[RFC PATCH v2 0/9] Add LSM access controls and auditing to io_uring

[Richard Guy Briggs]

On 2021-09-13 22:49, Paul Moore wrote:
> On Mon, Sep 13, 2021 at 9:50 PM Paul Moore <paul at paul-moore.com> wrote:
> > On Mon, Sep 13, 2021 at 3:23 PM Paul Moore <paul at paul-moore.com> wrote:
> > > On Thu, Sep 9, 2021 at 8:59 PM Richard Guy Briggs <rgb at redhat.com> wrote:
> > > > On 2021-09-01 15:21, Paul Moore wrote:
> > > > > On Sun, Aug 29, 2021 at 11:18 AM Paul Moore <paul at paul-moore.com> wrote:
> > > > > > On Sat, Aug 28, 2021 at 11:04 AM Richard Guy Briggs <rgb at redhat.com> wrote:
> > > > > > > I did set a syscall filter for
> > > > > > >         -a exit,always -F arch=b64 -S io_uring_enter,io_uring_setup,io_uring_register -F key=iouringsyscall
> > > > > > > and that yielded some records with a couple of orphans that surprised me
> > > > > > > a bit.
> > > > > >
> > > > > > Without looking too closely at the log you sent, you can expect URING
> > > > > > records without an associated SYSCALL record when the uring op is
> > > > > > being processed in the io-wq or sqpoll context.  In the io-wq case the
> > > > > > processing is happening after the thread finished the syscall but
> > > > > > before the execution context returns to userspace and in the case of
> > > > > > sqpoll the processing is handled by a separate kernel thread with no
> > > > > > association to a process thread.
> > > > >
> > > > > I spent some time this morning/afternoon playing with the io_uring
> > > > > audit filtering capability and with your audit userspace
> > > > > ghau-iouring-filtering.v1.0 branch it appears to work correctly.  Yes,
> > > > > the userspace tooling isn't quite 100% yet (e.g. `auditctl -l` doesn't
> > > > > map the io_uring ops correctly), but I know you mentioned you have a
> > > > > number of fixes/improvements still as a work-in-progress there so I'm
> > > > > not too concerned.  The important part is that the kernel pieces look
> > > > > to be working correctly.
> > > >
> > > > Ok, I have squashed and pushed the audit userspace support for iouring:
> > > >         https://github.com/rgbriggs/audit-userspace/commit/e8bd8d2ea8adcaa758024cb9b8fa93895ae35eea
> > > >         https://github.com/linux-audit/audit-userspace/compare/master...rgbriggs:ghak-iouring-filtering.v2.1
> > > > There are test rpms for f35 here:
> > > >         http://people.redhat.com/~rbriggs/ghak-iouring/git-e8bd8d2-fc35/
> > > >
> > > > userspace v2 changelog:
> > > > - check for watch before adding perm
> > > > - update manpage to include filesystem filter
> > > > - update support for the uring filter list: doc, -U op, op names
> > > > - add support for the AUDIT_URINGOP record type
> > > > - add uringop support to ausearch
> > > > - add uringop support to aureport
> > > > - lots of bug fixes
> > > >
> > > > "auditctl -a uring,always -S ..." will now throw an error and require
> > > > "-U" instead.
> > >
> > > Thanks Richard.
> > >
> > > FYI, I rebased the io_uring/LSM/audit patchset on top of v5.15-rc1
> > > today and tested both with your v1.0 and with your v2.1 branch and the
> > > various combinations seemed to work just fine (of course the v2.1
> > > userspace branch was more polished, less warts, etc.).  I'm going to
> > > go over the patch set one more time to make sure everything is still
> > > looking good, write up an updated cover letter, and post a v3 revision
> > > later tonight with the hope of merging it into -next later this week.
> >
> > Best laid plans of mice and men ...
> >
> > It turns out the LSM hook macros are full of warnings-now-errors that
> > should likely be resolved before sending anything LSM related to
> > Linus.  I'll post v3 once I fix this, which may not be until tomorrow.
> >
> > (To be clear, the warnings/errors aren't new to this patchset, I'm
> > likely just the first person to notice them.)
> 
> Actually, scratch that ... I'm thinking that might just be an oddity
> of the Intel 0day test robot building for the xtensa arch.  I'll post
> the v3 patchset tonight.

I was in the middle of reviewing the v2 patchset to add my acks when I
forgot to add the comment that you still haven't convinced me that ses=
isn't needed or relevant if we are including auid=.

> paul moore

- RGB

--
Richard Guy Briggs <rgb at redhat.com>
Sr. S/W Engineer, Kernel Security, Base Operating Systems
Remote, Ottawa, Red Hat Canada
IRC: rgb, SunRaycer
Voice: +1.647.777.2635, Internal: (81) 32635