Timeout bug in inactive handle

[omriyaacovi]

When reading packet data from an activated inactive handle to read packets off the wire, a positive timeout seems to be blocking forever.
Consider this code segment:

package main

import (
	"log"
	"time"

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

var (
	device  = "eno1"
	snaplen = 65535
	promisc = true
)

// A simple ZeroCopyReadPacketData() on a handle generated with gopacket.OpenLive()
func openLive() {
	handle, err := pcap.OpenLive(device, int32(snaplen), promisc, time.Second*2)
	if err != nil {
		log.Fatal(err)
	}
	err = handle.SetBPFFilter("src host 1.2.3.4") // A random filter to make sure no data is received
	if err != nil {
		log.Fatal(err)
	}
	data, _, err := handle.ZeroCopyReadPacketData()
	if err != nil {
		log.Fatal(err)
	}
	log.Println(data)
}

// A simple ZeroCopyReadPacketData() on a handle generated from an InactiveHandle
func inactive() {
	inactiveHandle, err := pcap.NewInactiveHandle(device)
	err = inactiveHandle.SetSnapLen(snaplen)
	if err != nil {
		log.Fatal(err)
	}
	err = inactiveHandle.SetPromisc(promisc)
	if err != nil {
		log.Fatal(err)
	}
	err = inactiveHandle.SetTimeout(time.Second * 2)
	if err != nil {
		log.Fatal(err)
	}
	handle, err := inactiveHandle.Activate()
	if err != nil {
		log.Fatal(err)
	}
	err = handle.SetBPFFilter("src host 1.2.3.4") // A random filter to make sure no data is received
	if err != nil {
		log.Fatal(err)
	}
	data, _, err := handle.ZeroCopyReadPacketData()
	if err != nil {
		log.Fatal(err)
	}
	log.Println(data)
}

With this main function:

func main() {
	log.Println("Started")
	openLive()
}

Will work just fine - output:

2018/12/27 00:00:02 started
2018/12/27 00:00:05 Timeout Expired

Whereas with this main function, calling inactive() instead of openLive():

func main() {
	log.Println("Started")
	inactive()
}

Will just hang/block forever (until killed):

2018/12/27 00:00:02 started

The same thing will happen if we replace ZeroReadPacketData() with ReadPacketData()
I've set timeout in both functions to two seconds.

[notti]

Could verify the behaviour. The reason why openLive works is, that OpenLive sets pcap to nonblocking and the waiting is done with pcap_wait (I think this was done because of a performance regression). The same mechanism wasn't included in Activate (no idea why).
strace shows that libpcap seems to want wo wait forever (poll([{fd=3, events=POLLIN}], 1, -1).
If I remove nonblocking from OpenLive it shows the same behaviour (adding to Activate removes it).

I'll investigate what the root cause is (=why libpcap doesn't want to wait 2 seconds).

[notti]

Hmm ok my bad: pcap_wait was introduced to counter this (in #253) - not due to performance regression