linux-kernel - Re: + prctl-pr_set_mm-introduce-pr_set_mm_map-operation-v3.patch added to -mm tree

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Sat, 23 Aug 2014 18:18:20 +0400
From:	Cyrill Gorcunov <gorcunov@...il.com>
To:	Oleg Nesterov <oleg@...hat.com>
Cc:	Kees Cook <keescook@...omium.org>, Tejun Heo <tj@...nel.org>,
	Andrew Vagin <avagin@...nvz.org>,
	"Eric W. Biederman" <ebiederm@...ssion.com>,
	"H. Peter Anvin" <hpa@...or.com>,
	Serge Hallyn <serge.hallyn@...onical.com>,
	Pavel Emelyanov <xemul@...allels.com>,
	Vasiliy Kulikov <segoon@...nwall.com>,
	KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>,
	Michael Kerrisk <mtk.manpages@...il.com>,
	Julien Tinnes <jln@...gle.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	linux-kernel@...r.kernel.org
Subject: Re: + prctl-pr_set_mm-introduce-pr_set_mm_map-operation-v3.patch
 added to -mm tree

On Sat, Aug 23, 2014 at 03:30:01PM +0200, Oleg Nesterov wrote:
> forgot to mention,
> 
> On 08/23, Oleg Nesterov wrote:
> >
> > On 08/23, Cyrill Gorcunov wrote:
> >
> > > Looks like I need
> > > to use cred_guard_mutex instead of task_lock here, no?
> >
> > Please don't. First of all, it can't help because proc_pid_auxv() doesn't hold
> > this lock. It does mm_access() which drops this lock after return. And to remind,
> > we are going to remove mm_access/lock_trace from sys_read() paths in proc.
> 
> Besides, it can't help anyway. cred_guard_mutex is per-process (not per-thread),
> suppose that a vfork()'ed child does prctl() while another thread reads the
> parent's /proc/pid/auxv.

Then either I need to use some other lock (not sure which one) either leave it
completely unlocked mentionin in the man page such lockless behaviour. Thoughts?

> Cyrill, I am sorry, but I am starting to think that this patch should be
> dropped and replaced by another version. Or do you think it would be better
> to send the fixes on top?

It's not a problem to drop this particular patch (together with all fixes on
top) one and replace it with new version (this looks like a better idea than
drowning lkml with series of small patches). I rather need to understand what
exactly should be done in new version. So from your previous email

 | > Stricktly speaking yes, but don't forget we might need to update
 | > exe::file as well which requires lock to be taken.
 |
 | For reading? I see prctl_set_mm_exe_file_locked() in this patch, probably
 | this function was added by another patch. But, if this function calls
 | set_mm_exe_file() (I guess it does?) then down_read() is not enough?
 | set_mm_exe_file() can race with itself.

yes, for reading, look in set_mm_exe_file we lookup for vma which should
be not present when we change the link, and yes, because of read-only lock
this call can race but only one caller success there because we allow
to change exe link only once.

 | But for what? Ignoring the (I think buggy) check in do_shmat() ->start_stack
 | is simply unused, we only report it via /proc/. The same for, say, mm->start_code.

that't the good question if this check in do_shmat is buggy or not, why do
you think it's a bug there?

Oleg, letme summarize all the concerns maybe there would be a way to handle
them gracefully

1) How code flows for now (with all fixes on top of current Andrew's queued patches)

   - obtain struct prctl_mm_map from userspace
   - copy saved_auxv from userspace
   - down-read mmap_sem
   - validate all the data passed from userspace
     - we need a reference to stack-vma for RLIMIT_STACK check (this is doable,
       as you said, but until we drop the RLIMIT_STACK from do_shmat I would
       prefer it to be here)
     - we need to be sure that start_brk, brk, arg_start, arg_end, env_start, env_end
       really point to existing VMAs, strictly speaking the probgram can unmap
       all own VMAs except executable one and continue running without problem
       but this is not that practical I think and at first iteration I prefer
       more severe tests here on VMAs
   - setup new mm::exe_file (we need to be sure the old exe_file is unmapped
     so mmap_sem read-lock is needed)
   - update mm::saved_auxv with new values
   - finally setup new members to struct mm_struct
   - up-read mmap_sem

2) The qustion is do we really need that read-lock would be taken for all this
   time? And my answer is yes because of how I implement the checks for start_brk
   and etc.

Oleg, check please if I undersnad you correctly, you propose

 - drop off mmap_sem completely
 - don't verify for RLIMIT_STACK
 - drop off task_lock when updating mm::saved_auxv but still invent
   how to prevent update/read race

right?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/