Re: [NBD] Use per-device semaphore instead of BKL

[Paul Clements <paul.clements@steeleye.com>]

Herbert Xu wrote:
> On Sun, Nov 20, 2005 at 09:34:19AM +1100, herbert wrote:
>
> > This is intentional actually.  nbd_clear_queue never races against
> > nbd_find_request because the ioctl is protected by the BKL.  If it
> > weren't, then we have much worse problems to worry about (e.g.,
> > while you're clearing the queue someone else could have set the
> > socket again and started queueing requests).
>
>
> Actually, we do have bigger problems :) The BKL is dropped every
> time you sleep, and nbd_do_it is definitely a frequent sleeper :)

The dropping of the lock in nbd_do_it is actually critical to the way 
nbd functions. nbd_do_it runs for the lifetime of the nbd device, so if 
nbd_do_it were holding some lock (BKL or otherwise), we'd have big problems.


> This isn't really an issue in practice though because the NBD
> client program is single-threaded and doesn't share its file
> descriptors with anyone else.

Right, there's no problem in practice.


> However, we shouldn't make it too easy for the user to shoot himself
> in the foot.  If he's going to do that, let him at least pay for the
> bullet :)
>
> So here is a patch to use a per-device semaphore instead of the
> BKL to protect the ioctl's against each other.

The problem with this patch is that no ioctls can come in once nbd_do_it 
starts because nbd_do_it runs for the lifetime of the device.

I think we really just need to add the acquiring of queue_lock in 
nbd_clear_que to your previous patch and leave it at that. I'll code 
that up and test it.

Thanks,
Paul
-
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/