1 /*
2 * Test functionality of BPF filters with SO_REUSEPORT. Same test as
3 * in reuseport_bpf_cpu, only as one socket per NUMA node.
4 */
5
6 #define _GNU_SOURCE
7
8 #include <arpa/inet.h>
9 #include <errno.h>
10 #include <error.h>
11 #include <linux/filter.h>
12 #include <linux/bpf.h>
13 #include <linux/in.h>
14 #include <linux/unistd.h>
15 #include <sched.h>
16 #include <stdio.h>
17 #include <stdlib.h>
18 #include <string.h>
19 #include <sys/epoll.h>
20 #include <sys/types.h>
21 #include <sys/socket.h>
22 #include <unistd.h>
23 #include <numa.h>
24
25 static const int PORT = 8888;
26
build_rcv_group(int * rcv_fd,size_t len,int family,int proto)27 static void build_rcv_group(int *rcv_fd, size_t len, int family, int proto)
28 {
29 struct sockaddr_storage addr;
30 struct sockaddr_in *addr4;
31 struct sockaddr_in6 *addr6;
32 size_t i;
33 int opt;
34
35 switch (family) {
36 case AF_INET:
37 addr4 = (struct sockaddr_in *)&addr;
38 addr4->sin_family = AF_INET;
39 addr4->sin_addr.s_addr = htonl(INADDR_ANY);
40 addr4->sin_port = htons(PORT);
41 break;
42 case AF_INET6:
43 addr6 = (struct sockaddr_in6 *)&addr;
44 addr6->sin6_family = AF_INET6;
45 addr6->sin6_addr = in6addr_any;
46 addr6->sin6_port = htons(PORT);
47 break;
48 default:
49 error(1, 0, "Unsupported family %d", family);
50 }
51
52 for (i = 0; i < len; ++i) {
53 rcv_fd[i] = socket(family, proto, 0);
54 if (rcv_fd[i] < 0)
55 error(1, errno, "failed to create receive socket");
56
57 opt = 1;
58 if (setsockopt(rcv_fd[i], SOL_SOCKET, SO_REUSEPORT, &opt,
59 sizeof(opt)))
60 error(1, errno, "failed to set SO_REUSEPORT");
61
62 if (bind(rcv_fd[i], (struct sockaddr *)&addr, sizeof(addr)))
63 error(1, errno, "failed to bind receive socket");
64
65 if (proto == SOCK_STREAM && listen(rcv_fd[i], len * 10))
66 error(1, errno, "failed to listen on receive port");
67 }
68 }
69
attach_bpf(int fd)70 static void attach_bpf(int fd)
71 {
72 static char bpf_log_buf[65536];
73 static const char bpf_license[] = "";
74
75 int bpf_fd;
76 const struct bpf_insn prog[] = {
77 /* R0 = bpf_get_numa_node_id() */
78 { BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_get_numa_node_id },
79 /* return R0 */
80 { BPF_JMP | BPF_EXIT, 0, 0, 0, 0 }
81 };
82 union bpf_attr attr;
83
84 memset(&attr, 0, sizeof(attr));
85 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
86 attr.insn_cnt = sizeof(prog) / sizeof(prog[0]);
87 attr.insns = (unsigned long) &prog;
88 attr.license = (unsigned long) &bpf_license;
89 attr.log_buf = (unsigned long) &bpf_log_buf;
90 attr.log_size = sizeof(bpf_log_buf);
91 attr.log_level = 1;
92
93 bpf_fd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr));
94 if (bpf_fd < 0)
95 error(1, errno, "ebpf error. log:\n%s\n", bpf_log_buf);
96
97 if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_REUSEPORT_EBPF, &bpf_fd,
98 sizeof(bpf_fd)))
99 error(1, errno, "failed to set SO_ATTACH_REUSEPORT_EBPF");
100
101 close(bpf_fd);
102 }
103
send_from_node(int node_id,int family,int proto)104 static void send_from_node(int node_id, int family, int proto)
105 {
106 struct sockaddr_storage saddr, daddr;
107 struct sockaddr_in *saddr4, *daddr4;
108 struct sockaddr_in6 *saddr6, *daddr6;
109 int fd;
110
111 switch (family) {
112 case AF_INET:
113 saddr4 = (struct sockaddr_in *)&saddr;
114 saddr4->sin_family = AF_INET;
115 saddr4->sin_addr.s_addr = htonl(INADDR_ANY);
116 saddr4->sin_port = 0;
117
118 daddr4 = (struct sockaddr_in *)&daddr;
119 daddr4->sin_family = AF_INET;
120 daddr4->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
121 daddr4->sin_port = htons(PORT);
122 break;
123 case AF_INET6:
124 saddr6 = (struct sockaddr_in6 *)&saddr;
125 saddr6->sin6_family = AF_INET6;
126 saddr6->sin6_addr = in6addr_any;
127 saddr6->sin6_port = 0;
128
129 daddr6 = (struct sockaddr_in6 *)&daddr;
130 daddr6->sin6_family = AF_INET6;
131 daddr6->sin6_addr = in6addr_loopback;
132 daddr6->sin6_port = htons(PORT);
133 break;
134 default:
135 error(1, 0, "Unsupported family %d", family);
136 }
137
138 if (numa_run_on_node(node_id) < 0)
139 error(1, errno, "failed to pin to node");
140
141 fd = socket(family, proto, 0);
142 if (fd < 0)
143 error(1, errno, "failed to create send socket");
144
145 if (bind(fd, (struct sockaddr *)&saddr, sizeof(saddr)))
146 error(1, errno, "failed to bind send socket");
147
148 if (connect(fd, (struct sockaddr *)&daddr, sizeof(daddr)))
149 error(1, errno, "failed to connect send socket");
150
151 if (send(fd, "a", 1, 0) < 0)
152 error(1, errno, "failed to send message");
153
154 close(fd);
155 }
156
157 static
receive_on_node(int * rcv_fd,int len,int epfd,int node_id,int proto)158 void receive_on_node(int *rcv_fd, int len, int epfd, int node_id, int proto)
159 {
160 struct epoll_event ev;
161 int i, fd;
162 char buf[8];
163
164 i = epoll_wait(epfd, &ev, 1, -1);
165 if (i < 0)
166 error(1, errno, "epoll_wait failed");
167
168 if (proto == SOCK_STREAM) {
169 fd = accept(ev.data.fd, NULL, NULL);
170 if (fd < 0)
171 error(1, errno, "failed to accept");
172 i = recv(fd, buf, sizeof(buf), 0);
173 close(fd);
174 } else {
175 i = recv(ev.data.fd, buf, sizeof(buf), 0);
176 }
177
178 if (i < 0)
179 error(1, errno, "failed to recv");
180
181 for (i = 0; i < len; ++i)
182 if (ev.data.fd == rcv_fd[i])
183 break;
184 if (i == len)
185 error(1, 0, "failed to find socket");
186 fprintf(stderr, "send node %d, receive socket %d\n", node_id, i);
187 if (node_id != i)
188 error(1, 0, "node id/receive socket mismatch");
189 }
190
test(int * rcv_fd,int len,int family,int proto)191 static void test(int *rcv_fd, int len, int family, int proto)
192 {
193 struct epoll_event ev;
194 int epfd, node;
195
196 build_rcv_group(rcv_fd, len, family, proto);
197 attach_bpf(rcv_fd[0]);
198
199 epfd = epoll_create(1);
200 if (epfd < 0)
201 error(1, errno, "failed to create epoll");
202 for (node = 0; node < len; ++node) {
203 ev.events = EPOLLIN;
204 ev.data.fd = rcv_fd[node];
205 if (epoll_ctl(epfd, EPOLL_CTL_ADD, rcv_fd[node], &ev))
206 error(1, errno, "failed to register sock epoll");
207 }
208
209 /* Forward iterate */
210 for (node = 0; node < len; ++node) {
211 send_from_node(node, family, proto);
212 receive_on_node(rcv_fd, len, epfd, node, proto);
213 }
214
215 /* Reverse iterate */
216 for (node = len - 1; node >= 0; --node) {
217 send_from_node(node, family, proto);
218 receive_on_node(rcv_fd, len, epfd, node, proto);
219 }
220
221 close(epfd);
222 for (node = 0; node < len; ++node)
223 close(rcv_fd[node]);
224 }
225
main(void)226 int main(void)
227 {
228 int *rcv_fd, nodes;
229
230 if (numa_available() < 0)
231 error(1, errno, "no numa api support");
232
233 nodes = numa_max_node() + 1;
234
235 rcv_fd = calloc(nodes, sizeof(int));
236 if (!rcv_fd)
237 error(1, 0, "failed to allocate array");
238
239 fprintf(stderr, "---- IPv4 UDP ----\n");
240 test(rcv_fd, nodes, AF_INET, SOCK_DGRAM);
241
242 fprintf(stderr, "---- IPv6 UDP ----\n");
243 test(rcv_fd, nodes, AF_INET6, SOCK_DGRAM);
244
245 fprintf(stderr, "---- IPv4 TCP ----\n");
246 test(rcv_fd, nodes, AF_INET, SOCK_STREAM);
247
248 fprintf(stderr, "---- IPv6 TCP ----\n");
249 test(rcv_fd, nodes, AF_INET6, SOCK_STREAM);
250
251 free(rcv_fd);
252
253 fprintf(stderr, "SUCCESS\n");
254 return 0;
255 }
256