Bug#657078: Patches to reduce footprint of nfs idmapper

[Jonathan Nieder <jrnieder@gmail.com>]

# patches are in mainline but not 3.2.y
tags 657078 - fixed-upstream
quit

Rik Theys wrote:

> NFSv4: Further reduce the footprint of the idmapper
> commit 685f50f9188ac1e8244d0340a9d6ea36b6136cec
>
> NFSv4: Reduce the footprint of the idmapper
> commit d073e9b541e1ac3f52d72c3a153855d9a9ee3278
>
> They are in the upstream kernel 3.4 kernel, were also backported by
> Fedora into their 3.2 kernel, and will be part of RHEL 6.3.
>
> A comment from Jeff Layton in the bug report seems to indicate the
> patches are relatively independent.

Yes, please.

> The patches don't meet the requirements for a stable update, but
> please consider applying these patches to the Debian kernel.

To expand on that: if I am reading correctly, this huge allocation has
been around since pre-historic times (more precisely, since
v2.5.68~112^2~133^2~1, "Add support for mapping NFSv4 remote
user/group names", 2003-04-08), which is part of why a patch to
decrease its size doesn't feel like -stable material.  I suspect the
only reasons the allocation would fail more often these days are that

  (i) NFSv4 is becoming more popular
 (ii) large contiguous allocations are harder to fulfill because of
      other bugs that produce memory fragmentation

I hope to see the patches applied in wheezy because they are not very
invasive, they improve performance generally, and they would take away
a source of distraction when trying to understand cases of (ii).

The patches apply cleanly to 3.2.y.

Hope that helps,
Jonathan