Re: [PATCH] mm: cachestat: avoid bogus workingset test during swapping & invalidation races

[Chengming Zhou]

On 2024/3/15 17:30, Johannes Weiner wrote:
> On Fri, Mar 15, 2024 at 11:16:35AM +0800, Chengming Zhou wrote:
>> On 2024/3/15 00:49, Johannes Weiner wrote:
>>> When cachestat against shmem races with swapping and invalidation, the
>>> shadow entry might not exist: swapout IO is still in progress and
>>> we're before __remove_mapping; or swapin/invalidation/swapoff has
>>> removed the shadow from swapcache after we saw a shmem swap entry.
>>>
>>> This will send a NULL to workingset_test_recent(). The latter purely
>>> operates on pointer bits, so it won't crash - node 0, memcg ID 0,
>>> eviction timestamp 0, etc. are all valid inputs - but it's a bogus
>>> test. In theory that could result in a false "recently evicted" count.
>>
>> Good catch!
>>
>>>
>>> Such a false positive wouldn't be the end of the world. But for code
>>> clarity and (future) robustness, be explicit about this case.
>>>
>>> Fixes: cf264e1329fb ("cachestat: implement cachestat syscall")
>>> Reported-by: Jann Horn <jannh@xxxxxxxxxx>
>>> Signed-off-by: Johannes Weiner <hannes@xxxxxxxxxxx>
>>> ---
>>>  mm/filemap.c | 3 +++
>>>  1 file changed, 3 insertions(+)
>>>
>>> diff --git a/mm/filemap.c b/mm/filemap.c
>>> index 222adac7c9c5..a07c27df7eab 100644
>>> --- a/mm/filemap.c
>>> +++ b/mm/filemap.c
>>> @@ -4199,6 +4199,9 @@ static void filemap_cachestat(struct address_space *mapping,
>>>  				swp_entry_t swp = radix_to_swp_entry(folio);
>>>  
>>
>> IIUC, we should first check if it's a real swap entry using non_swap_entry(), right?
>> Since there maybe other types of entries in shmem.
> 
> Good point, it could be a poisoned entry. I'll add the
> non_swap_entry() check on swp.
> 
>> And need to get_swap_device() to prevent concurrent swapoff here,
>> get_shadow_from_swap_cache() won't do it for us.
> 
> We're holding rcu_read_lock() for the xarray iteration, so if we see
> the swap entry in the shmem mapping, it means we beat shmem_unuse()
> and swapoff hasn't run synchronize_rcu() yet.

Ah, you are right, so it's safe.

> 
> So it's safe. But I think it could use a comment. Maybe the
> documentation of get_swap_device() should mention this option too?