ping方法與pending的處理，之前談到了pending是等待一個reply。 這裡通過程式碼來分析是如何實現等待reply的。
pending方法把pending結構體傳送給addpending. 然後等待訊息的處理和接收。

// ping sends a ping message to the given node and waits for a reply.
func (t *udp) ping(toid NodeID, toaddr *net.UDPAddr) error {
    // TODO: maybe check for ReplyTo field in callback to measure RTT
    errc := t.pending(toid, pongPacket, func(interface{}) bool { return true })
    t.send(toaddr, pingPacket, &ping{
        Version: Version,
        From: t.ourEndpoint,
        To: makeEndpoint(toaddr, 0), // TODO: maybe use known TCP port from DB
        Expiration: uint64(time.Now().Add(expiration).Unix()),
    })
    return <-errc
}
// pending adds a reply callback to the pending reply queue.
// see the documentation of type pending for a detailed explanation.
func (t *udp) pending(id NodeID, ptype byte, callback func(interface{}) bool) <-chan error {
    ch := make(chan error, 1)
    p := &pending{from: id, ptype: ptype, callback: callback, errc: ch}
    select {
    case t.addpending <- p:
        // loop will handle it
    case <-t.closing:
        ch <- errClosed
    }
    return ch
}
 

addpending訊息的處理。 之前建立udp的時候呼叫了newUDP方法。裡面啟動了兩個goroutine。 其中的loop()就是用來處理pending訊息的。

// loop runs in its own goroutine. it keeps track of
// the refresh timer and the pending reply queue.
func (t *udp) loop() {
    var (
        plist = list.New()
        timeout = time.NewTimer(0)
        nextTimeout *pending // head of plist when timeout was last reset
        contTimeouts = 0 // number of continuous timeouts to do NTP checks
        ntpWarnTime = time.Unix(0, 0)
    )
    <-timeout.C // ignore first timeout
    defer timeout.Stop()

    resetTimeout := func() {
        //這個方法的主要功能是檢視佇列裡面是否有需要超時的pending訊息。 如果有。那麼
        //根據最先超時的時間設定超時醒來。
        if plist.Front() == nil || nextTimeout == plist.Front().Value {
            return
        }
        // Start the timer so it fires when the next pending reply has expired.
        now := time.Now()
        for el := plist.Front(); el != nil; el = el.Next() {
            nextTimeout = el.Value.(*pending)
            if dist := nextTimeout.deadline.Sub(now); dist < 2*respTimeout {
                timeout.Reset(dist)
                return
            }
            // Remove pending replies whose deadline is too far in the
            // future. These can occur if the system clock jumped
            // backwards after the deadline was assigned.
            //如果有訊息的deadline在很遠的未來，那麼直接設定超時，然後移除。
            //這種情況在修改系統時間的時候有可能發生，如果不處理可能導致堵塞太長時間。
            nextTimeout.errc <- errClockWarp
            plist.Remove(el)
        }
        nextTimeout = nil
        timeout.Stop()
    }

    for {
        resetTimeout() //首先處理超時。

        select {
        case <-t.closing: //收到關閉資訊。 超時所有的堵塞的佇列
            for el := plist.Front(); el != nil; el = el.Next() {
                el.Value.(*pending).errc <- errClosed
            }
            return

        case p := <-t.addpending: //增加一個pending 設定deadline
            p.deadline = time.Now().Add(respTimeout)
            plist.PushBack(p)

        case r := <-t.gotreply: //收到一個reply 尋找匹配的pending
            var matched bool
            for el := plist.Front(); el != nil; el = el.Next() {
                p := el.Value.(*pending)
                if p.from == r.from && p.ptype == r.ptype { //如果來自同一個人。 而且型別相同
                    matched = true
                    // Remove the matcher if its callback indicates
                    // that all replies have been received. This is
                    // required for packet types that expect multiple
                    // reply packets.
                    if p.callback(r.data) { //如果callback返回值是true 。說明pending已經完成。 給p.errc寫入nil。 pending完成。
                        p.errc <- nil
                        plist.Remove(el)
                    }
                    // Reset the continuous timeout counter (time drift detection)
                    contTimeouts = 0
                }
            }
            r.matched <- matched //寫入reply的matched

        case now := <-timeout.C: //處理超時資訊
            nextTimeout = nil

            // Notify and remove callbacks whose deadline is in the past.
            for el := plist.Front(); el != nil; el = el.Next() {
                p := el.Value.(*pending)
                if now.After(p.deadline) || now.Equal(p.deadline) { //如果超時寫入超時資訊並移除
                    p.errc <- errTimeout
                    plist.Remove(el)
                    contTimeouts++
                }
            }
            // If we've accumulated too many timeouts, do an NTP time sync check
            if contTimeouts > ntpFailureThreshold {
                //如果連續超時很多次。 那麼檢視是否是時間不同步。 和NTP伺服器進行同步。
                if time.Since(ntpWarnTime) >= ntpWarningCooldown {
                    ntpWarnTime = time.Now()
                    go checkClockDrift()
                }
                contTimeouts = 0
            }
        }
    }
}
 

上面看到了pending的處理。 不過loop()方法種還有一個gotreply的處理。 這個實在readLoop()這個goroutine中產生的。

// readLoop runs in its own goroutine. it handles incoming UDP packets.
func (t *udp) readLoop() {
    defer t.conn.Close()
    // Discovery packets are defined to be no larger than 1280 bytes.
    // Packets larger than this size will be cut at the end and treated
    // as invalid because their hash won't match.
    buf := make([]byte, 1280)
    for {
        nbytes, from, err := t.conn.ReadFromUDP(buf)
        if netutil.IsTemporaryError(err) {
            // Ignore temporary read errors.
            log.Debug("Temporary UDP read error", "err", err)
            continue
        } else if err != nil {
            // Shut down the loop for permament errors.
            log.Debug("UDP read error", "err", err)
            return
        }
        t.handlePacket(from, buf[:nbytes])
    }
}

func (t *udp) handlePacket(from *net.UDPAddr, buf []byte) error {
    packet, fromID, hash, err := decodePacket(buf)
    if err != nil {
        log.Debug("Bad discv4 packet", "addr", from, "err", err)
        return err
    }
    err = packet.handle(t, from, fromID, hash)
    log.Trace("<< "+packet.name(), "addr", from, "err", err)
    return err
}

func (req *ping) handle(t *udp, from *net.UDPAddr, fromID NodeID, mac []byte) error {
    if expired(req.Expiration) {
        return errExpired
    }
    t.send(from, pongPacket, &pong{
        To: makeEndpoint(from, req.From.TCP),
        ReplyTok: mac,
        Expiration: uint64(time.Now().Add(expiration).Unix()),
    })
    if !t.handleReply(fromID, pingPacket, req) {
        // Note: we're ignoring the provided IP address right now
        go t.bond(true, fromID, from, req.From.TCP)
    }
    return nil
}

func (t *udp) handleReply(from NodeID, ptype byte, req packet) bool {
    matched := make(chan bool, 1)
    select {
    case t.gotreply <- reply{from, ptype, req, matched}:
        // loop will handle it
        return <-matched
    case <-t.closing:
        return false
    }
}
 

上面介紹了udp的大致處理的流程。 下面介紹下udp的主要處理的業務。 udp主要傳送兩種請求，對應的也會接收別人傳送的這兩種請求， 對應這兩種請求又會產生兩種迴應。

ping請求，可以看到ping請求希望得到一個pong回答。 然後返回。

// ping sends a ping message to the given node and waits for a reply.
func (t *udp) ping(toid NodeID, toaddr *net.UDPAddr) error {
    // TODO: maybe check for ReplyTo field in callback to measure RTT
    errc := t.pending(toid, pongPacket, func(interface{}) bool { return true })
    t.send(toaddr, pingPacket, &ping{
        Version: Version,
        From: t.ourEndpoint,
        To: makeEndpoint(toaddr, 0), // TODO: maybe use known TCP port from DB
        Expiration: uint64(time.Now().Add(expiration).Unix()),
    })
    return <-errc
}

pong回答,如果pong回答沒有匹配到一個對應的ping請求。那麼返回errUnsolicitedReply異常。

func (req *pong) handle(t *udp, from *net.UDPAddr, fromID NodeID, mac []byte) error {
    if expired(req.Expiration) {
        return errExpired
    }
    if !t.handleReply(fromID, pongPacket, req) {
        return errUnsolicitedReply
    }
    return nil
}

findnode請求, 傳送findnode請求，然後等待node迴應 k個鄰居。

// findnode sends a findnode request to the given node and waits until
// the node has sent up to k neighbors.
func (t *udp) findnode(toid NodeID, toaddr *net.UDPAddr, target NodeID) ([]*Node, error) {
    nodes := make([]*Node, 0, bucketSize)
    nreceived := 0
    errc := t.pending(toid, neighborsPacket, func(r interface{}) bool {
        reply := r.(*neighbors)
        for _, rn := range reply.Nodes {
            nreceived++
            n, err := t.nodeFromRPC(toaddr, rn)
            if err != nil {
                log.Trace("Invalid neighbor node received", "ip", rn.IP, "addr", toaddr, "err", err)
                continue
            }
            nodes = append(nodes, n)
        }
        return nreceived >= bucketSize
    })
    t.send(toaddr, findnodePacket, &findnode{
        Target: target,
        Expiration: uint64(time.Now().Add(expiration).Unix()),
    })
    err := <-errc
    return nodes, err
}

neighbors迴應, 很簡單。 把迴應傳送給gotreply佇列。 如果沒有找到匹配的findnode請求。返回errUnsolicitedReply錯誤

func (req *neighbors) handle(t *udp, from *net.UDPAddr, fromID NodeID, mac []byte) error {
    if expired(req.Expiration) {
        return errExpired
    }
    if !t.handleReply(fromID, neighborsPacket, req) {
        return errUnsolicitedReply
    }
    return nil
}

收到別的節點發送的ping請求，傳送pong回答。 如果沒有匹配上一個pending(說明不是自己方請求的結果)。 就呼叫bond方法把這個節點加入自己的bucket快取。(這部分原理在table.go裡面會詳細介紹)

func (req *ping) handle(t *udp, from *net.UDPAddr, fromID NodeID, mac []byte) error {
    if expired(req.Expiration) {
        return errExpired
    }
    t.send(from, pongPacket, &pong{
        To: makeEndpoint(from, req.From.TCP),
        ReplyTok: mac,
        Expiration: uint64(time.Now().Add(expiration).Unix()),
    })
    if !t.handleReply(fromID, pingPacket, req) {
        // Note: we're ignoring the provided IP address right now
        go t.bond(true, fromID, from, req.From.TCP)
    }
    return nil
}

收到別人傳送的findnode請求。這個請求希望把和target距離相近的k個節點發送回去。 演算法的詳細請參考references目錄下面的pdf文件。

func (req *findnode) handle(t *udp, from *net.UDPAddr, fromID NodeID, mac []byte) error {
    if expired(req.Expiration) {
        return errExpired
    }
    if t.db.node(fromID) == nil {
        // No bond exists, we don't process the packet. This prevents
        // an attack vector where the discovery protocol could be used
        // to amplify traffic in a DDOS attack. A malicious actor
        // would send a findnode request with the IP address and UDP
        // port of the target as the source address. The recipient of
        // the findnode packet would then send a neighbors packet
        // (which is a much bigger packet than findnode) to the victim.
        return errUnknownNode
    }
    target := crypto.Keccak256Hash(req.Target[:])
    t.mutex.Lock()
    //獲取bucketSize個和target距離相近的節點。 這個方法在table.go內部實現。後續會詳細介紹
    closest := t.closest(target, bucketSize).entries
    t.mutex.Unlock()

    p := neighbors{Expiration: uint64(time.Now().Add(expiration).Unix())}
    // Send neighbors in chunks with at most maxNeighbors per packet
    // to stay below the 1280 byte limit.
    for i, n := range closest {
        if netutil.CheckRelayIP(from.IP, n.IP) != nil {
            continue
        }
        p.Nodes = append(p.Nodes, nodeToRPC(n))
        if len(p.Nodes) == maxNeighbors || i == len(closest)-1 {
            t.send(from, neighborsPacket, &p)
            p.Nodes = p.Nodes[:0]
        }
    }
    return nil
}

udp資訊加密和安全問題

discover協議因為沒有承載什麼敏感資料，所以資料是以明文傳輸，但是為了確保資料的完整性和不被篡改，所以在資料包的包頭加上了數字簽名。

func encodePacket(priv *ecdsa.PrivateKey, ptype byte, req interface{}) ([]byte, error) {
    b := new(bytes.Buffer)
    b.Write(headSpace)
    b.WriteByte(ptype)
    if err := rlp.Encode(b, req); err != nil {
        log.Error("Can't encode discv4 packet", "err", err)
        return nil, err
    }
    packet := b.Bytes()
    sig, err := crypto.Sign(crypto.Keccak256(packet[headSize:]), priv)
    if err != nil {
        log.Error("Can't sign discv4 packet", "err", err)
        return nil, err
    }
    copy(packet[macSize:], sig)
    // add the hash to the front. Note: this doesn't protect the
    // packet in any way. Our public key will be part of this hash in
    // The future.
    copy(packet, crypto.Keccak256(packet[macSize:]))
    return packet, nil
}

func decodePacket(buf []byte) (packet, NodeID, []byte, error) {
    if len(buf) < headSize+1 {
        return nil, NodeID{}, nil, errPacketTooSmall
    }
    hash, sig, sigdata := buf[:macSize], buf[macSize:headSize], buf[headSize:]
    shouldhash := crypto.Keccak256(buf[macSize:])
    if !bytes.Equal(hash, shouldhash) {
        return nil, NodeID{}, nil, errBadHash
    }
    fromID, err := recoverNodeID(crypto.Keccak256(buf[headSize:]), sig)
    if err != nil {
        return nil, NodeID{}, hash, err
    }
    var req packet
    switch ptype := sigdata[0]; ptype {
    case pingPacket:
        req = new(ping)
    case pongPacket:
        req = new(pong)
    case findnodePacket:
        req = new(findnode)
    case neighborsPacket:
        req = new(neighbors)
    default:
        return nil, fromID, hash, fmt.Errorf("unknown type: %d", ptype)
    }
    s := rlp.NewStream(bytes.NewReader(sigdata[1:]), 0)
    err = s.Decode(req)
    return req, fromID, hash, err
}