Re: [PATCH v3] bcachefs: fix shift oob in alloc_lru_idx_fragmentation

[Alan Huang <mmpgouride@gmail.com>]

On Oct 22, 2024, at 11:05, Hongbo Li <lihongbo22@huawei.com> wrote:
> 
> 
> 
> On 2024/10/22 10:38, Alan Huang wrote:
>> On Oct 22, 2024, at 10:26, Hongbo Li <lihongbo22@huawei.com> wrote:
>>> 
>>> 
>>> 
>>> On 2024/10/21 23:43, Jeongjun Park wrote:
>>>> The size of a.data_type is set abnormally large, causing shift-out-of-bounds.
>>>> To fix this, we need to add validation on a.data_type in
>>>> alloc_lru_idx_fragmentation().
>>>> Reported-by: syzbot+7f45fa9805c40db3f108@syzkaller.appspotmail.com
>>>> Fixes: 260af1562ec1 ("bcachefs: Kill alloc_v4.fragmentation_lru")
>>>> Signed-off-by: Jeongjun Park <aha310510@gmail.com>
>>>> ---
>>>>  fs/bcachefs/alloc_background.h | 3 +++
>>>>  1 file changed, 3 insertions(+)
>>>> diff --git a/fs/bcachefs/alloc_background.h b/fs/bcachefs/alloc_background.h
>>>> index f8e87c6721b1..163a67b97a40 100644
>>>> --- a/fs/bcachefs/alloc_background.h
>>>> +++ b/fs/bcachefs/alloc_background.h
>>>> @@ -168,6 +168,9 @@ static inline bool data_type_movable(enum bch_data_type type)
>>>>  static inline u64 alloc_lru_idx_fragmentation(struct bch_alloc_v4 a,
>>>>         struct bch_dev *ca)
>>>>  {
>>>> + if (a.data_type >= BCH_DATA_NR)
>>>> + return 0;
>>>> +
>>> 
>>> oh, I found I have done the same thing in [1]("Re: [syzbot] [bcachefs?] UBSAN: shift-out-of-bounds in bch2_alloc_to_text"). But
>> Your patch there is still triggering the issue.
> Yeah, it just notify the issue and not prevent the issue. So I found it should add a.data_type condition indeed. :)
>>> in my humble opinion, the validation changes also should be added. And in addition, move the condition about a.data_type into
>> There is already the validation:
>> bkey_fsck_err_on(alloc_data_type(a, a.data_type) != a.data_type
> 
> This is actually not enough. This only do some transition check. For example, if a.data_type break when bch2_bucket_sectors_dirty (the data corruption can lead to various situations occurring) is true, the helper does noting.

Make sense.

> 
> Thanks,
> Hongbo
> 
>> And the unknown data type is already printed in bch2_prt_data_type, additional validation doesn’t help much.
>>> data_type_movable will be better. Just my personal opinion.:)
>> In my personal opinion, I don’t think so :)
>>> 
>>> [1] https://www.spinics.net/lists/kernel/msg5412619.html
>>> 
>>> Thanks,
>>> Hongbo
>>> 
>>>>   if (!data_type_movable(a.data_type) ||
>>>>       !bch2_bucket_sectors_fragmented(ca, a))
>>>>   return 0;
>>>> --