rbd: don't revalidate so much

Whenever a header object event causes a mapped rbd image to refresh
its header information, revalidate_disk() is being called.  This was
done in rbd_dev_refresh() outside the control mutex in order to
avoid a lock inversion.  Although a an event like this *might*
indicate the image has changed size, most of the time it does not.

Record the image size before and after the refresh, and only
call revalidate_disk() if it changes.

This resolves:
    http://tracker.ceph.com/issues/4867

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This commit is contained in:
Alex Elder 2013-04-30 00:44:32 -05:00 committed by Sage Weil
parent 96882f55c4
commit a3fbe5d447

View File

@ -3065,19 +3065,22 @@ static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver)
static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver)
{
u64 image_size;
int ret;
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
image_size = rbd_dev->header.image_size;
mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
if (rbd_dev->image_format == 1)
ret = rbd_dev_v1_refresh(rbd_dev, hver);
else
ret = rbd_dev_v2_refresh(rbd_dev, hver);
mutex_unlock(&ctl_mutex);
revalidate_disk(rbd_dev->disk);
if (ret)
rbd_warn(rbd_dev, "got notification but failed to "
" update snaps: %d\n", ret);
if (image_size != rbd_dev->header.image_size)
revalidate_disk(rbd_dev->disk);
return ret;
}