162306a36Sopenharmony_ci/* Copyright (c) 2017 Facebook
262306a36Sopenharmony_ci *
362306a36Sopenharmony_ci * This program is free software; you can redistribute it and/or
462306a36Sopenharmony_ci * modify it under the terms of version 2 of the GNU General Public
562306a36Sopenharmony_ci * License as published by the Free Software Foundation.
662306a36Sopenharmony_ci *
762306a36Sopenharmony_ci * BPF program to set congestion control to dctcp when both hosts are
862306a36Sopenharmony_ci * in the same datacenter (as deteremined by IPv6 prefix).
962306a36Sopenharmony_ci *
1062306a36Sopenharmony_ci * Use "bpftool cgroup attach $cg sock_ops $prog" to load this BPF program.
1162306a36Sopenharmony_ci */
1262306a36Sopenharmony_ci
1362306a36Sopenharmony_ci#include <uapi/linux/bpf.h>
1462306a36Sopenharmony_ci#include <uapi/linux/tcp.h>
1562306a36Sopenharmony_ci#include <uapi/linux/if_ether.h>
1662306a36Sopenharmony_ci#include <uapi/linux/if_packet.h>
1762306a36Sopenharmony_ci#include <uapi/linux/ip.h>
1862306a36Sopenharmony_ci#include <linux/socket.h>
1962306a36Sopenharmony_ci#include <bpf/bpf_helpers.h>
2062306a36Sopenharmony_ci#include <bpf/bpf_endian.h>
2162306a36Sopenharmony_ci
2262306a36Sopenharmony_ci#define DEBUG 1
2362306a36Sopenharmony_ci
2462306a36Sopenharmony_ciSEC("sockops")
2562306a36Sopenharmony_ciint bpf_cong(struct bpf_sock_ops *skops)
2662306a36Sopenharmony_ci{
2762306a36Sopenharmony_ci	char cong[] = "dctcp";
2862306a36Sopenharmony_ci	int rv = 0;
2962306a36Sopenharmony_ci	int op;
3062306a36Sopenharmony_ci
3162306a36Sopenharmony_ci	/* For testing purposes, only execute rest of BPF program
3262306a36Sopenharmony_ci	 * if neither port numberis 55601
3362306a36Sopenharmony_ci	 */
3462306a36Sopenharmony_ci	if (bpf_ntohl(skops->remote_port) != 55601 &&
3562306a36Sopenharmony_ci	    skops->local_port != 55601) {
3662306a36Sopenharmony_ci		skops->reply = -1;
3762306a36Sopenharmony_ci		return 1;
3862306a36Sopenharmony_ci	}
3962306a36Sopenharmony_ci
4062306a36Sopenharmony_ci	op = (int) skops->op;
4162306a36Sopenharmony_ci
4262306a36Sopenharmony_ci#ifdef DEBUG
4362306a36Sopenharmony_ci	bpf_printk("BPF command: %d\n", op);
4462306a36Sopenharmony_ci#endif
4562306a36Sopenharmony_ci
4662306a36Sopenharmony_ci	/* Check if both hosts are in the same datacenter. For this
4762306a36Sopenharmony_ci	 * example they are if the 1st 5.5 bytes in the IPv6 address
4862306a36Sopenharmony_ci	 * are the same.
4962306a36Sopenharmony_ci	 */
5062306a36Sopenharmony_ci	if (skops->family == AF_INET6 &&
5162306a36Sopenharmony_ci	    skops->local_ip6[0] == skops->remote_ip6[0] &&
5262306a36Sopenharmony_ci	    (bpf_ntohl(skops->local_ip6[1]) & 0xfff00000) ==
5362306a36Sopenharmony_ci	    (bpf_ntohl(skops->remote_ip6[1]) & 0xfff00000)) {
5462306a36Sopenharmony_ci		switch (op) {
5562306a36Sopenharmony_ci		case BPF_SOCK_OPS_NEEDS_ECN:
5662306a36Sopenharmony_ci			rv = 1;
5762306a36Sopenharmony_ci			break;
5862306a36Sopenharmony_ci		case BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB:
5962306a36Sopenharmony_ci			rv = bpf_setsockopt(skops, SOL_TCP, TCP_CONGESTION,
6062306a36Sopenharmony_ci					    cong, sizeof(cong));
6162306a36Sopenharmony_ci			break;
6262306a36Sopenharmony_ci		case BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB:
6362306a36Sopenharmony_ci			rv = bpf_setsockopt(skops, SOL_TCP, TCP_CONGESTION,
6462306a36Sopenharmony_ci					    cong, sizeof(cong));
6562306a36Sopenharmony_ci			break;
6662306a36Sopenharmony_ci		default:
6762306a36Sopenharmony_ci			rv = -1;
6862306a36Sopenharmony_ci		}
6962306a36Sopenharmony_ci	} else {
7062306a36Sopenharmony_ci		rv = -1;
7162306a36Sopenharmony_ci	}
7262306a36Sopenharmony_ci#ifdef DEBUG
7362306a36Sopenharmony_ci	bpf_printk("Returning %d\n", rv);
7462306a36Sopenharmony_ci#endif
7562306a36Sopenharmony_ci	skops->reply = rv;
7662306a36Sopenharmony_ci	return 1;
7762306a36Sopenharmony_ci}
7862306a36Sopenharmony_cichar _license[] SEC("license") = "GPL";
79