Re: [PATCH] security/selinux: fix /proc/sys/ labeling

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

On Tue, 2011-02-01 at 18:32 +0200, Lucian Adrian Grijincu wrote:
> On Tue, Feb 1, 2011 at 5:59 PM, Stephen Smalley <sds@tycho.nsa.gov> wrote:
> >> Just the earlier one. I added his sign-off because of this paragraph
> >> in SubmittingPatches:
> >> | The Signed-off-by: tag indicates that the signer was involved in the
> >> | development of the patch, or that he/she was in the patch's delivery path.
> 
> 
> So should I leave Eric's sign-off here?

I guess so, given that paragraph.

> >> Without we label all nodes in /proc/ through selinux_proc_get_sid.
> >>
> >> /proc/1/limits should not get it's sid from here, but from
> >> security_task_to_inode -> selinux_task_to_inode.
> >>
> >> Without the check we send "/1/limits" to selinux_proc_get_sid, which
> >> strips off "/1" leaving "/limits". This will be labeled with "proc_t"
> >> IIRC.
> >
> > Are you sure?  Those inodes should be labeled by proc_pid_make_inode()
> > -> security_task_to_inode() -> selinux_task_to_inode(), which will set
> > the inode SID to match the associated task SID, and set the
> > isec->initialized flag.  Then when inode_doinit_with_dentry gets called
> > later, it should bail immediately due to isec->initialized already being
> > set.
> 
> 
> 
> I'll post an updated patch without those checks. I tested and 'find
> /proc | xargs ls -Z' said the same thing with and without those
> checks.
> 
> I remember doing the same test yesterday and saw some differences, but
> I must have compared the wrong files.

Ok, good.  That gets rid of the last vestige of proc implementation
details in selinux.

-- 
Stephen Smalley
National Security Agency