Re: [PATCH v2] sbitmap: fix io hung due to race on sbitmap_word::cleared

[Pavel Begunkov <asml.silence@gmail.com>]

On 6/4/24 08:03, YangYang wrote:
> On 2024/6/4 14:12, Yu Kuai wrote:
>> Hi,
>>
>> 在 2024/06/04 11:25, Ming Lei 写道:
>>> On Tue, Jun 4, 2024 at 11:12 AM Yang Yang <yang.yang@vivo.com> wrote:
>>>>
>>>> Configuration for sbq:
>>>>    depth=64, wake_batch=6, shift=6, map_nr=1
>>>>
>>>> 1. There are 64 requests in progress:
>>>>    map->word = 0xFFFFFFFFFFFFFFFF
>>>> 2. After all the 64 requests complete, and no more requests come:
>>>>    map->word = 0xFFFFFFFFFFFFFFFF, map->cleared = 0xFFFFFFFFFFFFFFFF
>>>> 3. Now two tasks try to allocate requests:
>>>>    T1:                                       T2:
>>>>    __blk_mq_get_tag                          .
>>>>    __sbitmap_queue_get                       .
>>>>    sbitmap_get                               .
>>>>    sbitmap_find_bit                          .
>>>>    sbitmap_find_bit_in_word                  .
>>>>    __sbitmap_get_word  -> nr=-1              __blk_mq_get_tag
>>>>    sbitmap_deferred_clear                    __sbitmap_queue_get
>>>>    /* map->cleared=0xFFFFFFFFFFFFFFFF */     sbitmap_find_bit
>>>>      if (!READ_ONCE(map->cleared))           sbitmap_find_bit_in_word
>>>>        return false;                         __sbitmap_get_word -> nr=-1
>>>>      mask = xchg(&map->cleared, 0)           sbitmap_deferred_clear
>>>>      atomic_long_andnot()                    /* map->cleared=0 */
>>>>                                                if (!(map->cleared))
>>>>                                                  return false;
>>>>                                       /*
>>>>                                        * map->cleared is cleared by T1
>>>>                                        * T2 fail to acquire the tag
>>>>                                        */
>>>>
>>>> 4. T2 is the sole tag waiter. When T1 puts the tag, T2 cannot be woken
>>>> up due to the wake_batch being set at 6. If no more requests come, T1
>>>> will wait here indefinitely.
>>>>
>>>> To fix this issue, simply revert commit 661d4f55a794 ("sbitmap:
>>>> remove swap_lock"), which causes this issue.
>>>
>>> I'd suggest to add the following words in commit log:
>>>
>>> Check on ->cleared and update on both ->cleared and ->word need to be
>>> done atomically, and using spinlock could be the simplest solution.
>>>
>>> Otherwise, the patch looks fine for me.
>>
>> Maybe I'm noob, but I'm confused how can this fix the problem, looks
>> like the race condition doesn't change.
>>
>> In sbitmap_find_bit_in_word:
>>
>> 1) __sbitmap_get_word read word;
>> 2) sbitmap_deferred_clear clear cleared;
>> 3) sbitmap_deferred_clear update word;
>>
>> 2) and 3) are done atomically while 1) can still concurrent with 3):
>>
>> t1:
>> sbitmap_find_bit_in_word
>>   __sbitmap_get_word
>>   -> read old word, return -1 >          t2:
>>          sbitmap_find_bit_in_word
>>           __sbitmap_get_word
>>           -> read old word, return -1
>>   sbitmap_deferred_clear
>>   -> clear cleared and update word
>>          sbitmap_deferred_clear
>>          -> cleared is cleared, fail
> 
> Yes, you are right, this patch cannot fix this issue.

One other alternative is to kill ->cleared. It's not
immediately clear how important it is. Do we have any
numbers?


>> BYW, I still think it's fine to fix this problem by trying the
>> __sbitmap_get_word() at least one more time if __sbitmap_get_word()
>> failed.
> 
> Err, after trying one more time __sbitmap_get_word() may still fail.

-- 
Pavel Begunkov