Re: [PATCH] rcu_read_lock/unlock protect find_task_by_vpid call

[Sergey Senozhatsky <sergey.senozhatsky@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3659 bytes --]

On (10/29/10 13:16), Paul E. McKenney wrote:
> On Fri, Oct 29, 2010 at 03:55:50PM +0300, Sergey Senozhatsky wrote:
> > Commit 4221a9918e38b7494cee341dda7b7b4bb8c04bde "Add RCU check for 
> > find_task_by_vpid()" introduced rcu_lockdep_assert to find_task_by_pid_ns.
> > Assertion failed in sys_ioprio_get. The patch is fixing assertion
> > failure in ioprio_set as well. 
> > 
> >  ===================================================
> >  [ INFO: suspicious rcu_dereference_check() usage. ]
> >  ---------------------------------------------------
> >  kernel/pid.c:419 invoked rcu_dereference_check() without protection!
> > 
> >  rcu_scheduler_active = 1, debug_locks = 0
> >  1 lock held by iotop/4254:
> >  #0:  (tasklist_lock){.?.?..}, at: [<ffffffff811104b4>] sys_ioprio_get+0x22/0x2da
> > 
> >  stack backtrace:
> >  Pid: 4254, comm: iotop Not tainted
> >  Call Trace:
> >  [<ffffffff810656f2>] lockdep_rcu_dereference+0xaa/0xb2
> >  [<ffffffff81053c67>] find_task_by_pid_ns+0x4f/0x68
> >  [<ffffffff81053c9d>] find_task_by_vpid+0x1d/0x1f
> >  [<ffffffff811104e2>] sys_ioprio_get+0x50/0x2da
> >  [<ffffffff81002182>] system_call_fastpath+0x16/0x1b
> > 
> > 
> > Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
> > 
> > ---
> > 
> > diff --git a/fs/ioprio.c b/fs/ioprio.c
> > index 748cfb9..666343d 100644
> > --- a/fs/ioprio.c
> > +++ b/fs/ioprio.c
> > @@ -113,8 +113,11 @@ SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
> 
> Interesting...
> 
> The task-list lock is read-held at this point, which should mean that
> the PID mapping cannot change.  The lockdep_tasklist_lock_is_held()
> function does lockdep_is_held(&tasklist_lock), which must therefore
> only be checking for write-holding the lock.  The fix would be to
> make lockdep_tasklist_lock_is_held() check for either read-holding or
> write-holding tasklist lock.
> 
> Or is there some subtle reason that read-holding the tasklist lock is
> not sufficient?
>

Hello,

On the kernel/pid.c side we have the requirement that 
find_task_by_vpid -> find_task_by_pid_ns

should be called with rcu_read_lock.

/*
 * Must be called under rcu_read_lock().
 */
struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
{
	rcu_lockdep_assert(rcu_read_lock_held());
	return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
}

 
Should it be changed to (let's say)

struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
{
-	rcu_lockdep_assert(rcu_read_lock_held());
+	rcu_lockdep_assert(rcu_read_lock_held() || lockdep_tasklist_lock_is_held());
	return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
}


	Sergey

> 							Thanx, Paul
> 
> >  		case IOPRIO_WHO_PROCESS:
> >  			if (!who)
> >  				p = current;
> > -			else
> > +			else {
> > +				rcu_read_lock();
> >  				p = find_task_by_vpid(who);
> > +				rcu_read_unlock();
> > +			}
> >  			if (p)
> >  				ret = set_task_ioprio(p, ioprio);
> >  			break;
> > @@ -202,8 +205,11 @@ SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
> >  		case IOPRIO_WHO_PROCESS:
> >  			if (!who)
> >  				p = current;
> > -			else
> > +			else {
> > +				rcu_read_lock();
> >  				p = find_task_by_vpid(who);
> > +				rcu_read_unlock();
> > +			}
> >  			if (p)
> >  				ret = get_task_ioprio(p);
> >  			break;
> > 
> > --
> > 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/
> 

[-- Attachment #2: Type: application/pgp-signature, Size: 316 bytes --]