Received-SPF: pass (google.com: domain of linux-kernel+bounces-179367-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Message-ID: <150ccdc2-48c7-40ac-b027-e8f92e2a0500@linux.dev>
Date: Wed, 15 May 2024 11:10:32 +0800
Precedence: bulk
MIME-Version: 1.0
Subject: Re: [PATCH] mm/ksm: fix possible UAF of stable_node
To: David Hildenbrand <david@redhat.com>,
 Andrew Morton <akpm@linux-foundation.org>, Hugh Dickins <hughd@google.com>,
 Andrea Arcangeli <aarcange@redhat.com>, Stefan Roesch <shr@devkernel.io>
Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org,
 zhouchengming@bytedance.com
References: <20240513-b4-ksm-stable-node-uaf-v1-1-f687de76f452@linux.dev>
 <74a72eeb-122f-453e-baa7-63504e7c4bd8@redhat.com>
Content-Language: en-US
From: Chengming Zhou <chengming.zhou@linux.dev>
In-Reply-To: <74a72eeb-122f-453e-baa7-63504e7c4bd8@redhat.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

On 2024/5/15 04:46, David Hildenbrand wrote:
> On 13.05.24 05:07, Chengming Zhou wrote:
>> The commit 2c653d0ee2ae ("ksm: introduce ksm_max_page_sharing per page
>> deduplication limit") introduced a possible failure case in the
>> stable_tree_insert(), where we may free the new allocated stable_node_dup
>> if we fail to prepare the missing chain node.
>>
>> Then that kfolio return and unlock with a freed stable_node set... And
>> any MM activities can come in to access kfolio->mapping, so UAF.
>>
>> Fix it by moving folio_set_stable_node() to the end after stable_node
>> is inserted successfully.
>>
>> Fixes: 2c653d0ee2ae ("ksm: introduce ksm_max_page_sharing per page deduplication limit")
>> Signed-off-by: Chengming Zhou <chengming.zhou@linux.dev>
>> ---
>>   mm/ksm.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/mm/ksm.c b/mm/ksm.c
>> index e1034bf1c937..a8b76af5cf64 100644
>> --- a/mm/ksm.c
>> +++ b/mm/ksm.c
>> @@ -2153,7 +2153,6 @@ static struct ksm_stable_node *stable_tree_insert(struct folio *kfolio)
>>         INIT_HLIST_HEAD(&stable_node_dup->hlist);
>>       stable_node_dup->kpfn = kpfn;
>> -    folio_set_stable_node(kfolio, stable_node_dup);
>>       stable_node_dup->rmap_hlist_len = 0;
>>       DO_NUMA(stable_node_dup->nid = nid);
>>       if (!need_chain) {
>> @@ -2172,6 +2171,8 @@ static struct ksm_stable_node *stable_tree_insert(struct folio *kfolio)
>>           stable_node_chain_add_dup(stable_node_dup, stable_node);
>>       }
>>   +    folio_set_stable_node(kfolio, stable_node_dup);
>> +
>>       return stable_node_dup;
> 
> Looks correct to me.
> 
> We might now link the node before the folio->mapping is set up. Do we care? Don't think so.

Yeah, it shouldn't be a problem, although it doesn't look very nice.

Another way to fix maybe "folio_set_stable_node(folio, NULL)" in the failure case,
which is safe since we have held the folio lock.

> 
> Acked-by: David Hildenbrand <david@redhat.com>
> 

Thanks.