Re: [PATCH] netfilter: nf_conntrack_sip: fix the ct_sip_parse_numerical_param() return value.

[Gavrilov Ilia]

On 5/2/23 17:05, Simon Horman wrote:
> On Tue, May 02, 2023 at 11:43:19AM +0000, Gavrilov Ilia wrote:
>> On 4/28/23 22:24, Simon Horman wrote:
>>> On Wed, Apr 26, 2023 at 03:04:31PM +0000, Gavrilov Ilia wrote:
>>>> ct_sip_parse_numerical_param() returns only 0 or 1 now.
>>>> But process_register_request() and process_register_response() imply
>>>> checking for a negative value if parsing of a numerical header parameter
>>>> failed. Let's fix it.
>>>>
>>>> Found by InfoTeCS on behalf of Linux Verification Center
>>>> (linuxtesting.org) with SVACE.
>>>>
>>>> Fixes: 0f32a40fc91a ("[NETFILTER]: nf_conntrack_sip: create signalling expectations")
>>>> Signed-off-by: Ilia.Gavrilov <Ilia.Gavrilov@xxxxxxxxxxx>
>>>
>>> Hi Gavrilov,
>>>
>>
>> Hi Simon, thank you for your answer.
>>
>>> although it is a slightly unusual convention for kernel code,
>>> I believe the intention is that this function returns 0 when
>>> it fails (to parse) and 1 on success. So I think that part is fine.
>>>
>>> What seems a bit broken is the way that callers use the return value.
>>>
>>> 1. The call in process_register_response() looks like this:
>>>
>>> 	ret = ct_sip_parse_numerical_param(...)
>>> 	if (ret < 0) {
>>> 		nf_ct_helper_log(skb, ct, "cannot parse expires");
>>> 		return NF_DROP;
>>> 	}
>>>
>>>       But ret can only be 0 or 1, so the error handling is never inoked,
>>>       and a failure to parse is ignored. I guess failure doesn't occur in
>>>       practice.
>>>
>>>       I suspect this should be:
>>>
>>> 	ret = ct_sip_parse_numerical_param(...)
>>> 	if (!ret) {
>>> 		nf_ct_helper_log(skb, ct, "cannot parse expires");
>>> 		return NF_DROP;
>>> 	}
>>>
>>
>> ct_sip_parse_numerical_param() returns 0 in to cases 1) when the
>> parameter 'expires=' isn't found in the header or 2) it's incorrectly set.
>> In the first case, the return value should be ignored, since this is a
>> normal situation
>> In the second case, it's better to write to the log and return NF_DROP,
>> or ignore it too, then checking the return value can be removed as
>> unnecessary.
> 
> Sorry, I think I misunderstood the intention of your patch earlier.
> 
> Do I (now) understand correctly that you are proposing a tristate?
> 
> a) return 1 if value is found; *val is set
> b) return 0 if value is not found; *val is unchanged
> c) return -1 on error; *val is undefined

Yes, it seems to me that this was originally intended.