github.com/aswedchain/aswed@v1.0.1/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/aswedchain/aswed" 29 "github.com/aswedchain/aswed/common" 30 "github.com/aswedchain/aswed/core/rawdb" 31 "github.com/aswedchain/aswed/core/types" 32 "github.com/aswedchain/aswed/ethdb" 33 "github.com/aswedchain/aswed/event" 34 "github.com/aswedchain/aswed/log" 35 "github.com/aswedchain/aswed/metrics" 36 "github.com/aswedchain/aswed/params" 37 "github.com/aswedchain/aswed/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 MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request 46 MaxStateFetch = 384 // Amount of node state values to allow fetching per request 47 48 rttMinEstimate = 2 * time.Second // Minimum round-trip time to target for download requests 49 rttMaxEstimate = 20 * time.Second // Maximum round-trip time to target for download requests 50 rttMinConfidence = 0.1 // Worse confidence factor in our estimated RTT value 51 ttlScaling = 3 // Constant scaling factor for RTT -> TTL conversion 52 ttlLimit = time.Minute // Maximum TTL allowance to prevent reaching crazy timeouts 53 54 qosTuningPeers = 5 // Number of peers to tune based on (best peers) 55 qosConfidenceCap = 10 // Number of peers above which not to modify RTT confidence 56 qosTuningImpact = 0.25 // Impact that a new tuning target has on the previous value 57 58 maxQueuedHeaders = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection) 59 maxHeadersProcess = 2048 // Number of header download results to import at once into the chain 60 maxResultsProcess = 2048 // Number of content download results to import at once into the chain 61 fullMaxForkAncestry uint64 = params.FullImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 62 lightMaxForkAncestry uint64 = params.LightImmutabilityThreshold // 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 >= 63)") 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 uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode 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 and contract code 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 // State sync 142 pivotHeader *types.Header // Pivot block header to dynamically push the syncing state root 143 pivotLock sync.RWMutex // Lock protecting pivot header reads from updates 144 145 stateSyncStart chan *stateSync 146 trackStateReq chan *stateReq 147 stateCh chan dataPack // [eth/63] Channel receiving inbound node state data 148 149 // Cancellation and termination 150 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 151 cancelCh chan struct{} // Channel to cancel mid-flight syncs 152 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 153 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 154 155 quitCh chan struct{} // Quit channel to signal termination 156 quitLock sync.Mutex // Lock to prevent double closes 157 158 // Testing hooks 159 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 160 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 161 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 162 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 163 } 164 165 // LightChain encapsulates functions required to synchronise a light chain. 166 type LightChain interface { 167 // HasHeader verifies a header's presence in the local chain. 168 HasHeader(common.Hash, uint64) bool 169 170 // GetHeaderByHash retrieves a header from the local chain. 171 GetHeaderByHash(common.Hash) *types.Header 172 173 // CurrentHeader retrieves the head header from the local chain. 174 CurrentHeader() *types.Header 175 176 // GetTd returns the total difficulty of a local block. 177 GetTd(common.Hash, uint64) *big.Int 178 179 // InsertHeaderChain inserts a batch of headers into the local chain. 180 InsertHeaderChain([]*types.Header, int) (int, error) 181 182 // SetHead rewinds the local chain to a new head. 183 SetHead(uint64) error 184 } 185 186 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 187 type BlockChain interface { 188 LightChain 189 190 // HasBlock verifies a block's presence in the local chain. 191 HasBlock(common.Hash, uint64) bool 192 193 // HasFastBlock verifies a fast block's presence in the local chain. 194 HasFastBlock(common.Hash, uint64) bool 195 196 // GetBlockByHash retrieves a block from the local chain. 197 GetBlockByHash(common.Hash) *types.Block 198 199 // CurrentBlock retrieves the head block from the local chain. 200 CurrentBlock() *types.Block 201 202 // CurrentFastBlock retrieves the head fast block from the local chain. 203 CurrentFastBlock() *types.Block 204 205 // FastSyncCommitHead directly commits the head block to a certain entity. 206 FastSyncCommitHead(common.Hash) error 207 208 // InsertChain inserts a batch of blocks into the local chain. 209 InsertChain(types.Blocks) (int, error) 210 211 // InsertReceiptChain inserts a batch of receipts into the local chain. 212 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 213 } 214 215 // New creates a new downloader to fetch hashes and blocks from remote peers. 216 func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 217 if lightchain == nil { 218 lightchain = chain 219 } 220 dl := &Downloader{ 221 stateDB: stateDb, 222 stateBloom: stateBloom, 223 mux: mux, 224 checkpoint: checkpoint, 225 queue: newQueue(blockCacheMaxItems, blockCacheInitialItems), 226 peers: newPeerSet(), 227 rttEstimate: uint64(rttMaxEstimate), 228 rttConfidence: uint64(1000000), 229 blockchain: chain, 230 lightchain: lightchain, 231 dropPeer: dropPeer, 232 headerCh: make(chan dataPack, 1), 233 bodyCh: make(chan dataPack, 1), 234 receiptCh: make(chan dataPack, 1), 235 bodyWakeCh: make(chan bool, 1), 236 receiptWakeCh: make(chan bool, 1), 237 headerProcCh: make(chan []*types.Header, 1), 238 quitCh: make(chan struct{}), 239 stateCh: make(chan dataPack), 240 stateSyncStart: make(chan *stateSync), 241 syncStatsState: stateSyncStats{ 242 processed: rawdb.ReadFastTrieProgress(stateDb), 243 }, 244 trackStateReq: make(chan *stateReq), 245 } 246 go dl.qosTuner() 247 go dl.stateFetcher() 248 return dl 249 } 250 251 // Progress retrieves the synchronisation boundaries, specifically the origin 252 // block where synchronisation started at (may have failed/suspended); the block 253 // or header sync is currently at; and the latest known block which the sync targets. 254 // 255 // In addition, during the state download phase of fast synchronisation the number 256 // of processed and the total number of known states are also returned. Otherwise 257 // these are zero. 258 func (d *Downloader) Progress() ethereum.SyncProgress { 259 // Lock the current stats and return the progress 260 d.syncStatsLock.RLock() 261 defer d.syncStatsLock.RUnlock() 262 263 current := uint64(0) 264 mode := d.getMode() 265 switch { 266 case d.blockchain != nil && mode == FullSync: 267 current = d.blockchain.CurrentBlock().NumberU64() 268 case d.blockchain != nil && mode == FastSync: 269 current = d.blockchain.CurrentFastBlock().NumberU64() 270 case d.lightchain != nil: 271 current = d.lightchain.CurrentHeader().Number.Uint64() 272 default: 273 log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", mode) 274 } 275 return ethereum.SyncProgress{ 276 StartingBlock: d.syncStatsChainOrigin, 277 CurrentBlock: current, 278 HighestBlock: d.syncStatsChainHeight, 279 PulledStates: d.syncStatsState.processed, 280 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 281 } 282 } 283 284 // Synchronising returns whether the downloader is currently retrieving blocks. 285 func (d *Downloader) Synchronising() bool { 286 return atomic.LoadInt32(&d.synchronising) > 0 287 } 288 289 // SyncBloomContains tests if the syncbloom filter contains the given hash: 290 // - false: the bloom definitely does not contain hash 291 // - true: the bloom maybe contains hash 292 // 293 // While the bloom is being initialized (or is closed), all queries will return true. 294 func (d *Downloader) SyncBloomContains(hash []byte) bool { 295 return d.stateBloom == nil || d.stateBloom.Contains(hash) 296 } 297 298 // RegisterPeer injects a new download peer into the set of block source to be 299 // used for fetching hashes and blocks from. 300 func (d *Downloader) RegisterPeer(id string, version int, peer Peer) error { 301 logger := log.New("peer", id) 302 logger.Trace("Registering sync peer") 303 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 304 logger.Error("Failed to register sync peer", "err", err) 305 return err 306 } 307 d.qosReduceConfidence() 308 309 return nil 310 } 311 312 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 313 func (d *Downloader) RegisterLightPeer(id string, version int, peer LightPeer) error { 314 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 315 } 316 317 // UnregisterPeer remove a peer from the known list, preventing any action from 318 // the specified peer. An effort is also made to return any pending fetches into 319 // the queue. 320 func (d *Downloader) UnregisterPeer(id string) error { 321 // Unregister the peer from the active peer set and revoke any fetch tasks 322 logger := log.New("peer", id) 323 logger.Trace("Unregistering sync peer") 324 if err := d.peers.Unregister(id); err != nil { 325 logger.Error("Failed to unregister sync peer", "err", err) 326 return err 327 } 328 d.queue.Revoke(id) 329 330 return nil 331 } 332 333 // Synchronise tries to sync up our local block chain with a remote peer, both 334 // adding various sanity checks as well as wrapping it with various log entries. 335 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 336 err := d.synchronise(id, head, td, mode) 337 338 switch err { 339 case nil, errBusy, errCanceled: 340 return err 341 } 342 343 if errors.Is(err, errInvalidChain) || errors.Is(err, errBadPeer) || errors.Is(err, errTimeout) || 344 errors.Is(err, errStallingPeer) || errors.Is(err, errUnsyncedPeer) || errors.Is(err, errEmptyHeaderSet) || 345 errors.Is(err, errPeersUnavailable) || errors.Is(err, errTooOld) || errors.Is(err, errInvalidAncestor) { 346 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 347 if d.dropPeer == nil { 348 // The dropPeer method is nil when `--copydb` is used for a local copy. 349 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 350 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 351 } else { 352 d.dropPeer(id) 353 } 354 return err 355 } 356 log.Warn("Synchronisation failed, retrying", "err", err) 357 return err 358 } 359 360 // synchronise will select the peer and use it for synchronising. If an empty string is given 361 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 362 // checks fail an error will be returned. This method is synchronous 363 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 364 // Mock out the synchronisation if testing 365 if d.synchroniseMock != nil { 366 return d.synchroniseMock(id, hash) 367 } 368 // Make sure only one goroutine is ever allowed past this point at once 369 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 370 return errBusy 371 } 372 defer atomic.StoreInt32(&d.synchronising, 0) 373 374 // Post a user notification of the sync (only once per session) 375 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 376 log.Info("Block synchronisation started") 377 } 378 // If we are already full syncing, but have a fast-sync bloom filter laying 379 // around, make sure it doesn't use memory any more. This is a special case 380 // when the user attempts to fast sync a new empty network. 381 if mode == FullSync && d.stateBloom != nil { 382 d.stateBloom.Close() 383 } 384 // Reset the queue, peer set and wake channels to clean any internal leftover state 385 d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems) 386 d.peers.Reset() 387 388 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 389 select { 390 case <-ch: 391 default: 392 } 393 } 394 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 395 for empty := false; !empty; { 396 select { 397 case <-ch: 398 default: 399 empty = true 400 } 401 } 402 } 403 for empty := false; !empty; { 404 select { 405 case <-d.headerProcCh: 406 default: 407 empty = true 408 } 409 } 410 // Create cancel channel for aborting mid-flight and mark the master peer 411 d.cancelLock.Lock() 412 d.cancelCh = make(chan struct{}) 413 d.cancelPeer = id 414 d.cancelLock.Unlock() 415 416 defer d.Cancel() // No matter what, we can't leave the cancel channel open 417 418 // Atomically set the requested sync mode 419 atomic.StoreUint32(&d.mode, uint32(mode)) 420 421 // Retrieve the origin peer and initiate the downloading process 422 p := d.peers.Peer(id) 423 if p == nil { 424 return errUnknownPeer 425 } 426 return d.syncWithPeer(p, hash, td) 427 } 428 429 func (d *Downloader) getMode() SyncMode { 430 return SyncMode(atomic.LoadUint32(&d.mode)) 431 } 432 433 // syncWithPeer starts a block synchronization based on the hash chain from the 434 // specified peer and head hash. 435 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 436 d.mux.Post(StartEvent{}) 437 defer func() { 438 // reset on error 439 if err != nil { 440 d.mux.Post(FailedEvent{err}) 441 } else { 442 latest := d.lightchain.CurrentHeader() 443 d.mux.Post(DoneEvent{latest}) 444 } 445 }() 446 if p.version < 63 { 447 return errTooOld 448 } 449 mode := d.getMode() 450 451 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", mode) 452 defer func(start time.Time) { 453 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 454 }(time.Now()) 455 456 // Look up the sync boundaries: the common ancestor and the target block 457 latest, pivot, err := d.fetchHead(p) 458 if err != nil { 459 return err 460 } 461 if mode == FastSync && pivot == nil { 462 // If no pivot block was returned, the head is below the min full block 463 // threshold (i.e. new chian). In that case we won't really fast sync 464 // anyway, but still need a valid pivot block to avoid some code hitting 465 // nil panics on an access. 466 pivot = d.blockchain.CurrentBlock().Header() 467 } 468 height := latest.Number.Uint64() 469 470 origin, err := d.findAncestor(p, latest) 471 if err != nil { 472 return err 473 } 474 d.syncStatsLock.Lock() 475 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 476 d.syncStatsChainOrigin = origin 477 } 478 d.syncStatsChainHeight = height 479 d.syncStatsLock.Unlock() 480 481 // Ensure our origin point is below any fast sync pivot point 482 if mode == FastSync { 483 if height <= uint64(fsMinFullBlocks) { 484 origin = 0 485 } else { 486 pivotNumber := pivot.Number.Uint64() 487 if pivotNumber <= origin { 488 origin = pivotNumber - 1 489 } 490 // Write out the pivot into the database so a rollback beyond it will 491 // reenable fast sync 492 rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber) 493 } 494 } 495 d.committed = 1 496 if mode == FastSync && pivot.Number.Uint64() != 0 { 497 d.committed = 0 498 } 499 if mode == FastSync { 500 // Set the ancient data limitation. 501 // If we are running fast sync, all block data older than ancientLimit will be 502 // written to the ancient store. More recent data will be written to the active 503 // database and will wait for the freezer to migrate. 504 // 505 // If there is a checkpoint available, then calculate the ancientLimit through 506 // that. Otherwise calculate the ancient limit through the advertised height 507 // of the remote peer. 508 // 509 // The reason for picking checkpoint first is that a malicious peer can give us 510 // a fake (very high) height, forcing the ancient limit to also be very high. 511 // The peer would start to feed us valid blocks until head, resulting in all of 512 // the blocks might be written into the ancient store. A following mini-reorg 513 // could cause issues. 514 if d.checkpoint != 0 && d.checkpoint > fullMaxForkAncestry+1 { 515 d.ancientLimit = d.checkpoint 516 } else if height > fullMaxForkAncestry+1 { 517 d.ancientLimit = height - fullMaxForkAncestry - 1 518 } else { 519 d.ancientLimit = 0 520 } 521 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 522 523 // If a part of blockchain data has already been written into active store, 524 // disable the ancient style insertion explicitly. 525 if origin >= frozen && frozen != 0 { 526 d.ancientLimit = 0 527 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 528 } else if d.ancientLimit > 0 { 529 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 530 } 531 // Rewind the ancient store and blockchain if reorg happens. 532 if origin+1 < frozen { 533 if err := d.lightchain.SetHead(origin + 1); err != nil { 534 return err 535 } 536 } 537 } 538 // Initiate the sync using a concurrent header and content retrieval algorithm 539 d.queue.Prepare(origin+1, mode) 540 if d.syncInitHook != nil { 541 d.syncInitHook(origin, height) 542 } 543 fetchers := []func() error{ 544 func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved 545 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 546 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 547 func() error { return d.processHeaders(origin+1, td) }, 548 } 549 if mode == FastSync { 550 d.pivotLock.Lock() 551 d.pivotHeader = pivot 552 d.pivotLock.Unlock() 553 554 fetchers = append(fetchers, func() error { return d.processFastSyncContent() }) 555 } else if mode == FullSync { 556 fetchers = append(fetchers, d.processFullSyncContent) 557 } 558 return d.spawnSync(fetchers) 559 } 560 561 // spawnSync runs d.process and all given fetcher functions to completion in 562 // separate goroutines, returning the first error that appears. 563 func (d *Downloader) spawnSync(fetchers []func() error) error { 564 errc := make(chan error, len(fetchers)) 565 d.cancelWg.Add(len(fetchers)) 566 for _, fn := range fetchers { 567 fn := fn 568 go func() { defer d.cancelWg.Done(); errc <- fn() }() 569 } 570 // Wait for the first error, then terminate the others. 571 var err error 572 for i := 0; i < len(fetchers); i++ { 573 if i == len(fetchers)-1 { 574 // Close the queue when all fetchers have exited. 575 // This will cause the block processor to end when 576 // it has processed the queue. 577 d.queue.Close() 578 } 579 if err = <-errc; err != nil && err != errCanceled { 580 break 581 } 582 } 583 d.queue.Close() 584 d.Cancel() 585 return err 586 } 587 588 // cancel aborts all of the operations and resets the queue. However, cancel does 589 // not wait for the running download goroutines to finish. This method should be 590 // used when cancelling the downloads from inside the downloader. 591 func (d *Downloader) cancel() { 592 // Close the current cancel channel 593 d.cancelLock.Lock() 594 defer d.cancelLock.Unlock() 595 596 if d.cancelCh != nil { 597 select { 598 case <-d.cancelCh: 599 // Channel was already closed 600 default: 601 close(d.cancelCh) 602 } 603 } 604 } 605 606 // Cancel aborts all of the operations and waits for all download goroutines to 607 // finish before returning. 608 func (d *Downloader) Cancel() { 609 d.cancel() 610 d.cancelWg.Wait() 611 } 612 613 // Terminate interrupts the downloader, canceling all pending operations. 614 // The downloader cannot be reused after calling Terminate. 615 func (d *Downloader) Terminate() { 616 // Close the termination channel (make sure double close is allowed) 617 d.quitLock.Lock() 618 select { 619 case <-d.quitCh: 620 default: 621 close(d.quitCh) 622 } 623 if d.stateBloom != nil { 624 d.stateBloom.Close() 625 } 626 d.quitLock.Unlock() 627 628 // Cancel any pending download requests 629 d.Cancel() 630 } 631 632 // fetchHead retrieves the head header and prior pivot block (if available) from 633 // a remote peer. 634 func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *types.Header, err error) { 635 p.log.Debug("Retrieving remote chain head") 636 mode := d.getMode() 637 638 // Request the advertised remote head block and wait for the response 639 latest, _ := p.peer.Head() 640 fetch := 1 641 if mode == FastSync { 642 fetch = 2 // head + pivot headers 643 } 644 go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true) 645 646 ttl := d.requestTTL() 647 timeout := time.After(ttl) 648 for { 649 select { 650 case <-d.cancelCh: 651 return nil, nil, errCanceled 652 653 case packet := <-d.headerCh: 654 // Discard anything not from the origin peer 655 if packet.PeerId() != p.id { 656 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 657 break 658 } 659 // Make sure the peer gave us at least one and at most the requested headers 660 headers := packet.(*headerPack).headers 661 if len(headers) == 0 || len(headers) > fetch { 662 return nil, nil, fmt.Errorf("%w: returned headers %d != requested %d", errBadPeer, len(headers), fetch) 663 } 664 // The first header needs to be the head, validate against the checkpoint 665 // and request. If only 1 header was returned, make sure there's no pivot 666 // or there was not one requested. 667 head := headers[0] 668 if (mode == FastSync || mode == LightSync) && head.Number.Uint64() < d.checkpoint { 669 return nil, nil, fmt.Errorf("%w: remote head %d below checkpoint %d", errUnsyncedPeer, head.Number, d.checkpoint) 670 } 671 if len(headers) == 1 { 672 if mode == FastSync && head.Number.Uint64() > uint64(fsMinFullBlocks) { 673 return nil, nil, fmt.Errorf("%w: no pivot included along head header", errBadPeer) 674 } 675 p.log.Debug("Remote head identified, no pivot", "number", head.Number, "hash", head.Hash()) 676 return head, nil, nil 677 } 678 // At this point we have 2 headers in total and the first is the 679 // validated head of the chian. Check the pivot number and return, 680 pivot := headers[1] 681 if pivot.Number.Uint64() != head.Number.Uint64()-uint64(fsMinFullBlocks) { 682 return nil, nil, fmt.Errorf("%w: remote pivot %d != requested %d", errInvalidChain, pivot.Number, head.Number.Uint64()-uint64(fsMinFullBlocks)) 683 } 684 return head, pivot, nil 685 686 case <-timeout: 687 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 688 return nil, nil, errTimeout 689 690 case <-d.bodyCh: 691 case <-d.receiptCh: 692 // Out of bounds delivery, ignore 693 } 694 } 695 } 696 697 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 698 // common ancestor. 699 // It returns parameters to be used for peer.RequestHeadersByNumber: 700 // from - starting block number 701 // count - number of headers to request 702 // skip - number of headers to skip 703 // and also returns 'max', the last block which is expected to be returned by the remote peers, 704 // given the (from,count,skip) 705 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 706 var ( 707 from int 708 count int 709 MaxCount = MaxHeaderFetch / 16 710 ) 711 // requestHead is the highest block that we will ask for. If requestHead is not offset, 712 // the highest block that we will get is 16 blocks back from head, which means we 713 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 714 // between us and the peer is 1-2 blocks, which is most common 715 requestHead := int(remoteHeight) - 1 716 if requestHead < 0 { 717 requestHead = 0 718 } 719 // requestBottom is the lowest block we want included in the query 720 // Ideally, we want to include the one just below our own head 721 requestBottom := int(localHeight - 1) 722 if requestBottom < 0 { 723 requestBottom = 0 724 } 725 totalSpan := requestHead - requestBottom 726 span := 1 + totalSpan/MaxCount 727 if span < 2 { 728 span = 2 729 } 730 if span > 16 { 731 span = 16 732 } 733 734 count = 1 + totalSpan/span 735 if count > MaxCount { 736 count = MaxCount 737 } 738 if count < 2 { 739 count = 2 740 } 741 from = requestHead - (count-1)*span 742 if from < 0 { 743 from = 0 744 } 745 max := from + (count-1)*span 746 return int64(from), count, span - 1, uint64(max) 747 } 748 749 // findAncestor tries to locate the common ancestor link of the local chain and 750 // a remote peers blockchain. In the general case when our node was in sync and 751 // on the correct chain, checking the top N links should already get us a match. 752 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 753 // the head links match), we do a binary search to find the common ancestor. 754 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 755 // Figure out the valid ancestor range to prevent rewrite attacks 756 var ( 757 floor = int64(-1) 758 localHeight uint64 759 remoteHeight = remoteHeader.Number.Uint64() 760 ) 761 mode := d.getMode() 762 switch mode { 763 case FullSync: 764 localHeight = d.blockchain.CurrentBlock().NumberU64() 765 case FastSync: 766 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 767 default: 768 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 769 } 770 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 771 772 // Recap floor value for binary search 773 maxForkAncestry := fullMaxForkAncestry 774 if d.getMode() == LightSync { 775 maxForkAncestry = lightMaxForkAncestry 776 } 777 if localHeight >= maxForkAncestry { 778 // We're above the max reorg threshold, find the earliest fork point 779 floor = int64(localHeight - maxForkAncestry) 780 } 781 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 782 // all headers before that point will be missing. 783 if mode == LightSync { 784 // If we don't know the current CHT position, find it 785 if d.genesis == 0 { 786 header := d.lightchain.CurrentHeader() 787 for header != nil { 788 d.genesis = header.Number.Uint64() 789 if floor >= int64(d.genesis)-1 { 790 break 791 } 792 header = d.lightchain.GetHeaderByHash(header.ParentHash) 793 } 794 } 795 // We already know the "genesis" block number, cap floor to that 796 if floor < int64(d.genesis)-1 { 797 floor = int64(d.genesis) - 1 798 } 799 } 800 801 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 802 803 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 804 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 805 806 // Wait for the remote response to the head fetch 807 number, hash := uint64(0), common.Hash{} 808 809 ttl := d.requestTTL() 810 timeout := time.After(ttl) 811 812 for finished := false; !finished; { 813 select { 814 case <-d.cancelCh: 815 return 0, errCanceled 816 817 case packet := <-d.headerCh: 818 // Discard anything not from the origin peer 819 if packet.PeerId() != p.id { 820 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 821 break 822 } 823 // Make sure the peer actually gave something valid 824 headers := packet.(*headerPack).headers 825 if len(headers) == 0 { 826 p.log.Warn("Empty head header set") 827 return 0, errEmptyHeaderSet 828 } 829 // Make sure the peer's reply conforms to the request 830 for i, header := range headers { 831 expectNumber := from + int64(i)*int64(skip+1) 832 if number := header.Number.Int64(); number != expectNumber { 833 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 834 return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering")) 835 } 836 } 837 // Check if a common ancestor was found 838 finished = true 839 for i := len(headers) - 1; i >= 0; i-- { 840 // Skip any headers that underflow/overflow our requested set 841 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 842 continue 843 } 844 // Otherwise check if we already know the header or not 845 h := headers[i].Hash() 846 n := headers[i].Number.Uint64() 847 848 var known bool 849 switch mode { 850 case FullSync: 851 known = d.blockchain.HasBlock(h, n) 852 case FastSync: 853 known = d.blockchain.HasFastBlock(h, n) 854 default: 855 known = d.lightchain.HasHeader(h, n) 856 } 857 if known { 858 number, hash = n, h 859 break 860 } 861 } 862 863 case <-timeout: 864 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 865 return 0, errTimeout 866 867 case <-d.bodyCh: 868 case <-d.receiptCh: 869 // Out of bounds delivery, ignore 870 } 871 } 872 // If the head fetch already found an ancestor, return 873 if hash != (common.Hash{}) { 874 if int64(number) <= floor { 875 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 876 return 0, errInvalidAncestor 877 } 878 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 879 return number, nil 880 } 881 // Ancestor not found, we need to binary search over our chain 882 start, end := uint64(0), remoteHeight 883 if floor > 0 { 884 start = uint64(floor) 885 } 886 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 887 888 for start+1 < end { 889 // Split our chain interval in two, and request the hash to cross check 890 check := (start + end) / 2 891 892 ttl := d.requestTTL() 893 timeout := time.After(ttl) 894 895 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 896 897 // Wait until a reply arrives to this request 898 for arrived := false; !arrived; { 899 select { 900 case <-d.cancelCh: 901 return 0, errCanceled 902 903 case packet := <-d.headerCh: 904 // Discard anything not from the origin peer 905 if packet.PeerId() != p.id { 906 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 907 break 908 } 909 // Make sure the peer actually gave something valid 910 headers := packet.(*headerPack).headers 911 if len(headers) != 1 { 912 p.log.Warn("Multiple headers for single request", "headers", len(headers)) 913 return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers)) 914 } 915 arrived = true 916 917 // Modify the search interval based on the response 918 h := headers[0].Hash() 919 n := headers[0].Number.Uint64() 920 921 var known bool 922 switch mode { 923 case FullSync: 924 known = d.blockchain.HasBlock(h, n) 925 case FastSync: 926 known = d.blockchain.HasFastBlock(h, n) 927 default: 928 known = d.lightchain.HasHeader(h, n) 929 } 930 if !known { 931 end = check 932 break 933 } 934 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 935 if header.Number.Uint64() != check { 936 p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 937 return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number) 938 } 939 start = check 940 hash = h 941 942 case <-timeout: 943 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 944 return 0, errTimeout 945 946 case <-d.bodyCh: 947 case <-d.receiptCh: 948 // Out of bounds delivery, ignore 949 } 950 } 951 } 952 // Ensure valid ancestry and return 953 if int64(start) <= floor { 954 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 955 return 0, errInvalidAncestor 956 } 957 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 958 return start, nil 959 } 960 961 // fetchHeaders keeps retrieving headers concurrently from the number 962 // requested, until no more are returned, potentially throttling on the way. To 963 // facilitate concurrency but still protect against malicious nodes sending bad 964 // headers, we construct a header chain skeleton using the "origin" peer we are 965 // syncing with, and fill in the missing headers using anyone else. Headers from 966 // other peers are only accepted if they map cleanly to the skeleton. If no one 967 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 968 // the origin is dropped. 969 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error { 970 p.log.Debug("Directing header downloads", "origin", from) 971 defer p.log.Debug("Header download terminated") 972 973 // Create a timeout timer, and the associated header fetcher 974 skeleton := true // Skeleton assembly phase or finishing up 975 pivoting := false // Whether the next request is pivot verification 976 request := time.Now() // time of the last skeleton fetch request 977 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 978 <-timeout.C // timeout channel should be initially empty 979 defer timeout.Stop() 980 981 var ttl time.Duration 982 getHeaders := func(from uint64) { 983 request = time.Now() 984 985 ttl = d.requestTTL() 986 timeout.Reset(ttl) 987 988 if skeleton { 989 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 990 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 991 } else { 992 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 993 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 994 } 995 } 996 getNextPivot := func() { 997 pivoting = true 998 request = time.Now() 999 1000 ttl = d.requestTTL() 1001 timeout.Reset(ttl) 1002 1003 d.pivotLock.RLock() 1004 pivot := d.pivotHeader.Number.Uint64() 1005 d.pivotLock.RUnlock() 1006 1007 p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks)) 1008 go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep 1009 } 1010 // Start pulling the header chain skeleton until all is done 1011 ancestor := from 1012 getHeaders(from) 1013 1014 mode := d.getMode() 1015 for { 1016 select { 1017 case <-d.cancelCh: 1018 return errCanceled 1019 1020 case packet := <-d.headerCh: 1021 // Make sure the active peer is giving us the skeleton headers 1022 if packet.PeerId() != p.id { 1023 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 1024 break 1025 } 1026 headerReqTimer.UpdateSince(request) 1027 timeout.Stop() 1028 1029 // If the pivot is being checked, move if it became stale and run the real retrieval 1030 var pivot uint64 1031 1032 d.pivotLock.RLock() 1033 if d.pivotHeader != nil { 1034 pivot = d.pivotHeader.Number.Uint64() 1035 } 1036 d.pivotLock.RUnlock() 1037 1038 if pivoting { 1039 if packet.Items() == 2 { 1040 // Retrieve the headers and do some sanity checks, just in case 1041 headers := packet.(*headerPack).headers 1042 1043 if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want { 1044 log.Warn("Peer sent invalid next pivot", "have", have, "want", want) 1045 return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want) 1046 } 1047 if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want { 1048 log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want) 1049 return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want) 1050 } 1051 log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number) 1052 pivot = headers[0].Number.Uint64() 1053 1054 d.pivotLock.Lock() 1055 d.pivotHeader = headers[0] 1056 d.pivotLock.Unlock() 1057 1058 // Write out the pivot into the database so a rollback beyond 1059 // it will reenable fast sync and update the state root that 1060 // the state syncer will be downloading. 1061 rawdb.WriteLastPivotNumber(d.stateDB, pivot) 1062 } 1063 pivoting = false 1064 getHeaders(from) 1065 continue 1066 } 1067 // If the skeleton's finished, pull any remaining head headers directly from the origin 1068 if skeleton && packet.Items() == 0 { 1069 skeleton = false 1070 getHeaders(from) 1071 continue 1072 } 1073 // If no more headers are inbound, notify the content fetchers and return 1074 if packet.Items() == 0 { 1075 // Don't abort header fetches while the pivot is downloading 1076 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 1077 p.log.Debug("No headers, waiting for pivot commit") 1078 select { 1079 case <-time.After(fsHeaderContCheck): 1080 getHeaders(from) 1081 continue 1082 case <-d.cancelCh: 1083 return errCanceled 1084 } 1085 } 1086 // Pivot done (or not in fast sync) and no more headers, terminate the process 1087 p.log.Debug("No more headers available") 1088 select { 1089 case d.headerProcCh <- nil: 1090 return nil 1091 case <-d.cancelCh: 1092 return errCanceled 1093 } 1094 } 1095 headers := packet.(*headerPack).headers 1096 1097 // If we received a skeleton batch, resolve internals concurrently 1098 if skeleton { 1099 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1100 if err != nil { 1101 p.log.Debug("Skeleton chain invalid", "err", err) 1102 return fmt.Errorf("%w: %v", errInvalidChain, err) 1103 } 1104 headers = filled[proced:] 1105 from += uint64(proced) 1106 } else { 1107 // If we're closing in on the chain head, but haven't yet reached it, delay 1108 // the last few headers so mini reorgs on the head don't cause invalid hash 1109 // chain errors. 1110 if n := len(headers); n > 0 { 1111 // Retrieve the current head we're at 1112 var head uint64 1113 if mode == LightSync { 1114 head = d.lightchain.CurrentHeader().Number.Uint64() 1115 } else { 1116 head = d.blockchain.CurrentFastBlock().NumberU64() 1117 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1118 head = full 1119 } 1120 } 1121 // If the head is below the common ancestor, we're actually deduplicating 1122 // already existing chain segments, so use the ancestor as the fake head. 1123 // Otherwise we might end up delaying header deliveries pointlessly. 1124 if head < ancestor { 1125 head = ancestor 1126 } 1127 // If the head is way older than this batch, delay the last few headers 1128 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1129 delay := reorgProtHeaderDelay 1130 if delay > n { 1131 delay = n 1132 } 1133 headers = headers[:n-delay] 1134 } 1135 } 1136 } 1137 // Insert all the new headers and fetch the next batch 1138 if len(headers) > 0 { 1139 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1140 select { 1141 case d.headerProcCh <- headers: 1142 case <-d.cancelCh: 1143 return errCanceled 1144 } 1145 from += uint64(len(headers)) 1146 1147 // If we're still skeleton filling fast sync, check pivot staleness 1148 // before continuing to the next skeleton filling 1149 if skeleton && pivot > 0 { 1150 getNextPivot() 1151 } else { 1152 getHeaders(from) 1153 } 1154 } else { 1155 // No headers delivered, or all of them being delayed, sleep a bit and retry 1156 p.log.Trace("All headers delayed, waiting") 1157 select { 1158 case <-time.After(fsHeaderContCheck): 1159 getHeaders(from) 1160 continue 1161 case <-d.cancelCh: 1162 return errCanceled 1163 } 1164 } 1165 1166 case <-timeout.C: 1167 if d.dropPeer == nil { 1168 // The dropPeer method is nil when `--copydb` is used for a local copy. 1169 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1170 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1171 break 1172 } 1173 // Header retrieval timed out, consider the peer bad and drop 1174 p.log.Debug("Header request timed out", "elapsed", ttl) 1175 headerTimeoutMeter.Mark(1) 1176 d.dropPeer(p.id) 1177 1178 // Finish the sync gracefully instead of dumping the gathered data though 1179 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1180 select { 1181 case ch <- false: 1182 case <-d.cancelCh: 1183 } 1184 } 1185 select { 1186 case d.headerProcCh <- nil: 1187 case <-d.cancelCh: 1188 } 1189 return fmt.Errorf("%w: header request timed out", errBadPeer) 1190 } 1191 } 1192 } 1193 1194 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1195 // and maps them to the provided skeleton header chain. 1196 // 1197 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1198 // immediately to the header processor to keep the rest of the pipeline full even 1199 // in the case of header stalls. 1200 // 1201 // The method returns the entire filled skeleton and also the number of headers 1202 // already forwarded for processing. 1203 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1204 log.Debug("Filling up skeleton", "from", from) 1205 d.queue.ScheduleSkeleton(from, skeleton) 1206 1207 var ( 1208 deliver = func(packet dataPack) (int, error) { 1209 pack := packet.(*headerPack) 1210 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1211 } 1212 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 1213 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) { 1214 return d.queue.ReserveHeaders(p, count), false, false 1215 } 1216 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1217 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 1218 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1219 p.SetHeadersIdle(accepted, deliveryTime) 1220 } 1221 ) 1222 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1223 d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve, 1224 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1225 1226 log.Debug("Skeleton fill terminated", "err", err) 1227 1228 filled, proced := d.queue.RetrieveHeaders() 1229 return filled, proced, err 1230 } 1231 1232 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1233 // available peers, reserving a chunk of blocks for each, waiting for delivery 1234 // and also periodically checking for timeouts. 1235 func (d *Downloader) fetchBodies(from uint64) error { 1236 log.Debug("Downloading block bodies", "origin", from) 1237 1238 var ( 1239 deliver = func(packet dataPack) (int, error) { 1240 pack := packet.(*bodyPack) 1241 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1242 } 1243 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 1244 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1245 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 1246 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) } 1247 ) 1248 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1249 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies, 1250 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1251 1252 log.Debug("Block body download terminated", "err", err) 1253 return err 1254 } 1255 1256 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1257 // available peers, reserving a chunk of receipts for each, waiting for delivery 1258 // and also periodically checking for timeouts. 1259 func (d *Downloader) fetchReceipts(from uint64) error { 1260 log.Debug("Downloading transaction receipts", "origin", from) 1261 1262 var ( 1263 deliver = func(packet dataPack) (int, error) { 1264 pack := packet.(*receiptPack) 1265 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1266 } 1267 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1268 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1269 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1270 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1271 p.SetReceiptsIdle(accepted, deliveryTime) 1272 } 1273 ) 1274 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1275 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts, 1276 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1277 1278 log.Debug("Transaction receipt download terminated", "err", err) 1279 return err 1280 } 1281 1282 // fetchParts iteratively downloads scheduled block parts, taking any available 1283 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1284 // also periodically checking for timeouts. 1285 // 1286 // As the scheduling/timeout logic mostly is the same for all downloaded data 1287 // types, this method is used by each for data gathering and is instrumented with 1288 // various callbacks to handle the slight differences between processing them. 1289 // 1290 // The instrumentation parameters: 1291 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1292 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1293 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1294 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1295 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1296 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1297 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1298 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1299 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1300 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1301 // - fetch: network callback to actually send a particular download request to a physical remote peer 1302 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1303 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1304 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1305 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1306 // - kind: textual label of the type being downloaded to display in log messages 1307 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1308 expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool), 1309 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1310 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error { 1311 1312 // Create a ticker to detect expired retrieval tasks 1313 ticker := time.NewTicker(100 * time.Millisecond) 1314 defer ticker.Stop() 1315 1316 update := make(chan struct{}, 1) 1317 1318 // Prepare the queue and fetch block parts until the block header fetcher's done 1319 finished := false 1320 for { 1321 select { 1322 case <-d.cancelCh: 1323 return errCanceled 1324 1325 case packet := <-deliveryCh: 1326 deliveryTime := time.Now() 1327 // If the peer was previously banned and failed to deliver its pack 1328 // in a reasonable time frame, ignore its message. 1329 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1330 // Deliver the received chunk of data and check chain validity 1331 accepted, err := deliver(packet) 1332 if errors.Is(err, errInvalidChain) { 1333 return err 1334 } 1335 // Unless a peer delivered something completely else than requested (usually 1336 // caused by a timed out request which came through in the end), set it to 1337 // idle. If the delivery's stale, the peer should have already been idled. 1338 if !errors.Is(err, errStaleDelivery) { 1339 setIdle(peer, accepted, deliveryTime) 1340 } 1341 // Issue a log to the user to see what's going on 1342 switch { 1343 case err == nil && packet.Items() == 0: 1344 peer.log.Trace("Requested data not delivered", "type", kind) 1345 case err == nil: 1346 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1347 default: 1348 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1349 } 1350 } 1351 // Blocks assembled, try to update the progress 1352 select { 1353 case update <- struct{}{}: 1354 default: 1355 } 1356 1357 case cont := <-wakeCh: 1358 // The header fetcher sent a continuation flag, check if it's done 1359 if !cont { 1360 finished = true 1361 } 1362 // Headers arrive, try to update the progress 1363 select { 1364 case update <- struct{}{}: 1365 default: 1366 } 1367 1368 case <-ticker.C: 1369 // Sanity check update the progress 1370 select { 1371 case update <- struct{}{}: 1372 default: 1373 } 1374 1375 case <-update: 1376 // Short circuit if we lost all our peers 1377 if d.peers.Len() == 0 { 1378 return errNoPeers 1379 } 1380 // Check for fetch request timeouts and demote the responsible peers 1381 for pid, fails := range expire() { 1382 if peer := d.peers.Peer(pid); peer != nil { 1383 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1384 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1385 // out that sync wise we need to get rid of the peer. 1386 // 1387 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1388 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1389 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1390 if fails > 2 { 1391 peer.log.Trace("Data delivery timed out", "type", kind) 1392 setIdle(peer, 0, time.Now()) 1393 } else { 1394 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1395 1396 if d.dropPeer == nil { 1397 // The dropPeer method is nil when `--copydb` is used for a local copy. 1398 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1399 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1400 } else { 1401 d.dropPeer(pid) 1402 1403 // If this peer was the master peer, abort sync immediately 1404 d.cancelLock.RLock() 1405 master := pid == d.cancelPeer 1406 d.cancelLock.RUnlock() 1407 1408 if master { 1409 d.cancel() 1410 return errTimeout 1411 } 1412 } 1413 } 1414 } 1415 } 1416 // If there's nothing more to fetch, wait or terminate 1417 if pending() == 0 { 1418 if !inFlight() && finished { 1419 log.Debug("Data fetching completed", "type", kind) 1420 return nil 1421 } 1422 break 1423 } 1424 // Send a download request to all idle peers, until throttled 1425 progressed, throttled, running := false, false, inFlight() 1426 idles, total := idle() 1427 pendCount := pending() 1428 for _, peer := range idles { 1429 // Short circuit if throttling activated 1430 if throttled { 1431 break 1432 } 1433 // Short circuit if there is no more available task. 1434 if pendCount = pending(); pendCount == 0 { 1435 break 1436 } 1437 // Reserve a chunk of fetches for a peer. A nil can mean either that 1438 // no more headers are available, or that the peer is known not to 1439 // have them. 1440 request, progress, throttle := reserve(peer, capacity(peer)) 1441 if progress { 1442 progressed = true 1443 } 1444 if throttle { 1445 throttled = true 1446 throttleCounter.Inc(1) 1447 } 1448 if request == nil { 1449 continue 1450 } 1451 if request.From > 0 { 1452 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1453 } else { 1454 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1455 } 1456 // Fetch the chunk and make sure any errors return the hashes to the queue 1457 if fetchHook != nil { 1458 fetchHook(request.Headers) 1459 } 1460 if err := fetch(peer, request); err != nil { 1461 // Although we could try and make an attempt to fix this, this error really 1462 // means that we've double allocated a fetch task to a peer. If that is the 1463 // case, the internal state of the downloader and the queue is very wrong so 1464 // better hard crash and note the error instead of silently accumulating into 1465 // a much bigger issue. 1466 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1467 } 1468 running = true 1469 } 1470 // Make sure that we have peers available for fetching. If all peers have been tried 1471 // and all failed throw an error 1472 if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 { 1473 return errPeersUnavailable 1474 } 1475 } 1476 } 1477 } 1478 1479 // processHeaders takes batches of retrieved headers from an input channel and 1480 // keeps processing and scheduling them into the header chain and downloader's 1481 // queue until the stream ends or a failure occurs. 1482 func (d *Downloader) processHeaders(origin uint64, td *big.Int) error { 1483 // Keep a count of uncertain headers to roll back 1484 var ( 1485 rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis) 1486 rollbackErr error 1487 mode = d.getMode() 1488 ) 1489 defer func() { 1490 if rollback > 0 { 1491 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1492 if mode != LightSync { 1493 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1494 lastBlock = d.blockchain.CurrentBlock().Number() 1495 } 1496 if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block 1497 // We're already unwinding the stack, only print the error to make it more visible 1498 log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err) 1499 } 1500 curFastBlock, curBlock := common.Big0, common.Big0 1501 if mode != LightSync { 1502 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1503 curBlock = d.blockchain.CurrentBlock().Number() 1504 } 1505 log.Warn("Rolled back chain segment", 1506 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1507 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1508 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr) 1509 } 1510 }() 1511 // Wait for batches of headers to process 1512 gotHeaders := false 1513 1514 for { 1515 select { 1516 case <-d.cancelCh: 1517 rollbackErr = errCanceled 1518 return errCanceled 1519 1520 case headers := <-d.headerProcCh: 1521 // Terminate header processing if we synced up 1522 if len(headers) == 0 { 1523 // Notify everyone that headers are fully processed 1524 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1525 select { 1526 case ch <- false: 1527 case <-d.cancelCh: 1528 } 1529 } 1530 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1531 // better chain compared to ours. The only exception is if its promised blocks were 1532 // already imported by other means (e.g. fetcher): 1533 // 1534 // R <remote peer>, L <local node>: Both at block 10 1535 // R: Mine block 11, and propagate it to L 1536 // L: Queue block 11 for import 1537 // L: Notice that R's head and TD increased compared to ours, start sync 1538 // L: Import of block 11 finishes 1539 // L: Sync begins, and finds common ancestor at 11 1540 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1541 // R: Nothing to give 1542 if mode != LightSync { 1543 head := d.blockchain.CurrentBlock() 1544 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1545 return errStallingPeer 1546 } 1547 } 1548 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1549 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1550 // of delivering the post-pivot blocks that would flag the invalid content. 1551 // 1552 // This check cannot be executed "as is" for full imports, since blocks may still be 1553 // queued for processing when the header download completes. However, as long as the 1554 // peer gave us something useful, we're already happy/progressed (above check). 1555 if mode == FastSync || mode == LightSync { 1556 head := d.lightchain.CurrentHeader() 1557 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1558 return errStallingPeer 1559 } 1560 } 1561 // Disable any rollback and return 1562 rollback = 0 1563 return nil 1564 } 1565 // Otherwise split the chunk of headers into batches and process them 1566 gotHeaders = true 1567 for len(headers) > 0 { 1568 // Terminate if something failed in between processing chunks 1569 select { 1570 case <-d.cancelCh: 1571 rollbackErr = errCanceled 1572 return errCanceled 1573 default: 1574 } 1575 // Select the next chunk of headers to import 1576 limit := maxHeadersProcess 1577 if limit > len(headers) { 1578 limit = len(headers) 1579 } 1580 chunk := headers[:limit] 1581 1582 // In case of header only syncing, validate the chunk immediately 1583 if mode == FastSync || mode == LightSync { 1584 // If we're importing pure headers, verify based on their recentness 1585 var pivot uint64 1586 1587 d.pivotLock.RLock() 1588 if d.pivotHeader != nil { 1589 pivot = d.pivotHeader.Number.Uint64() 1590 } 1591 d.pivotLock.RUnlock() 1592 1593 frequency := fsHeaderCheckFrequency 1594 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1595 frequency = 1 1596 } 1597 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1598 rollbackErr = err 1599 1600 // If some headers were inserted, track them as uncertain 1601 if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 { 1602 rollback = chunk[0].Number.Uint64() 1603 } 1604 log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err) 1605 return fmt.Errorf("%w: %v", errInvalidChain, err) 1606 } 1607 // All verifications passed, track all headers within the alloted limits 1608 if mode == FastSync { 1609 head := chunk[len(chunk)-1].Number.Uint64() 1610 if head-rollback > uint64(fsHeaderSafetyNet) { 1611 rollback = head - uint64(fsHeaderSafetyNet) 1612 } else { 1613 rollback = 1 1614 } 1615 } 1616 } 1617 // Unless we're doing light chains, schedule the headers for associated content retrieval 1618 if mode == FullSync || mode == FastSync { 1619 // If we've reached the allowed number of pending headers, stall a bit 1620 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1621 select { 1622 case <-d.cancelCh: 1623 rollbackErr = errCanceled 1624 return errCanceled 1625 case <-time.After(time.Second): 1626 } 1627 } 1628 // Otherwise insert the headers for content retrieval 1629 inserts := d.queue.Schedule(chunk, origin) 1630 if len(inserts) != len(chunk) { 1631 rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk)) 1632 return fmt.Errorf("%w: stale headers", errBadPeer) 1633 } 1634 } 1635 headers = headers[limit:] 1636 origin += uint64(limit) 1637 } 1638 // Update the highest block number we know if a higher one is found. 1639 d.syncStatsLock.Lock() 1640 if d.syncStatsChainHeight < origin { 1641 d.syncStatsChainHeight = origin - 1 1642 } 1643 d.syncStatsLock.Unlock() 1644 1645 // Signal the content downloaders of the availablility of new tasks 1646 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1647 select { 1648 case ch <- true: 1649 default: 1650 } 1651 } 1652 } 1653 } 1654 } 1655 1656 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1657 func (d *Downloader) processFullSyncContent() error { 1658 for { 1659 results := d.queue.Results(true) 1660 if len(results) == 0 { 1661 return nil 1662 } 1663 if d.chainInsertHook != nil { 1664 d.chainInsertHook(results) 1665 } 1666 if err := d.importBlockResults(results); err != nil { 1667 return err 1668 } 1669 } 1670 } 1671 1672 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1673 // Check for any early termination requests 1674 if len(results) == 0 { 1675 return nil 1676 } 1677 select { 1678 case <-d.quitCh: 1679 return errCancelContentProcessing 1680 default: 1681 } 1682 // Retrieve the a batch of results to import 1683 first, last := results[0].Header, results[len(results)-1].Header 1684 log.Debug("Inserting downloaded chain", "items", len(results), 1685 "firstnum", first.Number, "firsthash", first.Hash(), 1686 "lastnum", last.Number, "lasthash", last.Hash(), 1687 ) 1688 blocks := make([]*types.Block, len(results)) 1689 for i, result := range results { 1690 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1691 } 1692 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1693 if index < len(results) { 1694 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1695 } else { 1696 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1697 // when it needs to preprocess blocks to import a sidechain. 1698 // The importer will put together a new list of blocks to import, which is a superset 1699 // of the blocks delivered from the downloader, and the indexing will be off. 1700 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1701 } 1702 return fmt.Errorf("%w: %v", errInvalidChain, err) 1703 } 1704 return nil 1705 } 1706 1707 // processFastSyncContent takes fetch results from the queue and writes them to the 1708 // database. It also controls the synchronisation of state nodes of the pivot block. 1709 func (d *Downloader) processFastSyncContent() error { 1710 // Start syncing state of the reported head block. This should get us most of 1711 // the state of the pivot block. 1712 d.pivotLock.RLock() 1713 sync := d.syncState(d.pivotHeader.Root) 1714 d.pivotLock.RUnlock() 1715 1716 defer func() { 1717 // The `sync` object is replaced every time the pivot moves. We need to 1718 // defer close the very last active one, hence the lazy evaluation vs. 1719 // calling defer sync.Cancel() !!! 1720 sync.Cancel() 1721 }() 1722 1723 closeOnErr := func(s *stateSync) { 1724 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled { 1725 d.queue.Close() // wake up Results 1726 } 1727 } 1728 go closeOnErr(sync) 1729 1730 // To cater for moving pivot points, track the pivot block and subsequently 1731 // accumulated download results separately. 1732 var ( 1733 oldPivot *fetchResult // Locked in pivot block, might change eventually 1734 oldTail []*fetchResult // Downloaded content after the pivot 1735 ) 1736 for { 1737 // Wait for the next batch of downloaded data to be available, and if the pivot 1738 // block became stale, move the goalpost 1739 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1740 if len(results) == 0 { 1741 // If pivot sync is done, stop 1742 if oldPivot == nil { 1743 return sync.Cancel() 1744 } 1745 // If sync failed, stop 1746 select { 1747 case <-d.cancelCh: 1748 sync.Cancel() 1749 return errCanceled 1750 default: 1751 } 1752 } 1753 if d.chainInsertHook != nil { 1754 d.chainInsertHook(results) 1755 } 1756 // If we haven't downloaded the pivot block yet, check pivot staleness 1757 // notifications from the header downloader 1758 d.pivotLock.RLock() 1759 pivot := d.pivotHeader 1760 d.pivotLock.RUnlock() 1761 1762 if oldPivot == nil { 1763 if pivot.Root != sync.root { 1764 sync.Cancel() 1765 sync = d.syncState(pivot.Root) 1766 1767 go closeOnErr(sync) 1768 } 1769 } else { 1770 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1771 } 1772 // Split around the pivot block and process the two sides via fast/full sync 1773 if atomic.LoadInt32(&d.committed) == 0 { 1774 latest := results[len(results)-1].Header 1775 // If the height is above the pivot block by 2 sets, it means the pivot 1776 // become stale in the network and it was garbage collected, move to a 1777 // new pivot. 1778 // 1779 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 1780 // need to be taken into account, otherwise we're detecting the pivot move 1781 // late and will drop peers due to unavailable state!!! 1782 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 1783 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 1784 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 1785 1786 d.pivotLock.Lock() 1787 d.pivotHeader = pivot 1788 d.pivotLock.Unlock() 1789 1790 // Write out the pivot into the database so a rollback beyond it will 1791 // reenable fast sync 1792 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 1793 } 1794 } 1795 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 1796 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1797 return err 1798 } 1799 if P != nil { 1800 // If new pivot block found, cancel old state retrieval and restart 1801 if oldPivot != P { 1802 sync.Cancel() 1803 sync = d.syncState(P.Header.Root) 1804 1805 go closeOnErr(sync) 1806 oldPivot = P 1807 } 1808 // Wait for completion, occasionally checking for pivot staleness 1809 select { 1810 case <-sync.done: 1811 if sync.err != nil { 1812 return sync.err 1813 } 1814 if err := d.commitPivotBlock(P); err != nil { 1815 return err 1816 } 1817 oldPivot = nil 1818 1819 case <-time.After(time.Second): 1820 oldTail = afterP 1821 continue 1822 } 1823 } 1824 // Fast sync done, pivot commit done, full import 1825 if err := d.importBlockResults(afterP); err != nil { 1826 return err 1827 } 1828 } 1829 } 1830 1831 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1832 if len(results) == 0 { 1833 return nil, nil, nil 1834 } 1835 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 1836 // the pivot is somewhere in the future 1837 return nil, results, nil 1838 } 1839 // This can also be optimized, but only happens very seldom 1840 for _, result := range results { 1841 num := result.Header.Number.Uint64() 1842 switch { 1843 case num < pivot: 1844 before = append(before, result) 1845 case num == pivot: 1846 p = result 1847 default: 1848 after = append(after, result) 1849 } 1850 } 1851 return p, before, after 1852 } 1853 1854 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1855 // Check for any early termination requests 1856 if len(results) == 0 { 1857 return nil 1858 } 1859 select { 1860 case <-d.quitCh: 1861 return errCancelContentProcessing 1862 case <-stateSync.done: 1863 if err := stateSync.Wait(); err != nil { 1864 return err 1865 } 1866 default: 1867 } 1868 // Retrieve the a batch of results to import 1869 first, last := results[0].Header, results[len(results)-1].Header 1870 log.Debug("Inserting fast-sync blocks", "items", len(results), 1871 "firstnum", first.Number, "firsthash", first.Hash(), 1872 "lastnumn", last.Number, "lasthash", last.Hash(), 1873 ) 1874 blocks := make([]*types.Block, len(results)) 1875 receipts := make([]types.Receipts, len(results)) 1876 for i, result := range results { 1877 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1878 receipts[i] = result.Receipts 1879 } 1880 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1881 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1882 return fmt.Errorf("%w: %v", errInvalidChain, err) 1883 } 1884 return nil 1885 } 1886 1887 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1888 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1889 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1890 1891 // Commit the pivot block as the new head, will require full sync from here on 1892 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1893 return err 1894 } 1895 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1896 return err 1897 } 1898 atomic.StoreInt32(&d.committed, 1) 1899 1900 // If we had a bloom filter for the state sync, deallocate it now. Note, we only 1901 // deallocate internally, but keep the empty wrapper. This ensures that if we do 1902 // a rollback after committing the pivot and restarting fast sync, we don't end 1903 // up using a nil bloom. Empty bloom is fine, it just returns that it does not 1904 // have the info we need, so reach down to the database instead. 1905 if d.stateBloom != nil { 1906 d.stateBloom.Close() 1907 } 1908 return nil 1909 } 1910 1911 // DeliverHeaders injects a new batch of block headers received from a remote 1912 // node into the download schedule. 1913 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) { 1914 return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1915 } 1916 1917 // DeliverBodies injects a new batch of block bodies received from a remote node. 1918 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) { 1919 return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1920 } 1921 1922 // DeliverReceipts injects a new batch of receipts received from a remote node. 1923 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) { 1924 return d.deliver(id, d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1925 } 1926 1927 // DeliverNodeData injects a new batch of node state data received from a remote node. 1928 func (d *Downloader) DeliverNodeData(id string, data [][]byte) (err error) { 1929 return d.deliver(id, d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1930 } 1931 1932 // deliver injects a new batch of data received from a remote node. 1933 func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1934 // Update the delivery metrics for both good and failed deliveries 1935 inMeter.Mark(int64(packet.Items())) 1936 defer func() { 1937 if err != nil { 1938 dropMeter.Mark(int64(packet.Items())) 1939 } 1940 }() 1941 // Deliver or abort if the sync is canceled while queuing 1942 d.cancelLock.RLock() 1943 cancel := d.cancelCh 1944 d.cancelLock.RUnlock() 1945 if cancel == nil { 1946 return errNoSyncActive 1947 } 1948 select { 1949 case destCh <- packet: 1950 return nil 1951 case <-cancel: 1952 return errNoSyncActive 1953 } 1954 } 1955 1956 // qosTuner is the quality of service tuning loop that occasionally gathers the 1957 // peer latency statistics and updates the estimated request round trip time. 1958 func (d *Downloader) qosTuner() { 1959 for { 1960 // Retrieve the current median RTT and integrate into the previoust target RTT 1961 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 1962 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 1963 1964 // A new RTT cycle passed, increase our confidence in the estimated RTT 1965 conf := atomic.LoadUint64(&d.rttConfidence) 1966 conf = conf + (1000000-conf)/2 1967 atomic.StoreUint64(&d.rttConfidence, conf) 1968 1969 // Log the new QoS values and sleep until the next RTT 1970 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1971 select { 1972 case <-d.quitCh: 1973 return 1974 case <-time.After(rtt): 1975 } 1976 } 1977 } 1978 1979 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 1980 // peer set, needing to reduce the confidence we have in out QoS estimates. 1981 func (d *Downloader) qosReduceConfidence() { 1982 // If we have a single peer, confidence is always 1 1983 peers := uint64(d.peers.Len()) 1984 if peers == 0 { 1985 // Ensure peer connectivity races don't catch us off guard 1986 return 1987 } 1988 if peers == 1 { 1989 atomic.StoreUint64(&d.rttConfidence, 1000000) 1990 return 1991 } 1992 // If we have a ton of peers, don't drop confidence) 1993 if peers >= uint64(qosConfidenceCap) { 1994 return 1995 } 1996 // Otherwise drop the confidence factor 1997 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 1998 if float64(conf)/1000000 < rttMinConfidence { 1999 conf = uint64(rttMinConfidence * 1000000) 2000 } 2001 atomic.StoreUint64(&d.rttConfidence, conf) 2002 2003 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2004 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 2005 } 2006 2007 // requestRTT returns the current target round trip time for a download request 2008 // to complete in. 2009 // 2010 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 2011 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 2012 // be adapted to, but smaller ones are preferred (stabler download stream). 2013 func (d *Downloader) requestRTT() time.Duration { 2014 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 2015 } 2016 2017 // requestTTL returns the current timeout allowance for a single download request 2018 // to finish under. 2019 func (d *Downloader) requestTTL() time.Duration { 2020 var ( 2021 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2022 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 2023 ) 2024 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 2025 if ttl > ttlLimit { 2026 ttl = ttlLimit 2027 } 2028 return ttl 2029 }