On 08/15/2018 09:19 AM, Markus Armbruster wrote:
When read() from the qtest socket or the QMP socket fails or EOFs, we report "Broken pipe" and exit(1). This commonly happens when QEMU crashes. It also happens when QEMU refuses to run because the test passed it bad arguments. Sadly, we neglect to report either.Improve this by calling abort() instead of exit(1), so kill_qemu() runs, and reports how QEMU died. This improves error reporting to something like /x86_64/device/introspect/list: Broken pipe tests/libqtest.c:129: kill_qemu() detected QEMU death from signal 6 (Aborted) (dumped core)
Matches your touchup to my v4 patch.
Three exit() remain in libqtest.c: * In qmp_response(), when we can't parse a QMP reply read from the QMP socket. Change to abort() for consistency. * In qtest_qemu_binary(), when QTEST_QEMU_BINARY isn't in the environment. This can only happen before we start QEMU. Leave alone. * In qtest_init_without_qmp_handshake(), when the fork()ed child fails to execlp(). Leave alone.
exit() after failed execlp() is often wrong (in particular, if you are fork()ed from a multi-threaded process, you REALLY want to be using _exit() instead). But as you say, not the goal of this particular patch.
exit() elsewhere are unlikely due to QEMU dying on us. If that should turn out to be wrong, we can move kill_qemu() from the abrt handler to atexit() or something. Signed-off-by: Markus Armbruster <[email protected]> --- tests/libqtest.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
Reviewed-by: Eric Blake <[email protected]> -- Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3266 Virtualization: qemu.org | libvirt.org
