Re: [PATCH] 2.6.0 NBD driver: remove send/recieve race for request

[Jens Axboe]

On Fri, Aug 08 2003, Paul Clements wrote:
> Jens Axboe wrote:
> > 
> > On Fri, Aug 08 2003, Lou Langholtz wrote:
> > > >@@ -499,12 +508,14 @@ static void do_nbd_request(request_queue
> > > >                                     lo->disk->disk_name);
> > > >                     spin_lock(&lo->queue_lock);
> > > >                     list_del_init(&req->queuelist);
> > > >+                    req->ref_count--;
> > > >                     spin_unlock(&lo->queue_lock);
> > > >                     nbd_end_request(req);
> > > >                     spin_lock_irq(q->queue_lock);
> > > >                     continue;
> > > >             }
> > > >
> > > >+            req->ref_count--;
> > > >             spin_lock_irq(q->queue_lock);
> > > >
> > > Since ref_count isn't atomic, shouldn't ref_count only be changed while
> > > the queue_lock is held???
> > 
> > Indeed, needs to be done after regrabbing the lock.
> 
> But req is pulled off of nbd's main request queue at this point, so I
> don't think anyone else could be touching it, could they?

In that case you are right, and it would probably be best not to touch
the reference count at all (just leave it at 1).

-- 
Jens Axboe

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