Change signal mask after vfork/clone system call

Change signal mask after vfork/clone system call

[Gregory Giguashvili]

For new processes created by vfork, while signals are blocked using pthread_sigmask: is there a way in a multi-threaded parent to safely unblock signals before the new process is started? Does pthread_atfork like functionality exist for vfork-ed processes maybe?

Ideally, the following code should work, except we're not allowed to call anything in the child process besides exec family of functions.

sigmask_t new;
sigmask_t old;
sigaddset (&new, SIGCHLD);
pthread_sigmask (SIG_BLOCK, &new, &old);

switch (vfork ()) {
case 0: // parent
	break;
case -1: // error
	break;
default: // child
	pthread_sigmask (SIG_SETMASK, &old, NULL);
	execv (...);
	_exit (-1);
}

Finally, if using fork should be efficient for starting processes from applications with heavy VM/RSS memory footprint, this question becomes obsolete.

Thanks
Giga

---------------------------------------------------------------------------------------------------------------
This e-mail, including any attached files, may contain confidential and privileged information for the sole use of the intended recipient. Any review, use, distribution, or disclosure by others is strictly prohibited. If you are not the intended recipient (or authorized to receive information for the intended recipient), please contact the sender by reply e-mail and delete all copies of this message.

RE: Change signal mask after vfork/clone system call

[Gregory Giguashvili]

Any comments on this issue? If you think this mailing list is not appropriate for submitting such questions, I would appreciate a pointer to the one that is.
To me, it seems to be related to "clone" system call functionality and flexibility.

Thanks
Giga

-----Original Message-----
From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-owner@vger.kernel.org] On Behalf Of Gregory Giguashvili
Sent: Tuesday, October 19, 2010 9:54 AM
To: linux-kernel@vger.kernel.org
Subject: Change signal mask after vfork/clone system call

For new processes created by vfork, while signals are blocked using pthread_sigmask: is there a way in a multi-threaded parent to safely unblock signals before the new process is started? Does pthread_atfork like functionality exist for vfork-ed processes maybe?

Ideally, the following code should work, except we're not allowed to call anything in the child process besides exec family of functions.

sigmask_t new;
sigmask_t old;
sigaddset (&new, SIGCHLD);
pthread_sigmask (SIG_BLOCK, &new, &old);

switch (vfork ()) {
case 0: // parent
	break;
case -1: // error
	break;
default: // child
	pthread_sigmask (SIG_SETMASK, &old, NULL);
	execv (...);
	_exit (-1);
}

Finally, if using fork should be efficient for starting processes from applications with heavy VM/RSS memory footprint, this question becomes obsolete.

Thanks
Giga

---------------------------------------------------------------------------------------------------------------
This e-mail, including any attached files, may contain confidential and privileged information for the sole use of the intended recipient. Any review, use, distribution, or disclosure by others is strictly prohibited. If you are not the intended recipient (or authorized to receive information for the intended recipient), please contact the sender by reply e-mail and delete all copies of this message.

RE: Change signal mask after vfork/clone system call

[Gregory Giguashvili]

Any comments on this issue? Is there a way to prevent signal mask inheritance when using vfork?

-----Original Message-----
From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-owner@vger.kernel.org] On Behalf Of Gregory Giguashvili
Sent: Tuesday, October 19, 2010 9:54 AM
To: linux-kernel@vger.kernel.org
Subject: Change signal mask after vfork/clone system call

For new processes created by vfork, while signals are blocked using pthread_sigmask: is there a way in a multi-threaded parent to safely unblock signals before the new process is started? Does pthread_atfork like functionality exist for vfork-ed processes maybe?

Ideally, the following code should work, except we're not allowed to call anything in the child process besides exec family of functions.

sigmask_t new;
sigmask_t old;
sigaddset (&new, SIGCHLD);
pthread_sigmask (SIG_BLOCK, &new, &old);

switch (vfork ()) {
case 0: // parent
	break;
case -1: // error
	break;
default: // child
	pthread_sigmask (SIG_SETMASK, &old, NULL);
	execv (...);
	_exit (-1);
}

Finally, if using fork should be efficient for starting processes from applications with heavy VM/RSS memory footprint, this question becomes obsolete.

Thanks
Giga

---------------------------------------------------------------------------------------------------------------
This e-mail, including any attached files, may contain confidential and privileged information for the sole use of the intended recipient. Any review, use, distribution, or disclosure by others is strictly prohibited. If you are not the intended recipient (or authorized to receive information for the intended recipient), please contact the sender by reply e-mail and delete all copies of this message.


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
---------------------------------------------------------------------------------------------------------------
This e-mail, including any attached files, may contain confidential and privileged information for the sole use of the intended recipient. Any review, use, distribution, or disclosure by others is strictly prohibited. If you are not the intended recipient (or authorized to receive information for the intended recipient), please contact the sender by reply e-mail and delete all copies of this message.

Re: Change signal mask after vfork/clone system call

[Alan Cox]

> Finally, if using fork should be efficient for starting processes from applications with heavy VM/RSS memory footprint, this question becomes obsolete.

Fork on pretty much any Unix like system around today does copy-on-write
so while not as efficient as vfork should be fine for most purposes.

Alan

RE: Change signal mask after vfork/clone system call

[Gregory Giguashvili]

> Fork on pretty much any Unix like system around today does copy-on-write
>so while not as efficient as vfork should be fine for most purposes.
Yes, but the problem is that fork simply does not work when a process has huge resident size and vm.overcommit_memory=0 as preferred by Linux distributions we run on. So, processes with large RSS footprint may occasionally fail fork, even if a tiny process is to be started.

Using vfork always works, but has the signal mask problem. Catch 22?

Giga

---------------------------------------------------------------------------------------------------------------
This e-mail, including any attached files, may contain confidential and privileged information for the sole use of the intended recipient. Any review, use, distribution, or disclosure by others is strictly prohibited. If you are not the intended recipient (or authorized to receive information for the intended recipient), please contact the sender by reply e-mail and delete all copies of this message.

Re: Change signal mask after vfork/clone system call

[Alan Cox]

On Thu, 4 Nov 2010 15:37:05 +0000
Gregory Giguashvili <Gregory.Giguashvili@PDGM.com> wrote:

> > Fork on pretty much any Unix like system around today does copy-on-write
> >so while not as efficient as vfork should be fine for most purposes.
> Yes, but the problem is that fork simply does not work when a process has huge resident size and vm.overcommit_memory=0 as preferred by Linux distributions we run on. So, processes with large RSS footprint may occasionally fail fork, even if a tiny process is to be started.
> 
> Using vfork always works, but has the signal mask problem. Catch 22?

And is there a reason you can't mask the signals, vfork and unmask them
again after the parent continues ?

RE: Change signal mask after vfork/clone system call

[Gregory Giguashvili]

> And is there a reason you can't mask the signals, vfork and unmask them
> again after the parent continues?
The signals are already masked in all threads and there's a dedicated thread, catching signals with sigwait. The problem is that vfork inherits signal mask to a child process and 3rd party executables are not processing any signals because they're not using sigwait technique.

Now, I could unmask signals for the thread calling vfork, but it seems not a signal-safe solution to me because signals may arrive to this thread until I reestablish the mask. pthread_atfork call comes to rescue to address such issues for fork, but there seems to be nothing for vfork call.

Giga
---------------------------------------------------------------------------------------------------------------
This e-mail, including any attached files, may contain confidential and privileged information for the sole use of the intended recipient. Any review, use, distribution, or disclosure by others is strictly prohibited. If you are not the intended recipient (or authorized to receive information for the intended recipient), please contact the sender by reply e-mail and delete all copies of this message.

Re: Change signal mask after vfork/clone system call

[Alan Cox]

> Now, I could unmask signals for the thread calling vfork, but it seems not a signal-safe solution to me because signals may arrive to this thread until I reestablish the mask. pthread_atfork call comes to rescue to address such issues for fork, but there seems to be nothing for vfork call.

Oh you are wrapping buggy apps ?

vfork
exec a tiny app which fixes the signals and execs the proper app

or fix the apps being run...

RE: Change signal mask after vfork/clone system call

[Gregory Giguashvili]

> Oh you are wrapping buggy apps ?
The applications are not buggy. They're just unaware of signals being blocked. It's a common practice to set a signal, but not to reset signal mask unfortunately.

> exec a tiny app which fixes the signals and execs the proper app
Yes, this is the last resort of course. I was wondering if I'm missing some functionality in Linux allowing me to avoid intermediate executables. It looks like I have to go via a wrapper exe.

Thank you for your comments,
Giga


---------------------------------------------------------------------------------------------------------------
This e-mail, including any attached files, may contain confidential and privileged information for the sole use of the intended recipient. Any review, use, distribution, or disclosure by others is strictly prohibited. If you are not the intended recipient (or authorized to receive information for the intended recipient), please contact the sender by reply e-mail and delete all copies of this message.

RE: Change signal mask after vfork/clone system call

[Gregory Giguashvili]

> Finally, if using fork should be efficient for starting processes from applications with heavy VM/RSS memory footprint, > this question becomes obsolete.

What about posix_spawn API? It allows for resetting signal mask for the child process, but does it belong to fork or vfork "family"?

Giga
---------------------------------------------------------------------------------------------------------------
This e-mail, including any attached files, may contain confidential and privileged information for the sole use of the intended recipient. Any review, use, distribution, or disclosure by others is strictly prohibited. If you are not the intended recipient (or authorized to receive information for the intended recipient), please contact the sender by reply e-mail and delete all copies of this message.