On 7/19/2024 10:08 AM, Paul Moore wrote:

On Jul 11, 2024 Xu Kuohai [off-list ref] wrote:

quoted

To be consistent with most LSM hooks, convert the return value of
hook audit_rule_match to 0 or a negative error code.

Before:
- Hook audit_rule_match returns 1 if the rule matches, 0 if it not,
   and negative error code otherwise.

After:
- Hook audit_rule_match returns 0 on success or a negative error
   code on failure. An output parameter @match is introduced to hold
   the match result on success.

Signed-off-by: Xu Kuohai <redacted>
---
  include/linux/lsm_hook_defs.h     |  3 +-
  security/apparmor/audit.c         | 22 ++++++-------
  security/apparmor/include/audit.h |  2 +-
  security/security.c               | 15 ++++++++-
  security/selinux/include/audit.h  |  8 +++--
  security/selinux/ss/services.c    | 54 +++++++++++++++++--------------
  security/smack/smack_lsm.c        | 19 +++++++----
  7 files changed, 75 insertions(+), 48 deletions(-)

This is another odd hook, and similar to some of the others in this
patchset, I'm not sure how applicable this would be to a BPF-based
LSM.  I suspect you could safely block this from a BPF LSM and no one
would notice or be upset.

However, if we did want to keep this hook for a BPF LSM, I think it
might be better to encode the "match" results in the return value, just
sticking with a more conventional 0/errno approach.  What do you think
about 0:found/ok, -ENOENT:missing/ok, -ERRNO:other/error?  Yes, some
of the existing LSM audit_match code uses -ENOENT but looking quickly
at those error conditions it seems that we could consider them
equivalent to a "missing" or "failed match" result and use -ENOENT for
both.  If you're really not happy with that overloading, we could use
something like -ENOMSG:missing/ok instead.

Thoughts?

I think we could just block it and see what happens.

--
paul-moore.com