On Thu, Nov 17, 2022 at 12:50 AM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> On Wed, Nov 16, 2022 at 05:04:05PM -0500, Paul Moore wrote:
> > On Wed, Nov 16, 2022 at 3:11 AM Roberto Sassu
> > <roberto.sassu@xxxxxxxxxxxxxxx> wrote:
> > > On Tue, 2022-11-15 at 21:27 -0500, Paul Moore wrote:
> > > > On Tue, Nov 15, 2022 at 12:58 PM Roberto Sassu
> > > > <roberto.sassu@xxxxxxxxxxxxxxx> wrote:
> > > > > From: Roberto Sassu <roberto.sassu@xxxxxxxxxx>
> > > > >
> > > > > Define four return value flags (LSM_RET_NEG, LSM_RET_ZERO, LSM_RET_ONE,
> > > > > LSM_RET_GT_ONE), one for each interval of interest (< 0, = 0, = 1, > 1).
> > > > >
> > > > > Redefine the LSM_HOOK() macro to add return value flags as argument, and
> > > > > set the correct flags for each LSM hook.
> > > > >
> > > > > Implementors of new LSM hooks should do the same as well.
> > > > >
> > > > > Cc: stable@xxxxxxxxxxxxxxx # 5.7.x
> > > > > Fixes: 9d3fdea789c8 ("bpf: lsm: Provide attachment points for BPF LSM programs")
> > > > > Signed-off-by: Roberto Sassu <roberto.sassu@xxxxxxxxxx>
> > > > > ---
> > > > > include/linux/bpf_lsm.h | 2 +-
> > > > > include/linux/lsm_hook_defs.h | 779 ++++++++++++++++++++--------------
> > > > > include/linux/lsm_hooks.h | 9 +-
> > > > > kernel/bpf/bpf_lsm.c | 5 +-
> > > > > security/bpf/hooks.c | 2 +-
> > > > > security/security.c | 4 +-
> > > > > 6 files changed, 466 insertions(+), 335 deletions(-)
> > > >
> > > > Just a quick note here that even if we wanted to do something like
> > > > this, it is absolutely not -stable kernel material. No way.
> > >
> > > I was unsure about that. We need a proper fix for this issue that needs
> > > to be backported to some kernels. I saw this more like a dependency.
> > > But I agree with you that it would be unlikely that this patch is
> > > applied to stable kernels.
> > >
> > > For stable kernels, what it would be the proper way? We still need to
> > > maintain an allow list of functions that allow a positive return value,
> > > at least. Should it be in the eBPF code only?
> >
> > Ideally the fix for -stable is the same as what is done for Linus'
> > kernel (ignoring backport fuzzing), so I would wait and see how that
> > ends up first. However, if the patchset for Linus' tree is
> > particularly large and touches a lot of code, you may need to work on
> > something a bit more targeted to the specific problem. I tend to be
> > more conservative than most kernel devs when it comes to -stable
> > patches, but if you can't backport the main upstream patchset, smaller
> > (both in terms of impact and lines changed) is almost always better.
>
> No, the mainline patch (what is in Linus's tree), is almost always
> better and preferred for stable backports. When you diverge, bugs
> happen, almost every time, and it makes later fixes harder to backport
> as well.
>
> But first work on solving the problem in Linus's tree. Don't worry
> about stable trees until after the correct solution is merged.
Perhaps you missed my very first sentence where I mentioned exactly
the same things: solve it in Linus' tree first, backports of patches
in Linus' tree is ideal.
--
paul-moore.com
