On Fri, Nov 14, 2014 at 1:31 PM, Chris Mason <clm@xxxxxx> wrote:
On Fri, Nov 14, 2014 at 1:23 PM, Patrick Schmid <schmid@xxxxxxxxxxxx>
wrote:
What kind of machine is this? I'm trying to reproduce here.
The "frontend" hardware is a 24 core Xeon E5-2620 on an Intel
S2600GZ board with 128 GiB RAM, the nic is a Intel X520-DA2 10Gb/s
SFP+, running Ubuntu 14.04LTS.
Do you need more infos?
No, I think I see the problem. I'm testing a patch here and I'll
send it later tonight.
Ok, I think this is related to the new fair read/writer lock
implementation in the generic kernel code. btrfs_clear_path_blocking()
will end up taking locks outside of the strict locking order the rest
of Btrfs uses. This used to be fine because we hold the blocking lock
while we do it, but with the new queued locks we're running into
trouble. We hit a similar bug earlier and I convinced myself the
problem was only with btrfs_next_leaf and our trylock. But it's a
bigger problem than I realized.
So for now I've changed btrfs_clear_path_blocking to honor the rules
and fixed up our trylock to make it faster.
I'm letting a test run on this patch over the weekend, since I don't
want any surprises with your backup farm.
-chris
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
