github.com/RobustRoundRobin/quorum@v20.10.0+incompatible/eth/downloader/downloader.go (about) 1 // Copyright 2015 The go-ethereum Authors 2 // This file is part of the go-ethereum library. 3 // 4 // The go-ethereum library is free software: you can redistribute it and/or modify 5 // it under the terms of the GNU Lesser General Public License as published by 6 // the Free Software Foundation, either version 3 of the License, or 7 // (at your option) any later version. 8 // 9 // The go-ethereum library is distributed in the hope that it will be useful, 10 // but WITHOUT ANY WARRANTY; without even the implied warranty of 11 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 // GNU Lesser General Public License for more details. 13 // 14 // You should have received a copy of the GNU Lesser General Public License 15 // along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>. 16 17 // Package downloader contains the manual full chain synchronisation. 18 package downloader 19 20 import ( 21 "errors" 22 "fmt" 23 "math/big" 24 "sync" 25 "sync/atomic" 26 "time" 27 28 "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 errCancelBlockFetch = errors.New("block download canceled (requested)") 89 errCancelHeaderFetch = errors.New("block header download canceled (requested)") 90 errCancelStateFetch = errors.New("state data download canceled (requested)") 91 errCancelContentProcessing = errors.New("content processing canceled (requested)") 92 errCanceled = errors.New("syncing canceled (requested)") 93 errNoSyncActive = errors.New("no sync active") 94 errTooOld = errors.New("peer doesn't speak recent enough protocol version (need version >= 62)") 95 ) 96 97 type Downloader struct { 98 // WARNING: The `rttEstimate` and `rttConfidence` fields are accessed atomically. 99 // On 32 bit platforms, only 64-bit aligned fields can be atomic. The struct is 100 // guaranteed to be so aligned, so take advantage of that. For more information, 101 // see https://golang.org/pkg/sync/atomic/#pkg-note-BUG. 102 rttEstimate uint64 // Round trip time to target for download requests 103 rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops) 104 105 mode SyncMode // Synchronisation mode defining the strategy used (per sync cycle) 106 mux *event.TypeMux // Event multiplexer to announce sync operation events 107 108 checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync) 109 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 110 queue *queue // Scheduler for selecting the hashes to download 111 peers *peerSet // Set of active peers from which download can proceed 112 113 stateDB ethdb.Database // Database to state sync into (and deduplicate via) 114 stateBloom *trie.SyncBloom // Bloom filter for fast trie node existence checks 115 116 // Statistics 117 syncStatsChainOrigin uint64 // Origin block number where syncing started at 118 syncStatsChainHeight uint64 // Highest block number known when syncing started 119 syncStatsState stateSyncStats 120 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 121 122 lightchain LightChain 123 blockchain BlockChain 124 125 // Callbacks 126 dropPeer peerDropFn // Drops a peer for misbehaving 127 128 // Status 129 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 130 synchronising int32 131 notified int32 132 committed int32 133 ancientLimit uint64 // The maximum block number which can be regarded as ancient data. 134 135 // Channels 136 headerCh chan dataPack // [eth/62] Channel receiving inbound block headers 137 bodyCh chan dataPack // [eth/62] Channel receiving inbound block bodies 138 receiptCh chan dataPack // [eth/63] Channel receiving inbound receipts 139 bodyWakeCh chan bool // [eth/62] Channel to signal the block body fetcher of new tasks 140 receiptWakeCh chan bool // [eth/63] Channel to signal the receipt fetcher of new tasks 141 headerProcCh chan []*types.Header // [eth/62] Channel to feed the header processor new tasks 142 143 // for stateFetcher 144 stateSyncStart chan *stateSync 145 trackStateReq chan *stateReq 146 stateCh chan dataPack // [eth/63] Channel receiving inbound node state data 147 148 // Cancellation and termination 149 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 150 cancelCh chan struct{} // Channel to cancel mid-flight syncs 151 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 152 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 153 154 quitCh chan struct{} // Quit channel to signal termination 155 quitLock sync.RWMutex // Lock to prevent double closes 156 157 // Testing hooks 158 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 159 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 160 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 161 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 162 } 163 164 // LightChain encapsulates functions required to synchronise a light chain. 165 type LightChain interface { 166 // HasHeader verifies a header's presence in the local chain. 167 HasHeader(common.Hash, uint64) bool 168 169 // GetHeaderByHash retrieves a header from the local chain. 170 GetHeaderByHash(common.Hash) *types.Header 171 172 // CurrentHeader retrieves the head header from the local chain. 173 CurrentHeader() *types.Header 174 175 // GetTd returns the total difficulty of a local block. 176 GetTd(common.Hash, uint64) *big.Int 177 178 // InsertHeaderChain inserts a batch of headers into the local chain. 179 InsertHeaderChain([]*types.Header, int) (int, error) 180 181 // Rollback removes a few recently added elements from the local chain. 182 Rollback([]common.Hash) 183 } 184 185 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 186 type BlockChain interface { 187 LightChain 188 189 // HasBlock verifies a block's presence in the local chain. 190 HasBlock(common.Hash, uint64) bool 191 192 // HasFastBlock verifies a fast block's presence in the local chain. 193 HasFastBlock(common.Hash, uint64) bool 194 195 // GetBlockByHash retrieves a block from the local chain. 196 GetBlockByHash(common.Hash) *types.Block 197 198 // CurrentBlock retrieves the head block from the local chain. 199 CurrentBlock() *types.Block 200 201 // CurrentFastBlock retrieves the head fast block from the local chain. 202 CurrentFastBlock() *types.Block 203 204 // FastSyncCommitHead directly commits the head block to a certain entity. 205 FastSyncCommitHead(common.Hash) error 206 207 // InsertChain inserts a batch of blocks into the local chain. 208 InsertChain(types.Blocks) (int, error) 209 210 // InsertReceiptChain inserts a batch of receipts into the local chain. 211 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 212 } 213 214 // New creates a new downloader to fetch hashes and blocks from remote peers. 215 func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 216 // reset the value of maxForkAncenstry for Quorum based 217 maxForkAncestry = uint64(params.GetImmutabilityThreshold()) 218 219 if lightchain == nil { 220 lightchain = chain 221 } 222 dl := &Downloader{ 223 stateDB: stateDb, 224 stateBloom: stateBloom, 225 mux: mux, 226 checkpoint: checkpoint, 227 queue: newQueue(), 228 peers: newPeerSet(), 229 rttEstimate: uint64(rttMaxEstimate), 230 rttConfidence: uint64(1000000), 231 blockchain: chain, 232 lightchain: lightchain, 233 dropPeer: dropPeer, 234 headerCh: make(chan dataPack, 1), 235 bodyCh: make(chan dataPack, 1), 236 receiptCh: make(chan dataPack, 1), 237 bodyWakeCh: make(chan bool, 1), 238 receiptWakeCh: make(chan bool, 1), 239 headerProcCh: make(chan []*types.Header, 1), 240 quitCh: make(chan struct{}), 241 stateCh: make(chan dataPack), 242 stateSyncStart: make(chan *stateSync), 243 syncStatsState: stateSyncStats{ 244 processed: rawdb.ReadFastTrieProgress(stateDb), 245 }, 246 trackStateReq: make(chan *stateReq), 247 } 248 go dl.qosTuner() 249 go dl.stateFetcher() 250 return dl 251 } 252 253 // Progress retrieves the synchronisation boundaries, specifically the origin 254 // block where synchronisation started at (may have failed/suspended); the block 255 // or header sync is currently at; and the latest known block which the sync targets. 256 // 257 // In addition, during the state download phase of fast synchronisation the number 258 // of processed and the total number of known states are also returned. Otherwise 259 // these are zero. 260 func (d *Downloader) Progress() ethereum.SyncProgress { 261 // Lock the current stats and return the progress 262 d.syncStatsLock.RLock() 263 defer d.syncStatsLock.RUnlock() 264 265 current := uint64(0) 266 switch { 267 case d.blockchain != nil && d.mode == FullSync: 268 current = d.blockchain.CurrentBlock().NumberU64() 269 case d.blockchain != nil && d.mode == FastSync: 270 current = d.blockchain.CurrentFastBlock().NumberU64() 271 case d.lightchain != nil: 272 current = d.lightchain.CurrentHeader().Number.Uint64() 273 default: 274 log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", d.mode) 275 } 276 return ethereum.SyncProgress{ 277 StartingBlock: d.syncStatsChainOrigin, 278 CurrentBlock: current, 279 HighestBlock: d.syncStatsChainHeight, 280 PulledStates: d.syncStatsState.processed, 281 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 282 } 283 } 284 285 // Synchronising returns whether the downloader is currently retrieving blocks. 286 func (d *Downloader) Synchronising() bool { 287 return atomic.LoadInt32(&d.synchronising) > 0 288 } 289 290 // RegisterPeer injects a new download peer into the set of block source to be 291 // used for fetching hashes and blocks from. 292 func (d *Downloader) RegisterPeer(id string, version int, peer Peer) error { 293 logger := log.New("peer", id) 294 logger.Trace("Registering sync peer") 295 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 296 logger.Error("Failed to register sync peer", "err", err) 297 return err 298 } 299 d.qosReduceConfidence() 300 301 return nil 302 } 303 304 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 305 func (d *Downloader) RegisterLightPeer(id string, version int, peer LightPeer) error { 306 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 307 } 308 309 // UnregisterPeer remove a peer from the known list, preventing any action from 310 // the specified peer. An effort is also made to return any pending fetches into 311 // the queue. 312 func (d *Downloader) UnregisterPeer(id string) error { 313 // Unregister the peer from the active peer set and revoke any fetch tasks 314 logger := log.New("peer", id) 315 logger.Trace("Unregistering sync peer") 316 if err := d.peers.Unregister(id); err != nil { 317 logger.Error("Failed to unregister sync peer", "err", err) 318 return err 319 } 320 d.queue.Revoke(id) 321 322 return nil 323 } 324 325 // Synchronise tries to sync up our local block chain with a remote peer, both 326 // adding various sanity checks as well as wrapping it with various log entries. 327 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 328 err := d.synchronise(id, head, td, mode) 329 switch err { 330 case nil: 331 case errBusy, errCanceled: 332 333 case errTimeout, errBadPeer, errStallingPeer, errUnsyncedPeer, 334 errEmptyHeaderSet, errPeersUnavailable, errTooOld, 335 errInvalidAncestor, errInvalidChain: 336 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 337 if d.dropPeer == nil { 338 // The dropPeer method is nil when `--copydb` is used for a local copy. 339 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 340 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 341 } else { 342 d.dropPeer(id) 343 } 344 default: 345 log.Warn("Synchronisation failed, retrying", "err", err) 346 } 347 return err 348 } 349 350 // synchronise will select the peer and use it for synchronising. If an empty string is given 351 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 352 // checks fail an error will be returned. This method is synchronous 353 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 354 // Mock out the synchronisation if testing 355 if d.synchroniseMock != nil { 356 return d.synchroniseMock(id, hash) 357 } 358 // Make sure only one goroutine is ever allowed past this point at once 359 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 360 return errBusy 361 } 362 363 // Quorum 364 // changes for permissions. added set sync status to indicate permissions that node sync has started 365 types.SetSyncStatus() 366 367 defer atomic.StoreInt32(&d.synchronising, 0) 368 369 // Post a user notification of the sync (only once per session) 370 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 371 log.Info("Block synchronisation started") 372 } 373 // If we are already full syncing, but have a fast-sync bloom filter laying 374 // around, make sure it does't use memory any more. This is a special case 375 // when the user attempts to fast sync a new empty network. 376 if mode == FullSync && d.stateBloom != nil { 377 d.stateBloom.Close() 378 } 379 // Reset the queue, peer set and wake channels to clean any internal leftover state 380 d.queue.Reset() 381 d.peers.Reset() 382 383 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 384 select { 385 case <-ch: 386 default: 387 } 388 } 389 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 390 for empty := false; !empty; { 391 select { 392 case <-ch: 393 default: 394 empty = true 395 } 396 } 397 } 398 for empty := false; !empty; { 399 select { 400 case <-d.headerProcCh: 401 default: 402 empty = true 403 } 404 } 405 // Create cancel channel for aborting mid-flight and mark the master peer 406 d.cancelLock.Lock() 407 d.cancelCh = make(chan struct{}) 408 d.cancelPeer = id 409 d.cancelLock.Unlock() 410 411 defer d.Cancel() // No matter what, we can't leave the cancel channel open 412 413 // Set the requested sync mode, unless it's forbidden 414 d.mode = mode 415 416 // Retrieve the origin peer and initiate the downloading process 417 p := d.peers.Peer(id) 418 if p == nil { 419 return errUnknownPeer 420 } 421 if d.mode == BoundedFullSync { 422 return d.syncWithPeerUntil(p, hash, td) 423 } 424 return d.syncWithPeer(p, hash, td) 425 } 426 427 // syncWithPeer starts a block synchronization based on the hash chain from the 428 // specified peer and head hash. 429 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 430 d.mux.Post(StartEvent{}) 431 defer func() { 432 // reset on error 433 if err != nil { 434 d.mux.Post(FailedEvent{err}) 435 } else { 436 latest := d.lightchain.CurrentHeader() 437 d.mux.Post(DoneEvent{latest}) 438 } 439 }() 440 if p.version < 62 { 441 return errTooOld 442 } 443 444 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", d.mode) 445 defer func(start time.Time) { 446 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 447 }(time.Now()) 448 449 // Look up the sync boundaries: the common ancestor and the target block 450 latest, err := d.fetchHeight(p) 451 if err != nil { 452 return err 453 } 454 height := latest.Number.Uint64() 455 456 origin, err := d.findAncestor(p, latest) 457 if err != nil { 458 return err 459 } 460 d.syncStatsLock.Lock() 461 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 462 d.syncStatsChainOrigin = origin 463 } 464 d.syncStatsChainHeight = height 465 d.syncStatsLock.Unlock() 466 467 // Ensure our origin point is below any fast sync pivot point 468 pivot := uint64(0) 469 if d.mode == FastSync { 470 if height <= uint64(fsMinFullBlocks) { 471 origin = 0 472 } else { 473 pivot = height - uint64(fsMinFullBlocks) 474 if pivot <= origin { 475 origin = pivot - 1 476 } 477 } 478 } 479 d.committed = 1 480 if d.mode == FastSync && pivot != 0 { 481 d.committed = 0 482 } 483 if d.mode == FastSync { 484 // Set the ancient data limitation. 485 // If we are running fast sync, all block data older than ancientLimit will be 486 // written to the ancient store. More recent data will be written to the active 487 // database and will wait for the freezer to migrate. 488 // 489 // If there is a checkpoint available, then calculate the ancientLimit through 490 // that. Otherwise calculate the ancient limit through the advertised height 491 // of the remote peer. 492 // 493 // The reason for picking checkpoint first is that a malicious peer can give us 494 // a fake (very high) height, forcing the ancient limit to also be very high. 495 // The peer would start to feed us valid blocks until head, resulting in all of 496 // the blocks might be written into the ancient store. A following mini-reorg 497 // could cause issues. 498 if d.checkpoint != 0 && d.checkpoint > maxForkAncestry+1 { 499 d.ancientLimit = d.checkpoint 500 } else if height > maxForkAncestry+1 { 501 d.ancientLimit = height - maxForkAncestry - 1 502 } 503 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 504 // If a part of blockchain data has already been written into active store, 505 // disable the ancient style insertion explicitly. 506 if origin >= frozen && frozen != 0 { 507 d.ancientLimit = 0 508 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 509 } else if d.ancientLimit > 0 { 510 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 511 } 512 // Rewind the ancient store and blockchain if reorg happens. 513 if origin+1 < frozen { 514 var hashes []common.Hash 515 for i := origin + 1; i < d.lightchain.CurrentHeader().Number.Uint64(); i++ { 516 hashes = append(hashes, rawdb.ReadCanonicalHash(d.stateDB, i)) 517 } 518 d.lightchain.Rollback(hashes) 519 } 520 } 521 // Initiate the sync using a concurrent header and content retrieval algorithm 522 d.queue.Prepare(origin+1, d.mode) 523 if d.syncInitHook != nil { 524 d.syncInitHook(origin, height) 525 } 526 fetchers := []func() error{ 527 func() error { return d.fetchHeaders(p, origin+1, pivot) }, // Headers are always retrieved 528 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 529 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 530 func() error { return d.processHeaders(origin+1, pivot, td) }, 531 } 532 if d.mode == FastSync { 533 fetchers = append(fetchers, func() error { return d.processFastSyncContent(latest) }) 534 } else if d.mode == FullSync { 535 fetchers = append(fetchers, d.processFullSyncContent) 536 } 537 return d.spawnSync(fetchers) 538 } 539 540 // spawnSync runs d.process and all given fetcher functions to completion in 541 // separate goroutines, returning the first error that appears. 542 func (d *Downloader) spawnSync(fetchers []func() error) error { 543 errc := make(chan error, len(fetchers)) 544 d.cancelWg.Add(len(fetchers)) 545 for _, fn := range fetchers { 546 fn := fn 547 go func() { defer d.cancelWg.Done(); errc <- fn() }() 548 } 549 // Wait for the first error, then terminate the others. 550 var err error 551 for i := 0; i < len(fetchers); i++ { 552 if i == len(fetchers)-1 { 553 // Close the queue when all fetchers have exited. 554 // This will cause the block processor to end when 555 // it has processed the queue. 556 d.queue.Close() 557 } 558 if err = <-errc; err != nil && err != errCanceled { 559 break 560 } 561 } 562 d.queue.Close() 563 d.Cancel() 564 return err 565 } 566 567 // cancel aborts all of the operations and resets the queue. However, cancel does 568 // not wait for the running download goroutines to finish. This method should be 569 // used when cancelling the downloads from inside the downloader. 570 func (d *Downloader) cancel() { 571 // Close the current cancel channel 572 d.cancelLock.Lock() 573 if d.cancelCh != nil { 574 select { 575 case <-d.cancelCh: 576 // Channel was already closed 577 default: 578 close(d.cancelCh) 579 } 580 } 581 d.cancelLock.Unlock() 582 } 583 584 // Cancel aborts all of the operations and waits for all download goroutines to 585 // finish before returning. 586 func (d *Downloader) Cancel() { 587 d.cancel() 588 d.cancelWg.Wait() 589 590 d.ancientLimit = 0 591 log.Debug("Reset ancient limit to zero") 592 } 593 594 // Terminate interrupts the downloader, canceling all pending operations. 595 // The downloader cannot be reused after calling Terminate. 596 func (d *Downloader) Terminate() { 597 // Close the termination channel (make sure double close is allowed) 598 d.quitLock.Lock() 599 select { 600 case <-d.quitCh: 601 default: 602 close(d.quitCh) 603 } 604 d.quitLock.Unlock() 605 606 // Cancel any pending download requests 607 d.Cancel() 608 } 609 610 // fetchHeight retrieves the head header of the remote peer to aid in estimating 611 // the total time a pending synchronisation would take. 612 func (d *Downloader) fetchHeight(p *peerConnection) (*types.Header, error) { 613 p.log.Debug("Retrieving remote chain height") 614 615 // Request the advertised remote head block and wait for the response 616 head, _ := p.peer.Head() 617 go p.peer.RequestHeadersByHash(head, 1, 0, false) 618 619 ttl := d.requestTTL() 620 timeout := time.After(ttl) 621 for { 622 select { 623 case <-d.cancelCh: 624 return nil, errCanceled 625 626 case packet := <-d.headerCh: 627 // Discard anything not from the origin peer 628 if packet.PeerId() != p.id { 629 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 630 break 631 } 632 // Make sure the peer actually gave something valid 633 headers := packet.(*headerPack).headers 634 if len(headers) != 1 { 635 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 636 return nil, errBadPeer 637 } 638 head := headers[0] 639 if (d.mode == FastSync || d.mode == LightSync) && head.Number.Uint64() < d.checkpoint { 640 p.log.Warn("Remote head below checkpoint", "number", head.Number, "hash", head.Hash()) 641 return nil, errUnsyncedPeer 642 } 643 p.log.Debug("Remote head header identified", "number", head.Number, "hash", head.Hash()) 644 return head, nil 645 646 case <-timeout: 647 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 648 return nil, errTimeout 649 650 case <-d.bodyCh: 651 case <-d.receiptCh: 652 // Out of bounds delivery, ignore 653 } 654 } 655 } 656 657 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 658 // common ancestor. 659 // It returns parameters to be used for peer.RequestHeadersByNumber: 660 // from - starting block number 661 // count - number of headers to request 662 // skip - number of headers to skip 663 // and also returns 'max', the last block which is expected to be returned by the remote peers, 664 // given the (from,count,skip) 665 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 666 var ( 667 from int 668 count int 669 MaxCount = MaxHeaderFetch / 16 670 ) 671 // requestHead is the highest block that we will ask for. If requestHead is not offset, 672 // the highest block that we will get is 16 blocks back from head, which means we 673 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 674 // between us and the peer is 1-2 blocks, which is most common 675 requestHead := int(remoteHeight) - 1 676 if requestHead < 0 { 677 requestHead = 0 678 } 679 // requestBottom is the lowest block we want included in the query 680 // Ideally, we want to include just below own head 681 requestBottom := int(localHeight - 1) 682 if requestBottom < 0 { 683 requestBottom = 0 684 } 685 totalSpan := requestHead - requestBottom 686 span := 1 + totalSpan/MaxCount 687 if span < 2 { 688 span = 2 689 } 690 if span > 16 { 691 span = 16 692 } 693 694 count = 1 + totalSpan/span 695 if count > MaxCount { 696 count = MaxCount 697 } 698 if count < 2 { 699 count = 2 700 } 701 from = requestHead - (count-1)*span 702 if from < 0 { 703 from = 0 704 } 705 max := from + (count-1)*span 706 return int64(from), count, span - 1, uint64(max) 707 } 708 709 // findAncestor tries to locate the common ancestor link of the local chain and 710 // a remote peers blockchain. In the general case when our node was in sync and 711 // on the correct chain, checking the top N links should already get us a match. 712 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 713 // the head links match), we do a binary search to find the common ancestor. 714 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 715 // Figure out the valid ancestor range to prevent rewrite attacks 716 var ( 717 floor = int64(-1) 718 localHeight uint64 719 remoteHeight = remoteHeader.Number.Uint64() 720 ) 721 switch d.mode { 722 case FullSync: 723 localHeight = d.blockchain.CurrentBlock().NumberU64() 724 case FastSync: 725 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 726 default: 727 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 728 } 729 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 730 731 // Recap floor value for binary search 732 if localHeight >= maxForkAncestry { 733 // We're above the max reorg threshold, find the earliest fork point 734 floor = int64(localHeight - maxForkAncestry) 735 } 736 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 737 // all headers before that point will be missing. 738 if d.mode == LightSync { 739 // If we dont know the current CHT position, find it 740 if d.genesis == 0 { 741 header := d.lightchain.CurrentHeader() 742 for header != nil { 743 d.genesis = header.Number.Uint64() 744 if floor >= int64(d.genesis)-1 { 745 break 746 } 747 header = d.lightchain.GetHeaderByHash(header.ParentHash) 748 } 749 } 750 // We already know the "genesis" block number, cap floor to that 751 if floor < int64(d.genesis)-1 { 752 floor = int64(d.genesis) - 1 753 } 754 } 755 756 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 757 758 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 759 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 760 761 // Wait for the remote response to the head fetch 762 number, hash := uint64(0), common.Hash{} 763 764 ttl := d.requestTTL() 765 timeout := time.After(ttl) 766 767 for finished := false; !finished; { 768 select { 769 case <-d.cancelCh: 770 return 0, errCanceled 771 772 case packet := <-d.headerCh: 773 // Discard anything not from the origin peer 774 if packet.PeerId() != p.id { 775 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 776 break 777 } 778 // Make sure the peer actually gave something valid 779 headers := packet.(*headerPack).headers 780 if len(headers) == 0 { 781 p.log.Warn("Empty head header set") 782 return 0, errEmptyHeaderSet 783 } 784 // Make sure the peer's reply conforms to the request 785 for i, header := range headers { 786 expectNumber := from + int64(i)*int64(skip+1) 787 if number := header.Number.Int64(); number != expectNumber { 788 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 789 return 0, errInvalidChain 790 } 791 } 792 // Check if a common ancestor was found 793 finished = true 794 for i := len(headers) - 1; i >= 0; i-- { 795 // Skip any headers that underflow/overflow our requested set 796 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 797 continue 798 } 799 // Otherwise check if we already know the header or not 800 h := headers[i].Hash() 801 n := headers[i].Number.Uint64() 802 803 var known bool 804 switch d.mode { 805 case FullSync: 806 known = d.blockchain.HasBlock(h, n) 807 case FastSync: 808 known = d.blockchain.HasFastBlock(h, n) 809 default: 810 known = d.lightchain.HasHeader(h, n) 811 } 812 if known { 813 number, hash = n, h 814 break 815 } 816 } 817 818 case <-timeout: 819 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 820 return 0, errTimeout 821 822 case <-d.bodyCh: 823 case <-d.receiptCh: 824 // Out of bounds delivery, ignore 825 } 826 } 827 // If the head fetch already found an ancestor, return 828 if hash != (common.Hash{}) { 829 if int64(number) <= floor { 830 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 831 return 0, errInvalidAncestor 832 } 833 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 834 return number, nil 835 } 836 // Ancestor not found, we need to binary search over our chain 837 start, end := uint64(0), remoteHeight 838 if floor > 0 { 839 start = uint64(floor) 840 } 841 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 842 843 for start+1 < end { 844 // Split our chain interval in two, and request the hash to cross check 845 check := (start + end) / 2 846 847 ttl := d.requestTTL() 848 timeout := time.After(ttl) 849 850 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 851 852 // Wait until a reply arrives to this request 853 for arrived := false; !arrived; { 854 select { 855 case <-d.cancelCh: 856 return 0, errCanceled 857 858 case packer := <-d.headerCh: 859 // Discard anything not from the origin peer 860 if packer.PeerId() != p.id { 861 log.Debug("Received headers from incorrect peer", "peer", packer.PeerId()) 862 break 863 } 864 // Make sure the peer actually gave something valid 865 headers := packer.(*headerPack).headers 866 if len(headers) != 1 { 867 p.log.Debug("Multiple headers for single request", "headers", len(headers)) 868 return 0, errBadPeer 869 } 870 arrived = true 871 872 // Modify the search interval based on the response 873 h := headers[0].Hash() 874 n := headers[0].Number.Uint64() 875 876 var known bool 877 switch d.mode { 878 case FullSync: 879 known = d.blockchain.HasBlock(h, n) 880 case FastSync: 881 known = d.blockchain.HasFastBlock(h, n) 882 default: 883 known = d.lightchain.HasHeader(h, n) 884 } 885 if !known { 886 end = check 887 break 888 } 889 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 890 if header.Number.Uint64() != check { 891 p.log.Debug("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 892 return 0, errBadPeer 893 } 894 start = check 895 hash = h 896 897 case <-timeout: 898 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 899 return 0, errTimeout 900 901 case <-d.bodyCh: 902 case <-d.receiptCh: 903 // Out of bounds delivery, ignore 904 } 905 } 906 } 907 // Ensure valid ancestry and return 908 if int64(start) <= floor { 909 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 910 return 0, errInvalidAncestor 911 } 912 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 913 return start, nil 914 } 915 916 // fetchHeaders keeps retrieving headers concurrently from the number 917 // requested, until no more are returned, potentially throttling on the way. To 918 // facilitate concurrency but still protect against malicious nodes sending bad 919 // headers, we construct a header chain skeleton using the "origin" peer we are 920 // syncing with, and fill in the missing headers using anyone else. Headers from 921 // other peers are only accepted if they map cleanly to the skeleton. If no one 922 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 923 // the origin is dropped. 924 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64, pivot uint64) error { 925 p.log.Debug("Directing header downloads", "origin", from) 926 defer p.log.Debug("Header download terminated") 927 928 // Create a timeout timer, and the associated header fetcher 929 skeleton := true // Skeleton assembly phase or finishing up 930 request := time.Now() // time of the last skeleton fetch request 931 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 932 <-timeout.C // timeout channel should be initially empty 933 defer timeout.Stop() 934 935 var ttl time.Duration 936 getHeaders := func(from uint64) { 937 request = time.Now() 938 939 ttl = d.requestTTL() 940 timeout.Reset(ttl) 941 942 if skeleton { 943 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 944 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 945 } else { 946 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 947 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 948 } 949 } 950 // Start pulling the header chain skeleton until all is done 951 ancestor := from 952 getHeaders(from) 953 954 for { 955 select { 956 case <-d.cancelCh: 957 return errCanceled 958 959 case packet := <-d.headerCh: 960 // Make sure the active peer is giving us the skeleton headers 961 if packet.PeerId() != p.id { 962 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 963 break 964 } 965 headerReqTimer.UpdateSince(request) 966 timeout.Stop() 967 968 // If the skeleton's finished, pull any remaining head headers directly from the origin 969 if packet.Items() == 0 && skeleton { 970 skeleton = false 971 getHeaders(from) 972 continue 973 } 974 // If no more headers are inbound, notify the content fetchers and return 975 if packet.Items() == 0 { 976 // Don't abort header fetches while the pivot is downloading 977 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 978 p.log.Debug("No headers, waiting for pivot commit") 979 select { 980 case <-time.After(fsHeaderContCheck): 981 getHeaders(from) 982 continue 983 case <-d.cancelCh: 984 return errCanceled 985 } 986 } 987 // Pivot done (or not in fast sync) and no more headers, terminate the process 988 p.log.Debug("No more headers available") 989 select { 990 case d.headerProcCh <- nil: 991 return nil 992 case <-d.cancelCh: 993 return errCanceled 994 } 995 } 996 headers := packet.(*headerPack).headers 997 998 // If we received a skeleton batch, resolve internals concurrently 999 if skeleton { 1000 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1001 if err != nil { 1002 p.log.Debug("Skeleton chain invalid", "err", err) 1003 return errInvalidChain 1004 } 1005 headers = filled[proced:] 1006 from += uint64(proced) 1007 } else { 1008 // If we're closing in on the chain head, but haven't yet reached it, delay 1009 // the last few headers so mini reorgs on the head don't cause invalid hash 1010 // chain errors. 1011 if n := len(headers); n > 0 { 1012 // Retrieve the current head we're at 1013 var head uint64 1014 if d.mode == LightSync { 1015 head = d.lightchain.CurrentHeader().Number.Uint64() 1016 } else { 1017 head = d.blockchain.CurrentFastBlock().NumberU64() 1018 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1019 head = full 1020 } 1021 } 1022 // If the head is below the common ancestor, we're actually deduplicating 1023 // already existing chain segments, so use the ancestor as the fake head. 1024 // Otherwise we might end up delaying header deliveries pointlessly. 1025 if head < ancestor { 1026 head = ancestor 1027 } 1028 // If the head is way older than this batch, delay the last few headers 1029 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1030 delay := reorgProtHeaderDelay 1031 if delay > n { 1032 delay = n 1033 } 1034 headers = headers[:n-delay] 1035 } 1036 } 1037 } 1038 // Insert all the new headers and fetch the next batch 1039 if len(headers) > 0 { 1040 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1041 select { 1042 case d.headerProcCh <- headers: 1043 case <-d.cancelCh: 1044 return errCanceled 1045 } 1046 from += uint64(len(headers)) 1047 getHeaders(from) 1048 } else { 1049 // No headers delivered, or all of them being delayed, sleep a bit and retry 1050 p.log.Trace("All headers delayed, waiting") 1051 select { 1052 case <-time.After(fsHeaderContCheck): 1053 getHeaders(from) 1054 continue 1055 case <-d.cancelCh: 1056 return errCanceled 1057 } 1058 } 1059 1060 case <-timeout.C: 1061 if d.dropPeer == nil { 1062 // The dropPeer method is nil when `--copydb` is used for a local copy. 1063 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1064 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1065 break 1066 } 1067 // Header retrieval timed out, consider the peer bad and drop 1068 p.log.Debug("Header request timed out", "elapsed", ttl) 1069 headerTimeoutMeter.Mark(1) 1070 d.dropPeer(p.id) 1071 1072 // Finish the sync gracefully instead of dumping the gathered data though 1073 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1074 select { 1075 case ch <- false: 1076 case <-d.cancelCh: 1077 } 1078 } 1079 select { 1080 case d.headerProcCh <- nil: 1081 case <-d.cancelCh: 1082 } 1083 return errBadPeer 1084 } 1085 } 1086 } 1087 1088 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1089 // and maps them to the provided skeleton header chain. 1090 // 1091 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1092 // immediately to the header processor to keep the rest of the pipeline full even 1093 // in the case of header stalls. 1094 // 1095 // The method returns the entire filled skeleton and also the number of headers 1096 // already forwarded for processing. 1097 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1098 log.Debug("Filling up skeleton", "from", from) 1099 d.queue.ScheduleSkeleton(from, skeleton) 1100 1101 var ( 1102 deliver = func(packet dataPack) (int, error) { 1103 pack := packet.(*headerPack) 1104 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1105 } 1106 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 1107 throttle = func() bool { return false } 1108 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, error) { 1109 return d.queue.ReserveHeaders(p, count), false, nil 1110 } 1111 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1112 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 1113 setIdle = func(p *peerConnection, accepted int) { p.SetHeadersIdle(accepted) } 1114 ) 1115 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1116 d.queue.PendingHeaders, d.queue.InFlightHeaders, throttle, reserve, 1117 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1118 1119 log.Debug("Skeleton fill terminated", "err", err) 1120 1121 filled, proced := d.queue.RetrieveHeaders() 1122 return filled, proced, err 1123 } 1124 1125 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1126 // available peers, reserving a chunk of blocks for each, waiting for delivery 1127 // and also periodically checking for timeouts. 1128 func (d *Downloader) fetchBodies(from uint64) error { 1129 log.Debug("Downloading block bodies", "origin", from) 1130 1131 var ( 1132 deliver = func(packet dataPack) (int, error) { 1133 pack := packet.(*bodyPack) 1134 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1135 } 1136 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 1137 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1138 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 1139 setIdle = func(p *peerConnection, accepted int) { p.SetBodiesIdle(accepted) } 1140 ) 1141 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1142 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ShouldThrottleBlocks, d.queue.ReserveBodies, 1143 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1144 1145 log.Debug("Block body download terminated", "err", err) 1146 return err 1147 } 1148 1149 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1150 // available peers, reserving a chunk of receipts for each, waiting for delivery 1151 // and also periodically checking for timeouts. 1152 func (d *Downloader) fetchReceipts(from uint64) error { 1153 log.Debug("Downloading transaction receipts", "origin", from) 1154 1155 var ( 1156 deliver = func(packet dataPack) (int, error) { 1157 pack := packet.(*receiptPack) 1158 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1159 } 1160 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1161 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1162 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1163 setIdle = func(p *peerConnection, accepted int) { p.SetReceiptsIdle(accepted) } 1164 ) 1165 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1166 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ShouldThrottleReceipts, d.queue.ReserveReceipts, 1167 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1168 1169 log.Debug("Transaction receipt download terminated", "err", err) 1170 return err 1171 } 1172 1173 // fetchParts iteratively downloads scheduled block parts, taking any available 1174 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1175 // also periodically checking for timeouts. 1176 // 1177 // As the scheduling/timeout logic mostly is the same for all downloaded data 1178 // types, this method is used by each for data gathering and is instrumented with 1179 // various callbacks to handle the slight differences between processing them. 1180 // 1181 // The instrumentation parameters: 1182 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1183 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1184 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1185 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1186 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1187 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1188 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1189 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1190 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1191 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1192 // - fetch: network callback to actually send a particular download request to a physical remote peer 1193 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1194 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1195 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1196 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1197 // - kind: textual label of the type being downloaded to display in log mesages 1198 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1199 expire func() map[string]int, pending func() int, inFlight func() bool, throttle func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, error), 1200 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1201 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int), kind string) error { 1202 1203 // Create a ticker to detect expired retrieval tasks 1204 ticker := time.NewTicker(100 * time.Millisecond) 1205 defer ticker.Stop() 1206 1207 update := make(chan struct{}, 1) 1208 1209 // Prepare the queue and fetch block parts until the block header fetcher's done 1210 finished := false 1211 for { 1212 select { 1213 case <-d.cancelCh: 1214 return errCanceled 1215 1216 case packet := <-deliveryCh: 1217 // If the peer was previously banned and failed to deliver its pack 1218 // in a reasonable time frame, ignore its message. 1219 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1220 // Deliver the received chunk of data and check chain validity 1221 accepted, err := deliver(packet) 1222 if err == errInvalidChain { 1223 return err 1224 } 1225 // Unless a peer delivered something completely else than requested (usually 1226 // caused by a timed out request which came through in the end), set it to 1227 // idle. If the delivery's stale, the peer should have already been idled. 1228 if err != errStaleDelivery { 1229 setIdle(peer, accepted) 1230 } 1231 // Issue a log to the user to see what's going on 1232 switch { 1233 case err == nil && packet.Items() == 0: 1234 peer.log.Trace("Requested data not delivered", "type", kind) 1235 case err == nil: 1236 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1237 default: 1238 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1239 } 1240 } 1241 // Blocks assembled, try to update the progress 1242 select { 1243 case update <- struct{}{}: 1244 default: 1245 } 1246 1247 case cont := <-wakeCh: 1248 // The header fetcher sent a continuation flag, check if it's done 1249 if !cont { 1250 finished = true 1251 } 1252 // Headers arrive, try to update the progress 1253 select { 1254 case update <- struct{}{}: 1255 default: 1256 } 1257 1258 case <-ticker.C: 1259 // Sanity check update the progress 1260 select { 1261 case update <- struct{}{}: 1262 default: 1263 } 1264 1265 case <-update: 1266 // Short circuit if we lost all our peers 1267 if d.peers.Len() == 0 { 1268 return errNoPeers 1269 } 1270 // Check for fetch request timeouts and demote the responsible peers 1271 for pid, fails := range expire() { 1272 if peer := d.peers.Peer(pid); peer != nil { 1273 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1274 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1275 // out that sync wise we need to get rid of the peer. 1276 // 1277 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1278 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1279 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1280 if fails > 2 { 1281 peer.log.Trace("Data delivery timed out", "type", kind) 1282 setIdle(peer, 0) 1283 } else { 1284 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1285 1286 if d.dropPeer == nil { 1287 // The dropPeer method is nil when `--copydb` is used for a local copy. 1288 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1289 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1290 } else { 1291 d.dropPeer(pid) 1292 1293 // If this peer was the master peer, abort sync immediately 1294 d.cancelLock.RLock() 1295 master := pid == d.cancelPeer 1296 d.cancelLock.RUnlock() 1297 1298 if master { 1299 d.cancel() 1300 return errTimeout 1301 } 1302 } 1303 } 1304 } 1305 } 1306 // If there's nothing more to fetch, wait or terminate 1307 if pending() == 0 { 1308 if !inFlight() && finished { 1309 log.Debug("Data fetching completed", "type", kind) 1310 return nil 1311 } 1312 break 1313 } 1314 // Send a download request to all idle peers, until throttled 1315 progressed, throttled, running := false, false, inFlight() 1316 idles, total := idle() 1317 1318 for _, peer := range idles { 1319 // Short circuit if throttling activated 1320 if throttle() { 1321 throttled = true 1322 break 1323 } 1324 // Short circuit if there is no more available task. 1325 if pending() == 0 { 1326 break 1327 } 1328 // Reserve a chunk of fetches for a peer. A nil can mean either that 1329 // no more headers are available, or that the peer is known not to 1330 // have them. 1331 request, progress, err := reserve(peer, capacity(peer)) 1332 if err != nil { 1333 return err 1334 } 1335 if progress { 1336 progressed = true 1337 } 1338 if request == nil { 1339 continue 1340 } 1341 if request.From > 0 { 1342 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1343 } else { 1344 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1345 } 1346 // Fetch the chunk and make sure any errors return the hashes to the queue 1347 if fetchHook != nil { 1348 fetchHook(request.Headers) 1349 } 1350 if err := fetch(peer, request); err != nil { 1351 // Although we could try and make an attempt to fix this, this error really 1352 // means that we've double allocated a fetch task to a peer. If that is the 1353 // case, the internal state of the downloader and the queue is very wrong so 1354 // better hard crash and note the error instead of silently accumulating into 1355 // a much bigger issue. 1356 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1357 } 1358 running = true 1359 } 1360 // Make sure that we have peers available for fetching. If all peers have been tried 1361 // and all failed throw an error 1362 if !progressed && !throttled && !running && len(idles) == total && pending() > 0 { 1363 return errPeersUnavailable 1364 } 1365 } 1366 } 1367 } 1368 1369 // processHeaders takes batches of retrieved headers from an input channel and 1370 // keeps processing and scheduling them into the header chain and downloader's 1371 // queue until the stream ends or a failure occurs. 1372 func (d *Downloader) processHeaders(origin uint64, pivot uint64, td *big.Int) error { 1373 // Keep a count of uncertain headers to roll back 1374 var rollback []*types.Header 1375 defer func() { 1376 if len(rollback) > 0 { 1377 // Flatten the headers and roll them back 1378 hashes := make([]common.Hash, len(rollback)) 1379 for i, header := range rollback { 1380 hashes[i] = header.Hash() 1381 } 1382 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1383 if d.mode != LightSync { 1384 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1385 lastBlock = d.blockchain.CurrentBlock().Number() 1386 } 1387 d.lightchain.Rollback(hashes) 1388 curFastBlock, curBlock := common.Big0, common.Big0 1389 if d.mode != LightSync { 1390 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1391 curBlock = d.blockchain.CurrentBlock().Number() 1392 } 1393 log.Warn("Rolled back headers", "count", len(hashes), 1394 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1395 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1396 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock)) 1397 } 1398 }() 1399 1400 // Wait for batches of headers to process 1401 gotHeaders := false 1402 1403 for { 1404 select { 1405 case <-d.cancelCh: 1406 return errCanceled 1407 1408 case headers := <-d.headerProcCh: 1409 // Terminate header processing if we synced up 1410 if len(headers) == 0 { 1411 // Notify everyone that headers are fully processed 1412 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1413 select { 1414 case ch <- false: 1415 case <-d.cancelCh: 1416 } 1417 } 1418 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1419 // better chain compared to ours. The only exception is if its promised blocks were 1420 // already imported by other means (e.g. fetcher): 1421 // 1422 // R <remote peer>, L <local node>: Both at block 10 1423 // R: Mine block 11, and propagate it to L 1424 // L: Queue block 11 for import 1425 // L: Notice that R's head and TD increased compared to ours, start sync 1426 // L: Import of block 11 finishes 1427 // L: Sync begins, and finds common ancestor at 11 1428 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1429 // R: Nothing to give 1430 if d.mode != LightSync { 1431 head := d.blockchain.CurrentBlock() 1432 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1433 return errStallingPeer 1434 } 1435 } 1436 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1437 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1438 // of delivering the post-pivot blocks that would flag the invalid content. 1439 // 1440 // This check cannot be executed "as is" for full imports, since blocks may still be 1441 // queued for processing when the header download completes. However, as long as the 1442 // peer gave us something useful, we're already happy/progressed (above check). 1443 if d.mode == FastSync || d.mode == LightSync { 1444 head := d.lightchain.CurrentHeader() 1445 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1446 return errStallingPeer 1447 } 1448 } 1449 // Disable any rollback and return 1450 rollback = nil 1451 return nil 1452 } 1453 // Otherwise split the chunk of headers into batches and process them 1454 gotHeaders = true 1455 for len(headers) > 0 { 1456 // Terminate if something failed in between processing chunks 1457 select { 1458 case <-d.cancelCh: 1459 return errCanceled 1460 default: 1461 } 1462 // Select the next chunk of headers to import 1463 limit := maxHeadersProcess 1464 if limit > len(headers) { 1465 limit = len(headers) 1466 } 1467 chunk := headers[:limit] 1468 // In case of header only syncing, validate the chunk immediately 1469 if d.mode == FastSync || d.mode == LightSync { 1470 // Collect the yet unknown headers to mark them as uncertain 1471 unknown := make([]*types.Header, 0, len(chunk)) 1472 for _, header := range chunk { 1473 if !d.lightchain.HasHeader(header.Hash(), header.Number.Uint64()) { 1474 unknown = append(unknown, header) 1475 } 1476 } 1477 // If we're importing pure headers, verify based on their recentness 1478 frequency := fsHeaderCheckFrequency 1479 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1480 frequency = 1 1481 } 1482 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1483 // If some headers were inserted, add them too to the rollback list 1484 if n > 0 { 1485 rollback = append(rollback, chunk[:n]...) 1486 } 1487 log.Debug("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "err", err) 1488 return errInvalidChain 1489 } 1490 // All verifications passed, store newly found uncertain headers 1491 rollback = append(rollback, unknown...) 1492 if len(rollback) > fsHeaderSafetyNet { 1493 rollback = append(rollback[:0], rollback[len(rollback)-fsHeaderSafetyNet:]...) 1494 } 1495 } 1496 // Unless we're doing light chains, schedule the headers for associated content retrieval 1497 if d.mode == FullSync || d.mode == FastSync || d.mode == BoundedFullSync { 1498 // If we've reached the allowed number of pending headers, stall a bit 1499 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1500 select { 1501 case <-d.cancelCh: 1502 return errCanceled 1503 case <-time.After(time.Second): 1504 } 1505 } 1506 // Otherwise insert the headers for content retrieval 1507 inserts := d.queue.Schedule(chunk, origin) 1508 if len(inserts) != len(chunk) { 1509 log.Debug("Stale headers") 1510 return errBadPeer 1511 } 1512 } 1513 headers = headers[limit:] 1514 origin += uint64(limit) 1515 } 1516 // Update the highest block number we know if a higher one is found. 1517 d.syncStatsLock.Lock() 1518 if d.syncStatsChainHeight < origin { 1519 d.syncStatsChainHeight = origin - 1 1520 } 1521 d.syncStatsLock.Unlock() 1522 1523 // Signal the content downloaders of the availablility of new tasks 1524 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1525 select { 1526 case ch <- true: 1527 default: 1528 } 1529 } 1530 } 1531 } 1532 } 1533 1534 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1535 func (d *Downloader) processFullSyncContent() error { 1536 for { 1537 results := d.queue.Results(true) 1538 if len(results) == 0 { 1539 return nil 1540 } 1541 if d.chainInsertHook != nil { 1542 d.chainInsertHook(results) 1543 } 1544 if err := d.importBlockResults(results); err != nil { 1545 return err 1546 } 1547 } 1548 } 1549 1550 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1551 // Check for any early termination requests 1552 if len(results) == 0 { 1553 return nil 1554 } 1555 select { 1556 case <-d.quitCh: 1557 return errCancelContentProcessing 1558 default: 1559 } 1560 // Retrieve the a batch of results to import 1561 first, last := results[0].Header, results[len(results)-1].Header 1562 log.Debug("Inserting downloaded chain", "items", len(results), 1563 "firstnum", first.Number, "firsthash", first.Hash(), 1564 "lastnum", last.Number, "lasthash", last.Hash(), 1565 ) 1566 blocks := make([]*types.Block, len(results)) 1567 for i, result := range results { 1568 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1569 } 1570 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1571 if index < len(results) { 1572 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1573 } else { 1574 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1575 // when it needs to preprocess blocks to import a sidechain. 1576 // The importer will put together a new list of blocks to import, which is a superset 1577 // of the blocks delivered from the downloader, and the indexing will be off. 1578 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1579 } 1580 return errInvalidChain 1581 } 1582 return nil 1583 } 1584 1585 // processFastSyncContent takes fetch results from the queue and writes them to the 1586 // database. It also controls the synchronisation of state nodes of the pivot block. 1587 func (d *Downloader) processFastSyncContent(latest *types.Header) error { 1588 // Start syncing state of the reported head block. This should get us most of 1589 // the state of the pivot block. 1590 sync := d.syncState(latest.Root) 1591 defer sync.Cancel() 1592 closeOnErr := func(s *stateSync) { 1593 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled { 1594 d.queue.Close() // wake up Results 1595 } 1596 } 1597 go closeOnErr(sync) 1598 // Figure out the ideal pivot block. Note, that this goalpost may move if the 1599 // sync takes long enough for the chain head to move significantly. 1600 pivot := uint64(0) 1601 if height := latest.Number.Uint64(); height > uint64(fsMinFullBlocks) { 1602 pivot = height - uint64(fsMinFullBlocks) 1603 } 1604 // To cater for moving pivot points, track the pivot block and subsequently 1605 // accumulated download results separately. 1606 var ( 1607 oldPivot *fetchResult // Locked in pivot block, might change eventually 1608 oldTail []*fetchResult // Downloaded content after the pivot 1609 ) 1610 for { 1611 // Wait for the next batch of downloaded data to be available, and if the pivot 1612 // block became stale, move the goalpost 1613 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1614 if len(results) == 0 { 1615 // If pivot sync is done, stop 1616 if oldPivot == nil { 1617 return sync.Cancel() 1618 } 1619 // If sync failed, stop 1620 select { 1621 case <-d.cancelCh: 1622 sync.Cancel() 1623 return errCanceled 1624 default: 1625 } 1626 } 1627 if d.chainInsertHook != nil { 1628 d.chainInsertHook(results) 1629 } 1630 if oldPivot != nil { 1631 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1632 } 1633 // Split around the pivot block and process the two sides via fast/full sync 1634 if atomic.LoadInt32(&d.committed) == 0 { 1635 latest = results[len(results)-1].Header 1636 if height := latest.Number.Uint64(); height > pivot+2*uint64(fsMinFullBlocks) { 1637 log.Warn("Pivot became stale, moving", "old", pivot, "new", height-uint64(fsMinFullBlocks)) 1638 pivot = height - uint64(fsMinFullBlocks) 1639 } 1640 } 1641 P, beforeP, afterP := splitAroundPivot(pivot, results) 1642 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1643 return err 1644 } 1645 if P != nil { 1646 // If new pivot block found, cancel old state retrieval and restart 1647 if oldPivot != P { 1648 sync.Cancel() 1649 1650 sync = d.syncState(P.Header.Root) 1651 defer sync.Cancel() 1652 go closeOnErr(sync) 1653 oldPivot = P 1654 } 1655 // Wait for completion, occasionally checking for pivot staleness 1656 select { 1657 case <-sync.done: 1658 if sync.err != nil { 1659 return sync.err 1660 } 1661 if err := d.commitPivotBlock(P); err != nil { 1662 return err 1663 } 1664 oldPivot = nil 1665 1666 case <-time.After(time.Second): 1667 oldTail = afterP 1668 continue 1669 } 1670 } 1671 // Fast sync done, pivot commit done, full import 1672 if err := d.importBlockResults(afterP); err != nil { 1673 return err 1674 } 1675 } 1676 } 1677 1678 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1679 for _, result := range results { 1680 num := result.Header.Number.Uint64() 1681 switch { 1682 case num < pivot: 1683 before = append(before, result) 1684 case num == pivot: 1685 p = result 1686 default: 1687 after = append(after, result) 1688 } 1689 } 1690 return p, before, after 1691 } 1692 1693 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1694 // Check for any early termination requests 1695 if len(results) == 0 { 1696 return nil 1697 } 1698 select { 1699 case <-d.quitCh: 1700 return errCancelContentProcessing 1701 case <-stateSync.done: 1702 if err := stateSync.Wait(); err != nil { 1703 return err 1704 } 1705 default: 1706 } 1707 // Retrieve the a batch of results to import 1708 first, last := results[0].Header, results[len(results)-1].Header 1709 log.Debug("Inserting fast-sync blocks", "items", len(results), 1710 "firstnum", first.Number, "firsthash", first.Hash(), 1711 "lastnumn", last.Number, "lasthash", last.Hash(), 1712 ) 1713 blocks := make([]*types.Block, len(results)) 1714 receipts := make([]types.Receipts, len(results)) 1715 for i, result := range results { 1716 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1717 receipts[i] = result.Receipts 1718 } 1719 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1720 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1721 return errInvalidChain 1722 } 1723 return nil 1724 } 1725 1726 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1727 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1728 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1729 1730 // Commit the pivot block as the new head, will require full sync from here on 1731 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1732 return err 1733 } 1734 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1735 return err 1736 } 1737 atomic.StoreInt32(&d.committed, 1) 1738 1739 // If we had a bloom filter for the state sync, deallocate it now. Note, we only 1740 // deallocate internally, but keep the empty wrapper. This ensures that if we do 1741 // a rollback after committing the pivot and restarting fast sync, we don't end 1742 // up using a nil bloom. Empty bloom is fine, it just returns that it does not 1743 // have the info we need, so reach down to the database instead. 1744 if d.stateBloom != nil { 1745 d.stateBloom.Close() 1746 } 1747 return nil 1748 } 1749 1750 // DeliverHeaders injects a new batch of block headers received from a remote 1751 // node into the download schedule. 1752 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) { 1753 return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1754 } 1755 1756 // DeliverBodies injects a new batch of block bodies received from a remote node. 1757 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) { 1758 return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1759 } 1760 1761 // DeliverReceipts injects a new batch of receipts received from a remote node. 1762 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) { 1763 return d.deliver(id, d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1764 } 1765 1766 // DeliverNodeData injects a new batch of node state data received from a remote node. 1767 func (d *Downloader) DeliverNodeData(id string, data [][]byte) (err error) { 1768 return d.deliver(id, d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1769 } 1770 1771 // deliver injects a new batch of data received from a remote node. 1772 func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1773 // Update the delivery metrics for both good and failed deliveries 1774 inMeter.Mark(int64(packet.Items())) 1775 defer func() { 1776 if err != nil { 1777 dropMeter.Mark(int64(packet.Items())) 1778 } 1779 }() 1780 // Deliver or abort if the sync is canceled while queuing 1781 d.cancelLock.RLock() 1782 cancel := d.cancelCh 1783 d.cancelLock.RUnlock() 1784 if cancel == nil { 1785 return errNoSyncActive 1786 } 1787 select { 1788 case destCh <- packet: 1789 return nil 1790 case <-cancel: 1791 return errNoSyncActive 1792 } 1793 } 1794 1795 // qosTuner is the quality of service tuning loop that occasionally gathers the 1796 // peer latency statistics and updates the estimated request round trip time. 1797 func (d *Downloader) qosTuner() { 1798 for { 1799 // Retrieve the current median RTT and integrate into the previoust target RTT 1800 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 1801 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 1802 1803 // A new RTT cycle passed, increase our confidence in the estimated RTT 1804 conf := atomic.LoadUint64(&d.rttConfidence) 1805 conf = conf + (1000000-conf)/2 1806 atomic.StoreUint64(&d.rttConfidence, conf) 1807 1808 // Log the new QoS values and sleep until the next RTT 1809 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1810 select { 1811 case <-d.quitCh: 1812 return 1813 case <-time.After(rtt): 1814 } 1815 } 1816 } 1817 1818 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 1819 // peer set, needing to reduce the confidence we have in out QoS estimates. 1820 func (d *Downloader) qosReduceConfidence() { 1821 // If we have a single peer, confidence is always 1 1822 peers := uint64(d.peers.Len()) 1823 if peers == 0 { 1824 // Ensure peer connectivity races don't catch us off guard 1825 return 1826 } 1827 if peers == 1 { 1828 atomic.StoreUint64(&d.rttConfidence, 1000000) 1829 return 1830 } 1831 // If we have a ton of peers, don't drop confidence) 1832 if peers >= uint64(qosConfidenceCap) { 1833 return 1834 } 1835 // Otherwise drop the confidence factor 1836 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 1837 if float64(conf)/1000000 < rttMinConfidence { 1838 conf = uint64(rttMinConfidence * 1000000) 1839 } 1840 atomic.StoreUint64(&d.rttConfidence, conf) 1841 1842 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1843 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1844 } 1845 1846 // requestRTT returns the current target round trip time for a download request 1847 // to complete in. 1848 // 1849 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 1850 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 1851 // be adapted to, but smaller ones are preferred (stabler download stream). 1852 func (d *Downloader) requestRTT() time.Duration { 1853 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 1854 } 1855 1856 // requestTTL returns the current timeout allowance for a single download request 1857 // to finish under. 1858 func (d *Downloader) requestTTL() time.Duration { 1859 var ( 1860 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 1861 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 1862 ) 1863 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 1864 if ttl > ttlLimit { 1865 ttl = ttlLimit 1866 } 1867 return ttl 1868 } 1869 1870 // Extra downloader functionality for non-proof-of-work consensus 1871 1872 // Synchronizes with a peer, but only up to the provided Hash 1873 func (d *Downloader) syncWithPeerUntil(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 1874 d.mux.Post(StartEvent{}) 1875 defer func() { 1876 // reset on error 1877 if err != nil { 1878 d.mux.Post(FailedEvent{err}) 1879 } else { 1880 // Raft syncWithPeerUntil never use the latest field in DoneEvent 1881 // therefore post empty DoneEvent only 1882 d.mux.Post(DoneEvent{}) 1883 } 1884 }() 1885 if p.version < 62 { 1886 return errTooOld 1887 } 1888 1889 log.Info("Synchronising with the network", "id", p.id, "version", p.version) 1890 defer func(start time.Time) { 1891 log.Info("Synchronisation terminated", "duration", time.Since(start)) 1892 }(time.Now()) 1893 1894 frozen, _ := d.stateDB.Ancients() 1895 localHeight := d.blockchain.CurrentBlock().NumberU64() 1896 1897 // check if recovering state db and only ancient db is present 1898 // in this case its possible that local hash is not the latest 1899 // as per raft wal. change header to remote header 1900 var remoteHeader *types.Header 1901 if localHeight == 0 && frozen > 0 { 1902 // statedb was removed and is being recovered now 1903 // we trust the peer height and sync upto that 1904 remoteHeader, err = d.fetchHeight(p) 1905 } else { 1906 remoteHeader, err = d.fetchHeader(p, hash) 1907 } 1908 if err != nil { 1909 return err 1910 } 1911 1912 remoteHeight := remoteHeader.Number.Uint64() 1913 1914 d.syncStatsLock.Lock() 1915 if d.syncStatsChainHeight <= localHeight || d.syncStatsChainOrigin > localHeight { 1916 d.syncStatsChainOrigin = localHeight 1917 } 1918 d.syncStatsChainHeight = remoteHeight 1919 d.syncStatsLock.Unlock() 1920 1921 d.queue.Prepare(localHeight+1, d.mode) 1922 if d.syncInitHook != nil { 1923 d.syncInitHook(localHeight, remoteHeight) 1924 } 1925 1926 pivot := uint64(0) 1927 1928 fetchers := []func() error{ 1929 func() error { return d.fetchBoundedHeaders(p, localHeight+1, remoteHeight) }, 1930 func() error { return d.fetchBodies(localHeight + 1) }, 1931 func() error { return d.fetchReceipts(localHeight + 1) }, // Receipts are only retrieved during fast sync 1932 func() error { return d.processHeaders(localHeight+1, pivot, td) }, 1933 d.processFullSyncContent, //This must be added to clear the buffer of downloaded content as it's being filled 1934 } 1935 return d.spawnSync(fetchers) 1936 } 1937 1938 // Fetches a single header from a peer 1939 func (d *Downloader) fetchHeader(p *peerConnection, hash common.Hash) (*types.Header, error) { 1940 log.Info("retrieving remote chain height", "peer", p) 1941 1942 go p.peer.RequestHeadersByHash(hash, 1, 0, false) 1943 1944 timeout := time.After(d.requestTTL()) 1945 for { 1946 select { 1947 case <-d.cancelCh: 1948 return nil, errCancelBlockFetch 1949 1950 case packet := <-d.headerCh: 1951 // Discard anything not from the origin peer 1952 if packet.PeerId() != p.id { 1953 log.Info("Received headers from incorrect peer", "peer id", packet.PeerId()) 1954 break 1955 } 1956 // Make sure the peer actually gave something valid 1957 headers := packet.(*headerPack).headers 1958 if len(headers) != 1 { 1959 log.Info("invalid number of head headers (!= 1)", "peer", p, "len(headers)", len(headers)) 1960 return nil, errBadPeer 1961 } 1962 return headers[0], nil 1963 1964 case <-timeout: 1965 log.Info("head header timeout", "peer", p) 1966 return nil, errTimeout 1967 1968 case <-d.bodyCh: 1969 case <-d.stateCh: 1970 case <-d.receiptCh: 1971 // Out of bounds delivery, ignore 1972 } 1973 } 1974 } 1975 1976 // Not defined in go's stdlib: 1977 func minInt(a, b int) int { 1978 if a < b { 1979 return a 1980 } 1981 return b 1982 } 1983 1984 // Fetches headers between `from` and `to`, inclusive. 1985 // Assumes invariant: from <= to. 1986 func (d *Downloader) fetchBoundedHeaders(p *peerConnection, from uint64, to uint64) error { 1987 log.Info("directing header downloads", "peer", p, "from", from, "to", to) 1988 defer log.Info("header download terminated", "peer", p) 1989 1990 // Create a timeout timer, and the associated header fetcher 1991 skeleton := true // Skeleton assembly phase or finishing up 1992 request := time.Now() // time of the last skeleton fetch request 1993 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 1994 <-timeout.C // timeout channel should be initially empty 1995 defer timeout.Stop() 1996 1997 getHeaders := func(from uint64) { 1998 request = time.Now() 1999 timeout.Reset(d.requestTTL()) 2000 2001 skeletonStart := from + uint64(MaxHeaderFetch) - 1 2002 2003 if skeleton { 2004 if skeletonStart > to { 2005 skeleton = false 2006 } 2007 } 2008 2009 if skeleton { 2010 numSkeletonHeaders := minInt(MaxSkeletonSize, (int(to-from)+1)/MaxHeaderFetch) 2011 log.Trace("fetching skeleton headers", "peer", p, "num skeleton headers", numSkeletonHeaders, "from", from) 2012 go p.peer.RequestHeadersByNumber(skeletonStart, numSkeletonHeaders, MaxHeaderFetch-1, false) 2013 } else { 2014 // There are not enough headers remaining to warrant a skeleton fetch. 2015 // Grab all of the remaining headers. 2016 2017 numHeaders := int(to-from) + 1 2018 log.Trace("fetching full headers", "peer", p, "num headers", numHeaders, "from", from) 2019 go p.peer.RequestHeadersByNumber(from, numHeaders, 0, false) 2020 } 2021 } 2022 // Start pulling the header chain skeleton until all is done 2023 getHeaders(from) 2024 2025 for { 2026 select { 2027 case <-d.cancelCh: 2028 return errCancelHeaderFetch 2029 2030 case packet := <-d.headerCh: 2031 // Make sure the active peer is giving us the skeleton headers 2032 if packet.PeerId() != p.id { 2033 log.Info("Received headers from incorrect peer", "peer id", packet.PeerId()) 2034 break 2035 } 2036 headerReqTimer.UpdateSince(request) 2037 timeout.Stop() 2038 2039 headers := packet.(*headerPack).headers 2040 2041 // If we received a skeleton batch, resolve internals concurrently 2042 if skeleton { 2043 filled, proced, err := d.fillHeaderSkeleton(from, headers) 2044 if err != nil { 2045 log.Debug("skeleton chain invalid", "peer", p, "err", err) 2046 return errInvalidChain 2047 } 2048 headers = filled[proced:] 2049 from += uint64(proced) 2050 } 2051 // Insert all the new headers and fetch the next batch 2052 if len(headers) > 0 { 2053 log.Trace("schedule headers", "peer", p, "num headers", len(headers), "from", from) 2054 select { 2055 case d.headerProcCh <- headers: 2056 case <-d.cancelCh: 2057 return errCancelHeaderFetch 2058 } 2059 from += uint64(len(headers)) 2060 } 2061 2062 if from <= to { 2063 getHeaders(from) 2064 } else { 2065 // Notify the content fetchers that no more headers are inbound and return. 2066 select { 2067 case d.headerProcCh <- nil: 2068 return nil 2069 case <-d.cancelCh: 2070 return errCancelHeaderFetch 2071 } 2072 } 2073 2074 case <-timeout.C: 2075 // Header retrieval timed out, consider the peer bad and drop 2076 log.Info("header request timed out", "peer", p) 2077 headerTimeoutMeter.Mark(1) 2078 d.dropPeer(p.id) 2079 2080 // Finish the sync gracefully instead of dumping the gathered data though 2081 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 2082 select { 2083 case ch <- false: 2084 case <-d.cancelCh: 2085 } 2086 } 2087 select { 2088 case d.headerProcCh <- nil: 2089 case <-d.cancelCh: 2090 } 2091 return errBadPeer 2092 } 2093 } 2094 }