From cd2fb0905109828028172c84f9c10f1343647ca6 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?P=C3=A9ter=20Szil=C3=A1gyi?= Date: Fri, 15 May 2015 01:40:16 +0300 Subject: [PATCH 1/8] eth, eth/downloader: prevent hash repeater attack --- eth/downloader/downloader.go | 10 ++++++---- eth/downloader/downloader_test.go | 29 +++++++++++++++++++++++++++++ eth/downloader/queue.go | 20 +++++++++++--------- eth/sync.go | 6 ++++-- 4 files changed, 50 insertions(+), 15 deletions(-) diff --git a/eth/downloader/downloader.go b/eth/downloader/downloader.go index 616971f73..3da606aef 100644 --- a/eth/downloader/downloader.go +++ b/eth/downloader/downloader.go @@ -27,7 +27,7 @@ var ( errLowTd = errors.New("peer's TD is too low") ErrBusy = errors.New("busy") errUnknownPeer = errors.New("peer's unknown or unhealthy") - errBadPeer = errors.New("action from bad peer ignored") + ErrBadPeer = errors.New("action from bad peer ignored") errNoPeers = errors.New("no peers to keep download active") ErrPendingQueue = errors.New("pending items in queue") ErrTimeout = errors.New("timeout") @@ -266,9 +266,11 @@ out: break } } - d.queue.Insert(hashPack.hashes) - - if !done { + // Insert all the new hashes, but only continue if got something useful + inserts := d.queue.Insert(hashPack.hashes) + if inserts == 0 && !done { + return ErrBadPeer + } else if !done { activePeer.getHashes(hash) continue } diff --git a/eth/downloader/downloader_test.go b/eth/downloader/downloader_test.go index 50fe00d42..1bba3ca51 100644 --- a/eth/downloader/downloader_test.go +++ b/eth/downloader/downloader_test.go @@ -336,3 +336,32 @@ func TestNonExistingParentAttack(t *testing.T) { t.Fatalf("tester doesn't know about the origin hash") } } + +// Tests that if a malicious peers keeps sending us repeating hashes, we don't +// loop indefinitely. +func TestRepeatingHashAttack(t *testing.T) { + // Create a valid chain, but drop the last link + hashes := createHashes(1000, 1) + blocks := createBlocksFromHashes(hashes) + + hashes = hashes[:len(hashes)-1] + + // Try and sync with the malicious node + tester := newTester(t, hashes, blocks) + tester.newPeer("attack", big.NewInt(10000), hashes[0]) + + errc := make(chan error) + go func() { + errc <- tester.sync("attack", hashes[0]) + }() + + // Make sure that syncing returns and does so with a failure + select { + case <-time.After(100 * time.Millisecond): + t.Fatalf("synchronisation blocked") + case err := <-errc: + if err == nil { + t.Fatalf("synchronisation succeeded") + } + } +} diff --git a/eth/downloader/queue.go b/eth/downloader/queue.go index 6ad915757..fdea1f63f 100644 --- a/eth/downloader/queue.go +++ b/eth/downloader/queue.go @@ -122,24 +122,26 @@ func (q *queue) Has(hash common.Hash) bool { return false } -// Insert adds a set of hashes for the download queue for scheduling. -func (q *queue) Insert(hashes []common.Hash) { +// Insert adds a set of hashes for the download queue for scheduling, returning +// the number of new hashes encountered. +func (q *queue) Insert(hashes []common.Hash) int { q.lock.Lock() defer q.lock.Unlock() // Insert all the hashes prioritized in the arrival order - for i, hash := range hashes { - index := q.hashCounter + i - + inserts := 0 + for _, hash := range hashes { + // Skip anything we already have if old, ok := q.hashPool[hash]; ok { glog.V(logger.Warn).Infof("Hash %x already scheduled at index %v", hash, old) continue } - q.hashPool[hash] = index - q.hashQueue.Push(hash, float32(index)) // Highest gets schedules first + // Update the counters and insert the hash + q.hashCounter, inserts = q.hashCounter+1, inserts+1 + q.hashPool[hash] = q.hashCounter + q.hashQueue.Push(hash, float32(q.hashCounter)) // Highest gets schedules first } - // Update the hash counter for the next batch of inserts - q.hashCounter += len(hashes) + return inserts } // GetHeadBlock retrieves the first block from the cache, or nil if it hasn't diff --git a/eth/sync.go b/eth/sync.go index c89f34596..73f6253be 100644 --- a/eth/sync.go +++ b/eth/sync.go @@ -101,11 +101,13 @@ func (pm *ProtocolManager) synchronise(peer *peer) { case downloader.ErrBusy: glog.V(logger.Debug).Infof("Synchronisation already in progress") - case downloader.ErrTimeout: - glog.V(logger.Debug).Infof("Removing peer %v due to sync timeout", peer.id) + case downloader.ErrTimeout, downloader.ErrBadPeer: + glog.V(logger.Debug).Infof("Removing peer %v: %v", peer.id, err) pm.removePeer(peer) + case downloader.ErrPendingQueue: glog.V(logger.Debug).Infoln("Synchronisation aborted:", err) + default: glog.V(logger.Warn).Infof("Synchronisation failed: %v", err) } From 366e9627e8cdddd8111789dc13050b1dd8ec6fb2 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?P=C3=A9ter=20Szil=C3=A1gyi?= Date: Fri, 15 May 2015 01:56:52 +0300 Subject: [PATCH 2/8] eth/downloader: add a test for detecting missing blocks --- eth/downloader/downloader.go | 4 +--- eth/downloader/downloader_test.go | 28 +++++++++++++++++++++++----- 2 files changed, 24 insertions(+), 8 deletions(-) diff --git a/eth/downloader/downloader.go b/eth/downloader/downloader.go index 3da606aef..c90d289df 100644 --- a/eth/downloader/downloader.go +++ b/eth/downloader/downloader.go @@ -2,7 +2,6 @@ package downloader import ( "errors" - "fmt" "sync" "sync/atomic" "time" @@ -398,8 +397,7 @@ out: // and all failed throw an error if d.queue.InFlight() == 0 { d.queue.Reset() - - return fmt.Errorf("%v peers available = %d. total peers = %d. hashes needed = %d", errPeersUnavailable, len(idlePeers), d.peers.Len(), d.queue.Pending()) + return errPeersUnavailable } } else if d.queue.InFlight() == 0 { diff --git a/eth/downloader/downloader_test.go b/eth/downloader/downloader_test.go index 1bba3ca51..46538d6b4 100644 --- a/eth/downloader/downloader_test.go +++ b/eth/downloader/downloader_test.go @@ -106,11 +106,12 @@ func (dl *downloadTester) getHashes(hash common.Hash) error { func (dl *downloadTester) getBlocks(id string) func([]common.Hash) error { return func(hashes []common.Hash) error { - blocks := make([]*types.Block, len(hashes)) - for i, hash := range hashes { - blocks[i] = dl.blocks[hash] + blocks := make([]*types.Block, 0, len(hashes)) + for _, hash := range hashes { + if block, ok := dl.blocks[hash]; ok { + blocks = append(blocks, block) + } } - go dl.downloader.DeliverBlocks(id, blocks) return nil @@ -341,7 +342,7 @@ func TestNonExistingParentAttack(t *testing.T) { // loop indefinitely. func TestRepeatingHashAttack(t *testing.T) { // Create a valid chain, but drop the last link - hashes := createHashes(1000, 1) + hashes := createHashes(0, 1000) blocks := createBlocksFromHashes(hashes) hashes = hashes[:len(hashes)-1] @@ -365,3 +366,20 @@ func TestRepeatingHashAttack(t *testing.T) { } } } + +// Tests that if a malicious peers returns a non-existent block hash, it should +// eventually time out and the sync reattempted. +func TestNonExistingBlockAttack(t *testing.T) { + // Create a valid chain, but forge the last link + hashes := createHashes(0, 10) + blocks := createBlocksFromHashes(hashes) + + hashes[len(hashes)/2] = unknownHash + + // Try and sync with the malicious node and check that it fails + tester := newTester(t, hashes, blocks) + tester.newPeer("attack", big.NewInt(10000), hashes[0]) + if err := tester.sync("attack", hashes[0]); err != errPeersUnavailable { + t.Fatalf("synchronisation error mismatch: have %v, want %v", err, errPeersUnavailable) + } +} From bcc2980179f39eea4825df72ad882274086d912e Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?P=C3=A9ter=20Szil=C3=A1gyi?= Date: Fri, 15 May 2015 02:14:21 +0300 Subject: [PATCH 3/8] eth/downloader: check sync after failed attacks --- eth/downloader/downloader.go | 6 +++++- eth/downloader/downloader_test.go | 24 ++++++++++++++++++------ 2 files changed, 23 insertions(+), 7 deletions(-) diff --git a/eth/downloader/downloader.go b/eth/downloader/downloader.go index c90d289df..30be0dde5 100644 --- a/eth/downloader/downloader.go +++ b/eth/downloader/downloader.go @@ -268,8 +268,12 @@ out: // Insert all the new hashes, but only continue if got something useful inserts := d.queue.Insert(hashPack.hashes) if inserts == 0 && !done { + glog.V(logger.Debug).Infof("Peer (%s) responded with stale hashes\n", activePeer.id) + d.queue.Reset() + return ErrBadPeer - } else if !done { + } + if !done { activePeer.getHashes(hash) continue } diff --git a/eth/downloader/downloader_test.go b/eth/downloader/downloader_test.go index 46538d6b4..7888bd1e0 100644 --- a/eth/downloader/downloader_test.go +++ b/eth/downloader/downloader_test.go @@ -342,14 +342,13 @@ func TestNonExistingParentAttack(t *testing.T) { // loop indefinitely. func TestRepeatingHashAttack(t *testing.T) { // Create a valid chain, but drop the last link - hashes := createHashes(0, 1000) + hashes := createHashes(0, blockCacheLimit) blocks := createBlocksFromHashes(hashes) - - hashes = hashes[:len(hashes)-1] + forged := hashes[:len(hashes)-1] // Try and sync with the malicious node - tester := newTester(t, hashes, blocks) - tester.newPeer("attack", big.NewInt(10000), hashes[0]) + tester := newTester(t, forged, blocks) + tester.newPeer("attack", big.NewInt(10000), forged[0]) errc := make(chan error) go func() { @@ -365,14 +364,21 @@ func TestRepeatingHashAttack(t *testing.T) { t.Fatalf("synchronisation succeeded") } } + // Ensure that a valid chain can still pass sync + tester.hashes = hashes + tester.newPeer("valid", big.NewInt(20000), hashes[0]) + if err := tester.sync("valid", hashes[0]); err != nil { + t.Fatalf("failed to synchronise blocks: %v", err) + } } // Tests that if a malicious peers returns a non-existent block hash, it should // eventually time out and the sync reattempted. func TestNonExistingBlockAttack(t *testing.T) { // Create a valid chain, but forge the last link - hashes := createHashes(0, 10) + hashes := createHashes(0, blockCacheLimit) blocks := createBlocksFromHashes(hashes) + origin := hashes[len(hashes)/2] hashes[len(hashes)/2] = unknownHash @@ -382,4 +388,10 @@ func TestNonExistingBlockAttack(t *testing.T) { if err := tester.sync("attack", hashes[0]); err != errPeersUnavailable { t.Fatalf("synchronisation error mismatch: have %v, want %v", err, errPeersUnavailable) } + // Ensure that a valid chain can still pass sync + hashes[len(hashes)/2] = origin + tester.newPeer("valid", big.NewInt(20000), hashes[0]) + if err := tester.sync("valid", hashes[0]); err != nil { + t.Fatalf("failed to synchronise blocks: %v", err) + } } From 72411eb24c47a6b41d8530e6057a88c60491f0e1 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?P=C3=A9ter=20Szil=C3=A1gyi?= Date: Fri, 15 May 2015 11:58:37 +0300 Subject: [PATCH 4/8] eth/downloader: circumvent hash reordering attacks --- eth/downloader/downloader.go | 36 +++++++------ eth/downloader/downloader_test.go | 90 +++++++++++++++++++++---------- eth/downloader/queue.go | 11 ++-- 3 files changed, 88 insertions(+), 49 deletions(-) diff --git a/eth/downloader/downloader.go b/eth/downloader/downloader.go index 30be0dde5..426da9beb 100644 --- a/eth/downloader/downloader.go +++ b/eth/downloader/downloader.go @@ -23,20 +23,20 @@ var ( minDesiredPeerCount = 5 // Amount of peers desired to start syncing blockTtl = 20 * time.Second // The amount of time it takes for a block request to time out - errLowTd = errors.New("peer's TD is too low") - ErrBusy = errors.New("busy") - errUnknownPeer = errors.New("peer's unknown or unhealthy") - ErrBadPeer = errors.New("action from bad peer ignored") - 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") - errBlockNumberOverflow = errors.New("received block which overflows") - errCancelHashFetch = errors.New("hash fetching cancelled (requested)") - errCancelBlockFetch = errors.New("block downloading cancelled (requested)") - errNoSyncActive = errors.New("no sync active") + errLowTd = errors.New("peer's TD is too low") + ErrBusy = errors.New("busy") + errUnknownPeer = errors.New("peer's unknown or unhealthy") + ErrBadPeer = errors.New("action from bad peer ignored") + 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") + 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 @@ -334,8 +334,14 @@ out: // 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, but drop the peer if invalid + // 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 + d.queue.Reset() + 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 diff --git a/eth/downloader/downloader_test.go b/eth/downloader/downloader_test.go index 7888bd1e0..4b8ee93d2 100644 --- a/eth/downloader/downloader_test.go +++ b/eth/downloader/downloader_test.go @@ -75,9 +75,40 @@ func newTester(t *testing.T, hashes []common.Hash, blocks map[common.Hash]*types return tester } -func (dl *downloadTester) sync(peerId string, hash common.Hash) error { +// sync is a simple wrapper around the downloader to start synchronisation and +// block until it returns +func (dl *downloadTester) sync(peerId string, head common.Hash) error { dl.activePeerId = peerId - return dl.downloader.Synchronise(peerId, hash) + return dl.downloader.Synchronise(peerId, head) +} + +// syncTake is starts synchronising with a remote peer, but concurrently it also +// starts fetching blocks that the downloader retrieved. IT blocks until both go +// routines terminate. +func (dl *downloadTester) syncTake(peerId string, head common.Hash) (types.Blocks, error) { + // Start a block collector to take blocks as they become available + done := make(chan struct{}) + took := []*types.Block{} + go func() { + for running := true; running; { + select { + case <-done: + running = false + default: + time.Sleep(time.Millisecond) + } + // Take a batch of blocks and accumulate + took = append(took, dl.downloader.TakeBlocks()...) + } + done <- struct{}{} + }() + // Start the downloading, sync the taker and return + err := dl.sync(peerId, head) + + done <- struct{}{} + <-done + + return took, err } func (dl *downloadTester) insertBlocks(blocks types.Blocks) { @@ -264,32 +295,7 @@ func TestThrottling(t *testing.T) { tester.badBlocksPeer("peer4", big.NewInt(0), common.Hash{}) // Concurrently download and take the blocks - errc := make(chan error, 1) - go func() { - errc <- tester.sync("peer1", hashes[0]) - }() - - done := make(chan struct{}) - took := []*types.Block{} - go func() { - for running := true; running; { - select { - case <-done: - running = false - default: - time.Sleep(time.Millisecond) - } - // Take a batch of blocks and accumulate - took = append(took, tester.downloader.TakeBlocks()...) - } - done <- struct{}{} - }() - - // Synchronise the two threads and verify - err := <-errc - done <- struct{}{} - <-done - + took, err := tester.syncTake("peer1", hashes[0]) if err != nil { t.Fatalf("failed to synchronise blocks: %v", err) } @@ -395,3 +401,31 @@ func TestNonExistingBlockAttack(t *testing.T) { t.Fatalf("failed to synchronise blocks: %v", err) } } + +// Tests that if a malicious peer is returning hashes in a weird order, that the +// sync throttler doesn't choke on them waiting for the valid blocks. +func TestInvalidHashOrderAttack(t *testing.T) { + // Create a valid long chain, but reverse some hashes within + hashes := createHashes(0, 4*blockCacheLimit) + blocks := createBlocksFromHashes(hashes) + + reverse := make([]common.Hash, len(hashes)) + copy(reverse, hashes) + + for i := len(hashes) / 4; i < 2*len(hashes)/4; i++ { + reverse[i], reverse[len(hashes)-i-1] = reverse[len(hashes)-i-1], reverse[i] + } + + // Try and sync with the malicious node and check that it fails + tester := newTester(t, reverse, blocks) + tester.newPeer("attack", big.NewInt(10000), reverse[0]) + if _, err := tester.syncTake("attack", reverse[0]); err != ErrInvalidChain { + t.Fatalf("synchronisation error mismatch: have %v, want %v", err, ErrInvalidChain) + } + // Ensure that a valid chain can still pass sync + tester.hashes = hashes + tester.newPeer("valid", big.NewInt(20000), hashes[0]) + if _, err := tester.syncTake("valid", hashes[0]); err != nil { + t.Fatalf("failed to synchronise blocks: %v", err) + } +} diff --git a/eth/downloader/queue.go b/eth/downloader/queue.go index fdea1f63f..aa48c521a 100644 --- a/eth/downloader/queue.go +++ b/eth/downloader/queue.go @@ -298,18 +298,17 @@ func (q *queue) Deliver(id string, blocks []*types.Block) (err error) { // Iterate over the downloaded blocks and add each of them errs := make([]error, 0) for _, block := range blocks { - // Skip any blocks that fall outside the cache range - index := int(block.NumberU64()) - q.blockOffset - if index >= len(q.blockCache) || index < 0 { - //fmt.Printf("block cache overflown (N=%v O=%v, C=%v)", block.Number(), q.blockOffset, len(q.blockCache)) - continue - } // Skip any blocks that were not requested hash := block.Hash() if _, ok := request.Hashes[hash]; !ok { errs = append(errs, fmt.Errorf("non-requested block %v", hash)) continue } + // If a requested block falls out of the range, the hash chain is invalid + index := int(block.NumberU64()) - q.blockOffset + if index >= len(q.blockCache) || index < 0 { + return ErrInvalidChain + } // Otherwise merge the block and mark the hash block q.blockCache[index] = block From 9ad515d2dc62c5d1ce1099efa89bd0a0b3f06a67 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?P=C3=A9ter=20Szil=C3=A1gyi?= Date: Fri, 15 May 2015 12:01:05 +0300 Subject: [PATCH 5/8] eth: drop a sync peer if it sends an invalid hash chain --- eth/sync.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/eth/sync.go b/eth/sync.go index 73f6253be..cc1fe80ea 100644 --- a/eth/sync.go +++ b/eth/sync.go @@ -101,7 +101,7 @@ func (pm *ProtocolManager) synchronise(peer *peer) { case downloader.ErrBusy: glog.V(logger.Debug).Infof("Synchronisation already in progress") - case downloader.ErrTimeout, downloader.ErrBadPeer: + case downloader.ErrTimeout, downloader.ErrBadPeer, downloader.ErrInvalidChain: glog.V(logger.Debug).Infof("Removing peer %v: %v", peer.id, err) pm.removePeer(peer) From 83226762c20dbf48939d76046ad32422a44feda0 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?P=C3=A9ter=20Szil=C3=A1gyi?= Date: Fri, 15 May 2015 13:14:46 +0300 Subject: [PATCH 6/8] eth, eth/downloader: detect and handle madeup hash attacks --- eth/downloader/downloader.go | 98 ++++++++++++++++++++----------- eth/downloader/downloader_test.go | 45 ++++++++++++-- eth/downloader/queue.go | 10 ++-- eth/sync.go | 2 +- 4 files changed, 111 insertions(+), 44 deletions(-) diff --git a/eth/downloader/downloader.go b/eth/downloader/downloader.go index 426da9beb..95dd37fd7 100644 --- a/eth/downloader/downloader.go +++ b/eth/downloader/downloader.go @@ -2,6 +2,7 @@ package downloader import ( "errors" + "math/rand" "sync" "sync/atomic" "time" @@ -14,15 +15,19 @@ import ( ) const ( - maxBlockFetch = 128 // Amount of max blocks to be fetched per chunk + maxHashFetch = 512 // Amount of hashes to be fetched per chunk + maxBlockFetch = 128 // Amount of blocks to be fetched per chunk peerCountTimeout = 12 * time.Second // Amount of time it takes for the peer handler to ignore minDesiredPeerCount - hashTtl = 20 * time.Second // The amount of time it takes for a hash request to time out + hashTTL = 5 * time.Second // Time it takes for a hash request to time out ) var ( - minDesiredPeerCount = 5 // Amount of peers desired to start syncing - blockTtl = 20 * time.Second // The amount of time it takes for a block request to time out + 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("peer's TD is too low") ErrBusy = errors.New("busy") errUnknownPeer = errors.New("peer's unknown or unhealthy") @@ -34,6 +39,7 @@ var ( 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") @@ -220,46 +226,47 @@ func (d *Downloader) fetchHashes(p *peer, h common.Hash) error { start := time.Now() - // Add the hash to the queue first + // Add the hash to the queue first, and start hash retrieval d.queue.Insert([]common.Hash{h}) - - // Get the first batch of hashes p.getHashes(h) var ( - failureResponseTimer = time.NewTimer(hashTtl) - attemptedPeers = make(map[string]bool) // attempted peers will help with retries - activePeer = p // active peer will help determine the current active peer - hash common.Hash // common and last hash - ) - attemptedPeers[p.id] = true + active = p // active peer will help determine the current active peer + head = common.Hash{} // common and last hash -out: - for { + timeout = time.NewTimer(hashTTL) // timer to dump a non-responsive active peer + attempted = make(map[string]bool) // attempted peers will help with retries + crossChecks = make(map[common.Hash]time.Time) // running cross checks and their deadline + 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 != activePeer.id { + if hashPack.peerId != active.id { glog.V(logger.Debug).Infof("Received hashes from incorrect peer(%s)\n", hashPack.peerId) break } - - failureResponseTimer.Reset(hashTtl) + 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", activePeer.id) + glog.V(logger.Debug).Infof("Peer (%s) responded with empty hash set\n", active.id) d.queue.Reset() return errEmptyHashSet } // Determine if we're done fetching hashes (queue up all pending), and continue if not done done, index := false, 0 - for index, hash = range hashPack.hashes { - if d.hasBlock(hash) || d.queue.GetBlock(hash) != nil { - glog.V(logger.Debug).Infof("Found common hash %x\n", hash[:4]) + 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 @@ -267,25 +274,50 @@ out: } // Insert all the new hashes, but only continue if got something useful inserts := d.queue.Insert(hashPack.hashes) - if inserts == 0 && !done { - glog.V(logger.Debug).Infof("Peer (%s) responded with stale hashes\n", activePeer.id) + if len(inserts) == 0 && !done { + glog.V(logger.Debug).Infof("Peer (%s) responded with stale hashes\n", active.id) d.queue.Reset() return ErrBadPeer } if !done { - activePeer.getHashes(hash) + // Try and fetch a random block to verify the hash batch + cross := inserts[rand.Intn(len(inserts))] + glog.V(logger.Detail).Infof("Cross checking (%s) with %x", active.id, cross) + + crossChecks[cross] = time.Now().Add(blockTTL) + active.getBlocks([]common.Hash{cross}) + + // 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(hash); block != nil { + if block := d.getBlock(head); block != nil { offset = int(block.NumberU64() + 1) } d.queue.Alloc(offset) - break out + finished = true - case <-failureResponseTimer.C: + case blockPack := <-d.blockCh: + // Cross check the block with the random verifications + if blockPack.peerId != active.id || len(blockPack.blocks) != 1 { + continue + } + hash := blockPack.blocks[0].Hash() + delete(crossChecks, hash) + + case <-crossTicker.C: + // Iterate over all the cross checks and fail the hash chain if they're not verified + for hash, deadline := range crossChecks { + if time.Now().After(deadline) { + 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 @@ -293,21 +325,21 @@ out: // 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) && !attemptedPeers[peer.id] { + 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 || (hash == common.Hash{}) { + if p == nil || (head == common.Hash{}) { d.queue.Reset() return ErrTimeout } // set p to the active peer. this will invalidate any hashes that may be returned // by our previous (delayed) peer. - activePeer = p - p.getHashes(hash) + active = p + p.getHashes(head) glog.V(logger.Debug).Infof("Hash fetching switched to new peer(%s)\n", p.id) } } @@ -359,7 +391,7 @@ out: // 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) + badPeers := d.queue.Expire(blockTTL) for _, pid := range badPeers { // XXX We could make use of a reputation system here ranking peers // in their performance diff --git a/eth/downloader/downloader_test.go b/eth/downloader/downloader_test.go index 4b8ee93d2..60dcc06cd 100644 --- a/eth/downloader/downloader_test.go +++ b/eth/downloader/downloader_test.go @@ -130,8 +130,23 @@ func (dl *downloadTester) getBlock(hash common.Hash) *types.Block { return dl.blocks[knownHash] } -func (dl *downloadTester) getHashes(hash common.Hash) error { - dl.downloader.DeliverHashes(dl.activePeerId, dl.hashes) +// getHashes retrieves a batch of hashes for reconstructing the chain. +func (dl *downloadTester) getHashes(head common.Hash) error { + // Gather the next batch of hashes + hashes := make([]common.Hash, 0, maxHashFetch) + for i, hash := range dl.hashes { + if hash == head { + for len(hashes) < cap(hashes) && i < len(dl.hashes) { + hashes = append(hashes, dl.hashes[i]) + i++ + } + break + } + } + // Delay delivery a bit to allow attacks to unfold + time.Sleep(time.Millisecond) + + dl.downloader.DeliverHashes(dl.activePeerId, hashes) return nil } @@ -166,7 +181,7 @@ func (dl *downloadTester) badBlocksPeer(id string, td *big.Int, hash common.Hash func TestDownload(t *testing.T) { minDesiredPeerCount = 4 - blockTtl = 1 * time.Second + blockTTL = 1 * time.Second targetBlocks := 1000 hashes := createHashes(0, targetBlocks) @@ -215,7 +230,7 @@ func TestMissing(t *testing.T) { func TestTaking(t *testing.T) { minDesiredPeerCount = 4 - blockTtl = 1 * time.Second + blockTTL = 1 * time.Second targetBlocks := 1000 hashes := createHashes(0, targetBlocks) @@ -256,7 +271,7 @@ func TestInactiveDownloader(t *testing.T) { func TestCancel(t *testing.T) { minDesiredPeerCount = 4 - blockTtl = 1 * time.Second + blockTTL = 1 * time.Second targetBlocks := 1000 hashes := createHashes(0, targetBlocks) @@ -282,7 +297,7 @@ func TestCancel(t *testing.T) { func TestThrottling(t *testing.T) { minDesiredPeerCount = 4 - blockTtl = 1 * time.Second + blockTTL = 1 * time.Second targetBlocks := 16 * blockCacheLimit hashes := createHashes(0, targetBlocks) @@ -429,3 +444,21 @@ func TestInvalidHashOrderAttack(t *testing.T) { t.Fatalf("failed to synchronise blocks: %v", err) } } + +// Tests that if a malicious peer makes up a random hash chain and tries to push +// indefinitely, it actually gets caught with it. +func TestMadeupHashChainAttack(t *testing.T) { + blockTTL = 100 * time.Millisecond + crossCheckCycle = 25 * time.Millisecond + + // Create a long chain of hashes without backing blocks + hashes := createHashes(0, 1024*blockCacheLimit) + hashes = hashes[:len(hashes)-1] + + // Try and sync with the malicious node and check that it fails + tester := newTester(t, hashes, nil) + tester.newPeer("attack", big.NewInt(10000), hashes[0]) + if _, err := tester.syncTake("attack", hashes[0]); err != ErrCrossCheckFailed { + t.Fatalf("synchronisation error mismatch: have %v, want %v", err, ErrCrossCheckFailed) + } +} diff --git a/eth/downloader/queue.go b/eth/downloader/queue.go index aa48c521a..13ec9a520 100644 --- a/eth/downloader/queue.go +++ b/eth/downloader/queue.go @@ -123,13 +123,13 @@ func (q *queue) Has(hash common.Hash) bool { } // Insert adds a set of hashes for the download queue for scheduling, returning -// the number of new hashes encountered. -func (q *queue) Insert(hashes []common.Hash) int { +// the new hashes encountered. +func (q *queue) Insert(hashes []common.Hash) []common.Hash { q.lock.Lock() defer q.lock.Unlock() // Insert all the hashes prioritized in the arrival order - inserts := 0 + inserts := make([]common.Hash, 0, len(hashes)) for _, hash := range hashes { // Skip anything we already have if old, ok := q.hashPool[hash]; ok { @@ -137,7 +137,9 @@ func (q *queue) Insert(hashes []common.Hash) int { continue } // Update the counters and insert the hash - q.hashCounter, inserts = q.hashCounter+1, inserts+1 + q.hashCounter = q.hashCounter + 1 + inserts = append(inserts, hash) + q.hashPool[hash] = q.hashCounter q.hashQueue.Push(hash, float32(q.hashCounter)) // Highest gets schedules first } diff --git a/eth/sync.go b/eth/sync.go index cc1fe80ea..aa7ebc77b 100644 --- a/eth/sync.go +++ b/eth/sync.go @@ -101,7 +101,7 @@ func (pm *ProtocolManager) synchronise(peer *peer) { case downloader.ErrBusy: glog.V(logger.Debug).Infof("Synchronisation already in progress") - case downloader.ErrTimeout, downloader.ErrBadPeer, downloader.ErrInvalidChain: + case downloader.ErrTimeout, downloader.ErrBadPeer, downloader.ErrInvalidChain, downloader.ErrCrossCheckFailed: glog.V(logger.Debug).Infof("Removing peer %v: %v", peer.id, err) pm.removePeer(peer) From b517967f8626d5a06772c488360f895c4186712f Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?P=C3=A9ter=20Szil=C3=A1gyi?= Date: Fri, 15 May 2015 13:32:21 +0300 Subject: [PATCH 7/8] eth/downloader: don't penalize for stale cross checks --- eth/downloader/downloader.go | 24 +++++++++++++++++------- 1 file changed, 17 insertions(+), 7 deletions(-) diff --git a/eth/downloader/downloader.go b/eth/downloader/downloader.go index 95dd37fd7..2807a4bcf 100644 --- a/eth/downloader/downloader.go +++ b/eth/downloader/downloader.go @@ -63,9 +63,10 @@ type hashPack struct { type Downloader struct { mux *event.TypeMux - mu sync.RWMutex - queue *queue - peers *peerSet + 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]time.Time // Pending cross checks to verify a hash chain // Callbacks hasBlock hashCheckFn @@ -89,6 +90,7 @@ func New(mux *event.TypeMux, hasBlock hashCheckFn, getBlock getBlockFn) *Downloa mux: mux, queue: newQueue(), peers: newPeerSet(), + checks: make(map[common.Hash]time.Time), hasBlock: hasBlock, getBlock: getBlock, newPeerCh: make(chan *peer, 1), @@ -236,7 +238,6 @@ func (d *Downloader) fetchHashes(p *peer, h common.Hash) error { timeout = time.NewTimer(hashTTL) // timer to dump a non-responsive active peer attempted = make(map[string]bool) // attempted peers will help with retries - crossChecks = make(map[common.Hash]time.Time) // running cross checks and their deadline crossTicker = time.NewTicker(crossCheckCycle) // ticker to periodically check expired cross checks ) defer crossTicker.Stop() @@ -285,7 +286,7 @@ func (d *Downloader) fetchHashes(p *peer, h common.Hash) error { cross := inserts[rand.Intn(len(inserts))] glog.V(logger.Detail).Infof("Cross checking (%s) with %x", active.id, cross) - crossChecks[cross] = time.Now().Add(blockTTL) + d.checks[cross] = time.Now().Add(blockTTL) active.getBlocks([]common.Hash{cross}) // Also fetch a fresh @@ -306,11 +307,11 @@ func (d *Downloader) fetchHashes(p *peer, h common.Hash) error { continue } hash := blockPack.blocks[0].Hash() - delete(crossChecks, hash) + delete(d.checks, hash) case <-crossTicker.C: // Iterate over all the cross checks and fail the hash chain if they're not verified - for hash, deadline := range crossChecks { + for hash, deadline := range d.checks { if time.Now().After(deadline) { glog.V(logger.Debug).Infof("Cross check timeout for %x", hash) return ErrCrossCheckFailed @@ -362,7 +363,16 @@ out: 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 { From 5c1a7b965ca7901d3b185d75205419b87163a4fa Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?P=C3=A9ter=20Szil=C3=A1gyi?= Date: Fri, 15 May 2015 14:48:57 +0300 Subject: [PATCH 8/8] eth/downloader: circumvent a fake blockchain attack --- eth/downloader/downloader.go | 21 +++++----- eth/downloader/downloader_test.go | 64 +++++++++++++++++++++++-------- 2 files changed, 58 insertions(+), 27 deletions(-) diff --git a/eth/downloader/downloader.go b/eth/downloader/downloader.go index 2807a4bcf..f9bd5a635 100644 --- a/eth/downloader/downloader.go +++ b/eth/downloader/downloader.go @@ -90,7 +90,6 @@ func New(mux *event.TypeMux, hasBlock hashCheckFn, getBlock getBlockFn) *Downloa mux: mux, queue: newQueue(), peers: newPeerSet(), - checks: make(map[common.Hash]time.Time), hasBlock: hasBlock, getBlock: getBlock, newPeerCh: make(chan *peer, 1), @@ -160,6 +159,7 @@ func (d *Downloader) Synchronise(id string, hash common.Hash) error { // Reset the queue and peer set to clean any internal leftover state d.queue.Reset() d.peers.Reset() + d.checks = make(map[common.Hash]time.Time) // Retrieve the origin peer and initiate the downloading process p := d.peers.Peer(id) @@ -184,7 +184,7 @@ func (d *Downloader) syncWithPeer(p *peer, hash common.Hash) (err error) { defer func() { // reset on error if err != nil { - d.queue.Reset() + d.Cancel() d.mux.Post(FailedEvent{err}) } else { d.mux.Post(DoneEvent{}) @@ -259,8 +259,6 @@ func (d *Downloader) fetchHashes(p *peer, h common.Hash) error { // 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) - d.queue.Reset() - return errEmptyHashSet } // Determine if we're done fetching hashes (queue up all pending), and continue if not done @@ -277,8 +275,6 @@ func (d *Downloader) fetchHashes(p *peer, h common.Hash) error { 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) - d.queue.Reset() - return ErrBadPeer } if !done { @@ -306,8 +302,13 @@ func (d *Downloader) fetchHashes(p *peer, h common.Hash) error { if blockPack.peerId != active.id || len(blockPack.blocks) != 1 { continue } - hash := blockPack.blocks[0].Hash() - delete(d.checks, hash) + block := blockPack.blocks[0] + if _, ok := d.checks[block.Hash()]; ok { + if !d.queue.Has(block.ParentHash()) { + 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 @@ -334,7 +335,6 @@ func (d *Downloader) fetchHashes(p *peer, h common.Hash) error { // if all peers have been tried, abort the process entirely or if the hash is // the zero hash. if p == nil || (head == common.Hash{}) { - d.queue.Reset() return ErrTimeout } // set p to the active peer. this will invalidate any hashes that may be returned @@ -380,7 +380,6 @@ out: 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 - d.queue.Reset() return err } // Peer did deliver, but some blocks were off, penalize @@ -414,7 +413,6 @@ out: } // After removing bad peers make sure we actually have sufficient peer left to keep downloading if d.peers.Len() == 0 { - d.queue.Reset() return errNoPeers } // If there are unrequested hashes left start fetching @@ -448,7 +446,6 @@ out: // 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 { - d.queue.Reset() return errPeersUnavailable } diff --git a/eth/downloader/downloader_test.go b/eth/downloader/downloader_test.go index 60dcc06cd..d55664314 100644 --- a/eth/downloader/downloader_test.go +++ b/eth/downloader/downloader_test.go @@ -23,25 +23,26 @@ func createHashes(start, amount int) (hashes []common.Hash) { for i := range hashes[:len(hashes)-1] { binary.BigEndian.PutUint64(hashes[i][:8], uint64(i+2)) } - return } -func createBlock(i int, prevHash, hash common.Hash) *types.Block { +func createBlock(i int, parent, hash common.Hash) *types.Block { header := &types.Header{Number: big.NewInt(int64(i))} block := types.NewBlockWithHeader(header) block.HeaderHash = hash - block.ParentHeaderHash = prevHash + block.ParentHeaderHash = parent return block } func createBlocksFromHashes(hashes []common.Hash) map[common.Hash]*types.Block { blocks := make(map[common.Hash]*types.Block) - - for i, hash := range hashes { - blocks[hash] = createBlock(len(hashes)-i, knownHash, hash) + for i := 0; i < len(hashes); i++ { + parent := knownHash + if i < len(hashes)-1 { + parent = hashes[i+1] + } + blocks[hashes[i]] = createBlock(len(hashes)-i, parent, hashes[i]) } - return blocks } @@ -136,6 +137,7 @@ func (dl *downloadTester) getHashes(head common.Hash) error { hashes := make([]common.Hash, 0, maxHashFetch) for i, hash := range dl.hashes { if hash == head { + i++ for len(hashes) < cap(hashes) && i < len(dl.hashes) { hashes = append(hashes, dl.hashes[i]) i++ @@ -144,9 +146,11 @@ func (dl *downloadTester) getHashes(head common.Hash) error { } } // Delay delivery a bit to allow attacks to unfold - time.Sleep(time.Millisecond) - - dl.downloader.DeliverHashes(dl.activePeerId, hashes) + id := dl.activePeerId + go func() { + time.Sleep(time.Millisecond) + dl.downloader.DeliverHashes(id, hashes) + }() return nil } @@ -424,12 +428,15 @@ func TestInvalidHashOrderAttack(t *testing.T) { hashes := createHashes(0, 4*blockCacheLimit) blocks := createBlocksFromHashes(hashes) + chunk1 := make([]common.Hash, blockCacheLimit) + chunk2 := make([]common.Hash, blockCacheLimit) + copy(chunk1, hashes[blockCacheLimit:2*blockCacheLimit]) + copy(chunk2, hashes[2*blockCacheLimit:3*blockCacheLimit]) + reverse := make([]common.Hash, len(hashes)) copy(reverse, hashes) - - for i := len(hashes) / 4; i < 2*len(hashes)/4; i++ { - reverse[i], reverse[len(hashes)-i-1] = reverse[len(hashes)-i-1], reverse[i] - } + copy(reverse[2*blockCacheLimit:], chunk1) + copy(reverse[blockCacheLimit:], chunk2) // Try and sync with the malicious node and check that it fails tester := newTester(t, reverse, blocks) @@ -453,7 +460,6 @@ func TestMadeupHashChainAttack(t *testing.T) { // Create a long chain of hashes without backing blocks hashes := createHashes(0, 1024*blockCacheLimit) - hashes = hashes[:len(hashes)-1] // Try and sync with the malicious node and check that it fails tester := newTester(t, hashes, nil) @@ -462,3 +468,31 @@ func TestMadeupHashChainAttack(t *testing.T) { t.Fatalf("synchronisation error mismatch: have %v, want %v", err, ErrCrossCheckFailed) } } + +// Tests that if a malicious peer makes up a random block chain, and tried to +// push indefinitely, it actually gets caught with it. +func TestMadeupBlockChainAttack(t *testing.T) { + blockTTL = 100 * time.Millisecond + crossCheckCycle = 25 * time.Millisecond + + // Create a long chain of blocks and simulate an invalid chain by dropping every second + hashes := createHashes(0, 32*blockCacheLimit) + blocks := createBlocksFromHashes(hashes) + + gapped := make([]common.Hash, len(hashes)/2) + for i := 0; i < len(gapped); i++ { + gapped[i] = hashes[2*i] + } + // Try and sync with the malicious node and check that it fails + tester := newTester(t, gapped, blocks) + tester.newPeer("attack", big.NewInt(10000), gapped[0]) + if _, err := tester.syncTake("attack", gapped[0]); err != ErrCrossCheckFailed { + t.Fatalf("synchronisation error mismatch: have %v, want %v", err, ErrCrossCheckFailed) + } + // Ensure that a valid chain can still pass sync + tester.hashes = hashes + tester.newPeer("valid", big.NewInt(20000), hashes[0]) + if _, err := tester.syncTake("valid", hashes[0]); err != nil { + t.Fatalf("failed to synchronise blocks: %v", err) + } +}