downloader.go 58.4 KB
Newer Older
F
Felix Lange 已提交
1
// Copyright 2015 The go-ethereum Authors
2
// This file is part of the go-ethereum library.
F
Felix Lange 已提交
3
//
4
// The go-ethereum library is free software: you can redistribute it and/or modify
F
Felix Lange 已提交
5 6 7 8
// it under the terms of the GNU Lesser General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
9
// The go-ethereum library is distributed in the hope that it will be useful,
F
Felix Lange 已提交
10
// but WITHOUT ANY WARRANTY; without even the implied warranty of
11
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
F
Felix Lange 已提交
12 13 14
// GNU Lesser General Public License for more details.
//
// 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/>.
F
Felix Lange 已提交
16

17
// Package downloader contains the manual full chain synchronisation.
18 19 20
package downloader

import (
21
	"crypto/rand"
22
	"errors"
23
	"fmt"
24
	"math"
25
	"math/big"
26 27 28 29
	"sync"
	"sync/atomic"
	"time"

30
	ethereum "github.com/ethereum/go-ethereum"
31 32
	"github.com/ethereum/go-ethereum/common"
	"github.com/ethereum/go-ethereum/core/types"
33
	"github.com/ethereum/go-ethereum/ethdb"
34
	"github.com/ethereum/go-ethereum/event"
35
	"github.com/ethereum/go-ethereum/log"
36
	"github.com/ethereum/go-ethereum/params"
37
	"github.com/rcrowley/go-metrics"
38 39
)

40
var (
41 42 43
	MaxHashFetch    = 512 // Amount of hashes to be fetched per retrieval request
	MaxBlockFetch   = 128 // Amount of blocks to be fetched per retrieval request
	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 46
	MaxBodyFetch    = 128 // Amount of block bodies to be fetched per retrieval request
	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 50 51 52 53 54
	MaxForkAncestry  = 3 * params.EpochDuration // Maximum chain reorganisation
	rttMinEstimate   = 2 * time.Second          // Minimum round-trip time to target for download requests
	rttMaxEstimate   = 20 * time.Second         // Maximum rount-trip time to target for download requests
	rttMinConfidence = 0.1                      // Worse confidence factor in our estimated RTT value
	ttlScaling       = 3                        // Constant scaling factor for RTT -> TTL conversion
	ttlLimit         = time.Minute              // Maximum TTL allowance to prevent reaching crazy timeouts
55 56 57 58

	qosTuningPeers   = 5    // Number of peers to tune based on (best peers)
	qosConfidenceCap = 10   // Number of peers above which not to modify RTT confidence
	qosTuningImpact  = 0.25 // Impact that a new tuning target has on the previous value
59

60 61
	maxQueuedHeaders  = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection)
	maxHeadersProcess = 2048      // Number of header download results to import at once into the chain
62
	maxResultsProcess = 2048      // Number of content download results to import at once into the chain
63

64 65 66
	fsHeaderCheckFrequency = 100        // Verification frequency of the downloaded headers during fast sync
	fsHeaderSafetyNet      = 2048       // Number of headers to discard in case a chain violation is detected
	fsHeaderForceVerify    = 24         // Number of headers to verify before and after the pivot to accept it
67 68
	fsPivotInterval        = 256        // Number of headers out of which to randomize the pivot point
	fsMinFullBlocks        = 64         // Number of blocks to retrieve fully even in fast sync
69
	fsCriticalTrials       = uint32(32) // Number of times to retry in the cricical section before bailing
70
)
71

72
var (
73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93
	errBusy                    = errors.New("busy")
	errUnknownPeer             = errors.New("peer is unknown or unhealthy")
	errBadPeer                 = errors.New("action from bad peer ignored")
	errStallingPeer            = errors.New("peer is stalling")
	errNoPeers                 = errors.New("no peers to keep download active")
	errTimeout                 = errors.New("timeout")
	errEmptyHeaderSet          = errors.New("empty header set by peer")
	errPeersUnavailable        = errors.New("no peers available or all tried for download")
	errInvalidAncestor         = errors.New("retrieved ancestor is invalid")
	errInvalidChain            = errors.New("retrieved hash chain is invalid")
	errInvalidBlock            = errors.New("retrieved block is invalid")
	errInvalidBody             = errors.New("retrieved block body is invalid")
	errInvalidReceipt          = errors.New("retrieved receipt is invalid")
	errCancelBlockFetch        = errors.New("block download canceled (requested)")
	errCancelHeaderFetch       = errors.New("block header download canceled (requested)")
	errCancelBodyFetch         = errors.New("block body download canceled (requested)")
	errCancelReceiptFetch      = errors.New("receipt download canceled (requested)")
	errCancelStateFetch        = errors.New("state data download canceled (requested)")
	errCancelHeaderProcessing  = errors.New("header processing canceled (requested)")
	errCancelContentProcessing = errors.New("content processing canceled (requested)")
	errNoSyncActive            = errors.New("no sync active")
94
	errTooOld                  = errors.New("peer doesn't speak recent enough protocol version (need version >= 62)")
95 96
)

97
type Downloader struct {
98 99
	mode SyncMode       // Synchronisation mode defining the strategy used (per sync cycle)
	mux  *event.TypeMux // Event multiplexer to announce sync operation events
100

101 102 103
	queue   *queue   // Scheduler for selecting the hashes to download
	peers   *peerSet // Set of active peers from which download can proceed
	stateDB ethdb.Database
104

105
	fsPivotLock  *types.Header // Pivot header on critical section entry (cannot change between retries)
106
	fsPivotFails uint32        // Number of subsequent fast sync failures in the critical section
107

108 109
	rttEstimate   uint64 // Round trip time to target for download requests
	rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops)
110

111
	// Statistics
112 113 114
	syncStatsChainOrigin uint64 // Origin block number where syncing started at
	syncStatsChainHeight uint64 // Highest block number known when syncing started
	syncStatsState       stateSyncStats
115
	syncStatsLock        sync.RWMutex // Lock protecting the sync stats fields
116

117 118 119
	lightchain LightChain
	chain      BlockChain

120
	// Callbacks
121
	dropPeer peerDropFn // Drops a peer for misbehaving
122

123
	// Status
124 125 126
	synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing
	synchronising   int32
	notified        int32
127 128

	// Channels
129 130 131 132 133 134
	headerCh      chan dataPack        // [eth/62] Channel receiving inbound block headers
	bodyCh        chan dataPack        // [eth/62] Channel receiving inbound block bodies
	receiptCh     chan dataPack        // [eth/63] Channel receiving inbound receipts
	bodyWakeCh    chan bool            // [eth/62] Channel to signal the block body fetcher of new tasks
	receiptWakeCh chan bool            // [eth/63] Channel to signal the receipt fetcher of new tasks
	headerProcCh  chan []*types.Header // [eth/62] Channel to feed the header processor new tasks
135

136 137 138 139 140
	// for stateFetcher
	stateSyncStart chan *stateSync
	trackStateReq  chan *stateReq
	stateCh        chan dataPack // [eth/63] Channel receiving inbound node state data

141 142
	// Cancellation and termination
	cancelPeer string        // Identifier of the peer currently being used as the master (cancel on drop)
143
	cancelCh   chan struct{} // Channel to cancel mid-flight syncs
144
	cancelLock sync.RWMutex  // Lock to protect the cancel channel and peer in delivers
145

146 147 148
	quitCh   chan struct{} // Quit channel to signal termination
	quitLock sync.RWMutex  // Lock to prevent double closes

149
	// Testing hooks
150 151 152 153
	syncInitHook     func(uint64, uint64)  // Method to call upon initiating a new sync run
	bodyFetchHook    func([]*types.Header) // Method to call upon starting a block body fetch
	receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch
	chainInsertHook  func([]*fetchResult)  // Method to call upon inserting a chain of blocks (possibly in multiple invocations)
154 155
}

156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200
type LightChain interface {
	// HasHeader verifies a header's presence in the local chain.
	HasHeader(common.Hash) bool

	// GetHeaderByHash retrieves a header from the local chain.
	GetHeaderByHash(common.Hash) *types.Header

	// CurrentHeader retrieves the head header from the local chain.
	CurrentHeader() *types.Header

	// GetTdByHash returns the total difficulty of a local block.
	GetTdByHash(common.Hash) *big.Int

	// InsertHeaderChain inserts a batch of headers into the local chain.
	InsertHeaderChain([]*types.Header, int) (int, error)

	// Rollback removes a few recently added elements from the local chain.
	Rollback([]common.Hash)
}

type BlockChain interface {
	LightChain

	// HasBlockAndState verifies block and associated states' presence in the local chain.
	HasBlockAndState(common.Hash) bool

	// GetBlockByHash retrieves a block from the local chain.
	GetBlockByHash(common.Hash) *types.Block

	// CurrentBlock retrieves the head block from the local chain.
	CurrentBlock() *types.Block

	// CurrentFastBlock retrieves the head fast block from the local chain.
	CurrentFastBlock() *types.Block

	// FastSyncCommitHead directly commits the head block to a certain entity.
	FastSyncCommitHead(common.Hash) error

	// InsertChain inserts a batch of blocks into the local chain.
	InsertChain(types.Blocks) (int, error)

	// InsertReceiptChain inserts a batch of receipts into the local chain.
	InsertReceiptChain(types.Blocks, []types.Receipts) (int, error)
}

201
// New creates a new downloader to fetch hashes and blocks from remote peers.
202 203 204 205
func New(mode SyncMode, stateDb ethdb.Database, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader {
	if lightchain == nil {
		lightchain = chain
	}
206

207
	dl := &Downloader{
208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
		mode:           mode,
		stateDB:        stateDb,
		mux:            mux,
		queue:          newQueue(),
		peers:          newPeerSet(),
		rttEstimate:    uint64(rttMaxEstimate),
		rttConfidence:  uint64(1000000),
		chain:          chain,
		lightchain:     lightchain,
		dropPeer:       dropPeer,
		headerCh:       make(chan dataPack, 1),
		bodyCh:         make(chan dataPack, 1),
		receiptCh:      make(chan dataPack, 1),
		bodyWakeCh:     make(chan bool, 1),
		receiptWakeCh:  make(chan bool, 1),
		headerProcCh:   make(chan []*types.Header, 1),
		quitCh:         make(chan struct{}),
		stateCh:        make(chan dataPack),
226 227
		stateSyncStart: make(chan *stateSync),
		trackStateReq:  make(chan *stateReq),
228
	}
229
	go dl.qosTuner()
230
	go dl.stateFetcher()
231
	return dl
232 233
}

234 235 236
// Progress retrieves the synchronisation boundaries, specifically the origin
// block where synchronisation started at (may have failed/suspended); the block
// or header sync is currently at; and the latest known block which the sync targets.
237
//
L
Leif Jurvetson 已提交
238
// In addition, during the state download phase of fast synchronisation the number
239 240
// of processed and the total number of known states are also returned. Otherwise
// these are zero.
241
func (d *Downloader) Progress() ethereum.SyncProgress {
242
	// Lock the current stats and return the progress
243 244
	d.syncStatsLock.RLock()
	defer d.syncStatsLock.RUnlock()
245

246 247 248
	current := uint64(0)
	switch d.mode {
	case FullSync:
249
		current = d.chain.CurrentBlock().NumberU64()
250
	case FastSync:
251
		current = d.chain.CurrentFastBlock().NumberU64()
252
	case LightSync:
253
		current = d.lightchain.CurrentHeader().Number.Uint64()
254
	}
255 256 257 258
	return ethereum.SyncProgress{
		StartingBlock: d.syncStatsChainOrigin,
		CurrentBlock:  current,
		HighestBlock:  d.syncStatsChainHeight,
259 260
		PulledStates:  d.syncStatsState.processed,
		KnownStates:   d.syncStatsState.processed + d.syncStatsState.pending,
261
	}
O
obscuren 已提交
262 263
}

264
// Synchronising returns whether the downloader is currently retrieving blocks.
265
func (d *Downloader) Synchronising() bool {
266
	return atomic.LoadInt32(&d.synchronising) > 0
267 268
}

269 270
// RegisterPeer injects a new download peer into the set of block source to be
// used for fetching hashes and blocks from.
271
func (d *Downloader) RegisterPeer(id string, version int, peer Peer) error {
272

273 274
	logger := log.New("peer", id)
	logger.Trace("Registering sync peer")
275
	if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil {
276
		logger.Error("Failed to register sync peer", "err", err)
277 278
		return err
	}
279 280
	d.qosReduceConfidence()

281 282 283
	return nil
}

284
// UnregisterPeer remove a peer from the known list, preventing any action from
285 286
// the specified peer. An effort is also made to return any pending fetches into
// the queue.
287
func (d *Downloader) UnregisterPeer(id string) error {
288
	// Unregister the peer from the active peer set and revoke any fetch tasks
289 290
	logger := log.New("peer", id)
	logger.Trace("Unregistering sync peer")
291
	if err := d.peers.Unregister(id); err != nil {
292
		logger.Error("Failed to unregister sync peer", "err", err)
293 294
		return err
	}
295
	d.queue.Revoke(id)
296 297 298 299 300 301 302

	// If this peer was the master peer, abort sync immediately
	d.cancelLock.RLock()
	master := id == d.cancelPeer
	d.cancelLock.RUnlock()

	if master {
303
		d.Cancel()
304
	}
305
	return nil
306 307
}

308 309
// Synchronise tries to sync up our local block chain with a remote peer, both
// adding various sanity checks as well as wrapping it with various log entries.
310 311 312
func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error {
	err := d.synchronise(id, head, td, mode)
	switch err {
313 314 315
	case nil:
	case errBusy:

316 317 318
	case errTimeout, errBadPeer, errStallingPeer,
		errEmptyHeaderSet, errPeersUnavailable, errTooOld,
		errInvalidAncestor, errInvalidChain:
319
		log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err)
320 321 322
		d.dropPeer(id)

	default:
323
		log.Warn("Synchronisation failed, retrying", "err", err)
324
	}
325
	return err
326 327 328
}

// synchronise will select the peer and use it for synchronising. If an empty string is given
329
// it will use the best peer possible and synchronize if it's TD is higher than our own. If any of the
330
// checks fail an error will be returned. This method is synchronous
331
func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error {
L
Leif Jurvetson 已提交
332
	// Mock out the synchronisation if testing
333 334 335
	if d.synchroniseMock != nil {
		return d.synchroniseMock(id, hash)
	}
336
	// Make sure only one goroutine is ever allowed past this point at once
337
	if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) {
338
		return errBusy
339
	}
340
	defer atomic.StoreInt32(&d.synchronising, 0)
341

342 343
	// Post a user notification of the sync (only once per session)
	if atomic.CompareAndSwapInt32(&d.notified, 0, 1) {
344
		log.Info("Block synchronisation started")
345
	}
346
	// Reset the queue, peer set and wake channels to clean any internal leftover state
347
	d.queue.Reset()
348
	d.peers.Reset()
349

350
	for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} {
351 352 353 354
		select {
		case <-ch:
		default:
		}
355
	}
356
	for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} {
357 358 359 360 361 362 363 364
		for empty := false; !empty; {
			select {
			case <-ch:
			default:
				empty = true
			}
		}
	}
365 366 367 368 369 370 371
	for empty := false; !empty; {
		select {
		case <-d.headerProcCh:
		default:
			empty = true
		}
	}
372
	// Create cancel channel for aborting mid-flight and mark the master peer
373 374
	d.cancelLock.Lock()
	d.cancelCh = make(chan struct{})
375
	d.cancelPeer = id
376 377
	d.cancelLock.Unlock()

378
	defer d.Cancel() // No matter what, we can't leave the cancel channel open
379

380 381
	// Set the requested sync mode, unless it's forbidden
	d.mode = mode
382
	if d.mode == FastSync && atomic.LoadUint32(&d.fsPivotFails) >= fsCriticalTrials {
383 384
		d.mode = FullSync
	}
385
	// Retrieve the origin peer and initiate the downloading process
386
	p := d.peers.Peer(id)
387
	if p == nil {
388
		return errUnknownPeer
389
	}
390
	return d.syncWithPeer(p, hash, td)
391 392
}

393 394
// syncWithPeer starts a block synchronization based on the hash chain from the
// specified peer and head hash.
395
func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) {
O
obscuren 已提交
396
	d.mux.Post(StartEvent{})
397 398 399
	defer func() {
		// reset on error
		if err != nil {
400 401 402
			d.mux.Post(FailedEvent{err})
		} else {
			d.mux.Post(DoneEvent{})
403 404
		}
	}()
405 406 407
	if p.version < 62 {
		return errTooOld
	}
408

409
	log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", d.mode)
410
	defer func(start time.Time) {
411
		log.Debug("Synchronisation terminated", "elapsed", time.Since(start))
412
	}(time.Now())
413

414 415 416 417 418 419
	// Look up the sync boundaries: the common ancestor and the target block
	latest, err := d.fetchHeight(p)
	if err != nil {
		return err
	}
	height := latest.Number.Uint64()
420

421 422 423 424 425 426 427 428 429 430
	origin, err := d.findAncestor(p, height)
	if err != nil {
		return err
	}
	d.syncStatsLock.Lock()
	if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin {
		d.syncStatsChainOrigin = origin
	}
	d.syncStatsChainHeight = height
	d.syncStatsLock.Unlock()
431

432 433 434 435 436 437 438 439 440 441 442
	// Initiate the sync using a concurrent header and content retrieval algorithm
	pivot := uint64(0)
	switch d.mode {
	case LightSync:
		pivot = height
	case FastSync:
		// Calculate the new fast/slow sync pivot point
		if d.fsPivotLock == nil {
			pivotOffset, err := rand.Int(rand.Reader, big.NewInt(int64(fsPivotInterval)))
			if err != nil {
				panic(fmt.Sprintf("Failed to access crypto random source: %v", err))
443
			}
444 445
			if height > uint64(fsMinFullBlocks)+pivotOffset.Uint64() {
				pivot = height - uint64(fsMinFullBlocks) - pivotOffset.Uint64()
446
			}
447 448 449
		} else {
			// Pivot point locked in, use this and do not pick a new one!
			pivot = d.fsPivotLock.Number.Uint64()
450
		}
451 452 453 454 455 456 457
		// If the point is below the origin, move origin back to ensure state download
		if pivot < origin {
			if pivot > 0 {
				origin = pivot - 1
			} else {
				origin = 0
			}
458
		}
459
		log.Debug("Fast syncing until pivot block", "pivot", pivot)
460 461 462 463
	}
	d.queue.Prepare(origin+1, d.mode, pivot, latest)
	if d.syncInitHook != nil {
		d.syncInitHook(origin, height)
464
	}
465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482

	fetchers := []func() error{
		func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved
		func() error { return d.fetchBodies(origin + 1) },   // Bodies are retrieved during normal and fast sync
		func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync
		func() error { return d.processHeaders(origin+1, td) },
	}
	if d.mode == FastSync {
		fetchers = append(fetchers, func() error { return d.processFastSyncContent(latest) })
	} else if d.mode == FullSync {
		fetchers = append(fetchers, d.processFullSyncContent)
	}
	err = d.spawnSync(fetchers)
	if err != nil && d.mode == FastSync && d.fsPivotLock != nil {
		// If sync failed in the critical section, bump the fail counter.
		atomic.AddUint32(&d.fsPivotFails, 1)
	}
	return err
483 484 485 486
}

// spawnSync runs d.process and all given fetcher functions to completion in
// separate goroutines, returning the first error that appears.
487
func (d *Downloader) spawnSync(fetchers []func() error) error {
488
	var wg sync.WaitGroup
489 490
	errc := make(chan error, len(fetchers))
	wg.Add(len(fetchers))
491 492 493 494 495 496
	for _, fn := range fetchers {
		fn := fn
		go func() { defer wg.Done(); errc <- fn() }()
	}
	// Wait for the first error, then terminate the others.
	var err error
497 498
	for i := 0; i < len(fetchers); i++ {
		if i == len(fetchers)-1 {
499 500 501 502 503 504 505 506 507 508
			// Close the queue when all fetchers have exited.
			// This will cause the block processor to end when
			// it has processed the queue.
			d.queue.Close()
		}
		if err = <-errc; err != nil {
			break
		}
	}
	d.queue.Close()
509
	d.Cancel()
510 511
	wg.Wait()
	return err
512 513
}

514
// Cancel cancels all of the operations and resets the queue. It returns true
515
// if the cancel operation was completed.
516
func (d *Downloader) Cancel() {
517
	// Close the current cancel channel
518
	d.cancelLock.Lock()
519 520 521 522 523 524 525
	if d.cancelCh != nil {
		select {
		case <-d.cancelCh:
			// Channel was already closed
		default:
			close(d.cancelCh)
		}
526 527
	}
	d.cancelLock.Unlock()
528 529
}

530
// Terminate interrupts the downloader, canceling all pending operations.
531
// The downloader cannot be reused after calling Terminate.
532
func (d *Downloader) Terminate() {
533 534 535 536 537 538 539 540 541 542
	// Close the termination channel (make sure double close is allowed)
	d.quitLock.Lock()
	select {
	case <-d.quitCh:
	default:
		close(d.quitCh)
	}
	d.quitLock.Unlock()

	// Cancel any pending download requests
543
	d.Cancel()
544 545
}

546 547
// fetchHeight retrieves the head header of the remote peer to aid in estimating
// the total time a pending synchronisation would take.
548
func (d *Downloader) fetchHeight(p *peerConnection) (*types.Header, error) {
P
Péter Szilágyi 已提交
549
	p.log.Debug("Retrieving remote chain height")
550 551

	// Request the advertised remote head block and wait for the response
552 553
	head, _ := p.peer.Head()
	go p.peer.RequestHeadersByHash(head, 1, 0, false)
554

555 556
	ttl := d.requestTTL()
	timeout := time.After(ttl)
557 558 559
	for {
		select {
		case <-d.cancelCh:
560
			return nil, errCancelBlockFetch
561

562
		case packet := <-d.headerCh:
563
			// Discard anything not from the origin peer
564
			if packet.PeerId() != p.id {
565
				log.Debug("Received headers from incorrect peer", "peer", packet.PeerId())
566 567 568
				break
			}
			// Make sure the peer actually gave something valid
569
			headers := packet.(*headerPack).headers
570
			if len(headers) != 1 {
P
Péter Szilágyi 已提交
571
				p.log.Debug("Multiple headers for single request", "headers", len(headers))
572
				return nil, errBadPeer
573
			}
574
			head := headers[0]
P
Péter Szilágyi 已提交
575
			p.log.Debug("Remote head header identified", "number", head.Number, "hash", head.Hash())
576
			return head, nil
577

578
		case <-timeout:
P
Péter Szilágyi 已提交
579
			p.log.Debug("Waiting for head header timed out", "elapsed", ttl)
580
			return nil, errTimeout
581

582
		case <-d.bodyCh:
583 584
		case <-d.receiptCh:
			// Out of bounds delivery, ignore
585 586 587 588
		}
	}
}

589
// findAncestor tries to locate the common ancestor link of the local chain and
590
// a remote peers blockchain. In the general case when our node was in sync and
591
// on the correct chain, checking the top N links should already get us a match.
592
// In the rare scenario when we ended up on a long reorganisation (i.e. none of
593
// the head links match), we do a binary search to find the common ancestor.
594
func (d *Downloader) findAncestor(p *peerConnection, height uint64) (uint64, error) {
595
	// Figure out the valid ancestor range to prevent rewrite attacks
596
	floor, ceil := int64(-1), d.lightchain.CurrentHeader().Number.Uint64()
597

P
Péter Szilágyi 已提交
598
	p.log.Debug("Looking for common ancestor", "local", ceil, "remote", height)
599
	if d.mode == FullSync {
600
		ceil = d.chain.CurrentBlock().NumberU64()
601
	} else if d.mode == FastSync {
602
		ceil = d.chain.CurrentFastBlock().NumberU64()
603 604 605 606 607 608 609 610
	}
	if ceil >= MaxForkAncestry {
		floor = int64(ceil - MaxForkAncestry)
	}
	// Request the topmost blocks to short circuit binary ancestor lookup
	head := ceil
	if head > height {
		head = height
611
	}
612
	from := int64(head) - int64(MaxHeaderFetch)
613 614 615
	if from < 0 {
		from = 0
	}
616 617 618 619 620 621
	// Span out with 15 block gaps into the future to catch bad head reports
	limit := 2 * MaxHeaderFetch / 16
	count := 1 + int((int64(ceil)-from)/16)
	if count > limit {
		count = limit
	}
622
	go p.peer.RequestHeadersByNumber(uint64(from), count, 15, false)
623 624 625

	// Wait for the remote response to the head fetch
	number, hash := uint64(0), common.Hash{}
626 627 628

	ttl := d.requestTTL()
	timeout := time.After(ttl)
629 630 631 632

	for finished := false; !finished; {
		select {
		case <-d.cancelCh:
633
			return 0, errCancelHeaderFetch
634

635
		case packet := <-d.headerCh:
636
			// Discard anything not from the origin peer
637
			if packet.PeerId() != p.id {
638
				log.Debug("Received headers from incorrect peer", "peer", packet.PeerId())
639 640 641
				break
			}
			// Make sure the peer actually gave something valid
642
			headers := packet.(*headerPack).headers
643
			if len(headers) == 0 {
P
Péter Szilágyi 已提交
644
				p.log.Warn("Empty head header set")
645 646
				return 0, errEmptyHeaderSet
			}
647 648
			// Make sure the peer's reply conforms to the request
			for i := 0; i < len(headers); i++ {
649
				if number := headers[i].Number.Int64(); number != from+int64(i)*16 {
P
Péter Szilágyi 已提交
650
					p.log.Warn("Head headers broke chain ordering", "index", i, "requested", from+int64(i)*16, "received", number)
651 652 653
					return 0, errInvalidChain
				}
			}
654 655 656
			// Check if a common ancestor was found
			finished = true
			for i := len(headers) - 1; i >= 0; i-- {
657
				// Skip any headers that underflow/overflow our requested set
658
				if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > ceil {
659 660 661
					continue
				}
				// Otherwise check if we already know the header or not
662
				if (d.mode == FullSync && d.chain.HasBlockAndState(headers[i].Hash())) || (d.mode != FullSync && d.lightchain.HasHeader(headers[i].Hash())) {
663
					number, hash = headers[i].Number.Uint64(), headers[i].Hash()
664 665 666

					// If every header is known, even future ones, the peer straight out lied about its head
					if number > height && i == limit-1 {
P
Péter Szilágyi 已提交
667
						p.log.Warn("Lied about chain head", "reported", height, "found", number)
668 669
						return 0, errStallingPeer
					}
670 671 672 673
					break
				}
			}

674
		case <-timeout:
P
Péter Szilágyi 已提交
675
			p.log.Debug("Waiting for head header timed out", "elapsed", ttl)
676 677
			return 0, errTimeout

678
		case <-d.bodyCh:
679 680
		case <-d.receiptCh:
			// Out of bounds delivery, ignore
681 682 683 684
		}
	}
	// If the head fetch already found an ancestor, return
	if !common.EmptyHash(hash) {
685
		if int64(number) <= floor {
P
Péter Szilágyi 已提交
686
			p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor)
687 688
			return 0, errInvalidAncestor
		}
P
Péter Szilágyi 已提交
689
		p.log.Debug("Found common ancestor", "number", number, "hash", hash)
690 691 692 693
		return number, nil
	}
	// Ancestor not found, we need to binary search over our chain
	start, end := uint64(0), head
694 695 696
	if floor > 0 {
		start = uint64(floor)
	}
697 698 699 700
	for start+1 < end {
		// Split our chain interval in two, and request the hash to cross check
		check := (start + end) / 2

701 702 703
		ttl := d.requestTTL()
		timeout := time.After(ttl)

704
		go p.peer.RequestHeadersByNumber(uint64(check), 1, 0, false)
705 706 707 708 709

		// Wait until a reply arrives to this request
		for arrived := false; !arrived; {
			select {
			case <-d.cancelCh:
710
				return 0, errCancelHeaderFetch
711

712
			case packer := <-d.headerCh:
713
				// Discard anything not from the origin peer
714
				if packer.PeerId() != p.id {
715
					log.Debug("Received headers from incorrect peer", "peer", packer.PeerId())
716 717 718
					break
				}
				// Make sure the peer actually gave something valid
719
				headers := packer.(*headerPack).headers
720
				if len(headers) != 1 {
P
Péter Szilágyi 已提交
721
					p.log.Debug("Multiple headers for single request", "headers", len(headers))
722 723 724 725 726
					return 0, errBadPeer
				}
				arrived = true

				// Modify the search interval based on the response
727
				if (d.mode == FullSync && !d.chain.HasBlockAndState(headers[0].Hash())) || (d.mode != FullSync && !d.lightchain.HasHeader(headers[0].Hash())) {
728 729 730
					end = check
					break
				}
731
				header := d.lightchain.GetHeaderByHash(headers[0].Hash()) // Independent of sync mode, header surely exists
732
				if header.Number.Uint64() != check {
P
Péter Szilágyi 已提交
733
					p.log.Debug("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check)
734 735 736 737
					return 0, errBadPeer
				}
				start = check

738
			case <-timeout:
P
Péter Szilágyi 已提交
739
				p.log.Debug("Waiting for search header timed out", "elapsed", ttl)
740 741
				return 0, errTimeout

742
			case <-d.bodyCh:
743 744
			case <-d.receiptCh:
				// Out of bounds delivery, ignore
745 746 747
			}
		}
	}
748 749
	// Ensure valid ancestry and return
	if int64(start) <= floor {
P
Péter Szilágyi 已提交
750
		p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor)
751 752
		return 0, errInvalidAncestor
	}
P
Péter Szilágyi 已提交
753
	p.log.Debug("Found common ancestor", "number", start, "hash", hash)
754 755 756
	return start, nil
}

757 758 759 760 761
// fetchHeaders keeps retrieving headers concurrently from the number
// requested, until no more are returned, potentially throttling on the way. To
// facilitate concurrency but still protect against malicious nodes sending bad
// headers, we construct a header chain skeleton using the "origin" peer we are
// syncing with, and fill in the missing headers using anyone else. Headers from
762
// other peers are only accepted if they map cleanly to the skeleton. If no one
763 764
// can fill in the skeleton - not even the origin peer - it's assumed invalid and
// the origin is dropped.
765
func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error {
P
Péter Szilágyi 已提交
766 767
	p.log.Debug("Directing header downloads", "origin", from)
	defer p.log.Debug("Header download terminated")
768

769 770 771
	// Create a timeout timer, and the associated header fetcher
	skeleton := true            // Skeleton assembly phase or finishing up
	request := time.Now()       // time of the last skeleton fetch request
772 773 774 775
	timeout := time.NewTimer(0) // timer to dump a non-responsive active peer
	<-timeout.C                 // timeout channel should be initially empty
	defer timeout.Stop()

776
	var ttl time.Duration
777
	getHeaders := func(from uint64) {
778
		request = time.Now()
779 780 781

		ttl = d.requestTTL()
		timeout.Reset(ttl)
782

783
		if skeleton {
P
Péter Szilágyi 已提交
784
			p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from)
785
			go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false)
786
		} else {
P
Péter Szilágyi 已提交
787
			p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from)
788
			go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false)
789
		}
790
	}
791
	// Start pulling the header chain skeleton until all is done
792 793 794 795 796 797 798
	getHeaders(from)

	for {
		select {
		case <-d.cancelCh:
			return errCancelHeaderFetch

799
		case packet := <-d.headerCh:
800
			// Make sure the active peer is giving us the skeleton headers
801
			if packet.PeerId() != p.id {
802
				log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId())
803 804
				break
			}
805
			headerReqTimer.UpdateSince(request)
806 807
			timeout.Stop()

808 809 810 811 812 813
			// If the skeleton's finished, pull any remaining head headers directly from the origin
			if packet.Items() == 0 && skeleton {
				skeleton = false
				getHeaders(from)
				continue
			}
814
			// If no more headers are inbound, notify the content fetchers and return
815
			if packet.Items() == 0 {
P
Péter Szilágyi 已提交
816
				p.log.Debug("No more headers available")
817 818 819 820 821 822
				select {
				case d.headerProcCh <- nil:
					return nil
				case <-d.cancelCh:
					return errCancelHeaderFetch
				}
823
			}
824
			headers := packet.(*headerPack).headers
825

826 827
			// If we received a skeleton batch, resolve internals concurrently
			if skeleton {
828
				filled, proced, err := d.fillHeaderSkeleton(from, headers)
829
				if err != nil {
P
Péter Szilágyi 已提交
830
					p.log.Debug("Skeleton chain invalid", "err", err)
831 832
					return errInvalidChain
				}
833 834
				headers = filled[proced:]
				from += uint64(proced)
835
			}
836
			// Insert all the new headers and fetch the next batch
837
			if len(headers) > 0 {
P
Péter Szilágyi 已提交
838
				p.log.Trace("Scheduling new headers", "count", len(headers), "from", from)
839 840 841 842 843 844
				select {
				case d.headerProcCh <- headers:
				case <-d.cancelCh:
					return errCancelHeaderFetch
				}
				from += uint64(len(headers))
845
			}
846 847 848 849
			getHeaders(from)

		case <-timeout.C:
			// Header retrieval timed out, consider the peer bad and drop
P
Péter Szilágyi 已提交
850
			p.log.Debug("Header request timed out", "elapsed", ttl)
851
			headerTimeoutMeter.Mark(1)
852 853 854
			d.dropPeer(p.id)

			// Finish the sync gracefully instead of dumping the gathered data though
855
			for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} {
856 857 858 859
				select {
				case ch <- false:
				case <-d.cancelCh:
				}
860
			}
861 862 863 864 865
			select {
			case d.headerProcCh <- nil:
			case <-d.cancelCh:
			}
			return errBadPeer
866 867 868 869
		}
	}
}

870 871
// fillHeaderSkeleton concurrently retrieves headers from all our available peers
// and maps them to the provided skeleton header chain.
872 873 874 875 876 877 878 879
//
// Any partial results from the beginning of the skeleton is (if possible) forwarded
// immediately to the header processor to keep the rest of the pipeline full even
// in the case of header stalls.
//
// The method returs the entire filled skeleton and also the number of headers
// already forwarded for processing.
func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) {
880
	log.Debug("Filling up skeleton", "from", from)
881 882 883 884 885
	d.queue.ScheduleSkeleton(from, skeleton)

	var (
		deliver = func(packet dataPack) (int, error) {
			pack := packet.(*headerPack)
886
			return d.queue.DeliverHeaders(pack.peerId, pack.headers, d.headerProcCh)
887
		}
888
		expire   = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) }
889
		throttle = func() bool { return false }
890
		reserve  = func(p *peerConnection, count int) (*fetchRequest, bool, error) {
891 892
			return d.queue.ReserveHeaders(p, count), false, nil
		}
893 894 895
		fetch    = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) }
		capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) }
		setIdle  = func(p *peerConnection, accepted int) { p.SetHeadersIdle(accepted) }
896 897 898
	)
	err := d.fetchParts(errCancelHeaderFetch, d.headerCh, deliver, d.queue.headerContCh, expire,
		d.queue.PendingHeaders, d.queue.InFlightHeaders, throttle, reserve,
899
		nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers")
900

901
	log.Debug("Skeleton fill terminated", "err", err)
902 903 904

	filled, proced := d.queue.RetrieveHeaders()
	return filled, proced, err
905 906
}

907 908 909 910
// fetchBodies iteratively downloads the scheduled block bodies, taking any
// available peers, reserving a chunk of blocks for each, waiting for delivery
// and also periodically checking for timeouts.
func (d *Downloader) fetchBodies(from uint64) error {
911
	log.Debug("Downloading block bodies", "origin", from)
912

913
	var (
914
		deliver = func(packet dataPack) (int, error) {
915
			pack := packet.(*bodyPack)
916
			return d.queue.DeliverBodies(pack.peerId, pack.transactions, pack.uncles)
917
		}
918
		expire   = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) }
919 920 921
		fetch    = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) }
		capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) }
		setIdle  = func(p *peerConnection, accepted int) { p.SetBodiesIdle(accepted) }
922
	)
923
	err := d.fetchParts(errCancelBodyFetch, d.bodyCh, deliver, d.bodyWakeCh, expire,
924
		d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ShouldThrottleBlocks, d.queue.ReserveBodies,
925
		d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies")
926

927
	log.Debug("Block body download terminated", "err", err)
928 929 930 931 932 933 934
	return err
}

// fetchReceipts iteratively downloads the scheduled block receipts, taking any
// available peers, reserving a chunk of receipts for each, waiting for delivery
// and also periodically checking for timeouts.
func (d *Downloader) fetchReceipts(from uint64) error {
935
	log.Debug("Downloading transaction receipts", "origin", from)
936 937

	var (
938
		deliver = func(packet dataPack) (int, error) {
939 940 941
			pack := packet.(*receiptPack)
			return d.queue.DeliverReceipts(pack.peerId, pack.receipts)
		}
942
		expire   = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) }
943 944 945
		fetch    = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) }
		capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) }
		setIdle  = func(p *peerConnection, accepted int) { p.SetReceiptsIdle(accepted) }
946
	)
947
	err := d.fetchParts(errCancelReceiptFetch, d.receiptCh, deliver, d.receiptWakeCh, expire,
948
		d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ShouldThrottleReceipts, d.queue.ReserveReceipts,
949
		d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts")
950

951
	log.Debug("Transaction receipt download terminated", "err", err)
952 953 954 955 956 957
	return err
}

// fetchParts iteratively downloads scheduled block parts, taking any available
// peers, reserving a chunk of fetch requests for each, waiting for delivery and
// also periodically checking for timeouts.
958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975
//
// As the scheduling/timeout logic mostly is the same for all downloaded data
// types, this method is used by each for data gathering and is instrumented with
// various callbacks to handle the slight differences between processing them.
//
// The instrumentation parameters:
//  - errCancel:   error type to return if the fetch operation is cancelled (mostly makes logging nicer)
//  - deliveryCh:  channel from which to retrieve downloaded data packets (merged from all concurrent peers)
//  - deliver:     processing callback to deliver data packets into type specific download queues (usually within `queue`)
//  - wakeCh:      notification channel for waking the fetcher when new tasks are available (or sync completed)
//  - expire:      task callback method to abort requests that took too long and return the faulty peers (traffic shaping)
//  - pending:     task callback for the number of requests still needing download (detect completion/non-completability)
//  - inFlight:    task callback for the number of in-progress requests (wait for all active downloads to finish)
//  - throttle:    task callback to check if the processing queue is full and activate throttling (bound memory use)
//  - reserve:     task callback to reserve new download tasks to a particular peer (also signals partial completions)
//  - fetchHook:   tester callback to notify of new tasks being initiated (allows testing the scheduling logic)
//  - fetch:       network callback to actually send a particular download request to a physical remote peer
//  - cancel:      task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer)
P
Péter Szilágyi 已提交
976
//  - capacity:    network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping)
977 978 979
//  - idle:        network callback to retrieve the currently (type specific) idle peers that can be assigned tasks
//  - setIdle:     network callback to set a peer back to idle and update its estimated capacity (traffic shaping)
//  - kind:        textual label of the type being downloaded to display in log mesages
980
func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool,
981 982 983
	expire func() map[string]int, pending func() int, inFlight func() bool, throttle func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, error),
	fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int,
	idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int), kind string) error {
984

985
	// Create a ticker to detect expired retrieval tasks
986 987 988 989 990
	ticker := time.NewTicker(100 * time.Millisecond)
	defer ticker.Stop()

	update := make(chan struct{}, 1)

991
	// Prepare the queue and fetch block parts until the block header fetcher's done
992 993 994 995
	finished := false
	for {
		select {
		case <-d.cancelCh:
996
			return errCancel
997

998
		case packet := <-deliveryCh:
999 1000
			// If the peer was previously banned and failed to deliver it's pack
			// in a reasonable time frame, ignore it's message.
1001
			if peer := d.peers.Peer(packet.PeerId()); peer != nil {
1002 1003 1004
				// Deliver the received chunk of data and check chain validity
				accepted, err := deliver(packet)
				if err == errInvalidChain {
1005
					return err
1006 1007 1008 1009 1010 1011 1012 1013 1014 1015
				}
				// Unless a peer delivered something completely else than requested (usually
				// caused by a timed out request which came through in the end), set it to
				// idle. If the delivery's stale, the peer should have already been idled.
				if err != errStaleDelivery {
					setIdle(peer, accepted)
				}
				// Issue a log to the user to see what's going on
				switch {
				case err == nil && packet.Items() == 0:
P
Péter Szilágyi 已提交
1016
					peer.log.Trace("Requested data not delivered", "type", kind)
1017
				case err == nil:
P
Péter Szilágyi 已提交
1018
					peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats())
1019
				default:
P
Péter Szilágyi 已提交
1020
					peer.log.Trace("Failed to deliver retrieved data", "type", kind, "err", err)
1021 1022 1023 1024 1025 1026 1027 1028
				}
			}
			// Blocks assembled, try to update the progress
			select {
			case update <- struct{}{}:
			default:
			}

1029
		case cont := <-wakeCh:
1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051
			// The header fetcher sent a continuation flag, check if it's done
			if !cont {
				finished = true
			}
			// Headers arrive, try to update the progress
			select {
			case update <- struct{}{}:
			default:
			}

		case <-ticker.C:
			// Sanity check update the progress
			select {
			case update <- struct{}{}:
			default:
			}

		case <-update:
			// Short circuit if we lost all our peers
			if d.peers.Len() == 0 {
				return errNoPeers
			}
1052
			// Check for fetch request timeouts and demote the responsible peers
1053
			for pid, fails := range expire() {
1054
				if peer := d.peers.Peer(pid); peer != nil {
1055 1056 1057 1058 1059 1060 1061 1062
					// If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps
					// ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times
					// out that sync wise we need to get rid of the peer.
					//
					// The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth
					// and latency of a peer separately, which requires pushing the measures capacity a bit and seeing
					// how response times reacts, to it always requests one more than the minimum (i.e. min 2).
					if fails > 2 {
P
Péter Szilágyi 已提交
1063
						peer.log.Trace("Data delivery timed out", "type", kind)
1064 1065
						setIdle(peer, 0)
					} else {
P
Péter Szilágyi 已提交
1066
						peer.log.Debug("Stalling delivery, dropping", "type", kind)
1067 1068
						d.dropPeer(pid)
					}
1069 1070
				}
			}
1071 1072
			// If there's nothing more to fetch, wait or terminate
			if pending() == 0 {
1073
				if !inFlight() && finished {
1074
					log.Debug("Data fetching completed", "type", kind)
1075 1076 1077 1078 1079
					return nil
				}
				break
			}
			// Send a download request to all idle peers, until throttled
1080
			progressed, throttled, running := false, false, inFlight()
1081 1082 1083
			idles, total := idle()

			for _, peer := range idles {
1084
				// Short circuit if throttling activated
1085
				if throttle() {
1086 1087 1088
					throttled = true
					break
				}
1089 1090
				// Reserve a chunk of fetches for a peer. A nil can mean either that
				// no more headers are available, or that the peer is known not to
1091
				// have them.
1092
				request, progress, err := reserve(peer, capacity(peer))
1093 1094 1095
				if err != nil {
					return err
				}
1096 1097
				if progress {
					progressed = true
1098 1099 1100 1101
				}
				if request == nil {
					continue
				}
1102
				if request.From > 0 {
P
Péter Szilágyi 已提交
1103
					peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From)
1104
				} else if len(request.Headers) > 0 {
P
Péter Szilágyi 已提交
1105
					peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number)
1106
				} else {
P
Péter Szilágyi 已提交
1107
					peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Hashes))
1108
				}
1109
				// Fetch the chunk and make sure any errors return the hashes to the queue
1110 1111
				if fetchHook != nil {
					fetchHook(request.Headers)
1112
				}
1113
				if err := fetch(peer, request); err != nil {
1114 1115 1116 1117 1118
					// Although we could try and make an attempt to fix this, this error really
					// means that we've double allocated a fetch task to a peer. If that is the
					// case, the internal state of the downloader and the queue is very wrong so
					// better hard crash and note the error instead of silently accumulating into
					// a much bigger issue.
1119
					panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind))
1120
				}
1121
				running = true
1122 1123 1124
			}
			// Make sure that we have peers available for fetching. If all peers have been tried
			// and all failed throw an error
1125
			if !progressed && !throttled && !running && len(idles) == total && pending() > 0 {
1126 1127 1128 1129 1130 1131
				return errPeersUnavailable
			}
		}
	}
}

1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147
// processHeaders takes batches of retrieved headers from an input channel and
// keeps processing and scheduling them into the header chain and downloader's
// queue until the stream ends or a failure occurs.
func (d *Downloader) processHeaders(origin uint64, td *big.Int) error {
	// Calculate the pivoting point for switching from fast to slow sync
	pivot := d.queue.FastSyncPivot()

	// Keep a count of uncertain headers to roll back
	rollback := []*types.Header{}
	defer func() {
		if len(rollback) > 0 {
			// Flatten the headers and roll them back
			hashes := make([]common.Hash, len(rollback))
			for i, header := range rollback {
				hashes[i] = header.Hash()
			}
1148 1149 1150 1151
			lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0
			if d.mode != LightSync {
				lastFastBlock = d.chain.CurrentFastBlock().Number()
				lastBlock = d.chain.CurrentBlock().Number()
1152
			}
1153
			d.lightchain.Rollback(hashes)
1154
			curFastBlock, curBlock := common.Big0, common.Big0
1155 1156 1157
			if d.mode != LightSync {
				curFastBlock = d.chain.CurrentFastBlock().Number()
				curBlock = d.chain.CurrentBlock().Number()
1158
			}
1159
			log.Warn("Rolled back headers", "count", len(hashes),
1160
				"header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number),
1161 1162
				"fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock),
				"block", fmt.Sprintf("%d->%d", lastBlock, curBlock))
1163

1164
			// If we're already past the pivot point, this could be an attack, thread carefully
1165
			if rollback[len(rollback)-1].Number.Uint64() > pivot {
1166
				// If we didn't ever fail, lock in te pivot header (must! not! change!)
1167
				if atomic.LoadUint32(&d.fsPivotFails) == 0 {
1168 1169
					for _, header := range rollback {
						if header.Number.Uint64() == pivot {
1170
							log.Warn("Fast-sync pivot locked in", "number", pivot, "hash", header.Hash())
1171 1172 1173 1174
							d.fsPivotLock = header
						}
					}
				}
1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190
			}
		}
	}()

	// Wait for batches of headers to process
	gotHeaders := false

	for {
		select {
		case <-d.cancelCh:
			return errCancelHeaderProcessing

		case headers := <-d.headerProcCh:
			// Terminate header processing if we synced up
			if len(headers) == 0 {
				// Notify everyone that headers are fully processed
1191
				for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} {
1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208
					select {
					case ch <- false:
					case <-d.cancelCh:
					}
				}
				// If no headers were retrieved at all, the peer violated it's TD promise that it had a
				// better chain compared to ours. The only exception is if it's promised blocks were
				// already imported by other means (e.g. fecher):
				//
				// R <remote peer>, L <local node>: Both at block 10
				// R: Mine block 11, and propagate it to L
				// L: Queue block 11 for import
				// L: Notice that R's head and TD increased compared to ours, start sync
				// L: Import of block 11 finishes
				// L: Sync begins, and finds common ancestor at 11
				// L: Request new headers up from 11 (R's TD was higher, it must have something)
				// R: Nothing to give
1209
				if d.mode != LightSync {
1210
					if !gotHeaders && td.Cmp(d.chain.GetTdByHash(d.chain.CurrentBlock().Hash())) > 0 {
1211 1212
						return errStallingPeer
					}
1213 1214 1215 1216 1217 1218 1219 1220 1221
				}
				// If fast or light syncing, ensure promised headers are indeed delivered. This is
				// needed to detect scenarios where an attacker feeds a bad pivot and then bails out
				// of delivering the post-pivot blocks that would flag the invalid content.
				//
				// This check cannot be executed "as is" for full imports, since blocks may still be
				// queued for processing when the header download completes. However, as long as the
				// peer gave us something useful, we're already happy/progressed (above check).
				if d.mode == FastSync || d.mode == LightSync {
1222
					if td.Cmp(d.lightchain.GetTdByHash(d.lightchain.CurrentHeader().Hash())) > 0 {
1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251
						return errStallingPeer
					}
				}
				// Disable any rollback and return
				rollback = nil
				return nil
			}
			// Otherwise split the chunk of headers into batches and process them
			gotHeaders = true

			for len(headers) > 0 {
				// Terminate if something failed in between processing chunks
				select {
				case <-d.cancelCh:
					return errCancelHeaderProcessing
				default:
				}
				// Select the next chunk of headers to import
				limit := maxHeadersProcess
				if limit > len(headers) {
					limit = len(headers)
				}
				chunk := headers[:limit]

				// In case of header only syncing, validate the chunk immediately
				if d.mode == FastSync || d.mode == LightSync {
					// Collect the yet unknown headers to mark them as uncertain
					unknown := make([]*types.Header, 0, len(headers))
					for _, header := range chunk {
1252
						if !d.lightchain.HasHeader(header.Hash()) {
1253 1254 1255 1256 1257 1258 1259 1260
							unknown = append(unknown, header)
						}
					}
					// If we're importing pure headers, verify based on their recentness
					frequency := fsHeaderCheckFrequency
					if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot {
						frequency = 1
					}
1261
					if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil {
1262 1263 1264 1265
						// If some headers were inserted, add them too to the rollback list
						if n > 0 {
							rollback = append(rollback, chunk[:n]...)
						}
1266
						log.Debug("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "err", err)
1267 1268 1269 1270 1271 1272 1273 1274
						return errInvalidChain
					}
					// All verifications passed, store newly found uncertain headers
					rollback = append(rollback, unknown...)
					if len(rollback) > fsHeaderSafetyNet {
						rollback = append(rollback[:0], rollback[len(rollback)-fsHeaderSafetyNet:]...)
					}
				}
1275 1276 1277
				// If we're fast syncing and just pulled in the pivot, make sure it's the one locked in
				if d.mode == FastSync && d.fsPivotLock != nil && chunk[0].Number.Uint64() <= pivot && chunk[len(chunk)-1].Number.Uint64() >= pivot {
					if pivot := chunk[int(pivot-chunk[0].Number.Uint64())]; pivot.Hash() != d.fsPivotLock.Hash() {
1278
						log.Warn("Pivot doesn't match locked in one", "remoteNumber", pivot.Number, "remoteHash", pivot.Hash(), "localNumber", d.fsPivotLock.Number, "localHash", d.fsPivotLock.Hash())
1279 1280 1281
						return errInvalidChain
					}
				}
1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294
				// Unless we're doing light chains, schedule the headers for associated content retrieval
				if d.mode == FullSync || d.mode == FastSync {
					// If we've reached the allowed number of pending headers, stall a bit
					for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders {
						select {
						case <-d.cancelCh:
							return errCancelHeaderProcessing
						case <-time.After(time.Second):
						}
					}
					// Otherwise insert the headers for content retrieval
					inserts := d.queue.Schedule(chunk, origin)
					if len(inserts) != len(chunk) {
1295
						log.Debug("Stale headers")
1296 1297 1298 1299 1300 1301 1302
						return errBadPeer
					}
				}
				headers = headers[limit:]
				origin += uint64(limit)
			}
			// Signal the content downloaders of the availablility of new tasks
1303
			for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} {
1304 1305 1306 1307 1308 1309 1310 1311 1312
				select {
				case ch <- true:
				default:
				}
			}
		}
	}
}

1313 1314
// processFullSyncContent takes fetch results from the queue and imports them into the chain.
func (d *Downloader) processFullSyncContent() error {
1315
	for {
1316
		results := d.queue.WaitResults()
1317
		if len(results) == 0 {
1318
			return nil
1319
		}
1320
		if d.chainInsertHook != nil {
1321
			d.chainInsertHook(results)
1322
		}
1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339
		if err := d.importBlockResults(results); err != nil {
			return err
		}
	}
}

func (d *Downloader) importBlockResults(results []*fetchResult) error {
	for len(results) != 0 {
		// Check for any termination requests. This makes clean shutdown faster.
		select {
		case <-d.quitCh:
			return errCancelContentProcessing
		default:
		}
		// Retrieve the a batch of results to import
		items := int(math.Min(float64(len(results)), float64(maxResultsProcess)))
		first, last := results[0].Header, results[items-1].Header
1340
		log.Debug("Inserting downloaded chain", "items", len(results),
1341 1342 1343
			"firstnum", first.Number, "firsthash", first.Hash(),
			"lastnum", last.Number, "lasthash", last.Hash(),
		)
1344 1345 1346 1347
		blocks := make([]*types.Block, items)
		for i, result := range results[:items] {
			blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles)
		}
1348
		if index, err := d.chain.InsertChain(blocks); err != nil {
1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387
			log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err)
			return errInvalidChain
		}
		// Shift the results to the next batch
		results = results[items:]
	}
	return nil
}

// processFastSyncContent takes fetch results from the queue and writes them to the
// database. It also controls the synchronisation of state nodes of the pivot block.
func (d *Downloader) processFastSyncContent(latest *types.Header) error {
	// Start syncing state of the reported head block.
	// This should get us most of the state of the pivot block.
	stateSync := d.syncState(latest.Root)
	defer stateSync.Cancel()
	go func() {
		if err := stateSync.Wait(); err != nil {
			d.queue.Close() // wake up WaitResults
		}
	}()

	pivot := d.queue.FastSyncPivot()
	for {
		results := d.queue.WaitResults()
		if len(results) == 0 {
			return stateSync.Cancel()
		}
		if d.chainInsertHook != nil {
			d.chainInsertHook(results)
		}
		P, beforeP, afterP := splitAroundPivot(pivot, results)
		if err := d.commitFastSyncData(beforeP, stateSync); err != nil {
			return err
		}
		if P != nil {
			stateSync.Cancel()
			if err := d.commitPivotBlock(P); err != nil {
				return err
1388

1389
			}
1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420
		}
		if err := d.importBlockResults(afterP); err != nil {
			return err
		}
	}
}

func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) {
	for _, result := range results {
		num := result.Header.Number.Uint64()
		switch {
		case num < pivot:
			before = append(before, result)
		case num == pivot:
			p = result
		default:
			after = append(after, result)
		}
	}
	return p, before, after
}

func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error {
	for len(results) != 0 {
		// Check for any termination requests.
		select {
		case <-d.quitCh:
			return errCancelContentProcessing
		case <-stateSync.done:
			if err := stateSync.Wait(); err != nil {
				return err
1421
			}
1422
		default:
1423
		}
1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436
		// Retrieve the a batch of results to import
		items := int(math.Min(float64(len(results)), float64(maxResultsProcess)))
		first, last := results[0].Header, results[items-1].Header
		log.Debug("Inserting fast-sync blocks", "items", len(results),
			"firstnum", first.Number, "firsthash", first.Hash(),
			"lastnumn", last.Number, "lasthash", last.Hash(),
		)
		blocks := make([]*types.Block, items)
		receipts := make([]types.Receipts, items)
		for i, result := range results[:items] {
			blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles)
			receipts[i] = result.Receipts
		}
1437
		if index, err := d.chain.InsertReceiptChain(blocks, receipts); err != nil {
1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454
			log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err)
			return errInvalidChain
		}
		// Shift the results to the next batch
		results = results[items:]
	}
	return nil
}

func (d *Downloader) commitPivotBlock(result *fetchResult) error {
	b := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles)
	// Sync the pivot block state. This should complete reasonably quickly because
	// we've already synced up to the reported head block state earlier.
	if err := d.syncState(b.Root()).Wait(); err != nil {
		return err
	}
	log.Debug("Committing fast sync pivot as new head", "number", b.Number(), "hash", b.Hash())
1455
	if _, err := d.chain.InsertReceiptChain([]*types.Block{b}, []types.Receipts{result.Receipts}); err != nil {
1456
		return err
1457
	}
1458
	return d.chain.FastSyncCommitHead(b.Hash())
1459 1460
}

L
Leif Jurvetson 已提交
1461
// DeliverHeaders injects a new batch of block headers received from a remote
1462
// node into the download schedule.
1463
func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) {
1464
	return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter)
1465 1466 1467
}

// DeliverBodies injects a new batch of block bodies received from a remote node.
1468
func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) {
1469
	return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter)
1470 1471 1472 1473
}

// DeliverReceipts injects a new batch of receipts received from a remote node.
func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) {
1474 1475 1476 1477 1478 1479 1480 1481 1482 1483
	return d.deliver(id, d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter)
}

// DeliverNodeData injects a new batch of node state data received from a remote node.
func (d *Downloader) DeliverNodeData(id string, data [][]byte) (err error) {
	return d.deliver(id, d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter)
}

// deliver injects a new batch of data received from a remote node.
func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) {
1484
	// Update the delivery metrics for both good and failed deliveries
1485
	inMeter.Mark(int64(packet.Items()))
1486 1487
	defer func() {
		if err != nil {
1488
			dropMeter.Mark(int64(packet.Items()))
1489 1490 1491 1492 1493 1494
		}
	}()
	// Deliver or abort if the sync is canceled while queuing
	d.cancelLock.RLock()
	cancel := d.cancelCh
	d.cancelLock.RUnlock()
1495 1496 1497
	if cancel == nil {
		return errNoSyncActive
	}
1498
	select {
1499
	case destCh <- packet:
1500 1501 1502 1503
		return nil
	case <-cancel:
		return errNoSyncActive
	}
1504
}
1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519

// qosTuner is the quality of service tuning loop that occasionally gathers the
// peer latency statistics and updates the estimated request round trip time.
func (d *Downloader) qosTuner() {
	for {
		// Retrieve the current median RTT and integrate into the previoust target RTT
		rtt := time.Duration(float64(1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT()))
		atomic.StoreUint64(&d.rttEstimate, uint64(rtt))

		// A new RTT cycle passed, increase our confidence in the estimated RTT
		conf := atomic.LoadUint64(&d.rttConfidence)
		conf = conf + (1000000-conf)/2
		atomic.StoreUint64(&d.rttConfidence, conf)

		// Log the new QoS values and sleep until the next RTT
1520
		log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL())
1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533
		select {
		case <-d.quitCh:
			return
		case <-time.After(rtt):
		}
	}
}

// qosReduceConfidence is meant to be called when a new peer joins the downloader's
// peer set, needing to reduce the confidence we have in out QoS estimates.
func (d *Downloader) qosReduceConfidence() {
	// If we have a single peer, confidence is always 1
	peers := uint64(d.peers.Len())
1534 1535 1536 1537
	if peers == 0 {
		// Ensure peer connectivity races don't catch us off guard
		return
	}
1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553
	if peers == 1 {
		atomic.StoreUint64(&d.rttConfidence, 1000000)
		return
	}
	// If we have a ton of peers, don't drop confidence)
	if peers >= uint64(qosConfidenceCap) {
		return
	}
	// Otherwise drop the confidence factor
	conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers
	if float64(conf)/1000000 < rttMinConfidence {
		conf = uint64(rttMinConfidence * 1000000)
	}
	atomic.StoreUint64(&d.rttConfidence, conf)

	rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate))
1554
	log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL())
1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579
}

// requestRTT returns the current target round trip time for a download request
// to complete in.
//
// Note, the returned RTT is .9 of the actually estimated RTT. The reason is that
// the downloader tries to adapt queries to the RTT, so multiple RTT values can
// be adapted to, but smaller ones are preffered (stabler download stream).
func (d *Downloader) requestRTT() time.Duration {
	return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10
}

// requestTTL returns the current timeout allowance for a single download request
// to finish under.
func (d *Downloader) requestTTL() time.Duration {
	var (
		rtt  = time.Duration(atomic.LoadUint64(&d.rttEstimate))
		conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0
	)
	ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf)
	if ttl > ttlLimit {
		ttl = ttlLimit
	}
	return ttl
}