lkml.org 
[lkml]   [2022]   [Mar]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] mm/mempolicy: fix potential mpol_new leak in shared_policy_replace
From
Date
On 2022/3/15 23:27, Michal Hocko wrote:
> On Tue 15-03-22 21:42:29, Miaohe Lin wrote:
>> On 2022/3/15 0:44, Michal Hocko wrote:
>>> On Fri 11-03-22 17:36:24, Miaohe Lin wrote:
>>>> If mpol_new is allocated but not used in restart loop, mpol_new will be
>>>> freed via mpol_put before returning to the caller. But refcnt is not
>>>> initialized yet, so mpol_put could not do the right things and might
>>>> leak the unused mpol_new.
>>>
>>> The code is really hideous but is there really any bug there? AFAICS the
>>> new policy is only allocated in if (n->end > end) branch and that one
>>> will set the reference count on the retry. Or am I missing something?
>>>
>>
>> Many thanks for your comment.
>> IIUC, new policy is allocated via the below code:
>>
>> shared_policy_replace:
>> alloc_new:
>> write_unlock(&sp->lock);
>> ret = -ENOMEM;
>> n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
>> if (!n_new)
>> goto err_out;
>> mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
>> if (!mpol_new)
>> goto err_out;
>> goto restart;
>>
>> And mpol_new' reference count will be set before used in n->end > end case. But
>> if that is "not" the case, i.e. mpol_new is not inserted into the rb_tree, mpol_new
>> will be freed via mpol_put before return:
>
> One thing I have missed previously is that the lock is dropped during
> the allocation so I guess the memory policy could have been changed
> during that time. Is this possible? Have you explored this possibility?
> Is this a theoretical problem or it can be triggered intentionally.
>

This is found via code investigation. I think this could be triggered if there
are many concurrent mpol_set_shared_policy in place. But the user-visible effect
might be obscure as only sizeof(struct mempolicy) bytes leaks possiblely every time.

> These details would be really interesting for the changelog so that we
> can judge how important this would be.

This might not be that important as this issue should have been well-concealed for
almost ten years (since commit 42288fe366c4 ("mm: mempolicy: Convert shared_policy mutex to spinlock")).

>

Thanks.

\
 
 \ /
  Last update: 2022-09-17 16:17    [W:0.095 / U:0.024 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site