[kernel-hardening] Re: [RFC 2/5 v4] procfs: add hidepid= and gid= mount options

[ebiederm@xmission.com (Eric W. Biederman)]

Vasiliy Kulikov <segoon@openwall.com> writes:

> (cc'ed Eric)
>
> On Mon, Jun 20, 2011 at 20:43 +1000, James Morris wrote:
>> > > > hidepid=1 means users may not access any /proc/<pid>/ directories, but their
>> > > > own.  Sensitive files like cmdline, io, sched*, status, wchan are now
>> > > > protected against other users.  As permission checking done in
>> > > > proc_pid_permission() and files' permissions are left untouched,
>> > > > programs expecting specific files' permissions are not confused.
>> > > 
>> > > IMHO such programs are beyond broken and have voided their kernel 
>> > > warranty.
>> > 
>> > Policykit, Debian's start-stop-daemon, util-linux use /proc/PID's uid.
>> > procps use both /proc/PID's uid and gid.  Are all of them totally broken?
>> 
>> If they depend on specific permissions, yes.
>
> Could you please then clarify why does this patch changes
> pid_revalidate() behaviour:
>
> http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=99f895518368252ba862cc15ce4eb98ebbe1bec6
>
> It changes files permissions to allow userspace apps to quickly stat
> files, not looking into /proc/PID/status.  So, uid and gid are explicit
> ABI.  Breaking procfs uid/gid attributes would break these apps.
>
> Or am I missing something?

No, you are not missing something.  The uid and gid of proc files are
part of the userspace ABI.  The files are owned by the uid and gid
of the respective process.  The commit in question did not change
the ABI it documented it because I found out the hardware way that
a small change there breaks userspace.

Eric