Messages in this thread | | | Subject | Re: [PATCH v3 06/11] rlimits: do security check under task_lock | From | Stephen Smalley <> | Date | Wed, 23 Jun 2010 14:37:43 -0400 |
| |
On Wed, 2010-06-23 at 19:44 +0200, Jiri Slaby wrote: > On 06/23/2010 06:12 PM, Oleg Nesterov wrote: > > On 06/23, Jiri Slaby wrote: > >> > >> On 06/07/2010 08:08 PM, Oleg Nesterov wrote: > >>> On 06/06, Jiri Slaby wrote: > >>>> @@ -1339,13 +1364,19 @@ int do_prlimit(struct task_struct *tsk, unsigned int resource, > >>>> > >>>> rlim = tsk->signal->rlim + resource; > >>>> task_lock(tsk->group_leader); > >>>> +again: > >>>> + retval = 0; > >>>> if (new_rlim) { > >>>> if ((new_rlim->rlim_max > rlim->rlim_max) && > >>>> !capable(CAP_SYS_RESOURCE)) > >> > >> BTW this capable() has the exactly same problem with being called with > >> task_lock held. Is it OK to move it completely out of critical section? > >> I'm asking because it sets a current->flags SU bit used for accounting. > >> If I move it out of the section, it will set the bit always. > > > > Well, with all these delays I do not know what "exactly same problem" > > means ;) Please explain? > > As I wrote: that the capable() is called with task_lock held. With > security enabled, capable() goes through all the avc_has_perm_noaudit, > avc_audit and similar (in selinux), the same as security_task_setrlimit > which we were writing about -- Andrew complaining about doing very long > security checks while holding spinlocks. > > I mean we should do either none of capable and selinux_task_setrlimit > under task_lock or both :).
IMO, just do them both under task_lock. All of the core SELinux permission checking code should be safe under task lock and the audit path is the exceptional case (only upon denial).
> >>> Finally. selinux_task_setrlimit(p) uses __task_cred(p) for the check. > >>> This looks a bit strange, different threads can have different creds > >>> but obviously rlimits are per-process. > >> > >> Sorry I can't see it. Could you point out in which function this is done? > > > > selinux_task_setrlimit()->current_has_perm()->current_sid()->current_cred() > > I still see no way how this is wrong. We want to check whether current > thread has capabilities to change (someone else's) rlimits. Maybe I'm > missing something? > > thanks,
-- Stephen Smalley National Security Agency
| |