GNU bug report logs - #73589
system* does not honor SIGINT restoration in child

Previous Next

Full log

View this message in rfc822 format

From: help-debbugs <at> gnu.org (GNU bug Tracking System)
To: Ludovic Courtès <ludo <at> gnu.org>
Cc: tracker <at> debbugs.gnu.org
Subject: bug#73589: closed (system* does not honor SIGINT restoration in
 child)
Date: Sun, 20 Oct 2024 18:59:02 +0000

[Message part 1 (text/plain, inline)]

Your message dated Sun, 20 Oct 2024 20:58:14 +0200
with message-id <875xpmmwjd.fsf <at> gnu.org>
and subject line Re: bug#73589: system* does not honor SIGINT restoration in child
has caused the debbugs.gnu.org bug report #73589,
regarding system* does not honor SIGINT restoration in child
to be marked as done.

(If you believe you have received this mail in error, please contact
help-debbugs <at> gnu.org.)


-- 
73589: https://debbugs.gnu.org/cgi/bugreport.cgi?bug=73589
GNU Bug Tracking System
Contact help-debbugs <at> gnu.org with problems

[Message part 2 (message/rfc822, inline)]

From: Olivier Dion <olivier.dion <at> polymtl.ca>
To: bug-guile <at> gnu.org
Subject: system* does not honor SIGINT restoration in child
Date: Tue, 01 Oct 2024 16:51:03 -0400

Hi,

Given the following C program:
--8<---------------cut here---------------start------------->8---
#include <stdio.h>
#include <signal.h>

int main(void)
{
	struct sigaction act;

	sigaction(SIGINT, NULL, &act);

	printf("%p %p %p\n", act.sa_handler, SIG_IGN, SIG_DFL);
}
--8<---------------cut here---------------end--------------->8---

and its ouput from various executions:

  1) $ ./a.out => (nil) 0x1 (nil)
  2) $ guile -c '(system "./a.out")' => (nil) 0x1 (nil)
  3) $ guile -c '(system* "./a.out")' => 0x1 0x1 (nil)

We can see that 3) does not honor restoration of `SIGINT' to `SIG_DFL'
like `system(3)' does.  This seems to be because of the following
sigaction before the creation of the process:

--8<---------------cut here---------------start------------->8---
1697  scm_dynwind_sigaction (SIGINT,
1698                         scm_from_uintptr_t ((uintptr_t) SIG_IGN),
1699                         SCM_UNDEFINED);
1700 #ifdef SIGQUIT
1701  scm_dynwind_sigaction (SIGQUIT,
1702                         scm_from_uintptr_t ((uintptr_t) SIG_IGN),
1703                         SCM_UNDEFINED);
1704 #endif
1705
1706  err = piped_process (&pid, prog, args,
1707                       SCM_UNDEFINED, SCM_UNDEFINED);
--8<---------------cut here---------------end--------------->8---

From execve(2):

   POSIX.1 specifies that the dispositions of any signals that are
   ignored or set to the default are left unchanged.  POSIX.1 specifies
   one exception: if SIGCHLD is being ignored, then an implementation
   may leave the disposition unchanged or reset it to the default; Linux
   does the former.

Therefore, setting `SIG_IGN' for `SIGINT' before the fork/execve results
in ignoring the signals in the child, which is unexpected as a user of
`system(3)'.  The solution would be to restore `SIG_DFL' before
`execve(2)' if before the call to `system*' the action was not
`SIG_IGN'.

In the mean time, I have a solution that works for single-threaded
application:
--8<---------------cut here---------------start------------->8---
(define (system* . args)
  (let ((handler+flags (sigaction SIGINT)))
    (dynamic-wind
      (lambda ()
        (sigaction SIGINT SIG_IGN))
      (lambda ()
        (let ((cpid (primitive-fork)))
          (if (zero? cpid)
              (catch #t
                (lambda ()
                  (sigaction SIGINT SIG_DFL)
                  (apply execlp (car args) args))
                (lambda _
                  (primitive-exit EXIT_FAILURE)))
              (waitpid cpid))))
      (lambda ()
        (sigaction SIGINT
          (car handler+flags)
          (cdr handler+flags))))))
--8<---------------cut here---------------end--------------->8---

Thanks,
Olivier

-- 
Olivier Dion
oldiob.ca

[Message part 3 (message/rfc822, inline)]

From: Ludovic Courtès <ludo <at> gnu.org>
To: Olivier Dion <olivier.dion <at> polymtl.ca>
Cc: 73589-done <at> debbugs.gnu.org
Subject: Re: bug#73589: system* does not honor SIGINT restoration in child
Date: Sun, 20 Oct 2024 20:58:14 +0200

Hi,

Olivier Dion <olivier.dion <at> polymtl.ca> skribis:

> and its ouput from various executions:
>
>   1) $ ./a.out => (nil) 0x1 (nil)
>   2) $ guile -c '(system "./a.out")' => (nil) 0x1 (nil)
>   3) $ guile -c '(system* "./a.out")' => 0x1 0x1 (nil)
>
> We can see that 3) does not honor restoration of `SIGINT' to `SIG_DFL'
> like `system(3)' does.  This seems to be because of the following
> sigaction before the creation of the process:
>
> 1697  scm_dynwind_sigaction (SIGINT,
> 1698                         scm_from_uintptr_t ((uintptr_t) SIG_IGN),
> 1699                         SCM_UNDEFINED);
> 1700 #ifdef SIGQUIT
> 1701  scm_dynwind_sigaction (SIGQUIT,
> 1702                         scm_from_uintptr_t ((uintptr_t) SIG_IGN),
> 1703                         SCM_UNDEFINED);
> 1704 #endif
> 1705
> 1706  err = piped_process (&pid, prog, args,
> 1707                       SCM_UNDEFINED, SCM_UNDEFINED);

This was fixed in commit 4ae33f76d6b33ea0bedfa36050d44c88d08c2823, which
is included in 3.0.10:

--8<---------------cut here---------------start------------->8---
$ cat t.c
#include <stdio.h>
#include <signal.h>

int main(void)
{
	struct sigaction act;

	sigaction(SIGINT, NULL, &act);

	printf("%p %p %p\n", act.sa_handler, SIG_IGN, SIG_DFL);
}
$ gcc -Wall t.c
$ ./a.out 
(nil) 0x1 (nil)
$ ./meta/guile -c '(system "./a.out")'
(nil) 0x1 (nil)
$ ./meta/guile -c '(system* "./a.out")'
(nil) 0x1 (nil)
--8<---------------cut here---------------end--------------->8---

Ludo’.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.