[LTP] [PATCH 2/2] mq_timedreceive01: Workaround segfault on libc variant on 32 bit

[Jan Stancek]

On Thu, Nov 14, 2024 at 3:53 PM Petr Vorel <pvorel@suse.cz> wrote:
>
> Hi all,
>
> > On Tue, Nov 12, 2024 at 06:18:31PM +0100, Petr Vorel wrote:
> > > EFAULT test segfaults on newer kernels (e.g. 6.4) on libc variant on
> > > 32bit.  Similarly to 1d4d5a0750 use typical LTP workaround to test by
> > > forked child + checking the terminating signal.
>
> ...
> > > +   unsigned int j;
> > > +   unsigned int prio;
>
> > >     if (tc->invalid_msg)
> > >             len -= 1;
> > > @@ -208,6 +192,60 @@ static void do_test(unsigned int i)
> > >                     TST_RET, prio, len);
> > >  }
>
> > > +static void test_bad_addr(unsigned int i)
> > > +{
> > > +   struct time64_variants *tv = &variants[tst_variant];
> > > +   pid_t pid;
> > > +   int status;
> > > +
> > > +   pid = SAFE_FORK();
> > > +   if (!pid) {
> > > +           verify_mqt_receive(i, pid);
> > > +           _exit(0);
> > nit:
> > If this is a normal exit, i suggest use s/_exit(0)/exit(0) ?
>
> I copy pasted this from similar tests. IMHO both should work. According to man
> exit() calls functions registered with atexit() and _exit(), _exit() just
> immediately terminates the process. Any open file descriptors belonging to the process are closed.
>
> @Jan, @Cyril Please correct me if I'm wrong.

Our main usage of _exit() is signal handlers, because exit() is not
async-signal-safe.
exit() will also flush open streams, which seems desirable here.

>
> Kind regards,
> Petr
>
> > > +   }
> > > +
> > > +   SAFE_WAITPID(pid, &status, 0);
> > > +
> > > +   if (WIFEXITED(status) && !WEXITSTATUS(status))
> > > +           return;
> > > +
> > > +   if (tv->ts_type == TST_LIBC_TIMESPEC &&
> > > +           WIFSIGNALED(status) && WTERMSIG(status) == SIGSEGV) {
> > > +           tst_res(TPASS, "Child killed by expected signal");
> > > +           return;
> > > +   }
> > > +
> > > +   tst_res(TFAIL, "Child %s", tst_strstatus(status));
> > > +}
> ...
>