Loading...
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 <linux/tcp.h>
11#include <linux/if.h>
12#include <errno.h>
13
14#include <bpf/bpf_helpers.h>
15#include <bpf/bpf_endian.h>
16
17#define SRC_REWRITE_IP4 0x7f000004U
18#define DST_REWRITE_IP4 0x7f000001U
19#define DST_REWRITE_PORT4 4444
20
21#ifndef TCP_CA_NAME_MAX
22#define TCP_CA_NAME_MAX 16
23#endif
24
25#ifndef TCP_NOTSENT_LOWAT
26#define TCP_NOTSENT_LOWAT 25
27#endif
28
29#ifndef IFNAMSIZ
30#define IFNAMSIZ 16
31#endif
32
33#ifndef SOL_TCP
34#define SOL_TCP 6
35#endif
36
37__attribute__ ((noinline)) __weak
38int do_bind(struct bpf_sock_addr *ctx)
39{
40 struct sockaddr_in sa = {};
41
42 sa.sin_family = AF_INET;
43 sa.sin_port = bpf_htons(0);
44 sa.sin_addr.s_addr = bpf_htonl(SRC_REWRITE_IP4);
45
46 if (bpf_bind(ctx, (struct sockaddr *)&sa, sizeof(sa)) != 0)
47 return 0;
48
49 return 1;
50}
51
52static __inline int verify_cc(struct bpf_sock_addr *ctx,
53 char expected[TCP_CA_NAME_MAX])
54{
55 char buf[TCP_CA_NAME_MAX];
56 int i;
57
58 if (bpf_getsockopt(ctx, SOL_TCP, TCP_CONGESTION, &buf, sizeof(buf)))
59 return 1;
60
61 for (i = 0; i < TCP_CA_NAME_MAX; i++) {
62 if (buf[i] != expected[i])
63 return 1;
64 if (buf[i] == 0)
65 break;
66 }
67
68 return 0;
69}
70
71static __inline int set_cc(struct bpf_sock_addr *ctx)
72{
73 char reno[TCP_CA_NAME_MAX] = "reno";
74 char cubic[TCP_CA_NAME_MAX] = "cubic";
75
76 if (bpf_setsockopt(ctx, SOL_TCP, TCP_CONGESTION, &reno, sizeof(reno)))
77 return 1;
78 if (verify_cc(ctx, reno))
79 return 1;
80
81 if (bpf_setsockopt(ctx, SOL_TCP, TCP_CONGESTION, &cubic, sizeof(cubic)))
82 return 1;
83 if (verify_cc(ctx, cubic))
84 return 1;
85
86 return 0;
87}
88
89static __inline int bind_to_device(struct bpf_sock_addr *ctx)
90{
91 char veth1[IFNAMSIZ] = "test_sock_addr1";
92 char veth2[IFNAMSIZ] = "test_sock_addr2";
93 char missing[IFNAMSIZ] = "nonexistent_dev";
94 char del_bind[IFNAMSIZ] = "";
95
96 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
97 &veth1, sizeof(veth1)))
98 return 1;
99 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
100 &veth2, sizeof(veth2)))
101 return 1;
102 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
103 &missing, sizeof(missing)) != -ENODEV)
104 return 1;
105 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
106 &del_bind, sizeof(del_bind)))
107 return 1;
108
109 return 0;
110}
111
112static __inline int set_keepalive(struct bpf_sock_addr *ctx)
113{
114 int zero = 0, one = 1;
115
116 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(one)))
117 return 1;
118 if (ctx->type == SOCK_STREAM) {
119 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPIDLE, &one, sizeof(one)))
120 return 1;
121 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPINTVL, &one, sizeof(one)))
122 return 1;
123 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPCNT, &one, sizeof(one)))
124 return 1;
125 if (bpf_setsockopt(ctx, SOL_TCP, TCP_SYNCNT, &one, sizeof(one)))
126 return 1;
127 if (bpf_setsockopt(ctx, SOL_TCP, TCP_USER_TIMEOUT, &one, sizeof(one)))
128 return 1;
129 }
130 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_KEEPALIVE, &zero, sizeof(zero)))
131 return 1;
132
133 return 0;
134}
135
136static __inline int set_notsent_lowat(struct bpf_sock_addr *ctx)
137{
138 int lowat = 65535;
139
140 if (ctx->type == SOCK_STREAM) {
141 if (bpf_setsockopt(ctx, SOL_TCP, TCP_NOTSENT_LOWAT, &lowat, sizeof(lowat)))
142 return 1;
143 }
144
145 return 0;
146}
147
148SEC("cgroup/connect4")
149int connect_v4_prog(struct bpf_sock_addr *ctx)
150{
151 struct bpf_sock_tuple tuple = {};
152 struct bpf_sock *sk;
153
154 /* Verify that new destination is available. */
155 memset(&tuple.ipv4.saddr, 0, sizeof(tuple.ipv4.saddr));
156 memset(&tuple.ipv4.sport, 0, sizeof(tuple.ipv4.sport));
157
158 tuple.ipv4.daddr = bpf_htonl(DST_REWRITE_IP4);
159 tuple.ipv4.dport = bpf_htons(DST_REWRITE_PORT4);
160
161 /* Bind to device and unbind it. */
162 if (bind_to_device(ctx))
163 return 0;
164
165 if (set_keepalive(ctx))
166 return 0;
167
168 if (set_notsent_lowat(ctx))
169 return 0;
170
171 if (ctx->type != SOCK_STREAM && ctx->type != SOCK_DGRAM)
172 return 0;
173 else if (ctx->type == SOCK_STREAM)
174 sk = bpf_sk_lookup_tcp(ctx, &tuple, sizeof(tuple.ipv4),
175 BPF_F_CURRENT_NETNS, 0);
176 else
177 sk = bpf_sk_lookup_udp(ctx, &tuple, sizeof(tuple.ipv4),
178 BPF_F_CURRENT_NETNS, 0);
179
180 if (!sk)
181 return 0;
182
183 if (sk->src_ip4 != tuple.ipv4.daddr ||
184 sk->src_port != DST_REWRITE_PORT4) {
185 bpf_sk_release(sk);
186 return 0;
187 }
188
189 bpf_sk_release(sk);
190
191 /* Rewrite congestion control. */
192 if (ctx->type == SOCK_STREAM && set_cc(ctx))
193 return 0;
194
195 /* Rewrite destination. */
196 ctx->user_ip4 = bpf_htonl(DST_REWRITE_IP4);
197 ctx->user_port = bpf_htons(DST_REWRITE_PORT4);
198
199 return do_bind(ctx) ? 1 : 0;
200}
201
202SEC("cgroup/connect4")
203int connect_v4_deny_prog(struct bpf_sock_addr *ctx)
204{
205 return 0;
206}
207
208char _license[] SEC("license") = "GPL";
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 IFNAMSIZ
27#define IFNAMSIZ 16
28#endif
29
30int _version SEC("version") = 1;
31
32__attribute__ ((noinline))
33int do_bind(struct bpf_sock_addr *ctx)
34{
35 struct sockaddr_in sa = {};
36
37 sa.sin_family = AF_INET;
38 sa.sin_port = bpf_htons(0);
39 sa.sin_addr.s_addr = bpf_htonl(SRC_REWRITE_IP4);
40
41 if (bpf_bind(ctx, (struct sockaddr *)&sa, sizeof(sa)) != 0)
42 return 0;
43
44 return 1;
45}
46
47static __inline int verify_cc(struct bpf_sock_addr *ctx,
48 char expected[TCP_CA_NAME_MAX])
49{
50 char buf[TCP_CA_NAME_MAX];
51 int i;
52
53 if (bpf_getsockopt(ctx, SOL_TCP, TCP_CONGESTION, &buf, sizeof(buf)))
54 return 1;
55
56 for (i = 0; i < TCP_CA_NAME_MAX; i++) {
57 if (buf[i] != expected[i])
58 return 1;
59 if (buf[i] == 0)
60 break;
61 }
62
63 return 0;
64}
65
66static __inline int set_cc(struct bpf_sock_addr *ctx)
67{
68 char reno[TCP_CA_NAME_MAX] = "reno";
69 char cubic[TCP_CA_NAME_MAX] = "cubic";
70
71 if (bpf_setsockopt(ctx, SOL_TCP, TCP_CONGESTION, &reno, sizeof(reno)))
72 return 1;
73 if (verify_cc(ctx, reno))
74 return 1;
75
76 if (bpf_setsockopt(ctx, SOL_TCP, TCP_CONGESTION, &cubic, sizeof(cubic)))
77 return 1;
78 if (verify_cc(ctx, cubic))
79 return 1;
80
81 return 0;
82}
83
84static __inline int bind_to_device(struct bpf_sock_addr *ctx)
85{
86 char veth1[IFNAMSIZ] = "test_sock_addr1";
87 char veth2[IFNAMSIZ] = "test_sock_addr2";
88 char missing[IFNAMSIZ] = "nonexistent_dev";
89 char del_bind[IFNAMSIZ] = "";
90
91 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
92 &veth1, sizeof(veth1)))
93 return 1;
94 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
95 &veth2, sizeof(veth2)))
96 return 1;
97 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
98 &missing, sizeof(missing)) != -ENODEV)
99 return 1;
100 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
101 &del_bind, sizeof(del_bind)))
102 return 1;
103
104 return 0;
105}
106
107static __inline int set_keepalive(struct bpf_sock_addr *ctx)
108{
109 int zero = 0, one = 1;
110
111 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(one)))
112 return 1;
113 if (ctx->type == SOCK_STREAM) {
114 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPIDLE, &one, sizeof(one)))
115 return 1;
116 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPINTVL, &one, sizeof(one)))
117 return 1;
118 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPCNT, &one, sizeof(one)))
119 return 1;
120 if (bpf_setsockopt(ctx, SOL_TCP, TCP_SYNCNT, &one, sizeof(one)))
121 return 1;
122 if (bpf_setsockopt(ctx, SOL_TCP, TCP_USER_TIMEOUT, &one, sizeof(one)))
123 return 1;
124 }
125 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_KEEPALIVE, &zero, sizeof(zero)))
126 return 1;
127
128 return 0;
129}
130
131SEC("cgroup/connect4")
132int connect_v4_prog(struct bpf_sock_addr *ctx)
133{
134 struct bpf_sock_tuple tuple = {};
135 struct bpf_sock *sk;
136
137 /* Verify that new destination is available. */
138 memset(&tuple.ipv4.saddr, 0, sizeof(tuple.ipv4.saddr));
139 memset(&tuple.ipv4.sport, 0, sizeof(tuple.ipv4.sport));
140
141 tuple.ipv4.daddr = bpf_htonl(DST_REWRITE_IP4);
142 tuple.ipv4.dport = bpf_htons(DST_REWRITE_PORT4);
143
144 /* Bind to device and unbind it. */
145 if (bind_to_device(ctx))
146 return 0;
147
148 if (set_keepalive(ctx))
149 return 0;
150
151 if (ctx->type != SOCK_STREAM && ctx->type != SOCK_DGRAM)
152 return 0;
153 else if (ctx->type == SOCK_STREAM)
154 sk = bpf_sk_lookup_tcp(ctx, &tuple, sizeof(tuple.ipv4),
155 BPF_F_CURRENT_NETNS, 0);
156 else
157 sk = bpf_sk_lookup_udp(ctx, &tuple, sizeof(tuple.ipv4),
158 BPF_F_CURRENT_NETNS, 0);
159
160 if (!sk)
161 return 0;
162
163 if (sk->src_ip4 != tuple.ipv4.daddr ||
164 sk->src_port != DST_REWRITE_PORT4) {
165 bpf_sk_release(sk);
166 return 0;
167 }
168
169 bpf_sk_release(sk);
170
171 /* Rewrite congestion control. */
172 if (ctx->type == SOCK_STREAM && set_cc(ctx))
173 return 0;
174
175 /* Rewrite destination. */
176 ctx->user_ip4 = bpf_htonl(DST_REWRITE_IP4);
177 ctx->user_port = bpf_htons(DST_REWRITE_PORT4);
178
179 return do_bind(ctx) ? 1 : 0;
180}
181
182char _license[] SEC("license") = "GPL";