On 10/29/2014 05:19 PM, Theodore Ts'o wrote:
On Wed, Oct 29, 2014 at 08:26:36PM +0100, Thomas Gleixner wrote:
For what it's worth, I'm currently testing a backport of commit b34090e from
mainline (which in turn required backporting commits e5a120a and f5113ef). It
switches from using the BJ_Shadow list to using the BH_Shadow flag on the
buffer head. More interestingly, waiters now get woken up from
journal_end_buffer_io_sync() instead of from
jbd2_journal_commit_transaction().
So far this seems to be helping a lot. It's lasted about 15x as long under
stress as without the patches.
I fear that this is just papering over the problem, but you have to
talk to the jbd2 folks about that.
No, it's a clean fix for the problem. The main issue is that what the
jbd2 commit was doing was starting inode writeback for those blocks
needed to guarantee data=ordered mode (so this is what caused various
pages to have writeback page set) as well as starting metadata writes
to the commit (which is what caused the shadow bit to be set on the
metadata buffers).
Now that we clear the shadow flag when the metadata writes is
complete, the writeback will eventually be allowed to complete and
this prevents the deadlock.
Thanks for the explanation.
A few questions:
1) Is this something that could hit mainline as well, or just the RT kernel?
2) If it can hit mainline, is this something that should be considered
for the various longterm-support kernels? (3.10, maybe 3.4)
3) For 3.4, do you think that it's sufficient to backport the three
commits I mentioned, or are you aware of others that I should be looking
at as well?
Chris
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
