send_packets: Avoid crashing on errors in sendto call.

[sruffell]

If sendto() returned an error, pthread_cleanup_pop() was not called before
return, resulting in a segfault on my CentOS 5.x platform.

Next, I need to look into why sendto is failing on CentOS 5.x but not on other
platforms.

[sruffell]

I also now see that this change wouldn't be necessary if Pull Request #57 is merged since it drops the use of the pthread_cleanup_push/pop calls.

[wdoekes]

Yes. I had the same problem with the pthread_cleanup stuff (on Ubuntu), so I had to hack it out too.

[sruffell]

You wouldn't know what the issue with sendto/raw sockets is would you? I think there must be some difference in glib on the platforms since just changing the kernel out on Centos 5.x to 3.2.x from the stock 2.6.18 didn't fix the "invalid arguement" error. But running under CentOS 6.x and Debian 6.0 did not have a problem.

I have the same failure running ossobv:wjd-no_root_for_pcap.

[wdoekes]

No, that I don't. The sendto issues I had were my own doing when I was hacking on the non-root play.

[tebriel]

I was seeing the same segfault issue, randomly after 30-70 calls. I'm running in two dockers communicating to each other over a bridged network.

dmesg shows:

[2860961.852761] traps: sipp[9672] general protection ip:7f6e211a32f0 sp:1f45fa96183e4bc9 error:0 in libc-2.23.so[7f6e2116e000+1bf000]

And the error I see in sipp is:

2016-11-09      20:36:04.498133 1478723764.498133: send_packets.c: sendto failed with error: Operation not permitted.

Sipp was exiting status code 139.

I can reproduce with great regularity, let me know if I can provide any more debugging output.

I created a patch for 2.5.1 since this branch has merge conflicts, patch attached.

send_packets.patch.txt

[wdoekes]

Fixed in recent commits