Re: [PATCH] f2fs: compress: fix UAF of f2fs_inode_info in f2fs_free_dic

[Chao Yu <chao@kernel.org>]

On 6/4/25 18:49, Zhiguo Niu wrote:
> Chao Yu <chao@kernel.org> 于2025年6月4日周三 17:48写道：
>>
>> On 6/4/25 13:54, Zhiguo Niu wrote:
>>> The decompress_io_ctx may be released asynchronously after
>>> I/O completion. If this file is deleted immediately after read,
>>> and the kworker of processing post_read_wq has not been executed yet
>>> due to high workloads, It is possible that the inode(f2fs_inode_info)
>>> is evicted and freed before it is used f2fs_free_dic.
>>>
>>>     The UAF case as below:
>>>     Thread A                                      Thread B
>>>     - f2fs_decompress_end_io
>>>      - f2fs_put_dic
>>>       - queue_work
>>>         add free_dic work to post_read_wq
>>>                                                    - do_unlink
>>>                                                     - iput
>>>                                                      - evict
>>>                                                       - call_rcu
>>>     This file is deleted after read.
>>>
>>>     Thread C                                 kworker to process post_read_wq
>>>     - rcu_do_batch
>>>      - f2fs_free_inode
>>>       - kmem_cache_free
>>>      inode is freed by rcu
>>>                                              - process_scheduled_works
>>>                                               - f2fs_late_free_dic
>>>                                                - f2fs_free_dic
>>>                                                 - f2fs_release_decomp_mem
>>>                                       read (dic->inode)->i_compress_algorithm
>>>
>>> This patch increase inode->i_count before f2fs_free_dic and decrease it
>>> after free the dic.
>>>
>>> Cc: Daeho Jeong <daehojeong@google.com>
>>> Fixes: bff139b49d9f ("f2fs: handle decompress only post processing in softirq")
>>> Signed-off-by: Zhiguo Niu <zhiguo.niu@unisoc.com>
>>> Signed-off-by: Baocong Liu <baocong.liu@unisoc.com>
>>> ---
>>>  fs/f2fs/compress.c | 19 ++++++++++++++-----
>>>  1 file changed, 14 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
>>> index b3c1df9..6b3b3a7 100644
>>> --- a/fs/f2fs/compress.c
>>> +++ b/fs/f2fs/compress.c
>>> @@ -1687,7 +1687,7 @@ static void f2fs_release_decomp_mem(struct decompress_io_ctx *dic,
>>>  }
>>>
>>>  static void f2fs_free_dic(struct decompress_io_ctx *dic,
>>> -             bool bypass_destroy_callback);
>>> +             bool bypass_destroy_callback, bool late_free);
>>>
>>>  struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
>>>  {
>>> @@ -1743,12 +1743,12 @@ struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
>>>       return dic;
>>>
>>>  out_free:
>>> -     f2fs_free_dic(dic, true);
>>> +     f2fs_free_dic(dic, true, false);
>>>       return ERR_PTR(ret);
>>>  }
>>>
>>>  static void f2fs_free_dic(struct decompress_io_ctx *dic,
>>> -             bool bypass_destroy_callback)
>>> +             bool bypass_destroy_callback, bool late_free)
>>>  {
>>>       int i;
>>>
>>> @@ -1775,6 +1775,11 @@ static void f2fs_free_dic(struct decompress_io_ctx *dic,
>>>       }
>>>
>>>       page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
>>> +     if (late_free) {
>>> +             spin_lock(&dic->inode->i_lock);
>>> +             atomic_dec(&dic->inode->i_count);
>>> +             spin_unlock(&dic->inode->i_lock);
>>
>> If it is the last one release i_count, it needs to call iput_final to evict inode
>> like what iput did, so we'd better to call iput() here?
> Hi Chao,
> Yes, we have also tested this method(iput/__iget), and it worked.
> Just think It is simpler and easier to read to directly operate
> i_count, and then free it
> by relying on the memory module when i_count=0.
> But It seems iput/__iget is better.
>>
>>> +     }
>>>       kmem_cache_free(dic_entry_slab, dic);
>>>  }
>>>
>>> @@ -1783,16 +1788,20 @@ static void f2fs_late_free_dic(struct work_struct *work)
>>>       struct decompress_io_ctx *dic =
>>>               container_of(work, struct decompress_io_ctx, free_work);
>>>
>>> -     f2fs_free_dic(dic, false);
>>> +     f2fs_free_dic(dic, false, true);
>>>  }
>>>
>>>  static void f2fs_put_dic(struct decompress_io_ctx *dic, bool in_task)
>>>  {
>>>       if (refcount_dec_and_test(&dic->refcnt)) {
>>>               if (in_task) {
>>> -                     f2fs_free_dic(dic, false);
>>> +                     f2fs_free_dic(dic, false, false);
>>>               } else {
>>>                       INIT_WORK(&dic->free_work, f2fs_late_free_dic);
>>> +                     /* to avoid inode is evicted simultaneously */
>>> +                     spin_lock(&dic->inode->i_lock);
>>> +                     atomic_inc(&dic->inode->i_count);
>>> +                     spin_unlock(&dic->inode->i_lock);
>>
>> iget()?
>>
>> BTW, can we store i_compress_algorithm in dic to avoid inode access?
> 
> Also thought of this method, but it would require more changes.
> dic->inode used in f2fs_free_dic are all needed to modify except
> i_compress_algorithm.
> such as  page_array_free(dic->inode),

Zhiguo,

page_array_free() parses dic->inode to get sbi only, so we can pass sbi to
page_array_free() directly to avoid using dic->inode.

> allow_memalloc_for_decomp(F2FS_I_SB(dic->inode)).
> 
> Do you have any other suggestions?

Using iget/iput looks fine to me, please go ahead.

Thanks,

> thanks！
> 
>>
>> Thanks,
>>
>>>                       queue_work(F2FS_I_SB(dic->inode)->post_read_wq,
>>>                                       &dic->free_work);
>>>               }
>>