Re: KMSAN: uninit-value in mpol_rebind_mm
From: Vlastimil Babka
Date: Thu Jan 03 2019 - 06:41:51 EST
On 1/3/19 12:14 PM, Alexander Potapenko wrote:
> On Thu, Jan 3, 2019 at 9:42 AM Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
>>
>> On Thu, Jan 3, 2019 at 9:36 AM Vlastimil Babka <vbabka@xxxxxxx> wrote:
>>>
>>>
>>> On 12/31/18 8:51 AM, syzbot wrote:
>>>> Hello,
>>>>
>>>> syzbot found the following crash on:
>>>>
>>>> HEAD commit: 79fc24ff6184 kmsan: highmem: use kmsan_clear_page() in cop..
>>>> git tree: kmsan
>>>> console output: https://syzkaller.appspot.com/x/log.txt?x=13c48b67400000
>>>> kernel config: https://syzkaller.appspot.com/x/.config?x=901dd030b2cc57e7
>>>> dashboard link: https://syzkaller.appspot.com/bug?extid=b19c2dc2c990ea657a71
>>>> compiler: clang version 8.0.0 (trunk 349734)
>>>>
>>>> Unfortunately, I don't have any reproducer for this crash yet.
>>>>
>>>> IMPORTANT: if you fix the bug, please add the following tag to the commit:
>>>> Reported-by: syzbot+b19c2dc2c990ea657a71@xxxxxxxxxxxxxxxxxxxxxxxxx
>>>>
>>>> ==================================================================
>>>> BUG: KMSAN: uninit-value in mpol_rebind_policy mm/mempolicy.c:353 [inline]
>>>> BUG: KMSAN: uninit-value in mpol_rebind_mm+0x249/0x370 mm/mempolicy.c:384
>>>
>>> The report doesn't seem to indicate where the uninit value resides in
>>> the mempolicy object.
>>
>> Yes, it doesn't and it's not trivial to do. The tool reports uses of
>> unint _values_. Values don't necessary reside in memory. It can be a
>> register, that come from another register that was calculated as a sum
>> of two other values, which may come from a function argument, etc.
>>
>>> I'll have to guess. mm/mempolicy.c:353 contains:
>>>
>>> if (!mpol_store_user_nodemask(pol) &&
>>> nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
>>>
>>> "mpol_store_user_nodemask(pol)" is testing pol->flags, which I couldn't
>>> see being uninitialized after leaving mpol_new(). So I'll guess it's
>>> actually about accessing pol->w.cpuset_mems_allowed on line 354.
>>>
>>> For w.cpuset_mems_allowed to be not initialized and the nodes_equal()
>>> reachable for a mempolicy where mpol_set_nodemask() is called in
>>> do_mbind(), it seems the only possibility is a MPOL_PREFERRED policy
>>> with empty set of nodes, i.e. MPOL_LOCAL equivalent. Let's see if the
>>> patch below helps. This code is a maze to me. Note the uninit access
>>> should be benign, rebinding this kind of policy is always a no-op.
> If I'm reading mempolicy.c right, `pol->flags & MPOL_F_LOCAL` doesn't
> imply `pol->mode == MPOL_PREFERRED`, shouldn't we check for both here?
I think it does? Only preferred mempolicies set it, including
default_policy, and MPOL_LOCAL is converted to MPOL_PREFERRED
internally. Anyway we would need the opposite implication here to be
safe, and that's also true.
>>> ----8<----
>>> From ff0ca29da6bc2572d7b267daa77ced6083e3f02d Mon Sep 17 00:00:00 2001
>>> From: Vlastimil Babka <vbabka@xxxxxxx>
>>> Date: Thu, 3 Jan 2019 09:31:59 +0100
>>> Subject: [PATCH] mm, mempolicy: fix uninit memory access
>>>
>>> ---
>>> mm/mempolicy.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/mm/mempolicy.c b/mm/mempolicy.c
>>> index d4496d9d34f5..a0b7487b9112 100644
>>> --- a/mm/mempolicy.c
>>> +++ b/mm/mempolicy.c
>>> @@ -350,7 +350,7 @@ static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask)
>>> {
>>> if (!pol)
>>> return;
>>> - if (!mpol_store_user_nodemask(pol) &&
>>> + if (!mpol_store_user_nodemask(pol) && !(pol->flags & MPOL_F_LOCAL) &&
>>> nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
>>> return;
>>>
>>> --
>>> 2.19.2
>>>
>>> --
>>> You received this message because you are subscribed to the Google Groups "syzkaller-bugs" group.
>>> To unsubscribe from this group and stop receiving emails from it, send an email to syzkaller-bugs+unsubscribe@xxxxxxxxxxxxxxxxx
>>> To view this discussion on the web visit https://groups.google.com/d/msgid/syzkaller-bugs/a71997c3-e8ae-a787-d5ce-3db05768b27c%40suse.cz.
>>> For more options, visit https://groups.google.com/d/optout.
>
>
>