kernel BUG at kernel/cred.c:434!

[Yang Yingliang]

On 2019/4/20 0:13, Paul Moore wrote:
> On Fri, Apr 19, 2019 at 10:34 AM Yang Yingliang
> <yangyingliang at huawei.com> wrote:
>> On 2019/4/19 21:24, Paul Moore wrote:
>>> On Thu, Apr 18, 2019 at 10:42 PM Yang Yingliang
>>> <yangyingliang at huawei.com> wrote:
>>>> On 2019/4/19 10:04, Paul Moore wrote:
>>>>> On Wed, Apr 17, 2019 at 10:50 PM Yang Yingliang
>>>>> <yangyingliang at huawei.com> wrote:
>>>>>> On 2019/4/18 8:24, Casey Schaufler wrote:
>>>>>>> On 4/17/2019 4:39 PM, Paul Moore wrote:
>>>>>>>> Since it looks like all three LSMs which implement the setprocattr
>>>>>>>> hook are vulnerable I'm open to the idea that proc_pid_attr_write() is
>>>>>>>> a better choice for the cred != read_cred check, but I would want to
>>>>>>>> make sure John and Casey are okay with that.
>>>>>>>>
>>>>>>>> John?
>>>>>>>>
>>>>>>>> Casey?
>>>>>>> I'm fine with the change going into proc_pid_attr_write().
>>>>>> The cred != real_cred checking is not enough.
>>>>>>
>>>>>> Consider this situation, when doing override, cred, real_cred and
>>>>>> new_cred are all same:
>>>>>>
>>>>>> after override_creds()    cred == real_cred == new1_cred
>>>>> I'm sorry, you've lost me.  After override_creds() returns
>>>>> current->cred and current->real_cred are not going to be the same,
>>>>> yes?
>>>> It's possible the new  cred is equal to current->real_cred and
>>>> current->cred,
>>>> so after overrides_creds(), they have the same value.
>>> Both task_struct.cred and task_struct.real_cred are pointer values,
>>> assuming that one uses prepare_creds() to allocate/initialize a new
>>> cred struct for use with override_creds() then the newly created cred
>>> should never be equal to task_struct.real_cred.  Am I missing
>>> something, or are you thinking of something else?
>> In do_acct_process(), file->f_cred may equal to current->real_cred, I
>> confirm
>> it by adding some debug message in do_acct_process() like this:
> I would expect that; real_cred is the task's objective DAC
> credentials, so using it for f_cred makes sense.
>
> What we are now talking about is the task's subjective credentials,
> which can be overridden via override_creds(), and are what the LSMs
> change via proc_pid_attr_write().
I'm not sure you got my point.

I was saying cred != real_cred check is not quite right, because the 
cred can
be overridden by a same pointer as my print messages showing.

"cred != real_cred" means override_creds() is called, but "cred == 
real_cred"
doesn't mean override_creds() is not called.

When we use "cred != real_cred" check, we may lost the situation that cred
is overridden by a same pointer. In this case, we will do 
override_creds() =>
commit_creds() => revert_creds(), this make cred != real_cred, when a new
commit_creds() is called, it also will trigger a BUG_ON().
>
>> --- a/kernel/acct.c
>> +++ b/kernel/acct.c
>> @@ -481,6 +481,7 @@ static void do_acct_process(struct bsd_acct_struct
>> *acct)
>>           flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
>>           current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
>>           /* Perform file operations on behalf of whoever enabled
>> accounting */
>> +       pr_info("task:%px new cred:%px real cred:%px cred:%px\n",
>> current, file->f_cred, current->real_cred, current->cred);
>>           orig_cred = override_creds(file->f_cred);