Bug#657078: patches to reduce the footprint of the nfs4 idmapper
On Wed, 25 Apr 2012 11:20:44 +0200
Rik Theys <Rik.Theys@esat.kuleuven.be> wrote:
> Hi,
>
> On 04/24/2012 11:16 PM, Jeff Layton wrote:
> > On Tue, 24 Apr 2012 13:35:13 -0700
> > Greg KH<gregkh@linuxfoundation.org> wrote:
> >
> >> On Tue, Apr 24, 2012 at 10:22:50PM +0200, Rik Theys wrote:
> >>> Hi,
> >>>
> >>> I'm experiencing the following Red Hat bug[1] on RHEL and also on
> >>> Debian.
> >>>
> >>> I've noticed Fedora has started to ship an update with two patches
> >>> that reduce the footprint of the nfs4 id mapper which should prevent
> >>> this (or seriously limit the chance).
> >>>
> >>> NFSv4: Further reduce the footprint of the idmapper
> >>> commit 685f50f9188ac1e8244d0340a9d6ea36b6136cec
> >>>
> >>> NFSv4: Reduce the footprint of the idmapper
> >>> commit d073e9b541e1ac3f52d72c3a153855d9a9ee3278
> >>>
> >>> I believe Red Hat will have those patches in an upcoming RHEL 6.x
> >>> release.
> >>>
> >>> I'm looking for feedback on if it would be possible to include
> >>> these patches in a stable 3.2 update (and/or 3.0.x) so they
> >>> will become part of the upcoming Debian 7.0 kernel (which is based
> >>> on 3.2).
> >>>
> >>> Have these patches made it into the 3.3 and/or 3.4-rc kernels?
> >>
> >> Both of these are in the 3.4-rc1 kernel release.
> >>
> >> As for stable kernels, I don't see how they fit the rules outlined in
> >> Documentation/stable_kernel_rules.txt, do you?
> >>
> >
> > There have been reports on linux-nfs of problems allocating this memory
> > in the past.
> >
> > I suspect most of those occurred when people attempt to do an NFS mount
> > after memory is already heavily fragmented. This allocation was
> > fricking huge before those patches...
> >
> > That said, I'm not sure that really qualifies as stable-kernel fodder...
> >
>
> After rereading the stable_kernel_rules.txt I agree that this might be
> too big for a stable update.
>
> I would still like to see these patches in the Debian 3.2 kernel (bug
> report in cc).
>
> Do you consider these patches something distributions could/should
> cherry pick for their kernels? I see from the RHEL 6.3 beta kernel
> (2.6.32-262.el6) changelog that it includes these fixes?
>
Yeah, it should go into RHEL6.3, but we also pulled in the patches for
the new idmapper too so we sort of had to take these...
> Maybe you and/or Trond could comment on whether you are aware of any
> dependencies or further fixes likely to be needed?
>
I think that set was relatively independent.
--
Jeff Layton <jlayton@redhat.com>
Reply to: