Re: [PATCH] net: sched: Fix an endian bug in tcf_proto_create

[Kunwu Chan <chentao@kylinos.cn>]

Hi Simon,

Thanks for your reply.
For a lot of newcomers who aren't proficient in this part of the code, 
like me, it might be confusing what is the  correct endien and width of 
a protocol.

In response to your question, I wonder if it is necessary to implement a 
unified checking mechanism with a strict parameter validation for all 
invocation parameters?

For example, add an input parameter to the 'tcf_proto_create' to 
represent the endien and width of the protocol, and check the validity 
of the input parameter at the beginning of the function.

I don't have a good idea of how to make sure that the right type is used 
in the call path.
This is just my personal opinion, welcome to discuss.

On 2023/11/20 18:04, Simon Horman wrote:
> On Fri, Nov 17, 2023 at 09:06:45AM -0300, Pedro Tammela wrote:
>> On 17/11/2023 06:31, Kunwu Chan wrote:
>>> net/sched/cls_api.c:390:22: warning: incorrect type in assignment (different base types)
>>> net/sched/cls_api.c:390:22:    expected restricted __be16 [usertype] protocol
>>> net/sched/cls_api.c:390:22:    got unsigned int [usertype] protocol
>>>
>>> Fixes: 33a48927c193 ("sched: push TC filter protocol creation into a separate function")
>>>
>>> Signed-off-by: Kunwu Chan <chentao@kylinos.cn>
>>> ---
>>>    net/sched/cls_api.c | 2 +-
>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c
>>> index 1976bd163986..f73f39f61f66 100644
>>> --- a/net/sched/cls_api.c
>>> +++ b/net/sched/cls_api.c
>>> @@ -387,7 +387,7 @@ static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
>>>    		goto errout;
>>>    	}
>>>    	tp->classify = tp->ops->classify;
>>> -	tp->protocol = protocol;
>>> +	tp->protocol = cpu_to_be16(protocol);
>>>    	tp->prio = prio;
>>>    	tp->chain = chain;
>>>    	spin_lock_init(&tp->lock);
>> I don't believe there's something to fix here either
> 
> Hi Pedro and Kunwu,
> 
> I suspect that updating the byte order of protocol isn't correct
> here - else I'd assume we would have seen a user-visible bug on
> little-endian systems buy now.
> 
> But nonetheless I think there is a problem, which is that the appropriate
> types aren't being used, which means the tooling isn't helping us wrt any
> bugs that might subsequently be added or already lurking. So I think an
> appropriate question is, what is the endien and width of protocol, and how
> can we use an appropriate type throughout the call-path?