On Tue, Jun 14, 2016 at 06:30:56PM -0700, Greg KH wrote:
> On Wed, Jun 15, 2016 at 10:02:41AM +1000, Dave Chinner wrote:
> > On Mon, Jun 13, 2016 at 11:57:53PM +0200, Thomas D. wrote:
> > > The bad commit according to grsec's statement is
> > >
> > > > From b1438f477934f5a4d5a44df26f3079a7575d5946 Mon Sep 17 00:00:00 2001
> > > > From: Dave Chinner <dchinner@xxxxxxxxxx>
> > > > Date: Wed, 18 May 2016 13:53:42 +1000
> > > > Subject: [PATCH] xfs: xfs_iflush_cluster fails to abort on error
> > >
> > > Would be nice to get some clarification.
> >
> > There's nothing wrong with that commit in the upstream kernel,
> > it's the backport that has a bug in it because it failed to take
> > into account changes outside the context of the upstream commit that
> > the older kernels don't have.
>
> Thanks for letting me know about this.
>
> As the patch was tagged with 3.10+, I assumed that it was safe to be
> merged to those older kernels, otherwise I would never have done so. We
> do have ways to mark external things like this for stable patches, it's
> a great help when doing backports.
Little things like this are very easy to forget about - those error
sign changes are ancient history as far as upstream development is
concerned. This is why we have regression tests - the zero-day
kernel robot can run xfstests - perhaps stable kernels should be
submitted to a full round of testing before release to catch
subtle "patch applies but ends up wrong" issues like this...
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
|