Re: [PATCH v4 1/2] mm/memory-failure.c: avoid calling invalidate_inode_page() with unexpected pages

[Miaohe Lin]

On 2022/3/20 0:20, Matthew Wilcox wrote:
> On Sun, Mar 20, 2022 at 01:13:33PM +0800, Miaohe Lin wrote:
>> invalidate_inode_page() can invalidate the pages in the swap cache because
>> the check of page->mapping != mapping is removed via Matthew's patch titled
>> "mm/truncate: Inline invalidate_complete_page() into its one caller". But
>> invalidate_inode_page() is not expected to deal with the pages in the swap
>> cache. Also non-lru movable page can reach here too. They're not page cache
>> pages. Skip these pages by checking PageSwapCache and PageLRU to fix this
>> unexpected issue.
> 
> I disagree with this changelog.
> 
> invalidate_inode_page() should not be called for pages which are not
> in the page cache.
>
> And then the patch shouldn't test PageLRU (which is actually wrong) or
> PageSwapCache().  It should simply be:
> > +	if (!PageHuge(page) && !PageAnon(page))

If we reach here, the page can be one of the PageHuge, __PageMovable and PageLRU.
If above code is used, __PageMovable can pass the check and enter the invalidate_inode_page
unexpectedly. So I think PageLRU check is necessary here. But it seems PageSwapCache
is only reliable when PageAnon, so might we should do:

+	if (!PageHuge(page) && PageLRU(page) && !PageAnon(page))

Am I miss something? Many thanks for comment.

> 
>> Signed-off-by: Miaohe Lin <linmiaohe@xxxxxxxxxx>
>> ---
>>  mm/memory-failure.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
>> index 5444a8ef4867..ecf45961f3b6 100644
>> --- a/mm/memory-failure.c
>> +++ b/mm/memory-failure.c
>> @@ -2178,7 +2178,7 @@ static int __soft_offline_page(struct page *page)
>>  		return 0;
>>  	}
>>  
>> -	if (!PageHuge(page))
>> +	if (!PageHuge(page) && PageLRU(page) && !PageSwapCache(page))
>>  		/*
>>  		 * Try to invalidate first. This should work for
>>  		 * non dirty unmapped page cache pages.
>> -- 
>> 2.23.0
>>
>>
> 
> .
>