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

[Stefano Brivio <sbrivio@redhat.com>]

On Mon, 5 Aug 2024 17:52:06 +0200
Laurent Vivier <lvivier@redhat.com> wrote:

> On 05/08/2024 16:34, Stefano Brivio wrote:
> > On Mon,  5 Aug 2024 16:10:27 +0200
> > Laurent Vivier <lvivier@redhat.com> 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).
> >> We don't have a coredump or a backtrace but we have at least the
> >> error message... only if logging is enabled.  
> > 
> > ...wait, err() should always end up somewhere (syslog, log file, or
> > stderr). If you pick stderr, and we closed stderr, then the error will
> > not be reported anywhere (or, possibly worse, on a socket that happens
> > to have number 2).
> > 
> > In which case are you losing messages?
> >   
> 
> Try something:
> 
> diff --git a/udp.c b/udp.c
> index a92014806a73..7e179d3b863b 100644
> --- a/udp.c
> +++ b/udp.c
> @@ -707,6 +707,8 @@ void udp_listen_sock_handler(const struct ctx *c, union epoll_ref ref,
>          if ((n = udp_sock_recv(c, ref.fd, events, mmh_recv)) <= 0)
>                  return;
> 
> +       ASSERT(0);
> +
>          /* We divide datagrams into batches based on how we need to send them,
>           * determined by udp_meta[i].tosidx.  To avoid either two passes through
>           * the array, or recalculating tosidx for a single entry, we have to
> 
> And then generate some UDP traffic.
> 
> This exits only with:
> 
> Bad system call
> 
> Oh, wait, I can see the error in the syslog... but it's not very intuitive as we have the 
> "Bad system call" displayed in the terminal.

Oops, this is an actual issue: if you run passt in foreground, then
err() should _also_ print to stderr, see commit bca0fefa32e0 ("conf,
passt: Make --stderr do nothing, and deprecate it")... but it doesn't.

That's because we set the log_runtime flag once we're ready (whether
we daemonised or not), and then we have:

	if (debug_print || !log_conf_parsed ||
	    (!log_runtime && (log_mask & LOG_MASK(LOG_PRI(pri))))) {
		(void)vfprintf(stderr, format, ap);
		if (newline && format[strlen(format)] != '\n')
			fprintf(stderr, "\n");
	}

See also https://archives.passt.top/passt-dev/ZnTUrhu8LNWi-krS@zatzit/.

Still, while we want to keep pasta usable, we should make sure we print
to stderr when passt is running in foreground. I'll send another
patch...

-- 
Stefano