[PATCH] mm/page_isolation: fix potential missing call to unset_migratetype_isolate()

[PATCH] mm/page_isolation: fix potential missing call to unset_migratetype_isolate()

[Miaohe Lin]

In start_isolate_page_range() undo path, pfn_to_online_page() just checks
the first pfn in a pageblock while __first_valid_page() will traverse the
pageblock until the first online pfn is found. So we may miss the call to
unset_migratetype_isolate() in undo path and pages will remain isolated
unexpectedly. Fix this by calling undo_isolate_page_range() and this will
also help to remove some duplicated codes.

Fixes: 2ce13640b3f4 ("mm: __first_valid_page skip over offline pages")
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
---
 mm/page_isolation.c | 9 +--------
 1 file changed, 1 insertion(+), 8 deletions(-)

diff --git a/mm/page_isolation.c b/mm/page_isolation.c
index 471e3a13b541..9bb562d5d194 100644
--- a/mm/page_isolation.c
+++ b/mm/page_isolation.c
@@ -202,14 +202,7 @@ int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
 	}
 	return 0;
 undo:
-	for (pfn = start_pfn;
-	     pfn < undo_pfn;
-	     pfn += pageblock_nr_pages) {
-		struct page *page = pfn_to_online_page(pfn);
-		if (!page)
-			continue;
-		unset_migratetype_isolate(page, migratetype);
-	}
+	undo_isolate_page_range(start_pfn, undo_pfn, migratetype);
 
 	return -EBUSY;
 }
-- 
2.23.0

Re: [PATCH] mm/page_isolation: fix potential missing call to unset_migratetype_isolate()

[David Hildenbrand]

On 04.09.21 11:20, Miaohe Lin wrote:
> In start_isolate_page_range() undo path, pfn_to_online_page() just checks
> the first pfn in a pageblock while __first_valid_page() will traverse the
> pageblock until the first online pfn is found. So we may miss the call to
> unset_migratetype_isolate() in undo path and pages will remain isolated
> unexpectedly. Fix this by calling undo_isolate_page_range() and this will
> also help to remove some duplicated codes.
> 
> Fixes: 2ce13640b3f4 ("mm: __first_valid_page skip over offline pages")

While that is true, we shouldn't ever trigger, neither via cma, 
virtio-mem nor memory offlining, because essentially all operate on 
MAX_ORDER -1 -aligned ranges without memory holes.

> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> ---
>   mm/page_isolation.c | 9 +--------
>   1 file changed, 1 insertion(+), 8 deletions(-)
> 
> diff --git a/mm/page_isolation.c b/mm/page_isolation.c
> index 471e3a13b541..9bb562d5d194 100644
> --- a/mm/page_isolation.c
> +++ b/mm/page_isolation.c
> @@ -202,14 +202,7 @@ int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
>   	}
>   	return 0;
>   undo:
> -	for (pfn = start_pfn;
> -	     pfn < undo_pfn;
> -	     pfn += pageblock_nr_pages) {
> -		struct page *page = pfn_to_online_page(pfn);
> -		if (!page)
> -			continue;
> -		unset_migratetype_isolate(page, migratetype);
> -	}
> +	undo_isolate_page_range(start_pfn, undo_pfn, migratetype);
>   

It'd be even cleaner to drop the label completely and call it from the 
single callsite. We can even avoid undo_pfn ...

if (page && set_migratetype_isolate(page, migratetype, flags)) {
	undo_isolate_page_range(start_pfn, pfn, migratetype);
	return -EBUSY;
}

If pfn == start_pfn, undo_isolate_page_range() will simply do nothing.

>   	return -EBUSY;
>   }
> 

-- 
Thanks,

David / dhildenb

Re: [PATCH] mm/page_isolation: fix potential missing call to unset_migratetype_isolate()

[Miaohe Lin]

On 2021/9/6 17:27, David Hildenbrand wrote:
> On 04.09.21 11:20, Miaohe Lin wrote:
>> In start_isolate_page_range() undo path, pfn_to_online_page() just checks
>> the first pfn in a pageblock while __first_valid_page() will traverse the
>> pageblock until the first online pfn is found. So we may miss the call to
>> unset_migratetype_isolate() in undo path and pages will remain isolated
>> unexpectedly. Fix this by calling undo_isolate_page_range() and this will
>> also help to remove some duplicated codes.
>>
>> Fixes: 2ce13640b3f4 ("mm: __first_valid_page skip over offline pages")
> 
> While that is true, we shouldn't ever trigger, neither via cma, virtio-mem nor memory offlining, because essentially all operate on MAX_ORDER -1 -aligned ranges without memory holes.

I think this should never trigger too. It's a theoretical issue. So is the Fixes tag necessary ?

> 
>> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
>> ---
>>   mm/page_isolation.c | 9 +--------
>>   1 file changed, 1 insertion(+), 8 deletions(-)
>>
>> diff --git a/mm/page_isolation.c b/mm/page_isolation.c
>> index 471e3a13b541..9bb562d5d194 100644
>> --- a/mm/page_isolation.c
>> +++ b/mm/page_isolation.c
>> @@ -202,14 +202,7 @@ int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
>>       }
>>       return 0;
>>   undo:
>> -    for (pfn = start_pfn;
>> -         pfn < undo_pfn;
>> -         pfn += pageblock_nr_pages) {
>> -        struct page *page = pfn_to_online_page(pfn);
>> -        if (!page)
>> -            continue;
>> -        unset_migratetype_isolate(page, migratetype);
>> -    }
>> +    undo_isolate_page_range(start_pfn, undo_pfn, migratetype);
>>   
> 
> It'd be even cleaner to drop the label completely and call it from the single callsite. We can even avoid undo_pfn ...
> 
> if (page && set_migratetype_isolate(page, migratetype, flags)) {
>     undo_isolate_page_range(start_pfn, pfn, migratetype);
>     return -EBUSY;
> }
> 

Looks much better. Will do it later. Many thanks. :)

> If pfn == start_pfn, undo_isolate_page_range() will simply do nothing.
> 
>>       return -EBUSY;
>>   }
>>
> 

Re: [PATCH] mm/page_isolation: fix potential missing call to unset_migratetype_isolate()

[David Hildenbrand]

On 06.09.21 11:38, Miaohe Lin wrote:
> On 2021/9/6 17:27, David Hildenbrand wrote:
>> On 04.09.21 11:20, Miaohe Lin wrote:
>>> In start_isolate_page_range() undo path, pfn_to_online_page() just checks
>>> the first pfn in a pageblock while __first_valid_page() will traverse the
>>> pageblock until the first online pfn is found. So we may miss the call to
>>> unset_migratetype_isolate() in undo path and pages will remain isolated
>>> unexpectedly. Fix this by calling undo_isolate_page_range() and this will
>>> also help to remove some duplicated codes.
>>>
>>> Fixes: 2ce13640b3f4 ("mm: __first_valid_page skip over offline pages")
>>
>> While that is true, we shouldn't ever trigger, neither via cma, virtio-mem nor memory offlining, because essentially all operate on MAX_ORDER -1 -aligned ranges without memory holes.
> 
> I think this should never trigger too. It's a theoretical issue. So is the Fixes tag necessary ?
> 

I think it's one of these "let's add Fixes: but no need for Cc: stable".

BUT in older kernels we could have triggered this via memory offlining 
eventually ... before c5e79ef561b0 ("mm/memory_hotplug.c: don't allow to 
online/offline memory blocks with holes") ... so maybe even a Cc: stable?

-- 
Thanks,

David / dhildenb

Re: [PATCH] mm/page_isolation: fix potential missing call to unset_migratetype_isolate()

[Miaohe Lin]

On 2021/9/6 17:46, David Hildenbrand wrote:
> On 06.09.21 11:38, Miaohe Lin wrote:
>> On 2021/9/6 17:27, David Hildenbrand wrote:
>>> On 04.09.21 11:20, Miaohe Lin wrote:
>>>> In start_isolate_page_range() undo path, pfn_to_online_page() just checks
>>>> the first pfn in a pageblock while __first_valid_page() will traverse the
>>>> pageblock until the first online pfn is found. So we may miss the call to
>>>> unset_migratetype_isolate() in undo path and pages will remain isolated
>>>> unexpectedly. Fix this by calling undo_isolate_page_range() and this will
>>>> also help to remove some duplicated codes.
>>>>
>>>> Fixes: 2ce13640b3f4 ("mm: __first_valid_page skip over offline pages")
>>>
>>> While that is true, we shouldn't ever trigger, neither via cma, virtio-mem nor memory offlining, because essentially all operate on MAX_ORDER -1 -aligned ranges without memory holes.
>>
>> I think this should never trigger too. It's a theoretical issue. So is the Fixes tag necessary ?
>>
> 
> I think it's one of these "let's add Fixes: but no need for Cc: stable".
> 
> BUT in older kernels we could have triggered this via memory offlining eventually ... before c5e79ef561b0 ("mm/memory_hotplug.c: don't allow to online/offline memory blocks with holes") ... so maybe even a Cc: stable?

Looks like it could happen in older kernels. Maybe a Cc: stable is necessary.
Many thanks.

>