Re: [PATCH v1 0/9] do not use s_dirt in ext4

On Thu, 2012-03-22 at 09:35 -0400, Ted Ts'o wrote:

Just as a quick FYI, I tried applying your patch series on top of my
development tree, and ran into problems when I ran the regression test
(using xfstests).  When I backed out your changes and reran, the tests
completed without any problems.

Thanks Ted, I'll take a look at xfstests and run them next time before
sending out v2.

I'm rerunning the tests since the first failure looks like it might
not be related to your patch series (and yet it went away once I
backed out your patch).  The second failure however looks definitely
related to your changes.  It looks like you don't wait to make sure
the workqueue is flushed out before the file system gets unmounted,
and that can lead to a panic.

Hmm, I thought the whole DIO workqueue would be flushed so I do not have
to do anything. I'll take a look.

Since we're already in the 3.3 has already been released, I suspect
this patch series will probably need to wait until the next merge
window.  We might be able to pull in some of the obviously safe
patches, however.

Sure, that's fine.

But I wonder, since this is cross-FS story, where I need to first do
small VFS change (export the variable), then change all file-systems,
and then remove whole 's_dirt'/'write_supers()' stuff from VFS, how this
would be handled?

IMO, the best way would be to make everything go in via one single tree,
granted I could get all the acks, do you feel like ext4 tree could be
the one?

Also, I am working on top of vanilla 3.3, do you prefer me to work with
your tree instead? I guess this tree:

git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4.git

but which branch?

-- 
Best Regards,
Artem Bityutskiy