Re: [PATCH] mm/hugetlbfs: Unmap pages if page fault raced with hole punch

[Mike Kravetz]

On 01/11/2016 02:35 PM, Andrew Morton wrote:
> On Wed,  6 Jan 2016 14:37:04 -0800 Mike Kravetz <mike.kravetz@xxxxxxxxxx> wrote:
> 
>> Page faults can race with fallocate hole punch.  If a page fault happens
>> between the unmap and remove operations, the page is not removed and
>> remains within the hole.  This is not the desired behavior.  The race
>> is difficult to detect in user level code as even in the non-race
>> case, a page within the hole could be faulted back in before fallocate
>> returns.  If userfaultfd is expanded to support hugetlbfs in the future,
>> this race will be easier to observe.
>>
>> If this race is detected and a page is mapped, the remove operation
>> (remove_inode_hugepages) will unmap the page before removing.  The unmap
>> within remove_inode_hugepages occurs with the hugetlb_fault_mutex held
>> so that no other faults will be processed until the page is removed.
>>
>> The (unmodified) routine hugetlb_vmdelete_list was moved ahead of
>> remove_inode_hugepages to satisfy the new reference.
>>
>> ...
>>
>> --- a/fs/hugetlbfs/inode.c
>> +++ b/fs/hugetlbfs/inode.c
>>
>> ...
>>
>> @@ -395,37 +431,43 @@ static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
>>  							mapping, next, 0);
>>  			mutex_lock(&hugetlb_fault_mutex_table[hash]);
>>  
>> -			lock_page(page);
>> -			if (likely(!page_mapped(page))) {
> 
> hm, what are the locking requirements for page_mapped()?

page_mapped is just reading/evaluating an atomic within the struct page
which we have a referene on from the pagevec_lookup.  But, I think the
question is what prevents page_mapped from changing after we check it?

The patch takes the hugetlb_fault_mutex_table lock before checking
page_mapped.  If the page is unmapped and the hugetlb_fault_mutex_table
is held, it can not be faulted in and change from unmapped to mapped.

The new comment in the patch about taking hugetlb_fault_mutex_table is
right before the check for page_mapped.

> 
>> -				bool rsv_on_error = !PagePrivate(page);
>> -				/*
>> -				 * We must free the huge page and remove
>> -				 * from page cache (remove_huge_page) BEFORE
>> -				 * removing the region/reserve map
>> -				 * (hugetlb_unreserve_pages).  In rare out
>> -				 * of memory conditions, removal of the
>> -				 * region/reserve map could fail.  Before
>> -				 * free'ing the page, note PagePrivate which
>> -				 * is used in case of error.
>> -				 */
>> -				remove_huge_page(page);
> 
> And remove_huge_page().

The page must be locked before calling remove_huge_page, since it will
call delete_from_page_cache.  It currently is locked.  Would you prefer
a comment stating this before the call?

-- 
Mike Kravetz

> 
>> -				freed++;
>> -				if (!truncate_op) {
>> -					if (unlikely(hugetlb_unreserve_pages(
>> -							inode, next,
>> -							next + 1, 1)))
>> -						hugetlb_fix_reserve_counts(
>> -							inode, rsv_on_error);
>> -				}
>>
>> ...
>>