RE: [PATCH v3 1/3] ptrace: Provide ___ptrace_may_access() that can be applied on arbitrary tasks

[Schaufler, Casey]

> -----Original Message-----
> From: Peter Zijlstra [mailto:peterz@xxxxxxxxxxxxx]
> Sent: Wednesday, September 05, 2018 12:03 PM
> To: Andrea Arcangeli <aarcange@xxxxxxxxxx>
> Cc: Jiri Kosina <jikos@xxxxxxxxxx>; Andi Kleen <ak@xxxxxxxxxxxxxxx>; Tim Chen
> <tim.c.chen@xxxxxxxxxxxxxxx>; Schaufler, Casey <casey.schaufler@xxxxxxxxx>;
> Thomas Gleixner <tglx@xxxxxxxxxxxxx>; Ingo Molnar <mingo@xxxxxxxxxx>;
> Josh Poimboeuf <jpoimboe@xxxxxxxxxx>; Woodhouse, David
> <dwmw@xxxxxxxxxxxx>; Oleg Nesterov <oleg@xxxxxxxxxx>; linux-
> kernel@xxxxxxxxxxxxxxx; x86@xxxxxxxxxx
> Subject: Re: [PATCH v3 1/3] ptrace: Provide ___ptrace_may_access() that can
> be applied on arbitrary tasks
> 
> On Wed, Sep 05, 2018 at 02:40:18PM -0400, Andrea Arcangeli wrote:
> > [ 1838.769917]  <<EOE>>  [<ffffffff816391e5>]
> avc_compute_av+0x126/0x1b5
> 
> That does read_lock(), which is not allowed from scheduler context.
> 
> > [ 1838.777125]  [<ffffffff810b842e>] ? walk_tg_tree_from+0xbe/0x110
> > [ 1838.783828]  [<ffffffff8128b9c4>] avc_has_perm_noaudit+0xc4/0x110
> 
> In current code this can end up in avc_update_node() which uses
> spin_lock(), which is a bug from scheduler context.o
> 
> > [ 1838.790628]  [<ffffffff8128f1fb>] cred_has_capability+0x6b/0x120
> > [ 1838.797331]  [<ffffffff810db71c>] ? ktime_get+0x4c/0xd0
> > [ 1838.803160]  [<ffffffff810e167b>] ?
> clockevents_program_event+0x6b/0xf0
> > [ 1838.810532]  [<ffffffff8128f2de>] selinux_capable+0x2e/0x40
> > [ 1838.816748]  [<ffffffff81288f65>] security_capable_noaudit+0x15/0x20
> > [ 1838.823829]  [<ffffffff8108b975>] has_ns_capability_noaudit+0x15/0x20
> > [ 1838.831014]  [<ffffffff8108bc55>] ptrace_has_cap+0x35/0x40
> > [ 1838.837126]  [<ffffffff8108c717>] ___ptrace_may_access+0xa7/0x1e0
> > [ 1838.843925]  [<ffffffff8163f0ae>] __schedule+0x26e/0xa00
> > [ 1838.849855]  [<ffffffff81640949>] schedule_preempt_disabled+0x29/0x70
> > [ 1838.857041]  [<ffffffff810d9324>] cpu_startup_entry+0x184/0x290
> > [ 1838.863637]  [<ffffffff8104891a>] start_secondary+0x1da/0x250
> 
> So yes, looks like all that security LSM nonsense isn't going to work
> here.

What won't work is using the ptrace code. That is one of the reasons why
you can't just blindly use it. Look at the patch set I submitted and you'll see
that the SELinux selinux_task_safe_sidechannel() hook does not do the things
that cause the lockup.