Re: [Bug 210655] ptrace.2: documentation is incorrect about access checking threads in same thread group

[Jann Horn]

On Wed, Dec 16, 2020 at 12:25 AM Alejandro Colomar (man-pages)
<alx.manpages@xxxxxxxxx> wrote:
> On 12/16/20 12:23 AM, Alejandro Colomar (man-pages) wrote:
> > On 12/16/20 12:07 AM, Jann Horn wrote:
> >> Am Tue, Dec 15, 2020 at 06:01:25PM +0100 schrieb Alejandro Colomar (man-pages):
> >>> There's a bug report: https://bugzilla.kernel.org/show_bug.cgi?id=210655
> >>>
> >>> [[
> >>> Under "Ptrace access mode checking", the documentation states:
> >>>   "1. If the calling thread and the target thread are in the same thread
> >>> group, access is always allowed."
> >>>
> >>> This is incorrect. A thread may never attach to another in the same group.
> >>
> >> No, that is correct. ptrace-mode access checks do always short-circuit for
> >> tasks in the same thread group:
> >>
> >> /* Returns 0 on success, -errno on denial. */
> >> static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
> >> {
> >> [...]
> >>         /* May we inspect the given task?
> >>          * This check is used both for attaching with ptrace
> >>          * and for allowing access to sensitive information in /proc.
> >>          *
> >>          * ptrace_attach denies several cases that /proc allows
> >>          * because setting up the necessary parent/child relationship
> >>          * or halting the specified task is impossible.
> >>          */
> >>
> >>         /* Don't let security modules deny introspection */
> >>         if (same_thread_group(task, current))
> >>                 return 0;
> >> [...]
> >> }
> >
> > AFAICS, that code always returns non-zero,
>
> Sorry, I should have said "that code never returns 0".
>
> > at least when called from ptrace_attach().

Yes.

> > As you can see below,
> > __ptrace_may_access() is called some lines after
> > the code pointed to by the bug report.
> >
> >
> > static int ptrace_attach(struct task_struct *task, long request,
> >                        unsigned long addr,
> >                        unsigned long flags)
> > {
> > [...]
> >       if (same_thread_group(task, current))
> >               goto out;
> >
> >       /*
> >        * Protect exec's credential calculations against our interference;
> >        * SUID, SGID and LSM creds get determined differently
> >        * under ptrace.
> >        */
> >       retval = -ERESTARTNOINTR;
> >       if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
> >               goto out;
> >
> >       task_lock(task);
> >       retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
> > [...]
> > }

I said exactly that in my last mail:

> >> As the comment explains, you can't actually *attach*
> >> to another task in the same thread group; but that's
> >> not because of the ptrace-style access check rules,
> >> but because specifically *attaching* to another task
> >> in the same thread group doesn't work.

As I said, attaching indeed doesn't work. But that's not what "Ptrace
access mode checking" means. As the first sentence of that section
says:

| Various parts of the kernel-user-space API (not just ptrace()
| operations), require so-called "ptrace access mode" checks,
| whose outcome determines whether an operation is
| permitted (or, in a  few cases,  causes  a "read" operation
| to return sanitized data).

You can find these places by grepping for \bptrace_may_access\b -
operations like e.g. the get_robust_list() syscall will always succeed
when inspecting other tasks in the caller's thread group thanks to
this rule.