lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:	Fri, 11 May 2007 17:19:01 +0200
From:	Tejun Heo <htejun@...il.com>
To:	Oleg Nesterov <oleg@...sign.ru>
CC:	Andrew Morton <akpm@...ux-foundation.org>,
	David Chinner <dgc@....com>,
	David Howells <dhowells@...hat.com>,
	Gautham Shenoy <ego@...ibm.com>,
	Jarek Poplawski <jarkao2@...pl>, Ingo Molnar <mingo@...e.hu>,
	Srivatsa Vaddagiri <vatsa@...ibm.com>,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] make cancel_rearming_delayed_work() reliable

Oleg Nesterov wrote:
>> and this.  After grabbing cwq lock, compare it to get_wq_data() first,
>> if they are the same, both are using the same lock so there's no
>> reason for the barriers.  If they are different, it doesn't matter
>> anyway.  We need to retry the locking.
> 
> I think this is not right. The problem is that work->data (its
> WORK_STRUCT_WQ_DATA_MASK part, cwq) could be changed _without_ holding
> the old cwq->lock.
> 
> Suppose that CPU_0 does queue_delayed_work(dwork). We start the timer,
> work->data points to cwq_0. CPU_DEAD comes, the timer migrates to
> CPU_1, but work->data was not changed yet.
> 
>>   retry:
>> 	cwq = get_sw_data(work);
>> 	if (!cwq)
>> 		return ret;
>>
>> 	spin_lock_irq(&cwq->lock);
>> 	if (unlikely(cwq != get_wq_data(work))) {
>> 		/* oops wrong cwq */
>> 		spin_unlock_irq(&cwq->lock);
>> 		goto retry;	/* or just return 0; */
>> 	}
> 
> dwork->timer fires, delayed_work_timer_fn() changes work->data to cwq_1
> and queues this work on cwq_1.
> 
>> 	if (!list_empty(&work->entry)) {
>> 		list_del_init(&work->entry);
> 
> Oops! we are holding cwq_0->lock, but modify cwq_1->worklist.
> 
> Actually, we have the same problem with a plain queue_work() if _cpu_down()
> comes at "right" time.
>
> However, I agree, this smp_wmb() in insert_work() should die. We can
> introduce "smp_mb__before_spinlock()" (no-op on x86 at least) to kill it.

Yeah, right, we allow cwq pointer to change without holding the lock.
Although I still think that is where we should fix the problem.  Taking
down CPU is a cold cold path.  We can afford a lot of overhead there.
IMHO, if we can do that, it would be far better than memory barrier
dance which tends to be difficult to understand and thus prove/maintain
correctness.  I'll think about it more.

>>> + * It is possible to use this function if the work re-queues itself. It can
>>> + * cancel the work even if it migrates to another workqueue, however in that
>>> + * case it only garantees that work->func() has completed on the last queued
>>> + * workqueue.
>> We first prevent requeueing from happening; then, wait for each cwq to
>> finish the work, so I think we're guaranteed that they're finished on
>> all cpus.  Otherwise, the 'sync' part isn't too useful as it means all
>> rearming tasks might be running on completion of cancel_work_sync().
> 
> Yes, sure, you are right. What I meant was:
> 
> 	struct workqueue_struct *WQ1, *WQ1;
> 
> 	void work_func(struct work_struct *self)
> 	{
> 		// migrate on another workqueue
> 		queue_work(WQ2, self);
> 
> 		do_something();
> 	}
> 
> 	queue_work(WQ1, work);
> 
> Now, cancel_work_sync() can't guarantee that this work has finished
> its execution on WQ1.

Right again, I misunderstood that the migration was between cwqs.
Things definitely won't work if it's between different wqs.

Thanks.

-- 
tejun
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists