mbox series

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

Message ID 162163367115.8379.8459012634106035341.stgit@sifl (mailing list archive)
Headers show
Series Add LSM access controls and auditing to io_uring | expand

Message

Paul Moore May 21, 2021, 9:49 p.m. UTC
Earlier this week Kumar Kartikeya Dwivedi posted a patchset switching
io_uring over to the anonymous inode variant that allows for LSM
controls.  While nice, the patchset left the actual LSM controls as
an exercise for the reader.  The posting can be found using the lore
link below:

https://lore.kernel.org/io-uring/CAHC9VhS=PDxx=MzZnGGNLwo-o5Og-HGZe84=+BBtBCZgaGSn4A@mail.gmail.com/T/#mde8c5120f3b8e34a5a3b18229b8c563a7855fd20

As fate would have it, I had been working on something very similar,
in fact the two patches from Kumar mirrored two in my own patchset.
This patchset, while still a bit crude, does include an attempt at
adding the LSM and audit support necessary to properly implement LSM
based access controls for io_uring.  I've provided the SELinux
implementation, Casey has been nice enough to provide a Smack patch,
and John is working on an AppArmor patch as I write this.  I've
mentioned this work to the other LSM maintainers that I believe might
be affected but I have not heard back from anyone else at this point.
If any of the other LSMs would like to contribute a patch to this
patchset I will happily accept it; I only ask that you post it to the
LSM list and make sure I am on the To/CC line.  I think it would be
nice to try and wrap this up as soon as possible for the obvious
reasons.

The individual patches provide an explanation of the changes involved
so I'm not going to repeat that here, but I will caution you that
these patches are still rather crude, perhaps more than a RFC patchset
should be, but it seemed prudent to move this along so I'm posting
these now.  Any review that you can provide would be helpful.

Also, any pointers to easy-to-run io_uring tests would be helpful.  I
am particularly interested in tests which make use of the personality
option, share urings across process boundaries, and make use of the
sqpoll functionality.

As a point of reference, this patchset is based on v5.13-rc2 and if
you want to follow along via git I'll be making updates to the git
tree/branch below (warning I will be force-pushing on this branch given
the early/rough nature of these patches).

git://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/selinux.git
 (checkout branch "working-io_uring")

Thanks in advance,
-Paul

---

Casey Schaufler (1):
      Smack: Brutalist io_uring support with debug

Paul Moore (8):
      audit: prepare audit_context for use in calling contexts beyond syscalls
      audit,io_uring,io-wq: add some basic audit support to io_uring
      audit: dev/test patch to force io_uring auditing
      audit: add filtering for io_uring records
      fs: add anon_inode_getfile_secure() similar to anon_inode_getfd_secure()
      io_uring: convert io_uring to the secure anon inode interface
      lsm,io_uring: add LSM hooks to io_uring
      selinux: add support for the io_uring access controls


 fs/anon_inodes.c                    |  29 ++
 fs/io-wq.c                          |   4 +
 fs/io_uring.c                       |  25 +-
 include/linux/anon_inodes.h         |   4 +
 include/linux/audit.h               |  17 +
 include/linux/lsm_hook_defs.h       |   5 +
 include/linux/lsm_hooks.h           |  13 +
 include/linux/security.h            |  16 +
 include/uapi/linux/audit.h          |   4 +-
 kernel/audit.h                      |   7 +-
 kernel/auditfilter.c                |   4 +-
 kernel/auditsc.c                    | 481 ++++++++++++++++++++++------
 security/security.c                 |  12 +
 security/selinux/hooks.c            |  67 ++++
 security/selinux/include/classmap.h |   2 +
 security/smack/smack_lsm.c          |  64 ++++
 16 files changed, 650 insertions(+), 104 deletions(-)

Comments

Tetsuo Handa May 22, 2021, 12:53 a.m. UTC | #1
On 2021/05/22 6:49, Paul Moore wrote:
> I've provided the SELinux
> implementation, Casey has been nice enough to provide a Smack patch,
> and John is working on an AppArmor patch as I write this.  I've
> mentioned this work to the other LSM maintainers that I believe might
> be affected but I have not heard back from anyone else at this point.

I don't think any change is required for TOMOYO, for TOMOYO does not
use "struct cred"->security where [RFC PATCH 8/9] and [RFC PATCH 9/9]
are addressing, and TOMOYO does not call kernel/audit*.c functions.
Paul Moore May 22, 2021, 2:06 a.m. UTC | #2
On Fri, May 21, 2021 at 8:53 PM Tetsuo Handa
<penguin-kernel@i-love.sakura.ne.jp> wrote:
> On 2021/05/22 6:49, Paul Moore wrote:
> > I've provided the SELinux
> > implementation, Casey has been nice enough to provide a Smack patch,
> > and John is working on an AppArmor patch as I write this.  I've
> > mentioned this work to the other LSM maintainers that I believe might
> > be affected but I have not heard back from anyone else at this point.
>
> I don't think any change is required for TOMOYO, for TOMOYO does not
> use "struct cred"->security where [RFC PATCH 8/9] and [RFC PATCH 9/9]
> are addressing, and TOMOYO does not call kernel/audit*.c functions.

Good to know, thank you for checking.
Jeff Moyer May 26, 2021, 3 p.m. UTC | #3
Paul Moore <paul@paul-moore.com> writes:

> Also, any pointers to easy-to-run io_uring tests would be helpful.  I
> am particularly interested in tests which make use of the personality
> option, share urings across process boundaries, and make use of the
> sqpoll functionality.

liburing contains a test suite:
  https://git.kernel.dk/cgit/liburing/

You can run it via 'make runtests'.

Cheers,
Jeff
Paul Moore May 26, 2021, 6:49 p.m. UTC | #4
On Wed, May 26, 2021 at 10:59 AM Jeff Moyer <jmoyer@redhat.com> wrote:
> Paul Moore <paul@paul-moore.com> writes:
>
> > Also, any pointers to easy-to-run io_uring tests would be helpful.  I
> > am particularly interested in tests which make use of the personality
> > option, share urings across process boundaries, and make use of the
> > sqpoll functionality.
>
> liburing contains a test suite:
>   https://git.kernel.dk/cgit/liburing/
>
> You can run it via 'make runtests'.

Thanks Jeff, I'll take a look.  Quick question as I start sifting
through the tests, are there any tests in here which share a single
ring across process boundaries?
Jeff Moyer May 26, 2021, 7:07 p.m. UTC | #5
Paul Moore <paul@paul-moore.com> writes:

> On Wed, May 26, 2021 at 10:59 AM Jeff Moyer <jmoyer@redhat.com> wrote:
>> Paul Moore <paul@paul-moore.com> writes:
>>
>> > Also, any pointers to easy-to-run io_uring tests would be helpful.  I
>> > am particularly interested in tests which make use of the personality
>> > option, share urings across process boundaries, and make use of the
>> > sqpoll functionality.
>>
>> liburing contains a test suite:
>>   https://git.kernel.dk/cgit/liburing/
>>
>> You can run it via 'make runtests'.
>
> Thanks Jeff, I'll take a look.  Quick question as I start sifting
> through the tests, are there any tests in here which share a single
> ring across process boundaries?

Yes.  At the very least, this one:

test/across-fork.c

-Jeff
Paul Moore May 26, 2021, 7:10 p.m. UTC | #6
On Wed, May 26, 2021 at 3:06 PM Jeff Moyer <jmoyer@redhat.com> wrote:
> Paul Moore <paul@paul-moore.com> writes:
>
> > On Wed, May 26, 2021 at 10:59 AM Jeff Moyer <jmoyer@redhat.com> wrote:
> >> Paul Moore <paul@paul-moore.com> writes:
> >>
> >> > Also, any pointers to easy-to-run io_uring tests would be helpful.  I
> >> > am particularly interested in tests which make use of the personality
> >> > option, share urings across process boundaries, and make use of the
> >> > sqpoll functionality.
> >>
> >> liburing contains a test suite:
> >>   https://git.kernel.dk/cgit/liburing/
> >>
> >> You can run it via 'make runtests'.
> >
> > Thanks Jeff, I'll take a look.  Quick question as I start sifting
> > through the tests, are there any tests in here which share a single
> > ring across process boundaries?
>
> Yes.  At the very least, this one:
>
> test/across-fork.c

Great, thanks!