aboutsummaryrefslogtreecommitdiffstats
path: root/eth/downloader/downloader.go
blob: fd588d2f30e1e37830669ee7a0f262e0d92b7c39 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
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
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
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
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
package downloader

import (
    "errors"
    "math/rand"
    "sync"
    "sync/atomic"
    "time"

    "github.com/ethereum/go-ethereum/common"
    "github.com/ethereum/go-ethereum/core/types"
    "github.com/ethereum/go-ethereum/event"
    "github.com/ethereum/go-ethereum/logger"
    "github.com/ethereum/go-ethereum/logger/glog"
)

const (
    MinHashFetch  = 512  // Minimum amount of hashes to not consider a peer stalling
    MaxHashFetch  = 2048 // Amount of hashes to be fetched per retrieval request
    MaxBlockFetch = 128  // Amount of blocks to be fetched per retrieval request

    peerCountTimeout = 12 * time.Second // Amount of time it takes for the peer handler to ignore minDesiredPeerCount
    hashTTL          = 5 * time.Second  // Time it takes for a hash request to time out
)

var (
    blockTTL            = 5 * time.Second // Time it takes for a block request to time out
    crossCheckCycle     = time.Second     // Period after which to check for expired cross checks
    minDesiredPeerCount = 5               // Amount of peers desired to start syncing
)

var (
    errLowTd            = errors.New("peers TD is too low")
    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")
    ErrPendingQueue     = errors.New("pending items in queue")
    ErrTimeout          = errors.New("timeout")
    errEmptyHashSet     = errors.New("empty hash set by peer")
    errPeersUnavailable = errors.New("no peers available or all peers tried for block download process")
    errAlreadyInPool    = errors.New("hash already in pool")
    ErrInvalidChain     = errors.New("retrieved hash chain is invalid")
    ErrCrossCheckFailed = errors.New("block cross-check failed")
    errCancelHashFetch  = errors.New("hash fetching cancelled (requested)")
    errCancelBlockFetch = errors.New("block downloading cancelled (requested)")
    errNoSyncActive     = errors.New("no sync active")
)

type hashCheckFn func(common.Hash) bool
type getBlockFn func(common.Hash) *types.Block
type chainInsertFn func(types.Blocks) (int, error)
type hashIterFn func() (common.Hash, error)

type blockPack struct {
    peerId string
    blocks []*types.Block
}

type hashPack struct {
    peerId string
    hashes []common.Hash
}

type crossCheck struct {
    expire time.Time
    parent common.Hash
}

type Downloader struct {
    mux *event.TypeMux

    mu     sync.RWMutex
    queue  *queue                      // Scheduler for selecting the hashes to download
    peers  *peerSet                    // Set of active peers from which download can proceed
    checks map[common.Hash]*crossCheck // Pending cross checks to verify a hash chain

    // Callbacks
    hasBlock hashCheckFn
    getBlock getBlockFn

    // Status
    synchronising int32
    notified      int32

    // Channels
    newPeerCh chan *peer
    hashCh    chan hashPack
    blockCh   chan blockPack

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

func New(mux *event.TypeMux, hasBlock hashCheckFn, getBlock getBlockFn) *Downloader {
    downloader := &Downloader{
        mux:       mux,
        queue:     newQueue(),
        peers:     newPeerSet(),
        hasBlock:  hasBlock,
        getBlock:  getBlock,
        newPeerCh: make(chan *peer, 1),
        hashCh:    make(chan hashPack, 1),
        blockCh:   make(chan blockPack, 1),
    }
    return downloader
}

func (d *Downloader) Stats() (current int, max int) {
    return d.queue.Size()
}

// Synchronising returns the state of the downloader
func (d *Downloader) Synchronising() bool {
    return atomic.LoadInt32(&d.synchronising) > 0
}

// RegisterPeer injects a new download peer into the set of block source to be
// used for fetching hashes and blocks from.
func (d *Downloader) RegisterPeer(id string, head common.Hash, getHashes hashFetcherFn, getBlocks blockFetcherFn) error {
    glog.V(logger.Detail).Infoln("Registering peer", id)
    if err := d.peers.Register(newPeer(id, head, getHashes, getBlocks)); err != nil {
        glog.V(logger.Error).Infoln("Register failed:", err)
        return err
    }
    return nil
}

// UnregisterPeer remove a peer from the known list, preventing any action from
// the specified peer.
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
    }
    return nil
}

// Synchronise will select the peer and use it for synchronising. If an empty string is given
// it will use the best peer possible and synchronize if it's TD is higher than our own. If any of the
// checks fail an error will be returned. This method is synchronous
func (d *Downloader) Synchronise(id string, hash common.Hash) error {
    // Make sure only one goroutine is ever allowed past this point at once
    if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) {
        return ErrBusy
    }
    defer atomic.StoreInt32(&d.synchronising, 0)

    // 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")
    }

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

    // Abort if the queue still contains some leftover data
    if _, cached := d.queue.Size(); cached > 0 && d.queue.GetHeadBlock() != nil {
        return ErrPendingQueue
    }
    // Reset the queue and peer set to clean any internal leftover state
    d.queue.Reset()
    d.peers.Reset()
    d.checks = make(map[common.Hash]*crossCheck)

    // Retrieve the origin peer and initiate the downloading process
    p := d.peers.Peer(id)
    if p == nil {
        return errUnknownPeer
    }

    return d.syncWithPeer(p, hash)
}

// TakeBlocks takes blocks from the queue and yields them to the caller.
func (d *Downloader) TakeBlocks() types.Blocks {
    return d.queue.TakeBlocks()
}

func (d *Downloader) Has(hash common.Hash) bool {
    return d.queue.Has(hash)
}

// syncWithPeer starts a block synchronization based on the hash chain from the
// specified peer and head hash.
func (d *Downloader) syncWithPeer(p *peer, hash common.Hash) (err error) {
    d.mux.Post(StartEvent{})
    defer func() {
        // reset on error
        if err != nil {
            d.Cancel()
            d.mux.Post(FailedEvent{err})
        } else {
            d.mux.Post(DoneEvent{})
        }
    }()

    glog.V(logger.Debug).Infoln("Synchronizing with the network using:", p.id)
    if err = d.fetchHashes(p, hash); err != nil {
        return err
    }
    if err = d.fetchBlocks(); err != nil {
        return err
    }
    glog.V(logger.Debug).Infoln("Synchronization completed")

    return nil
}

// Cancel cancels all of the operations and resets the queue. It returns true
// if the cancel operation was completed.
func (d *Downloader) Cancel() bool {
    // If we're not syncing just return.
    hs, bs := d.queue.Size()
    if atomic.LoadInt32(&d.synchronising) == 0 && hs == 0 && bs == 0 {
        return false
    }
    // Close the current cancel channel
    d.cancelLock.Lock()
    select {
    case <-d.cancelCh:
        // Channel was already closed
    default:
        close(d.cancelCh)
    }
    d.cancelLock.Unlock()

    // reset the queue
    d.queue.Reset()

    return true
}

// XXX Make synchronous
func (d *Downloader) fetchHashes(p *peer, h common.Hash) error {
    glog.V(logger.Debug).Infof("Downloading hashes (%x) from %s", h[:4], p.id)

    start := time.Now()

    // Add the hash to the queue first, and start hash retrieval
    d.queue.Insert([]common.Hash{h})
    p.getHashes(h)

    var (
        active = p             // active peer will help determine the current active peer
        head   = common.Hash{} // common and last hash

        timeout     = time.NewTimer(hashTTL)          // timer to dump a non-responsive active peer
        attempted   = make(map[string]bool)           // attempted peers will help with retries
        crossTicker = time.NewTicker(crossCheckCycle) // ticker to periodically check expired cross checks
    )
    defer crossTicker.Stop()

    attempted[p.id] = true
    for finished := false; !finished; {
        select {
        case <-d.cancelCh:
            return errCancelHashFetch

        case hashPack := <-d.hashCh:
            // Make sure the active peer is giving us the hashes
            if hashPack.peerId != active.id {
                glog.V(logger.Debug).Infof("Received hashes from incorrect peer(%s)\n", hashPack.peerId)
                break
            }
            timeout.Reset(hashTTL)

            // Make sure the peer actually gave something valid
            if len(hashPack.hashes) == 0 {
                glog.V(logger.Debug).Infof("Peer (%s) responded with empty hash set\n", active.id)
                return errEmptyHashSet
            }
            // Determine if we're done fetching hashes (queue up all pending), and continue if not done
            done, index := false, 0
            for index, head = range hashPack.hashes {
                if d.hasBlock(head) || d.queue.GetBlock(head) != nil {
                    glog.V(logger.Debug).Infof("Found common hash %x\n", head[:4])
                    hashPack.hashes = hashPack.hashes[:index]
                    done = true
                    break
                }
            }
            // Insert all the new hashes, but only continue if got something useful
            inserts := d.queue.Insert(hashPack.hashes)
            if len(inserts) == 0 && !done {
                glog.V(logger.Debug).Infof("Peer (%s) responded with stale hashes\n", active.id)
                return ErrBadPeer
            }
            if !done {
                // Check that the peer is not stalling the sync
                if len(inserts) < MinHashFetch {
                    return ErrStallingPeer
                }
                // Try and fetch a random block to verify the hash batch
                // Skip the last hash as the cross check races with the next hash fetch
                cross := rand.Intn(len(inserts) - 1)
                origin, parent := inserts[cross], inserts[cross+1]
                glog.V(logger.Detail).Infof("Cross checking (%s) with %x/%x", active.id, origin, parent)

                d.checks[origin] = &crossCheck{
                    expire: time.Now().Add(blockTTL),
                    parent: parent,
                }
                active.getBlocks([]common.Hash{origin})

                // Also fetch a fresh
                active.getHashes(head)
                continue
            }
            // We're done, allocate the download cache and proceed pulling the blocks
            offset := 0
            if block := d.getBlock(head); block != nil {
                offset = int(block.NumberU64() + 1)
            }
            d.queue.Alloc(offset)
            finished = true

        case blockPack := <-d.blockCh:
            // Cross check the block with the random verifications
            if blockPack.peerId != active.id || len(blockPack.blocks) != 1 {
                continue
            }
            block := blockPack.blocks[0]
            if check, ok := d.checks[block.Hash()]; ok {
                if block.ParentHash() != check.parent {
                    return ErrCrossCheckFailed
                }
                delete(d.checks, block.Hash())
            }

        case <-crossTicker.C:
            // Iterate over all the cross checks and fail the hash chain if they're not verified
            for hash, check := range d.checks {
                if time.Now().After(check.expire) {
                    glog.V(logger.Debug).Infof("Cross check timeout for %x", hash)
                    return ErrCrossCheckFailed
                }
            }

        case <-timeout.C:
            glog.V(logger.Debug).Infof("Peer (%s) didn't respond in time for hash request\n", p.id)

            var p *peer // p will be set if a peer can be found
            // Attempt to find a new peer by checking inclusion of peers best hash in our
            // already fetched hash list. This can't guarantee 100% correctness but does
            // a fair job. This is always either correct or false incorrect.
            for _, peer := range d.peers.AllPeers() {
                if d.queue.Has(peer.head) && !attempted[peer.id] {
                    p = peer
                    break
                }
            }
            // if all peers have been tried, abort the process entirely or if the hash is
            // the zero hash.
            if p == nil || (head == common.Hash{}) {
                return ErrTimeout
            }
            // set p to the active peer. this will invalidate any hashes that may be returned
            // by our previous (delayed) peer.
            active = p
            p.getHashes(head)
            glog.V(logger.Debug).Infof("Hash fetching switched to new peer(%s)\n", p.id)
        }
    }
    glog.V(logger.Debug).Infof("Downloaded hashes (%d) in %v\n", d.queue.Pending(), time.Since(start))

    return nil
}

// fetchBlocks iteratively downloads the entire schedules block-chain, taking
// any available peers, reserving a chunk of blocks for each, wait for delivery
// and periodically checking for timeouts.
func (d *Downloader) fetchBlocks() error {
    glog.V(logger.Debug).Infoln("Downloading", d.queue.Pending(), "block(s)")
    start := time.Now()

    // default ticker for re-fetching blocks every now and then
    ticker := time.NewTicker(20 * time.Millisecond)
out:
    for {
        select {
        case <-d.cancelCh:
            return errCancelBlockFetch

        case blockPack := <-d.blockCh:
            // Short circuit if it's a stale cross check
            if len(blockPack.blocks) == 1 {
                block := blockPack.blocks[0]
                if _, ok := d.checks[block.Hash()]; ok {
                    delete(d.checks, block.Hash())
                    continue
                }
            }
            // If the peer was previously banned and failed to deliver it's pack
            // in a reasonable time frame, ignore it's message.
            if peer := d.peers.Peer(blockPack.peerId); peer != nil {
                // Deliver the received chunk of blocks
                if err := d.queue.Deliver(blockPack.peerId, blockPack.blocks); err != nil {
                    if err == ErrInvalidChain {
                        // The hash chain is invalid (blocks are not ordered properly), abort
                        return err
                    }
                    // Peer did deliver, but some blocks were off, penalize
                    glog.V(logger.Debug).Infof("Failed delivery for peer %s: %v\n", blockPack.peerId, err)
                    peer.Demote()
                    break
                }
                if glog.V(logger.Debug) {
                    glog.Infof("Added %d blocks from: %s\n", len(blockPack.blocks), blockPack.peerId)
                }
                // Promote the peer and update it's idle state
                peer.Promote()
                peer.SetIdle()
            }
        case <-ticker.C:
            // Check for bad peers. Bad peers may indicate a peer not responding
            // to a `getBlocks` message. A timeout of 5 seconds is set. Peers
            // that badly or poorly behave are removed from the peer set (not banned).
            // Bad peers are excluded from the available peer set and therefor won't be
            // reused. XXX We could re-introduce peers after X time.
            badPeers := d.queue.Expire(blockTTL)
            for _, pid := range badPeers {
                // XXX We could make use of a reputation system here ranking peers
                // in their performance
                // 1) Time for them to respond;
                // 2) Measure their speed;
                // 3) Amount and availability.
                if peer := d.peers.Peer(pid); peer != nil {
                    peer.Demote()
                }
            }
            // After removing bad peers make sure we actually have sufficient peer left to keep downloading
            if d.peers.Len() == 0 {
                return errNoPeers
            }
            // If there are unrequested hashes left start fetching
            // from the available peers.
            if d.queue.Pending() > 0 {
                // Throttle the download if block cache is full and waiting processing
                if d.queue.Throttle() {
                    continue
                }
                // Send a download request to all idle peers, until throttled
                idlePeers := d.peers.IdlePeers()
                for _, peer := range idlePeers {
                    // Short circuit if throttling activated since above
                    if d.queue.Throttle() {
                        break
                    }
                    // Get a possible chunk. If nil is returned no chunk
                    // could be returned due to no hashes available.
                    request := d.queue.Reserve(peer, MaxBlockFetch)
                    if request == nil {
                        continue
                    }
                    // Fetch the chunk and check for error. If the peer was somehow
                    // already fetching a chunk due to a bug, it will be returned to
                    // the queue
                    if err := peer.Fetch(request); err != nil {
                        glog.V(logger.Error).Infof("Peer %s received double work\n", peer.id)
                        d.queue.Cancel(request)
                    }
                }
                // Make sure that we have peers available for fetching. If all peers have been tried
                // and all failed throw an error
                if d.queue.InFlight() == 0 {
                    return errPeersUnavailable
                }

            } else if d.queue.InFlight() == 0 {
                // When there are no more queue and no more in flight, We can
                // safely assume we're done. Another part of the process will  check
                // for parent errors and will re-request anything that's missing
                break out
            }
        }
    }
    glog.V(logger.Detail).Infoln("Downloaded block(s) in", time.Since(start))

    return nil
}

// DeliverBlocks injects a new batch of blocks received from a remote node.
// This is usually invoked through the BlocksMsg by the protocol handler.
func (d *Downloader) DeliverBlocks(id string, blocks []*types.Block) error {
    // Make sure the downloader is active
    if atomic.LoadInt32(&d.synchronising) == 0 {
        return errNoSyncActive
    }
    // Deliver or abort if the sync is canceled while queuing
    d.cancelLock.RLock()
    cancel := d.cancelCh
    d.cancelLock.RUnlock()

    select {
    case d.blockCh <- blockPack{id, blocks}:
        return nil

    case <-cancel:
        return errNoSyncActive
    }
}

// DeliverHashes injects a new batch of hashes received from a remote node into
// the download schedule. This is usually invoked through the BlockHashesMsg by
// the protocol handler.
func (d *Downloader) DeliverHashes(id string, hashes []common.Hash) error {
    // Make sure the downloader is active
    if atomic.LoadInt32(&d.synchronising) == 0 {
        return errNoSyncActive
    }
    // Deliver or abort if the sync is canceled while queuing
    d.cancelLock.RLock()
    cancel := d.cancelCh
    d.cancelLock.RUnlock()

    select {
    case d.hashCh <- hashPack{id, hashes}:
        return nil

    case <-cancel:
        return errNoSyncActive
    }
}