Re: [PATCH] seccomp: Release fp pointer when leaving from seccomp_attach_filter().

[Masami Ichikawa]

On Tue, Apr 15, 2014 at 2:08 AM, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> On Mon, Apr 14, 2014 at 9:18 AM, Alexei Starovoitov <ast@xxxxxxxxxxxx> wrote:
>> On Mon, Apr 14, 2014 at 9:02 AM, Masami Ichikawa <masami256@xxxxxxxxx> wrote:
>>> kmemleak reported some memory leak as below.
>>
>> grrr. yes. sorry.
>>
>>> unreferenced object 0xffff8800d6ea4000 (size 512):
>>>   comm "sshd", pid 278, jiffies 4294898315 (age 46.653s)
>>>   hex dump (first 32 bytes):
>>>     21 00 00 00 04 00 00 00 15 00 01 00 3e 00 00 c0  !...........>...
>>>     06 00 00 00 00 00 00 00 21 00 00 00 00 00 00 00  ........!.......
>>>   backtrace:
>>>     [<ffffffff8151414e>] kmemleak_alloc+0x4e/0xb0
>>>     [<ffffffff811a3a40>] __kmalloc+0x280/0x320
>>>     [<ffffffff8110842e>] prctl_set_seccomp+0x11e/0x3b0
>>>     [<ffffffff8107bb6b>] SyS_prctl+0x3bb/0x4a0
>>>     [<ffffffff8152ef2d>] system_call_fastpath+0x1a/0x1f
>>>     [<ffffffffffffffff>] 0xffffffffffffffff
>>>
>>> This memory leak happend in seccomp_attach_filter().
>>> The fp pointer was allocated via kzalloc so that it needs to realase memory
>>> when leaving from function.
>
> Thanks for the catch!
>
>>> This patch changed two things.
>>> One is set -ENOMEM to ret, if fp is unable to get memory.
>>> The other is removes "return 0" statement, and frees fp pointer before
>>> leaving.
>>>
>>> Signed-off-by: Masami Ichikawa <masami256@xxxxxxxxx>
>>> ---
>>>  kernel/seccomp.c | 8 +++++---
>>>  1 file changed, 5 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/kernel/seccomp.c b/kernel/seccomp.c
>>> index d8d046c..a9ce7a9 100644
>>> --- a/kernel/seccomp.c
>>> +++ b/kernel/seccomp.c
>>> @@ -259,8 +259,10 @@ static long seccomp_attach_filter(struct sock_fprog *fprog)
>>>         filter = kzalloc(sizeof(struct seccomp_filter) +
>>>                          sizeof(struct sock_filter_int) * new_len,
>>>                          GFP_KERNEL|__GFP_NOWARN);
>>> -       if (!filter)
>>> +       if (!filter) {
>>> +               ret = -ENOMEM;
>>>                 goto free_prog;
>>> +       }
>>
>> agree. that's a good addition.
>>
>>>         ret = sk_convert_filter(fp, fprog->len, filter->insnsi, &new_len);
>>>         if (ret)
>>> @@ -275,10 +277,10 @@ static long seccomp_attach_filter(struct sock_fprog *fprog)
>>>          */
>>>         filter->prev = current->seccomp.filter;
>>>         current->seccomp.filter = filter;
>>> -       return 0;
>>
>> I think mixing error and ok return paths is ugly.
I troubled that have two kfree(fp) path or merge error and success path.
But separate paths is easy to read.

>> Can you add kfree(fp) here instead of removing return 0?
>>
>> Thanks!
>>
>>>  free_filter:
>>> -       kfree(filter);
>>> +       if (ret)
>>> +               kfree(filter);
>>>  free_prog:
>>>         kfree(fp);
>>>         return ret;
>>> --
>>> 1.9.1
>>>
>
> Yeah, I'd prefer a different approach that follows the existing
> conventions in the code. I'll send a separate patch.
I see.

Thank you for considering it.

> -Kees
>
> --
> Kees Cook
> Chrome OS Security


Cheeers,

-- 
Masami Ichikawa
--
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/