fence_virtd can't be killed cleanly

[andyprice]

Since commit 2c909b7, control doesn't return to the dispatch loop in main() after select() returns with EINTR:

while (run && lp->dispatch(listener_ctx, NULL) >= 0);

The signal handler zeroes run when SIGINT, SIGTERM or SIGQUIT is caught but now it never gets checked after the first time because the loop now continues inside ->dispatch on EINTR.

This is causing fence_virtd.service to take 1.5 minutes to stop after SIGTERM is sent as that is the time limit before it gets sent a SIGKILL.

Tested with the multicast listener.

[andyprice]

Thanks!

[ryan-mccabe]

Thanks for the catch. Glad at least somebody is running off master.

[andyprice]

I found the bug in rawhide but I did check master before I filed the issue.