Re: [PATCH 2.6.13.1] Patch for invisible threads

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Sripathi Kodi <[email protected]> wrote:
>
> Hi,
> 
> When the main thread of a multi-threaded program calls 'pthread_exit' before
> other threads have exited, it results in the other threads becoming
> 'invisible' to commands like 'ps'.

This stuff is subtle.   Let me cc some subtle people.

> This problem was discussed here :
> http://lkml.org/lkml/2004/10/5/234 and http://kerneltrap.org/node/3930, but
> I can't find a patch or explanation for it anywhere. This problem is only
> seen with NPTL and not with LinuxThreads, because Linuxthreads does not let
> the main thread exit (puts it to sleep) until all other threads have exited.
> 
> The problem can be easily recreated with this simple program:
> #include <pthread.h>
> #include <sys/types.h>
> #include <unistd.h>
> void *run(void *arg)
> {
>      sleep(60);
>      printf("Thread: Exiting\n");
>      pthread_exit(NULL);
> }
> int main()
> {
>      pthread_t t;
>      pthread_create(&t, NULL, run, NULL);
>      sleep(20);
>      printf("Main: exiting\n");
>      pthread_exit(NULL);
> }
> 
> After the main thread calls 'pthread_exit', it is shown to be defunct. We
> can still see the directory /proc/<pid_of_main_thread>/task using 'ls',
> 'stat' on it returns success, but 'open' on that directory returns ENOENT.
> Hence though the other thread is still running, it can't be seen.
> 
> The reason appears to be the call to __exit_fs from do_exit when the main
> thread exits. This sets the 'fs' pointer in the task struct to NULL. It also
> decrements the reference count on the fs structure, but does not release the
> memory because the other thread still holds a reference (__put_fs_struct).
> When we do open() on /proc/<pid>/task, proc_root_link() (flow is open_namei
> - may_open - proc_permission - proc_check_root - proc_root_link) tries to
> obtain the task_struct->fs of the main thread, which is now NULL. So it
> returns ENOENT.
> 
> I think we can fix this problem by the following patch. We set the fs
> pointer to NULL only if either the thread is not a thread group leader or if
> the whole thread group has exited. If the main thread is the last to exit,
> it will set the fs pointer to NULL. However, if it is not the last, it won't
> set fs pointer to NULL so that other threads can still use it. Behavior of
> __put_fs_struct is not affected.
> 
> Please let me know if this is reasonable or if there are other ways to fix
> the problem.
> 
> Thanks and regards,
> Sripathi.
> 
> 
> Signed-off-by: Sripathi Kodi <[email protected]>
> 
> --- linux-2.6.13.1/kernel/exit.c	2005-09-12 02:46:26.000000000 -0500
> +++ /home/sripathi/17794/patch_2.6.13.1/exit.c	2005-09-12 02:46:15.000000000 
> -0500
> @@ -463,9 +463,11 @@ static inline void __exit_fs(struct task
>   	struct fs_struct * fs = tsk->fs;
> 
>   	if (fs) {
> -		task_lock(tsk);
> -		tsk->fs = NULL;
> -		task_unlock(tsk);
> +		if (!thread_group_leader(tsk) || !atomic_read(&tsk->signal->live)) {
> +			task_lock(tsk);
> +			tsk->fs = NULL;
> +			task_unlock(tsk);
> +		}
>   		__put_fs_struct(fs);
>   	}
>   }

A comment in there would be nice.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[Index of Archives]     [Kernel Newbies]     [Netfilter]     [Bugtraq]     [Photo]     [Gimp]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Video 4 Linux]     [Linux for the blind]
  Powered by Linux