1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (c) 2018 Facebook
3
4 #include <string.h>
5
6 #include <linux/stddef.h>
7 #include <linux/bpf.h>
8 #include <linux/in.h>
9 #include <linux/in6.h>
10 #include <sys/socket.h>
11 #include <netinet/tcp.h>
12 #include <linux/if.h>
13 #include <errno.h>
14
15 #include <bpf/bpf_helpers.h>
16 #include <bpf/bpf_endian.h>
17
18 #define SRC_REWRITE_IP4 0x7f000004U
19 #define DST_REWRITE_IP4 0x7f000001U
20 #define DST_REWRITE_PORT4 4444
21
22 #ifndef TCP_CA_NAME_MAX
23 #define TCP_CA_NAME_MAX 16
24 #endif
25
26 #ifndef TCP_NOTSENT_LOWAT
27 #define TCP_NOTSENT_LOWAT 25
28 #endif
29
30 #ifndef IFNAMSIZ
31 #define IFNAMSIZ 16
32 #endif
33
34 int _version SEC("version") = 1;
35
36 __attribute__ ((noinline))
do_bind(struct bpf_sock_addr * ctx)37 int do_bind(struct bpf_sock_addr *ctx)
38 {
39 struct sockaddr_in sa = {};
40
41 sa.sin_family = AF_INET;
42 sa.sin_port = bpf_htons(0);
43 sa.sin_addr.s_addr = bpf_htonl(SRC_REWRITE_IP4);
44
45 if (bpf_bind(ctx, (struct sockaddr *)&sa, sizeof(sa)) != 0)
46 return 0;
47
48 return 1;
49 }
50
verify_cc(struct bpf_sock_addr * ctx,char expected[TCP_CA_NAME_MAX])51 static __inline int verify_cc(struct bpf_sock_addr *ctx,
52 char expected[TCP_CA_NAME_MAX])
53 {
54 char buf[TCP_CA_NAME_MAX];
55 int i;
56
57 if (bpf_getsockopt(ctx, SOL_TCP, TCP_CONGESTION, &buf, sizeof(buf)))
58 return 1;
59
60 for (i = 0; i < TCP_CA_NAME_MAX; i++) {
61 if (buf[i] != expected[i])
62 return 1;
63 if (buf[i] == 0)
64 break;
65 }
66
67 return 0;
68 }
69
set_cc(struct bpf_sock_addr * ctx)70 static __inline int set_cc(struct bpf_sock_addr *ctx)
71 {
72 char reno[TCP_CA_NAME_MAX] = "reno";
73 char cubic[TCP_CA_NAME_MAX] = "cubic";
74
75 if (bpf_setsockopt(ctx, SOL_TCP, TCP_CONGESTION, &reno, sizeof(reno)))
76 return 1;
77 if (verify_cc(ctx, reno))
78 return 1;
79
80 if (bpf_setsockopt(ctx, SOL_TCP, TCP_CONGESTION, &cubic, sizeof(cubic)))
81 return 1;
82 if (verify_cc(ctx, cubic))
83 return 1;
84
85 return 0;
86 }
87
bind_to_device(struct bpf_sock_addr * ctx)88 static __inline int bind_to_device(struct bpf_sock_addr *ctx)
89 {
90 char veth1[IFNAMSIZ] = "test_sock_addr1";
91 char veth2[IFNAMSIZ] = "test_sock_addr2";
92 char missing[IFNAMSIZ] = "nonexistent_dev";
93 char del_bind[IFNAMSIZ] = "";
94
95 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
96 &veth1, sizeof(veth1)))
97 return 1;
98 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
99 &veth2, sizeof(veth2)))
100 return 1;
101 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
102 &missing, sizeof(missing)) != -ENODEV)
103 return 1;
104 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
105 &del_bind, sizeof(del_bind)))
106 return 1;
107
108 return 0;
109 }
110
set_keepalive(struct bpf_sock_addr * ctx)111 static __inline int set_keepalive(struct bpf_sock_addr *ctx)
112 {
113 int zero = 0, one = 1;
114
115 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(one)))
116 return 1;
117 if (ctx->type == SOCK_STREAM) {
118 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPIDLE, &one, sizeof(one)))
119 return 1;
120 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPINTVL, &one, sizeof(one)))
121 return 1;
122 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPCNT, &one, sizeof(one)))
123 return 1;
124 if (bpf_setsockopt(ctx, SOL_TCP, TCP_SYNCNT, &one, sizeof(one)))
125 return 1;
126 if (bpf_setsockopt(ctx, SOL_TCP, TCP_USER_TIMEOUT, &one, sizeof(one)))
127 return 1;
128 }
129 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_KEEPALIVE, &zero, sizeof(zero)))
130 return 1;
131
132 return 0;
133 }
134
set_notsent_lowat(struct bpf_sock_addr * ctx)135 static __inline int set_notsent_lowat(struct bpf_sock_addr *ctx)
136 {
137 int lowat = 65535;
138
139 if (ctx->type == SOCK_STREAM) {
140 if (bpf_setsockopt(ctx, SOL_TCP, TCP_NOTSENT_LOWAT, &lowat, sizeof(lowat)))
141 return 1;
142 }
143
144 return 0;
145 }
146
147 SEC("cgroup/connect4")
connect_v4_prog(struct bpf_sock_addr * ctx)148 int connect_v4_prog(struct bpf_sock_addr *ctx)
149 {
150 struct bpf_sock_tuple tuple = {};
151 struct bpf_sock *sk;
152
153 /* Verify that new destination is available. */
154 memset(&tuple.ipv4.saddr, 0, sizeof(tuple.ipv4.saddr));
155 memset(&tuple.ipv4.sport, 0, sizeof(tuple.ipv4.sport));
156
157 tuple.ipv4.daddr = bpf_htonl(DST_REWRITE_IP4);
158 tuple.ipv4.dport = bpf_htons(DST_REWRITE_PORT4);
159
160 /* Bind to device and unbind it. */
161 if (bind_to_device(ctx))
162 return 0;
163
164 if (set_keepalive(ctx))
165 return 0;
166
167 if (set_notsent_lowat(ctx))
168 return 0;
169
170 if (ctx->type != SOCK_STREAM && ctx->type != SOCK_DGRAM)
171 return 0;
172 else if (ctx->type == SOCK_STREAM)
173 sk = bpf_sk_lookup_tcp(ctx, &tuple, sizeof(tuple.ipv4),
174 BPF_F_CURRENT_NETNS, 0);
175 else
176 sk = bpf_sk_lookup_udp(ctx, &tuple, sizeof(tuple.ipv4),
177 BPF_F_CURRENT_NETNS, 0);
178
179 if (!sk)
180 return 0;
181
182 if (sk->src_ip4 != tuple.ipv4.daddr ||
183 sk->src_port != DST_REWRITE_PORT4) {
184 bpf_sk_release(sk);
185 return 0;
186 }
187
188 bpf_sk_release(sk);
189
190 /* Rewrite congestion control. */
191 if (ctx->type == SOCK_STREAM && set_cc(ctx))
192 return 0;
193
194 /* Rewrite destination. */
195 ctx->user_ip4 = bpf_htonl(DST_REWRITE_IP4);
196 ctx->user_port = bpf_htons(DST_REWRITE_PORT4);
197
198 return do_bind(ctx) ? 1 : 0;
199 }
200
201 char _license[] SEC("license") = "GPL";
202
