After kill_procs(), tk will be freed without being removed from the to_kill
list. In the next iteration, the freed list entry in the to_kill list will
be accessed, thus leading to use-after-free issue. Fix it by reinitializing
the to_kill list after unmap_and_kill().

Fixes: c36e20249571 ("mm: introduce mf_dax_kill_procs() for fsdax case")
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
---
 mm/memory-failure.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 7023c3d81273..a2f4e8b00a26 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -1658,6 +1658,8 @@ int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
 		collect_procs_fsdax(page, mapping, index, &to_kill);
 		unmap_and_kill(&to_kill, page_to_pfn(page), mapping,
 				index, mf_flags);
+		/* Reinitialize to_kill list for later resuing. */
+		INIT_LIST_HEAD(&to_kill);
 unlock:
 		dax_unlock_mapping_entry(mapping, index, cookie);
 	}
-- 
2.23.0

On Thu, Aug 18, 2022 at 09:00:14PM +0800, Miaohe Lin wrote:
> After kill_procs(), tk will be freed without being removed from the to_kill
> list. In the next iteration, the freed list entry in the to_kill list will
> be accessed, thus leading to use-after-free issue.

kill_procs() runs over the to_kill list and frees all listed items in each
iteration.  So just after returning from unmap_and_kill(), to_kill->next and
to_kill->prev still point to the addresses of struct to_kill which was the
first or last item (already freed!).  This is bad-manered, but
collect_procs_fsdax() in the next iteration calls list_add_tail() and
overwrites the dangling pointers with newly allocated item.  So this problem
should not be so critical?  Anyway, I agree with fixing this fragile code.

> Fix it by reinitializing
> the to_kill list after unmap_and_kill().
> 
> Fixes: c36e20249571 ("mm: introduce mf_dax_kill_procs() for fsdax case")
> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>

> ---
>  mm/memory-failure.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> index 7023c3d81273..a2f4e8b00a26 100644
> --- a/mm/memory-failure.c
> +++ b/mm/memory-failure.c
> @@ -1658,6 +1658,8 @@ int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
>  		collect_procs_fsdax(page, mapping, index, &to_kill);
>  		unmap_and_kill(&to_kill, page_to_pfn(page), mapping,
>  				index, mf_flags);
> +		/* Reinitialize to_kill list for later resuing. */

s/resuing/reusing/ ?

> +		INIT_LIST_HEAD(&to_kill);

How about adding list_del() in kill_procs()?  Other callers now use
to_kill only once, but fixing generally looks tidier to me.

Thanks,
Naoya Horiguchi

On 2022/8/19 13:23, HORIGUCHI NAOYA(堀口 直也) wrote:
> On Thu, Aug 18, 2022 at 09:00:14PM +0800, Miaohe Lin wrote:
>> After kill_procs(), tk will be freed without being removed from the to_kill
>> list. In the next iteration, the freed list entry in the to_kill list will
>> be accessed, thus leading to use-after-free issue.
> 
> kill_procs() runs over the to_kill list and frees all listed items in each
> iteration.  So just after returning from unmap_and_kill(), to_kill->next and
> to_kill->prev still point to the addresses of struct to_kill which was the
> first or last item (already freed!).  This is bad-manered, but
> collect_procs_fsdax() in the next iteration calls list_add_tail() and
> overwrites the dangling pointers with newly allocated item.  So this problem

list_add_tail will do WRITE_ONCE(prev->next, new) where prev is already freed!
Or am I miss something?

> should not be so critical?  Anyway, I agree with fixing this fragile code.
> 
>> Fix it by reinitializing
>> the to_kill list after unmap_and_kill().
>>
>> Fixes: c36e20249571 ("mm: introduce mf_dax_kill_procs() for fsdax case")
>> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> 
>> ---
>>  mm/memory-failure.c | 2 ++
>>  1 file changed, 2 insertions(+)
>>
>> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
>> index 7023c3d81273..a2f4e8b00a26 100644
>> --- a/mm/memory-failure.c
>> +++ b/mm/memory-failure.c
>> @@ -1658,6 +1658,8 @@ int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
>>  		collect_procs_fsdax(page, mapping, index, &to_kill);
>>  		unmap_and_kill(&to_kill, page_to_pfn(page), mapping,
>>  				index, mf_flags);
>> +		/* Reinitialize to_kill list for later resuing. */
> 
> s/resuing/reusing/ ?

OK.

> 
>> +		INIT_LIST_HEAD(&to_kill);
> 
> How about adding list_del() in kill_procs()?  Other callers now use
> to_kill only once, but fixing generally looks tidier to me.

That's a good idea. Will do it in v2. Many thanks for your review, Naoya!

Thanks,
Miaohe Lin

> 
> Thanks,
> Naoya Horiguchi
> 

On Fri, Aug 19, 2022 at 03:32:27PM +0800, Miaohe Lin wrote:
> On 2022/8/19 13:23, HORIGUCHI NAOYA(堀口 直也) wrote:
> > On Thu, Aug 18, 2022 at 09:00:14PM +0800, Miaohe Lin wrote:
> >> After kill_procs(), tk will be freed without being removed from the to_kill
> >> list. In the next iteration, the freed list entry in the to_kill list will
> >> be accessed, thus leading to use-after-free issue.
> > 
> > kill_procs() runs over the to_kill list and frees all listed items in each
> > iteration.  So just after returning from unmap_and_kill(), to_kill->next and
> > to_kill->prev still point to the addresses of struct to_kill which was the
> > first or last item (already freed!).  This is bad-manered, but
> > collect_procs_fsdax() in the next iteration calls list_add_tail() and
> > overwrites the dangling pointers with newly allocated item.  So this problem
> 
> list_add_tail will do WRITE_ONCE(prev->next, new) where prev is already freed!
> Or am I miss something?

No, you're right. Thank you for explanation.

Thanks,
Naoya Horiguchi