[PATCH v4 00/14] security: digest_cache LSM

[Paul Moore]

On Thu, Jun 20, 2024 at 1:06 PM Roberto Sassu
<roberto.sassu at huaweicloud.com> wrote:
> On Thu, 2024-06-20 at 12:51 -0400, Paul Moore wrote:
> > On Thu, Jun 20, 2024 at 12:31 PM Roberto Sassu
> > <roberto.sassu at huaweicloud.com> wrote:
> > > On Thu, 2024-06-20 at 12:08 -0400, Paul Moore wrote:
> > > > On Thu, Jun 20, 2024 at 11:14 AM Roberto Sassu
> > > > <roberto.sassu at huaweicloud.com> wrote:
> > > > > On Thu, 2024-06-20 at 10:48 -0400, Paul Moore wrote:
> > > > > > On Thu, Jun 20, 2024 at 5:12 AM Roberto Sassu
> > > > > > <roberto.sassu at huaweicloud.com> wrote:
> > > > > > > On Wed, 2024-06-19 at 14:43 -0400, Paul Moore wrote:
> > > > > > > > On Wed, Jun 19, 2024 at 12:38 PM Roberto Sassu
> > > > > > > > <roberto.sassu at huaweicloud.com> wrote:
> > > > > > > > >
> > > > > > > > > Making it a kernel subsystem would likely mean replicating what the LSM
> > > > > > > > > infrastructure is doing, inode (security) blob and being notified about
> > > > > > > > > file/directory changes.
> > > > > > > >
> > > > > > > > Just because the LSM framework can be used for something, perhaps it
> > > > > > > > even makes the implementation easier, it doesn't mean the framework
> > > > > > > > should be used for everything.
> > > > > > >
> > > > > > > It is supporting 3 LSMs: IMA, IPE and BPF LSM.
> > > > > > >
> > > > > > > That makes it a clear target for the security subsystem, and as you
> > > > > > > suggested to start for IMA, if other kernel subsystems require them, we
> > > > > > > can make it as an independent subsystem.
> > > > > >
> > > > > > Have you discussed the file digest cache functionality with either the
> > > > > > IPE or BPF LSM maintainers?  While digest_cache may support these
> > > > >
> > > > > Well, yes. I was in a discussion since long time ago with Deven and
> > > > > Fan. The digest_cache LSM is listed in the Use Case section of the IPE
> > > > > cover letter:
> > > > >
> > > > > https://lore.kernel.org/linux-integrity/1716583609-21790-1-git-send-email-wufan@linux.microsoft.com/
> > > >
> > > > I would hope to see more than one sentence casually mentioning that
> > > > there might be some integration in the future.
> > >
> > > Sure, I can work more with Fan to do a proper integration.
> >
> > That seems like a good pre-requisite for turning digest_cache into a
> > general purpose subsystem.
> >
> > > > > I also developed an IPE module back in the DIGLIM days:
> > > > >
> > > > > https://lore.kernel.org/linux-integrity/a16a628b9e21433198c490500a987121@huawei.com/
> > > >
> > > > That looks like more of an fs-verity integration to me.  Yes, of
> > > > course there would be IPE changes to accept a signature/digest from a
> > > > digest cache, but that should be minor.
> > >
> > > True, but IPE will also benefit from not needing to specify every
> > > digest in the policy.
> >
> > Sure, but that isn't really that important from a code integration
> > perspective, that's an admin policy issue.  I expect there would be
> > much more integration work with fs-verity than with IPE, and I think
> > the fs-verity related work might be a challenge.
>
> Uhm, not sure what you mean, but I don't plan to touch fsverity. There
> was already work to get the fsverity digest. All I would need to do
> from my side is to request a digest cache for the inode being verified
> by IPE and to query the fsverity digest.

So your proposed file digest cache wouldn't be used as a replacement
for the fs-verity digest?  Hmm.  I'll leave this up to you and Fan
(current IPE maintainer), but I'm not sure how much value this would
have for IPE, especially since I believe IPE's fs-verity support is
mostly around fs-verity signatures.

> Of course IPE should also capture kernel reads and verify the file
> containing the reference digests, used to build the digest cache.
>
> > > Also, the design choice of attaching the digest cache to the inode
> > > helps LSMs like IPE that don't have a per inode cache on their own.
> > > Sure, IPE would have to do a digest lookup every time, but at least on
> > > an already populated hash table.
> >
> > Just because you need to attach some state to an inode does not mean a
> > file digest cache must be a LSM.  It could be integrated into the VFS
> > or it could be a separate subsystem; either way it could provide an
> > API (either through well defined data structures or functions) that
> > could be used by various LSMs and filesystems that provide integrity
> > protection.
>
> Given that IMA solved the same problem after 15 years, when it became
> an LSM, I'm not super optimistic on that. But if VFS people or other
> subsystem maintainers would be open for such alternative, I can give it
> a try.

I think you should, because I'm not currently supportive of
digest_cache as a standalone LSM.

-- 
paul-moore.com