Re: [PATCH] ocfs2: use get_task_comm

[Kees Cook]

On Tue, Dec 5, 2017 at 12:27 PM, Arnd Bergmann <arnd@xxxxxxxx> wrote:
> On Tue, Dec 5, 2017 at 8:19 PM, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
>> On Tue, Dec 5, 2017 at 7:20 AM, Arnd Bergmann <arnd@xxxxxxxx> wrote:
>>> While reviewing all callers of get_task_comm(), I stumbled
>>> over this one that claimed it was not exported, when in fact
>>> it is. Accessing task->comm directly is not safe, so better
>>> convert this one to using get_task_comm as well.
>>
>> Using get_task_comm() in cases like this is actually overkill (i.e.
>> using up stack space), since there's (currently) no benefit. Nothing
>> protects getting a "correct" view of task->comm (i.e. it could get
>> updated in the middle of a copy), but it _is_ always NULL terminated,
>> so it's safe to use with %s like this. While it does make me slightly
>> uncomfortable to _depend_ on this NULL termination, but there are lots
>> of open-coded %s users of task->comm. When we're trying to save a
>> _copy_ of task->comm, then we want get_task_comm(), just to make sure
>> we're doing it right.
>>
>> So, while I don't oppose this patch, it might be seen as a wasteful
>> use of stack space.
>
> It's only a few bytes of stack space in a leaf function, I'd not be
> worried about that.
>
> More generally speaking though, how exactly do we guarantee that
> there is NUL-termination on tsk->comm during a concurrent update?
> Could we ever get into a situation where overwrite the NUL byte
> while setting tsk->comm to a longer string, and read the new start
> of the string together with an unterminated end, or do we strictly
> guarantee that the last byte is still NUL? I assume the latter is
> true, just haven't found exactly where that guarantee is made.

strncpy will zero pad with the trailing NULL, so it's supposed to
always be safe... still gives me the creeps, though.

-Kees

-- 
Kees Cook
Pixel Security