ifdestroy can block

[djnz00]

Currently ifdestroy blocks indefinitely in pthread_join().
The file descriptor needs to be closed before the pthread_join(), since the receiver thread is probably blocked in poll() and will not unblock until the FD is closed and poll() returns with an error.
Another method would be to add a shutdown pipe to the poll FD set, but that is overkill since closing the FD aborts the poll anyway.
The above fixes the hang, but there is still a memory leak since closing the FD does not unmap the ring buffer. Once the thread has safely exited, the ring buffer must be unmapped using munmap().

[anttikantee]

Closing the fd first exposes a race condition. Something might open another fd before the receiver sees "dying". I think the current code has its heritage in another implementation where there was a poll timeout, so no hang would happen.

I've actually fixed this the way you describe here:
http://nxr.netbsd.org/xref/src/sys/rump/net/lib/libvirtif/virtif_user.c#254

[anttikantee]

But this is not fixed yet, right? Let's keep an issue open until it is.

[anttikantee]

Hrmph, apparently I can't reopen this. I'll file a new issue.

[djnz00]

Sorry about that. I closed it because my patch wasn't going in and I
thought we'd end up with the pipe solution under a different ticket. -Huw

On Wed, May 7, 2014 at 8:35 PM, Antti Kantee [email protected]:

Hrmph, apparently I can't reopen this. I'll file a new issue.

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/5#issuecomment-42416355
.

[anttikantee]

Sure, it might be, but until it is, it's nice to explicitly list open problems, just in case someone else meanwhile runs into the same problem. Overdocumentation is rarely a problem ;-)

Anyway, no biggie.