Re: [PATCH net-next 2/6] netdevsim: psp: remove unnecessary UDP checksum computation

[Daniel Zahka <daniel.zahka@gmail.com>]

On 5/11/26 3:01 PM, Willem de Bruijn wrote:
> Daniel Zahka wrote:
>    
> @@ -81,36 +79,6 @@ nsim_do_psp(struct sk_buff *skb, struct netdevsim *ns,
>    			      skb->len - skb_inner_transport_offset(skb));
>    		u64_stats_update_end(&ns->psp.syncp);
>    	} else {
> -		struct ipv6hdr *ip6h __maybe_unused;
> -		struct iphdr *iph;
> -		struct udphdr *uh;
> -		__wsum csum;
> -
> -		/* Do not decapsulate. Receive the skb with the udp and psp
> -		 * headers still there as if this is a normal udp packet.
> -		 * psp_dev_encapsulate() sets udp checksum to 0, so we need to
> -		 * provide a valid checksum here, so the skb isn't dropped.
> -		 */
>>> Perhaps this was here as IPv6 does not allow zero checksums except for
>>> tunneling in specific cases (RFC 6936)?
>>
>> Yes it was originally here for IPv6. It was needed to make a test case
>> pass that ultimately never got upstreamed (yet). The test basically used
>> the psp_dev_ops::set_config() function to turn psp rx off midflow, and
>> then tried to catch packets with a udp socket listening on port 1000. At
>> the time I didn't realize that I could probably make the test work by
>> opting the socket into a setting for RFC 6936 with setsockopt().
> Just curious: which setsockopt is this?

I was thinking of UDP_NO_CHECK6_RX, though I probably shouldn't have 
characterized it as relating to RFC 6936. I haven't dug through the code 
or history to figure out if that was added for tunnels or to implement 
that rfc. I can see that tunnel driver callers of udp_sock_create() can 
configure the 0 checksum behavior, which seems to actually implement 
what's described in the rfc.

>> though as I mentioned, none of the current tests cover that the
>> psp-udp csum is correct.
> .. actually based on this, preferable even


I agree. My plan would be to add the psp-udp checksum calculation back 
when we have a test that actually depends on it to pass.