Re: [PATCH 2/2] RFC: selinux: sysctl: fix selinux labeling broken by last patch

[Stephen Smalley <sds@tycho.nsa.gov>]

On Mon, 2011-01-31 at 05:26 +0200, Lucian Adrian Grijincu wrote:
> Eric's patch was rejected because it broke selinux labeling:
> http://thread.gmane.org/gmane.linux.kernel.lsm/9807/focus=9841
> 
>        This seems to break labeling.  Prior to this patch, I see:
> 
>        # ls -lZ /proc/1/net/rpc/nfsd.fh
>        -rw-------. root root system_u:object_r:sysctl_rpc_t:s0 channel
> 
>        with the patch:
> 
>        # ls -lZ /proc/1/net/rpc/nfsd.fh
>        -rw-------. root root system_u:object_r:proc_t:s0      channel
> 
> My patch seems to have fixed this problem (it correctly reports
> sysctl_rpc_t in this case), but my selinux experience is ε > 0 and I
> have no ideea what else it may have broken.
> 
> I ran 'find /proc | xargs ls -Z > f' on a kernel with an without
> these patches:
> * http://swarm.cs.pub.ro/~lucian/store/ls-Z-with-patch
> * http://swarm.cs.pub.ro/~lucian/store/ls-Z-without-patch
> 
> My setup is a custom busybox live CD with selinux enabled, with
> /etc/selinux and /usr/share/selinux/default copied from Ubuntu 10.10's
> selinux-policy-default package. I'm sure there are lots of reasons why
> this is not correcly configured, etc., but I have no experience
> regarding selinux. I can make all the scripts/sources/configs/etc
> available to anyone interested.
> 
> NOTE: this patch will be merged with:
>   security/selinux: Simplify proc inode to security label mapping
> 
> I'm only prividing this patch separately to point out the differences
> to Eric W. Biederman's patch.
> 
> Both of these patches apply cleanly agains Linux 2.6.37.
> 
> Signed-off-by: Lucian Adrian Grijincu <lucian.grijincu@gmail.com>
> ---
>  fs/proc/proc_sysctl.c    |    1 -
>  security/selinux/hooks.c |   20 ++++++++++++++++----
>  2 files changed, 16 insertions(+), 5 deletions(-)

It would be better to include the patch inline for review.  In any
event, a few observations on your patch:
- We don't want to replace "incestuous" knowledge of proc with
"incestous" knowledge of the dcache.  So rather than encoding knowledge
of the magical "//deleted" suffix into selinux, use an interface to the
dcache (or add one if none exists) that does not append that suffix at
all.  I think apparmor did something similar to deal with the (deleted)
suffix for d_path.

- You don't need special handling of /proc/PID entries.  Those are
labeled via the security_task_to_inode -> selinux_task_to_inode hook,
called from proc_pid_make_inode and the _revalidate functions.

- Don't remove the IS_PRIVATE() test from inode_has_perm(), as other
inodes beyond just the /proc/sys ones are marked with that flag
(original usage was for reiserfs xattr inodes).

-- 
Stephen Smalley
National Security Agency