eth: port the synchronisation algo to eth/62

This commit is contained in:
Péter Szilágyi
2015-08-14 21:25:41 +03:00
parent ca88e18f59
commit 47a7fe5d22
10 changed files with 2012 additions and 343 deletions

View File

@ -26,12 +26,10 @@ import (
"time"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core"
"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"
"gopkg.in/fatih/set.v0"
)
const (
@ -40,40 +38,44 @@ const (
)
var (
MinHashFetch = 512 // Minimum amount of hashes to not consider a peer stalling
MaxHashFetch = 512 // Amount of hashes to be fetched per retrieval request
MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request
MaxHeaderFetch = 256 // Amount of block headers 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
MaxStateFetch = 384 // Amount of node state values to allow fetching per request
MaxReceiptsFetch = 384 // Amount of transaction receipts to allow fetching per request
hashTTL = 5 * time.Second // Time it takes for a hash request to time out
blockSoftTTL = 3 * time.Second // Request completion threshold for increasing or decreasing a peer's bandwidth
blockHardTTL = 3 * blockSoftTTL // Maximum time allowance before a block request is considered expired
crossCheckCycle = time.Second // Period after which to check for expired cross checks
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
maxQueuedHashes = 256 * 1024 // Maximum number of hashes to queue for import (DOS protection)
maxBannedHashes = 4096 // Number of bannable hashes before phasing old ones out
maxBlockProcess = 256 // Number of blocks to import at once into the chain
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)
maxBlockProcess = 256 // Number of blocks to import at once into the chain
)
var (
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")
errBannedHead = errors.New("peer head hash already banned")
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 canceled (requested)")
errCancelBlockFetch = errors.New("block downloading canceled (requested)")
errNoSyncActive = errors.New("no sync active")
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")
errEmptyHeaderSet = errors.New("empty header 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")
errInvalidBody = errors.New("retrieved block body is invalid")
errCancelHashFetch = errors.New("hash fetching canceled (requested)")
errCancelBlockFetch = errors.New("block downloading canceled (requested)")
errCancelHeaderFetch = errors.New("block header fetching canceled (requested)")
errCancelBodyFetch = errors.New("block body downloading canceled (requested)")
errNoSyncActive = errors.New("no sync active")
)
// hashCheckFn is a callback type for verifying a hash's presence in the local chain.
@ -91,28 +93,36 @@ type chainInsertFn func(types.Blocks) (int, error)
// peerDropFn is a callback type for dropping a peer detected as malicious.
type peerDropFn func(id string)
type blockPack struct {
peerId string
blocks []*types.Block
}
// hashPack is a batch of block hashes returned by a peer (eth/61).
type hashPack struct {
peerId string
hashes []common.Hash
}
type crossCheck struct {
expire time.Time
parent common.Hash
// blockPack is a batch of blocks returned by a peer (eth/61).
type blockPack struct {
peerId string
blocks []*types.Block
}
// headerPack is a batch of block headers returned by a peer.
type headerPack struct {
peerId string
headers []*types.Header
}
// bodyPack is a batch of block bodies returned by a peer.
type bodyPack struct {
peerId string
transactions [][]*types.Transaction
uncles [][]*types.Header
}
type Downloader struct {
mux *event.TypeMux
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
banned *set.Set // Set of hashes we've received and banned
queue *queue // Scheduler for selecting the hashes to download
peers *peerSet // Set of active peers from which download can proceed
interrupt int32 // Atomic boolean to signal termination
@ -137,12 +147,18 @@ type Downloader struct {
// Channels
newPeerCh chan *peer
hashCh chan hashPack // Channel receiving inbound hashes
blockCh chan blockPack // Channel receiving inbound blocks
processCh chan bool // Channel to signal the block fetcher of new or finished work
hashCh chan hashPack // [eth/61] Channel receiving inbound hashes
blockCh chan blockPack // [eth/61] Channel receiving inbound blocks
headerCh chan headerPack // [eth/62] Channel receiving inbound block headers
bodyCh chan bodyPack // [eth/62] Channel receiving inbound block bodies
processCh chan bool // Channel to signal the block fetcher of new or finished work
cancelCh chan struct{} // Channel to cancel mid-flight syncs
cancelLock sync.RWMutex // Lock to protect the cancel channel in delivers
// Testing hooks
bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch
chainInsertHook func([]*Block) // Method to call upon inserting a chain of blocks (possibly in multiple invocations)
}
// Block is an origin-tagged blockchain block.
@ -153,8 +169,7 @@ type Block struct {
// New creates a new downloader to fetch hashes and blocks from remote peers.
func New(mux *event.TypeMux, hasBlock hashCheckFn, getBlock blockRetrievalFn, headBlock headRetrievalFn, insertChain chainInsertFn, dropPeer peerDropFn) *Downloader {
// Create the base downloader
downloader := &Downloader{
return &Downloader{
mux: mux,
queue: newQueue(),
peers: newPeerSet(),
@ -166,14 +181,10 @@ func New(mux *event.TypeMux, hasBlock hashCheckFn, getBlock blockRetrievalFn, he
newPeerCh: make(chan *peer, 1),
hashCh: make(chan hashPack, 1),
blockCh: make(chan blockPack, 1),
headerCh: make(chan headerPack, 1),
bodyCh: make(chan bodyPack, 1),
processCh: make(chan bool, 1),
}
// Inject all the known bad hashes
downloader.banned = set.New()
for hash, _ := range core.BadHashes {
downloader.banned.Add(hash)
}
return downloader
}
// Stats retrieves the current status of the downloader.
@ -206,15 +217,12 @@ func (d *Downloader) Synchronising() bool {
// 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, version int, head common.Hash, getRelHashes relativeHashFetcherFn, getAbsHashes absoluteHashFetcherFn, getBlocks blockFetcherFn) error {
// If the peer wants to send a banned hash, reject
if d.banned.Has(head) {
glog.V(logger.Debug).Infoln("Register rejected, head hash banned:", id)
return errBannedHead
}
// Otherwise try to construct and register the peer
func (d *Downloader) RegisterPeer(id string, version int, head common.Hash,
getRelHashes relativeHashFetcherFn, getAbsHashes absoluteHashFetcherFn, getBlocks blockFetcherFn, // eth/61 callbacks, remove when upgrading
getRelHeaders relativeHeaderFetcherFn, getAbsHeaders absoluteHeaderFetcherFn, getBlockBodies blockBodyFetcherFn) error {
glog.V(logger.Detail).Infoln("Registering peer", id)
if err := d.peers.Register(newPeer(id, version, head, getRelHashes, getAbsHashes, getBlocks)); err != nil {
if err := d.peers.Register(newPeer(id, version, head, getRelHashes, getAbsHashes, getBlocks, getRelHeaders, getAbsHeaders, getBlockBodies)); err != nil {
glog.V(logger.Error).Infoln("Register failed:", err)
return err
}
@ -235,7 +243,7 @@ func (d *Downloader) UnregisterPeer(id string) error {
// 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.
func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int) {
glog.V(logger.Detail).Infof("Attempting synchronisation: %v, head 0x%x, TD %v", id, head[:4], td)
glog.V(logger.Detail).Infof("Attempting synchronisation: %v, head [%x…], TD %v", id, head[:4], td)
switch err := d.synchronise(id, head, td); err {
case nil:
@ -244,7 +252,7 @@ func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int) {
case errBusy:
glog.V(logger.Detail).Infof("Synchronisation already in progress")
case errTimeout, errBadPeer, errStallingPeer, errBannedHead, errEmptyHashSet, errPeersUnavailable, errInvalidChain, errCrossCheckFailed:
case errTimeout, errBadPeer, errStallingPeer, errEmptyHashSet, errEmptyHeaderSet, errPeersUnavailable, errInvalidChain:
glog.V(logger.Debug).Infof("Removing peer %v: %v", id, err)
d.dropPeer(id)
@ -270,10 +278,6 @@ func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int) error
}
defer atomic.StoreInt32(&d.synchronising, 0)
// If the head hash is banned, terminate immediately
if d.banned.Has(hash) {
return errBannedHead
}
// 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")
@ -285,7 +289,6 @@ func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int) 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]*crossCheck)
// Create cancel channel for aborting mid-flight
d.cancelLock.Lock()
@ -320,17 +323,37 @@ func (d *Downloader) syncWithPeer(p *peer, hash common.Hash, td *big.Int) (err e
}
}()
glog.V(logger.Debug).Infof("Synchronizing with the network using: %s, eth/%d", p.id, p.version)
switch p.version {
case eth61:
glog.V(logger.Debug).Infof("Synchronising with the network using: %s [eth/%d]", p.id, p.version)
defer glog.V(logger.Debug).Infof("Synchronisation terminated")
switch {
case p.version == eth61:
// Old eth/61, use forward, concurrent hash and block retrieval algorithm
number, err := d.findAncestor61(p)
if err != nil {
return err
}
errc := make(chan error, 2)
go func() { errc <- d.fetchHashes61(p, td, number+1) }()
go func() { errc <- d.fetchBlocks61(number + 1) }()
// If any fetcher fails, cancel the other
if err := <-errc; err != nil {
d.cancel()
<-errc
return err
}
return <-errc
case p.version >= eth62:
// New eth/62, use forward, concurrent header and block body retrieval algorithm
number, err := d.findAncestor(p)
if err != nil {
return err
}
errc := make(chan error, 2)
go func() { errc <- d.fetchHashes(p, td, number+1) }()
go func() { errc <- d.fetchBlocks(number + 1) }()
go func() { errc <- d.fetchHeaders(p, td, number+1) }()
go func() { errc <- d.fetchBodies(number + 1) }()
// If any fetcher fails, cancel the other
if err := <-errc; err != nil {
@ -373,17 +396,17 @@ func (d *Downloader) Terminate() {
d.cancel()
}
// findAncestor tries to locate the common ancestor block of the local chain and
// findAncestor61 tries to locate the common ancestor block of the local chain and
// 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.
// In the rare scenario when we ended up on a long soft fork (i.e. none of the
// head blocks match), we do a binary search to find the common ancestor.
func (d *Downloader) findAncestor(p *peer) (uint64, error) {
// 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) {
glog.V(logger.Debug).Infof("%v: looking for common ancestor", p)
// Request out head blocks to short circuit ancestor location
head := d.headBlock().NumberU64()
from := int64(head) - int64(MaxHashFetch)
from := int64(head) - int64(MaxHashFetch) + 1
if from < 0 {
from = 0
}
@ -422,6 +445,12 @@ func (d *Downloader) findAncestor(p *peer) (uint64, error) {
case <-d.blockCh:
// Out of bounds blocks received, ignore them
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 <-timeout:
glog.V(logger.Debug).Infof("%v: head hash timeout", p)
return 0, errTimeout
@ -429,7 +458,7 @@ func (d *Downloader) findAncestor(p *peer) (uint64, error) {
}
// 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])
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
@ -468,7 +497,7 @@ func (d *Downloader) findAncestor(p *peer) (uint64, error) {
break
}
if block.NumberU64() != check {
glog.V(logger.Debug).Infof("%v: non requested hash #%d [%x], instead of #%d", p, block.NumberU64(), block.Hash().Bytes()[:4], check)
glog.V(logger.Debug).Infof("%v: non requested hash #%d [%x], instead of #%d", p, block.NumberU64(), block.Hash().Bytes()[:4], check)
return 0, errBadPeer
}
start = check
@ -476,6 +505,12 @@ func (d *Downloader) findAncestor(p *peer) (uint64, error) {
case <-d.blockCh:
// Out of bounds blocks received, ignore them
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 <-timeout:
glog.V(logger.Debug).Infof("%v: search hash timeout", p)
return 0, errTimeout
@ -485,9 +520,9 @@ func (d *Downloader) findAncestor(p *peer) (uint64, error) {
return start, nil
}
// fetchHashes keeps retrieving hashes from the requested number, until no more
// fetchHashes61 keeps retrieving hashes from the requested number, until no more
// are returned, potentially throttling on the way.
func (d *Downloader) fetchHashes(p *peer, td *big.Int, from uint64) error {
func (d *Downloader) fetchHashes61(p *peer, td *big.Int, from uint64) error {
glog.V(logger.Debug).Infof("%v: downloading hashes from #%d", p, from)
// Create a timeout timer, and the associated hash fetcher
@ -510,6 +545,12 @@ func (d *Downloader) fetchHashes(p *peer, td *big.Int, from uint64) error {
case <-d.cancelCh:
return errCancelHashFetch
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 hashPack := <-d.hashCh:
// Make sure the active peer is giving us the hashes
if hashPack.peerId != p.id {
@ -548,7 +589,7 @@ func (d *Downloader) fetchHashes(p *peer, td *big.Int, from uint64) error {
// Otherwise insert all the new hashes, aborting in case of junk
glog.V(logger.Detail).Infof("%v: inserting %d hashes from #%d", p, len(hashPack.hashes), from)
inserts := d.queue.Insert(hashPack.hashes, true)
inserts := d.queue.Insert61(hashPack.hashes, true)
if len(inserts) != len(hashPack.hashes) {
glog.V(logger.Debug).Infof("%v: stale hashes", p)
return errBadPeer
@ -573,10 +614,10 @@ func (d *Downloader) fetchHashes(p *peer, td *big.Int, from uint64) error {
}
}
// fetchBlocks iteratively downloads the scheduled hashes, taking any available
// fetchBlocks61 iteratively downloads the scheduled hashes, taking any available
// peers, reserving a chunk of blocks for each, waiting for delivery and also
// periodically checking for timeouts.
func (d *Downloader) fetchBlocks(from uint64) error {
func (d *Downloader) fetchBlocks61(from uint64) error {
glog.V(logger.Debug).Infof("Downloading blocks from #%d", from)
defer glog.V(logger.Debug).Infof("Block download terminated")
@ -595,24 +636,30 @@ func (d *Downloader) fetchBlocks(from uint64) error {
case <-d.cancelCh:
return 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 blockPack := <-d.blockCh:
// 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, and demote in case of errors
err := d.queue.Deliver(blockPack.peerId, blockPack.blocks)
err := d.queue.Deliver61(blockPack.peerId, blockPack.blocks)
switch err {
case nil:
// If no blocks were delivered, demote the peer (need the delivery above)
if len(blockPack.blocks) == 0 {
peer.Demote()
peer.SetIdle()
peer.SetIdle61()
glog.V(logger.Detail).Infof("%s: no blocks delivered", peer)
break
}
// All was successful, promote the peer and potentially start processing
peer.Promote()
peer.SetIdle()
peer.SetIdle61()
glog.V(logger.Detail).Infof("%s: delivered %d blocks", peer, len(blockPack.blocks))
go d.process()
@ -624,7 +671,7 @@ func (d *Downloader) fetchBlocks(from uint64) error {
// Peer probably timed out with its delivery but came through
// in the end, demote, but allow to to pull from this peer.
peer.Demote()
peer.SetIdle()
peer.SetIdle61()
glog.V(logger.Detail).Infof("%s: out of bound delivery", peer)
case errStaleDelivery:
@ -638,7 +685,7 @@ func (d *Downloader) fetchBlocks(from uint64) error {
default:
// Peer did something semi-useful, demote but keep it around
peer.Demote()
peer.SetIdle()
peer.SetIdle61()
glog.V(logger.Detail).Infof("%s: delivery partially failed: %v", peer, err)
go d.process()
}
@ -696,7 +743,7 @@ func (d *Downloader) fetchBlocks(from uint64) error {
// 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.
request := d.queue.Reserve(peer, peer.Capacity())
request := d.queue.Reserve61(peer, peer.Capacity())
if request == nil {
continue
}
@ -704,7 +751,7 @@ func (d *Downloader) fetchBlocks(from uint64) error {
glog.Infof("%s: requesting %d blocks", peer, len(request.Hashes))
}
// Fetch the chunk and make sure any errors return the hashes to the queue
if err := peer.Fetch(request); err != nil {
if err := peer.Fetch61(request); err != nil {
glog.V(logger.Error).Infof("%v: fetch failed, rescheduling", peer)
d.queue.Cancel(request)
}
@ -718,6 +765,401 @@ func (d *Downloader) fetchBlocks(from uint64) error {
}
}
// findAncestor tries to locate the common ancestor block of the local chain and
// 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.
// 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) findAncestor(p *peer) (uint64, error) {
glog.V(logger.Debug).Infof("%v: looking for common ancestor", p)
// Request our head blocks to short circuit ancestor location
head := d.headBlock().NumberU64()
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
case headerPack := <-d.headerCh:
// Discard anything not from the origin peer
if headerPack.peerId != p.id {
glog.V(logger.Debug).Infof("Received headers from incorrect peer(%s)", headerPack.peerId)
break
}
// Make sure the peer actually gave something valid
headers := headerPack.headers
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-- {
if d.hasBlock(headers[i].Hash()) {
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
case headerPack := <-d.headerCh:
// Discard anything not from the origin peer
if headerPack.peerId != p.id {
glog.V(logger.Debug).Infof("Received headers from incorrect peer(%s)", headerPack.peerId)
break
}
// Make sure the peer actually gave something valid
headers := headerPack.headers
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
block := d.getBlock(headers[0].Hash())
if block == nil {
end = check
break
}
if block.NumberU64() != check {
glog.V(logger.Debug).Infof("%v: non requested header #%d [%x…], instead of #%d", p, block.NumberU64(), block.Hash().Bytes()[:4], check)
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.
func (d *Downloader) fetchHeaders(p *peer, td *big.Int, from uint64) error {
glog.V(logger.Debug).Infof("%v: downloading headers from #%d", p, from)
defer glog.V(logger.Debug).Infof("%v: header download terminated", p)
// Create a timeout timer, and the associated hash fetcher
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)
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
case headerPack := <-d.headerCh:
// Make sure the active peer is giving us the headers
if headerPack.peerId != p.id {
glog.V(logger.Debug).Infof("Received headers from incorrect peer (%s)", headerPack.peerId)
break
}
timeout.Stop()
// If no more headers are inbound, notify the body fetcher and return
if len(headerPack.headers) == 0 {
glog.V(logger.Debug).Infof("%v: no available headers", p)
select {
case d.processCh <- false:
case <-d.cancelCh:
}
// If no headers were retrieved at all, the peer violated it's TD promise that it had a
// better chain compared to ours. The only exception is if it's promised blocks were
// already imported by other means (e.g. fecher):
//
// R <remote peer>, L <local node>: Both at block 10
// R: Mine block 11, and propagate it to L
// L: Queue block 11 for import
// L: Notice that R's head and TD increased compared to ours, start sync
// L: Import of block 11 finishes
// L: Sync begins, and finds common ancestor at 11
// L: Request new headers up from 11 (R's TD was higher, it must have something)
// R: Nothing to give
if !gotHeaders && td.Cmp(d.headBlock().Td) > 0 {
return errStallingPeer
}
return nil
}
gotHeaders = true
// Otherwise insert all the new headers, aborting in case of junk
glog.V(logger.Detail).Infof("%v: inserting %d headers from #%d", p, len(headerPack.headers), from)
inserts := d.queue.Insert(headerPack.headers)
if len(inserts) != len(headerPack.headers) {
glog.V(logger.Debug).Infof("%v: stale headers", p)
return errBadPeer
}
// Notify the block fetcher of new headers, but stop if queue is full
cont := d.queue.Pending() < maxQueuedHeaders
select {
case d.processCh <- cont:
default:
}
if !cont {
return nil
}
// Queue not yet full, fetch the next batch
from += uint64(len(headerPack.headers))
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)
d.dropPeer(p.id)
// Finish the sync gracefully instead of dumping the gathered data though
select {
case d.processCh <- false:
default:
}
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)
defer glog.V(logger.Debug).Infof("Block body download terminated")
// Create a timeout timer for scheduling expiration tasks
ticker := time.NewTicker(100 * time.Millisecond)
defer ticker.Stop()
update := make(chan struct{}, 1)
// Prepare the queue and fetch block bodies until the block header fetcher's done
d.queue.Prepare(from)
finished := false
for {
select {
case <-d.cancelCh:
return errCancelBlockFetch
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 bodyPack := <-d.bodyCh:
// 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(bodyPack.peerId); peer != nil {
// Deliver the received chunk of bodies, and demote in case of errors
err := d.queue.Deliver(bodyPack.peerId, bodyPack.transactions, bodyPack.uncles)
switch err {
case nil:
// If no blocks were delivered, demote the peer (need the delivery above)
if len(bodyPack.transactions) == 0 || len(bodyPack.uncles) == 0 {
peer.Demote()
peer.SetIdle()
glog.V(logger.Detail).Infof("%s: no block bodies delivered", peer)
break
}
// All was successful, promote the peer and potentially start processing
peer.Promote()
peer.SetIdle()
glog.V(logger.Detail).Infof("%s: delivered %d:%d block bodies", peer, len(bodyPack.transactions), len(bodyPack.uncles))
go d.process()
case errInvalidChain:
// The hash chain is invalid (blocks are not ordered properly), abort
return err
case errInvalidBody:
// The peer delivered something very bad, drop immediately
glog.V(logger.Error).Infof("%s: delivered invalid block, dropping", peer)
d.dropPeer(peer.id)
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()
peer.SetIdle()
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()
peer.SetIdle()
glog.V(logger.Detail).Infof("%s: delivery partially failed: %v", peer, err)
go d.process()
}
}
// Blocks assembled, try to update the progress
select {
case update <- struct{}{}:
default:
}
case cont := <-d.processCh:
// 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
}
// Check for block body request timeouts and demote the responsible peers
for _, pid := range d.queue.Expire(bodyHardTTL) {
if peer := d.peers.Peer(pid); peer != nil {
peer.Demote()
glog.V(logger.Detail).Infof("%s: block body delivery timeout", peer)
}
}
// If there's noting more to fetch, wait or terminate
if d.queue.Pending() == 0 {
if d.queue.InFlight() == 0 && finished {
glog.V(logger.Debug).Infof("Block body fetching completed")
return nil
}
break
}
// Send a download request to all idle peers, until throttled
queuedEmptyBlocks, throttled := false, false
for _, peer := range d.peers.IdlePeers() {
// Short circuit if throttling activated
if d.queue.Throttle() {
throttled = true
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.
request, process, err := d.queue.Reserve(peer, peer.Capacity())
if err != nil {
return err
}
if process {
queuedEmptyBlocks = true
go d.process()
}
if request == nil {
continue
}
if glog.V(logger.Detail) {
glog.Infof("%s: requesting %d block bodies", peer, len(request.Headers))
}
// Fetch the chunk and make sure any errors return the hashes to the queue
if d.bodyFetchHook != nil {
d.bodyFetchHook(request.Headers)
}
if err := peer.Fetch(request); err != nil {
glog.V(logger.Error).Infof("%v: fetch failed, rescheduling", peer)
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 !queuedEmptyBlocks && !throttled && d.queue.InFlight() == 0 {
return errPeersUnavailable
}
}
}
}
// process takes blocks from the queue and tries to import them into the chain.
//
// The algorithmic flow is as follows:
@ -763,6 +1205,9 @@ func (d *Downloader) process() {
if len(blocks) == 0 {
return
}
if d.chainInsertHook != nil {
d.chainInsertHook(blocks)
}
// Reset the import statistics
d.importLock.Lock()
d.importStart = time.Now()
@ -796,9 +1241,31 @@ func (d *Downloader) process() {
}
}
// DeliverBlocks injects a new batch of blocks received from a remote node.
// DeliverHashes61 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) DeliverHashes61(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
}
}
// DeliverBlocks61 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 {
func (d *Downloader) DeliverBlocks61(id string, blocks []*types.Block) error {
// Make sure the downloader is active
if atomic.LoadInt32(&d.synchronising) == 0 {
return errNoSyncActive
@ -817,10 +1284,9 @@ func (d *Downloader) DeliverBlocks(id string, blocks []*types.Block) error {
}
}
// 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 {
// DeliverHeaders injects a new batch of blck headers received from a remote
// node into the download schedule.
func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) error {
// Make sure the downloader is active
if atomic.LoadInt32(&d.synchronising) == 0 {
return errNoSyncActive
@ -831,7 +1297,27 @@ func (d *Downloader) DeliverHashes(id string, hashes []common.Hash) error {
d.cancelLock.RUnlock()
select {
case d.hashCh <- hashPack{id, hashes}:
case d.headerCh <- headerPack{id, headers}:
return nil
case <-cancel:
return errNoSyncActive
}
}
// DeliverBodies injects a new batch of block bodies received from a remote node.
func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) 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.bodyCh <- bodyPack{id, transactions, uncles}:
return nil
case <-cancel: