThreshold = 48 // Threshold number of recent blocks to disable mini reorg protection 65 reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs 66 67 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 68 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 69 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 70 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 71 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 72 ) 73 74 var ( 75 errBusy = errors.New("busy") 76 errUnknownPeer = errors.New("peer is unknown or unhealthy") 77 errBadPeer = errors.New("action from bad peer ignored") 78 errStallingPeer = errors.New("peer is stalling") 79 errUnsyncedPeer = errors.New("unsynced peer") 80 errNoPeers = errors.New("no peers to keep download active") 81 errTimeout = errors.New("timeout") 82 errEmptyHeaderSet = errors.New("empty header set by peer") 83 errPeersUnavailable = errors.New("no peers available or all tried for download") 84 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 85 errInvalidChain = errors.New("retrieved hash chain is invalid") 86 errInvalidBody = errors.New("retrieved block body is invalid") 87 errInvalidReceipt = errors.New("retrieved receipt is invalid") 88 errCancelStateFetch = errors.New("state data download canceled (requested)") 89 errCancelContentProcessing = errors.New("content processing canceled (requested)") 90 errCanceled = errors.New("syncing 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 // WARNING: The `rttEstimate` and `rttConfidence` fields are accessed atomically. 97 // On 32 bit platforms, only 64-bit aligned fields can be atomic. The struct is 98 // guaranteed to be so aligned, so take advantage of that. For more information, 99 // see https://golang.org/pkg/sync/atomic/#pkg-note-BUG. 100 rttEstimate uint64 // Round trip time to target for download requests 101 rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops) 102 103 mode SyncMode // Synchronisation mode defining the strategy used (per sync cycle) 104 mux *event.TypeMux // Event multiplexer to announce sync operation events 105 106 checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync) 107 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 108 queue *queue // Scheduler for selecting the hashes to download 109 peers *peerSet // Set of active peers from which download can proceed 110 111 stateDB ethdb.Database // Database to state sync into (and deduplicate via) 112 stateBloom *trie.SyncBloom // Bloom filter for fast trie node existence checks 113 114 // Statistics 115 syncStatsChainOrigin uint64 // Origin block number where syncing started at 116 syncStatsChainHeight uint64 // Highest block number known when syncing started 117 syncStatsState stateSyncStats 118 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 119 120 lightchain LightChain 121 blockchain BlockChain 122 123 // Callbacks 124 dropPeer peerDropFn // Drops a peer for misbehaving 125 126 // Status 127 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 128 synchronising int32 129 notified int32 130 committed int32 131 ancientLimit uint64 // The maximum block number which can be regarded as ancient data. 132 133 // Channels 134 headerCh chan dataPack // [eth/62] Channel receiving inbound block headers 135 bodyCh chan dataPack // [eth/62] Channel receiving inbound block bodies 136 receiptCh chan dataPack // [eth/63] Channel receiving inbound receipts 137 bodyWakeCh chan bool // [eth/62] Channel to signal the block body fetcher of new tasks 138 receiptWakeCh chan bool // [eth/63] Channel to signal the receipt fetcher of new tasks 139 headerProcCh chan []*types.Header // [eth/62] Channel to feed the header processor new tasks 140 141 // for stateFetcher 142 stateSyncStart chan *stateSync 143 trackStateReq chan *stateReq 144 stateCh chan dataPack // [eth/63] Channel receiving inbound node state data 145 146 // Cancellation and termination 147 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 148 cancelCh chan struct{} // Channel to cancel mid-flight syncs 149 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 150 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 151 152 quitCh chan struct{} // Quit channel to signal termination 153 quitLock sync.RWMutex // Lock to prevent double closes 154 155 // Testing hooks 156 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 157 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 158 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 159 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 160 } 161 162 // LightChain encapsulates functions required to synchronise a light chain. 163 type LightChain interface { 164 // HasHeader verifies a header's presence in the local chain. 165 HasHeader(common.Hash, uint64) bool 166 167 // GetHeaderByHash retrieves a header from the local chain. 168 GetHeaderByHash(common.Hash) *types.Header 169 170 // CurrentHeader retrieves the head header from the local chain. 171 CurrentHeader() *types.Header 172 173 // GetTd returns the total difficulty of a local block. 174 GetTd(common.Hash, uint64) *big.Int 175 176 // InsertHeaderChain inserts a batch of headers into the local chain. 177 InsertHeaderChain([]*types.Header, int) (int, error) 178 179 // Rollback removes a few recently added elements from the local chain. 180 Rollback([]common.Hash) 181 } 182 183 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 184 type BlockChain interface { 185 LightChain 186 187 // HasBlock verifies a block's presence in the local chain. 188 HasBlock(common.Hash, uint64) bool 189 190 // HasFastBlock verifies a fast block's presence in the local chain. 191 HasFastBlock(common.Hash, uint64) bool 192 193 // GetBlockByHash retrieves a block from the local chain. 194 GetBlockByHash(common.Hash) *types.Block 195 196 // CurrentBlock retrieves the head block from the local chain. 197 CurrentBlock() *types.Block 198 199 // CurrentFastBlock retrieves the head fast block from the local chain. 200 CurrentFastBlock() *types.Block 201 202 // FastSyncCommitHead directly commits the head block to a certain entity. 203 FastSyncCommitHead(common.Hash) error 204 205 // InsertChain inserts a batch of blocks into the local chain. 206 InsertChain(types.Blocks) (int, error) 207 208 // InsertReceiptChain inserts a batch of receipts into the local chain. 209 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 210 } 211 212 // New creates a new downloader to fetch hashes and blocks from remote peers. 213 func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 214 if lightchain == nil { 215 lightchain = chain 216 } 217 dl := &Downloader{ 218 stateDB: stateDb, 219 stateBloom: stateBloom, 220 mux: mux, 221 checkpoint: checkpoint, 222 queue: newQueue(), 223 peers: newPeerSet(), 224 rttEstimate: uint64(rttMaxEstimate), 225 rttConfidence: uint64(1000000), 226 blockchain: chain, 227 lightchain: lightchain, 228 dropPeer: dropPeer, 229 headerCh: make(chan dataPack, 1), 230 bodyCh: make(chan dataPack, 1), 231 receiptCh: make(chan dataPack, 1), 232 bodyWakeCh: make(chan bool, 1), 233 receiptWakeCh: make(chan bool, 1), 234 headerProcCh: make(chan []*types.Header, 1), 235 quitCh: make(chan struct{}), 236 stateCh: make(chan dataPack), 237 stateSyncStart: make(chan *stateSync), 238 syncStatsState: stateSyncStats{ 239 processed: rawdb.ReadFastTrieProgress(stateDb), 240 }, 241 trackStateReq: make(chan *stateReq), 242 } 243 go dl.qosTuner() 244 go dl.stateFetcher() 245 return dl 246 } 247 248 // Progress retrieves the synchronisation boundaries, specifically the origin 249 // block where synchronisation started at (may have failed/suspended); the block 250 // or header sync is currently at; and the latest known block which the sync targets. 251 // 252 // In addition, during the state download phase of fast synchronisation the number 253 // of processed and the total number of known states are also returned. Otherwise 254 // these are zero. 255 func (d *Downloader) Progress() ethereum.SyncProgress { 256 // Lock the current stats and return the progress 257 d.syncStatsLock.RLock() 258 defer d.syncStatsLock.RUnlock() 259 260 current := uint64(0) 261 switch { 262 case d.blockchain != nil && d.mode == FullSync: 263 current = d.blockchain.CurrentBlock().NumberU64() 264 case d.blockchain != nil && d.mode == FastSync: 265 current = d.blockchain.CurrentFastBlock().NumberU64() 266 case d.lightchain != nil: 267 current = d.lightchain.CurrentHeader().Number.Uint64() 268 default: 269 log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", d.mode) 270 } 271 return ethereum.SyncProgress{ 272 StartingBlock: d.syncStatsChainOrigin, 273 CurrentBlock: current, 274 HighestBlock: d.syncStatsChainHeight, 275 PulledStates: d.syncStatsState.processed, 276 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 277 } 278 } 279 280 // Synchronising returns whether the downloader is currently retrieving blocks. 281 func (d *Downloader) Synchronising() bool { 282 return atomic.LoadInt32(&d.synchronising) > 0 283 } 284 285 // RegisterPeer injects a new download peer into the set of block source to be 286 // used for fetching hashes and blocks from. 287 func (d *Downloader) RegisterPeer(id string, version int, peer Peer) error { 288 logger := log.New("peer", id) 289 logger.Trace("Registering sync peer") 290 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 291 logger.Error("Failed to register sync peer", "err", err) 292 return err 293 } 294 d.qosReduceConfidence() 295 296 return nil 297 } 298 299 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 300 func (d *Downloader) RegisterLightPeer(id string, version int, peer LightPeer) error { 301 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 302 } 303 304 // UnregisterPeer remove a peer from the known list, preventing any action from 305 // the specified peer. An effort is also made to return any pending fetches into 306 // the queue. 307 func (d *Downloader) UnregisterPeer(id string) error { 308 // Unregister the peer from the active peer set and revoke any fetch tasks 309 logger := log.New("peer", id) 310 logger.Trace("Unregistering sync peer") 311 if err := d.peers.Unregister(id); err != nil { 312 logger.Error("Failed to unregister sync peer", "err", err) 313 return err 314 } 315 d.queue.Revoke(id) 316 317 return nil 318 } 319 320 // Synchronise tries to sync up our local block chain with a remote peer, both 321 // adding various sanity checks as well as wrapping it with various log entries. 322 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 323 err := d.synchronise(id, head, td, mode) 324 switch err { 325 case nil: 326 case errBusy, errCanceled: 327 328 case errTimeout, errBadPeer, errStallingPeer, errUnsyncedPeer, 329 errEmptyHeaderSet, errPeersUnavailable, errTooOld, 330 errInvalidAncestor, errInvalidChain: 331 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 332 if d.dropPeer == nil { 333 // The dropPeer method is nil when `--copydb` is used for a local copy. 334 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 335 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 336 } else { 337 d.dropPeer(id) 338 } 339 default: 340 log.Warn("Synchronisation failed, retrying", "err", err) 341 } 342 return err 343 } 344 345 // synchronise will select the peer and use it for synchronising. If an empty string is given 346 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 347 // checks fail an error will be returned. This method is synchronous 348 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 349 // Mock out the synchronisation if testing 350 if d.synchroniseMock != nil { 351 return d.synchroniseMock(id, hash) 352 } 353 // Make sure only one goroutine is ever allowed past this point at once 354 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 355 return errBusy 356 } 357 defer atomic.StoreInt32(&d.synchronising, 0) 358 359 // Post a user notification of the sync (only once per session) 360 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 361 log.Info("Block synchronisation started") 362 } 363 // If we are already full syncing, but have a fast-sync bloom filter laying 364 // around, make sure it does't use memory any more. This is a special case 365 // when the user attempts to fast sync a new empty network. 366 if mode == FullSync && d.stateBloom != nil { 367 d.stateBloom.Close() 368 } 369 // Reset the queue, peer set and wake channels to clean any internal leftover state 370 d.queue.Reset() 371 d.peers.Reset() 372 373 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 374 select { 375 case <-ch: 376 default: 377 } 378 } 379 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 380 for empty := false; !empty; { 381 select { 382 case <-ch: 383 default: 384 empty = true 385 } 386 } 387 } 388 for empty := false; !empty; { 389 select { 390 case <-d.headerProcCh: 391 default: 392 empty = true 393 } 394 } 395 // Create cancel channel for aborting mid-flight and mark the master peer 396 d.cancelLock.Lock() 397 d.cancelCh = make(chan struct{}) 398 d.cancelPeer = id 399 d.cancelLock.Unlock() 400 401 defer d.Cancel() // No matter what, we can't leave the cancel channel open 402 403 // Set the requested sync mode, unless it's forbidden 404 d.mode = mode 405 406 // Retrieve the origin peer and initiate the downloading process 407 p := d.peers.Peer(id) 408 if p == nil { 409 return errUnknownPeer 410 } 411 return d.syncWithPeer(p, hash, td) 412 } 413 414 // syncWithPeer starts a block synchronization based on the hash chain from the 415 // specified peer and head hash. 416 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 417 d.mux.Post(StartEvent{}) 418 defer func() { 419 // reset on error 420 if err != nil { 421 d.mux.Post(FailedEvent{err}) 422 } else { 423 latest := d.lightchain.CurrentHeader() 424 d.mux.Post(DoneEvent{latest}) 425 } 426 }() 427 if p.version < 62 { 428 return errTooOld 429 } 430 431 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", d.mode) 432 defer func(start time.Time) { 433 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 434 }(time.Now()) 435 436 // Look up the sync boundaries: the common ancestor and the target block 437 latest, err := d.fetchHeight(p) 438 if err != nil { 439 return err 440 } 441 height := latest.Number.Uint64() 442 443 origin, err := d.findAncestor(p, latest) 444 if err != nil { 445 return err 446 } 447 d.syncStatsLock.Lock() 448 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 449 d.syncStatsChainOrigin = origin 450 } 451 d.syncStatsChainHeight = height 452 d.syncStatsLock.Unlock() 453 454 // Ensure our origin point is below any fast sync pivot point 455 pivot := uint64(0) 456 if d.mode == FastSync { 457 if height <= uint64(fsMinFullBlocks) { 458 origin = 0 459 } else { 460 pivot = height - uint64(fsMinFullBlocks) 461 if pivot <= origin { 462 origin = pivot - 1 463 } 464 } 465 } 466 d.committed = 1 467 if d.mode == FastSync && pivot != 0 { 468 d.committed = 0 469 } 470 if d.mode == FastSync { 471 // Set the ancient data limitation. 472 // If we are running fast sync, all block data older than ancientLimit will be 473 // written to the ancient store. More recent data will be written to the active 474 // database and will wait for the freezer to migrate. 475 // 476 // If there is a checkpoint available, then calculate the ancientLimit through 477 // that. Otherwise calculate the ancient limit through the advertised height 478 // of the remote peer. 479 // 480 // The reason for picking checkpoint first is that a malicious peer can give us 481 // a fake (very high) height, forcing the ancient limit to also be very high. 482 // The peer would start to feed us valid blocks until head, resulting in all of 483 // the blocks might be written into the ancient store. A following mini-reorg 484 // could cause issues. 485 if d.checkpoint != 0 && d.checkpoint > maxForkAncestry+1 { 486 d.ancientLimit = d.checkpoint 487 } else if height > maxForkAncestry+1 { 488 d.ancientLimit = height - maxForkAncestry - 1 489 } 490 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 491 // If a part of blockchain data has already been written into active store, 492 // disable the ancient style insertion explicitly. 493 if origin >= frozen && frozen != 0 { 494 d.ancientLimit = 0 495 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 496 } else if d.ancientLimit > 0 { 497 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 498 } 499 // Rewind the ancient store and blockchain if reorg happens. 500 if origin+1 < frozen { 501 var hashes []common.Hash 502 for i := origin + 1; i < d.lightchain.CurrentHeader().Number.Uint64(); i++ { 503 hashes = append(hashes, rawdb.ReadCanonicalHash(d.stateDB, i)) 504 } 505 d.lightchain.Rollback(hashes) 506 } 507 } 508 // Initiate the sync using a concurrent header and content retrieval algorithm 509 d.queue.Prepare(origin+1, d.mode) 510 if d.syncInitHook != nil { 511 d.syncInitHook(origin, height) 512 } 513 fetchers := []func() error{ 514 func() error { return d.fetchHeaders(p, origin+1, pivot) }, // Headers are always retrieved 515 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 516 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 517 func() error { return d.processHeaders(origin+1, pivot, td) }, 518 } 519 if d.mode == FastSync { 520 fetchers = append(fetchers, func() error { return d.processFastSyncContent(latest) }) 521 } else if d.mode == FullSync { 522 fetchers = append(fetchers, d.processFullSyncContent) 523 } 524 return d.spawnSync(fetchers) 525 } 526 527 // spawnSync runs d.process and all given fetcher functions to completion in 528 // separate goroutines, returning the first error that appears. 529 func (d *Downloader) spawnSync(fetchers []func() error) error { 530 errc := make(chan error, len(fetchers)) 531 d.cancelWg.Add(len(fetchers)) 532 for _, fn := range fetchers { 533 fn := fn 534 go func() { defer d.cancelWg.Done(); errc <- fn() }() 535 } 536 // Wait for the first error, then terminate the others. 537 var err error 538 for i := 0; i < len(fetchers); i++ { 539 if i == len(fetchers)-1 { 540 // Close the queue when all fetchers have exited. 541 // This will cause the block processor to end when 542 // it has processed the queue. 543 d.queue.Close() 544 } 545 if err = <-errc; err != nil && err != errCanceled { 546 break 547 } 548 } 549 d.queue.Close() 550 d.Cancel() 551 return err 552 } 553 554 // cancel aborts all of the operations and resets the queue. However, cancel does 555 // not wait for the running download goroutines to finish. This method should be 556 // used when cancelling the downloads from inside the downloader. 557 func (d *Downloader) cancel() { 558 // Close the current cancel channel 559 d.cancelLock.Lock() 560 if d.cancelCh != nil { 561 select { 562 case <-d.cancelCh: 563 // Channel was already closed 564 default: 565 close(d.cancelCh) 566 } 567 } 568 d.cancelLock.Unlock() 569 } 570 571 // Cancel aborts all of the operations and waits for all download goroutines to 572 // finish before returning. 573 func (d *Downloader) Cancel() { 574 d.cancel() 575 d.cancelWg.Wait() 576 577 d.ancientLimit = 0 578 log.Debug("Reset ancient limit to zero") 579 } 580 581 // Terminate interrupts the downloader, canceling all pending operations. 582 // The downloader cannot be reused after calling Terminate. 583 func (d *Downloader) Terminate() { 584 // Close the termination channel (make sure double close is allowed) 585 d.quitLock.Lock() 586 select { 587 case <-d.quitCh: 588 default: 589 close(d.quitCh) 590 } 591 d.quitLock.Unlock() 592 593 // Cancel any pending download requests 594 d.Cancel() 595 } 596 597 // fetchHeight retrieves the head header of the remote peer to aid in estimating 598 // the total time a pending synchronisation would take. 599 func (d *Downloader) fetchHeight(p *peerConnection) (*types.Header, error) { 600 p.log.Debug("Retrieving remote chain height") 601 602 // Request the advertised remote head block and wait for the response 603 head, _ := p.peer.Head() 604 go p.peer.RequestHeadersByHash(head, 1, 0, false) 605 606 ttl := d.requestTTL() 607 timeout := time.After(ttl) 608 for { 609 select { 610 case <-d.cancelCh: 611 return nil, errCanceled 612 613 case packet := <-d.headerCh: 614 // Discard anything not from the origin peer 615 if packet.PeerId() != p.id { 616 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 617 break 618 } 619 // Make sure the peer actually gave something valid 620 headers := packet.(*headerPack).headers 621 if len(headers) != 1 { 622 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 623 return nil, errBadPeer 624 } 625 head := headers[0] 626 if (d.mode == FastSync || d.mode == LightSync) && head.Number.Uint64() < d.checkpoint { 627 p.log.Warn("Remote head below checkpoint", "number", head.Number, "hash", head.Hash()) 628 return nil, errUnsyncedPeer 629 } 630 p.log.Debug("Remote head header identified", "number", head.Number, "hash", head.Hash()) 631 return head, nil 632 633 case <-timeout: 634 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 635 return nil, errTimeout 636 637 case <-d.bodyCh: 638 case <-d.receiptCh: 639 // Out of bounds delivery, ignore 640 } 641 } 642 } 643 644 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 645 // common ancestor. 646 // It returns parameters to be used for peer.RequestHeadersByNumber: 647 // from - starting block number 648 // count - number of headers to request 649 // skip - number of headers to skip 650 // and also returns 'max', the last block which is expected to be returned by the remote peers, 651 // given the (from,count,skip) 652 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 653 var ( 654 from int 655 count int 656 MaxCount = MaxHeaderFetch / 16 657 ) 658 // requestHead is the highest block that we will ask for. If requestHead is not offset, 659 // the highest block that we will get is 16 blocks back from head, which means we 660 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 661 // between us and the peer is 1-2 blocks, which is most common 662 requestHead := int(remoteHeight) - 1 663 if requestHead < 0 { 664 requestHead = 0 665 } 666 // requestBottom is the lowest block we want included in the query 667 // Ideally, we want to include just below own head 668 requestBottom := int(localHeight - 1) 669 if requestBottom < 0 { 670 requestBottom = 0 671 } 672 totalSpan := requestHead - requestBottom 673 span := 1 + totalSpan/MaxCount 674 if span < 2 { 675 span = 2 676 } 677 if span > 16 { 678 span = 16 679 } 680 681 count = 1 + totalSpan/span 682 if count > MaxCount { 683 count = MaxCount 684 } 685 if count < 2 { 686 count = 2 687 } 688 from = requestHead - (count-1)*span 689 if from < 0 { 690 from = 0 691 } 692 max := from + (count-1)*span 693 return int64(from), count, span - 1, uint64(max) 694 } 695 696 // findAncestor tries to locate the common ancestor link of the local chain and 697 // a remote peers blockchain. In the general case when our node was in sync and 698 // on the correct chain, checking the top N links should already get us a match. 699 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 700 // the head links match), we do a binary search to find the common ancestor. 701 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 702 // Figure out the valid ancestor range to prevent rewrite attacks 703 var ( 704 floor = int64(-1) 705 localHeight uint64 706 remoteHeight = remoteHeader.Number.Uint64() 707 ) 708 switch d.mode { 709 case FullSync: 710 localHeight = d.blockchain.CurrentBlock().NumberU64() 711 case FastSync: 712 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 713 default: 714 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 715 } 716 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 717 718 // Recap floor value for binary search 719 if localHeight >= maxForkAncestry { 720 // We're above the max reorg threshold, find the earliest fork point 721 floor = int64(localHeight - maxForkAncestry) 722 } 723 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 724 // all headers before that point will be missing. 725 if d.mode == LightSync { 726 // If we dont know the current CHT position, find it 727 if d.genesis == 0 { 728 header := d.lightchain.CurrentHeader() 729 for header != nil { 730 d.genesis = header.Number.Uint64() 731 if floor >= int64(d.genesis)-1 { 732 break 733 } 734 header = d.lightchain.GetHeaderByHash(header.ParentHash) 735 } 736 } 737 // We already know the "genesis" block number, cap floor to that 738 if floor < int64(d.genesis)-1 { 739 floor = int64(d.genesis) - 1 740 } 741 } 742 743 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 744 745 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 746 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 747 748 // Wait for the remote response to the head fetch 749 number, hash := uint64(0), common.Hash{} 750 751 ttl := d.requestTTL() 752 timeout := time.After(ttl) 753 754 for finished := false; !finished; { 755 select { 756 case <-d.cancelCh: 757 return 0, errCanceled 758 759 case packet := <-d.headerCh: 760 // Discard anything not from the origin peer 761 if packet.PeerId() != p.id { 762 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 763 break 764 } 765 // Make sure the peer actually gave something valid 766 headers := packet.(*headerPack).headers 767 if len(headers) == 0 { 768 p.log.Warn("Empty head header set") 769 return 0, errEmptyHeaderSet 770 } 771 // Make sure the peer's reply conforms to the request 772 for i, header := range headers { 773 expectNumber := from + int64(i)*int64(skip+1) 774 if number := header.Number.Int64(); number != expectNumber { 775 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 776 return 0, errInvalidChain 777 } 778 } 779 // Check if a common ancestor was found 780 finished = true 781 for i := len(headers) - 1; i >= 0; i-- { 782 // Skip any headers that underflow/overflow our requested set 783 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 784 continue 785 } 786 // Otherwise check if we already know the header or not 787 h := headers[i].Hash() 788 n := headers[i].Number.Uint64() 789 790 var known bool 791 switch d.mode { 792 case FullSync: 793 known = d.blockchain.HasBlock(h, n) 794 case FastSync: 795 known = d.blockchain.HasFastBlock(h, n) 796 default: 797 known = d.lightchain.HasHeader(h, n) 798 } 799 if known { 800 number, hash = n, h 801 break 802 } 803 } 804 805 case <-timeout: 806 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 807 return 0, errTimeout 808 809 case <-d.bodyCh: 810 case <-d.receiptCh: 811 // Out of bounds delivery, ignore 812 } 813 } 814 // If the head fetch already found an ancestor, return 815 if hash != (common.Hash{}) { 816 if int64(number) <= floor { 817 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 818 return 0, errInvalidAncestor 819 } 820 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 821 return number, nil 822 } 823 // Ancestor not found, we need to binary search over our chain 824 start, end := uint64(0), remoteHeight 825 if floor > 0 { 826 start = uint64(floor) 827 } 828 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 829 830 for start+1 < end { 831 // Split our chain interval in two, and request the hash to cross check 832 check := (start + end) / 2 833 834 ttl := d.requestTTL() 835 timeout := time.After(ttl) 836 837 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 838 839 // Wait until a reply arrives to this request 840 for arrived := false; !arrived; { 841 select { 842 case <-d.cancelCh: 843 return 0, errCanceled 844 845 case packer := <-d.headerCh: 846 // Discard anything not from the origin peer 847 if packer.PeerId() != p.id { 848 log.Debug("Received headers from incorrect peer", "peer", packer.PeerId()) 849 break 850 } 851 // Make sure the peer actually gave something valid 852 headers := packer.(*headerPack).headers 853 if len(headers) != 1 { 854 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 855 return 0, errBadPeer 856 } 857 arrived = true 858 859 // Modify the search interval based on the response 860 h := headers[0].Hash() 861 n := headers[0].Number.Uint64() 862 863 var known bool 864 switch d.mode { 865 case FullSync: 866 known = d.blockchain.HasBlock(h, n) 867 case FastSync: 868 known = d.blockchain.HasFastBlock(h, n) 869 default: 870 known = d.lightchain.HasHeader(h, n) 871 } 872 if !known { 873 end = check 874 break 875 } 876 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 877 if header.Number.Uint64() != check { 878 p.log.Debug("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 879 return 0, errBadPeer 880 } 881 start = check 882 hash = h 883 884 case <-timeout: 885 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 886 return 0, errTimeout 887 888 case <-d.bodyCh: 889 case <-d.receiptCh: 890 // Out of bounds delivery, ignore 891 } 892 } 893 } 894 // Ensure valid ancestry and return 895 if int64(start) <= floor { 896 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 897 return 0, errInvalidAncestor 898 } 899 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 900 return start, nil 901 } 902 903 // fetchHeaders keeps retrieving headers concurrently from the number 904 // requested, until no more are returned, potentially throttling on the way. To 905 // facilitate concurrency but still protect against malicious nodes sending bad 906 // headers, we construct a header chain skeleton using the "origin" peer we are 907 // syncing with, and fill in the missing headers using anyone else. Headers from 908 // other peers are only accepted if they map cleanly to the skeleton. If no one 909 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 910 // the origin is dropped. 911 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64, pivot uint64) error { 912 p.log.Debug("Directing header downloads", "origin", from) 913 defer p.log.Debug("Header download terminated") 914 915 // Create a timeout timer, and the associated header fetcher 916 skeleton := true // Skeleton assembly phase or finishing up 917 request := time.Now() // time of the last skeleton fetch request 918 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 919 <-timeout.C // timeout channel should be initially empty 920 defer timeout.Stop() 921 922 var ttl time.Duration 923 getHeaders := func(from uint64) { 924 request = time.Now() 925 926 ttl = d.requestTTL() 927 timeout.Reset(ttl) 928 929 if skeleton { 930 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 931 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 932 } else { 933 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 934 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 935 } 936 } 937 // Start pulling the header chain skeleton until all is done 938 ancestor := from 939 getHeaders(from) 940 941 for { 942 select { 943 case <-d.cancelCh: 944 return errCanceled 945 946 case packet := <-d.headerCh: 947 // Make sure the active peer is giving us the skeleton headers 948 if packet.PeerId() != p.id { 949 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 950 break 951 } 952 headerReqTimer.UpdateSince(request) 953 timeout.Stop() 954 955 // If the skeleton's finished, pull any remaining head headers directly from the origin 956 if packet.Items() == 0 && skeleton { 957 skeleton = false 958 getHeaders(from) 959 continue 960 } 961 // If no more headers are inbound, notify the content fetchers and return 962 if packet.Items() == 0 { 963 // Don't abort header fetches while the pivot is downloading 964 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 965 p.log.Debug("No headers, waiting for pivot commit") 966 select { 967 case <-time.After(fsHeaderContCheck): 968 getHeaders(from) 969 continue 970 case <-d.cancelCh: 971 return errCanceled 972 } 973 } 974 // Pivot done (or not in fast sync) and no more headers, terminate the process 975 p.log.Debug("No more headers available") 976 select { 977 case d.headerProcCh <- nil: 978 return nil 979 case <-d.cancelCh: 980 return errCanceled 981 } 982 } 983 headers := packet.(*headerPack).headers 984 985 // If we received a skeleton batch, resolve internals concurrently 986 if skeleton { 987 filled, proced, err := d.fillHeaderSkeleton(from, headers) 988 if err != nil { 989 p.log.Debug("Skeleton chain invalid", "err", err) 990 return errInvalidChain 991 } 992 headers = filled[proced:] 993 from += uint64(proced) 994 } else { 995 // If we're closing in on the chain head, but haven't yet reached it, delay 996 // the last few headers so mini reorgs on the head don't cause invalid hash 997 // chain errors. 998 if n := len(headers); n > 0 { 999 // Retrieve the current head we're at 1000 var head uint64 1001 if d.mode == LightSync { 1002 head = d.lightchain.CurrentHeader().Number.Uint64() 1003 } else { 1004 head = d.blockchain.CurrentFastBlock().NumberU64() 1005 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1006 head = full 1007 } 1008 } 1009 // If the head is below the common ancestor, we're actually deduplicating 1010 // already existing chain segments, so use the ancestor as the fake head. 1011 // Otherwise we might end up delaying header deliveries pointlessly. 1012 if head < ancestor { 1013 head = ancestor 1014 } 1015 // If the head is way older than this batch, delay the last few headers 1016 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1017 delay := reorgProtHeaderDelay 1018 if delay > n { 1019 delay = n 1020 } 1021 headers = headers[:n-delay] 1022 } 1023 } 1024 } 1025 // Insert all the new headers and fetch the next batch 1026 if len(headers) > 0 { 1027 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1028 select { 1029 case d.headerProcCh <- headers: 1030 case <-d.cancelCh: 1031 return errCanceled 1032 } 1033 from += uint64(len(headers)) 1034 getHeaders(from) 1035 } else { 1036 // No headers delivered, or all of them being delayed, sleep a bit and retry 1037 p.log.Trace("All headers delayed, waiting") 1038 select { 1039 case <-time.After(fsHeaderContCheck): 1040 getHeaders(from) 1041 continue 1042 case <-d.cancelCh: 1043 return errCanceled 1044 } 1045 } 1046 1047 case <-timeout.C: 1048 if d.dropPeer == nil { 1049 // The dropPeer method is nil when `--copydb` is used for a local copy. 1050 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1051 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1052 break 1053 } 1054 // Header retrieval timed out, consider the peer bad and drop 1055 p.log.Debug("Header request timed out", "elapsed", ttl) 1056 headerTimeoutMeter.Mark(1) 1057 d.dropPeer(p.id) 1058 1059 // Finish the sync gracefully instead of dumping the gathered data though 1060 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1061 select { 1062 case ch <- false: 1063 case <-d.cancelCh: 1064 } 1065 } 1066 select { 1067 case d.headerProcCh <- nil: 1068 case <-d.cancelCh: 1069 } 1070 return errBadPeer 1071 } 1072 } 1073 } 1074 1075 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1076 // and maps them to the provided skeleton header chain. 1077 // 1078 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1079 // immediately to the header processor to keep the rest of the pipeline full even 1080 // in the case of header stalls. 1081 // 1082 // The method returns the entire filled skeleton and also the number of headers 1083 // already forwarded for processing. 1084 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1085 log.Debug("Filling up skeleton", "from", from) 1086 d.queue.ScheduleSkeleton(from, skeleton) 1087 1088 var ( 1089 deliver = func(packet dataPack) (int, error) { 1090 pack := packet.(*headerPack) 1091 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1092 } 1093 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 1094 throttle = func() bool { return false } 1095 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, error) { 1096 return d.queue.ReserveHeaders(p, count), false, nil 1097 } 1098 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1099 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 1100 setIdle = func(p *peerConnection, accepted int) { p.SetHeadersIdle(accepted) } 1101 ) 1102 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1103 d.queue.PendingHeaders, d.queue.InFlightHeaders, throttle, reserve, 1104 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1105 1106 log.Debug("Skeleton fill terminated", "err", err) 1107 1108 filled, proced := d.queue.RetrieveHeaders() 1109 return filled, proced, err 1110 } 1111 1112 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1113 // available peers, reserving a chunk of blocks for each, waiting for delivery 1114 // and also periodically checking for timeouts. 1115 func (d *Downloader) fetchBodies(from uint64) error { 1116 log.Debug("Downloading block bodies", "origin", from) 1117 1118 var ( 1119 deliver = func(packet dataPack) (int, error) { 1120 pack := packet.(*bodyPack) 1121 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1122 } 1123 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 1124 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1125 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 1126 setIdle = func(p *peerConnection, accepted int) { p.SetBodiesIdle(accepted) } 1127 ) 1128 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1129 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ShouldThrottleBlocks, d.queue.ReserveBodies, 1130 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1131 1132 log.Debug("Block body download terminated", "err", err) 1133 return err 1134 } 1135 1136 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1137 // available peers, reserving a chunk of receipts for each, waiting for delivery 1138 // and also periodically checking for timeouts. 1139 func (d *Downloader) fetchReceipts(from uint64) error { 1140 log.Debug("Downloading transaction receipts", "origin", from) 1141 1142 var ( 1143 deliver = func(packet dataPack) (int, error) { 1144 pack := packet.(*receiptPack) 1145 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1146 } 1147 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1148 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1149 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1150 setIdle = func(p *peerConnection, accepted int) { p.SetReceiptsIdle(accepted) } 1151 ) 1152 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1153 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ShouldThrottleReceipts, d.queue.ReserveReceipts, 1154 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1155 1156 log.Debug("Transaction receipt download terminated", "err", err) 1157 return err 1158 } 1159 1160 // fetchParts iteratively downloads scheduled block parts, taking any available 1161 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1162 // also periodically checking for timeouts. 1163 // 1164 // As the scheduling/timeout logic mostly is the same for all downloaded data 1165 // types, this method is used by each for data gathering and is instrumented with 1166 // various callbacks to handle the slight differences between processing them. 1167 // 1168 // The instrumentation parameters: 1169 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1170 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1171 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1172 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1173 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1174 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1175 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1176 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1177 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1178 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1179 // - fetch: network callback to actually send a particular download request to a physical remote peer 1180 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1181 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1182 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1183 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1184 // - kind: textual label of the type being downloaded to display in log mesages 1185 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1186 expire func() map[string]int, pending func() int, inFlight func() bool, throttle func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, error), 1187 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1188 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int), kind string) error { 1189 1190 // Create a ticker to detect expired retrieval tasks 1191 ticker := time.NewTicker(100 * time.Millisecond) 1192 defer ticker.Stop() 1193 1194 update := make(chan struct{}, 1) 1195 1196 // Prepare the queue and fetch block parts until the block header fetcher's done 1197 finished := false 1198 for { 1199 select { 1200 case <-d.cancelCh: 1201 return errCanceled 1202 1203 case packet := <-deliveryCh: 1204 // If the peer was previously banned and failed to deliver its pack 1205 // in a reasonable time frame, ignore its message. 1206 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1207 // Deliver the received chunk of data and check chain validity 1208 accepted, err := deliver(packet) 1209 if err == errInvalidChain { 1210 return err 1211 } 1212 // Unless a peer delivered something completely else than requested (usually 1213 // caused by a timed out request which came through in the end), set it to 1214 // idle. If the delivery's stale, the peer should have already been idled. 1215 if err != errStaleDelivery { 1216 setIdle(peer, accepted) 1217 } 1218 // Issue a log to the user to see what's going on 1219 switch { 1220 case err == nil && packet.Items() == 0: 1221 peer.log.Trace("Requested data not delivered", "type", kind) 1222 case err == nil: 1223 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1224 default: 1225 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1226 } 1227 } 1228 // Blocks assembled, try to update the progress 1229 select { 1230 case update <- struct{}{}: 1231 default: 1232 } 1233 1234 case cont := <-wakeCh: 1235 // The header fetcher sent a continuation flag, check if it's done 1236 if !cont { 1237 finished = true 1238 } 1239 // Headers arrive, try to update the progress 1240 select { 1241 case update <- struct{}{}: 1242 default: 1243 } 1244 1245 case <-ticker.C: 1246 // Sanity check update the progress 1247 select { 1248 case update <- struct{}{}: 1249 default: 1250 } 1251 1252 case <-update: 1253 // Short circuit if we lost all our peers 1254 if d.peers.Len() == 0 { 1255 return errNoPeers 1256 } 1257 // Check for fetch request timeouts and demote the responsible peers 1258 for pid, fails := range expire() { 1259 if peer := d.peers.Peer(pid); peer != nil { 1260 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1261 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1262 // out that sync wise we need to get rid of the peer. 1263 // 1264 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1265 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1266 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1267 if fails > 2 { 1268 peer.log.Trace("Data delivery timed out", "type", kind) 1269 setIdle(peer, 0) 1270 } else { 1271 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1272 1273 if d.dropPeer == nil { 1274 // The dropPeer method is nil when `--copydb` is used for a local copy. 1275 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1276 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1277 } else { 1278 d.dropPeer(pid) 1279 1280 // If this peer was the master peer, abort sync immediately 1281 d.cancelLock.RLock() 1282 master := pid == d.cancelPeer 1283 d.cancelLock.RUnlock() 1284 1285 if master { 1286 d.cancel() 1287 return errTimeout 1288 } 1289 } 1290 } 1291 } 1292 } 1293 // If there's nothing more to fetch, wait or terminate 1294 if pending() == 0 { 1295 if !inFlight() && finished { 1296 log.Debug("Data fetching completed", "type", kind) 1297 return nil 1298 } 1299 break 1300 } 1301 // Send a download request to all idle peers, until throttled 1302 progressed, throttled, running := false, false, inFlight() 1303 idles, total := idle() 1304 1305 for _, peer := range idles { 1306 // Short circuit if throttling activated 1307 if throttle() { 1308 throttled = true 1309 break 1310 } 1311 // Short circuit if there is no more available task. 1312 if pending() == 0 { 1313 break 1314 } 1315 // Reserve a chunk of fetches for a peer. A nil can mean either that 1316 // no more headers are available, or that the peer is known not to 1317 // have them. 1318 request, progress, err := reserve(peer, capacity(peer)) 1319 if err != nil { 1320 return err 1321 } 1322 if progress { 1323 progressed = true 1324 } 1325 if request == nil { 1326 continue 1327 } 1328 if request.From > 0 { 1329 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1330 } else { 1331 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1332 } 1333 // Fetch the chunk and make sure any errors return the hashes to the queue 1334 if fetchHook != nil { 1335 fetchHook(request.Headers) 1336 } 1337 if err := fetch(peer, request); err != nil { 1338 // Although we could try and make an attempt to fix this, this error really 1339 // means that we've double allocated a fetch task to a peer. If that is the 1340 // case, the internal state of the downloader and the queue is very wrong so 1341 // better hard crash and note the error instead of silently accumulating into 1342 // a much bigger issue. 1343 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1344 } 1345 running = true 1346 } 1347 // Make sure that we have peers available for fetching. If all peers have been tried 1348 // and all failed throw an error 1349 if !progressed && !throttled && !running && len(idles) == total && pending() > 0 { 1350 return errPeersUnavailable 1351 } 1352 } 1353 } 1354 } 1355 1356 // processHeaders takes batches of retrieved headers from an input channel and 1357 // keeps processing and scheduling them into the header chain and downloader's 1358 // queue until the stream ends or a failure occurs. 1359 func (d *Downloader) processHeaders(origin uint64, pivot uint64, td *big.Int) error { 1360 // Keep a count of uncertain headers to roll back 1361 var rollback []*types.Header 1362 defer func() { 1363 if len(rollback) > 0 { 1364 // Flatten the headers and roll them back 1365 hashes := make([]common.Hash, len(rollback)) 1366 for i, header := range rollback { 1367 hashes[i] = header.Hash() 1368 } 1369 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1370 if d.mode != LightSync { 1371 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1372 lastBlock = d.blockchain.CurrentBlock().Number() 1373 } 1374 d.lightchain.Rollback(hashes) 1375 curFastBlock, curBlock := common.Big0, common.Big0 1376 if d.mode != LightSync { 1377 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1378 curBlock = d.blockchain.CurrentBlock().Number() 1379 } 1380 log.Warn("Rolled back headers", "count", len(hashes), 1381 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1382 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1383 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock)) 1384 } 1385 }() 1386 1387 // Wait for batches of headers to process 1388 gotHeaders := false 1389 1390 for { 1391 select { 1392 case <-d.cancelCh: 1393 return errCanceled 1394 1395 case headers := <-d.headerProcCh: 1396 // Terminate header processing if we synced up 1397 if len(headers) == 0 { 1398 // Notify everyone that headers are fully processed 1399 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1400 select { 1401 case ch <- false: 1402 case <-d.cancelCh: 1403 } 1404 } 1405 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1406 // better chain compared to ours. The only exception is if its promised blocks were 1407 // already imported by other means (e.g. fetcher): 1408 // 1409 // R <remote peer>, L <local node>: Both at block 10 1410 // R: Mine block 11, and propagate it to L 1411 // L: Queue block 11 for import 1412 // L: Notice that R's head and TD increased compared to ours, start sync 1413 // L: Import of block 11 finishes 1414 // L: Sync begins, and finds common ancestor at 11 1415 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1416 // R: Nothing to give 1417 if d.mode != LightSync { 1418 head := d.blockchain.CurrentBlock() 1419 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1420 return errStallingPeer 1421 } 1422 } 1423 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1424 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1425 // of delivering the post-pivot blocks that would flag the invalid content. 1426 // 1427 // This check cannot be executed "as is" for full imports, since blocks may still be 1428 // queued for processing when the header download completes. However, as long as the 1429 // peer gave us something useful, we're already happy/progressed (above check). 1430 if d.mode == FastSync || d.mode == LightSync { 1431 head := d.lightchain.CurrentHeader() 1432 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1433 return errStallingPeer 1434 } 1435 } 1436 // Disable any rollback and return 1437 rollback = nil 1438 return nil 1439 } 1440 // Otherwise split the chunk of headers into batches and process them 1441 gotHeaders = true 1442 for len(headers) > 0 { 1443 // Terminate if something failed in between processing chunks 1444 select { 1445 case <-d.cancelCh: 1446 return errCanceled 1447 default: 1448 } 1449 // Select the next chunk of headers to import 1450 limit := maxHeadersProcess 1451 if limit > len(headers) { 1452 limit = len(headers) 1453 } 1454 chunk := headers[:limit] 1455 // In case of header only syncing, validate the chunk immediately 1456 if d.mode == FastSync || d.mode == LightSync { 1457 // Collect the yet unknown headers to mark them as uncertain 1458 unknown := make([]*types.Header, 0, len(chunk)) 1459 for _, header := range chunk { 1460 if !d.lightchain.HasHeader(header.Hash(), header.Number.Uint64()) { 1461 unknown = append(unknown, header) 1462 } 1463 } 1464 // If we're importing pure headers, verify based on their recentness 1465 frequency := fsHeaderCheckFrequency 1466 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1467 frequency = 1 1468 } 1469 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1470 // If some headers were inserted, add them too to the rollback list 1471 if n > 0 { 1472 rollback = append(rollback, chunk[:n]...) 1473 } 1474 log.Debug("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "err", err) 1475 return errInvalidChain 1476 } 1477 // All verifications passed, store newly found uncertain headers 1478 rollback = append(rollback, unknown...) 1479 if len(rollback) > fsHeaderSafetyNet { 1480 rollback = append(rollback[:0], rollback[len(rollback)-fsHeaderSafetyNet:]...) 1481 } 1482 } 1483 // Unless we're doing light chains, schedule the headers for associated content retrieval 1484 if d.mode == FullSync || d.mode == FastSync { 1485 // If we've reached the allowed number of pending headers, stall a bit 1486 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1487 select { 1488 case <-d.cancelCh: 1489 return errCanceled 1490 case <-time.After(time.Second): 1491 } 1492 } 1493 // Otherwise insert the headers for content retrieval 1494 inserts := d.queue.Schedule(chunk, origin) 1495 if len(inserts) != len(chunk) { 1496 log.Debug("Stale headers") 1497 return errBadPeer 1498 } 1499 } 1500 headers = headers[limit:] 1501 origin += uint64(limit) 1502 } 1503 // Update the highest block number we know if a higher one is found. 1504 d.syncStatsLock.Lock() 1505 if d.syncStatsChainHeight < origin { 1506 d.syncStatsChainHeight = origin - 1 1507 } 1508 d.syncStatsLock.Unlock() 1509 1510 // Signal the content downloaders of the availablility of new tasks 1511 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1512 select { 1513 case ch <- true: 1514 default: 1515 } 1516 } 1517 } 1518 } 1519 } 1520 1521 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1522 func (d *Downloader) processFullSyncContent() error { 1523 for { 1524 results := d.queue.Results(true) 1525 if len(results) == 0 { 1526 return nil 1527 } 1528 if d.chainInsertHook != nil { 1529 d.chainInsertHook(results) 1530 } 1531 if err := d.importBlockResults(results); err != nil { 1532 return err 1533 } 1534 } 1535 } 1536 1537 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1538 // Check for any early termination requests 1539 if len(results) == 0 { 1540 return nil 1541 } 1542 select { 1543 case <-d.quitCh: 1544 return errCancelContentProcessing 1545 default: 1546 } 1547 // Retrieve the a batch of results to import 1548 first, last := results[0].Header, results[len(results)-1].Header 1549 log.Debug("Inserting downloaded chain", "items", len(results), 1550 "firstnum", first.Number, "firsthash", first.Hash(), 1551 "lastnum", last.Number, "lasthash", last.Hash(), 1552 ) 1553 blocks := make([]*types.Block, len(results)) 1554 for i, result := range results { 1555 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1556 } 1557 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1558 if index < len(results) { 1559 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1560 } else { 1561 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1562 // when it needs to preprocess blocks to import a sidechain. 1563 // The importer will put together a new list of blocks to import, which is a superset 1564 // of the blocks delivered from the downloader, and the indexing will be off. 1565 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1566 } 1567 return errInvalidChain 1568 } 1569 return nil 1570 } 1571 1572 // processFastSyncContent takes fetch results from the queue and writes them to the 1573 // database. It also controls the synchronisation of state nodes of the pivot block. 1574 func (d *Downloader) processFastSyncContent(latest *types.Header) error { 1575 // Start syncing state of the reported head block. This should get us most of 1576 // the state of the pivot block. 1577 sync := d.syncState(latest.Root) 1578 defer sync.Cancel() 1579 closeOnErr := func(s *stateSync) { 1580 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled { 1581 d.queue.Close() // wake up Results 1582 } 1583 } 1584 go closeOnErr(sync) 1585 // Figure out the ideal pivot block. Note, that this goalpost may move if the 1586 // sync takes long enough for the chain head to move significantly. 1587 pivot := uint64(0) 1588 if height := latest.Number.Uint64(); height > uint64(fsMinFullBlocks) { 1589 pivot = height - uint64(fsMinFullBlocks) 1590 } 1591 // To cater for moving pivot points, track the pivot block and subsequently 1592 // accumulated download results separately. 1593 var ( 1594 oldPivot *fetchResult // Locked in pivot block, might change eventually 1595 oldTail []*fetchResult // Downloaded content after the pivot 1596 ) 1597 for { 1598 // Wait for the next batch of downloaded data to be available, and if the pivot 1599 // block became stale, move the goalpost 1600 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1601 if len(results) == 0 { 1602 // If pivot sync is done, stop 1603 if oldPivot == nil { 1604 return sync.Cancel() 1605 } 1606 // If sync failed, stop 1607 select { 1608 case <-d.cancelCh: 1609 sync.Cancel() 1610 return errCanceled 1611 default: 1612 } 1613 } 1614 if d.chainInsertHook != nil { 1615 d.chainInsertHook(results) 1616 } 1617 if oldPivot != nil { 1618 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1619 } 1620 // Split around the pivot block and process the two sides via fast/full sync 1621 if atomic.LoadInt32(&d.committed) == 0 { 1622 latest = results[len(results)-1].Header 1623 if height := latest.Number.Uint64(); height > pivot+2*uint64(fsMinFullBlocks) { 1624 log.Warn("Pivot became stale, moving", "old", pivot, "new", height-uint64(fsMinFullBlocks)) 1625 pivot = height - uint64(fsMinFullBlocks) 1626 } 1627 } 1628 P, beforeP, afterP := splitAroundPivot(pivot, results) 1629 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1630 return err 1631 } 1632 if P != nil { 1633 // If new pivot block found, cancel old state retrieval and restart 1634 if oldPivot != P { 1635 sync.Cancel() 1636 1637 sync = d.syncState(P.Header.Root) 1638 defer sync.Cancel() 1639 go closeOnErr(sync) 1640 oldPivot = P 1641 } 1642 // Wait for completion, occasionally checking for pivot staleness 1643 select { 1644 case <-sync.done: 1645 if sync.err != nil { 1646 return sync.err 1647 } 1648 if err := d.commitPivotBlock(P); err != nil { 1649 return err 1650 } 1651 oldPivot = nil 1652 1653 case <-time.After(time.Second): 1654 oldTail = afterP 1655 continue 1656 } 1657 } 1658 // Fast sync done, pivot commit done, full import 1659 if err := d.importBlockResults(afterP); err != nil { 1660 return err 1661 } 1662 } 1663 } 1664 1665 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1666 for _, result := range results { 1667 num := result.Header.Number.Uint64() 1668 switch { 1669 case num < pivot: 1670 before = append(before, result) 1671 case num == pivot: 1672 p = result 1673 default: 1674 after = append(after, result) 1675 } 1676 } 1677 return p, before, after 1678 } 1679 1680 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1681 // Check for any early termination requests 1682 if len(results) == 0 { 1683 return nil 1684 } 1685 select { 1686 case <-d.quitCh: 1687 return errCancelContentProcessing 1688 case <-stateSync.done: 1689 if err := stateSync.Wait(); err != nil { 1690 return err 1691 } 1692 default: 1693 } 1694 // Retrieve the a batch of results to import 1695 first, last := results[0].Header, results[len(results)-1].Header 1696 log.Debug("Inserting fast-sync blocks", "items", len(results), 1697 "firstnum", first.Number, "firsthash", first.Hash(), 1698 "lastnumn", last.Number, "lasthash", last.Hash(), 1699 ) 1700 blocks := make([]*types.Block, len(results)) 1701 receipts := make([]types.Receipts, len(results)) 1702 for i, result := range results { 1703 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1704 receipts[i] = result.Receipts 1705 } 1706 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1707 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1708 return errInvalidChain 1709 } 1710 return nil 1711 } 1712 1713 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1714 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1715 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1716 1717 // Commit the pivot block as the new head, will require full sync from here on 1718 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1719 return err 1720 } 1721 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1722 return err 1723 } 1724 atomic.StoreInt32(&d.committed, 1) 1725 1726 // If we had a bloom filter for the state sync, deallocate it now. Note, we only 1727 // deallocate internally, but keep the empty wrapper. This ensures that if we do 1728 // a rollback after committing the pivot and restarting fast sync, we don't end 1729 // up using a nil bloom. Empty bloom is fine, it just returns that it does not 1730 // have the info we need, so reach down to the database instead. 1731 if d.stateBloom != nil { 1732 d.stateBloom.Close() 1733 } 1734 return nil 1735 } 1736 1737 // DeliverHeaders injects a new batch of block headers received from a remote 1738 // node into the download schedule. 1739 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) { 1740 return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1741 } 1742 1743 // DeliverBodies injects a new batch of block bodies received from a remote node. 1744 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) { 1745 return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1746 } 1747 1748 // DeliverReceipts injects a new batch of receipts received from a remote node. 1749 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) { 1750 return d.deliver(id, d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1751 } 1752 1753 // DeliverNodeData injects a new batch of node state data received from a remote node. 1754 func (d *Downloader) DeliverNodeData(id string, data [][]byte) (err error) { 1755 return d.deliver(id, d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1756 } 1757 1758 // deliver injects a new batch of data received from a remote node. 1759 func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1760 // Update the delivery metrics for both good and failed deliveries 1761 inMeter.Mark(int64(packet.Items())) 1762 defer func() { 1763 if err != nil { 1764 dropMeter.Mark(int64(packet.Items())) 1765 } 1766 }() 1767 // Deliver or abort if the sync is canceled while queuing 1768 d.cancelLock.RLock() 1769 cancel := d.cancelCh 1770 d.cancelLock.RUnlock() 1771 if cancel == nil { 1772 return errNoSyncActive 1773 } 1774 select { 1775 case destCh <- packet: 1776 return nil 1777 case <-cancel: 1778 return errNoSyncActive 1779 } 1780 } 1781 1782 // qosTuner is the quality of service tuning loop that occasionally gathers the 1783 // peer latency statistics and updates the estimated request round trip time. 1784 func (d *Downloader) qosTuner() { 1785 for { 1786 // Retrieve the current median RTT and integrate into the previoust target RTT 1787 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 1788 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 1789 1790 // A new RTT cycle passed, increase our confidence in the estimated RTT 1791 conf := atomic.LoadUint64(&d.rttConfidence) 1792 conf = conf + (1000000-conf)/2 1793 atomic.StoreUint64(&d.rttConfidence, conf) 1794 1795 // Log the new QoS values and sleep until the next RTT 1796 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1797 select { 1798 case <-d.quitCh: 1799 return 1800 case <-time.After(rtt): 1801 } 1802 } 1803 } 1804 1805 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 1806 // peer set, needing to reduce the confidence we have in out QoS estimates. 1807 func (d *Downloader) qosReduceConfidence() { 1808 // If we have a single peer, confidence is always 1 1809 peers := uint64(d.peers.Len()) 1810 if peers == 0 { 1811 // Ensure peer connectivity races don't catch us off guard 1812 return 1813 } 1814 if peers == 1 { 1815 atomic.StoreUint64(&d.rttConfidence, 1000000) 1816 return 1817 } 1818 // If we have a ton of peers, don't drop confidence) 1819 if peers >= uint64(qosConfidenceCap) { 1820 return 1821 } 1822 // Otherwise drop the confidence factor 1823 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 1824 if float64(conf)/1000000 < rttMinConfidence { 1825 conf = uint64(rttMinConfidence * 1000000) 1826 } 1827 atomic.StoreUint64(&d.rttConfidence, conf) 1828 1829 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1830 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1831 } 1832 1833 // requestRTT returns the current target round trip time for a download request 1834 // to complete in. 1835 // 1836 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 1837 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 1838 // be adapted to, but smaller ones are preferred (stabler download stream). 1839 func (d *Downloader) requestRTT() time.Duration { 1840 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 1841 } 1842 1843 // requestTTL returns the current timeout allowance for a single download request 1844 // to finish under. 1845 func (d *Downloader) requestTTL() time.Duration { 1846 var ( 1847 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1848 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 1849 ) 1850 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 1851 if ttl > ttlLimit { 1852 ttl = ttlLimit 1853 } 1854 return ttl 1855 }