Problem with 9ba09998baa9 ("selinux: Implement the watch_key security hook") in linux-next

[Richard Haines]

On Fri, 2020-04-17 at 11:48 -0400, Paul Moore wrote:
> I just notice that the "selinux: Implement the watch_key security
> hook" patch made it's way into linux-next via 9ba09998baa9:
> 
>   commit 9ba09998baa995518d94c9a32e6329b28ccb9045
>   Author: David Howells <dhowells at redhat.com>
>   Date:   Tue Jan 14 17:07:13 2020 +0000
> 
>    selinux: Implement the watch_key security hook
> 
>    Implement the watch_key security hook to make sure that a key
> grants the
>    caller View permission in order to set a watch on a key.
> 
>    For the moment, the watch_devices security hook is left
> unimplemented as
>    it's not obvious what the object should be since the queue is
> global and
>    didn't previously exist.
> 
>    Signed-off-by: David Howells <dhowells at redhat.com>
>    Acked-by: Stephen Smalley <sds at tycho.nsa.gov>
> 
> I'm reasonably confident that this code hasn't been tested as I
> expect
> it would fail, or at the very least behave in unintended ways.  The
> problem is the selinux_watch_key(...) function, shown below:

I built an selinx-testsuite test for this last year and it worked fine
then. I'll send the test as an RFC patch as its been some time since I
ran it. David also has a test in kernel
samples/watch_queue/watch_test.c

> 
> +static int selinux_watch_key(struct key *key)
> +{
> +       struct key_security_struct *ksec = key->security;
> +       u32 sid = current_sid();
> +
> +       return avc_has_perm(&selinux_state,
> +                           sid, ksec->sid, SECCLASS_KEY,
> KEY_NEED_VIEW, NULL);
> +}
> 
> ... in particular it is the fifth argument to avc_has_perm(),
> "KEY_NEED_VIEW" which is a problem.  KEY_NEED_VIEW is not a SELinux

True, however by magic the KEY_NEED_* perms match with the bits defined
in classmap.h. I did some work on this during the 'keys' saga, see
various emails in list like [1]

[1] 
https://lore.kernel.org/selinux/20200220181031.156674-2-richard_c_haines@btinternet.com/

> permission and would likely result in odd behavior when passed to
> avc_has_perm().  Given that the keyring permission to SELinux object
> class permission is variable depending on the key_perms policy
> capability, it probably makes the most sense to pull the permission
> mapping in selinux_key_permission() out into a separate function,
> e.g.
> key_perm_to_av(...) (see the other XXX_to_av() functions in
> security/selinux/hooks.c), and then use this newly created mapping
> function in both selinux_key_permission() and
> selinux_watch_key().  Or
> you could just duplicate the KEY_NEED_VIEW mapping code in both
> functions, but I would advise against that.
>