On Mon, Mar 10, 2008 at 12:18:00PM +0000, Alasdair G Kergon wrote: > The question is, are there any > *actual* code paths which could lead to that happening, where existing > locking/serialisation is inadequate? The final dm_put that triggers the deletion (md->holders dropping to 0) is in __hash_remove() while holding _hash_lock - so I don't think the dm_get() calls you mention can run in parallel. There are other dm_get() calls outside _hash_lock, but I think the argument is that the code paths containing them can't be reached if the final dm_put() has started - they all require something to have the device open. Alasdair -- agk@xxxxxxxxxx -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel