github.com/cryptogateway/go-paymex@v0.0.0-20210204174735-96277fb1e602/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/cryptogateway/go-paymex" 29 "github.com/cryptogateway/go-paymex/common" 30 "github.com/cryptogateway/go-paymex/core/rawdb" 31 "github.com/cryptogateway/go-paymex/core/types" 32 "github.com/cryptogateway/go-paymex/eth/protocols/snap" 33 "github.com/cryptogateway/go-paymex/ethdb" 34 "github.com/cryptogateway/go-paymex/event" 35 "github.com/cryptogateway/go-paymex/log" 36 "github.com/cryptogateway/go-paymex/metrics" 37 "github.com/cryptogateway/go-paymex/params" 38 "github.com/cryptogateway/go-paymex/trie" 39 ) 40 41 var ( 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's protocol version too old") 93 errNoAncestorFound = errors.New("no common ancestor found") 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 ethdb.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 // Channel receiving inbound block headers 136 bodyCh chan dataPack // Channel receiving inbound block bodies 137 receiptCh chan dataPack // Channel receiving inbound receipts 138 bodyWakeCh chan bool // Channel to signal the block body fetcher of new tasks 139 receiptWakeCh chan bool // Channel to signal the receipt fetcher of new tasks 140 headerProcCh chan []*types.Header // 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 snapSync bool // Whether to run state sync over the snap protocol 147 SnapSyncer *snap.Syncer // TODO(karalabe): make private! hack for now 148 stateSyncStart chan *stateSync 149 trackStateReq chan *stateReq 150 stateCh chan dataPack // Channel receiving inbound node state data 151 152 // Cancellation and termination 153 cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop) 154 cancelCh chan struct{} // Channel to cancel mid-flight syncs 155 cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers 156 cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited. 157 158 quitCh chan struct{} // Quit channel to signal termination 159 quitLock sync.Mutex // Lock to prevent double closes 160 161 // Testing hooks 162 syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run 163 bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch 164 receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch 165 chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations) 166 } 167 168 // LightChain encapsulates functions required to synchronise a light chain. 169 type LightChain interface { 170 // HasHeader verifies a header's presence in the local chain. 171 HasHeader(common.Hash, uint64) bool 172 173 // GetHeaderByHash retrieves a header from the local chain. 174 GetHeaderByHash(common.Hash) *types.Header 175 176 // CurrentHeader retrieves the head header from the local chain. 177 CurrentHeader() *types.Header 178 179 // GetTd returns the total difficulty of a local block. 180 GetTd(common.Hash, uint64) *big.Int 181 182 // InsertHeaderChain inserts a batch of headers into the local chain. 183 InsertHeaderChain([]*types.Header, int) (int, error) 184 185 // SetHead rewinds the local chain to a new head. 186 SetHead(uint64) error 187 } 188 189 // BlockChain encapsulates functions required to sync a (full or fast) blockchain. 190 type BlockChain interface { 191 LightChain 192 193 // HasBlock verifies a block's presence in the local chain. 194 HasBlock(common.Hash, uint64) bool 195 196 // HasFastBlock verifies a fast block's presence in the local chain. 197 HasFastBlock(common.Hash, uint64) bool 198 199 // GetBlockByHash retrieves a block from the local chain. 200 GetBlockByHash(common.Hash) *types.Block 201 202 // CurrentBlock retrieves the head block from the local chain. 203 CurrentBlock() *types.Block 204 205 // CurrentFastBlock retrieves the head fast block from the local chain. 206 CurrentFastBlock() *types.Block 207 208 // FastSyncCommitHead directly commits the head block to a certain entity. 209 FastSyncCommitHead(common.Hash) error 210 211 // InsertChain inserts a batch of blocks into the local chain. 212 InsertChain(types.Blocks) (int, error) 213 214 // InsertReceiptChain inserts a batch of receipts into the local chain. 215 InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error) 216 } 217 218 // New creates a new downloader to fetch hashes and blocks from remote peers. 219 func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader { 220 if lightchain == nil { 221 lightchain = chain 222 } 223 dl := &Downloader{ 224 stateDB: stateDb, 225 stateBloom: stateBloom, 226 mux: mux, 227 checkpoint: checkpoint, 228 queue: newQueue(blockCacheMaxItems, blockCacheInitialItems), 229 peers: newPeerSet(), 230 rttEstimate: uint64(rttMaxEstimate), 231 rttConfidence: uint64(1000000), 232 blockchain: chain, 233 lightchain: lightchain, 234 dropPeer: dropPeer, 235 headerCh: make(chan dataPack, 1), 236 bodyCh: make(chan dataPack, 1), 237 receiptCh: make(chan dataPack, 1), 238 bodyWakeCh: make(chan bool, 1), 239 receiptWakeCh: make(chan bool, 1), 240 headerProcCh: make(chan []*types.Header, 1), 241 quitCh: make(chan struct{}), 242 stateCh: make(chan dataPack), 243 SnapSyncer: snap.NewSyncer(stateDb, stateBloom), 244 stateSyncStart: make(chan *stateSync), 245 syncStatsState: stateSyncStats{ 246 processed: rawdb.ReadFastTrieProgress(stateDb), 247 }, 248 trackStateReq: make(chan *stateReq), 249 } 250 go dl.qosTuner() 251 go dl.stateFetcher() 252 return dl 253 } 254 255 // Progress retrieves the synchronisation boundaries, specifically the origin 256 // block where synchronisation started at (may have failed/suspended); the block 257 // or header sync is currently at; and the latest known block which the sync targets. 258 // 259 // In addition, during the state download phase of fast synchronisation the number 260 // of processed and the total number of known states are also returned. Otherwise 261 // these are zero. 262 func (d *Downloader) Progress() ethereum.SyncProgress { 263 // Lock the current stats and return the progress 264 d.syncStatsLock.RLock() 265 defer d.syncStatsLock.RUnlock() 266 267 current := uint64(0) 268 mode := d.getMode() 269 switch { 270 case d.blockchain != nil && mode == FullSync: 271 current = d.blockchain.CurrentBlock().NumberU64() 272 case d.blockchain != nil && mode == FastSync: 273 current = d.blockchain.CurrentFastBlock().NumberU64() 274 case d.lightchain != nil: 275 current = d.lightchain.CurrentHeader().Number.Uint64() 276 default: 277 log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", mode) 278 } 279 return ethereum.SyncProgress{ 280 StartingBlock: d.syncStatsChainOrigin, 281 CurrentBlock: current, 282 HighestBlock: d.syncStatsChainHeight, 283 PulledStates: d.syncStatsState.processed, 284 KnownStates: d.syncStatsState.processed + d.syncStatsState.pending, 285 } 286 } 287 288 // Synchronising returns whether the downloader is currently retrieving blocks. 289 func (d *Downloader) Synchronising() bool { 290 return atomic.LoadInt32(&d.synchronising) > 0 291 } 292 293 // RegisterPeer injects a new download peer into the set of block source to be 294 // used for fetching hashes and blocks from. 295 func (d *Downloader) RegisterPeer(id string, version uint, peer Peer) error { 296 var logger log.Logger 297 if len(id) < 16 { 298 // Tests use short IDs, don't choke on them 299 logger = log.New("peer", id) 300 } else { 301 logger = log.New("peer", id[:8]) 302 } 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 uint, 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 var logger log.Logger 324 if len(id) < 16 { 325 // Tests use short IDs, don't choke on them 326 logger = log.New("peer", id) 327 } else { 328 logger = log.New("peer", id[:8]) 329 } 330 logger.Trace("Unregistering sync peer") 331 if err := d.peers.Unregister(id); err != nil { 332 logger.Error("Failed to unregister sync peer", "err", err) 333 return err 334 } 335 d.queue.Revoke(id) 336 337 return nil 338 } 339 340 // Synchronise tries to sync up our local block chain with a remote peer, both 341 // adding various sanity checks as well as wrapping it with various log entries. 342 func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error { 343 err := d.synchronise(id, head, td, mode) 344 345 switch err { 346 case nil, errBusy, errCanceled: 347 return err 348 } 349 350 if errors.Is(err, errInvalidChain) || errors.Is(err, errBadPeer) || errors.Is(err, errTimeout) || 351 errors.Is(err, errStallingPeer) || errors.Is(err, errUnsyncedPeer) || errors.Is(err, errEmptyHeaderSet) || 352 errors.Is(err, errPeersUnavailable) || errors.Is(err, errTooOld) || errors.Is(err, errInvalidAncestor) { 353 log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err) 354 if d.dropPeer == nil { 355 // The dropPeer method is nil when `--copydb` is used for a local copy. 356 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 357 log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id) 358 } else { 359 d.dropPeer(id) 360 } 361 return err 362 } 363 log.Warn("Synchronisation failed, retrying", "err", err) 364 return err 365 } 366 367 // synchronise will select the peer and use it for synchronising. If an empty string is given 368 // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the 369 // checks fail an error will be returned. This method is synchronous 370 func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error { 371 // Mock out the synchronisation if testing 372 if d.synchroniseMock != nil { 373 return d.synchroniseMock(id, hash) 374 } 375 // Make sure only one goroutine is ever allowed past this point at once 376 if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) { 377 return errBusy 378 } 379 defer atomic.StoreInt32(&d.synchronising, 0) 380 381 // Post a user notification of the sync (only once per session) 382 if atomic.CompareAndSwapInt32(&d.notified, 0, 1) { 383 log.Info("Block synchronisation started") 384 } 385 // If we are already full syncing, but have a fast-sync bloom filter laying 386 // around, make sure it doesn't use memory any more. This is a special case 387 // when the user attempts to fast sync a new empty network. 388 if mode == FullSync && d.stateBloom != nil { 389 d.stateBloom.Close() 390 } 391 // If snap sync was requested, create the snap scheduler and switch to fast 392 // sync mode. Long term we could drop fast sync or merge the two together, 393 // but until snap becomes prevalent, we should support both. TODO(karalabe). 394 if mode == SnapSync { 395 if !d.snapSync { 396 log.Warn("Enabling snapshot sync prototype") 397 d.snapSync = true 398 } 399 mode = FastSync 400 } 401 // Reset the queue, peer set and wake channels to clean any internal leftover state 402 d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems) 403 d.peers.Reset() 404 405 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 406 select { 407 case <-ch: 408 default: 409 } 410 } 411 for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} { 412 for empty := false; !empty; { 413 select { 414 case <-ch: 415 default: 416 empty = true 417 } 418 } 419 } 420 for empty := false; !empty; { 421 select { 422 case <-d.headerProcCh: 423 default: 424 empty = true 425 } 426 } 427 // Create cancel channel for aborting mid-flight and mark the master peer 428 d.cancelLock.Lock() 429 d.cancelCh = make(chan struct{}) 430 d.cancelPeer = id 431 d.cancelLock.Unlock() 432 433 defer d.Cancel() // No matter what, we can't leave the cancel channel open 434 435 // Atomically set the requested sync mode 436 atomic.StoreUint32(&d.mode, uint32(mode)) 437 438 // Retrieve the origin peer and initiate the downloading process 439 p := d.peers.Peer(id) 440 if p == nil { 441 return errUnknownPeer 442 } 443 return d.syncWithPeer(p, hash, td) 444 } 445 446 func (d *Downloader) getMode() SyncMode { 447 return SyncMode(atomic.LoadUint32(&d.mode)) 448 } 449 450 // syncWithPeer starts a block synchronization based on the hash chain from the 451 // specified peer and head hash. 452 func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) { 453 d.mux.Post(StartEvent{}) 454 defer func() { 455 // reset on error 456 if err != nil { 457 d.mux.Post(FailedEvent{err}) 458 } else { 459 latest := d.lightchain.CurrentHeader() 460 d.mux.Post(DoneEvent{latest}) 461 } 462 }() 463 if p.version < 64 { 464 return fmt.Errorf("%w: advertized %d < required %d", errTooOld, p.version, 64) 465 } 466 mode := d.getMode() 467 468 log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", mode) 469 defer func(start time.Time) { 470 log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start))) 471 }(time.Now()) 472 473 // Look up the sync boundaries: the common ancestor and the target block 474 latest, pivot, err := d.fetchHead(p) 475 if err != nil { 476 return err 477 } 478 if mode == FastSync && pivot == nil { 479 // If no pivot block was returned, the head is below the min full block 480 // threshold (i.e. new chian). In that case we won't really fast sync 481 // anyway, but still need a valid pivot block to avoid some code hitting 482 // nil panics on an access. 483 pivot = d.blockchain.CurrentBlock().Header() 484 } 485 height := latest.Number.Uint64() 486 487 origin, err := d.findAncestor(p, latest) 488 if err != nil { 489 return err 490 } 491 d.syncStatsLock.Lock() 492 if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin { 493 d.syncStatsChainOrigin = origin 494 } 495 d.syncStatsChainHeight = height 496 d.syncStatsLock.Unlock() 497 498 // Ensure our origin point is below any fast sync pivot point 499 if mode == FastSync { 500 if height <= uint64(fsMinFullBlocks) { 501 origin = 0 502 } else { 503 pivotNumber := pivot.Number.Uint64() 504 if pivotNumber <= origin { 505 origin = pivotNumber - 1 506 } 507 // Write out the pivot into the database so a rollback beyond it will 508 // reenable fast sync 509 rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber) 510 } 511 } 512 d.committed = 1 513 if mode == FastSync && pivot.Number.Uint64() != 0 { 514 d.committed = 0 515 } 516 if mode == FastSync { 517 // Set the ancient data limitation. 518 // If we are running fast sync, all block data older than ancientLimit will be 519 // written to the ancient store. More recent data will be written to the active 520 // database and will wait for the freezer to migrate. 521 // 522 // If there is a checkpoint available, then calculate the ancientLimit through 523 // that. Otherwise calculate the ancient limit through the advertised height 524 // of the remote peer. 525 // 526 // The reason for picking checkpoint first is that a malicious peer can give us 527 // a fake (very high) height, forcing the ancient limit to also be very high. 528 // The peer would start to feed us valid blocks until head, resulting in all of 529 // the blocks might be written into the ancient store. A following mini-reorg 530 // could cause issues. 531 if d.checkpoint != 0 && d.checkpoint > fullMaxForkAncestry+1 { 532 d.ancientLimit = d.checkpoint 533 } else if height > fullMaxForkAncestry+1 { 534 d.ancientLimit = height - fullMaxForkAncestry - 1 535 } else { 536 d.ancientLimit = 0 537 } 538 frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here. 539 540 // If a part of blockchain data has already been written into active store, 541 // disable the ancient style insertion explicitly. 542 if origin >= frozen && frozen != 0 { 543 d.ancientLimit = 0 544 log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1) 545 } else if d.ancientLimit > 0 { 546 log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit) 547 } 548 // Rewind the ancient store and blockchain if reorg happens. 549 if origin+1 < frozen { 550 if err := d.lightchain.SetHead(origin + 1); err != nil { 551 return err 552 } 553 } 554 } 555 // Initiate the sync using a concurrent header and content retrieval algorithm 556 d.queue.Prepare(origin+1, mode) 557 if d.syncInitHook != nil { 558 d.syncInitHook(origin, height) 559 } 560 fetchers := []func() error{ 561 func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved 562 func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync 563 func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync 564 func() error { return d.processHeaders(origin+1, td) }, 565 } 566 if mode == FastSync { 567 d.pivotLock.Lock() 568 d.pivotHeader = pivot 569 d.pivotLock.Unlock() 570 571 fetchers = append(fetchers, func() error { return d.processFastSyncContent() }) 572 } else if mode == FullSync { 573 fetchers = append(fetchers, d.processFullSyncContent) 574 } 575 return d.spawnSync(fetchers) 576 } 577 578 // spawnSync runs d.process and all given fetcher functions to completion in 579 // separate goroutines, returning the first error that appears. 580 func (d *Downloader) spawnSync(fetchers []func() error) error { 581 errc := make(chan error, len(fetchers)) 582 d.cancelWg.Add(len(fetchers)) 583 for _, fn := range fetchers { 584 fn := fn 585 go func() { defer d.cancelWg.Done(); errc <- fn() }() 586 } 587 // Wait for the first error, then terminate the others. 588 var err error 589 for i := 0; i < len(fetchers); i++ { 590 if i == len(fetchers)-1 { 591 // Close the queue when all fetchers have exited. 592 // This will cause the block processor to end when 593 // it has processed the queue. 594 d.queue.Close() 595 } 596 if err = <-errc; err != nil && err != errCanceled { 597 break 598 } 599 } 600 d.queue.Close() 601 d.Cancel() 602 return err 603 } 604 605 // cancel aborts all of the operations and resets the queue. However, cancel does 606 // not wait for the running download goroutines to finish. This method should be 607 // used when cancelling the downloads from inside the downloader. 608 func (d *Downloader) cancel() { 609 // Close the current cancel channel 610 d.cancelLock.Lock() 611 defer d.cancelLock.Unlock() 612 613 if d.cancelCh != nil { 614 select { 615 case <-d.cancelCh: 616 // Channel was already closed 617 default: 618 close(d.cancelCh) 619 } 620 } 621 } 622 623 // Cancel aborts all of the operations and waits for all download goroutines to 624 // finish before returning. 625 func (d *Downloader) Cancel() { 626 d.cancel() 627 d.cancelWg.Wait() 628 } 629 630 // Terminate interrupts the downloader, canceling all pending operations. 631 // The downloader cannot be reused after calling Terminate. 632 func (d *Downloader) Terminate() { 633 // Close the termination channel (make sure double close is allowed) 634 d.quitLock.Lock() 635 select { 636 case <-d.quitCh: 637 default: 638 close(d.quitCh) 639 } 640 if d.stateBloom != nil { 641 d.stateBloom.Close() 642 } 643 d.quitLock.Unlock() 644 645 // Cancel any pending download requests 646 d.Cancel() 647 } 648 649 // fetchHead retrieves the head header and prior pivot block (if available) from 650 // a remote peer. 651 func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *types.Header, err error) { 652 p.log.Debug("Retrieving remote chain head") 653 mode := d.getMode() 654 655 // Request the advertised remote head block and wait for the response 656 latest, _ := p.peer.Head() 657 fetch := 1 658 if mode == FastSync { 659 fetch = 2 // head + pivot headers 660 } 661 go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true) 662 663 ttl := d.requestTTL() 664 timeout := time.After(ttl) 665 for { 666 select { 667 case <-d.cancelCh: 668 return nil, nil, errCanceled 669 670 case packet := <-d.headerCh: 671 // Discard anything not from the origin peer 672 if packet.PeerId() != p.id { 673 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 674 break 675 } 676 // Make sure the peer gave us at least one and at most the requested headers 677 headers := packet.(*headerPack).headers 678 if len(headers) == 0 || len(headers) > fetch { 679 return nil, nil, fmt.Errorf("%w: returned headers %d != requested %d", errBadPeer, len(headers), fetch) 680 } 681 // The first header needs to be the head, validate against the checkpoint 682 // and request. If only 1 header was returned, make sure there's no pivot 683 // or there was not one requested. 684 head := headers[0] 685 if (mode == FastSync || mode == LightSync) && head.Number.Uint64() < d.checkpoint { 686 return nil, nil, fmt.Errorf("%w: remote head %d below checkpoint %d", errUnsyncedPeer, head.Number, d.checkpoint) 687 } 688 if len(headers) == 1 { 689 if mode == FastSync && head.Number.Uint64() > uint64(fsMinFullBlocks) { 690 return nil, nil, fmt.Errorf("%w: no pivot included along head header", errBadPeer) 691 } 692 p.log.Debug("Remote head identified, no pivot", "number", head.Number, "hash", head.Hash()) 693 return head, nil, nil 694 } 695 // At this point we have 2 headers in total and the first is the 696 // validated head of the chian. Check the pivot number and return, 697 pivot := headers[1] 698 if pivot.Number.Uint64() != head.Number.Uint64()-uint64(fsMinFullBlocks) { 699 return nil, nil, fmt.Errorf("%w: remote pivot %d != requested %d", errInvalidChain, pivot.Number, head.Number.Uint64()-uint64(fsMinFullBlocks)) 700 } 701 return head, pivot, nil 702 703 case <-timeout: 704 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 705 return nil, nil, errTimeout 706 707 case <-d.bodyCh: 708 case <-d.receiptCh: 709 // Out of bounds delivery, ignore 710 } 711 } 712 } 713 714 // calculateRequestSpan calculates what headers to request from a peer when trying to determine the 715 // common ancestor. 716 // It returns parameters to be used for peer.RequestHeadersByNumber: 717 // from - starting block number 718 // count - number of headers to request 719 // skip - number of headers to skip 720 // and also returns 'max', the last block which is expected to be returned by the remote peers, 721 // given the (from,count,skip) 722 func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) { 723 var ( 724 from int 725 count int 726 MaxCount = MaxHeaderFetch / 16 727 ) 728 // requestHead is the highest block that we will ask for. If requestHead is not offset, 729 // the highest block that we will get is 16 blocks back from head, which means we 730 // will fetch 14 or 15 blocks unnecessarily in the case the height difference 731 // between us and the peer is 1-2 blocks, which is most common 732 requestHead := int(remoteHeight) - 1 733 if requestHead < 0 { 734 requestHead = 0 735 } 736 // requestBottom is the lowest block we want included in the query 737 // Ideally, we want to include the one just below our own head 738 requestBottom := int(localHeight - 1) 739 if requestBottom < 0 { 740 requestBottom = 0 741 } 742 totalSpan := requestHead - requestBottom 743 span := 1 + totalSpan/MaxCount 744 if span < 2 { 745 span = 2 746 } 747 if span > 16 { 748 span = 16 749 } 750 751 count = 1 + totalSpan/span 752 if count > MaxCount { 753 count = MaxCount 754 } 755 if count < 2 { 756 count = 2 757 } 758 from = requestHead - (count-1)*span 759 if from < 0 { 760 from = 0 761 } 762 max := from + (count-1)*span 763 return int64(from), count, span - 1, uint64(max) 764 } 765 766 // findAncestor tries to locate the common ancestor link of the local chain and 767 // a remote peers blockchain. In the general case when our node was in sync and 768 // on the correct chain, checking the top N links should already get us a match. 769 // In the rare scenario when we ended up on a long reorganisation (i.e. none of 770 // the head links match), we do a binary search to find the common ancestor. 771 func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) { 772 // Figure out the valid ancestor range to prevent rewrite attacks 773 var ( 774 floor = int64(-1) 775 localHeight uint64 776 remoteHeight = remoteHeader.Number.Uint64() 777 ) 778 mode := d.getMode() 779 switch mode { 780 case FullSync: 781 localHeight = d.blockchain.CurrentBlock().NumberU64() 782 case FastSync: 783 localHeight = d.blockchain.CurrentFastBlock().NumberU64() 784 default: 785 localHeight = d.lightchain.CurrentHeader().Number.Uint64() 786 } 787 p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight) 788 789 // Recap floor value for binary search 790 maxForkAncestry := fullMaxForkAncestry 791 if d.getMode() == LightSync { 792 maxForkAncestry = lightMaxForkAncestry 793 } 794 if localHeight >= maxForkAncestry { 795 // We're above the max reorg threshold, find the earliest fork point 796 floor = int64(localHeight - maxForkAncestry) 797 } 798 // If we're doing a light sync, ensure the floor doesn't go below the CHT, as 799 // all headers before that point will be missing. 800 if mode == LightSync { 801 // If we don't know the current CHT position, find it 802 if d.genesis == 0 { 803 header := d.lightchain.CurrentHeader() 804 for header != nil { 805 d.genesis = header.Number.Uint64() 806 if floor >= int64(d.genesis)-1 { 807 break 808 } 809 header = d.lightchain.GetHeaderByHash(header.ParentHash) 810 } 811 } 812 // We already know the "genesis" block number, cap floor to that 813 if floor < int64(d.genesis)-1 { 814 floor = int64(d.genesis) - 1 815 } 816 } 817 818 ancestor, err := d.findAncestorSpanSearch(p, mode, remoteHeight, localHeight, floor) 819 if err == nil { 820 return ancestor, nil 821 } 822 // The returned error was not nil. 823 // If the error returned does not reflect that a common ancestor was not found, return it. 824 // If the error reflects that a common ancestor was not found, continue to binary search, 825 // where the error value will be reassigned. 826 if !errors.Is(err, errNoAncestorFound) { 827 return 0, err 828 } 829 830 ancestor, err = d.findAncestorBinarySearch(p, mode, remoteHeight, floor) 831 if err != nil { 832 return 0, err 833 } 834 return ancestor, nil 835 } 836 837 func (d *Downloader) findAncestorSpanSearch(p *peerConnection, mode SyncMode, remoteHeight, localHeight uint64, floor int64) (commonAncestor uint64, err error) { 838 from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight) 839 840 p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip) 841 go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false) 842 843 // Wait for the remote response to the head fetch 844 number, hash := uint64(0), common.Hash{} 845 846 ttl := d.requestTTL() 847 timeout := time.After(ttl) 848 849 for finished := false; !finished; { 850 select { 851 case <-d.cancelCh: 852 return 0, errCanceled 853 854 case packet := <-d.headerCh: 855 // Discard anything not from the origin peer 856 if packet.PeerId() != p.id { 857 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 858 break 859 } 860 // Make sure the peer actually gave something valid 861 headers := packet.(*headerPack).headers 862 if len(headers) == 0 { 863 p.log.Warn("Empty head header set") 864 return 0, errEmptyHeaderSet 865 } 866 // Make sure the peer's reply conforms to the request 867 for i, header := range headers { 868 expectNumber := from + int64(i)*int64(skip+1) 869 if number := header.Number.Int64(); number != expectNumber { 870 p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number) 871 return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering")) 872 } 873 } 874 // Check if a common ancestor was found 875 finished = true 876 for i := len(headers) - 1; i >= 0; i-- { 877 // Skip any headers that underflow/overflow our requested set 878 if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max { 879 continue 880 } 881 // Otherwise check if we already know the header or not 882 h := headers[i].Hash() 883 n := headers[i].Number.Uint64() 884 885 var known bool 886 switch mode { 887 case FullSync: 888 known = d.blockchain.HasBlock(h, n) 889 case FastSync: 890 known = d.blockchain.HasFastBlock(h, n) 891 default: 892 known = d.lightchain.HasHeader(h, n) 893 } 894 if known { 895 number, hash = n, h 896 break 897 } 898 } 899 900 case <-timeout: 901 p.log.Debug("Waiting for head header timed out", "elapsed", ttl) 902 return 0, errTimeout 903 904 case <-d.bodyCh: 905 case <-d.receiptCh: 906 // Out of bounds delivery, ignore 907 } 908 } 909 // If the head fetch already found an ancestor, return 910 if hash != (common.Hash{}) { 911 if int64(number) <= floor { 912 p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor) 913 return 0, errInvalidAncestor 914 } 915 p.log.Debug("Found common ancestor", "number", number, "hash", hash) 916 return number, nil 917 } 918 return 0, errNoAncestorFound 919 } 920 921 func (d *Downloader) findAncestorBinarySearch(p *peerConnection, mode SyncMode, remoteHeight uint64, floor int64) (commonAncestor uint64, err error) { 922 hash := common.Hash{} 923 924 // Ancestor not found, we need to binary search over our chain 925 start, end := uint64(0), remoteHeight 926 if floor > 0 { 927 start = uint64(floor) 928 } 929 p.log.Trace("Binary searching for common ancestor", "start", start, "end", end) 930 931 for start+1 < end { 932 // Split our chain interval in two, and request the hash to cross check 933 check := (start + end) / 2 934 935 ttl := d.requestTTL() 936 timeout := time.After(ttl) 937 938 go p.peer.RequestHeadersByNumber(check, 1, 0, false) 939 940 // Wait until a reply arrives to this request 941 for arrived := false; !arrived; { 942 select { 943 case <-d.cancelCh: 944 return 0, errCanceled 945 946 case packet := <-d.headerCh: 947 // Discard anything not from the origin peer 948 if packet.PeerId() != p.id { 949 log.Debug("Received headers from incorrect peer", "peer", packet.PeerId()) 950 break 951 } 952 // Make sure the peer actually gave something valid 953 headers := packet.(*headerPack).headers 954 if len(headers) != 1 { 955 p.log.Warn("Multiple headers for single request", "headers", len(headers)) 956 return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers)) 957 } 958 arrived = true 959 960 // Modify the search interval based on the response 961 h := headers[0].Hash() 962 n := headers[0].Number.Uint64() 963 964 var known bool 965 switch mode { 966 case FullSync: 967 known = d.blockchain.HasBlock(h, n) 968 case FastSync: 969 known = d.blockchain.HasFastBlock(h, n) 970 default: 971 known = d.lightchain.HasHeader(h, n) 972 } 973 if !known { 974 end = check 975 break 976 } 977 header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists 978 if header.Number.Uint64() != check { 979 p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check) 980 return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number) 981 } 982 start = check 983 hash = h 984 985 case <-timeout: 986 p.log.Debug("Waiting for search header timed out", "elapsed", ttl) 987 return 0, errTimeout 988 989 case <-d.bodyCh: 990 case <-d.receiptCh: 991 // Out of bounds delivery, ignore 992 } 993 } 994 } 995 // Ensure valid ancestry and return 996 if int64(start) <= floor { 997 p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor) 998 return 0, errInvalidAncestor 999 } 1000 p.log.Debug("Found common ancestor", "number", start, "hash", hash) 1001 return start, nil 1002 } 1003 1004 // fetchHeaders keeps retrieving headers concurrently from the number 1005 // requested, until no more are returned, potentially throttling on the way. To 1006 // facilitate concurrency but still protect against malicious nodes sending bad 1007 // headers, we construct a header chain skeleton using the "origin" peer we are 1008 // syncing with, and fill in the missing headers using anyone else. Headers from 1009 // other peers are only accepted if they map cleanly to the skeleton. If no one 1010 // can fill in the skeleton - not even the origin peer - it's assumed invalid and 1011 // the origin is dropped. 1012 func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error { 1013 p.log.Debug("Directing header downloads", "origin", from) 1014 defer p.log.Debug("Header download terminated") 1015 1016 // Create a timeout timer, and the associated header fetcher 1017 skeleton := true // Skeleton assembly phase or finishing up 1018 pivoting := false // Whether the next request is pivot verification 1019 request := time.Now() // time of the last skeleton fetch request 1020 timeout := time.NewTimer(0) // timer to dump a non-responsive active peer 1021 <-timeout.C // timeout channel should be initially empty 1022 defer timeout.Stop() 1023 1024 var ttl time.Duration 1025 getHeaders := func(from uint64) { 1026 request = time.Now() 1027 1028 ttl = d.requestTTL() 1029 timeout.Reset(ttl) 1030 1031 if skeleton { 1032 p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from) 1033 go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false) 1034 } else { 1035 p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from) 1036 go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false) 1037 } 1038 } 1039 getNextPivot := func() { 1040 pivoting = true 1041 request = time.Now() 1042 1043 ttl = d.requestTTL() 1044 timeout.Reset(ttl) 1045 1046 d.pivotLock.RLock() 1047 pivot := d.pivotHeader.Number.Uint64() 1048 d.pivotLock.RUnlock() 1049 1050 p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks)) 1051 go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep 1052 } 1053 // Start pulling the header chain skeleton until all is done 1054 ancestor := from 1055 getHeaders(from) 1056 1057 mode := d.getMode() 1058 for { 1059 select { 1060 case <-d.cancelCh: 1061 return errCanceled 1062 1063 case packet := <-d.headerCh: 1064 // Make sure the active peer is giving us the skeleton headers 1065 if packet.PeerId() != p.id { 1066 log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId()) 1067 break 1068 } 1069 headerReqTimer.UpdateSince(request) 1070 timeout.Stop() 1071 1072 // If the pivot is being checked, move if it became stale and run the real retrieval 1073 var pivot uint64 1074 1075 d.pivotLock.RLock() 1076 if d.pivotHeader != nil { 1077 pivot = d.pivotHeader.Number.Uint64() 1078 } 1079 d.pivotLock.RUnlock() 1080 1081 if pivoting { 1082 if packet.Items() == 2 { 1083 // Retrieve the headers and do some sanity checks, just in case 1084 headers := packet.(*headerPack).headers 1085 1086 if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want { 1087 log.Warn("Peer sent invalid next pivot", "have", have, "want", want) 1088 return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want) 1089 } 1090 if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want { 1091 log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want) 1092 return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want) 1093 } 1094 log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number) 1095 pivot = headers[0].Number.Uint64() 1096 1097 d.pivotLock.Lock() 1098 d.pivotHeader = headers[0] 1099 d.pivotLock.Unlock() 1100 1101 // Write out the pivot into the database so a rollback beyond 1102 // it will reenable fast sync and update the state root that 1103 // the state syncer will be downloading. 1104 rawdb.WriteLastPivotNumber(d.stateDB, pivot) 1105 } 1106 pivoting = false 1107 getHeaders(from) 1108 continue 1109 } 1110 // If the skeleton's finished, pull any remaining head headers directly from the origin 1111 if skeleton && packet.Items() == 0 { 1112 skeleton = false 1113 getHeaders(from) 1114 continue 1115 } 1116 // If no more headers are inbound, notify the content fetchers and return 1117 if packet.Items() == 0 { 1118 // Don't abort header fetches while the pivot is downloading 1119 if atomic.LoadInt32(&d.committed) == 0 && pivot <= from { 1120 p.log.Debug("No headers, waiting for pivot commit") 1121 select { 1122 case <-time.After(fsHeaderContCheck): 1123 getHeaders(from) 1124 continue 1125 case <-d.cancelCh: 1126 return errCanceled 1127 } 1128 } 1129 // Pivot done (or not in fast sync) and no more headers, terminate the process 1130 p.log.Debug("No more headers available") 1131 select { 1132 case d.headerProcCh <- nil: 1133 return nil 1134 case <-d.cancelCh: 1135 return errCanceled 1136 } 1137 } 1138 headers := packet.(*headerPack).headers 1139 1140 // If we received a skeleton batch, resolve internals concurrently 1141 if skeleton { 1142 filled, proced, err := d.fillHeaderSkeleton(from, headers) 1143 if err != nil { 1144 p.log.Debug("Skeleton chain invalid", "err", err) 1145 return fmt.Errorf("%w: %v", errInvalidChain, err) 1146 } 1147 headers = filled[proced:] 1148 from += uint64(proced) 1149 } else { 1150 // If we're closing in on the chain head, but haven't yet reached it, delay 1151 // the last few headers so mini reorgs on the head don't cause invalid hash 1152 // chain errors. 1153 if n := len(headers); n > 0 { 1154 // Retrieve the current head we're at 1155 var head uint64 1156 if mode == LightSync { 1157 head = d.lightchain.CurrentHeader().Number.Uint64() 1158 } else { 1159 head = d.blockchain.CurrentFastBlock().NumberU64() 1160 if full := d.blockchain.CurrentBlock().NumberU64(); head < full { 1161 head = full 1162 } 1163 } 1164 // If the head is below the common ancestor, we're actually deduplicating 1165 // already existing chain segments, so use the ancestor as the fake head. 1166 // Otherwise we might end up delaying header deliveries pointlessly. 1167 if head < ancestor { 1168 head = ancestor 1169 } 1170 // If the head is way older than this batch, delay the last few headers 1171 if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() { 1172 delay := reorgProtHeaderDelay 1173 if delay > n { 1174 delay = n 1175 } 1176 headers = headers[:n-delay] 1177 } 1178 } 1179 } 1180 // Insert all the new headers and fetch the next batch 1181 if len(headers) > 0 { 1182 p.log.Trace("Scheduling new headers", "count", len(headers), "from", from) 1183 select { 1184 case d.headerProcCh <- headers: 1185 case <-d.cancelCh: 1186 return errCanceled 1187 } 1188 from += uint64(len(headers)) 1189 1190 // If we're still skeleton filling fast sync, check pivot staleness 1191 // before continuing to the next skeleton filling 1192 if skeleton && pivot > 0 { 1193 getNextPivot() 1194 } else { 1195 getHeaders(from) 1196 } 1197 } else { 1198 // No headers delivered, or all of them being delayed, sleep a bit and retry 1199 p.log.Trace("All headers delayed, waiting") 1200 select { 1201 case <-time.After(fsHeaderContCheck): 1202 getHeaders(from) 1203 continue 1204 case <-d.cancelCh: 1205 return errCanceled 1206 } 1207 } 1208 1209 case <-timeout.C: 1210 if d.dropPeer == nil { 1211 // The dropPeer method is nil when `--copydb` is used for a local copy. 1212 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1213 p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id) 1214 break 1215 } 1216 // Header retrieval timed out, consider the peer bad and drop 1217 p.log.Debug("Header request timed out", "elapsed", ttl) 1218 headerTimeoutMeter.Mark(1) 1219 d.dropPeer(p.id) 1220 1221 // Finish the sync gracefully instead of dumping the gathered data though 1222 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1223 select { 1224 case ch <- false: 1225 case <-d.cancelCh: 1226 } 1227 } 1228 select { 1229 case d.headerProcCh <- nil: 1230 case <-d.cancelCh: 1231 } 1232 return fmt.Errorf("%w: header request timed out", errBadPeer) 1233 } 1234 } 1235 } 1236 1237 // fillHeaderSkeleton concurrently retrieves headers from all our available peers 1238 // and maps them to the provided skeleton header chain. 1239 // 1240 // Any partial results from the beginning of the skeleton is (if possible) forwarded 1241 // immediately to the header processor to keep the rest of the pipeline full even 1242 // in the case of header stalls. 1243 // 1244 // The method returns the entire filled skeleton and also the number of headers 1245 // already forwarded for processing. 1246 func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) { 1247 log.Debug("Filling up skeleton", "from", from) 1248 d.queue.ScheduleSkeleton(from, skeleton) 1249 1250 var ( 1251 deliver = func(packet dataPack) (int, error) { 1252 pack := packet.(*headerPack) 1253 return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh) 1254 } 1255 expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) } 1256 reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) { 1257 return d.queue.ReserveHeaders(p, count), false, false 1258 } 1259 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) } 1260 capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) } 1261 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1262 p.SetHeadersIdle(accepted, deliveryTime) 1263 } 1264 ) 1265 err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire, 1266 d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve, 1267 nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers") 1268 1269 log.Debug("Skeleton fill terminated", "err", err) 1270 1271 filled, proced := d.queue.RetrieveHeaders() 1272 return filled, proced, err 1273 } 1274 1275 // fetchBodies iteratively downloads the scheduled block bodies, taking any 1276 // available peers, reserving a chunk of blocks for each, waiting for delivery 1277 // and also periodically checking for timeouts. 1278 func (d *Downloader) fetchBodies(from uint64) error { 1279 log.Debug("Downloading block bodies", "origin", from) 1280 1281 var ( 1282 deliver = func(packet dataPack) (int, error) { 1283 pack := packet.(*bodyPack) 1284 return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles) 1285 } 1286 expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) } 1287 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) } 1288 capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) } 1289 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) } 1290 ) 1291 err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire, 1292 d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies, 1293 d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies") 1294 1295 log.Debug("Block body download terminated", "err", err) 1296 return err 1297 } 1298 1299 // fetchReceipts iteratively downloads the scheduled block receipts, taking any 1300 // available peers, reserving a chunk of receipts for each, waiting for delivery 1301 // and also periodically checking for timeouts. 1302 func (d *Downloader) fetchReceipts(from uint64) error { 1303 log.Debug("Downloading transaction receipts", "origin", from) 1304 1305 var ( 1306 deliver = func(packet dataPack) (int, error) { 1307 pack := packet.(*receiptPack) 1308 return d.queue.DeliverReceipts(pack.peerID, pack.receipts) 1309 } 1310 expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) } 1311 fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) } 1312 capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) } 1313 setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { 1314 p.SetReceiptsIdle(accepted, deliveryTime) 1315 } 1316 ) 1317 err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire, 1318 d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts, 1319 d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts") 1320 1321 log.Debug("Transaction receipt download terminated", "err", err) 1322 return err 1323 } 1324 1325 // fetchParts iteratively downloads scheduled block parts, taking any available 1326 // peers, reserving a chunk of fetch requests for each, waiting for delivery and 1327 // also periodically checking for timeouts. 1328 // 1329 // As the scheduling/timeout logic mostly is the same for all downloaded data 1330 // types, this method is used by each for data gathering and is instrumented with 1331 // various callbacks to handle the slight differences between processing them. 1332 // 1333 // The instrumentation parameters: 1334 // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer) 1335 // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers) 1336 // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`) 1337 // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed) 1338 // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping) 1339 // - pending: task callback for the number of requests still needing download (detect completion/non-completability) 1340 // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish) 1341 // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use) 1342 // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions) 1343 // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic) 1344 // - fetch: network callback to actually send a particular download request to a physical remote peer 1345 // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer) 1346 // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping) 1347 // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks 1348 // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping) 1349 // - kind: textual label of the type being downloaded to display in log messages 1350 func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool, 1351 expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool), 1352 fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int, 1353 idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error { 1354 1355 // Create a ticker to detect expired retrieval tasks 1356 ticker := time.NewTicker(100 * time.Millisecond) 1357 defer ticker.Stop() 1358 1359 update := make(chan struct{}, 1) 1360 1361 // Prepare the queue and fetch block parts until the block header fetcher's done 1362 finished := false 1363 for { 1364 select { 1365 case <-d.cancelCh: 1366 return errCanceled 1367 1368 case packet := <-deliveryCh: 1369 deliveryTime := time.Now() 1370 // If the peer was previously banned and failed to deliver its pack 1371 // in a reasonable time frame, ignore its message. 1372 if peer := d.peers.Peer(packet.PeerId()); peer != nil { 1373 // Deliver the received chunk of data and check chain validity 1374 accepted, err := deliver(packet) 1375 if errors.Is(err, errInvalidChain) { 1376 return err 1377 } 1378 // Unless a peer delivered something completely else than requested (usually 1379 // caused by a timed out request which came through in the end), set it to 1380 // idle. If the delivery's stale, the peer should have already been idled. 1381 if !errors.Is(err, errStaleDelivery) { 1382 setIdle(peer, accepted, deliveryTime) 1383 } 1384 // Issue a log to the user to see what's going on 1385 switch { 1386 case err == nil && packet.Items() == 0: 1387 peer.log.Trace("Requested data not delivered", "type", kind) 1388 case err == nil: 1389 peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats()) 1390 default: 1391 peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err) 1392 } 1393 } 1394 // Blocks assembled, try to update the progress 1395 select { 1396 case update <- struct{}{}: 1397 default: 1398 } 1399 1400 case cont := <-wakeCh: 1401 // The header fetcher sent a continuation flag, check if it's done 1402 if !cont { 1403 finished = true 1404 } 1405 // Headers arrive, try to update the progress 1406 select { 1407 case update <- struct{}{}: 1408 default: 1409 } 1410 1411 case <-ticker.C: 1412 // Sanity check update the progress 1413 select { 1414 case update <- struct{}{}: 1415 default: 1416 } 1417 1418 case <-update: 1419 // Short circuit if we lost all our peers 1420 if d.peers.Len() == 0 { 1421 return errNoPeers 1422 } 1423 // Check for fetch request timeouts and demote the responsible peers 1424 for pid, fails := range expire() { 1425 if peer := d.peers.Peer(pid); peer != nil { 1426 // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps 1427 // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times 1428 // out that sync wise we need to get rid of the peer. 1429 // 1430 // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth 1431 // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing 1432 // how response times reacts, to it always requests one more than the minimum (i.e. min 2). 1433 if fails > 2 { 1434 peer.log.Trace("Data delivery timed out", "type", kind) 1435 setIdle(peer, 0, time.Now()) 1436 } else { 1437 peer.log.Debug("Stalling delivery, dropping", "type", kind) 1438 1439 if d.dropPeer == nil { 1440 // The dropPeer method is nil when `--copydb` is used for a local copy. 1441 // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored 1442 peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid) 1443 } else { 1444 d.dropPeer(pid) 1445 1446 // If this peer was the master peer, abort sync immediately 1447 d.cancelLock.RLock() 1448 master := pid == d.cancelPeer 1449 d.cancelLock.RUnlock() 1450 1451 if master { 1452 d.cancel() 1453 return errTimeout 1454 } 1455 } 1456 } 1457 } 1458 } 1459 // If there's nothing more to fetch, wait or terminate 1460 if pending() == 0 { 1461 if !inFlight() && finished { 1462 log.Debug("Data fetching completed", "type", kind) 1463 return nil 1464 } 1465 break 1466 } 1467 // Send a download request to all idle peers, until throttled 1468 progressed, throttled, running := false, false, inFlight() 1469 idles, total := idle() 1470 pendCount := pending() 1471 for _, peer := range idles { 1472 // Short circuit if throttling activated 1473 if throttled { 1474 break 1475 } 1476 // Short circuit if there is no more available task. 1477 if pendCount = pending(); pendCount == 0 { 1478 break 1479 } 1480 // Reserve a chunk of fetches for a peer. A nil can mean either that 1481 // no more headers are available, or that the peer is known not to 1482 // have them. 1483 request, progress, throttle := reserve(peer, capacity(peer)) 1484 if progress { 1485 progressed = true 1486 } 1487 if throttle { 1488 throttled = true 1489 throttleCounter.Inc(1) 1490 } 1491 if request == nil { 1492 continue 1493 } 1494 if request.From > 0 { 1495 peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From) 1496 } else { 1497 peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number) 1498 } 1499 // Fetch the chunk and make sure any errors return the hashes to the queue 1500 if fetchHook != nil { 1501 fetchHook(request.Headers) 1502 } 1503 if err := fetch(peer, request); err != nil { 1504 // Although we could try and make an attempt to fix this, this error really 1505 // means that we've double allocated a fetch task to a peer. If that is the 1506 // case, the internal state of the downloader and the queue is very wrong so 1507 // better hard crash and note the error instead of silently accumulating into 1508 // a much bigger issue. 1509 panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind)) 1510 } 1511 running = true 1512 } 1513 // Make sure that we have peers available for fetching. If all peers have been tried 1514 // and all failed throw an error 1515 if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 { 1516 return errPeersUnavailable 1517 } 1518 } 1519 } 1520 } 1521 1522 // processHeaders takes batches of retrieved headers from an input channel and 1523 // keeps processing and scheduling them into the header chain and downloader's 1524 // queue until the stream ends or a failure occurs. 1525 func (d *Downloader) processHeaders(origin uint64, td *big.Int) error { 1526 // Keep a count of uncertain headers to roll back 1527 var ( 1528 rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis) 1529 rollbackErr error 1530 mode = d.getMode() 1531 ) 1532 defer func() { 1533 if rollback > 0 { 1534 lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0 1535 if mode != LightSync { 1536 lastFastBlock = d.blockchain.CurrentFastBlock().Number() 1537 lastBlock = d.blockchain.CurrentBlock().Number() 1538 } 1539 if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block 1540 // We're already unwinding the stack, only print the error to make it more visible 1541 log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err) 1542 } 1543 curFastBlock, curBlock := common.Big0, common.Big0 1544 if mode != LightSync { 1545 curFastBlock = d.blockchain.CurrentFastBlock().Number() 1546 curBlock = d.blockchain.CurrentBlock().Number() 1547 } 1548 log.Warn("Rolled back chain segment", 1549 "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number), 1550 "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock), 1551 "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr) 1552 } 1553 }() 1554 // Wait for batches of headers to process 1555 gotHeaders := false 1556 1557 for { 1558 select { 1559 case <-d.cancelCh: 1560 rollbackErr = errCanceled 1561 return errCanceled 1562 1563 case headers := <-d.headerProcCh: 1564 // Terminate header processing if we synced up 1565 if len(headers) == 0 { 1566 // Notify everyone that headers are fully processed 1567 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1568 select { 1569 case ch <- false: 1570 case <-d.cancelCh: 1571 } 1572 } 1573 // If no headers were retrieved at all, the peer violated its TD promise that it had a 1574 // better chain compared to ours. The only exception is if its promised blocks were 1575 // already imported by other means (e.g. fetcher): 1576 // 1577 // R <remote peer>, L <local node>: Both at block 10 1578 // R: Mine block 11, and propagate it to L 1579 // L: Queue block 11 for import 1580 // L: Notice that R's head and TD increased compared to ours, start sync 1581 // L: Import of block 11 finishes 1582 // L: Sync begins, and finds common ancestor at 11 1583 // L: Request new headers up from 11 (R's TD was higher, it must have something) 1584 // R: Nothing to give 1585 if mode != LightSync { 1586 head := d.blockchain.CurrentBlock() 1587 if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 { 1588 return errStallingPeer 1589 } 1590 } 1591 // If fast or light syncing, ensure promised headers are indeed delivered. This is 1592 // needed to detect scenarios where an attacker feeds a bad pivot and then bails out 1593 // of delivering the post-pivot blocks that would flag the invalid content. 1594 // 1595 // This check cannot be executed "as is" for full imports, since blocks may still be 1596 // queued for processing when the header download completes. However, as long as the 1597 // peer gave us something useful, we're already happy/progressed (above check). 1598 if mode == FastSync || mode == LightSync { 1599 head := d.lightchain.CurrentHeader() 1600 if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 { 1601 return errStallingPeer 1602 } 1603 } 1604 // Disable any rollback and return 1605 rollback = 0 1606 return nil 1607 } 1608 // Otherwise split the chunk of headers into batches and process them 1609 gotHeaders = true 1610 for len(headers) > 0 { 1611 // Terminate if something failed in between processing chunks 1612 select { 1613 case <-d.cancelCh: 1614 rollbackErr = errCanceled 1615 return errCanceled 1616 default: 1617 } 1618 // Select the next chunk of headers to import 1619 limit := maxHeadersProcess 1620 if limit > len(headers) { 1621 limit = len(headers) 1622 } 1623 chunk := headers[:limit] 1624 1625 // In case of header only syncing, validate the chunk immediately 1626 if mode == FastSync || mode == LightSync { 1627 // If we're importing pure headers, verify based on their recentness 1628 var pivot uint64 1629 1630 d.pivotLock.RLock() 1631 if d.pivotHeader != nil { 1632 pivot = d.pivotHeader.Number.Uint64() 1633 } 1634 d.pivotLock.RUnlock() 1635 1636 frequency := fsHeaderCheckFrequency 1637 if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot { 1638 frequency = 1 1639 } 1640 if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil { 1641 rollbackErr = err 1642 1643 // If some headers were inserted, track them as uncertain 1644 if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 { 1645 rollback = chunk[0].Number.Uint64() 1646 } 1647 log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err) 1648 return fmt.Errorf("%w: %v", errInvalidChain, err) 1649 } 1650 // All verifications passed, track all headers within the alloted limits 1651 if mode == FastSync { 1652 head := chunk[len(chunk)-1].Number.Uint64() 1653 if head-rollback > uint64(fsHeaderSafetyNet) { 1654 rollback = head - uint64(fsHeaderSafetyNet) 1655 } else { 1656 rollback = 1 1657 } 1658 } 1659 } 1660 // Unless we're doing light chains, schedule the headers for associated content retrieval 1661 if mode == FullSync || mode == FastSync { 1662 // If we've reached the allowed number of pending headers, stall a bit 1663 for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders { 1664 select { 1665 case <-d.cancelCh: 1666 rollbackErr = errCanceled 1667 return errCanceled 1668 case <-time.After(time.Second): 1669 } 1670 } 1671 // Otherwise insert the headers for content retrieval 1672 inserts := d.queue.Schedule(chunk, origin) 1673 if len(inserts) != len(chunk) { 1674 rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk)) 1675 return fmt.Errorf("%w: stale headers", errBadPeer) 1676 } 1677 } 1678 headers = headers[limit:] 1679 origin += uint64(limit) 1680 } 1681 // Update the highest block number we know if a higher one is found. 1682 d.syncStatsLock.Lock() 1683 if d.syncStatsChainHeight < origin { 1684 d.syncStatsChainHeight = origin - 1 1685 } 1686 d.syncStatsLock.Unlock() 1687 1688 // Signal the content downloaders of the availablility of new tasks 1689 for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} { 1690 select { 1691 case ch <- true: 1692 default: 1693 } 1694 } 1695 } 1696 } 1697 } 1698 1699 // processFullSyncContent takes fetch results from the queue and imports them into the chain. 1700 func (d *Downloader) processFullSyncContent() error { 1701 for { 1702 results := d.queue.Results(true) 1703 if len(results) == 0 { 1704 return nil 1705 } 1706 if d.chainInsertHook != nil { 1707 d.chainInsertHook(results) 1708 } 1709 if err := d.importBlockResults(results); err != nil { 1710 return err 1711 } 1712 } 1713 } 1714 1715 func (d *Downloader) importBlockResults(results []*fetchResult) error { 1716 // Check for any early termination requests 1717 if len(results) == 0 { 1718 return nil 1719 } 1720 select { 1721 case <-d.quitCh: 1722 return errCancelContentProcessing 1723 default: 1724 } 1725 // Retrieve the a batch of results to import 1726 first, last := results[0].Header, results[len(results)-1].Header 1727 log.Debug("Inserting downloaded chain", "items", len(results), 1728 "firstnum", first.Number, "firsthash", first.Hash(), 1729 "lastnum", last.Number, "lasthash", last.Hash(), 1730 ) 1731 blocks := make([]*types.Block, len(results)) 1732 for i, result := range results { 1733 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1734 } 1735 if index, err := d.blockchain.InsertChain(blocks); err != nil { 1736 if index < len(results) { 1737 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1738 } else { 1739 // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index, 1740 // when it needs to preprocess blocks to import a sidechain. 1741 // The importer will put together a new list of blocks to import, which is a superset 1742 // of the blocks delivered from the downloader, and the indexing will be off. 1743 log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err) 1744 } 1745 return fmt.Errorf("%w: %v", errInvalidChain, err) 1746 } 1747 return nil 1748 } 1749 1750 // processFastSyncContent takes fetch results from the queue and writes them to the 1751 // database. It also controls the synchronisation of state nodes of the pivot block. 1752 func (d *Downloader) processFastSyncContent() error { 1753 // Start syncing state of the reported head block. This should get us most of 1754 // the state of the pivot block. 1755 d.pivotLock.RLock() 1756 sync := d.syncState(d.pivotHeader.Root) 1757 d.pivotLock.RUnlock() 1758 1759 defer func() { 1760 // The `sync` object is replaced every time the pivot moves. We need to 1761 // defer close the very last active one, hence the lazy evaluation vs. 1762 // calling defer sync.Cancel() !!! 1763 sync.Cancel() 1764 }() 1765 1766 closeOnErr := func(s *stateSync) { 1767 if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled { 1768 d.queue.Close() // wake up Results 1769 } 1770 } 1771 go closeOnErr(sync) 1772 1773 // To cater for moving pivot points, track the pivot block and subsequently 1774 // accumulated download results separately. 1775 var ( 1776 oldPivot *fetchResult // Locked in pivot block, might change eventually 1777 oldTail []*fetchResult // Downloaded content after the pivot 1778 ) 1779 for { 1780 // Wait for the next batch of downloaded data to be available, and if the pivot 1781 // block became stale, move the goalpost 1782 results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness 1783 if len(results) == 0 { 1784 // If pivot sync is done, stop 1785 if oldPivot == nil { 1786 return sync.Cancel() 1787 } 1788 // If sync failed, stop 1789 select { 1790 case <-d.cancelCh: 1791 sync.Cancel() 1792 return errCanceled 1793 default: 1794 } 1795 } 1796 if d.chainInsertHook != nil { 1797 d.chainInsertHook(results) 1798 } 1799 // If we haven't downloaded the pivot block yet, check pivot staleness 1800 // notifications from the header downloader 1801 d.pivotLock.RLock() 1802 pivot := d.pivotHeader 1803 d.pivotLock.RUnlock() 1804 1805 if oldPivot == nil { 1806 if pivot.Root != sync.root { 1807 sync.Cancel() 1808 sync = d.syncState(pivot.Root) 1809 1810 go closeOnErr(sync) 1811 } 1812 } else { 1813 results = append(append([]*fetchResult{oldPivot}, oldTail...), results...) 1814 } 1815 // Split around the pivot block and process the two sides via fast/full sync 1816 if atomic.LoadInt32(&d.committed) == 0 { 1817 latest := results[len(results)-1].Header 1818 // If the height is above the pivot block by 2 sets, it means the pivot 1819 // become stale in the network and it was garbage collected, move to a 1820 // new pivot. 1821 // 1822 // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those 1823 // need to be taken into account, otherwise we're detecting the pivot move 1824 // late and will drop peers due to unavailable state!!! 1825 if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) { 1826 log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay)) 1827 pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted 1828 1829 d.pivotLock.Lock() 1830 d.pivotHeader = pivot 1831 d.pivotLock.Unlock() 1832 1833 // Write out the pivot into the database so a rollback beyond it will 1834 // reenable fast sync 1835 rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64()) 1836 } 1837 } 1838 P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results) 1839 if err := d.commitFastSyncData(beforeP, sync); err != nil { 1840 return err 1841 } 1842 if P != nil { 1843 // If new pivot block found, cancel old state retrieval and restart 1844 if oldPivot != P { 1845 sync.Cancel() 1846 sync = d.syncState(P.Header.Root) 1847 1848 go closeOnErr(sync) 1849 oldPivot = P 1850 } 1851 // Wait for completion, occasionally checking for pivot staleness 1852 select { 1853 case <-sync.done: 1854 if sync.err != nil { 1855 return sync.err 1856 } 1857 if err := d.commitPivotBlock(P); err != nil { 1858 return err 1859 } 1860 oldPivot = nil 1861 1862 case <-time.After(time.Second): 1863 oldTail = afterP 1864 continue 1865 } 1866 } 1867 // Fast sync done, pivot commit done, full import 1868 if err := d.importBlockResults(afterP); err != nil { 1869 return err 1870 } 1871 } 1872 } 1873 1874 func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) { 1875 if len(results) == 0 { 1876 return nil, nil, nil 1877 } 1878 if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot { 1879 // the pivot is somewhere in the future 1880 return nil, results, nil 1881 } 1882 // This can also be optimized, but only happens very seldom 1883 for _, result := range results { 1884 num := result.Header.Number.Uint64() 1885 switch { 1886 case num < pivot: 1887 before = append(before, result) 1888 case num == pivot: 1889 p = result 1890 default: 1891 after = append(after, result) 1892 } 1893 } 1894 return p, before, after 1895 } 1896 1897 func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error { 1898 // Check for any early termination requests 1899 if len(results) == 0 { 1900 return nil 1901 } 1902 select { 1903 case <-d.quitCh: 1904 return errCancelContentProcessing 1905 case <-stateSync.done: 1906 if err := stateSync.Wait(); err != nil { 1907 return err 1908 } 1909 default: 1910 } 1911 // Retrieve the a batch of results to import 1912 first, last := results[0].Header, results[len(results)-1].Header 1913 log.Debug("Inserting fast-sync blocks", "items", len(results), 1914 "firstnum", first.Number, "firsthash", first.Hash(), 1915 "lastnumn", last.Number, "lasthash", last.Hash(), 1916 ) 1917 blocks := make([]*types.Block, len(results)) 1918 receipts := make([]types.Receipts, len(results)) 1919 for i, result := range results { 1920 blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1921 receipts[i] = result.Receipts 1922 } 1923 if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil { 1924 log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err) 1925 return fmt.Errorf("%w: %v", errInvalidChain, err) 1926 } 1927 return nil 1928 } 1929 1930 func (d *Downloader) commitPivotBlock(result *fetchResult) error { 1931 block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles) 1932 log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash()) 1933 1934 // Commit the pivot block as the new head, will require full sync from here on 1935 if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil { 1936 return err 1937 } 1938 if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil { 1939 return err 1940 } 1941 atomic.StoreInt32(&d.committed, 1) 1942 1943 // If we had a bloom filter for the state sync, deallocate it now. Note, we only 1944 // deallocate internally, but keep the empty wrapper. This ensures that if we do 1945 // a rollback after committing the pivot and restarting fast sync, we don't end 1946 // up using a nil bloom. Empty bloom is fine, it just returns that it does not 1947 // have the info we need, so reach down to the database instead. 1948 if d.stateBloom != nil { 1949 d.stateBloom.Close() 1950 } 1951 return nil 1952 } 1953 1954 // DeliverHeaders injects a new batch of block headers received from a remote 1955 // node into the download schedule. 1956 func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) error { 1957 return d.deliver(d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter) 1958 } 1959 1960 // DeliverBodies injects a new batch of block bodies received from a remote node. 1961 func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) error { 1962 return d.deliver(d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter) 1963 } 1964 1965 // DeliverReceipts injects a new batch of receipts received from a remote node. 1966 func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) error { 1967 return d.deliver(d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter) 1968 } 1969 1970 // DeliverNodeData injects a new batch of node state data received from a remote node. 1971 func (d *Downloader) DeliverNodeData(id string, data [][]byte) error { 1972 return d.deliver(d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter) 1973 } 1974 1975 // DeliverSnapPacket is invoked from a peer's message handler when it transmits a 1976 // data packet for the local node to consume. 1977 func (d *Downloader) DeliverSnapPacket(peer *snap.Peer, packet snap.Packet) error { 1978 switch packet := packet.(type) { 1979 case *snap.AccountRangePacket: 1980 hashes, accounts, err := packet.Unpack() 1981 if err != nil { 1982 return err 1983 } 1984 return d.SnapSyncer.OnAccounts(peer, packet.ID, hashes, accounts, packet.Proof) 1985 1986 case *snap.StorageRangesPacket: 1987 hashset, slotset := packet.Unpack() 1988 return d.SnapSyncer.OnStorage(peer, packet.ID, hashset, slotset, packet.Proof) 1989 1990 case *snap.ByteCodesPacket: 1991 return d.SnapSyncer.OnByteCodes(peer, packet.ID, packet.Codes) 1992 1993 case *snap.TrieNodesPacket: 1994 return d.SnapSyncer.OnTrieNodes(peer, packet.ID, packet.Nodes) 1995 1996 default: 1997 return fmt.Errorf("unexpected snap packet type: %T", packet) 1998 } 1999 } 2000 2001 // deliver injects a new batch of data received from a remote node. 2002 func (d *Downloader) deliver(destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) { 2003 // Update the delivery metrics for both good and failed deliveries 2004 inMeter.Mark(int64(packet.Items())) 2005 defer func() { 2006 if err != nil { 2007 dropMeter.Mark(int64(packet.Items())) 2008 } 2009 }() 2010 // Deliver or abort if the sync is canceled while queuing 2011 d.cancelLock.RLock() 2012 cancel := d.cancelCh 2013 d.cancelLock.RUnlock() 2014 if cancel == nil { 2015 return errNoSyncActive 2016 } 2017 select { 2018 case destCh <- packet: 2019 return nil 2020 case <-cancel: 2021 return errNoSyncActive 2022 } 2023 } 2024 2025 // qosTuner is the quality of service tuning loop that occasionally gathers the 2026 // peer latency statistics and updates the estimated request round trip time. 2027 func (d *Downloader) qosTuner() { 2028 for { 2029 // Retrieve the current median RTT and integrate into the previoust target RTT 2030 rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT())) 2031 atomic.StoreUint64(&d.rttEstimate, uint64(rtt)) 2032 2033 // A new RTT cycle passed, increase our confidence in the estimated RTT 2034 conf := atomic.LoadUint64(&d.rttConfidence) 2035 conf = conf + (1000000-conf)/2 2036 atomic.StoreUint64(&d.rttConfidence, conf) 2037 2038 // Log the new QoS values and sleep until the next RTT 2039 log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 2040 select { 2041 case <-d.quitCh: 2042 return 2043 case <-time.After(rtt): 2044 } 2045 } 2046 } 2047 2048 // qosReduceConfidence is meant to be called when a new peer joins the downloader's 2049 // peer set, needing to reduce the confidence we have in out QoS estimates. 2050 func (d *Downloader) qosReduceConfidence() { 2051 // If we have a single peer, confidence is always 1 2052 peers := uint64(d.peers.Len()) 2053 if peers == 0 { 2054 // Ensure peer connectivity races don't catch us off guard 2055 return 2056 } 2057 if peers == 1 { 2058 atomic.StoreUint64(&d.rttConfidence, 1000000) 2059 return 2060 } 2061 // If we have a ton of peers, don't drop confidence) 2062 if peers >= uint64(qosConfidenceCap) { 2063 return 2064 } 2065 // Otherwise drop the confidence factor 2066 conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers 2067 if float64(conf)/1000000 < rttMinConfidence { 2068 conf = uint64(rttMinConfidence * 1000000) 2069 } 2070 atomic.StoreUint64(&d.rttConfidence, conf) 2071 2072 rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2073 log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL()) 2074 } 2075 2076 // requestRTT returns the current target round trip time for a download request 2077 // to complete in. 2078 // 2079 // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that 2080 // the downloader tries to adapt queries to the RTT, so multiple RTT values can 2081 // be adapted to, but smaller ones are preferred (stabler download stream). 2082 func (d *Downloader) requestRTT() time.Duration { 2083 return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10 2084 } 2085 2086 // requestTTL returns the current timeout allowance for a single download request 2087 // to finish under. 2088 func (d *Downloader) requestTTL() time.Duration { 2089 var ( 2090 rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate)) 2091 conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0 2092 ) 2093 ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf) 2094 if ttl > ttlLimit { 2095 ttl = ttlLimit 2096 } 2097 return ttl 2098 }