Re: [added to the 4.1 stable tree] xfs: Don't wrap growfs AGFL indexes

On 06/03/2016 08:02 PM, Dave Chinner wrote:

On Fri, Jun 03, 2016 at 07:10:26PM -0400, Sasha Levin wrote:

quoted

On 06/03/2016 07:05 PM, Dave Chinner wrote:

quoted

On Fri, Jun 03, 2016 at 05:35:03PM -0400, Sasha Levin wrote:

quoted

quoted

From: Dave Chinner <redacted>

This patch has been added to the 4.1 stable tree. If you have any
objections, please let us know.

This should not go to the 4.1 tree. Check the stable notification
information, please:

quoted

quoted

===============

[ Upstream commit ad747e3b299671e1a53db74963cc6c5f6cdb9f6d ]

....

quoted

quoted

cc: <redacted> # 4.4-4.5

It says 4.4 and 4.5 only.

That's correct, but the commit that this patch fixes was:

commit 96f859d52bcb1c6ea6f3388d39862bf7143e2f30
Author: Darrick J. Wong [off-list ref]
Date:   Mon Jan 4 16:13:21 2016 +1100

    libxfs: pack the agfl header structure so XFS_AGFL_SIZE is correct

....

quoted

    cc: [off-list ref] # 3.10 - 4.4
    Signed-off-by: Darrick J. Wong [off-list ref]
    Reviewed-by: Dave Chinner [off-list ref]
    Signed-off-by: Dave Chinner [off-list ref]

Which, as you see, was marked for and backported to -stable versions
prior to 4.4, which is why I've preferred to queue this commit even though
it's not supposed to go in and see if anyone objects.

Ugh, which means that you're going to have another major tranche of
fixes to backport once I fix the problem properly and handle the
problems automatically. The stable patxh for 4.4-4.5 was really just
meant as a stop-gap measure (i.e.  prevent the reported vector)
while we fixed the problem once and for all. That means you're
looking at a another set of patches in the hundreds of lines needing
to be backported to the stable kernels.

This doesn't sound like stable material though, for stable it should
be trivially correct (and <100 LOC), which isn't the case here.

I get rather concerned about the stable kernel process when this
starts happening - no-one I know of on the XFS side does any testing
on *any* of the stable kernel releases, and now we're in the
dangerous territory of layered on-disk format fixes being backported
by developers without any specific XFS expertise and there's no
real safety net....

Maybe I'm being overly paranoid, but this *scares me*. The chance of
something going wrong and lots of users having problems is much
higher than I'm comfortable with.... 

I agree. This isn't true only for XFS and these cases where backports
are less than trivial either because the code is complex, or because
the commit dependency chain makes the combination tricky probably
happen more often than they should.

Maybe we need to put our foot down more often and not take anything
that is not a strictly trivial fix?


Thanks,
Sasha