Re: [Fwd: [PATCH]: Fix crash in virtio_blk during modprobe ; rmmod ; modprobe]

[Jens Axboe <jens.axboe@oracle.com>]

On Thu, May 22 2008, Chris Lalancette wrote:
> Jens Axboe wrote:
> > On Thu, May 22 2008, Rusty Russell wrote:
> >> On Wednesday 21 May 2008 23:12:39 Chris Lalancette wrote:
> >>> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> >>> index 4962e62..c678ac5 100644
> >>> --- a/drivers/block/virtio_blk.c
> >>> +++ b/drivers/block/virtio_blk.c
> >>> @@ -294,6 +294,7 @@ static void virtblk_remove(struct virtio_device *vdev)
> >>>         vdev->config->reset(vdev);
> >>>  
> >>>         blk_cleanup_queue(vblk->disk->queue);
> >>> +       del_gendisk(vblk->disk);
> >>>         put_disk(vblk->disk);
> >>>         unregister_blkdev(major, "virtblk");
> >>>         mempool_destroy(vblk->pool);
> >> Thanks Chris, it seems reasonable and I'm sure it works (kinda hard to
> >> test here, it's my root block dev).  Other drivers seem to do
> >> blk_cleanup_queue() *after* del_gendisk: does it matter?
> >>
> >> Jens CC'd: he's gentle with my dumb questions...  Rusty.
> > 
> > del_gendisk() can generate IO, so it would seem safer to do that
> > _before_ putting the queue reference :-)
> > 
> 
> Ah, good to know.  Out of curiousity, why would del_gendisk() generate
> additional I/O?

It does invalidate+sync.

-- 
Jens Axboe