2006-10-11 11:18:17

by Vasily Tarasov

Subject: [PATCH] block layer: elv_iosched_show should get elv_list_lock

elv_iosched_show function iterates other elv_list,
hence elv_list_lock should be got.

Also the question is: in elv_iosched_show, elv_iosched_store
q->elevator->elevator_type construction is used without locking q->queue_lock.
Is it expected?..

Signed-off-by: Vasily Tarasov <[email protected]>

--

--- linux-2.6.18/block/elevator.c.orig 2006-10-11 11:00:34.000000000 +0400
+++ linux-2.6.18/block/elevator.c 2006-10-11 15:08:20.000000000 +0400
@@ -892,7 +892,7 @@ ssize_t elv_iosched_show(request_queue_t
struct list_head *entry;
int len = 0;

- spin_lock_irq(q->queue_lock);
+ spin_lock_irq(&elv_list_lock);
list_for_each(entry, &elv_list) {
struct elevator_type *__e;

@@ -902,7 +902,7 @@ ssize_t elv_iosched_show(request_queue_t
else
len += sprintf(name+len, "%s ", __e->elevator_name);
}
- spin_unlock_irq(q->queue_lock);
+ spin_unlock_irq(&elv_list_lock);

len += sprintf(len+name, "\n");
return len;

2006-10-11 11:23:10

by Jens Axboe

Subject: Re: [PATCH] block layer: elv_iosched_show should get elv_list_lock

On Wed, Oct 11 2006, Vasily Tarasov wrote:
> elv_iosched_show function iterates other elv_list,
> hence elv_list_lock should be got.

Indeed, that's a bug. Thanks!

> Also the question is: in elv_iosched_show, elv_iosched_store
> q->elevator->elevator_type construction is used without locking
> q->queue_lock. Is it expected?..

As long as the queue doesn't go away, we should be safe enough.

--
Jens Axboe