Re: [PATCH v4] ksmbd: fix possible memory leak in smb2_lock()

[Hangyu Hua <hbh25y@gmail.com>]

On 15/2/2023 09:04, Namjae Jeon wrote:
> 2023-02-14 15:36 GMT+09:00, Hangyu Hua <hbh25y@gmail.com>:
>> argv needs to be free when setup_async_work fails or when the current
>> process is woken up.
>>
>> Fixes: e2f34481b24d ("cifsd: add server-side procedures for SMB3")
>> Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
>> ---
>>
>> v2: avoid NULL pointer dereference in set_close_state_blocked_works()
>> v3: avoid race condition between smb2_lock() and smb2_cancel()
>> v4: use another way to avoid race condition
>>
>>   fs/ksmbd/smb2pdu.c   | 20 +++++++++++++-------
>>   fs/ksmbd/vfs_cache.c |  2 ++
>>   2 files changed, 15 insertions(+), 7 deletions(-)
>>
>> diff --git a/fs/ksmbd/smb2pdu.c b/fs/ksmbd/smb2pdu.c
>> index d681f91947d9..1b517d3ca2ef 100644
>> --- a/fs/ksmbd/smb2pdu.c
>> +++ b/fs/ksmbd/smb2pdu.c
>> @@ -6666,7 +6666,8 @@ int smb2_cancel(struct ksmbd_work *work)
>>   				    "smb2 with AsyncId %llu cancelled command = 0x%x\n",
>>   				    le64_to_cpu(hdr->Id.AsyncId),
>>   				    le16_to_cpu(chdr->Command));
>> -			cancel_work = iter;
>> +			if (iter->cancel_fn)
>> +				cancel_work = iter;
> You have freed ->cancel_argv in smb2_lock().  Wouldn't this cause UAF
> issue the below?
> 
>          if (cancel_work) {
>                  cancel_work->state = KSMBD_WORK_CANCELLED;
>                  if (cancel_work->cancel_fn)
>                          cancel_work->cancel_fn(cancel_work->cancel_argv);
>          }
I think this can't cause UAF beacause cancel_fn will be set to NULL 
before releasing argv in smb2_lock(). There is no way to use a argv.

>>   			break;
>>   		}
>>   		spin_unlock(&conn->request_lock);
>> @@ -7050,6 +7051,7 @@ int smb2_lock(struct ksmbd_work *work)
>>   						      smb2_remove_blocked_lock,
>>   						      argv);
>>   				if (rc) {
>> +					kfree(argv);
>>   					err = -ENOMEM;
>>   					goto out;
>>   				}
>> @@ -7061,6 +7063,16 @@ int smb2_lock(struct ksmbd_work *work)
>>
>>   				ksmbd_vfs_posix_lock_wait(flock);
>>
>> +				spin_lock(&work->conn->request_lock);
>> +				spin_lock(&fp->f_lock);
>> +				if (work->state != KSMBD_WORK_CLOSED) {
> I think that this check is not needed if we don't delete entry in
> set_close_state_blocked_works().
>> +					list_del(&work->fp_entry);
>> +					work->cancel_fn = NULL;
>> +					kfree(argv);
>> +				}
>> +				spin_unlock(&fp->f_lock);
>> +				spin_unlock(&work->conn->request_lock);
>> +
>>   				if (work->state != KSMBD_WORK_ACTIVE) {
>>   					list_del(&smb_lock->llist);
>>   					spin_lock(&work->conn->llist_lock);
>> @@ -7069,9 +7081,6 @@ int smb2_lock(struct ksmbd_work *work)
>>   					locks_free_lock(flock);
>>
>>   					if (work->state == KSMBD_WORK_CANCELLED) {
>> -						spin_lock(&fp->f_lock);
>> -						list_del(&work->fp_entry);
>> -						spin_unlock(&fp->f_lock);
>>   						rsp->hdr.Status =
>>   							STATUS_CANCELLED;
>>   						kfree(smb_lock);
>> @@ -7093,9 +7102,6 @@ int smb2_lock(struct ksmbd_work *work)
>>   				list_del(&smb_lock->clist);
>>   				spin_unlock(&work->conn->llist_lock);
>>
>> -				spin_lock(&fp->f_lock);
>> -				list_del(&work->fp_entry);
>> -				spin_unlock(&fp->f_lock);
>>   				goto retry;
>>   			} else if (!rc) {
>>   				spin_lock(&work->conn->llist_lock);
>> diff --git a/fs/ksmbd/vfs_cache.c b/fs/ksmbd/vfs_cache.c
>> index da9163b00350..761a8aa540ce 100644
>> --- a/fs/ksmbd/vfs_cache.c
>> +++ b/fs/ksmbd/vfs_cache.c
>> @@ -372,6 +372,8 @@ static void set_close_state_blocked_works(struct
>> ksmbd_file *fp)
>>   		list_del(&cancel_work->fp_entry);
>>   		cancel_work->state = KSMBD_WORK_CLOSED;
>>   		cancel_work->cancel_fn(cancel_work->cancel_argv);
>> +		cancel_work->cancel_fn = NULL;
>> +		kfree(cancel_work->cancel_argv);
> If we remove list_del, we don't need to do them here ? and we can
> change _safe version to list_for_each().

Agreed. I will move this list_del() to smb2_lock() in v5.

Thanks,
Hangyu

>>   	}
>>   	spin_unlock(&fp->f_lock);
>>   }
>> --
>> 2.34.1
>>
>>