lkml.org 
[lkml]   [2016]   [Aug]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH] kasan: avoid overflowing quarantine size on low memory systems
On Tue, Aug 2, 2016 at 12:07 PM, Alexander Potapenko <glider@google.com> wrote:
> On Tue, Aug 2, 2016 at 12:05 PM, Dmitry Vyukov <dvyukov@google.com> wrote:
>> On Tue, Aug 2, 2016 at 12:00 PM, Andrey Ryabinin
>> <aryabinin@virtuozzo.com> wrote:
>>>
>>>
>>> On 08/01/2016 05:59 PM, Alexander Potapenko wrote:
>>>> If the total amount of memory assigned to quarantine is less than the
>>>> amount of memory assigned to per-cpu quarantines, |new_quarantine_size|
>>>> may overflow. Instead, set it to zero.
>>>>
>>>
>>> Just curious, how did find this?
>>> Overflow is possible if system has more than 32 cpus per GB of memory. AFIAK this quite unusual.
>>
>> I was reading code for unrelated reason.
>>
>>>> Reported-by: Dmitry Vyukov <dvyukov@google.com>
>>>> Fixes: 55834c59098d ("mm: kasan: initial memory quarantine
>>>> implementation")
>>>> Signed-off-by: Alexander Potapenko <glider@google.com>
>>>> ---
>>>> mm/kasan/quarantine.c | 12 ++++++++++--
>>>> 1 file changed, 10 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/mm/kasan/quarantine.c b/mm/kasan/quarantine.c
>>>> index 65793f1..416d3b0 100644
>>>> --- a/mm/kasan/quarantine.c
>>>> +++ b/mm/kasan/quarantine.c
>>>> @@ -196,7 +196,7 @@ void quarantine_put(struct kasan_free_meta *info, struct kmem_cache *cache)
>>>>
>>>> void quarantine_reduce(void)
>>>> {
>>>> - size_t new_quarantine_size;
>>>> + size_t new_quarantine_size, percpu_quarantines;
>>>> unsigned long flags;
>>>> struct qlist_head to_free = QLIST_INIT;
>>>> size_t size_to_free = 0;
>>>> @@ -214,7 +214,15 @@ void quarantine_reduce(void)
>>>> */
>>>> new_quarantine_size = (READ_ONCE(totalram_pages) << PAGE_SHIFT) /
>>>> QUARANTINE_FRACTION;
>>>> - new_quarantine_size -= QUARANTINE_PERCPU_SIZE * num_online_cpus();
>>>> + percpu_quarantines = QUARANTINE_PERCPU_SIZE * num_online_cpus();
>>>> + if (new_quarantine_size < percpu_quarantines) {
>>>> + WARN_ONCE(1,
>>>> + "Too little memory, disabling global KASAN quarantine.\n",
>>>> + );
>>>
>>> Why WARN? I'd suggest pr_warn_once();
>>
>>
>> I would suggest to just do something useful. Setting quarantine
>> new_quarantine_size to 0 looks fine.
>> What would user do with this warning? Number of CPUs and amount of
>> memory are generally fixed. Why is it an issue for end user at all? We
>> still have some quarantine per-cpu. A WARNING means a [non-critical]
>> kernel bug. E.g. syzkaller will catch each and every boot of such
>> system as a bug.
> How about printk_once then?
> Silently setting the quarantine size to zero may puzzle the user.


We still have per-cpu quarantine.
new_quarantine_size==0 is not radically different from
new_quarantine_size==1. Both limit KASAN ability to detect UAF. Why do
we WARN in the former case but not in the latter?
We can print per-cpu/global quarantine sizes to console. Then if we
got any complaints we can figure out what happens from the log.



>>>> + new_quarantine_size = 0;
>>>> + } else {
>>>> + new_quarantine_size -= percpu_quarantines;
>>>> + }
>>>> WRITE_ONCE(quarantine_size, new_quarantine_size);
>>>>
>>>> last = global_quarantine.head;
>>>>
>
>
>
> --
> Alexander Potapenko
> Software Engineer
>
> Google Germany GmbH
> Erika-Mann-Straße, 33
> 80636 München
>
> Geschäftsführer: Matthew Scott Sucherman, Paul Terence Manicle
> Registergericht und -nummer: Hamburg, HRB 86891
> Sitz der Gesellschaft: Hamburg

\
 
 \ /
  Last update: 2016-08-02 13:01    [W:0.335 / U:0.124 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site