1// SPDX-License-Identifier: GPL-2.0
  2/* Copyright (c) 2020, Tessares SA. */
  3/* Copyright (c) 2024, Kylin Software */
  4
  5/* vmlinux.h, bpf_helpers.h and other 'define' */
  6#include "bpf_tracing_net.h"
  7#include "mptcp_bpf.h"
  8
  9char _license[] SEC("license") = "GPL";
 10
 11char cc[TCP_CA_NAME_MAX] = "reno";
 12int pid;
 13
 14/* Associate a subflow counter to each token */
 15struct {
 16	__uint(type, BPF_MAP_TYPE_HASH);
 17	__uint(key_size, sizeof(__u32));
 18	__uint(value_size, sizeof(__u32));
 19	__uint(max_entries, 100);
 20} mptcp_sf SEC(".maps");
 21
 22SEC("sockops")
 23int mptcp_subflow(struct bpf_sock_ops *skops)
 24{
 25	__u32 init = 1, key, mark, *cnt;
 26	struct mptcp_sock *msk;
 27	struct bpf_sock *sk;
 28	int err;
 29
 30	if (skops->op != BPF_SOCK_OPS_TCP_CONNECT_CB)
 31		return 1;
 32
 33	sk = skops->sk;
 34	if (!sk)
 35		return 1;
 36
 37	msk = bpf_skc_to_mptcp_sock(sk);
 38	if (!msk)
 39		return 1;
 40
 41	key = msk->token;
 42	cnt = bpf_map_lookup_elem(&mptcp_sf, &key);
 43	if (cnt) {
 44		/* A new subflow is added to an existing MPTCP connection */
 45		__sync_fetch_and_add(cnt, 1);
 46		mark = *cnt;
 47	} else {
 48		/* A new MPTCP connection is just initiated and this is its primary subflow */
 49		bpf_map_update_elem(&mptcp_sf, &key, &init, BPF_ANY);
 50		mark = init;
 51	}
 52
 53	/* Set the mark of the subflow's socket based on appearance order */
 54	err = bpf_setsockopt(skops, SOL_SOCKET, SO_MARK, &mark, sizeof(mark));
 55	if (err < 0)
 56		return 1;
 57	if (mark == 2)
 58		err = bpf_setsockopt(skops, SOL_TCP, TCP_CONGESTION, cc, TCP_CA_NAME_MAX);
 59
 60	return 1;
 61}
 62
 63static int _check_getsockopt_subflow_mark(struct mptcp_sock *msk, struct bpf_sockopt *ctx)
 64{
 65	struct mptcp_subflow_context *subflow;
 66	int i = 0;
 67
 68	mptcp_for_each_subflow(msk, subflow) {
 69		struct sock *ssk;
 70
 71		ssk = mptcp_subflow_tcp_sock(bpf_core_cast(subflow,
 72							   struct mptcp_subflow_context));
 73
 74		if (ssk->sk_mark != ++i) {
 75			ctx->retval = -2;
 76			break;
 77		}
 78	}
 79
 80	return 1;
 81}
 82
 83static int _check_getsockopt_subflow_cc(struct mptcp_sock *msk, struct bpf_sockopt *ctx)
 84{
 85	struct mptcp_subflow_context *subflow;
 86
 87	mptcp_for_each_subflow(msk, subflow) {
 88		struct inet_connection_sock *icsk;
 89		struct sock *ssk;
 90
 91		ssk = mptcp_subflow_tcp_sock(bpf_core_cast(subflow,
 92							   struct mptcp_subflow_context));
 93		icsk = bpf_core_cast(ssk, struct inet_connection_sock);
 94
 95		if (ssk->sk_mark == 2 &&
 96		    __builtin_memcmp(icsk->icsk_ca_ops->name, cc, TCP_CA_NAME_MAX)) {
 97			ctx->retval = -2;
 98			break;
 99		}
100	}
101
102	return 1;
103}
104
105SEC("cgroup/getsockopt")
106int _getsockopt_subflow(struct bpf_sockopt *ctx)
107{
108	struct bpf_sock *sk = ctx->sk;
109	struct mptcp_sock *msk;
110
111	if (bpf_get_current_pid_tgid() >> 32 != pid)
112		return 1;
113
114	if (!sk || sk->protocol != IPPROTO_MPTCP ||
115	    (!(ctx->level == SOL_SOCKET && ctx->optname == SO_MARK) &&
116	     !(ctx->level == SOL_TCP && ctx->optname == TCP_CONGESTION)))
117		return 1;
118
119	msk = bpf_core_cast(sk, struct mptcp_sock);
120	if (msk->pm.subflows != 1) {
121		ctx->retval = -1;
122		return 1;
123	}
124
125	if (ctx->optname == SO_MARK)
126		return _check_getsockopt_subflow_mark(msk, ctx);
127	return _check_getsockopt_subflow_cc(msk, ctx);
128}