Re: [nft PATCH v3] configure: Implement --enable-profiling option

[Phil Sutter <phil@nwl.cc>]

On Thu, Feb 12, 2026 at 01:08:17AM +0100, Florian Westphal wrote:
> Phil Sutter <phil@nwl.cc> wrote:
> > This will set compiler flag --coverage so code coverage may be inspected
> > using gcov.
> > 
> > In order to successfully profile processes which are killed or
> > interrupted as well, add a signal handler for those cases which calls
> > exit(). This is relevant for test cases invoking nft monitor.
> >
> > index 0000000000000..912ead9d7eb94
> > --- /dev/null
> > +++ b/src/profiling.c
> > @@ -0,0 +1,36 @@
> > +/*
> > + * Copyright (c) Red Hat GmbH.  Author: Phil Sutter <phil@nwl.cc>
> > + *
> > + * This program is free software; you can redistribute it and/or modify
> > + * it under the terms of the GNU General Public License version 2 (or any
> > + * later) as published by the Free Software Foundation.
> > + */
> > +
> > +#include <nft.h>
> > +#include <profiling.h>
> > +
> > +#include <signal.h>
> > +#include <stdio.h>
> > +
> > +static void termhandler(int signo)
> > +{
> > +	switch (signo) {
> > +	case SIGTERM:
> > +		exit(143);
> > +	case SIGINT:
> > +		exit(130);
> 
> Unfortunately I can't find exit(3) in the list of async-signal safe
> functions, so I have to assume this isn't allowed.

You're right. At least since exit() runs atexit() callbacks, it is very
unlikely this is async-signal-safe.

> From a quick glance, I would suggest to either use self-pipe-trick, or,
> given nft is linux specific anyway, use signalfd(2) instead of a
> traditional handler; then, stuff the fd into mnl_nft_event_listener
> select().

I'll go with signalfd() as it's simpler than the self-pipe and hooking
into mnl_nft_event_listener is required in both cases.

> Sorry, I did not think of this earlier.  If I'm wrong and this is safe,

Thanks for the pointers! I'm not very familiar with writing signal
handlers.

Cheers, Phil