Re: [Qemu-devel] [PATCH 3/3] qtest: kill QEMU process on g_assert() failure

[Stefan Hajnoczi <stefanha@gmail.com>]

On Tue, Feb 18, 2014 at 11:02:52AM +0100, Markus Armbruster wrote:
> Markus Armbruster <armbru@redhat.com> writes:
> 
> > Stefan Hajnoczi <stefanha@redhat.com> writes:
> >
> >> The QEMU process stays running if the test case fails.  This patch fixes
> >> the leak by installing a SIGABRT signal handler which invokes
> >> qtest_end().
> >>
> >> In order to make that work for assertion failures during qtest_init(),
> >> we need to initialize QTestState fields including file descriptors and
> >> pids carefully.  qtest_quit() is then safe to call even during
> >> qtest_init().
> >>
> >> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> >> ---
> >>  tests/libqtest.c | 29 ++++++++++++++++++++++++++---
> >>  1 file changed, 26 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/tests/libqtest.c b/tests/libqtest.c
> >> index 8b2b2d7..09a0481 100644
> >> --- a/tests/libqtest.c
> >> +++ b/tests/libqtest.c
> >> @@ -44,6 +44,7 @@ struct QTestState
> >>      bool irq_level[MAX_IRQ];
> >>      GString *rx;
> >>      pid_t qemu_pid;  /* our child QEMU process */
> >> +    struct sigaction sigact_old; /* restored on exit */
> >>  };
> >>  
> >>  #define g_assert_no_errno(ret) do { \
> >> @@ -88,6 +89,11 @@ static int socket_accept(int sock)
> >>      return ret;
> >>  }
> >>  
> >> +static void sigabrt_handler(int signo)
> >> +{
> >> +    qtest_end();
> 
> Don't you have to re-raise SIGABRT here, to actually terminate the
> process?

No.  POSIX says:

  RETURN VALUE
       The abort() function shall not return.

(BTW the way to avoid that is using longjmp.)

The Linux man page is more explicit:

  If  the  SIGABRT  signal  is ignored, or caught by a handler that returns,
  the abort() function will still terminate the process.  It does this by
  restoring the default disposition for  SIGABRT and then raising the signal
  for a second time.