On Tue, Apr 21, 2020 at 07:06:58PM +0200, gregkh@xxxxxxxxxxxxxxxxxxx wrote:
The patch below does not apply to the 5.4-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable@xxxxxxxxxxxxxxx>.
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
From 0e4e1de5b63fa423b13593337a27fd2d2b0bcf77 Mon Sep 17 00:00:00 2001
From: Ilya Dryomov <idryomov@xxxxxxxxx>
Date: Fri, 13 Mar 2020 11:20:51 +0100
Subject: [PATCH] rbd: avoid a deadlock on header_rwsem when flushing notifies
rbd_unregister_watch() flushes notifies and therefore cannot be called
under header_rwsem because a header update notify takes header_rwsem to
synchronize with "rbd map". If mapping an image fails after the watch
is established and a header update notify sneaks in, we deadlock when
erroring out from rbd_dev_image_probe().
Move watch registration and unregistration out of the critical section.
The only reason they were put there was to make header_rwsem management
slightly more obvious.
Fixes: 811c66887746 ("rbd: fix rbd map vs notify races")
Signed-off-by: Ilya Dryomov <idryomov@xxxxxxxxx>
Reviewed-by: Jason Dillaman <dillaman@xxxxxxxxxx>
There was a conflict with:
b9ef2b8858a0 ("rbd: don't establish watch for read-only mappings")
And I ended up with a funny resolution, but given the conflict at hand I
think it makes sense:
@@ -6135,6 +6145,8 @@ static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
err_out_probe:
rbd_dev_unprobe(rbd_dev);
err_out_watch:
+ if (!depth)
+ up_write(&rbd_dev->header_rwsem);
if (!depth)
rbd_unregister_watch(rbd_dev);
err_out_format:
Queued for 5.4-4.14.
--
Thanks,
Sasha
