Re: [PATCH] util: In ASSERT(), use "fprintf(stderr, ...)" rather than err(...)

[Stefano Brivio <sbrivio@redhat.com>]

On Tue, 6 Aug 2024 11:27:40 +0200
Laurent Vivier <lvivier@redhat.com> wrote:

> On 06/08/2024 02:27, David Gibson wrote:
> > On Mon, Aug 05, 2024 at 04:10:27PM +0200, Laurent Vivier wrote:  
> >> We use our own implementation of assert() because the glic implementation
> >> uses syscalls that aren't in our seccomp filter, see 7a8ed9459dfe
> >> ("Make assertions actually useful"). And we replaced it by an err(),
> >> followed by an abort() (that is also catched by seccomp).  
> > 
> > I think Stefano's said everything I would on the change itself, but..
> >   
> >> We don't have a coredump or a backtrace but we have at least the
> >> error message... only if logging is enabled.  
> > 
> > Whether we get a coredump shouldn't be affected by our weird ASSERT()
> > here.  If coredumps are enabled (which they're not by default on
> > current distros, AFAICT), we should still get a coredump with the
> > SIGSYS here, just as we would for a SIGABRT.
> >   
> 
> I don't know what, but there is something that prevent passts to generate a core dump
> (I tried abort(), (char *)0 = 0, it works with a simple program, not with passt).

This:

  isolation.c:374:	prctl(PR_SET_DUMPABLE, 0);

> Moreover, if we use gdb it doesn't stop on the syscall but exit and we cannot locate the 
> exit point. And as strace doesn't report the syscall that generates the SIGSYS, it makes 
> hard to find which one it is.

This is probably caused by the seccomp filter instead. You could skip
applying that by commenting this out:

        if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) ||
            prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog))
                die_perror("Failed to apply seccomp filter");

and find out which system call causes that... maybe we could add it to
the 'make valgrind' target, then, or add a 'make debug' one. It's just
important to clearly show that users should *not* use any such target
for regular operation.

-- 
Stefano