2008/688 Sun Cluster TCP/IP Hooks Update

[Huafeng Lu]

James Carlson 写道:
> Erik Nordmark writes:
>> If you understand udp.c then you should be able to convince yourself 
>> that your use of udp_v6lastdst doesn't interfere with its use in TX.
>> If you don't understand udp.c sufficiently well to do that, then it 
>> might be that you shouldn't touch udp.c in the first place.
> 
> I agree with Erik on this point, but I don't think this is actually an
> architectural issue, but rather a matter of design.  I'd encourage the
> folks involved to handle it off of the ARC mailing list.
> 
> The other issue -- lack of IP Instances information -- is
> architectural, as it involves the support of one project by another.
> As long as we're suffering a pair of linked patches to get to the new
> API here, it does seem reasonable to say that we could include the
> required arguments to make stack instances work in the future.
> 
> As long as we're paying the patching complexity price now, is there a
> specific reason to _avoid_ doing that?

To support the IP instance information, and to support future changes, 
we could change the API to use a TLV (type, length, value). We can put 
all paramters in a structure like:
	typedef cl_hook_param {
		int	cl_type;
		size_t	cl_len;
		uint8_t	cl_value[];
	} cl_hook_param_t;
Thus, all future changes can be handled.

This will of course change the signature of the hook; but since we're 
already here, we think it's ok to make the change.

--
Huafeng