Re: mm, something wring in page_lock_anon_vma_read()?

From: Xishi Qiu
Date: Fri May 19 2017 - 22:22:35 EST


On 2017/5/20 10:02, Hugh Dickins wrote:

> On Sat, 20 May 2017, Xishi Qiu wrote:
>> On 2017/5/20 6:00, Hugh Dickins wrote:
>>>
>>> You're ignoring the rcu_read_lock() on entry to page_lock_anon_vma_read(),
>>> and the SLAB_DESTROY_BY_RCU (recently renamed SLAB_TYPESAFE_BY_RCU) nature
>>> of the anon_vma_cachep kmem cache. It is not safe to muck with anon_vma->
>>> root in anon_vma_free(), others could still be looking at it.
>>>
>>> Hugh
>>>
>>
>> Hi Hugh,
>>
>> Thanks for your reply.
>>
>> SLAB_DESTROY_BY_RCU will let it call call_rcu() in free_slab(), but if the
>> anon_vma *reuse* by someone again, access root_anon_vma is not safe, right?
>
> That is safe, on reuse it is still a struct anon_vma; then the test for
> !page_mapped(page) will show that it's no longer a reliable anon_vma for
> this page, so page_lock_anon_vma_read() returns NULL.
>
> But of course, if page->_mapcount has been corrupted or misaccounted,
> it may think page_mapped(page) when actually page is not mapped,
> and the anon_vma is not good for it.
>

Hi Hugh,

Here is a bug report form redhat: https://bugzilla.redhat.com/show_bug.cgi?id=1305620
And I meet the bug too. However it is hard to reproduce, and
624483f3ea82598("mm: rmap: fix use-after-free in __put_anon_vma") is not help.

>From the vmcore, it seems that the page is still mapped(_mapcount=0 and _count=2),
and the value of mapping is a valid address(mapping = 0xffff8801b3e2a101),
but anon_vma has been corrupted.

Any ideas?

Thanks,
Xishi Qiu

>>
>> e.g. if I clean the root pointer before free it, then access root_anon_vma
>> in page_lock_anon_vma_read() is NULL pointer access, right?
>
> Yes, cleaning root pointer before free may result in NULL pointer access.
>
> Hugh
>
>>
>> anon_vma_free()
>> ...
>> anon_vma->root = NULL;
>> kmem_cache_free(anon_vma_cachep, anon_vma);
>> ...
>>
>> Thanks,
>> Xishi Qiu
>
> .
>