Re: [PATCH] pids: sys_getsid: make sure we don't return 0 instead of -ESRCH

[Oleg Nesterov <oleg@tv-sign.ru>]

Sorry! please ignore this patch and
"[PATCH] pids: sys_getpgid: use rcu_read_lock() instead of tasklist_lock",

I'll send the updated version.

It turns out that sys_getsid/sys_getpgid have other problems which should
be fixed.

On 03/23, Oleg Nesterov wrote:
>
> The task can exit between find_task_by_vpid() and task_session_vnr(), in
> that unlikely case sys_getsid() returns 0 instead of -ESRCH.
> 
> The bug was introduced by me in ac9a8e3f0f43d20fc316162e8e5f9186d295ff49.
> 
> Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
> 
> --- 25/kernel/sys.c~1_GETSID	2008-03-20 18:21:17.000000000 +0300
> +++ 25/kernel/sys.c	2008-03-23 19:25:36.000000000 +0300
> @@ -1019,19 +1019,19 @@ asmlinkage long sys_getsid(pid_t pid)
>  	if (!pid)
>  		return task_session_vnr(current);
>  	else {
> -		int retval;
> +		int retval = 0;
>  		struct task_struct *p;
>  
>  		rcu_read_lock();
>  		p = find_task_by_vpid(pid);
> -		retval = -ESRCH;
>  		if (p) {
>  			retval = security_task_getsid(p);
>  			if (!retval)
>  				retval = task_session_vnr(p);
>  		}
>  		rcu_read_unlock();
> -		return retval;
> +
> +		return retval ?: -ESRCH;
>  	}
>  }
>