On Tue, Mar 05, 2013 at 09:11:00AM +1100, Dave Chinner wrote:
> On Mon, Mar 04, 2013 at 03:52:34AM -0500, CAI Qian wrote:
> > This is to request to apply the below commit for the stable releases
> > in order to fix a regression introduced by 055388a (xfs: dynamic
> > speculative EOF preallocation) that caused fsync() took long time during
> > the sparse file testing.
> > For stable-3.4 and stable-3.8, it can be applied as it is. For stable-3.0,
> > please see the below patch which fixed the context and used xfs_bmapi()
> > instead of xfs_bmapi_read() which yet in the tree. Also tested on the
> > stable-3.0 to confirmed the original fsync() slowness regression is now
> > gone. Please review and ACK.
> I've already said no to -stable in another discussion thread, and
> that discussion has not yet played out. please do not try to preempt
> any discussion by sending patches to @stable before it is even
> decided if it is something we *need* to fix in 2 year old kernels.
> Yes, you have input into the discussion, but please do not take it
> upon yourself to determine what should be backported to -stable and
> what shouldn't be - that is for the subsystem maintainers to decide.
> FWIW, is your memory so short that you don't remember what happened
> a couple of weeks ago with the last XFS bugfix backport you
> requested directly to @stable and was accepted based on "it applies
> and builds, so it's OK?" i.e. without proper review, discussion or
> That's right - it caused a major functional regression and that
> wasted a heap of time for quite a few people in sorting it out.
> So right now this request gets a big, fat, loud NACK from me while
> the aforementioned discussion takes place.
I appreciate that you've been willing to do the legwork on this. That's really
nice work, but I agree with Dave that it needs a closer look before we request
that it be picked up in -stable. Lets get this reviewed and tested on
xfs@xxxxxxxxxxx before bringing it to the attention of the -stable folk. We
can continue to work through this in the other thread. Thanks for spending the