github.com/core-coin/go-core/v2@v2.1.9/xcb/downloader/downloader.go (about) 1 // Copyright 2015 by the Authors 2 // This file is part of the go-core library. 3 // 4 // The go-core 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-core 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-core 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 core "github.com/core-coin/go-core/v2" 29 30 "github.com/core-coin/go-core/v2/common" 31 "github.com/core-coin/go-core/v2/core/rawdb" 32 "github.com/core-coin/go-core/v2/core/types" 33 "github.com/core-coin/go-core/v2/event" 34 "github.com/core-coin/go-core/v2/log" 35 "github.com/core-coin/go-core/v2/metrics" 36 "github.com/core-coin/go-core/v2/params" 37 "github.com/core-coin/go-core/v2/trie" 38 "github.com/core-coin/go-core/v2/xcbdb" 39 ) 40 41 var ( 42 MaxHashFetch = 512 // Amount of hashes to be fetched per retrieval request 43 MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request 44 MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request 45 MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly 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 // [xcb/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 fullMaxForkAncestry uint64 = params.FullImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 63 lightMaxForkAncestry uint64 = params.LightImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it) 64 65 reorgProtThreshold = 48 // Threshold number of recent blocks to disable mini reorg protection 66 reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs 67 68 fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync 69 fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected 70 fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it 71 fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download 72 fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync 73 ) 74 75 var ( 76 errBusy = errors.New("busy") 77 errUnknownPeer = errors.New("peer is unknown or unhealthy") 78 errBadPeer = errors.New("action from bad peer ignored") 79 errStallingPeer = errors.New("peer is stalling") 80 errUnsyncedPeer = errors.New("unsynced peer") 81 errNoPeers = errors.New("no peers to keep download active") 82 errTimeout = errors.New("timeout") 83 errEmptyHeaderSet = errors.New("empty header set by peer") 84 errPeersUnavailable = errors.New("no peers available or all tried for download") 85 errInvalidAncestor = errors.New("retrieved ancestor is invalid") 86 errInvalidChain = errors.New("retrieved hash chain is invalid") 87 errInvalidBody = errors.New("retrieved block body is invalid") 88 errInvalidReceipt = errors.New("retrieved receipt is invalid") 89 errCancelStateFetch = errors.New("state data download canceled (requested)") 90 errCancelContentProcessing = errors.New("content processing canceled (requested)") 91 errCanceled = errors.New("syncing canceled (requested)") 92 errNoSyncActive = errors.New("no sync active") 93 errTooOld = errors.New("peer doesn't speak recent enough protocol version (need version >= 63)") 94 ) 95 96 type Downloader struct { 97 // WARNING: The `rttEstimate` and `rttConfidence` fields are accessed atomically. 98 // On 32 bit platforms, only 64-bit aligned fields can be atomic. The struct is 99 // guaranteed to be so aligned, so take advantage of that. For more information, 100 // see https://golang.org/pkg/sync/atomic/#pkg-note-BUG. 101 rttEstimate uint64 // Round trip time to target for download requests 102 rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops) 103 104 mode uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode 105 mux *event.TypeMux // Event multiplexer to announce sync operation events 106 107 checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync) 108 genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT) 109 queue *queue // Scheduler for selecting the hashes to download 110 peers *peerSet // Set of active peers from which download can proceed 111 112 stateDB xcbdb.Database // Database to state sync into (and deduplicate via) 113 stateBloom *trie.SyncBloom // Bloom filter for fast trie node and contract code existence checks 114 115 // Statistics 116 syncStatsChainOrigin uint64 // Origin block number where syncing started at 117 syncStatsChainHeight uint64 // Highest block number known when syncing started 118 syncStatsState stateSyncStats 119 syncStatsLock sync.RWMutex // Lock protecting the sync stats fields 120 121 lightchain LightChain 122 blockchain BlockChain 123 124 // Callbacks 125 dropPeer peerDropFn // Drops a peer for misbehaving 126 127 // Status 128 synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing 129 synchronising int32 130 notified int32 131 committed int32 132 ancientLimit uint64 // The maximum block number which can be regarded as ancient data. 133 134 // Channels 135 headerCh chan dataPack // [xcb/62] Channel receiving inbound block headers 136 bodyCh chan dataPack // [xcb/62] Channel receiving inbound block bodies 137 receiptCh chan dataPack // [xcb/63] Channel receiving inbound receipts 138 bodyWakeCh chan bool // [xcb/62] Channel to signal the block body fetcher of new tasks 139 receiptWakeCh chan bool // [xcb/63] Channel to signal the receipt fetcher of new tasks 140 headerProcCh chan []*types.Header // [xcb/62] Channel to feed the header processor new tasks 141 142 // State sync 143 pivotHeader *types.Header // Pivot block header to dynamically push the syncing state root 144 pivotLock sync.RWMutex // Lock protecting pivot header reads from updates 145 146 stateSyncStart chan *stateSync 147 trackStateReq chan *stateReq 148 stateCh chan dataPack // [xcb/63] Channel receiving inbound node state data 149 150 // Cancellation and termination 151 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 152 cancelCh chan struct{} // Channel to cancel mid-flight syncs 153 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 154 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 155 156 quitCh chan struct{} // Quit channel to signal termination 157 quitLock sync.Mutex // Lock to prevent double closes 158 159 // Testing hooks 160 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 161 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 162 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 163 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 164 } 165 166 // LightChain encapsulates functions required to synchronise a light chain. 167 type LightChain interface { 168 // HasHeader verifies a header's presence in the local chain. 169 HasHeader(common.Hash, uint64) bool 170 171 // GetHeaderByHash retrieves a header from the local chain. 172 GetHeaderByHash(common.Hash) *types.Header 173 174 // CurrentHeader retrieves the head header from the local chain. 175 CurrentHeader() *types.Header 176 177 // GetTd returns the total difficulty of a local block. 178 GetTd(common.Hash, uint64) *big.Int 179 180 // InsertHeaderChain inserts a batch of headers into the local chain. 181 InsertHeaderChain([]*types.Header, int) (int, error) 182 183 // SetHead rewinds the local chain to a new head. 184 SetHead(uint64) error 185 } 186 187 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 188 type BlockChain interface { 189 LightChain 190 191 // HasBlock verifies a block's presence in the local chain. 192 HasBlock(common.Hash, uint64) bool 193 194 // HasFastBlock verifies a fast block's presence in the local chain. 195 HasFastBlock(common.Hash, uint64) bool 196 197 // GetBlockByHash retrieves a block from the local chain. 198 GetBlockByHash(common.Hash) *types.Block 199 200 // CurrentBlock retrieves the head block from the local chain. 201 CurrentBlock() *types.Block 202 203 // CurrentFastBlock retrieves the head fast block from the local chain. 204 CurrentFastBlock() *types.Block 205 206 // FastSyncCommitHead directly commits the head block to a certain entity. 207 FastSyncCommitHead(common.Hash) error 208 209 // InsertChain inserts a batch of blocks into the local chain. 210 InsertChain(types.Blocks) (int, error) 211 212 // InsertReceiptChain inserts a batch of receipts into the local chain. 213 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 214 } 215 216 // New creates a new downloader to fetch hashes and blocks from remote peers. 217 func New(checkpoint uint64, stateDb xcbdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 218 if lightchain == nil { 219 lightchain = chain 220 } 221 dl := &Downloader{ 222 stateDB: stateDb, 223 stateBloom: stateBloom, 224 mux: mux, 225 checkpoint: checkpoint, 226 queue: newQueue(blockCacheMaxItems, blockCacheInitialItems), 227 peers: newPeerSet(), 228 rttEstimate: uint64(rttMaxEstimate), 229 rttConfidence: uint64(1000000), 230 blockchain: chain, 231 lightchain: lightchain, 232 dropPeer: dropPeer, 233 headerCh: make(chan dataPack, 1), 234 bodyCh: make(chan dataPack, 1), 235 receiptCh: make(chan dataPack, 1), 236 bodyWakeCh: make(chan bool, 1), 237 receiptWakeCh: make(chan bool, 1), 238 headerProcCh: make(chan []*types.Header, 1), 239 quitCh: make(chan struct{}), 240 stateCh: make(chan dataPack), 241 stateSyncStart: make(chan *stateSync), 242 syncStatsState: stateSyncStats{ 243 processed: rawdb.ReadFastTrieProgress(stateDb), 244 }, 245 trackStateReq: make(chan *stateReq), 246 } 247 go dl.qosTuner() 248 go dl.stateFetcher() 249 return dl 250 } 251 252 // Progress retrieves the synchronisation boundaries, specifically the origin 253 // block where synchronisation started at (may have failed/suspended); the block 254 // or header sync is currently at; and the latest known block which the sync targets. 255 // 256 // In addition, during the state download phase of fast synchronisation the number 257 // of processed and the total number of known states are also returned. Otherwise 258 // these are zero. 259 func (d *Downloader) Progress() core.SyncProgress { 260 // Lock the current stats and return the progress 261 d.syncStatsLock.RLock() 262 defer d.syncStatsLock.RUnlock() 263 264 current := uint64(0) 265 mode := d.getMode() 266 switch { 267 case d.blockchain != nil && mode == FullSync: 268 current = d.blockchain.CurrentBlock().NumberU64() 269 case d.blockchain != nil && 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", mode) 275 } 276 return core.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 // SyncBloomContains tests if the syncbloom filter contains the given hash: 291 // - false: the bloom definitely does not contain hash 292 // - true: the bloom maybe contains hash 293 // 294 // While the bloom is being initialized (or is closed), all queries will return true. 295 func (d *Downloader) SyncBloomContains(hash []byte) bool { 296 return d.stateBloom == nil || d.stateBloom.Contains(hash) 297 } 298 299 // RegisterPeer injects a new download peer into the set of block source to be 300 // used for fetching hashes and blocks from. 301 func (d *Downloader) RegisterPeer(id string, version int, peer Peer) error { 302 logger := log.New("peer", id) 303 logger.Trace("Registering sync peer") 304 if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil { 305 logger.Error("Failed to register sync peer", "err", err) 306 return err 307 } 308 d.qosReduceConfidence() 309 310 return nil 311 } 312 313 // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer. 314 func (d *Downloader) RegisterLightPeer(id string, version int, peer LightPeer) error { 315 return d.RegisterPeer(id, version, &lightPeerWrapper{peer}) 316 } 317 318 // UnregisterPeer remove a peer from the known list, preventing any action from 319 // the specified peer. An effort is also made to return any pending fetches into 320 // the queue. 321 func (d *Downloader) UnregisterPeer(id string) error { 322 // Unregister the peer from the active peer set and revoke any fetch tasks 323 logger := log.New("peer", id) 324 logger.Trace("Unregistering sync peer") 325 if err := d.peers.Unregister(id); err != nil { 326 logger.Error("Failed to unregister sync peer", "err", err) 327 return err 328 } 329 d.queue.Revoke(id) 330 331 return nil 332 } 333 334 // Synchronise tries to sync up our local block chain with a remote peer, both 335 // adding various sanity checks as well as wrapping it with various log entries. 336 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 337 err := d.synchronise(id, head, td, mode) 338 339 switch err { 340 case nil, errBusy, errCanceled: 341 return err 342 } 343 344 if errors.Is(err, errInvalidChain) || errors.Is(err, errBadPeer) || errors.Is(err, errTimeout) || 345 errors.Is(err, errStallingPeer) || errors.Is(err, errUnsyncedPeer) || errors.Is(err, errEmptyHeaderSet) || 346 errors.Is(err, errPeersUnavailable) || errors.Is(err, errTooOld) || errors.Is(err, errInvalidAncestor) { 347 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 348 if d.dropPeer == nil { 349 // The dropPeer method is nil when `--copydb` is used for a local copy. 350 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 351 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 352 } else { 353 d.dropPeer(id) 354 } 355 return err 356 } 357 log.Warn("Synchronisation failed, retrying", "err", err) 358 return err 359 } 360 361 // synchronise will select the peer and use it for synchronising. If an empty string is given 362 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 363 // checks fail an error will be returned. This method is synchronous 364 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 365 // Mock out the synchronisation if testing 366 if d.synchroniseMock != nil { 367 return d.synchroniseMock(id, hash) 368 } 369 // Make sure only one goroutine is ever allowed past this point at once 370 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 371 return errBusy 372 } 373 defer atomic.StoreInt32(&d.synchronising, 0) 374 375 // Post a user notification of the sync (only once per session) 376 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 377 log.Info("Block synchronisation started") 378 } 379 // If we are already full syncing, but have a fast-sync bloom filter laying 380 // around, make sure it doesn't use memory any more. This is a special case 381 // when the user attempts to fast sync a new empty network. 382 if mode == FullSync && d.stateBloom != nil { 383 d.stateBloom.Close() 384 } 385 // Reset the queue, peer set and wake channels to clean any internal leftover state 386 d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems) 387 d.peers.Reset() 388 389 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 390 select { 391 case <-ch: 392 default: 393 } 394 } 395 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 396 for empty := false; !empty; { 397 select { 398 case <-ch: 399 default: 400 empty = true 401 } 402 } 403 } 404 for empty := false; !empty; { 405 select { 406 case <-d.headerProcCh: 407 default: 408 empty = true 409 } 410 } 411 // Create cancel channel for aborting mid-flight and mark the master peer 412 d.cancelLock.Lock() 413 d.cancelCh = make(chan struct{}) 414 d.cancelPeer = id 415 d.cancelLock.Unlock() 416 417 defer d.Cancel() // No matter what, we can't leave the cancel channel open 418 419 // Atomically set the requested sync mode 420 atomic.StoreUint32(&d.mode, uint32(mode)) 421 422 // Retrieve the origin peer and initiate the downloading process 423 p := d.peers.Peer(id) 424 if p == nil { 425 return errUnknownPeer 426 } 427 return d.syncWithPeer(p, hash, td) 428 } 429 430 func (d *Downloader) getMode() SyncMode { 431 return SyncMode(atomic.LoadUint32(&d.mode)) 432 } 433 434 // syncWithPeer starts a block synchronization based on the hash chain from the 435 // specified peer and head hash. 436 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 437 d.mux.Post(StartEvent{}) 438 defer func() { 439 // reset on error 440 if err != nil { 441 d.mux.Post(FailedEvent{err}) 442 } else { 443 latest := d.lightchain.CurrentHeader() 444 d.mux.Post(DoneEvent{latest}) 445 } 446 }() 447 if p.version < 63 { 448 return errTooOld 449 } 450 mode := d.getMode() 451 452 log.Debug("Synchronising with the network", "peer", p.id, "xcb", p.version, "head", hash, "td", td, "mode", mode) 453 defer func(start time.Time) { 454 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 455 }(time.Now()) 456 457 // Look up the sync boundaries: the common ancestor and the target block 458 latest, pivot, err := d.fetchHead(p) 459 if err != nil { 460 return err 461 } 462 if mode == FastSync && pivot == nil { 463 // If no pivot block was returned, the head is below the min full block 464 // threshold (i.e. new chian). In that case we won't really fast sync 465 // anyway, but still need a valid pivot block to avoid some code hitting 466 // nil panics on an access. 467 pivot = d.blockchain.CurrentBlock().Header() 468 } 469 height := latest.Number.Uint64() 470 471 origin, err := d.findAncestor(p, latest) 472 if err != nil { 473 return err 474 } 475 d.syncStatsLock.Lock() 476 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 477 d.syncStatsChainOrigin = origin 478 } 479 d.syncStatsChainHeight = height 480 d.syncStatsLock.Unlock() 481 482 // Ensure our origin point is below any fast sync pivot point 483 if mode == FastSync { 484 if height <= uint64(fsMinFullBlocks) { 485 origin = 0 486 } else { 487 pivotNumber := pivot.Number.Uint64() 488 if pivotNumber <= origin { 489 origin = pivotNumber - 1 490 } 491 // Write out the pivot into the database so a rollback beyond it will 492 // reenable fast sync 493 rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber) 494 } 495 } 496 d.committed = 1 497 if mode == FastSync && pivot.Number.Uint64() != 0 { 498 d.committed = 0 499 } 500 if mode == FastSync { 501 // Set the ancient data limitation. 502 // If we are running fast sync, all block data older than ancientLimit will be 503 // written to the ancient store. More recent data will be written to the active 504 // database and will wait for the freezer to migrate. 505 // 506 // If there is a checkpoint available, then calculate the ancientLimit through 507 // that. Otherwise calculate the ancient limit through the advertised height 508 // of the remote peer. 509 // 510 // The reason for picking checkpoint first is that a malicious peer can give us 511 // a fake (very high) height, forcing the ancient limit to also be very high. 512 // The peer would start to feed us valid blocks until head, resulting in all of 513 // the blocks might be written into the ancient store. A following mini-reorg 514 // could cause issues. 515 if d.checkpoint != 0 && d.checkpoint > fullMaxForkAncestry+1 { 516 d.ancientLimit = d.checkpoint 517 } else if height > fullMaxForkAncestry+1 { 518 d.ancientLimit = height - fullMaxForkAncestry - 1 519 } else { 520 d.ancientLimit = 0 521 } 522 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 523 524 // If a part of blockchain data has already been written into active store, 525 // disable the ancient style insertion explicitly. 526 if origin >= frozen && frozen != 0 { 527 d.ancientLimit = 0 528 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 529 } else if d.ancientLimit > 0 { 530 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 531 } 532 // Rewind the ancient store and blockchain if reorg happens. 533 if origin+1 < frozen { 534 if err := d.lightchain.SetHead(origin + 1); err != nil { 535 return err 536 } 537 } 538 } 539 // Initiate the sync using a concurrent header and content retrieval algorithm 540 d.queue.Prepare(origin+1, mode) 541 if d.syncInitHook != nil { 542 d.syncInitHook(origin, height) 543 } 544 fetchers := []func() error{ 545 func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved 546 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 547 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 548 func() error { return d.processHeaders(origin+1, td) }, 549 } 550 if mode == FastSync { 551 d.pivotLock.Lock() 552 d.pivotHeader = pivot 553 d.pivotLock.Unlock() 554 555 fetchers = append(fetchers, func() error { return d.processFastSyncContent() }) 556 } else if mode == FullSync { 557 fetchers = append(fetchers, d.processFullSyncContent) 558 } 559 return d.spawnSync(fetchers) 560 } 561 562 // spawnSync runs d.process and all given fetcher functions to completion in 563 // separate goroutines, returning the first error that appears. 564 func (d *Downloader) spawnSync(fetchers []func() error) error { 565 errc := make(chan error, len(fetchers)) 566 d.cancelWg.Add(len(fetchers)) 567 for _, fn := range fetchers { 568 fn := fn 569 go func() { defer d.cancelWg.Done(); errc <- fn() }() 570 } 571 // Wait for the first error, then terminate the others. 572 var err error 573 for i := 0; i < len(fetchers); i++ { 574 if i == len(fetchers)-1 { 575 // Close the queue when all fetchers have exited. 576 // This will cause the block processor to end when 577 // it has processed the queue. 578 d.queue.Close() 579 } 580 if err = <-errc; err != nil && err != errCanceled { 581 break 582 } 583 } 584 d.queue.Close() 585 d.Cancel() 586 return err 587 } 588 589 // cancel aborts all of the operations and resets the queue. However, cancel does 590 // not wait for the running download goroutines to finish. This method should be 591 // used when cancelling the downloads from inside the downloader. 592 func (d *Downloader) cancel() { 593 // Close the current cancel channel 594 d.cancelLock.Lock() 595 defer d.cancelLock.Unlock() 596 597 if d.cancelCh != nil { 598 select { 599 case <-d.cancelCh: 600 // Channel was already closed 601 default: 602 close(d.cancelCh) 603 } 604 } 605 } 606 607 // Cancel aborts all of the operations and waits for all download goroutines to 608 // finish before returning. 609 func (d *Downloader) Cancel() { 610 d.cancel() 611 d.cancelWg.Wait() 612 } 613 614 // Terminate interrupts the downloader, canceling all pending operations. 615 // The downloader cannot be reused after calling Terminate. 616 func (d *Downloader) Terminate() { 617 // Close the termination channel (make sure double close is allowed) 618 d.quitLock.Lock() 619 select { 620 case <-d.quitCh: 621 default: 622 close(d.quitCh) 623 } 624 if d.stateBloom != nil { 625 d.stateBloom.Close() 626 } 627 d.quitLock.Unlock() 628 629 // Cancel any pending download requests 630 d.Cancel() 631 } 632 633 // fetchHead retrieves the head header and prior pivot block (if available) from 634 // a remote peer. 635 func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *types.Header, err error) { 636 p.log.Debug("Retrieving remote chain head") 637 mode := d.getMode() 638 639 // Request the advertised remote head block and wait for the response 640 latest, _ := p.peer.Head() 641 fetch := 1 642 if mode == FastSync { 643 fetch = 2 // head + pivot headers 644 } 645 go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true) 646 647 ttl := d.requestTTL() 648 timeout := time.After(ttl) 649 for { 650 select { 651 case <-d.cancelCh: 652 return nil, nil, errCanceled 653 654 case packet := <-d.headerCh: 655 // Discard anything not from the origin peer 656 if packet.PeerId() != p.id { 657 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 658 break 659 } 660 // Make sure the peer gave us at least one and at most the requested headers 661 headers := packet.(*headerPack).headers 662 if len(headers) == 0 || len(headers) > fetch { 663 return nil, nil, fmt.Errorf("%w: returned headers %d != requested %d", errBadPeer, len(headers), fetch) 664 } 665 // The first header needs to be the head, validate against the checkpoint 666 // and request. If only 1 header was returned, make sure there's no pivot 667 // or there was not one requested. 668 head := headers[0] 669 if (mode == FastSync || mode == LightSync) && head.Number.Uint64() < d.checkpoint { 670 return nil, nil, fmt.Errorf("%w: remote head %d below checkpoint %d", errUnsyncedPeer, head.Number, d.checkpoint) 671 } 672 if len(headers) == 1 { 673 if mode == FastSync && head.Number.Uint64() > uint64(fsMinFullBlocks) { 674 return nil, nil, fmt.Errorf("%w: no pivot included along head header", errBadPeer) 675 } 676 p.log.Debug("Remote head identified, no pivot", "number", head.Number, "hash", head.Hash()) 677 return head, nil, nil 678 } 679 // At this point we have 2 headers in total and the first is the 680 // validated head of the chian. Check the pivot number and return, 681 pivot := headers[1] 682 if pivot.Number.Uint64() != head.Number.Uint64()-uint64(fsMinFullBlocks) { 683 return nil, nil, fmt.Errorf("%w: remote pivot %d != requested %d", errInvalidChain, pivot.Number, head.Number.Uint64()-uint64(fsMinFullBlocks)) 684 } 685 return head, pivot, nil 686 687 case <-timeout: 688 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 689 return nil, nil, errTimeout 690 691 case <-d.bodyCh: 692 case <-d.receiptCh: 693 // Out of bounds delivery, ignore 694 } 695 } 696 } 697 698 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 699 // common ancestor. 700 // It returns parameters to be used for peer.RequestHeadersByNumber: 701 // 702 // from - starting block number 703 // count - number of headers to request 704 // skip - number of headers to skip 705 // 706 // and also returns 'max', the last block which is expected to be returned by the remote peers, 707 // given the (from,count,skip) 708 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 709 var ( 710 from int 711 count int 712 MaxCount = MaxHeaderFetch / 16 713 ) 714 // requestHead is the highest block that we will ask for. If requestHead is not offset, 715 // the highest block that we will get is 16 blocks back from head, which means we 716 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 717 // between us and the peer is 1-2 blocks, which is most common 718 requestHead := int(remoteHeight) - 1 719 if requestHead < 0 { 720 requestHead = 0 721 } 722 // requestBottom is the lowest block we want included in the query 723 // Ideally, we want to include the one just below our own head 724 requestBottom := int(localHeight - 1) 725 if requestBottom < 0 { 726 requestBottom = 0 727 } 728 totalSpan := requestHead - requestBottom 729 span := 1 + totalSpan/MaxCount 730 if span < 2 { 731 span = 2 732 } 733 if span > 16 { 734 span = 16 735 } 736 737 count = 1 + totalSpan/span 738 if count > MaxCount { 739 count = MaxCount 740 } 741 if count < 2 { 742 count = 2 743 } 744 from = requestHead - (count-1)*span 745 if from < 0 { 746 from = 0 747 } 748 max := from + (count-1)*span 749 return int64(from), count, span - 1, uint64(max) 750 } 751 752 // findAncestor tries to locate the common ancestor link of the local chain and 753 // a remote peers blockchain. In the general case when our node was in sync and 754 // on the correct chain, checking the top N links should already get us a match. 755 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 756 // the head links match), we do a binary search to find the common ancestor. 757 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 758 // Figure out the valid ancestor range to prevent rewrite attacks 759 var ( 760 floor = int64(-1) 761 localHeight uint64 762 remoteHeight = remoteHeader.Number.Uint64() 763 ) 764 mode := d.getMode() 765 switch mode { 766 case FullSync: 767 localHeight = d.blockchain.CurrentBlock().NumberU64() 768 case FastSync: 769 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 770 default: 771 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 772 } 773 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 774 775 // Recap floor value for binary search 776 maxForkAncestry := fullMaxForkAncestry 777 if d.getMode() == LightSync { 778 maxForkAncestry = lightMaxForkAncestry 779 } 780 if localHeight >= maxForkAncestry { 781 // We're above the max reorg threshold, find the earliest fork point 782 floor = int64(localHeight - maxForkAncestry) 783 } 784 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 785 // all headers before that point will be missing. 786 if mode == LightSync { 787 // If we don't know the current CHT position, find it 788 if d.genesis == 0 { 789 header := d.lightchain.CurrentHeader() 790 for header != nil { 791 d.genesis = header.Number.Uint64() 792 if floor >= int64(d.genesis)-1 { 793 break 794 } 795 header = d.lightchain.GetHeaderByHash(header.ParentHash) 796 } 797 } 798 // We already know the "genesis" block number, cap floor to that 799 if floor < int64(d.genesis)-1 { 800 floor = int64(d.genesis) - 1 801 } 802 } 803 804 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 805 806 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 807 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 808 809 // Wait for the remote response to the head fetch 810 number, hash := uint64(0), common.Hash{} 811 812 ttl := d.requestTTL() 813 timeout := time.After(ttl) 814 815 for finished := false; !finished; { 816 select { 817 case <-d.cancelCh: 818 return 0, errCanceled 819 820 case packet := <-d.headerCh: 821 // Discard anything not from the origin peer 822 if packet.PeerId() != p.id { 823 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 824 break 825 } 826 // Make sure the peer actually gave something valid 827 headers := packet.(*headerPack).headers 828 if len(headers) == 0 { 829 p.log.Warn("Empty head header set") 830 return 0, errEmptyHeaderSet 831 } 832 // Make sure the peer's reply conforms to the request 833 for i, header := range headers { 834 expectNumber := from + int64(i)*int64(skip+1) 835 if number := header.Number.Int64(); number != expectNumber { 836 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 837 return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering")) 838 } 839 } 840 // Check if a common ancestor was found 841 finished = true 842 for i := len(headers) - 1; i >= 0; i-- { 843 // Skip any headers that underflow/overflow our requested set 844 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 845 continue 846 } 847 // Otherwise check if we already know the header or not 848 h := headers[i].Hash() 849 n := headers[i].Number.Uint64() 850 851 var known bool 852 switch mode { 853 case FullSync: 854 known = d.blockchain.HasBlock(h, n) 855 case FastSync: 856 known = d.blockchain.HasFastBlock(h, n) 857 default: 858 known = d.lightchain.HasHeader(h, n) 859 } 860 if known { 861 number, hash = n, h 862 break 863 } 864 } 865 866 case <-timeout: 867 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 868 return 0, errTimeout 869 870 case <-d.bodyCh: 871 case <-d.receiptCh: 872 // Out of bounds delivery, ignore 873 } 874 } 875 // If the head fetch already found an ancestor, return 876 if hash != (common.Hash{}) { 877 if int64(number) <= floor { 878 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 879 return 0, errInvalidAncestor 880 } 881 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 882 return number, nil 883 } 884 // Ancestor not found, we need to binary search over our chain 885 start, end := uint64(0), remoteHeight 886 if floor > 0 { 887 start = uint64(floor) 888 } 889 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 890 891 for start+1 < end { 892 // Split our chain interval in two, and request the hash to cross check 893 check := (start + end) / 2 894 895 ttl := d.requestTTL() 896 timeout := time.After(ttl) 897 898 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 899 900 // Wait until a reply arrives to this request 901 for arrived := false; !arrived; { 902 select { 903 case <-d.cancelCh: 904 return 0, errCanceled 905 906 case packet := <-d.headerCh: 907 // Discard anything not from the origin peer 908 if packet.PeerId() != p.id { 909 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 910 break 911 } 912 // Make sure the peer actually gave something valid 913 headers := packet.(*headerPack).headers 914 if len(headers) != 1 { 915 p.log.Warn("Multiple headers for single request", "headers", len(headers)) 916 return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers)) 917 } 918 arrived = true 919 920 // Modify the search interval based on the response 921 h := headers[0].Hash() 922 n := headers[0].Number.Uint64() 923 924 var known bool 925 switch mode { 926 case FullSync: 927 known = d.blockchain.HasBlock(h, n) 928 case FastSync: 929 known = d.blockchain.HasFastBlock(h, n) 930 default: 931 known = d.lightchain.HasHeader(h, n) 932 } 933 if !known { 934 end = check 935 break 936 } 937 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 938 if header.Number.Uint64() != check { 939 p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 940 return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number) 941 } 942 start = check 943 hash = h 944 945 case <-timeout: 946 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 947 return 0, errTimeout 948 949 case <-d.bodyCh: 950 case <-d.receiptCh: 951 // Out of bounds delivery, ignore 952 } 953 } 954 } 955 // Ensure valid ancestry and return 956 if int64(start) <= floor { 957 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 958 return 0, errInvalidAncestor 959 } 960 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 961 return start, nil 962 } 963 964 // fetchHeaders keeps retrieving headers concurrently from the number 965 // requested, until no more are returned, potentially throttling on the way. To 966 // facilitate concurrency but still protect against malicious nodes sending bad 967 // headers, we construct a header chain skeleton using the "origin" peer we are 968 // syncing with, and fill in the missing headers using anyone else. Headers from 969 // other peers are only accepted if they map cleanly to the skeleton. If no one 970 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 971 // the origin is dropped. 972 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error { 973 p.log.Debug("Directing header downloads", "origin", from) 974 defer p.log.Debug("Header download terminated") 975 976 // Create a timeout timer, and the associated header fetcher 977 skeleton := true // Skeleton assembly phase or finishing up 978 pivoting := false // Whether the next request is pivot verification 979 request := time.Now() // time of the last skeleton fetch request 980 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 981 <-timeout.C // timeout channel should be initially empty 982 defer timeout.Stop() 983 984 var ttl time.Duration 985 getHeaders := func(from uint64) { 986 request = time.Now() 987 988 ttl = d.requestTTL() 989 timeout.Reset(ttl) 990 991 if skeleton { 992 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 993 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 994 } else { 995 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 996 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 997 } 998 } 999 getNextPivot := func() { 1000 pivoting = true 1001 request = time.Now() 1002 1003 ttl = d.requestTTL() 1004 timeout.Reset(ttl) 1005 1006 d.pivotLock.RLock() 1007 pivot := d.pivotHeader.Number.Uint64() 1008 d.pivotLock.RUnlock() 1009 1010 p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks)) 1011 go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep 1012 } 1013 // Start pulling the header chain skeleton until all is done 1014 ancestor := from 1015 getHeaders(from) 1016 1017 mode := d.getMode() 1018 for { 1019 select { 1020 case <-d.cancelCh: 1021 return errCanceled 1022 1023 case packet := <-d.headerCh: 1024 // Make sure the active peer is giving us the skeleton headers 1025 if packet.PeerId() != p.id { 1026 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 1027 break 1028 } 1029 headerReqTimer.UpdateSince(request) 1030 timeout.Stop() 1031 1032 // If the pivot is being checked, move if it became stale and run the real retrieval 1033 var pivot uint64 1034 1035 d.pivotLock.RLock() 1036 if d.pivotHeader != nil { 1037 pivot = d.pivotHeader.Number.Uint64() 1038 } 1039 d.pivotLock.RUnlock() 1040 1041 if pivoting { 1042 if packet.Items() == 2 { 1043 // Retrieve the headers and do some sanity checks, just in case 1044 headers := packet.(*headerPack).headers 1045 1046 if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want { 1047 log.Warn("Peer sent invalid next pivot", "have", have, "want", want) 1048 return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want) 1049 } 1050 if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want { 1051 log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want) 1052 return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want) 1053 } 1054 log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number) 1055 pivot = headers[0].Number.Uint64() 1056 1057 d.pivotLock.Lock() 1058 d.pivotHeader = headers[0] 1059 d.pivotLock.Unlock() 1060 1061 // Write out the pivot into the database so a rollback beyond 1062 // it will reenable fast sync and update the state root that 1063 // the state syncer will be downloading. 1064 rawdb.WriteLastPivotNumber(d.stateDB, pivot) 1065 } 1066 pivoting = false 1067 getHeaders(from) 1068 continue 1069 } 1070 // If the skeleton's finished, pull any remaining head headers directly from the origin 1071 if skeleton && packet.Items() == 0 { 1072 skeleton = false 1073 getHeaders(from) 1074 continue 1075 } 1076 // If no more headers are inbound, notify the content fetchers and return 1077 if packet.Items() == 0 { 1078 // Don't abort header fetches while the pivot is downloading 1079 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 1080 p.log.Debug("No headers, waiting for pivot commit") 1081 select { 1082 case <-time.After(fsHeaderContCheck): 1083 getHeaders(from) 1084 continue 1085 case <-d.cancelCh: 1086 return errCanceled 1087 } 1088 } 1089 // Pivot done (or not in fast sync) and no more headers, terminate the process 1090 p.log.Debug("No more headers available") 1091 select { 1092 case d.headerProcCh <- nil: 1093 return nil 1094 case <-d.cancelCh: 1095 return errCanceled 1096 } 1097 } 1098 headers := packet.(*headerPack).headers 1099 1100 // If we received a skeleton batch, resolve internals concurrently 1101 if skeleton { 1102 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1103 if err != nil { 1104 p.log.Debug("Skeleton chain invalid", "err", err) 1105 return fmt.Errorf("%w: %v", errInvalidChain, err) 1106 } 1107 headers = filled[proced:] 1108 from += uint64(proced) 1109 } else { 1110 // If we're closing in on the chain head, but haven't yet reached it, delay 1111 // the last few headers so mini reorgs on the head don't cause invalid hash 1112 // chain errors. 1113 if n := len(headers); n > 0 { 1114 // Retrieve the current head we're at 1115 var head uint64 1116 if mode == LightSync { 1117 head = d.lightchain.CurrentHeader().Number.Uint64() 1118 } else { 1119 head = d.blockchain.CurrentFastBlock().NumberU64() 1120 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1121 head = full 1122 } 1123 } 1124 // If the head is below the common ancestor, we're actually deduplicating 1125 // already existing chain segments, so use the ancestor as the fake head. 1126 // Otherwise we might end up delaying header deliveries pointlessly. 1127 if head < ancestor { 1128 head = ancestor 1129 } 1130 // If the head is way older than this batch, delay the last few headers 1131 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1132 delay := reorgProtHeaderDelay 1133 if delay > n { 1134 delay = n 1135 } 1136 headers = headers[:n-delay] 1137 } 1138 } 1139 } 1140 // Insert all the new headers and fetch the next batch 1141 if len(headers) > 0 { 1142 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1143 select { 1144 case d.headerProcCh <- headers: 1145 case <-d.cancelCh: 1146 return errCanceled 1147 } 1148 from += uint64(len(headers)) 1149 1150 // If we're still skeleton filling fast sync, check pivot staleness 1151 // before continuing to the next skeleton filling 1152 if skeleton && pivot > 0 { 1153 getNextPivot() 1154 } else { 1155 getHeaders(from) 1156 } 1157 } else { 1158 // No headers delivered, or all of them being delayed, sleep a bit and retry 1159 p.log.Trace("All headers delayed, waiting") 1160 select { 1161 case <-time.After(fsHeaderContCheck): 1162 getHeaders(from) 1163 continue 1164 case <-d.cancelCh: 1165 return errCanceled 1166 } 1167 } 1168 1169 case <-timeout.C: 1170 if d.dropPeer == nil { 1171 // The dropPeer method is nil when `--copydb` is used for a local copy. 1172 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1173 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1174 break 1175 } 1176 // Header retrieval timed out, consider the peer bad and drop 1177 p.log.Debug("Header request timed out", "elapsed", ttl) 1178 headerTimeoutMeter.Mark(1) 1179 d.dropPeer(p.id) 1180 1181 // Finish the sync gracefully instead of dumping the gathered data though 1182 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1183 select { 1184 case ch <- false: 1185 case <-d.cancelCh: 1186 } 1187 } 1188 select { 1189 case d.headerProcCh <- nil: 1190 case <-d.cancelCh: 1191 } 1192 return fmt.Errorf("%w: header request timed out", errBadPeer) 1193 } 1194 } 1195 } 1196 1197 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1198 // and maps them to the provided skeleton header chain. 1199 // 1200 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1201 // immediately to the header processor to keep the rest of the pipeline full even 1202 // in the case of header stalls. 1203 // 1204 // The method returns the entire filled skeleton and also the number of headers 1205 // already forwarded for processing. 1206 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1207 log.Debug("Filling up skeleton", "from", from) 1208 d.queue.ScheduleSkeleton(from, skeleton) 1209 1210 var ( 1211 deliver = func(packet dataPack) (int, error) { 1212 pack := packet.(*headerPack) 1213 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1214 } 1215 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 1216 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) { 1217 return d.queue.ReserveHeaders(p, count), false, false 1218 } 1219 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1220 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 1221 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1222 p.SetHeadersIdle(accepted, deliveryTime) 1223 } 1224 ) 1225 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1226 d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve, 1227 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1228 1229 log.Debug("Skeleton fill terminated", "err", err) 1230 1231 filled, proced := d.queue.RetrieveHeaders() 1232 return filled, proced, err 1233 } 1234 1235 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1236 // available peers, reserving a chunk of blocks for each, waiting for delivery 1237 // and also periodically checking for timeouts. 1238 func (d *Downloader) fetchBodies(from uint64) error { 1239 log.Debug("Downloading block bodies", "origin", from) 1240 1241 var ( 1242 deliver = func(packet dataPack) (int, error) { 1243 pack := packet.(*bodyPack) 1244 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1245 } 1246 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 1247 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1248 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 1249 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) } 1250 ) 1251 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1252 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies, 1253 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1254 1255 log.Debug("Block body download terminated", "err", err) 1256 return err 1257 } 1258 1259 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1260 // available peers, reserving a chunk of receipts for each, waiting for delivery 1261 // and also periodically checking for timeouts. 1262 func (d *Downloader) fetchReceipts(from uint64) error { 1263 log.Debug("Downloading transaction receipts", "origin", from) 1264 1265 var ( 1266 deliver = func(packet dataPack) (int, error) { 1267 pack := packet.(*receiptPack) 1268 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1269 } 1270 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1271 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1272 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1273 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1274 p.SetReceiptsIdle(accepted, deliveryTime) 1275 } 1276 ) 1277 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1278 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts, 1279 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1280 1281 log.Debug("Transaction receipt download terminated", "err", err) 1282 return err 1283 } 1284 1285 // fetchParts iteratively downloads scheduled block parts, taking any available 1286 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1287 // also periodically checking for timeouts. 1288 // 1289 // As the scheduling/timeout logic mostly is the same for all downloaded data 1290 // types, this method is used by each for data gathering and is instrumented with 1291 // various callbacks to handle the slight differences between processing them. 1292 // 1293 // The instrumentation parameters: 1294 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1295 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1296 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1297 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1298 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1299 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1300 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1301 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1302 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1303 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1304 // - fetch: network callback to actually send a particular download request to a physical remote peer 1305 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1306 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1307 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1308 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1309 // - kind: textual label of the type being downloaded to display in log messages 1310 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1311 expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool), 1312 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1313 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error { 1314 1315 // Create a ticker to detect expired retrieval tasks 1316 ticker := time.NewTicker(100 * time.Millisecond) 1317 defer ticker.Stop() 1318 1319 update := make(chan struct{}, 1) 1320 1321 // Prepare the queue and fetch block parts until the block header fetcher's done 1322 finished := false 1323 for { 1324 select { 1325 case <-d.cancelCh: 1326 return errCanceled 1327 1328 case packet := <-deliveryCh: 1329 deliveryTime := time.Now() 1330 // If the peer was previously banned and failed to deliver its pack 1331 // in a reasonable time frame, ignore its message. 1332 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1333 // Deliver the received chunk of data and check chain validity 1334 accepted, err := deliver(packet) 1335 if errors.Is(err, errInvalidChain) { 1336 return err 1337 } 1338 // Unless a peer delivered something completely else than requested (usually 1339 // caused by a timed out request which came through in the end), set it to 1340 // idle. If the delivery's stale, the peer should have already been idled. 1341 if !errors.Is(err, errStaleDelivery) { 1342 setIdle(peer, accepted, deliveryTime) 1343 } 1344 // Issue a log to the user to see what's going on 1345 switch { 1346 case err == nil && packet.Items() == 0: 1347 peer.log.Trace("Requested data not delivered", "type", kind) 1348 case err == nil: 1349 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1350 default: 1351 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1352 } 1353 } 1354 // Blocks assembled, try to update the progress 1355 select { 1356 case update <- struct{}{}: 1357 default: 1358 } 1359 1360 case cont := <-wakeCh: 1361 // The header fetcher sent a continuation flag, check if it's done 1362 if !cont { 1363 finished = true 1364 } 1365 // Headers arrive, try to update the progress 1366 select { 1367 case update <- struct{}{}: 1368 default: 1369 } 1370 1371 case <-ticker.C: 1372 // Sanity check update the progress 1373 select { 1374 case update <- struct{}{}: 1375 default: 1376 } 1377 1378 case <-update: 1379 // Short circuit if we lost all our peers 1380 if d.peers.Len() == 0 { 1381 return errNoPeers 1382 } 1383 // Check for fetch request timeouts and demote the responsible peers 1384 for pid, fails := range expire() { 1385 if peer := d.peers.Peer(pid); peer != nil { 1386 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1387 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1388 // out that sync wise we need to get rid of the peer. 1389 // 1390 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1391 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1392 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1393 if fails > 2 { 1394 peer.log.Trace("Data delivery timed out", "type", kind) 1395 setIdle(peer, 0, time.Now()) 1396 } else { 1397 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1398 1399 if d.dropPeer == nil { 1400 // The dropPeer method is nil when `--copydb` is used for a local copy. 1401 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1402 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1403 } else { 1404 d.dropPeer(pid) 1405 1406 // If this peer was the master peer, abort sync immediately 1407 d.cancelLock.RLock() 1408 master := pid == d.cancelPeer 1409 d.cancelLock.RUnlock() 1410 1411 if master { 1412 d.cancel() 1413 return errTimeout 1414 } 1415 } 1416 } 1417 } 1418 } 1419 // If there's nothing more to fetch, wait or terminate 1420 if pending() == 0 { 1421 if !inFlight() && finished { 1422 log.Debug("Data fetching completed", "type", kind) 1423 return nil 1424 } 1425 break 1426 } 1427 // Send a download request to all idle peers, until throttled 1428 progressed, throttled, running := false, false, inFlight() 1429 idles, total := idle() 1430 pendCount := pending() 1431 for _, peer := range idles { 1432 // Short circuit if throttling activated 1433 if throttled { 1434 break 1435 } 1436 // Short circuit if there is no more available task. 1437 if pendCount = pending(); pendCount == 0 { 1438 break 1439 } 1440 // Reserve a chunk of fetches for a peer. A nil can mean either that 1441 // no more headers are available, or that the peer is known not to 1442 // have them. 1443 request, progress, throttle := reserve(peer, capacity(peer)) 1444 if progress { 1445 progressed = true 1446 } 1447 if throttle { 1448 throttled = true 1449 throttleCounter.Inc(1) 1450 } 1451 if request == nil { 1452 continue 1453 } 1454 if request.From > 0 { 1455 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1456 } else { 1457 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1458 } 1459 // Fetch the chunk and make sure any errors return the hashes to the queue 1460 if fetchHook != nil { 1461 fetchHook(request.Headers) 1462 } 1463 if err := fetch(peer, request); err != nil { 1464 // Although we could try and make an attempt to fix this, this error really 1465 // means that we've double allocated a fetch task to a peer. If that is the 1466 // case, the internal state of the downloader and the queue is very wrong so 1467 // better hard crash and note the error instead of silently accumulating into 1468 // a much bigger issue. 1469 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1470 } 1471 running = true 1472 } 1473 // Make sure that we have peers available for fetching. If all peers have been tried 1474 // and all failed throw an error 1475 if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 { 1476 return errPeersUnavailable 1477 } 1478 } 1479 } 1480 } 1481 1482 // processHeaders takes batches of retrieved headers from an input channel and 1483 // keeps processing and scheduling them into the header chain and downloader's 1484 // queue until the stream ends or a failure occurs. 1485 func (d *Downloader) processHeaders(origin uint64, td *big.Int) error { 1486 // Keep a count of uncertain headers to roll back 1487 var ( 1488 rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis) 1489 rollbackErr error 1490 mode = d.getMode() 1491 ) 1492 defer func() { 1493 if rollback > 0 { 1494 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1495 if mode != LightSync { 1496 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1497 lastBlock = d.blockchain.CurrentBlock().Number() 1498 } 1499 if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block 1500 // We're already unwinding the stack, only print the error to make it more visible 1501 log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err) 1502 } 1503 curFastBlock, curBlock := common.Big0, common.Big0 1504 if mode != LightSync { 1505 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1506 curBlock = d.blockchain.CurrentBlock().Number() 1507 } 1508 log.Warn("Rolled back chain segment", 1509 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1510 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1511 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr) 1512 } 1513 }() 1514 // Wait for batches of headers to process 1515 gotHeaders := false 1516 1517 for { 1518 select { 1519 case <-d.cancelCh: 1520 rollbackErr = errCanceled 1521 return errCanceled 1522 1523 case headers := <-d.headerProcCh: 1524 // Terminate header processing if we synced up 1525 if len(headers) == 0 { 1526 // Notify everyone that headers are fully processed 1527 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1528 select { 1529 case ch <- false: 1530 case <-d.cancelCh: 1531 } 1532 } 1533 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1534 // better chain compared to ours. The only exception is if its promised blocks were 1535 // already imported by other means (e.g. fetcher): 1536 // 1537 // R <remote peer>, L <local node>: Both at block 10 1538 // R: Mine block 11, and propagate it to L 1539 // L: Queue block 11 for import 1540 // L: Notice that R's head and TD increased compared to ours, start sync 1541 // L: Import of block 11 finishes 1542 // L: Sync begins, and finds common ancestor at 11 1543 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1544 // R: Nothing to give 1545 if mode != LightSync { 1546 head := d.blockchain.CurrentBlock() 1547 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1548 return errStallingPeer 1549 } 1550 } 1551 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1552 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1553 // of delivering the post-pivot blocks that would flag the invalid content. 1554 // 1555 // This check cannot be executed "as is" for full imports, since blocks may still be 1556 // queued for processing when the header download completes. However, as long as the 1557 // peer gave us something useful, we're already happy/progressed (above check). 1558 if mode == FastSync || mode == LightSync { 1559 head := d.lightchain.CurrentHeader() 1560 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1561 return errStallingPeer 1562 } 1563 } 1564 // Disable any rollback and return 1565 rollback = 0 1566 return nil 1567 } 1568 // Otherwise split the chunk of headers into batches and process them 1569 gotHeaders = true 1570 for len(headers) > 0 { 1571 // Terminate if something failed in between processing chunks 1572 select { 1573 case <-d.cancelCh: 1574 rollbackErr = errCanceled 1575 return errCanceled 1576 default: 1577 } 1578 // Select the next chunk of headers to import 1579 limit := maxHeadersProcess 1580 if limit > len(headers) { 1581 limit = len(headers) 1582 } 1583 chunk := headers[:limit] 1584 1585 // In case of header only syncing, validate the chunk immediately 1586 if mode == FastSync || mode == LightSync { 1587 // If we're importing pure headers, verify based on their recentness 1588 var pivot uint64 1589 1590 d.pivotLock.RLock() 1591 if d.pivotHeader != nil { 1592 pivot = d.pivotHeader.Number.Uint64() 1593 } 1594 d.pivotLock.RUnlock() 1595 1596 frequency := fsHeaderCheckFrequency 1597 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1598 frequency = 1 1599 } 1600 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1601 rollbackErr = err 1602 1603 // If some headers were inserted, track them as uncertain 1604 if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 { 1605 rollback = chunk[0].Number.Uint64() 1606 } 1607 log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err) 1608 return fmt.Errorf("%w: %v", errInvalidChain, err) 1609 } 1610 // All verifications passed, track all headers within the alloted limits 1611 if mode == FastSync { 1612 head := chunk[len(chunk)-1].Number.Uint64() 1613 if head-rollback > uint64(fsHeaderSafetyNet) { 1614 rollback = head - uint64(fsHeaderSafetyNet) 1615 } else { 1616 rollback = 1 1617 } 1618 } 1619 } 1620 // Unless we're doing light chains, schedule the headers for associated content retrieval 1621 if mode == FullSync || mode == FastSync { 1622 // If we've reached the allowed number of pending headers, stall a bit 1623 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1624 select { 1625 case <-d.cancelCh: 1626 rollbackErr = errCanceled 1627 return errCanceled 1628 case <-time.After(time.Second): 1629 } 1630 } 1631 // Otherwise insert the headers for content retrieval 1632 inserts := d.queue.Schedule(chunk, origin) 1633 if len(inserts) != len(chunk) { 1634 rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk)) 1635 return fmt.Errorf("%w: stale headers", errBadPeer) 1636 } 1637 } 1638 headers = headers[limit:] 1639 origin += uint64(limit) 1640 } 1641 // Update the highest block number we know if a higher one is found. 1642 d.syncStatsLock.Lock() 1643 if d.syncStatsChainHeight < origin { 1644 d.syncStatsChainHeight = origin - 1 1645 } 1646 d.syncStatsLock.Unlock() 1647 1648 // Signal the content downloaders of the availablility of new tasks 1649 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1650 select { 1651 case ch <- true: 1652 default: 1653 } 1654 } 1655 } 1656 } 1657 } 1658 1659 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1660 func (d *Downloader) processFullSyncContent() error { 1661 for { 1662 results := d.queue.Results(true) 1663 if len(results) == 0 { 1664 return nil 1665 } 1666 if d.chainInsertHook != nil { 1667 d.chainInsertHook(results) 1668 } 1669 if err := d.importBlockResults(results); err != nil { 1670 return err 1671 } 1672 } 1673 } 1674 1675 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1676 // Check for any early termination requests 1677 if len(results) == 0 { 1678 return nil 1679 } 1680 select { 1681 case <-d.quitCh: 1682 return errCancelContentProcessing 1683 default: 1684 } 1685 // Retrieve the a batch of results to import 1686 first, last := results[0].Header, results[len(results)-1].Header 1687 log.Debug("Inserting downloaded chain", "items", len(results), 1688 "firstnum", first.Number, "firsthash", first.Hash(), 1689 "lastnum", last.Number, "lasthash", last.Hash(), 1690 ) 1691 blocks := make([]*types.Block, len(results)) 1692 for i, result := range results { 1693 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1694 } 1695 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1696 if index < len(results) { 1697 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1698 } else { 1699 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1700 // when it needs to preprocess blocks to import a sidechain. 1701 // The importer will put together a new list of blocks to import, which is a superset 1702 // of the blocks delivered from the downloader, and the indexing will be off. 1703 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1704 } 1705 return fmt.Errorf("%w: %v", errInvalidChain, err) 1706 } 1707 return nil 1708 } 1709 1710 // processFastSyncContent takes fetch results from the queue and writes them to the 1711 // database. It also controls the synchronisation of state nodes of the pivot block. 1712 func (d *Downloader) processFastSyncContent() error { 1713 // Start syncing state of the reported head block. This should get us most of 1714 // the state of the pivot block. 1715 d.pivotLock.RLock() 1716 sync := d.syncState(d.pivotHeader.Root) 1717 d.pivotLock.RUnlock() 1718 1719 defer func() { 1720 // The `sync` object is replaced every time the pivot moves. We need to 1721 // defer close the very last active one, hence the lazy evaluation vs. 1722 // calling defer sync.Cancel() !!! 1723 sync.Cancel() 1724 }() 1725 1726 closeOnErr := func(s *stateSync) { 1727 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled { 1728 d.queue.Close() // wake up Results 1729 } 1730 } 1731 go closeOnErr(sync) 1732 1733 // To cater for moving pivot points, track the pivot block and subsequently 1734 // accumulated download results separately. 1735 var ( 1736 oldPivot *fetchResult // Locked in pivot block, might change eventually 1737 oldTail []*fetchResult // Downloaded content after the pivot 1738 ) 1739 for { 1740 // Wait for the next batch of downloaded data to be available, and if the pivot 1741 // block became stale, move the goalpost 1742 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1743 if len(results) == 0 { 1744 // If pivot sync is done, stop 1745 if oldPivot == nil { 1746 return sync.Cancel() 1747 } 1748 // If sync failed, stop 1749 select { 1750 case <-d.cancelCh: 1751 sync.Cancel() 1752 return errCanceled 1753 default: 1754 } 1755 } 1756 if d.chainInsertHook != nil { 1757 d.chainInsertHook(results) 1758 } 1759 // If we haven't downloaded the pivot block yet, check pivot staleness 1760 // notifications from the header downloader 1761 d.pivotLock.RLock() 1762 pivot := d.pivotHeader 1763 d.pivotLock.RUnlock() 1764 1765 if oldPivot == nil { 1766 if pivot.Root != sync.root { 1767 sync.Cancel() 1768 sync = d.syncState(pivot.Root) 1769 1770 go closeOnErr(sync) 1771 } 1772 } else { 1773 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1774 } 1775 // Split around the pivot block and process the two sides via fast/full sync 1776 if atomic.LoadInt32(&d.committed) == 0 { 1777 latest := results[len(results)-1].Header 1778 // If the height is above the pivot block by 2 sets, it means the pivot 1779 // become stale in the network and it was garbage collected, move to a 1780 // new pivot. 1781 // 1782 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 1783 // need to be taken into account, otherwise we're detecting the pivot move 1784 // late and will drop peers due to unavailable state!!! 1785 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 1786 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 1787 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 1788 1789 d.pivotLock.Lock() 1790 d.pivotHeader = pivot 1791 d.pivotLock.Unlock() 1792 1793 // Write out the pivot into the database so a rollback beyond it will 1794 // reenable fast sync 1795 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 1796 } 1797 } 1798 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 1799 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1800 return err 1801 } 1802 if P != nil { 1803 // If new pivot block found, cancel old state retrieval and restart 1804 if oldPivot != P { 1805 sync.Cancel() 1806 sync = d.syncState(P.Header.Root) 1807 1808 go closeOnErr(sync) 1809 oldPivot = P 1810 } 1811 // Wait for completion, occasionally checking for pivot staleness 1812 select { 1813 case <-sync.done: 1814 if sync.err != nil { 1815 return sync.err 1816 } 1817 if err := d.commitPivotBlock(P); err != nil { 1818 return err 1819 } 1820 oldPivot = nil 1821 1822 case <-time.After(time.Second): 1823 oldTail = afterP 1824 continue 1825 } 1826 } 1827 // Fast sync done, pivot commit done, full import 1828 if err := d.importBlockResults(afterP); err != nil { 1829 return err 1830 } 1831 } 1832 } 1833 1834 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1835 if len(results) == 0 { 1836 return nil, nil, nil 1837 } 1838 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 1839 // the pivot is somewhere in the future 1840 return nil, results, nil 1841 } 1842 // This can also be optimized, but only happens very seldom 1843 for _, result := range results { 1844 num := result.Header.Number.Uint64() 1845 switch { 1846 case num < pivot: 1847 before = append(before, result) 1848 case num == pivot: 1849 p = result 1850 default: 1851 after = append(after, result) 1852 } 1853 } 1854 return p, before, after 1855 } 1856 1857 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1858 // Check for any early termination requests 1859 if len(results) == 0 { 1860 return nil 1861 } 1862 select { 1863 case <-d.quitCh: 1864 return errCancelContentProcessing 1865 case <-stateSync.done: 1866 if err := stateSync.Wait(); err != nil { 1867 return err 1868 } 1869 default: 1870 } 1871 // Retrieve the a batch of results to import 1872 first, last := results[0].Header, results[len(results)-1].Header 1873 log.Debug("Inserting fast-sync blocks", "items", len(results), 1874 "firstnum", first.Number, "firsthash", first.Hash(), 1875 "lastnumn", last.Number, "lasthash", last.Hash(), 1876 ) 1877 blocks := make([]*types.Block, len(results)) 1878 receipts := make([]types.Receipts, len(results)) 1879 for i, result := range results { 1880 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1881 receipts[i] = result.Receipts 1882 } 1883 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1884 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1885 return fmt.Errorf("%w: %v", errInvalidChain, err) 1886 } 1887 return nil 1888 } 1889 1890 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1891 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1892 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1893 1894 // Commit the pivot block as the new head, will require full sync from here on 1895 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1896 return err 1897 } 1898 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1899 return err 1900 } 1901 atomic.StoreInt32(&d.committed, 1) 1902 1903 // If we had a bloom filter for the state sync, deallocate it now. Note, we only 1904 // deallocate internally, but keep the empty wrapper. This ensures that if we do 1905 // a rollback after committing the pivot and restarting fast sync, we don't end 1906 // up using a nil bloom. Empty bloom is fine, it just returns that it does not 1907 // have the info we need, so reach down to the database instead. 1908 if d.stateBloom != nil { 1909 d.stateBloom.Close() 1910 } 1911 return nil 1912 } 1913 1914 // DeliverHeaders injects a new batch of block headers received from a remote 1915 // node into the download schedule. 1916 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) { 1917 return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1918 } 1919 1920 // DeliverBodies injects a new batch of block bodies received from a remote node. 1921 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) { 1922 return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1923 } 1924 1925 // DeliverReceipts injects a new batch of receipts received from a remote node. 1926 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) { 1927 return d.deliver(id, d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1928 } 1929 1930 // DeliverNodeData injects a new batch of node state data received from a remote node. 1931 func (d *Downloader) DeliverNodeData(id string, data [][]byte) (err error) { 1932 return d.deliver(id, d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1933 } 1934 1935 // deliver injects a new batch of data received from a remote node. 1936 func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 1937 // Update the delivery metrics for both good and failed deliveries 1938 inMeter.Mark(int64(packet.Items())) 1939 defer func() { 1940 if err != nil { 1941 dropMeter.Mark(int64(packet.Items())) 1942 } 1943 }() 1944 // Deliver or abort if the sync is canceled while queuing 1945 d.cancelLock.RLock() 1946 cancel := d.cancelCh 1947 d.cancelLock.RUnlock() 1948 if cancel == nil { 1949 return errNoSyncActive 1950 } 1951 select { 1952 case destCh <- packet: 1953 return nil 1954 case <-cancel: 1955 return errNoSyncActive 1956 } 1957 } 1958 1959 // qosTuner is the quality of service tuning loop that occasionally gathers the 1960 // peer latency statistics and updates the estimated request round trip time. 1961 func (d *Downloader) qosTuner() { 1962 for { 1963 // Retrieve the current median RTT and integrate into the previoust target RTT 1964 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 1965 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 1966 1967 // A new RTT cycle passed, increase our confidence in the estimated RTT 1968 conf := atomic.LoadUint64(&d.rttConfidence) 1969 conf = conf + (1000000-conf)/2 1970 atomic.StoreUint64(&d.rttConfidence, conf) 1971 1972 // Log the new QoS values and sleep until the next RTT 1973 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 1974 select { 1975 case <-d.quitCh: 1976 return 1977 case <-time.After(rtt): 1978 } 1979 } 1980 } 1981 1982 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 1983 // peer set, needing to reduce the confidence we have in out QoS estimates. 1984 func (d *Downloader) qosReduceConfidence() { 1985 // If we have a single peer, confidence is always 1 1986 peers := uint64(d.peers.Len()) 1987 if peers == 0 { 1988 // Ensure peer connectivity races don't catch us off guard 1989 return 1990 } 1991 if peers == 1 { 1992 atomic.StoreUint64(&d.rttConfidence, 1000000) 1993 return 1994 } 1995 // If we have a ton of peers, don't drop confidence) 1996 if peers >= uint64(qosConfidenceCap) { 1997 return 1998 } 1999 // Otherwise drop the confidence factor 2000 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 2001 if float64(conf)/1000000 < rttMinConfidence { 2002 conf = uint64(rttMinConfidence * 1000000) 2003 } 2004 atomic.StoreUint64(&d.rttConfidence, conf) 2005 2006 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2007 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 2008 } 2009 2010 // requestRTT returns the current target round trip time for a download request 2011 // to complete in. 2012 // 2013 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 2014 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 2015 // be adapted to, but smaller ones are preferred (stabler download stream). 2016 func (d *Downloader) requestRTT() time.Duration { 2017 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 2018 } 2019 2020 // requestTTL returns the current timeout allowance for a single download request 2021 // to finish under. 2022 func (d *Downloader) requestTTL() time.Duration { 2023 var ( 2024 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2025 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 2026 ) 2027 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 2028 if ttl > ttlLimit { 2029 ttl = ttlLimit 2030 } 2031 return ttl 2032 }