[Qemu-devel] qemu-ga: possible race while suspending the guest

[Luiz Capitulino]

Hi Michael,

There's a possible race condition in the bios_supports_mode() function used
by all suspend commands in qemu-ga: if the parent process receives a SIGCHLD
while executing:

    close(pipefds[1]);
    g_free(pmutils_path);

Or:

    ret = read(pipefds[0], &status, sizeof(status));

Then bad things can happen, like reporting that the guest doesn't support
suspend or a resource leak (a segfault may be possible too).

The easy & obvious way to fix this is just to block SIGCLHD while in close()
and g_free() and to loop read() on EINTR.

However, the real problem here is that the double fork schema used in the
suspend commands got complex. It's this way because I was trying to avoid
causing a conflict with your series that adds support for running commands
in the guest.

The ideal solution is to add an internal API to qemu-ga to create & wait for
child processes, like we discussed during the suspend commands review.

Now, what it's best way to go with this bug? Should I fix it the easy way
or should I wait for the new API (which I believe you're going to work on)?

Thanks.