NetBSD Problem Report #58919

From www@netbsd.org  Wed Dec 18 22:12:56 2024
Return-Path: <www@netbsd.org>
Received: from mail.netbsd.org (mail.netbsd.org [199.233.217.200])
	(using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits)
	 key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256
	 client-signature RSA-PSS (2048 bits) client-digest SHA256)
	(Client CN "mail.NetBSD.org", Issuer "mail.NetBSD.org CA" (not verified))
	by mollari.NetBSD.org (Postfix) with ESMTPS id BBA371A9238
	for <gnats-bugs@gnats.NetBSD.org>; Wed, 18 Dec 2024 22:12:56 +0000 (UTC)
Message-Id: <20241218221255.0CB4D1A923A@mollari.NetBSD.org>
Date: Wed, 18 Dec 2024 22:12:54 +0000 (UTC)
From: campbell+netbsd@mumble.net
Reply-To: campbell+netbsd@mumble.net
To: gnats-bugs@NetBSD.org
Subject: timer_settime fails to trigger for past times
X-Send-Pr-Version: www-1.0

>Number:         58919
>Category:       kern
>Synopsis:       timer_settime fails to trigger for past times
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Wed Dec 18 22:15:00 +0000 2024
>Last-Modified:  Thu Dec 19 22:15:01 +0000 2024
>Originator:     Taylor R Campbell
>Release:        current, 10, 9, ...
>Organization:
The NetBSD Tachyondation
>Environment:
>Description:
POSIX says:

> If the specified time has already passed, the function shall succeed and the expiration notification shall be made.

https://pubs.opengroup.org/onlinepubs/9799919799.2024edition/functions/timer_settime.html

But for relative timers, timer_settime fails with ETIMEDOUT, and for absolute timers, nothing happens at all.
>How-To-Repeat:
$ cat timerpastabs.c
#include <err.h>
#include <signal.h>
#include <time.h>

sig_atomic_t woken;

static void
wakeup(int signo)
{

	woken = 1;
}

int
main(void)
{
	timer_t t;
	struct itimerspec it = { .it_value = { .tv_sec = -1, .tv_nsec = 0 } };
	struct timespec t0;

	if (signal(SIGALRM, wakeup) == SIG_ERR)
		err(1, "signal(SIGALRM)");
	if (timer_create(CLOCK_MONOTONIC, NULL, &t) == -1)
		err(1, "timer_create");
	if (clock_gettime(CLOCK_MONOTONIC, &t0) == -1)
		err(1, "clock_gettime");
	timespecadd(&t0, &it.it_value, &it.it_value);
	if (timer_settime(t, TIMER_ABSTIME, &it, NULL) == -1)
		err(1, "timer_settime");
	return woken ? 0 : 1;
}
$ cat timerpastrel
cat: timerpastrel: No such file or directory
$ cat timerpastrel.c
#include <err.h>
#include <signal.h>
#include <time.h>

sig_atomic_t woken;

static void
wakeup(int signo)
{

	woken = 1;
}

int
main(void)
{
	timer_t t;
	struct itimerspec it = { .it_value = { .tv_sec = -1, .tv_nsec = 0 } };

	if (signal(SIGALRM, wakeup) == SIG_ERR)
		err(1, "signal(SIGALRM)");
	if (timer_create(CLOCK_MONOTONIC, NULL, &t) == -1)
		err(1, "timer_create");
	if (timer_settime(t, 0, &it, NULL) == -1)
		err(1, "timer_settime");
	return woken ? 0 : 1;
}
$ make timerpastabs timerpastrel
cc -O2   -o timerpastabs timerpastabs.c 
cc -O2   -o timerpastrel timerpastrel.c 
$ ./timerpastabs
$ echo $?
1
$ ./timerpastrel
timerpastrel: timer_settime: Connection timed out

>Fix:
Yes, please!

>Audit-Trail:
From: "Taylor R Campbell" <riastradh@netbsd.org>
To: gnats-bugs@gnats.NetBSD.org
Cc: 
Subject: PR/58919 CVS commit: src/tests/lib/libc/sys
Date: Wed, 18 Dec 2024 22:26:53 +0000

 Module Name:	src
 Committed By:	riastradh
 Date:		Wed Dec 18 22:26:53 UTC 2024

 Modified Files:
 	src/tests/lib/libc/sys: t_timer_create.c

 Log Message:
 t_timer_create: Add some more test cases.

 PR kern/58917: timer_settime and timerfd_settime return absolute time
 of next event

 PR kern/58919: timer_settime fails to trigger for past times


 To generate a diff of this commit:
 cvs rdiff -u -r1.5 -r1.6 src/tests/lib/libc/sys/t_timer_create.c

 Please note that diffs are not public domain; they are subject to the
 copyright notices on the relevant files.

From: "Taylor R Campbell" <riastradh@netbsd.org>
To: gnats-bugs@gnats.NetBSD.org
Cc: 
Subject: PR/58919 CVS commit: src/tests/lib/libc/sys
Date: Thu, 19 Dec 2024 20:07:29 +0000

 Module Name:	src
 Committed By:	riastradh
 Date:		Thu Dec 19 20:07:29 UTC 2024

 Modified Files:
 	src/tests/lib/libc/sys: t_timer_create.c

 Log Message:
 t_timer_create: Fix up tests for edge cases.

 While here, save and restore errno in signal handler.

 PR kern/58919: timer_settime fails to trigger for past times

 PR kern/58920: timer_settime fails ETIMEDOUT on negative interval, not
 EINVAL


 To generate a diff of this commit:
 cvs rdiff -u -r1.6 -r1.7 src/tests/lib/libc/sys/t_timer_create.c

 Please note that diffs are not public domain; they are subject to the
 copyright notices on the relevant files.

From: Taylor R Campbell <riastradh@NetBSD.org>
To: gnats-bugs@NetBSD.org, netbsd-bugs@NetBSD.org
Cc: 
Subject: Re: kern/58919: timer_settime fails to trigger for past times
Date: Thu, 19 Dec 2024 22:12:16 +0000

 For relative timers, the language in POSIX.1-2024 is not very clear
 (does the part I quoted apply to the whole paragraph, or only to the
 half of the paragraph about absolute timers?), but by analogy with
 more explicit language about setitimer in POSIX.1-2008, we might
 reasonably conclude that negative relative timers are intended to fail
 with EINVAL (not ETIMEDOUT) rather than be treated as having fired
 already:

 > The setitimer() function shall fail if:
 > 
 > [EINVAL]
 >     The value argument is not in canonical form. (In canonical form,
 >     the number of microseconds is a non-negative integer less than
 >     1000000 and the number of seconds is a non-negative integer.)
 > 
 https://pubs.opengroup.org/onlinepubs/9699919799/functions/setitimer.html

 For absolute timers, the issue is that the timer expiration occurs on
 the _next tick_ rather than immediately.

 This strikes me as wrong: the timer should have fired already, so it
 seems to me the signal should be delivered synchronously with the
 timer_settime syscall, just like a pending signal is delivered
 synchronously with the sigprocmask(SIG_UNBLOCK) syscall.  But we're
 not alone in this behaviour (at least Linux defers it a tiny bit too).

 The following variant of the how-to-repeat program, which sleeps for
 the smallest possible nonzero increment of time (rounded up to one
 tick), shows this (and I've incorporated it into t_timer_create.c and
 t_timerfd.c):

 $ cat timerpastabs1.c
 #include <err.h>
 #include <signal.h>
 #include <time.h>

 sig_atomic_t woken;

 static void
 wakeup(int signo)
 {

 	woken = 1;
 }

 int
 main(void)
 {
 	timer_t t;
 	struct itimerspec it = {.it_value = {-1, 0}};
 	struct timespec t0;

 	if (signal(SIGALRM, wakeup) == SIG_ERR)
 		err(1, "signal(SIGALRM)");
 	if (timer_create(CLOCK_MONOTONIC, NULL, &t) == -1)
 		err(1, "timer_create");
 	if (clock_gettime(CLOCK_MONOTONIC, &t0) == -1)
 		err(1, "clock_gettime");
 	timespecadd(&t0, &it.it_value, &it.it_value);
 	if (timer_settime(t, TIMER_ABSTIME, &it, NULL) == -1)
 		err(1, "timer_settime");
 	if (clock_nanosleep(CLOCK_MONOTONIC, 0,
 		&(const struct timespec){0,1}, NULL) == -1)
 		err(1, "clock_nanosleep");
 	return woken ? 0 : 1;
 }
 $ make timerpastabs1
 cc -O2   -o timerpastabs1 timerpastabs1.c 
 $ ./timerpastabs1
 0

Home	PR Database Search