refactor: optimize TCP and UDP checksum handling with eBPF

[tzssangglass]

udp and tcp ingress:

1. use bpf_l4_csum_replace bpf_l3_csum_replace to recalculate the checksums;
2. and bpf_skb_store_bytes to update dest ip and port;

Signed-off-by: tzssangglass [email protected]

[k8s-ci-robot]

Welcome @tzssangglass!

It looks like this is your first PR to kubernetes-sigs/blixt 🎉. Please refer to our pull request process documentation to help your PR have a smooth ride to approval.

You will be prompted by a bot to use commands during the review process. Do not be afraid to follow the prompts! It is okay to experiment. Here is the bot commands documentation.

You can also check if kubernetes-sigs/blixt has its own contribution guidelines.

You may want to refer to our testing guide if you run into trouble with your tests not passing.

If you are having difficulty getting your pull request seen, please follow the recommended escalation practices. Also, for tips and tricks in the contribution process you may want to read the Kubernetes contributor cheat sheet. We want to make sure your contribution gets all the attention it needs!

Thank you, and welcome to Kubernetes. 😃

[astoycos]

Looking good @tzssangglass Thanks so much for doing this! Just some questions and comments, also looks like you need to run cargo fmt

Otherwise my general comments are

• Try and keep the unsafe closures as minimal as possible
• Can we condense alot of this code into a separate helper function with a good comment describing what it does? Something like (packet_redirect_cksum_update) or something like that?

[astoycos]

Should this be moved to after the cksum calculation code block as it was prior to this patch?

[tzssangglass]

in fact, I found that I cannot put

    if tcp_hdr_ref.rst() == 1 {
        unsafe {
            LB_CONNECTIONS.remove(&client_key)?;
        }
    }

    let mut lb_mapping = LoadBalancerMapping {
        backend,
        backend_key,
        tcp_state,
    };

    update_tcp_conns(tcp_hdr_ref, &client_key, &mut lb_mapping)?;

behind

    let backend_ip = backend.daddr.to_be();
    let ret = set_ipv4_ip_dst(&ctx, TCP_CSUM_OFF, &original_daddr, backend_ip);
    if ret != 0 {
        return Ok(TC_ACT_OK);
    }

    let backend_port = (backend.dport as u16).to_be();
    let ret = set_ipv4_dest_port(&ctx, TCP_CSUM_OFF, &original_dport, backend_port);
    if ret != 0 {
        return Ok(TC_ACT_OK);
    }

otherwise eBPF program would load failed, eBPF verifier shows that access out of bounds

[tzssangglass]

I think accessing tcp_hdr after bpf_redirect_neigh will cause an out-of-bounds error, but I don't know why the original code did not have this issue.

[astoycos]

What is the IS_PSEUDO const doing here?

[tzssangglass]

According to my understanding, IS_PSEUDO | (mem::size_of_val(&backend_ip) as u64) as the last argument of bpf_l4_csum_replace(), it can simultaneously set the IS_PSEUDO flag and specify the size of the field to be updated, allowing the function to correctly recalculate the checksum of the transport layer and consider the impact of the pseudo header.

see more:

1. result of IS_PSEUDO | (mem::size_of_val(&backend_ip) as u64): 00010000 | 0100 -> 00010100

2. in bpf_l4_csum_replace()

   	switch (flags & BPF_F_HDR_FIELD_MASK) {
   	case 0:
   		if (unlikely(from != 0))
   			return -EINVAL;
   
   		inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
   		break;
   	case 2:
   		inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
   		break;
   	case 4:
   		inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
   		break;

   here BPF_F_HDR_FIELD_MASK is

   	enum {
   	BPF_F_HDR_FIELD_MASK		= 0xfULL,
   	};

   flags & BPF_F_HDR_FIELD_MASK: 00010100 & 00001111 -> 00000100, and would choose case 4 to update l4 checksum.

[tzssangglass]

ok, I will update based on the comments.

[shaneutt]

@tzssangglass needs a rebase on main

@astoycos when you're done with your review, if you'd like me to take a pass as well please do ping me

[tzssangglass]

needs a rebase on main

done

[astoycos]

/lgtm
/approve

Huge improvement here thanks and sorry for the slowness @tzssangglass !!!

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: astoycos, tzssangglass

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [astoycos]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[tzssangglass]

oops, it looks like k8s-ci-robot didn't squash the commit.