downloader.go 60.3 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
	"strings"
27 28 29 30 31 32
	"sync"
	"sync/atomic"
	"time"

	"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 36
	"github.com/ethereum/go-ethereum/logger"
	"github.com/ethereum/go-ethereum/logger/glog"
37
	"github.com/rcrowley/go-metrics"
38 39
)

40
var (
41 42 43 44 45
	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
	MaxBodyFetch    = 128 // Amount of block bodies to be fetched per retrieval request
	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 49 50 51 52 53 54

	hashTTL        = 5 * time.Second    // [eth/61] Time it takes for a hash request to time out
	blockSoftTTL   = 3 * time.Second    // [eth/61] Request completion threshold for increasing or decreasing a peer's bandwidth
	blockHardTTL   = 3 * blockSoftTTL   // [eth/61] Maximum time allowance before a block request is considered expired
	headerTTL      = 5 * time.Second    // [eth/62] Time it takes for a header request to time out
	bodySoftTTL    = 3 * time.Second    // [eth/62] Request completion threshold for increasing or decreasing a peer's bandwidth
	bodyHardTTL    = 3 * bodySoftTTL    // [eth/62] Maximum time allowance before a block body request is considered expired
	receiptSoftTTL = 3 * time.Second    // [eth/63] Request completion threshold for increasing or decreasing a peer's bandwidth
55 56 57
	receiptHardTTL = 3 * receiptSoftTTL // [eth/63] Maximum time allowance before a receipt request is considered expired
	stateSoftTTL   = 2 * time.Second    // [eth/63] Request completion threshold for increasing or decreasing a peer's bandwidth
	stateHardTTL   = 3 * stateSoftTTL   // [eth/63] Maximum time allowance before a node data request is considered expired
58 59 60

	maxQueuedHashes   = 256 * 1024 // [eth/61] Maximum number of hashes to queue for import (DOS protection)
	maxQueuedHeaders  = 256 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection)
61
	maxQueuedStates   = 256 * 1024 // [eth/63] Maximum number of state requests to queue (DOS protection)
62 63
	maxResultsProcess = 256        // Number of download results to import at once into the chain

64 65 66 67 68
	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
	fsPivotInterval        = 512  // Number of headers out of which to randomize the pivot point
	fsMinFullBlocks        = 1024 // Number of blocks to retrieve fully even in fast sync
69
)
70

71
var (
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
	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")
	errEmptyHashSet       = errors.New("empty hash set by peer")
	errEmptyHeaderSet     = errors.New("empty header set by peer")
	errPeersUnavailable   = errors.New("no peers available or all tried for download")
	errAlreadyInPool      = errors.New("hash already in pool")
	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")
	errCancelHashFetch    = errors.New("hash download canceled (requested)")
	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)")
91
	errCancelStateFetch   = errors.New("state data download canceled (requested)")
92
	errCancelProcessing   = errors.New("processing canceled (requested)")
93
	errNoSyncActive       = errors.New("no sync active")
94 95
)

96
type Downloader struct {
97 98 99
	mode   SyncMode       // Synchronisation mode defining the strategy used (per sync cycle)
	noFast bool           // Flag to disable fast syncing in case of a security error
	mux    *event.TypeMux // Event multiplexer to announce sync operation events
100

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

104 105
	interrupt int32 // Atomic boolean to signal termination

106
	// Statistics
107 108 109 110 111
	syncStatsChainOrigin uint64       // Origin block number where syncing started at
	syncStatsChainHeight uint64       // Highest block number known when syncing started
	syncStatsStateTotal  uint64       // Total number of node state entries known so far
	syncStatsStateDone   uint64       // Number of state trie entries already pulled
	syncStatsLock        sync.RWMutex // Lock protecting the sync stats fields
112

113
	// Callbacks
114 115 116 117 118 119 120 121 122 123 124 125
	hasHeader       headerCheckFn            // Checks if a header is present in the chain
	hasBlock        blockCheckFn             // Checks if a block is present in the chain
	getHeader       headerRetrievalFn        // Retrieves a header from the chain
	getBlock        blockRetrievalFn         // Retrieves a block from the chain
	headHeader      headHeaderRetrievalFn    // Retrieves the head header from the chain
	headBlock       headBlockRetrievalFn     // Retrieves the head block from the chain
	headFastBlock   headFastBlockRetrievalFn // Retrieves the head fast-sync block from the chain
	commitHeadBlock headBlockCommitterFn     // Commits a manually assembled block as the chain head
	getTd           tdRetrievalFn            // Retrieves the TD of a block from the chain
	insertHeaders   headerChainInsertFn      // Injects a batch of headers into the chain
	insertBlocks    blockChainInsertFn       // Injects a batch of blocks into the chain
	insertReceipts  receiptChainInsertFn     // Injects a batch of blocks and their receipts into the chain
126
	rollback        chainRollbackFn          // Removes a batch of recently added chain links
127
	dropPeer        peerDropFn               // Drops a peer for misbehaving
128

129
	// Status
130 131 132
	synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing
	synchronising   int32
	notified        int32
133 134

	// Channels
135
	newPeerCh     chan *peer
136 137 138 139 140 141 142 143 144 145
	hashCh        chan dataPack // [eth/61] Channel receiving inbound hashes
	blockCh       chan dataPack // [eth/61] Channel receiving inbound blocks
	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
	stateCh       chan dataPack // [eth/63] Channel receiving inbound node state data
	blockWakeCh   chan bool     // [eth/61] Channel to signal the block fetcher of new tasks
	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
	stateWakeCh   chan bool     // [eth/63] Channel to signal the state fetcher of new tasks
146 147 148

	cancelCh   chan struct{} // Channel to cancel mid-flight syncs
	cancelLock sync.RWMutex  // Lock to protect the cancel channel in delivers
149 150

	// Testing hooks
151 152 153 154
	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)
155 156
}

157
// New creates a new downloader to fetch hashes and blocks from remote peers.
158
func New(stateDb ethdb.Database, mux *event.TypeMux, hasHeader headerCheckFn, hasBlock blockCheckFn, getHeader headerRetrievalFn,
159 160
	getBlock blockRetrievalFn, headHeader headHeaderRetrievalFn, headBlock headBlockRetrievalFn, headFastBlock headFastBlockRetrievalFn,
	commitHeadBlock headBlockCommitterFn, getTd tdRetrievalFn, insertHeaders headerChainInsertFn, insertBlocks blockChainInsertFn,
161
	insertReceipts receiptChainInsertFn, rollback chainRollbackFn, dropPeer peerDropFn) *Downloader {
162

163
	return &Downloader{
164
		mode:            FullSync,
165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
		mux:             mux,
		queue:           newQueue(stateDb),
		peers:           newPeerSet(),
		hasHeader:       hasHeader,
		hasBlock:        hasBlock,
		getHeader:       getHeader,
		getBlock:        getBlock,
		headHeader:      headHeader,
		headBlock:       headBlock,
		headFastBlock:   headFastBlock,
		commitHeadBlock: commitHeadBlock,
		getTd:           getTd,
		insertHeaders:   insertHeaders,
		insertBlocks:    insertBlocks,
		insertReceipts:  insertReceipts,
180
		rollback:        rollback,
181 182 183 184 185 186 187 188 189 190 191 192
		dropPeer:        dropPeer,
		newPeerCh:       make(chan *peer, 1),
		hashCh:          make(chan dataPack, 1),
		blockCh:         make(chan dataPack, 1),
		headerCh:        make(chan dataPack, 1),
		bodyCh:          make(chan dataPack, 1),
		receiptCh:       make(chan dataPack, 1),
		stateCh:         make(chan dataPack, 1),
		blockWakeCh:     make(chan bool, 1),
		bodyWakeCh:      make(chan bool, 1),
		receiptWakeCh:   make(chan bool, 1),
		stateWakeCh:     make(chan bool, 1),
193 194 195
	}
}

196 197 198 199
// 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.
func (d *Downloader) Progress() (uint64, uint64, uint64) {
200 201
	d.syncStatsLock.RLock()
	defer d.syncStatsLock.RUnlock()
202

203 204 205 206 207 208 209 210 211 212
	current := uint64(0)
	switch d.mode {
	case FullSync:
		current = d.headBlock().NumberU64()
	case FastSync:
		current = d.headFastBlock().NumberU64()
	case LightSync:
		current = d.headHeader().Number.Uint64()
	}
	return d.syncStatsChainOrigin, current, d.syncStatsChainHeight
O
obscuren 已提交
213 214
}

215
// Synchronising returns whether the downloader is currently retrieving blocks.
216
func (d *Downloader) Synchronising() bool {
217
	return atomic.LoadInt32(&d.synchronising) > 0
218 219
}

220 221
// RegisterPeer injects a new download peer into the set of block source to be
// used for fetching hashes and blocks from.
222 223
func (d *Downloader) RegisterPeer(id string, version int, head common.Hash,
	getRelHashes relativeHashFetcherFn, getAbsHashes absoluteHashFetcherFn, getBlocks blockFetcherFn, // eth/61 callbacks, remove when upgrading
224 225
	getRelHeaders relativeHeaderFetcherFn, getAbsHeaders absoluteHeaderFetcherFn, getBlockBodies blockBodyFetcherFn,
	getReceipts receiptFetcherFn, getNodeData stateFetcherFn) error {
226

227
	glog.V(logger.Detail).Infoln("Registering peer", id)
228
	if err := d.peers.Register(newPeer(id, version, head, getRelHashes, getAbsHashes, getBlocks, getRelHeaders, getAbsHeaders, getBlockBodies, getReceipts, getNodeData)); err != nil {
229 230 231
		glog.V(logger.Error).Infoln("Register failed:", err)
		return err
	}
232 233 234
	return nil
}

235
// UnregisterPeer remove a peer from the known list, preventing any action from
236 237
// the specified peer. An effort is also made to return any pending fetches into
// the queue.
238 239 240 241 242 243
func (d *Downloader) UnregisterPeer(id string) error {
	glog.V(logger.Detail).Infoln("Unregistering peer", id)
	if err := d.peers.Unregister(id); err != nil {
		glog.V(logger.Error).Infoln("Unregister failed:", err)
		return err
	}
244
	d.queue.Revoke(id)
245
	return nil
246 247
}

248 249
// 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.
250
func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error {
251
	glog.V(logger.Detail).Infof("Attempting synchronisation: %v, head [%x…], TD %v", id, head[:4], td)
252

253 254
	err := d.synchronise(id, head, td, mode)
	switch err {
255 256 257 258 259 260
	case nil:
		glog.V(logger.Detail).Infof("Synchronisation completed")

	case errBusy:
		glog.V(logger.Detail).Infof("Synchronisation already in progress")

261
	case errTimeout, errBadPeer, errStallingPeer, errEmptyHashSet, errEmptyHeaderSet, errPeersUnavailable, errInvalidChain:
262 263 264 265 266 267
		glog.V(logger.Debug).Infof("Removing peer %v: %v", id, err)
		d.dropPeer(id)

	default:
		glog.V(logger.Warn).Infof("Synchronisation failed: %v", err)
	}
268
	return err
269 270 271
}

// synchronise will select the peer and use it for synchronising. If an empty string is given
272
// it will use the best peer possible and synchronize if it's TD is higher than our own. If any of the
273
// checks fail an error will be returned. This method is synchronous
274
func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error {
275 276 277 278
	// Mock out the synchonisation if testing
	if d.synchroniseMock != nil {
		return d.synchroniseMock(id, hash)
	}
279
	// Make sure only one goroutine is ever allowed past this point at once
280
	if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) {
281
		return errBusy
282
	}
283
	defer atomic.StoreInt32(&d.synchronising, 0)
284

285 286 287 288
	// Post a user notification of the sync (only once per session)
	if atomic.CompareAndSwapInt32(&d.notified, 0, 1) {
		glog.V(logger.Info).Infoln("Block synchronisation started")
	}
289
	// Reset the queue, peer set and wake channels to clean any internal leftover state
290
	d.queue.Reset()
291
	d.peers.Reset()
292

293
	for _, ch := range []chan bool{d.blockWakeCh, d.bodyWakeCh, d.receiptWakeCh, d.stateWakeCh} {
294 295 296 297
		select {
		case <-ch:
		default:
		}
298
	}
299 300 301 302 303 304
	// Reset and ephemeral sync statistics
	d.syncStatsLock.Lock()
	d.syncStatsStateTotal = 0
	d.syncStatsStateDone = 0
	d.syncStatsLock.Unlock()

305 306 307 308 309
	// Create cancel channel for aborting mid-flight
	d.cancelLock.Lock()
	d.cancelCh = make(chan struct{})
	d.cancelLock.Unlock()

310 311 312 313 314
	// Set the requested sync mode, unless it's forbidden
	d.mode = mode
	if d.mode == FastSync && d.noFast {
		d.mode = FullSync
	}
315
	// Retrieve the origin peer and initiate the downloading process
316
	p := d.peers.Peer(id)
317
	if p == nil {
318
		return errUnknownPeer
319
	}
320
	return d.syncWithPeer(p, hash, td)
321 322
}

323 324
// syncWithPeer starts a block synchronization based on the hash chain from the
// specified peer and head hash.
325
func (d *Downloader) syncWithPeer(p *peer, hash common.Hash, td *big.Int) (err error) {
O
obscuren 已提交
326
	d.mux.Post(StartEvent{})
327 328 329
	defer func() {
		// reset on error
		if err != nil {
330 331 332
			d.mux.Post(FailedEvent{err})
		} else {
			d.mux.Post(DoneEvent{})
333 334
		}
	}()
335

336
	glog.V(logger.Debug).Infof("Synchronising with the network using: %s [eth/%d]", p.id, p.version)
337 338 339
	defer func(start time.Time) {
		glog.V(logger.Debug).Infof("Synchronisation terminated after %v", time.Since(start))
	}(time.Now())
340 341

	switch {
342
	case p.version == 61:
343 344 345 346 347 348
		// Look up the sync boundaries: the common ancestor and the target block
		latest, err := d.fetchHeight61(p)
		if err != nil {
			return err
		}
		origin, err := d.findAncestor61(p)
349 350 351
		if err != nil {
			return err
		}
352
		d.syncStatsLock.Lock()
353 354
		if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin {
			d.syncStatsChainOrigin = origin
355
		}
356
		d.syncStatsChainHeight = latest
357 358
		d.syncStatsLock.Unlock()

359 360
		// Initiate the sync using a concurrent hash and block retrieval algorithm
		d.queue.Prepare(origin+1, d.mode, 0)
361 362 363
		if d.syncInitHook != nil {
			d.syncInitHook(origin, latest)
		}
364 365 366 367
		return d.spawnSync(
			func() error { return d.fetchHashes61(p, td, origin+1) },
			func() error { return d.fetchBlocks61(origin + 1) },
		)
368

369
	case p.version >= 62:
370 371
		// Look up the sync boundaries: the common ancestor and the target block
		latest, err := d.fetchHeight(p)
372 373 374
		if err != nil {
			return err
		}
375 376 377 378 379
		origin, err := d.findAncestor(p)
		if err != nil {
			return err
		}
		d.syncStatsLock.Lock()
380 381
		if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin {
			d.syncStatsChainOrigin = origin
382
		}
383
		d.syncStatsChainHeight = latest
384 385
		d.syncStatsLock.Unlock()

386
		// Initiate the sync using a concurrent header and content retrieval algorithm
387
		pivot := uint64(0)
388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408
		switch d.mode {
		case LightSync:
			pivot = latest
		case FastSync:
			// Calculate the new fast/slow sync pivot point
			pivotOffset, err := rand.Int(rand.Reader, big.NewInt(int64(fsPivotInterval)))
			if err != nil {
				panic(fmt.Sprintf("Failed to access crypto random source: %v", err))
			}
			if latest > uint64(fsMinFullBlocks)+pivotOffset.Uint64() {
				pivot = latest - uint64(fsMinFullBlocks) - pivotOffset.Uint64()
			}
			// 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
				}
			}
			glog.V(logger.Debug).Infof("Fast syncing until pivot block #%d", pivot)
409
		}
410
		d.queue.Prepare(origin+1, d.mode, pivot)
411 412 413
		if d.syncInitHook != nil {
			d.syncInitHook(origin, latest)
		}
414 415 416 417 418 419
		return d.spawnSync(
			func() error { return d.fetchHeaders(p, td, 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.fetchNodeData() },               // Node state data is retrieved during fast sync
		)
420 421 422 423 424

	default:
		// Something very wrong, stop right here
		glog.V(logger.Error).Infof("Unsupported eth protocol: %d", p.version)
		return errBadPeer
425
	}
426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455
}

// spawnSync runs d.process and all given fetcher functions to completion in
// separate goroutines, returning the first error that appears.
func (d *Downloader) spawnSync(fetchers ...func() error) error {
	var wg sync.WaitGroup
	errc := make(chan error, len(fetchers)+1)
	wg.Add(len(fetchers) + 1)
	go func() { defer wg.Done(); errc <- d.process() }()
	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
	for i := 0; i < len(fetchers)+1; i++ {
		if i == len(fetchers) {
			// 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()
	d.cancel()
	wg.Wait()
	return err
456 457
}

458
// cancel cancels all of the operations and resets the queue. It returns true
459
// if the cancel operation was completed.
460
func (d *Downloader) cancel() {
461
	// Close the current cancel channel
462
	d.cancelLock.Lock()
463 464 465 466 467 468 469
	if d.cancelCh != nil {
		select {
		case <-d.cancelCh:
			// Channel was already closed
		default:
			close(d.cancelCh)
		}
470 471
	}
	d.cancelLock.Unlock()
472 473
}

474
// Terminate interrupts the downloader, canceling all pending operations.
475
// The downloader cannot be reused after calling Terminate.
476 477 478 479 480
func (d *Downloader) Terminate() {
	atomic.StoreInt32(&d.interrupt, 1)
	d.cancel()
}

481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503
// fetchHeight61 retrieves the head block of the remote peer to aid in estimating
// the total time a pending synchronisation would take.
func (d *Downloader) fetchHeight61(p *peer) (uint64, error) {
	glog.V(logger.Debug).Infof("%v: retrieving remote chain height", p)

	// Request the advertised remote head block and wait for the response
	go p.getBlocks([]common.Hash{p.head})

	timeout := time.After(blockSoftTTL)
	for {
		select {
		case <-d.cancelCh:
			return 0, errCancelBlockFetch

		case <-d.headerCh:
			// Out of bounds eth/62 block headers received, ignore them

		case <-d.bodyCh:
			// Out of bounds eth/62 block bodies received, ignore them

		case <-d.hashCh:
			// Out of bounds hashes received, ignore them

504
		case packet := <-d.blockCh:
505
			// Discard anything not from the origin peer
506 507
			if packet.PeerId() != p.id {
				glog.V(logger.Debug).Infof("Received blocks from incorrect peer(%s)", packet.PeerId())
508 509 510
				break
			}
			// Make sure the peer actually gave something valid
511
			blocks := packet.(*blockPack).blocks
512 513 514 515 516 517 518 519 520 521 522 523 524
			if len(blocks) != 1 {
				glog.V(logger.Debug).Infof("%v: invalid number of head blocks: %d != 1", p, len(blocks))
				return 0, errBadPeer
			}
			return blocks[0].NumberU64(), nil

		case <-timeout:
			glog.V(logger.Debug).Infof("%v: head block timeout", p)
			return 0, errTimeout
		}
	}
}

525
// findAncestor61 tries to locate the common ancestor block of the local chain and
526 527
// a remote peers blockchain. In the general case when our node was in sync and
// on the correct chain, checking the top N blocks should already get us a match.
528 529 530
// In the rare scenario when we ended up on a long reorganization (i.e. none of
// the head blocks match), we do a binary search to find the common ancestor.
func (d *Downloader) findAncestor61(p *peer) (uint64, error) {
531 532 533 534
	glog.V(logger.Debug).Infof("%v: looking for common ancestor", p)

	// Request out head blocks to short circuit ancestor location
	head := d.headBlock().NumberU64()
535
	from := int64(head) - int64(MaxHashFetch) + 1
536 537 538 539 540 541 542 543 544 545 546 547 548 549
	if from < 0 {
		from = 0
	}
	go p.getAbsHashes(uint64(from), MaxHashFetch)

	// Wait for the remote response to the head fetch
	number, hash := uint64(0), common.Hash{}
	timeout := time.After(hashTTL)

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

550
		case packet := <-d.hashCh:
551
			// Discard anything not from the origin peer
552 553
			if packet.PeerId() != p.id {
				glog.V(logger.Debug).Infof("Received hashes from incorrect peer(%s)", packet.PeerId())
554 555 556
				break
			}
			// Make sure the peer actually gave something valid
557
			hashes := packet.(*hashPack).hashes
558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573
			if len(hashes) == 0 {
				glog.V(logger.Debug).Infof("%v: empty head hash set", p)
				return 0, errEmptyHashSet
			}
			// Check if a common ancestor was found
			finished = true
			for i := len(hashes) - 1; i >= 0; i-- {
				if d.hasBlock(hashes[i]) {
					number, hash = uint64(from)+uint64(i), hashes[i]
					break
				}
			}

		case <-d.blockCh:
			// Out of bounds blocks received, ignore them

574 575 576 577 578 579
		case <-d.headerCh:
			// Out of bounds eth/62 block headers received, ignore them

		case <-d.bodyCh:
			// Out of bounds eth/62 block bodies received, ignore them

580 581 582 583 584 585 586
		case <-timeout:
			glog.V(logger.Debug).Infof("%v: head hash timeout", p)
			return 0, errTimeout
		}
	}
	// If the head fetch already found an ancestor, return
	if !common.EmptyHash(hash) {
587
		glog.V(logger.Debug).Infof("%v: common ancestor: #%d [%x…]", p, number, hash[:4])
588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
		return number, nil
	}
	// Ancestor not found, we need to binary search over our chain
	start, end := uint64(0), head
	for start+1 < end {
		// Split our chain interval in two, and request the hash to cross check
		check := (start + end) / 2

		timeout := time.After(hashTTL)
		go p.getAbsHashes(uint64(check), 1)

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

605
			case packet := <-d.hashCh:
606
				// Discard anything not from the origin peer
607 608
				if packet.PeerId() != p.id {
					glog.V(logger.Debug).Infof("Received hashes from incorrect peer(%s)", packet.PeerId())
609 610 611
					break
				}
				// Make sure the peer actually gave something valid
612
				hashes := packet.(*hashPack).hashes
613 614 615 616 617 618 619 620 621 622 623 624 625
				if len(hashes) != 1 {
					glog.V(logger.Debug).Infof("%v: invalid search hash set (%d)", p, len(hashes))
					return 0, errBadPeer
				}
				arrived = true

				// Modify the search interval based on the response
				block := d.getBlock(hashes[0])
				if block == nil {
					end = check
					break
				}
				if block.NumberU64() != check {
626
					glog.V(logger.Debug).Infof("%v: non requested hash #%d [%x…], instead of #%d", p, block.NumberU64(), block.Hash().Bytes()[:4], check)
627 628 629 630 631 632 633
					return 0, errBadPeer
				}
				start = check

			case <-d.blockCh:
				// Out of bounds blocks received, ignore them

634 635 636 637 638 639
			case <-d.headerCh:
				// Out of bounds eth/62 block headers received, ignore them

			case <-d.bodyCh:
				// Out of bounds eth/62 block bodies received, ignore them

640 641 642 643 644 645 646 647 648
			case <-timeout:
				glog.V(logger.Debug).Infof("%v: search hash timeout", p)
				return 0, errTimeout
			}
		}
	}
	return start, nil
}

649
// fetchHashes61 keeps retrieving hashes from the requested number, until no more
650
// are returned, potentially throttling on the way.
651
func (d *Downloader) fetchHashes61(p *peer, td *big.Int, from uint64) error {
652 653 654
	glog.V(logger.Debug).Infof("%v: downloading hashes from #%d", p, from)

	// Create a timeout timer, and the associated hash fetcher
655
	request := time.Now()       // time of the last fetch request
656 657 658 659 660
	timeout := time.NewTimer(0) // timer to dump a non-responsive active peer
	<-timeout.C                 // timeout channel should be initially empty
	defer timeout.Stop()

	getHashes := func(from uint64) {
661 662
		glog.V(logger.Detail).Infof("%v: fetching %d hashes from #%d", p, MaxHashFetch, from)

663
		go p.getAbsHashes(from, MaxHashFetch)
664
		request = time.Now()
665 666 667 668
		timeout.Reset(hashTTL)
	}
	// Start pulling hashes, until all are exhausted
	getHashes(from)
669 670
	gotHashes := false

671 672 673 674 675
	for {
		select {
		case <-d.cancelCh:
			return errCancelHashFetch

676 677 678 679 680 681
		case <-d.headerCh:
			// Out of bounds eth/62 block headers received, ignore them

		case <-d.bodyCh:
			// Out of bounds eth/62 block bodies received, ignore them

682
		case packet := <-d.hashCh:
683
			// Make sure the active peer is giving us the hashes
684 685
			if packet.PeerId() != p.id {
				glog.V(logger.Debug).Infof("Received hashes from incorrect peer(%s)", packet.PeerId())
686 687
				break
			}
688
			hashReqTimer.UpdateSince(request)
689 690 691
			timeout.Stop()

			// If no more hashes are inbound, notify the block fetcher and return
692
			if packet.Items() == 0 {
693 694 695
				glog.V(logger.Debug).Infof("%v: no available hashes", p)

				select {
696
				case d.blockWakeCh <- false:
697 698
				case <-d.cancelCh:
				}
699 700 701 702 703 704 705 706 707 708 709 710
				// If no hashes 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 hashes up from 11 (R's TD was higher, it must have something)
				// R: Nothing to give
711
				if !gotHashes && td.Cmp(d.getTd(d.headBlock().Hash())) > 0 {
712 713
					return errStallingPeer
				}
714 715
				return nil
			}
716
			gotHashes = true
717
			hashes := packet.(*hashPack).hashes
718

719
			// Otherwise insert all the new hashes, aborting in case of junk
720
			glog.V(logger.Detail).Infof("%v: scheduling %d hashes from #%d", p, len(hashes), from)
721

722 723
			inserts := d.queue.Schedule61(hashes, true)
			if len(inserts) != len(hashes) {
724 725 726
				glog.V(logger.Debug).Infof("%v: stale hashes", p)
				return errBadPeer
			}
727
			// Notify the block fetcher of new hashes, but stop if queue is full
728
			if d.queue.PendingBlocks() < maxQueuedHashes {
729 730
				// We still have hashes to fetch, send continuation wake signal (potential)
				select {
731
				case d.blockWakeCh <- true:
732 733 734 735 736
				default:
				}
			} else {
				// Hash limit reached, send a termination wake signal (enforced)
				select {
737
				case d.blockWakeCh <- false:
738 739
				case <-d.cancelCh:
				}
740 741 742
				return nil
			}
			// Queue not yet full, fetch the next batch
743
			from += uint64(len(hashes))
744 745 746 747
			getHashes(from)

		case <-timeout.C:
			glog.V(logger.Debug).Infof("%v: hash request timed out", p)
748
			hashTimeoutMeter.Mark(1)
749 750 751 752 753
			return errTimeout
		}
	}
}

754
// fetchBlocks61 iteratively downloads the scheduled hashes, taking any available
755 756
// peers, reserving a chunk of blocks for each, waiting for delivery and also
// periodically checking for timeouts.
757
func (d *Downloader) fetchBlocks61(from uint64) error {
758 759 760 761 762 763 764 765 766
	glog.V(logger.Debug).Infof("Downloading blocks from #%d", from)
	defer glog.V(logger.Debug).Infof("Block download terminated")

	// Create a timeout timer for scheduling expiration tasks
	ticker := time.NewTicker(100 * time.Millisecond)
	defer ticker.Stop()

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

767
	// Fetch blocks until the hash fetcher's done
768 769 770 771 772 773
	finished := false
	for {
		select {
		case <-d.cancelCh:
			return errCancelBlockFetch

774 775 776 777 778 779
		case <-d.headerCh:
			// Out of bounds eth/62 block headers received, ignore them

		case <-d.bodyCh:
			// Out of bounds eth/62 block bodies received, ignore them

780
		case packet := <-d.blockCh:
781 782
			// If the peer was previously banned and failed to deliver it's pack
			// in a reasonable time frame, ignore it's message.
783
			if peer := d.peers.Peer(packet.PeerId()); peer != nil {
784
				// Deliver the received chunk of blocks, and demote in case of errors
785 786
				blocks := packet.(*blockPack).blocks
				err := d.queue.DeliverBlocks(peer.id, blocks)
787 788 789
				switch err {
				case nil:
					// If no blocks were delivered, demote the peer (need the delivery above)
790
					if len(blocks) == 0 {
791
						peer.Demote()
792
						peer.SetBlocksIdle()
793 794 795 796 797
						glog.V(logger.Detail).Infof("%s: no blocks delivered", peer)
						break
					}
					// All was successful, promote the peer and potentially start processing
					peer.Promote()
798
					peer.SetBlocksIdle()
799
					glog.V(logger.Detail).Infof("%s: delivered %d blocks", peer, len(blocks))
800 801 802 803 804 805 806 807 808

				case errInvalidChain:
					// The hash chain is invalid (blocks are not ordered properly), abort
					return err

				case errNoFetchesPending:
					// Peer probably timed out with its delivery but came through
					// in the end, demote, but allow to to pull from this peer.
					peer.Demote()
809
					peer.SetBlocksIdle()
810 811 812 813 814 815 816 817 818 819 820 821 822
					glog.V(logger.Detail).Infof("%s: out of bound delivery", peer)

				case errStaleDelivery:
					// Delivered something completely else than requested, usually
					// caused by a timeout and delivery during a new sync cycle.
					// Don't set it to idle as the original request should still be
					// in flight.
					peer.Demote()
					glog.V(logger.Detail).Infof("%s: stale delivery", peer)

				default:
					// Peer did something semi-useful, demote but keep it around
					peer.Demote()
823
					peer.SetBlocksIdle()
824 825 826 827 828 829 830 831 832
					glog.V(logger.Detail).Infof("%s: delivery partially failed: %v", peer, err)
				}
			}
			// Blocks arrived, try to update the progress
			select {
			case update <- struct{}{}:
			default:
			}

833
		case cont := <-d.blockWakeCh:
834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856
			// The hash fetcher sent a continuation flag, check if it's done
			if !cont {
				finished = true
			}
			// Hashes 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
			}
			// Check for block request timeouts and demote the responsible peers
857
			for _, pid := range d.queue.ExpireBlocks(blockHardTTL) {
858 859 860 861 862
				if peer := d.peers.Peer(pid); peer != nil {
					peer.Demote()
					glog.V(logger.Detail).Infof("%s: block delivery timeout", peer)
				}
			}
863 864
			// If there's nothing more to fetch, wait or terminate
			if d.queue.PendingBlocks() == 0 {
865
				if !d.queue.InFlightBlocks() && finished {
866 867 868 869 870 871
					glog.V(logger.Debug).Infof("Block fetching completed")
					return nil
				}
				break
			}
			// Send a download request to all idle peers, until throttled
872
			throttled := false
873
			idles, total := d.peers.BlockIdlePeers()
874 875

			for _, peer := range idles {
876
				// Short circuit if throttling activated
877
				if d.queue.ShouldThrottleBlocks() {
878
					throttled = true
879 880 881 882 883
					break
				}
				// Reserve a chunk of hashes for a peer. A nil can mean either that
				// no more hashes are available, or that the peer is known not to
				// have them.
884
				request := d.queue.ReserveBlocks(peer, peer.BlockCapacity())
885 886 887 888 889 890 891
				if request == nil {
					continue
				}
				if glog.V(logger.Detail) {
					glog.Infof("%s: requesting %d blocks", peer, len(request.Hashes))
				}
				// Fetch the chunk and make sure any errors return the hashes to the queue
892
				if err := peer.Fetch61(request); err != nil {
893 894 895 896 897 898 899
					// 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.
					panic(fmt.Sprintf("%v: fetch assignment failed, hard panic", peer))
					d.queue.CancelBlocks(request) // noop for now
900 901 902 903
				}
			}
			// Make sure that we have peers available for fetching. If all peers have been tried
			// and all failed throw an error
904
			if !throttled && !d.queue.InFlightBlocks() && len(idles) == total {
905 906 907 908 909 910
				return errPeersUnavailable
			}
		}
	}
}

911 912 913 914 915 916 917 918 919 920 921 922 923 924
// fetchHeight retrieves the head header of the remote peer to aid in estimating
// the total time a pending synchronisation would take.
func (d *Downloader) fetchHeight(p *peer) (uint64, error) {
	glog.V(logger.Debug).Infof("%v: retrieving remote chain height", p)

	// Request the advertised remote head block and wait for the response
	go p.getRelHeaders(p.head, 1, 0, false)

	timeout := time.After(headerTTL)
	for {
		select {
		case <-d.cancelCh:
			return 0, errCancelBlockFetch

925
		case packet := <-d.headerCh:
926
			// Discard anything not from the origin peer
927 928
			if packet.PeerId() != p.id {
				glog.V(logger.Debug).Infof("Received headers from incorrect peer(%s)", packet.PeerId())
929 930 931
				break
			}
			// Make sure the peer actually gave something valid
932
			headers := packet.(*headerPack).headers
933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954
			if len(headers) != 1 {
				glog.V(logger.Debug).Infof("%v: invalid number of head headers: %d != 1", p, len(headers))
				return 0, errBadPeer
			}
			return headers[0].Number.Uint64(), nil

		case <-d.bodyCh:
			// Out of bounds block bodies received, ignore them

		case <-d.hashCh:
			// Out of bounds eth/61 hashes received, ignore them

		case <-d.blockCh:
			// Out of bounds eth/61 blocks received, ignore them

		case <-timeout:
			glog.V(logger.Debug).Infof("%v: head header timeout", p)
			return 0, errTimeout
		}
	}
}

955
// findAncestor tries to locate the common ancestor link of the local chain and
956
// a remote peers blockchain. In the general case when our node was in sync and
957
// on the correct chain, checking the top N links should already get us a match.
958
// In the rare scenario when we ended up on a long reorganization (i.e. none of
959
// the head links match), we do a binary search to find the common ancestor.
960 961 962
func (d *Downloader) findAncestor(p *peer) (uint64, error) {
	glog.V(logger.Debug).Infof("%v: looking for common ancestor", p)

963 964 965 966
	// Request our head headers to short circuit ancestor location
	head := d.headHeader().Number.Uint64()
	if d.mode == FullSync {
		head = d.headBlock().NumberU64()
967 968
	} else if d.mode == FastSync {
		head = d.headFastBlock().NumberU64()
969
	}
970 971 972 973 974 975 976 977 978 979 980 981 982 983 984
	from := int64(head) - int64(MaxHeaderFetch) + 1
	if from < 0 {
		from = 0
	}
	go p.getAbsHeaders(uint64(from), MaxHeaderFetch, 0, false)

	// Wait for the remote response to the head fetch
	number, hash := uint64(0), common.Hash{}
	timeout := time.After(hashTTL)

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

985
		case packet := <-d.headerCh:
986
			// Discard anything not from the origin peer
987 988
			if packet.PeerId() != p.id {
				glog.V(logger.Debug).Infof("Received headers from incorrect peer(%s)", packet.PeerId())
989 990 991
				break
			}
			// Make sure the peer actually gave something valid
992
			headers := packet.(*headerPack).headers
993 994 995 996 997 998 999
			if len(headers) == 0 {
				glog.V(logger.Debug).Infof("%v: empty head header set", p)
				return 0, errEmptyHeaderSet
			}
			// Check if a common ancestor was found
			finished = true
			for i := len(headers) - 1; i >= 0; i-- {
1000
				if (d.mode != LightSync && d.hasBlock(headers[i].Hash())) || (d.mode == LightSync && d.hasHeader(headers[i].Hash())) {
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039
					number, hash = headers[i].Number.Uint64(), headers[i].Hash()
					break
				}
			}

		case <-d.bodyCh:
			// Out of bounds block bodies received, ignore them

		case <-d.hashCh:
			// Out of bounds eth/61 hashes received, ignore them

		case <-d.blockCh:
			// Out of bounds eth/61 blocks received, ignore them

		case <-timeout:
			glog.V(logger.Debug).Infof("%v: head header timeout", p)
			return 0, errTimeout
		}
	}
	// If the head fetch already found an ancestor, return
	if !common.EmptyHash(hash) {
		glog.V(logger.Debug).Infof("%v: common ancestor: #%d [%x…]", p, number, hash[:4])
		return number, nil
	}
	// Ancestor not found, we need to binary search over our chain
	start, end := uint64(0), head
	for start+1 < end {
		// Split our chain interval in two, and request the hash to cross check
		check := (start + end) / 2

		timeout := time.After(hashTTL)
		go p.getAbsHeaders(uint64(check), 1, 0, false)

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

1040
			case packer := <-d.headerCh:
1041
				// Discard anything not from the origin peer
1042 1043
				if packer.PeerId() != p.id {
					glog.V(logger.Debug).Infof("Received headers from incorrect peer(%s)", packer.PeerId())
1044 1045 1046
					break
				}
				// Make sure the peer actually gave something valid
1047
				headers := packer.(*headerPack).headers
1048 1049 1050 1051 1052 1053 1054
				if len(headers) != 1 {
					glog.V(logger.Debug).Infof("%v: invalid search header set (%d)", p, len(headers))
					return 0, errBadPeer
				}
				arrived = true

				// Modify the search interval based on the response
1055
				if (d.mode == FullSync && !d.hasBlock(headers[0].Hash())) || (d.mode != FullSync && !d.hasHeader(headers[0].Hash())) {
1056 1057 1058
					end = check
					break
				}
1059 1060 1061
				header := d.getHeader(headers[0].Hash()) // Independent of sync mode, header surely exists
				if header.Number.Uint64() != check {
					glog.V(logger.Debug).Infof("%v: non requested header #%d [%x…], instead of #%d", p, header.Number, header.Hash().Bytes()[:4], check)
1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085
					return 0, errBadPeer
				}
				start = check

			case <-d.bodyCh:
				// Out of bounds block bodies received, ignore them

			case <-d.hashCh:
				// Out of bounds eth/61 hashes received, ignore them

			case <-d.blockCh:
				// Out of bounds eth/61 blocks received, ignore them

			case <-timeout:
				glog.V(logger.Debug).Infof("%v: search header timeout", p)
				return 0, errTimeout
			}
		}
	}
	return start, nil
}

// fetchHeaders keeps retrieving headers from the requested number, until no more
// are returned, potentially throttling on the way.
1086 1087 1088
//
// The queue parameter can be used to switch between queuing headers for block
// body download too, or directly import as pure header chains.
1089
func (d *Downloader) fetchHeaders(p *peer, td *big.Int, from uint64) error {
1090 1091 1092
	glog.V(logger.Debug).Infof("%v: downloading headers from #%d", p, from)
	defer glog.V(logger.Debug).Infof("%v: header download terminated", p)

1093 1094 1095
	// Calculate the pivoting point for switching from fast to slow sync
	pivot := d.queue.FastSyncPivot()

1096 1097 1098 1099
	// Keep a count of uncertain headers to roll back
	rollback := []*types.Header{}
	defer func() {
		if len(rollback) > 0 {
1100
			// Flatten the headers and roll them back
1101 1102 1103 1104
			hashes := make([]common.Hash, len(rollback))
			for i, header := range rollback {
				hashes[i] = header.Hash()
			}
1105
			lh, lfb, lb := d.headHeader().Number, d.headFastBlock().Number(), d.headBlock().Number()
1106
			d.rollback(hashes)
1107 1108 1109 1110 1111 1112 1113
			glog.V(logger.Warn).Infof("Rolled back %d headers (LH: %d->%d, FB: %d->%d, LB: %d->%d)",
				len(hashes), lh, d.headHeader().Number, lfb, d.headFastBlock().Number(), lb, d.headBlock().Number())

			// If we're already past the pivot point, this could be an attack, disable fast sync
			if rollback[len(rollback)-1].Number.Uint64() > pivot {
				d.noFast = true
			}
1114 1115
		}
	}()
1116

1117
	// Create a timeout timer, and the associated hash fetcher
1118
	request := time.Now()       // time of the last fetch request
1119 1120 1121 1122 1123 1124 1125 1126
	timeout := time.NewTimer(0) // timer to dump a non-responsive active peer
	<-timeout.C                 // timeout channel should be initially empty
	defer timeout.Stop()

	getHeaders := func(from uint64) {
		glog.V(logger.Detail).Infof("%v: fetching %d headers from #%d", p, MaxHeaderFetch, from)

		go p.getAbsHeaders(from, MaxHeaderFetch, 0, false)
1127
		request = time.Now()
1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144
		timeout.Reset(headerTTL)
	}
	// Start pulling headers, until all are exhausted
	getHeaders(from)
	gotHeaders := false

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

		case <-d.hashCh:
			// Out of bounds eth/61 hashes received, ignore them

		case <-d.blockCh:
			// Out of bounds eth/61 blocks received, ignore them

1145
		case packet := <-d.headerCh:
1146
			// Make sure the active peer is giving us the headers
1147 1148
			if packet.PeerId() != p.id {
				glog.V(logger.Debug).Infof("Received headers from incorrect peer (%s)", packet.PeerId())
1149 1150
				break
			}
1151
			headerReqTimer.UpdateSince(request)
1152 1153
			timeout.Stop()

1154
			// If no more headers are inbound, notify the content fetchers and return
1155
			if packet.Items() == 0 {
1156 1157
				glog.V(logger.Debug).Infof("%v: no available headers", p)

1158
				for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh, d.stateWakeCh} {
1159 1160 1161 1162
					select {
					case ch <- false:
					case <-d.cancelCh:
					}
1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175
				}
				// 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
1176
				if !gotHeaders && td.Cmp(d.getTd(d.headBlock().Hash())) > 0 {
1177 1178
					return errStallingPeer
				}
1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191
				// 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 {
					if td.Cmp(d.getTd(d.headHeader().Hash())) > 0 {
						return errStallingPeer
					}
				}
				rollback = nil
1192 1193 1194
				return nil
			}
			gotHeaders = true
1195
			headers := packet.(*headerPack).headers
1196 1197

			// Otherwise insert all the new headers, aborting in case of junk
1198
			glog.V(logger.Detail).Infof("%v: schedule %d headers from #%d", p, len(headers), from)
1199

1200
			if d.mode == FastSync || d.mode == LightSync {
1201 1202 1203 1204 1205 1206 1207 1208
				// Collect the yet unknown headers to mark them as uncertain
				unknown := make([]*types.Header, 0, len(headers))
				for _, header := range headers {
					if !d.hasHeader(header.Hash()) {
						unknown = append(unknown, header)
					}
				}
				// If we're importing pure headers, verify based on their recentness
1209 1210
				frequency := fsHeaderCheckFrequency
				if headers[len(headers)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot {
1211 1212 1213
					frequency = 1
				}
				if n, err := d.insertHeaders(headers, frequency); err != nil {
1214
					glog.V(logger.Debug).Infof("%v: invalid header #%d [%x…]: %v", p, headers[n].Number, headers[n].Hash().Bytes()[:4], err)
1215 1216
					return errInvalidChain
				}
1217 1218
				// All verifications passed, store newly found uncertain headers
				rollback = append(rollback, unknown...)
1219 1220
				if len(rollback) > fsHeaderSafetyNet {
					rollback = append(rollback[:0], rollback[len(rollback)-fsHeaderSafetyNet:]...)
1221
				}
1222
			}
1223
			if d.mode == FullSync || d.mode == FastSync {
1224 1225
				inserts := d.queue.Schedule(headers, from)
				if len(inserts) != len(headers) {
1226 1227
					glog.V(logger.Debug).Infof("%v: stale headers", p)
					return errBadPeer
1228
				}
1229 1230 1231
			}
			// Notify the content fetchers of new headers, but stop if queue is full
			cont := d.queue.PendingBlocks() < maxQueuedHeaders || d.queue.PendingReceipts() < maxQueuedHeaders
1232
			for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh, d.stateWakeCh} {
1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245
				if cont {
					// We still have headers to fetch, send continuation wake signal (potential)
					select {
					case ch <- true:
					default:
					}
				} else {
					// Header limit reached, send a termination wake signal (enforced)
					select {
					case ch <- false:
					case <-d.cancelCh:
					}
					return nil
1246
				}
1247 1248
			}
			// Queue not yet full, fetch the next batch
1249
			from += uint64(len(headers))
1250 1251 1252 1253 1254
			getHeaders(from)

		case <-timeout.C:
			// Header retrieval timed out, consider the peer bad and drop
			glog.V(logger.Debug).Infof("%v: header request timed out", p)
1255
			headerTimeoutMeter.Mark(1)
1256 1257 1258
			d.dropPeer(p.id)

			// Finish the sync gracefully instead of dumping the gathered data though
1259
			for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh, d.stateWakeCh} {
1260 1261 1262 1263
				select {
				case ch <- false:
				case <-d.cancelCh:
				}
1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275
			}
			return nil
		}
	}
}

// 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 {
	glog.V(logger.Debug).Infof("Downloading block bodies from #%d", from)

1276
	var (
1277
		deliver = func(packet dataPack) error {
1278
			pack := packet.(*bodyPack)
1279
			return d.queue.DeliverBodies(pack.peerId, pack.transactions, pack.uncles)
1280
		}
1281
		expire   = func() []string { return d.queue.ExpireBodies(bodyHardTTL) }
1282 1283
		fetch    = func(p *peer, req *fetchRequest) error { return p.FetchBodies(req) }
		capacity = func(p *peer) int { return p.BlockCapacity() }
1284
		setIdle  = func(p *peer) { p.SetBodiesIdle() }
1285
	)
1286
	err := d.fetchParts(errCancelBodyFetch, d.bodyCh, deliver, d.bodyWakeCh, expire,
1287 1288
		d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ShouldThrottleBlocks, d.queue.ReserveBodies,
		d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "Body")
1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300

	glog.V(logger.Debug).Infof("Block body download terminated: %v", err)
	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 {
	glog.V(logger.Debug).Infof("Downloading receipts from #%d", from)

	var (
1301
		deliver = func(packet dataPack) error {
1302 1303 1304
			pack := packet.(*receiptPack)
			return d.queue.DeliverReceipts(pack.peerId, pack.receipts)
		}
1305
		expire   = func() []string { return d.queue.ExpireReceipts(receiptHardTTL) }
1306 1307 1308 1309
		fetch    = func(p *peer, req *fetchRequest) error { return p.FetchReceipts(req) }
		capacity = func(p *peer) int { return p.ReceiptCapacity() }
		setIdle  = func(p *peer) { p.SetReceiptsIdle() }
	)
1310
	err := d.fetchParts(errCancelReceiptFetch, d.receiptCh, deliver, d.receiptWakeCh, expire,
1311
		d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ShouldThrottleReceipts, d.queue.ReserveReceipts,
1312
		d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "Receipt")
1313 1314 1315 1316 1317

	glog.V(logger.Debug).Infof("Receipt download terminated: %v", err)
	return err
}

1318 1319 1320 1321 1322 1323 1324 1325 1326
// fetchNodeData iteratively downloads the scheduled state trie nodes, taking any
// available peers, reserving a chunk of nodes for each, waiting for delivery and
// also periodically checking for timeouts.
func (d *Downloader) fetchNodeData() error {
	glog.V(logger.Debug).Infof("Downloading node state data")

	var (
		deliver = func(packet dataPack) error {
			start := time.Now()
1327 1328 1329 1330 1331 1332 1333
			return d.queue.DeliverNodeData(packet.PeerId(), packet.(*statePack).states, func(err error, delivered int) {
				if err != nil {
					// If the node data processing failed, the root hash is very wrong, abort
					glog.V(logger.Error).Infof("peer %d: state processing failed: %v", packet.PeerId(), err)
					d.cancel()
					return
				}
1334 1335
				// Processing succeeded, notify state fetcher of continuation
				if d.queue.PendingNodeData() > 0 {
1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346
					select {
					case d.stateWakeCh <- true:
					default:
					}
				}
				// Log a message to the user and return
				d.syncStatsLock.Lock()
				defer d.syncStatsLock.Unlock()
				d.syncStatsStateDone += uint64(delivered)
				glog.V(logger.Info).Infof("imported %d state entries in %v: processed %d in total", delivered, time.Since(start), d.syncStatsStateDone)
			})
1347 1348 1349 1350 1351 1352 1353 1354 1355 1356
		}
		expire   = func() []string { return d.queue.ExpireNodeData(stateHardTTL) }
		throttle = func() bool { return false }
		reserve  = func(p *peer, count int) (*fetchRequest, bool, error) {
			return d.queue.ReserveNodeData(p, count), false, nil
		}
		fetch    = func(p *peer, req *fetchRequest) error { return p.FetchNodeData(req) }
		capacity = func(p *peer) int { return p.NodeDataCapacity() }
		setIdle  = func(p *peer) { p.SetNodeDataIdle() }
	)
1357
	err := d.fetchParts(errCancelStateFetch, d.stateCh, deliver, d.stateWakeCh, expire,
1358
		d.queue.PendingNodeData, d.queue.InFlightNodeData, throttle, reserve, nil, fetch,
1359
		d.queue.CancelNodeData, capacity, d.peers.NodeDataIdlePeers, setIdle, "State")
1360 1361 1362 1363 1364

	glog.V(logger.Debug).Infof("Node state data download terminated: %v", err)
	return err
}

1365 1366 1367
// 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.
1368
func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliver func(packet dataPack) error, wakeCh chan bool,
1369 1370 1371
	expire func() []string, pending func() int, inFlight func() bool, throttle func() bool, reserve func(*peer, int) (*fetchRequest, bool, error),
	fetchHook func([]*types.Header), fetch func(*peer, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peer) int,
	idle func() ([]*peer, int), setIdle func(*peer), kind string) error {
1372

1373
	// Create a ticker to detect expired retrieval tasks
1374 1375 1376 1377 1378
	ticker := time.NewTicker(100 * time.Millisecond)
	defer ticker.Stop()

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

1379
	// Prepare the queue and fetch block parts until the block header fetcher's done
1380 1381 1382 1383
	finished := false
	for {
		select {
		case <-d.cancelCh:
1384
			return errCancel
1385 1386 1387 1388 1389 1390 1391

		case <-d.hashCh:
			// Out of bounds eth/61 hashes received, ignore them

		case <-d.blockCh:
			// Out of bounds eth/61 blocks received, ignore them

1392
		case packet := <-deliveryCh:
1393 1394
			// If the peer was previously banned and failed to deliver it's pack
			// in a reasonable time frame, ignore it's message.
1395 1396 1397
			if peer := d.peers.Peer(packet.PeerId()); peer != nil {
				// Deliver the received chunk of data, and demote in case of errors
				switch err := deliver(packet); err {
1398
				case nil:
1399
					// If no blocks were delivered, demote the peer (need the delivery above to clean internal queue!)
1400
					if packet.Items() == 0 {
1401
						peer.Demote()
1402 1403
						setIdle(peer)
						glog.V(logger.Detail).Infof("%s: no %s delivered", peer, strings.ToLower(kind))
1404 1405 1406 1407
						break
					}
					// All was successful, promote the peer and potentially start processing
					peer.Promote()
1408 1409
					setIdle(peer)
					glog.V(logger.Detail).Infof("%s: delivered %s %s(s)", peer, packet.Stats(), strings.ToLower(kind))
1410 1411 1412 1413 1414 1415 1416 1417 1418

				case errInvalidChain:
					// The hash chain is invalid (blocks are not ordered properly), abort
					return err

				case errNoFetchesPending:
					// Peer probably timed out with its delivery but came through
					// in the end, demote, but allow to to pull from this peer.
					peer.Demote()
1419 1420
					setIdle(peer)
					glog.V(logger.Detail).Infof("%s: out of bound %s delivery", peer, strings.ToLower(kind))
1421 1422 1423 1424 1425 1426 1427

				case errStaleDelivery:
					// Delivered something completely else than requested, usually
					// caused by a timeout and delivery during a new sync cycle.
					// Don't set it to idle as the original request should still be
					// in flight.
					peer.Demote()
1428
					glog.V(logger.Detail).Infof("%s: %s stale delivery", peer, strings.ToLower(kind))
1429 1430 1431 1432

				default:
					// Peer did something semi-useful, demote but keep it around
					peer.Demote()
1433 1434
					setIdle(peer)
					glog.V(logger.Detail).Infof("%s: %s delivery partially failed: %v", peer, strings.ToLower(kind), err)
1435 1436 1437 1438 1439 1440 1441 1442
				}
			}
			// Blocks assembled, try to update the progress
			select {
			case update <- struct{}{}:
			default:
			}

1443
		case cont := <-wakeCh:
1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465
			// 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
			}
1466 1467
			// Check for fetch request timeouts and demote the responsible peers
			for _, pid := range expire() {
1468 1469
				if peer := d.peers.Peer(pid); peer != nil {
					peer.Demote()
1470
					setIdle(peer)
1471
					glog.V(logger.Detail).Infof("%s: %s delivery timeout", peer, strings.ToLower(kind))
1472 1473
				}
			}
1474 1475
			// If there's nothing more to fetch, wait or terminate
			if pending() == 0 {
1476
				if !inFlight() && finished {
1477
					glog.V(logger.Debug).Infof("%s fetching completed", kind)
1478 1479 1480 1481 1482
					return nil
				}
				break
			}
			// Send a download request to all idle peers, until throttled
1483
			progressed, throttled, running := false, false, inFlight()
1484 1485 1486
			idles, total := idle()

			for _, peer := range idles {
1487
				// Short circuit if throttling activated
1488
				if throttle() {
1489 1490 1491
					throttled = true
					break
				}
1492 1493
				// 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
1494
				// have them.
1495
				request, progress, err := reserve(peer, capacity(peer))
1496 1497 1498
				if err != nil {
					return err
				}
1499 1500
				if progress {
					progressed = true
1501 1502 1503 1504 1505
				}
				if request == nil {
					continue
				}
				if glog.V(logger.Detail) {
1506 1507 1508 1509 1510
					if len(request.Headers) > 0 {
						glog.Infof("%s: requesting %d %s(s), first at #%d", peer, len(request.Headers), strings.ToLower(kind), request.Headers[0].Number)
					} else {
						glog.Infof("%s: requesting %d %s(s)", peer, len(request.Hashes), strings.ToLower(kind))
					}
1511 1512
				}
				// Fetch the chunk and make sure any errors return the hashes to the queue
1513 1514
				if fetchHook != nil {
					fetchHook(request.Headers)
1515
				}
1516
				if err := fetch(peer, request); err != nil {
1517 1518 1519 1520 1521 1522 1523
					// 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.
					panic(fmt.Sprintf("%v: %s fetch assignment failed, hard panic", peer, strings.ToLower(kind)))
					cancel(request) // noop for now
1524
				}
1525
				running = true
1526 1527 1528
			}
			// Make sure that we have peers available for fetching. If all peers have been tried
			// and all failed throw an error
1529
			if !progressed && !throttled && !running && len(idles) == total && pending() > 0 {
1530 1531 1532 1533 1534 1535
				return errPeersUnavailable
			}
		}
	}
}

1536
// process takes fetch results from the queue and tries to import them into the
1537 1538 1539
// chain. The type of import operation will depend on the result contents.
func (d *Downloader) process() error {
	pivot := d.queue.FastSyncPivot()
1540
	for {
1541
		results := d.queue.WaitResults()
1542
		if len(results) == 0 {
1543
			return nil // queue empty
1544
		}
1545
		if d.chainInsertHook != nil {
1546
			d.chainInsertHook(results)
1547
		}
1548
		// Actually import the blocks
1549 1550
		if glog.V(logger.Debug) {
			first, last := results[0].Header, results[len(results)-1].Header
1551
			glog.Infof("Inserting chain with %d items (#%d [%x…] - #%d [%x…])", len(results), first.Number, first.Hash().Bytes()[:4], last.Number, last.Hash().Bytes()[:4])
1552 1553
		}
		for len(results) != 0 {
1554
			// Check for any termination requests
1555
			if atomic.LoadInt32(&d.interrupt) == 1 {
1556
				return errCancelProcessing
1557
			}
1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569
			// Retrieve the a batch of results to import
			var (
				blocks   = make([]*types.Block, 0, maxResultsProcess)
				receipts = make([]types.Receipts, 0, maxResultsProcess)
			)
			items := int(math.Min(float64(len(results)), float64(maxResultsProcess)))
			for _, result := range results[:items] {
				switch {
				case d.mode == FullSync:
					blocks = append(blocks, types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles))
				case d.mode == FastSync:
					blocks = append(blocks, types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles))
1570
					if result.Header.Number.Uint64() <= pivot {
1571 1572
						receipts = append(receipts, result.Receipts)
					}
1573 1574 1575 1576 1577 1578 1579 1580
				}
			}
			// Try to process the results, aborting if there's an error
			var (
				err   error
				index int
			)
			switch {
1581 1582
			case len(receipts) > 0:
				index, err = d.insertReceipts(blocks, receipts)
1583 1584
				if err == nil && blocks[len(blocks)-1].NumberU64() == pivot {
					glog.V(logger.Debug).Infof("Committing block #%d [%x…] as the new head", blocks[len(blocks)-1].Number(), blocks[len(blocks)-1].Hash().Bytes()[:4])
1585
					index, err = len(blocks)-1, d.commitHeadBlock(blocks[len(blocks)-1].Hash())
1586 1587 1588
				}
			default:
				index, err = d.insertBlocks(blocks)
1589 1590
			}
			if err != nil {
1591
				glog.V(logger.Debug).Infof("Result #%d [%x…] processing failed: %v", results[index].Header.Number, results[index].Header.Hash().Bytes()[:4], err)
1592
				return err
1593
			}
1594 1595
			// Shift the results to the next batch
			results = results[items:]
1596 1597 1598 1599
		}
	}
}

1600
// DeliverHashes injects a new batch of hashes received from a remote node into
1601 1602
// the download schedule. This is usually invoked through the BlockHashesMsg by
// the protocol handler.
1603 1604
func (d *Downloader) DeliverHashes(id string, hashes []common.Hash) (err error) {
	return d.deliver(id, d.hashCh, &hashPack{id, hashes}, hashInMeter, hashDropMeter)
1605 1606
}

1607
// DeliverBlocks injects a new batch of blocks received from a remote node.
1608
// This is usually invoked through the BlocksMsg by the protocol handler.
1609 1610
func (d *Downloader) DeliverBlocks(id string, blocks []*types.Block) (err error) {
	return d.deliver(id, d.blockCh, &blockPack{id, blocks}, blockInMeter, blockDropMeter)
O
moved  
obscuren 已提交
1611 1612
}

1613 1614
// DeliverHeaders injects a new batch of blck headers received from a remote
// node into the download schedule.
1615
func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) (err error) {
1616
	return d.deliver(id, d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter)
1617 1618 1619
}

// DeliverBodies injects a new batch of block bodies received from a remote node.
1620
func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) (err error) {
1621
	return d.deliver(id, d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter)
1622 1623 1624 1625
}

// DeliverReceipts injects a new batch of receipts received from a remote node.
func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) (err error) {
1626 1627 1628 1629 1630 1631 1632 1633 1634 1635
	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) {
1636
	// Update the delivery metrics for both good and failed deliveries
1637
	inMeter.Mark(int64(packet.Items()))
1638 1639
	defer func() {
		if err != nil {
1640
			dropMeter.Mark(int64(packet.Items()))
1641 1642 1643 1644 1645 1646
		}
	}()
	// Deliver or abort if the sync is canceled while queuing
	d.cancelLock.RLock()
	cancel := d.cancelCh
	d.cancelLock.RUnlock()
1647 1648 1649
	if cancel == nil {
		return errNoSyncActive
	}
1650
	select {
1651
	case destCh <- packet:
1652 1653 1654 1655
		return nil
	case <-cancel:
		return errNoSyncActive
	}
1656
}