Re: [PATCH net-next 1/3] bpf: introduce current->pid, tgid, uid, gid, comm accessors

From: Andy Lutomirski
Date: Fri Jun 12 2015 - 20:25:38 EST


On Fri, Jun 12, 2015 at 5:15 PM, Alexei Starovoitov <ast@xxxxxxxxxxxx> wrote:
> On 6/12/15 5:03 PM, Andy Lutomirski wrote:
>>
>> On Fri, Jun 12, 2015 at 4:55 PM, Alexei Starovoitov <ast@xxxxxxxxxxxx>
>> wrote:
>>>
>>> On 6/12/15 4:47 PM, Andy Lutomirski wrote:
>>>>
>>>>
>>>> On Fri, Jun 12, 2015 at 4:38 PM, Alexei Starovoitov <ast@xxxxxxxxxxxx>
>>>> wrote:
>>>>>
>>>>>
>>>>> On 6/12/15 4:25 PM, Andy Lutomirski wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>> It's a dangerous tool. Also, shouldn't the returned uid match the
>>>>>> namespace of the task that installed the probe, not the task that's
>>>>>> being probed?
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> so leaking info to unprivileged apps is the concern?
>>>>> The whole thing is for root only as you know.
>>>>> The non-root is still far away. Today root needs to see the whole
>>>>> kernel. That was the goal from the beginning.
>>>>>
>>>>
>>>> This is more of a correctness issue than a security issue. ISTM using
>>>> current_user_ns() in a kprobe is asking for trouble. It certainly
>>>> allows any unprivilege user to show any uid it wants to the probe,
>>>> which is probably not what the installer of the probe expects.
>>>
>>>
>>>
>>> probe doesn't expect anything. it doesn't make any decisions.
>>> bpf is read only. it's _visibility_ into the kernel.
>>> It's not used for security.
>>> When we start connecting eBPF to seccomp I would agree that uid
>>> handling needs to be done carefully, but we're not there yet.
>>> I don't want to kill _visibility_ because in some distant future
>>> bpf becomes a decision making tool in security area and
>>> get_current_uid() will return numbers that shouldn't be blindly
>>> used to reject/accept a user requesting something. That's far away.
>>>
>>
>> All that is true, but the code that *installed* the bpf probe might
>> get might confused when it logs that uid 0 did such-and-such when
>> really some unprivileged userns root did it.
>
>
> so what specifically you proposing?
> Use from_kuid(&init_user_ns,...) instead?

That seems reasonable to me. After all, you can't install one of
these probes from a non-init userns.

--Andy
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/