diff --git a/eth/block_pool.go b/eth/block_pool.go index 2c5ebb1e35..64d1e73fa2 100644 --- a/eth/block_pool.go +++ b/eth/block_pool.go @@ -1,96 +1,107 @@ package eth import ( - "bytes" - "fmt" "math" "math/big" + "math/rand" + "sort" "sync" "time" "github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/ethutil" - "github.com/ethereum/go-ethereum/event" ethlogger "github.com/ethereum/go-ethereum/logger" ) var poolLogger = ethlogger.NewLogger("Blockpool") const ( - blockHashesBatchSize = 256 - blockBatchSize = 64 - blockRequestInterval = 10 // seconds - blockRequestRepetition = 1 - cacheTimeout = 3 // minutes - blockTimeout = 5 // minutes + blockHashesBatchSize = 256 + blockBatchSize = 64 + blocksRequestInterval = 10 // seconds + blocksRequestRepetition = 1 + blockHashesRequestInterval = 10 // seconds + blocksRequestMaxIdleRounds = 10 + cacheTimeout = 3 // minutes + blockTimeout = 5 // minutes ) type poolNode struct { - hash []byte - block *types.Block - child *poolNode - parent *poolNode - root *nodePointer - knownParent bool - suicide chan bool - peer string - source string - blockRequestRoot bool - blockRequestControl *bool - blockRequestQuit *(chan bool) -} - -// the minimal interface for chain manager -type chainManager interface { - KnownBlock(hash []byte) bool - AddBlock(*types.Block) error - CheckPoW(*types.Block) bool + lock sync.RWMutex + hash []byte + block *types.Block + child *poolNode + parent *poolNode + section *section + knownParent bool + peer string + source string + complete bool } type BlockPool struct { - chainManager chainManager - eventer event.TypeMux - - // pool Pool - lock sync.Mutex + lock sync.RWMutex pool map[string]*poolNode - peersLock sync.Mutex + peersLock sync.RWMutex peers map[string]*peerInfo peer *peerInfo quit chan bool wg sync.WaitGroup running bool + + // the minimal interface with blockchain + hasBlock func(hash []byte) bool + insertChain func(types.Blocks) error + verifyPoW func(*types.Block) bool } type peerInfo struct { - td *big.Int - currentBlock []byte - id string + lock sync.RWMutex + + td *big.Int + currentBlock []byte + id string + requestBlockHashes func([]byte) error requestBlocks func([][]byte) error - invalidBlock func(error) + peerError func(int, string, ...interface{}) + + sections map[string]*section + roots []*poolNode + quitC chan bool } -type nodePointer struct { - hash []byte -} - -type peerChangeEvent struct { - *peerInfo -} - -func NewBlockPool(chMgr chainManager) *BlockPool { +func NewBlockPool(hasBlock func(hash []byte) bool, insertChain func(types.Blocks) error, verifyPoW func(*types.Block) bool, +) *BlockPool { return &BlockPool{ - chainManager: chMgr, - pool: make(map[string]*poolNode), - peers: make(map[string]*peerInfo), - quit: make(chan bool), - running: true, + hasBlock: hasBlock, + insertChain: insertChain, + verifyPoW: verifyPoW, } } +// allows restart +func (self *BlockPool) Start() { + self.lock.Lock() + if self.running { + self.lock.Unlock() + return + } + self.running = true + self.quit = make(chan bool) + self.pool = make(map[string]*poolNode) + self.lock.Unlock() + + self.peersLock.Lock() + self.peers = make(map[string]*peerInfo) + self.peersLock.Unlock() + + poolLogger.Infoln("Started") + +} + func (self *BlockPool) Stop() { self.lock.Lock() if !self.running { @@ -103,267 +114,46 @@ func (self *BlockPool) Stop() { poolLogger.Infoln("Stopping") close(self.quit) + self.lock.Lock() + self.peersLock.Lock() + self.peers = nil + self.pool = nil + self.peer = nil self.wg.Wait() + self.lock.Unlock() + self.peersLock.Unlock() poolLogger.Infoln("Stopped") } -// Entry point for eth protocol to add block hashes received via BlockHashesMsg -// only hashes from the best peer is handled -// this method is always responsible to initiate further hash requests until -// a known parent is reached unless cancelled by a peerChange event -func (self *BlockPool) AddBlockHashes(next func() ([]byte, bool), peerId string) { - // subscribe to peerChangeEvent before we check for best peer - peerChange := self.eventer.Subscribe(peerChangeEvent{}) - defer peerChange.Unsubscribe() - // check if this peer is the best - peer, best := self.getPeer(peerId) - if !best { - return - } - root := &nodePointer{} - // peer is still the best - hashes := make(chan []byte) - var lastPoolNode *poolNode - - // using a for select loop so that peer change (new best peer) can abort the parallel thread that processes hashes of the earlier best peer - for { - hash, ok := next() - if ok { - hashes <- hash - } else { - break - } - select { - case <-self.quit: - return - case <-peerChange.Chan(): - // remember where we left off with this peer - if lastPoolNode != nil { - root.hash = lastPoolNode.hash - go self.killChain(lastPoolNode) - } - case hash := <-hashes: - self.lock.Lock() - defer self.lock.Unlock() - // check if known block connecting the downloaded chain to our blockchain - if self.chainManager.KnownBlock(hash) { - poolLogger.Infof("known block (%x...)\n", hash[0:4]) - if lastPoolNode != nil { - lastPoolNode.knownParent = true - go self.requestBlocksLoop(lastPoolNode) - } else { - // all hashes known if topmost one is in blockchain - } - return - } - // - var currentPoolNode *poolNode - // check if lastPoolNode has the correct parent node (hash matching), - // then just assign to currentPoolNode - if lastPoolNode != nil && lastPoolNode.parent != nil && bytes.Compare(lastPoolNode.parent.hash, hash) == 0 { - currentPoolNode = lastPoolNode.parent - } else { - // otherwise look up in pool - currentPoolNode = self.pool[string(hash)] - // if node does not exist, create it and index in the pool - if currentPoolNode == nil { - currentPoolNode = &poolNode{ - hash: hash, - } - self.pool[string(hash)] = currentPoolNode - } - } - // set up parent-child nodes (doubly linked list) - self.link(currentPoolNode, lastPoolNode) - // ! we trust the node iff - // (1) node marked as by the same peer or - // (2) it has a PoW valid block retrieved - if currentPoolNode.peer == peer.id || currentPoolNode.block != nil { - // the trusted checkpoint from which we request hashes down to known head - lastPoolNode = self.pool[string(currentPoolNode.root.hash)] - break - } - currentPoolNode.peer = peer.id - currentPoolNode.root = root - lastPoolNode = currentPoolNode - } - } - // lastPoolNode is nil if and only if the node with stored root hash is already cleaned up - // after valid block insertion, therefore in this case the blockpool active chain is connected to the blockchain, so no need to request further hashes or request blocks - if lastPoolNode != nil { - root.hash = lastPoolNode.hash - peer.requestBlockHashes(lastPoolNode.hash) - go self.requestBlocksLoop(lastPoolNode) - } - return -} - -func (self *BlockPool) requestBlocksLoop(node *poolNode) { - suicide := time.After(blockTimeout * time.Minute) - requestTimer := time.After(0) - var controlChan chan bool - closedChan := make(chan bool) - quit := make(chan bool) - close(closedChan) - requestBlocks := true - origNode := node - self.lock.Lock() - node.blockRequestRoot = true - b := false - control := &b - node.blockRequestControl = control - node.blockRequestQuit = &quit - self.lock.Unlock() - blocks := 0 - self.wg.Add(1) -loop: - for { - if requestBlocks { - controlChan = closedChan - } else { - self.lock.Lock() - if *node.blockRequestControl { - controlChan = closedChan - *node.blockRequestControl = false - } - self.lock.Unlock() - } - select { - case <-quit: - break loop - case <-suicide: - go self.killChain(origNode) - break loop - - case <-requestTimer: - requestBlocks = true - - case <-controlChan: - controlChan = nil - // this iteration takes care of requesting blocks only starting from the first node with a missing block (moving target), - // max up to the next checkpoint (n.blockRequestRoot true) - nodes := []*poolNode{} - n := node - next := node - self.lock.Lock() - for n != nil && (n == node || !n.blockRequestRoot) && (requestBlocks || n.block != nil) { - if n.block != nil { - if len(nodes) == 0 { - // nil control indicates that node is not needed anymore - // block can be inserted to blockchain and deleted if knownParent - n.blockRequestControl = nil - blocks++ - next = next.child - } else { - // this is needed to indicate that when a new chain forks from an existing one - // triggering a reorg will ? renew the blockTimeout period ??? - // if there is a block but control == nil should start fetching blocks, see link function - n.blockRequestControl = control - } - } else { - nodes = append(nodes, n) - n.blockRequestControl = control - } - n = n.child - } - // if node is connected to the blockchain, we can immediately start inserting - // blocks to the blockchain and delete nodes - if node.knownParent { - go self.insertChainFrom(node) - } - if next.blockRequestRoot && next != node { - // no more missing blocks till the checkpoint, quitting - poolLogger.Debugf("fetched %v blocks on active chain, batch %v-%v", blocks, origNode, n) - break loop - } - self.lock.Unlock() - - // reset starting node to the first descendant node with missing block - node = next - if !requestBlocks { - continue - } - go self.requestBlocks(nodes) - requestTimer = time.After(blockRequestInterval * time.Second) - } - } - self.wg.Done() - return -} - -func (self *BlockPool) requestBlocks(nodes []*poolNode) { - // distribute block request among known peers - self.peersLock.Lock() - peerCount := len(self.peers) - poolLogger.Debugf("requesting %v missing blocks from %v peers", len(nodes), peerCount) - blockHashes := make([][][]byte, peerCount) - repetitions := int(math.Max(float64(peerCount)/2.0, float64(blockRequestRepetition))) - for n, node := range nodes { - for i := 0; i < repetitions; i++ { - blockHashes[n%peerCount] = append(blockHashes[n%peerCount], node.hash) - n++ - } - } - i := 0 - for _, peer := range self.peers { - peer.requestBlocks(blockHashes[i]) - i++ - } - self.peersLock.Unlock() -} - -func (self *BlockPool) insertChainFrom(node *poolNode) { - self.lock.Lock() - defer self.lock.Unlock() - for node != nil && node.blockRequestControl == nil { - err := self.chainManager.AddBlock(node.block) - if err != nil { - poolLogger.Debugf("invalid block %v", node.hash) - poolLogger.Debugf("penalise peers %v (hash), %v (block)", node.peer, node.source) - // penalise peer in node.source - go self.killChain(node) - return - } - poolLogger.Debugf("insert block %v into blockchain", node.hash) - node = node.child - } - // if block insertion succeeds, mark the child as knownParent - // trigger request blocks reorg - if node != nil { - node.knownParent = true - *(node.blockRequestControl) = true - } -} - // AddPeer is called by the eth protocol instance running on the peer after // the status message has been received with total difficulty and current block hash // AddPeer can only be used once, RemovePeer needs to be called when the peer disconnects -func (self *BlockPool) AddPeer(td *big.Int, currentBlock []byte, peerId string, requestBlockHashes func([]byte) error, requestBlocks func([][]byte) error, invalidBlock func(error)) bool { +func (self *BlockPool) AddPeer(td *big.Int, currentBlock []byte, peerId string, requestBlockHashes func([]byte) error, requestBlocks func([][]byte) error, peerError func(int, string, ...interface{})) bool { self.peersLock.Lock() defer self.peersLock.Unlock() if self.peers[peerId] != nil { panic("peer already added") } - info := &peerInfo{ + peer := &peerInfo{ td: td, currentBlock: currentBlock, id: peerId, //peer.Identity().Pubkey() requestBlockHashes: requestBlockHashes, requestBlocks: requestBlocks, - invalidBlock: invalidBlock, + peerError: peerError, } - self.peers[peerId] = info + self.peers[peerId] = peer poolLogger.Debugf("add new peer %v with td %v", peerId, td) currentTD := ethutil.Big0 if self.peer != nil { currentTD = self.peer.td } if td.Cmp(currentTD) > 0 { - self.peer = info - self.eventer.Post(peerChangeEvent{info}) + self.peer.stop(peer) + peer.start(self.peer) poolLogger.Debugf("peer %v promoted to best peer", peerId) - requestBlockHashes(currentBlock) + self.peer = peer return true } return false @@ -373,14 +163,15 @@ func (self *BlockPool) AddPeer(td *big.Int, currentBlock []byte, peerId string, func (self *BlockPool) RemovePeer(peerId string) { self.peersLock.Lock() defer self.peersLock.Unlock() - if self.peers[peerId] != nil { - panic("peer already removed") + peer := self.peers[peerId] + if peer == nil { + return } self.peers[peerId] = nil poolLogger.Debugf("remove peer %v", peerId[0:4]) // if current best peer is removed, need find a better one - if peerId == self.peer.id { + if self.peer != nil && peerId == self.peer.id { var newPeer *peerInfo max := ethutil.Big0 // peer with the highest self-acclaimed TD is chosen @@ -390,21 +181,499 @@ func (self *BlockPool) RemovePeer(peerId string) { newPeer = info } } - self.peer = newPeer - self.eventer.Post(peerChangeEvent{newPeer}) + self.peer.stop(peer) + peer.start(self.peer) if newPeer != nil { - poolLogger.Debugf("peer %v with td %v spromoted to best peer", newPeer.id[0:4], newPeer.td) - newPeer.requestBlockHashes(newPeer.currentBlock) + poolLogger.Debugf("peer %v with td %v promoted to best peer", newPeer.id[0:4], newPeer.td) } else { poolLogger.Warnln("no peers left") } } } -func (self *BlockPool) getPeer(peerId string) (*peerInfo, bool) { +// Entry point for eth protocol to add block hashes received via BlockHashesMsg +// only hashes from the best peer is handled +// this method is always responsible to initiate further hash requests until +// a known parent is reached unless cancelled by a peerChange event +// this process also launches all request processes on each chain section +// this function needs to run asynchronously for one peer since the message is discarded??? +func (self *BlockPool) AddBlockHashes(next func() ([]byte, bool), peerId string) { + + // check if this peer is the best + peer, best := self.getPeer(peerId) + if !best { + return + } + // peer is still the best + + var child *poolNode + var depth int + + // iterate using next (rlp stream lazy decoder) feeding hashesC + self.wg.Add(1) + go func() { + for { + select { + case <-self.quit: + return + case <-peer.quitC: + // if the peer is demoted, no more hashes taken + break + default: + hash, ok := next() + if !ok { + // message consumed chain skeleton built + break + } + // check if known block connecting the downloaded chain to our blockchain + if self.hasBlock(hash) { + poolLogger.Infof("known block (%x...)\n", hash[0:4]) + if child != nil { + child.Lock() + // mark child as absolute pool root with parent known to blockchain + child.knownParent = true + child.Unlock() + } + break + } + // + var parent *poolNode + // look up node in pool + parent = self.get(hash) + if parent != nil { + // reached a known chain in the pool + // request blocks on the newly added part of the chain + if child != nil { + self.link(parent, child) + + // activate the current chain + self.activateChain(parent, peer, true) + poolLogger.Debugf("potential chain of %v blocks added, reached blockpool, activate chain", depth) + break + } + // if this is the first hash, we expect to find it + parent.RLock() + grandParent := parent.parent + parent.RUnlock() + if grandParent != nil { + // activate the current chain + self.activateChain(parent, peer, true) + poolLogger.Debugf("block hash found, activate chain") + break + } + // the first node is the root of a chain in the pool, rejoice and continue + } + // if node does not exist, create it and index in the pool + section := §ion{} + if child == nil { + section.top = parent + } + parent = &poolNode{ + hash: hash, + child: child, + section: section, + peer: peerId, + } + self.set(hash, parent) + poolLogger.Debugf("create potential block for %x...", hash[0:4]) + + depth++ + child = parent + } + } + if child != nil { + poolLogger.Debugf("chain of %v hashes added", depth) + // start a processSection on the last node, but switch off asking + // hashes and blocks until next peer confirms this chain + section := self.processSection(child) + peer.addSection(child.hash, section) + section.start() + } + }() +} + +// AddBlock is the entry point for the eth protocol when blockmsg is received upon requests +// It has a strict interpretation of the protocol in that if the block received has not been requested, it results in an error (which can be ignored) +// block is checked for PoW +// only the first PoW-valid block for a hash is considered legit +func (self *BlockPool) AddBlock(block *types.Block, peerId string) { + hash := block.Hash() + node := self.get(hash) + node.RLock() + b := node.block + node.RUnlock() + if b != nil { + return + } + if node == nil && !self.hasBlock(hash) { + self.peerError(peerId, ErrUnrequestedBlock, "%x", hash) + return + } + // validate block for PoW + if !self.verifyPoW(block) { + self.peerError(peerId, ErrInvalidPoW, "%x", hash) + } + node.Lock() + node.block = block + node.source = peerId + node.Unlock() +} + +// iterates down a known poolchain and activates fetching processes +// on each chain section for the peer +// stops if the peer is demoted +// registers last section root as root for the peer (in case peer is promoted a second time, to remember) +func (self *BlockPool) activateChain(node *poolNode, peer *peerInfo, on bool) { + self.wg.Add(1) + go func() { + for { + node.sectionRLock() + bottom := node.section.bottom + if bottom == nil { // the chain section is being created or killed + break + } + // register this section with the peer + if peer != nil { + peer.addSection(bottom.hash, bottom.section) + if on { + bottom.section.start() + } else { + bottom.section.start() + } + } + if bottom.parent == nil { + node = bottom + break + } + // if peer demoted stop activation + select { + case <-peer.quitC: + break + default: + } + + node = bottom.parent + bottom.sectionRUnlock() + } + // remember root for this peer + peer.addRoot(node) + self.wg.Done() + }() +} + +// main worker thread on each section in the poolchain +// - kills the section if there are blocks missing after an absolute time +// - kills the section if there are maxIdleRounds of idle rounds of block requests with no response +// - periodically polls the chain section for missing blocks which are then requested from peers +// - registers the process controller on the peer so that if the peer is promoted as best peer the second time (after a disconnect of a better one), all active processes are switched back on unless they expire and killed () +// - when turned off (if peer disconnects and new peer connects with alternative chain), no blockrequests are made but absolute expiry timer is ticking +// - when turned back on it recursively calls itself on the root of the next chain section +// - when exits, signals to +func (self *BlockPool) processSection(node *poolNode) *section { + // absolute time after which sub-chain is killed if not complete (some blocks are missing) + suicideTimer := time.After(blockTimeout * time.Minute) + var blocksRequestTimer, blockHashesRequestTimer <-chan time.Time + var nodeC, missingC, processC chan *poolNode + controlC := make(chan bool) + resetC := make(chan bool) + var hashes [][]byte + var i, total, missing, lastMissing, depth int + var blockHashesRequests, blocksRequests int + var idle int + var init, alarm, done, same, running, once bool + orignode := node + hash := node.hash + + node.sectionLock() + defer node.sectionUnlock() + section := §ion{controlC: controlC, resetC: resetC} + node.section = section + + go func() { + self.wg.Add(1) + for { + node.sectionRLock() + controlC = node.section.controlC + node.sectionRUnlock() + + if init { + // missing blocks read from nodeC + // initialized section + if depth == 0 { + break + } + // enable select case to read missing block when ready + processC = missingC + missingC = make(chan *poolNode, lastMissing) + nodeC = nil + // only do once + init = false + } else { + if !once { + missingC = nil + processC = nil + i = 0 + total = 0 + lastMissing = 0 + } + } + + // went through all blocks in section + if i != 0 && i == lastMissing { + if len(hashes) > 0 { + // send block requests to peers + self.requestBlocks(blocksRequests, hashes) + } + blocksRequests++ + poolLogger.Debugf("[%x] block request attempt %v: missing %v/%v/%v", hash[0:4], blocksRequests, missing, total, depth) + if missing == lastMissing { + // idle round + if same { + // more than once + idle++ + // too many idle rounds + if idle > blocksRequestMaxIdleRounds { + poolLogger.Debugf("[%x] block requests had %v idle rounds (%v total attempts): missing %v/%v/%v\ngiving up...", hash[0:4], idle, blocksRequests, missing, total, depth) + self.killChain(node, nil) + break + } + } else { + idle = 0 + } + same = true + } else { + if missing == 0 { + // no missing nodes + poolLogger.Debugf("block request process complete on section %x... (%v total blocksRequests): missing %v/%v/%v", hash[0:4], blockHashesRequests, blocksRequests, missing, total, depth) + node.Lock() + orignode.complete = true + node.Unlock() + blocksRequestTimer = nil + if blockHashesRequestTimer == nil { + // not waiting for hashes any more + poolLogger.Debugf("hash request on root %x... successful (%v total attempts)\nquitting...", hash[0:4], blockHashesRequests) + break + } // otherwise suicide if no hashes coming + } + same = false + } + lastMissing = missing + i = 0 + missing = 0 + // ready for next round + done = true + } + if done && alarm { + poolLogger.Debugf("start checking if new blocks arrived (attempt %v): missing %v/%v/%v", blocksRequests, missing, total, depth) + blocksRequestTimer = time.After(blocksRequestInterval * time.Second) + alarm = false + done = false + // processC supposed to be empty and never closed so just swap, no need to allocate + tempC := processC + processC = missingC + missingC = tempC + } + select { + case <-self.quit: + break + case <-suicideTimer: + self.killChain(node, nil) + poolLogger.Warnf("[%x] timeout. (%v total attempts): missing %v/%v/%v", hash[0:4], blocksRequests, missing, total, depth) + break + case <-blocksRequestTimer: + alarm = true + case <-blockHashesRequestTimer: + orignode.RLock() + parent := orignode.parent + orignode.RUnlock() + if parent != nil { + // if not root of chain, switch off + poolLogger.Debugf("[%x] parent found, hash requests deactivated (after %v total attempts)\n", hash[0:4], blockHashesRequests) + blockHashesRequestTimer = nil + } else { + blockHashesRequests++ + poolLogger.Debugf("[%x] hash request on root (%v total attempts)\n", hash[0:4], blockHashesRequests) + self.requestBlockHashes(parent.hash) + blockHashesRequestTimer = time.After(blockHashesRequestInterval * time.Second) + } + case r, ok := <-controlC: + if !ok { + break + } + if running && !r { + poolLogger.Debugf("process on section %x... (%v total attempts): missing %v/%v/%v", hash[0:4], blocksRequests, missing, total, depth) + + alarm = false + blocksRequestTimer = nil + blockHashesRequestTimer = nil + processC = nil + } + if !running && r { + poolLogger.Debugf("[%x] on", hash[0:4]) + + orignode.RLock() + parent := orignode.parent + complete := orignode.complete + knownParent := orignode.knownParent + orignode.RUnlock() + if !complete { + poolLogger.Debugf("[%x] activate block requests", hash[0:4]) + blocksRequestTimer = time.After(0) + } + if parent == nil && !knownParent { + // if no parent but not connected to blockchain + poolLogger.Debugf("[%x] activate block hashes requests", hash[0:4]) + blockHashesRequestTimer = time.After(0) + } else { + blockHashesRequestTimer = nil + } + alarm = true + processC = missingC + if !once { + // if not run at least once fully, launch iterator + processC = make(chan *poolNode) + missingC = make(chan *poolNode) + self.foldUp(orignode, processC) + once = true + } + } + total = lastMissing + case <-resetC: + once = false + init = false + done = false + case node, ok := <-processC: + if !ok { + // channel closed, first iteration finished + init = true + once = true + continue + } + i++ + // if node has no block + node.RLock() + block := node.block + nhash := node.hash + knownParent := node.knownParent + node.RUnlock() + if !init { + depth++ + } + if block == nil { + missing++ + if !init { + total++ + } + hashes = append(hashes, nhash) + if len(hashes) == blockBatchSize { + self.requestBlocks(blocksRequests, hashes) + hashes = nil + } + missingC <- node + } else { + // block is found + if knownParent { + // connected to the blockchain, insert the longest chain of blocks + var blocks types.Blocks + child := node + parent := node + node.sectionRLock() + for child != nil && child.block != nil { + parent = child + blocks = append(blocks, parent.block) + child = parent.child + } + node.sectionRUnlock() + poolLogger.Debugf("[%x] insert %v blocks into blockchain", hash[0:4], len(blocks)) + if err := self.insertChain(blocks); err != nil { + // TODO: not clear which peer we need to address + // peerError should dispatch to peer if still connected and disconnect + self.peerError(node.source, ErrInvalidBlock, "%v", err) + poolLogger.Debugf("invalid block %v", node.hash) + poolLogger.Debugf("penalise peers %v (hash), %v (block)", node.peer, node.source) + // penalise peer in node.source + self.killChain(node, nil) + // self.disconnect() + break + } + // if suceeded mark the next one (no block yet) as connected to blockchain + if child != nil { + child.Lock() + child.knownParent = true + child.Unlock() + } + // reset starting node to first node with missing block + orignode = child + // pop the inserted ancestors off the channel + for i := 1; i < len(blocks); i++ { + <-processC + } + // delink inserted chain section + self.killChain(node, parent) + } + } + } + } + poolLogger.Debugf("[%x] quit after\n%v block hashes requests\n%v block requests: missing %v/%v/%v", hash[0:4], blockHashesRequests, blocksRequests, missing, total, depth) + + self.wg.Done() + node.sectionLock() + node.section.controlC = nil + node.sectionUnlock() + // this signals that controller not available + }() + return section + +} + +func (self *BlockPool) peerError(peerId string, code int, format string, params ...interface{}) { + self.peersLock.RLock() + defer self.peersLock.RUnlock() + peer, ok := self.peers[peerId] + if ok { + peer.peerError(code, format, params...) + } +} + +func (self *BlockPool) requestBlockHashes(hash []byte) { self.peersLock.Lock() defer self.peersLock.Unlock() - if self.peer.id == peerId { + if self.peer != nil { + self.peer.requestBlockHashes(hash) + } +} + +func (self *BlockPool) requestBlocks(attempts int, hashes [][]byte) { + // distribute block request among known peers + self.peersLock.Lock() + defer self.peersLock.Unlock() + peerCount := len(self.peers) + // on first attempt use the best peer + if attempts == 0 { + self.peer.requestBlocks(hashes) + return + } + repetitions := int(math.Min(float64(peerCount), float64(blocksRequestRepetition))) + poolLogger.Debugf("request %v missing blocks from %v/%v peers", len(hashes), repetitions, peerCount) + i := 0 + indexes := rand.Perm(peerCount)[0:(repetitions - 1)] + sort.Ints(indexes) + for _, peer := range self.peers { + if i == indexes[0] { + peer.requestBlocks(hashes) + indexes = indexes[1:] + if len(indexes) == 0 { + break + } + } + i++ + } +} + +func (self *BlockPool) getPeer(peerId string) (*peerInfo, bool) { + self.peersLock.RLock() + defer self.peersLock.RUnlock() + if self.peer != nil && self.peer.id == peerId { return self.peer, true } info, ok := self.peers[peerId] @@ -414,101 +683,332 @@ func (self *BlockPool) getPeer(peerId string) (*peerInfo, bool) { return info, false } -// if same peer gave different chain before, this will overwrite it -// if currentPoolNode existed as a non-leaf node the earlier fork is delinked -// if same parent hash is found, we can abort, we do not allow the same peer to change minds about parent of same hash, if errored first time round, will get penalized. -// if lastPoolNode had a different parent the earlier parent (with entire subtree) is delinked, this situation cannot normally arise though -// just in case reset lastPoolNode as non-root (unlikely) +func (self *peerInfo) addSection(hash []byte, section *section) { + self.lock.Lock() + defer self.lock.Unlock() + self.sections[string(hash)] = section +} +func (self *peerInfo) addRoot(node *poolNode) { + self.lock.Lock() + defer self.lock.Unlock() + self.roots = append(self.roots, node) +} + +// (re)starts processes registered for this peer (self) +func (self *peerInfo) start(peer *peerInfo) { + self.lock.Lock() + defer self.lock.Unlock() + self.quitC = make(chan bool) + for _, root := range self.roots { + root.sectionRLock() + if root.section.bottom != nil { + if root.parent == nil { + self.requestBlockHashes(root.hash) + } + } + root.sectionRUnlock() + } + self.roots = nil + self.controlSections(peer, true) +} + +// (re)starts process without requests, only suicide timer +func (self *peerInfo) stop(peer *peerInfo) { + self.lock.RLock() + defer self.lock.RUnlock() + close(self.quitC) + self.controlSections(peer, false) +} + +func (self *peerInfo) controlSections(peer *peerInfo, on bool) { + if peer != nil { + peer.lock.RLock() + defer peer.lock.RUnlock() + } + for hash, section := range peer.sections { + if section.done() { + delete(self.sections, hash) + } + _, exists := peer.sections[hash] + if on || peer == nil || exists { + if on { + // self is best peer + section.start() + } else { + // (re)starts process without requests, only suicide timer + section.stop() + } + } + } +} + +// called when parent is found in pool +// parent and child are guaranteed to be on different sections func (self *BlockPool) link(parent, child *poolNode) { - // reactivate node scheduled for suicide - if parent.suicide != nil { - close(parent.suicide) - parent.suicide = nil + var top bool + parent.sectionLock() + if child != nil { + child.sectionLock() + } + if parent == parent.section.top && parent.section.top != nil { + top = true + } + var bottom bool + + if child == child.section.bottom { + bottom = true } if parent.child != child { orphan := parent.child - orphan.parent = nil - go self.killChain(orphan) + if orphan != nil { + // got a fork in the chain + if top { + orphan.lock.Lock() + // make old child orphan + orphan.parent = nil + orphan.lock.Unlock() + } else { // we are under section lock + // make old child orphan + orphan.parent = nil + // reset section objects above the fork + nchild := orphan.child + node := orphan + section := §ion{bottom: orphan} + for node.section == nchild.section { + node = nchild + node.section = section + nchild = node.child + } + section.top = node + // set up a suicide + self.processSection(orphan).stop() + } + } else { + // child is on top of a chain need to close section + child.section.bottom = child + } + // adopt new child parent.child = child + if !top { + parent.section.top = parent + // restart section process so that shorter section is scanned for blocks + parent.section.reset() + } } + if child != nil { if child.parent != parent { - orphan := child.parent - orphan.child = nil - go func() { - // if it is a aberrant reverse fork, zip down to bottom - for orphan.parent != nil { - orphan = orphan.parent - } - self.killChain(orphan) - }() - child.parent = parent - } - child.knownParent = false - } -} - -func (self *BlockPool) killChain(node *poolNode) { - if node == nil { - return - } - poolLogger.Debugf("suicide scheduled on node %v", node) - suicide := make(chan bool) - self.lock.Lock() - node.suicide = suicide - self.lock.Unlock() - timer := time.After(cacheTimeout * time.Minute) - self.wg.Add(1) - select { - case <-self.quit: - case <-suicide: - // cancel suicide = close node.suicide to reactivate node - case <-timer: - poolLogger.Debugf("suicide on node %v", node) - self.lock.Lock() - defer self.lock.Unlock() - // proceed up via child links until another suicide root found or chain ends - // abort request blocks loops that start above - // and delete nodes from pool then quit the suicide process - okToAbort := node.blockRequestRoot - for node != nil && (node.suicide == suicide || node.suicide == nil) { - self.pool[string(node.hash)] = nil - if okToAbort && node.blockRequestQuit != nil { - quit := *(node.blockRequestQuit) - if quit != nil { // not yet closed - *(node.blockRequestQuit) = nil - close(quit) + stepParent := child.parent + if stepParent != nil { + if bottom { + stepParent.Lock() + stepParent.child = nil + stepParent.Unlock() + } else { + // we are on the same section + // if it is a aberrant reverse fork, + stepParent.child = nil + node := stepParent + nparent := stepParent.child + section := §ion{top: stepParent} + for node.section == nparent.section { + node = nparent + node.section = section + node = node.parent + } } } else { - okToAbort = true + // linking to a root node, ie. parent is under the root of a chain + parent.section.top = parent } - node = node.child } + child.parent = parent + child.section.bottom = child + } + // this needed if someone lied about the parent before + child.knownParent = false + + parent.sectionUnlock() + if child != nil { + child.sectionUnlock() } - self.wg.Done() } -// AddBlock is the entry point for the eth protocol when blockmsg is received upon requests -// It has a strict interpretation of the protocol in that if the block received has not been requested, it results in an error (which can be ignored) -// block is checked for PoW -// only the first PoW-valid block for a hash is considered legit -func (self *BlockPool) AddBlock(block *types.Block, peerId string) (err error) { - hash := block.Hash() +// this immediately kills the chain from node to end (inclusive) section by section +func (self *BlockPool) killChain(node *poolNode, end *poolNode) { + poolLogger.Debugf("kill chain section with root node %v", node) + + node.sectionLock() + node.section.abort() + self.set(node.hash, nil) + child := node.child + top := node.section.top + i := 1 + self.wg.Add(1) + go func() { + var quit bool + for node != top && node != end && child != nil { + node = child + select { + case <-self.quit: + quit = true + break + default: + } + self.set(node.hash, nil) + child = node.child + } + poolLogger.Debugf("killed chain section of %v blocks with root node %v", i, node) + if !quit { + if node == top { + if node != end && child != nil && end != nil { + // + self.killChain(child, end) + } + } else { + if child != nil { + // delink rest of this section if ended midsection + child.section.bottom = child + child.parent = nil + } + } + } + node.section.bottom = nil + node.sectionUnlock() + self.wg.Done() + }() +} + +// structure to store long range links on chain to skip along +type section struct { + lock sync.RWMutex + bottom *poolNode + top *poolNode + controlC chan bool + resetC chan bool +} + +func (self *section) start() { + self.lock.RLock() + defer self.lock.RUnlock() + if self.controlC != nil { + self.controlC <- true + } +} + +func (self *section) stop() { + self.lock.RLock() + defer self.lock.RUnlock() + if self.controlC != nil { + self.controlC <- false + } +} + +func (self *section) reset() { + self.lock.RLock() + defer self.lock.RUnlock() + if self.controlC != nil { + self.resetC <- true + self.controlC <- false + } +} + +func (self *section) abort() { self.lock.Lock() defer self.lock.Unlock() - node, ok := self.pool[string(hash)] - if !ok && !self.chainManager.KnownBlock(hash) { - return fmt.Errorf("unrequested block %x", hash) + if self.controlC != nil { + close(self.controlC) + self.controlC = nil } - if node.block != nil { - return - } - // validate block for PoW - if !self.chainManager.CheckPoW(block) { - return fmt.Errorf("invalid pow on block %x", hash) - } - node.block = block - node.source = peerId - return nil +} + +func (self *section) done() bool { + self.lock.Lock() + defer self.lock.Unlock() + if self.controlC != nil { + return true + } + return false +} + +func (self *BlockPool) get(hash []byte) (node *poolNode) { + self.lock.Lock() + defer self.lock.Unlock() + return self.pool[string(hash)] +} + +func (self *BlockPool) set(hash []byte, node *poolNode) { + self.lock.Lock() + defer self.lock.Unlock() + self.pool[string(hash)] = node +} + +// first time for block request, this iteration retrieves nodes of the chain +// from node up to top (all the way if nil) via child links +// copies the controller +// and feeds nodeC channel +// this is performed under section readlock to prevent top from going away +// when +func (self *BlockPool) foldUp(node *poolNode, nodeC chan *poolNode) { + self.wg.Add(1) + go func() { + node.sectionRLock() + defer node.sectionRUnlock() + for node != nil { + select { + case <-self.quit: + break + case nodeC <- node: + if node == node.section.top { + break + } + node = node.child + } + } + close(nodeC) + self.wg.Done() + }() +} + +func (self *poolNode) Lock() { + self.sectionLock() + self.lock.Lock() +} + +func (self *poolNode) Unlock() { + self.lock.Unlock() + self.sectionUnlock() +} + +func (self *poolNode) RLock() { + self.lock.RLock() +} + +func (self *poolNode) RUnlock() { + self.lock.RUnlock() +} + +func (self *poolNode) sectionLock() { + self.lock.RLock() + defer self.lock.RUnlock() + self.section.lock.Lock() +} + +func (self *poolNode) sectionUnlock() { + self.lock.RLock() + defer self.lock.RUnlock() + self.section.lock.Unlock() +} + +func (self *poolNode) sectionRLock() { + self.lock.RLock() + defer self.lock.RUnlock() + self.section.lock.RLock() +} + +func (self *poolNode) sectionRUnlock() { + self.lock.RLock() + defer self.lock.RUnlock() + self.section.lock.RUnlock() }