Skip to content

Deadlock when trying to quit #253

@jcrussell

Description

@jcrussell

This program will deadlock with the latest version of libpcap (1.8.1):

package main

import (
    "log"
    "os"
    "os/exec"
    "os/signal"
    "time"

    "github.com/google/gopacket/pcap"
)

// Tap is the name of the tap we'll capture traffic from, shouldn't exist
// already otherwise the ip command will fail
const Tap = "testtap"

func main() {
    // Create tap that won't receive any traffic, ever
    cmd := exec.Command("ip", "tuntap", "add", "mode", "tap", Tap)
    if _, err := cmd.CombinedOutput(); err != nil {
        panic(err)
    }
    cmd = exec.Command("ip", "link", "set", Tap, "up")
    if _, err := cmd.CombinedOutput(); err != nil {
        panic(err)
    }   

    // And delete it when we're done
    defer func() {
        cmd := exec.Command("ip", "link", "del", Tap)
        if _, err := cmd.CombinedOutput(); err != nil {
            panic(err)
        }   
    }() 

    // Channel to signal when the pcap reading goroutine is finished
    done := make(chan bool)

    handle, err := pcap.OpenLive(Tap, 1600, true, time.Second)
    if err != nil {
        panic(err)
    }   

    go func() {
        defer close(done)

        var err error
        for err == nil {
            _, _, err = handle.ReadPacketData()
            if err == pcap.NextErrorTimeoutExpired {
                // non-terminating...
                err = nil 
            }   
        }   

        log.Printf("done reading packets: %v", err)
    }() 

    // Wait for Ctrl-C
    sig := make(chan os.Signal, 1024)
    signal.Notify(sig, os.Interrupt)

    <-sig
    log.Print("got SIGTERM, closing handle")

    // Close the handle
    handle.Close()

    // Wait for goroutine to terminate
    <-done
    log.Print("all done")
}

Expected output (with libpcap 1.7.4):

$ sudo ./test 
^C2016/11/09 14:36:22 got SIGTERM, closing handle
2016/11/09 14:36:23 done reading packets: EOF
2016/11/09 14:36:23 all done

Actual output (with libpcap 1.8.1):

$ sudo ./test 
^C2016/11/09 14:36:22 got SIGTERM, closing handle
<hangs>

The problem is related to the read timeout with pcap_next_ex: [0]

This means that the read timeout should NOT be used, for example, in an interactive application to allow the packet capture loop to ``poll'' for user input periodically, as there's no guarantee that a call reading packets will return after the timeout expires even if no packets have arrived.

pcap.Handle.Close should probably call pcap_breakloop in order to signal that pcap_next_ex should return. I looked at the pcap code and there may need to be some refactoring of the locking to allow for this.

[0] http://www.tcpdump.org/manpages/pcap.3pcap.html

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions