[PATCH v2 01/15] mm/memory_hotplug: fix possible race in scan_movable_pages()

[David Hildenbrand (Arm)]

On 3/23/26 14:26, Lorenzo Stoakes (Oracle) wrote:
> On Fri, Mar 20, 2026 at 11:13:33PM +0100, David Hildenbrand (Arm) wrote:
>> If a hugetlb folio gets freed while we are in scan_movable_pages(),
>> folio_nr_pages() could return 0, resulting in or'ing "0 - 1 = -1"
>> to the PFN, resulting in PFN = -1. We're not holding any locks or
>> references that would prevent that.
>>
>> for_each_valid_pfn() would then search for the next valid PFN, and could
>> return a PFN that is outside of the range of the original requested
>> range. do_migrate_page() would then try to migrate quite a big range,
>> which is certainly undesirable.
>>
>> To fix it, simply test for valid folio_nr_pages() values. While at it,
>> as PageHuge() really just does a page_folio() internally, we can just
>> use folio_test_hugetlb() on the folio directly.
>>
>> scan_movable_pages() is expected to be fast, and we try to avoid taking
>> locks or grabbing references. We cannot use folio_try_get() as that does
>> not work for free hugetlb folios. We could grab the hugetlb_lock, but
>> that just adds complexity.
>>
>> The race is unlikely to trigger in practice, so we won't be CCing
>> stable.
>>
>> Fixes: 16540dae959d ("mm/hugetlb: mm/memory_hotplug: use a folio in scan_movable_pages()")
>> Signed-off-by: David Hildenbrand (Arm) <david at kernel.org>
> 
> Logic looks right to me, though some nits below. With those accounted for:
> 
> Reviewed-by: Lorenzo Stoakes (Oracle) <ljs at kernel.org>
> 
>> ---
>>  mm/memory_hotplug.c | 11 ++++++++---
>>  1 file changed, 8 insertions(+), 3 deletions(-)
>>
>> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
>> index 86d3faf50453..969cd7ddf68f 100644
>> --- a/mm/memory_hotplug.c
>> +++ b/mm/memory_hotplug.c
>> @@ -1747,6 +1747,7 @@ static int scan_movable_pages(unsigned long start, unsigned long end,
>>  	unsigned long pfn;
>>
>>  	for_each_valid_pfn(pfn, start, end) {
>> +		unsigned long nr_pages;
>>  		struct page *page;
>>  		struct folio *folio;
>>
>> @@ -1763,9 +1764,9 @@ static int scan_movable_pages(unsigned long start, unsigned long end,
>>  		if (PageOffline(page) && page_count(page))
>>  			return -EBUSY;
>>
>> -		if (!PageHuge(page))
> 
> Yeah interesting to see this is folio_test_hugetlb(page_folio(page)) :))
> 
> So this is a nice change for sure.
> 
>> -			continue;
>>  		folio = page_folio(page);
>> +		if (!folio_test_hugetlb(folio))
>> +			continue;
>>  		/*
>>  		 * This test is racy as we hold no reference or lock.  The
>>  		 * hugetlb page could have been free'ed and head is no longer
>> @@ -1775,7 +1776,11 @@ static int scan_movable_pages(unsigned long start, unsigned long end,
>>  		 */
>>  		if (folio_test_hugetlb_migratable(folio))
>>  			goto found;
>> -		pfn |= folio_nr_pages(folio) - 1;
>> +		nr_pages = folio_nr_pages(folio);
>> +		if (unlikely(nr_pages < 1 || nr_pages > MAX_FOLIO_NR_PAGES ||
> 
> NIT: since nr_pages is an unsigned long, would this be better as !nr_pages || ...?

It's easier on the brain when spotting that only a given range is
allowed, without having to remember the exact type of the variable :)

So I guess it doesn't really make a difference in the end.

> 
>> +			     !is_power_of_2(nr_pages)))
> 
> Could the latter two conditions ever really happen? I guess some weird tearing
> or something maybe?

Yes, or when the fields gets reused for something else.

> 
> It would also be nice to maybe separate this out as is_valid_nr_pages() or
> something, but then again, I suppose given this is a rare case of us
> checking this under circumstances where the value might not be valid, maybe
> not worth it.

I had the same thought. But this code is way too special regarding
raciness that I hope nobody else will really require this ... and if
they do, they might be doing something wrong :)

-- 
Cheers,

David