1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Test functionality of BPF filters with SO_REUSEPORT. Same test as
4  * in reuseport_bpf_cpu, only as one socket per NUMA node.
5  */
6 
7 #define _GNU_SOURCE
8 
9 #include <arpa/inet.h>
10 #include <errno.h>
11 #include <error.h>
12 #include <linux/filter.h>
13 #include <linux/bpf.h>
14 #include <linux/in.h>
15 #include <linux/unistd.h>
16 #include <sched.h>
17 #include <stdio.h>
18 #include <stdlib.h>
19 #include <string.h>
20 #include <sys/epoll.h>
21 #include <sys/types.h>
22 #include <sys/socket.h>
23 #include <unistd.h>
24 #include <numa.h>
25 
26 #include "../kselftest.h"
27 
28 static const int PORT = 8888;
29 
30 static void build_rcv_group(int *rcv_fd, size_t len, int family, int proto)
31 {
32 	struct sockaddr_storage addr;
33 	struct sockaddr_in  *addr4;
34 	struct sockaddr_in6 *addr6;
35 	size_t i;
36 	int opt;
37 
38 	switch (family) {
39 	case AF_INET:
40 		addr4 = (struct sockaddr_in *)&addr;
41 		addr4->sin_family = AF_INET;
42 		addr4->sin_addr.s_addr = htonl(INADDR_ANY);
43 		addr4->sin_port = htons(PORT);
44 		break;
45 	case AF_INET6:
46 		addr6 = (struct sockaddr_in6 *)&addr;
47 		addr6->sin6_family = AF_INET6;
48 		addr6->sin6_addr = in6addr_any;
49 		addr6->sin6_port = htons(PORT);
50 		break;
51 	default:
52 		error(1, 0, "Unsupported family %d", family);
53 	}
54 
55 	for (i = 0; i < len; ++i) {
56 		rcv_fd[i] = socket(family, proto, 0);
57 		if (rcv_fd[i] < 0)
58 			error(1, errno, "failed to create receive socket");
59 
60 		opt = 1;
61 		if (setsockopt(rcv_fd[i], SOL_SOCKET, SO_REUSEPORT, &opt,
62 			       sizeof(opt)))
63 			error(1, errno, "failed to set SO_REUSEPORT");
64 
65 		if (bind(rcv_fd[i], (struct sockaddr *)&addr, sizeof(addr)))
66 			error(1, errno, "failed to bind receive socket");
67 
68 		if (proto == SOCK_STREAM && listen(rcv_fd[i], len * 10))
69 			error(1, errno, "failed to listen on receive port");
70 	}
71 }
72 
73 static void attach_bpf(int fd)
74 {
75 	static char bpf_log_buf[65536];
76 	static const char bpf_license[] = "";
77 
78 	int bpf_fd;
79 	const struct bpf_insn prog[] = {
80 		/* R0 = bpf_get_numa_node_id() */
81 		{ BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_get_numa_node_id },
82 		/* return R0 */
83 		{ BPF_JMP | BPF_EXIT, 0, 0, 0, 0 }
84 	};
85 	union bpf_attr attr;
86 
87 	memset(&attr, 0, sizeof(attr));
88 	attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
89 	attr.insn_cnt = ARRAY_SIZE(prog);
90 	attr.insns = (unsigned long) &prog;
91 	attr.license = (unsigned long) &bpf_license;
92 	attr.log_buf = (unsigned long) &bpf_log_buf;
93 	attr.log_size = sizeof(bpf_log_buf);
94 	attr.log_level = 1;
95 
96 	bpf_fd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr));
97 	if (bpf_fd < 0)
98 		error(1, errno, "ebpf error. log:\n%s\n", bpf_log_buf);
99 
100 	if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_REUSEPORT_EBPF, &bpf_fd,
101 			sizeof(bpf_fd)))
102 		error(1, errno, "failed to set SO_ATTACH_REUSEPORT_EBPF");
103 
104 	close(bpf_fd);
105 }
106 
107 static void send_from_node(int node_id, int family, int proto)
108 {
109 	struct sockaddr_storage saddr, daddr;
110 	struct sockaddr_in  *saddr4, *daddr4;
111 	struct sockaddr_in6 *saddr6, *daddr6;
112 	int fd;
113 
114 	switch (family) {
115 	case AF_INET:
116 		saddr4 = (struct sockaddr_in *)&saddr;
117 		saddr4->sin_family = AF_INET;
118 		saddr4->sin_addr.s_addr = htonl(INADDR_ANY);
119 		saddr4->sin_port = 0;
120 
121 		daddr4 = (struct sockaddr_in *)&daddr;
122 		daddr4->sin_family = AF_INET;
123 		daddr4->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
124 		daddr4->sin_port = htons(PORT);
125 		break;
126 	case AF_INET6:
127 		saddr6 = (struct sockaddr_in6 *)&saddr;
128 		saddr6->sin6_family = AF_INET6;
129 		saddr6->sin6_addr = in6addr_any;
130 		saddr6->sin6_port = 0;
131 
132 		daddr6 = (struct sockaddr_in6 *)&daddr;
133 		daddr6->sin6_family = AF_INET6;
134 		daddr6->sin6_addr = in6addr_loopback;
135 		daddr6->sin6_port = htons(PORT);
136 		break;
137 	default:
138 		error(1, 0, "Unsupported family %d", family);
139 	}
140 
141 	if (numa_run_on_node(node_id) < 0)
142 		error(1, errno, "failed to pin to node");
143 
144 	fd = socket(family, proto, 0);
145 	if (fd < 0)
146 		error(1, errno, "failed to create send socket");
147 
148 	if (bind(fd, (struct sockaddr *)&saddr, sizeof(saddr)))
149 		error(1, errno, "failed to bind send socket");
150 
151 	if (connect(fd, (struct sockaddr *)&daddr, sizeof(daddr)))
152 		error(1, errno, "failed to connect send socket");
153 
154 	if (send(fd, "a", 1, 0) < 0)
155 		error(1, errno, "failed to send message");
156 
157 	close(fd);
158 }
159 
160 static
161 void receive_on_node(int *rcv_fd, int len, int epfd, int node_id, int proto)
162 {
163 	struct epoll_event ev;
164 	int i, fd;
165 	char buf[8];
166 
167 	i = epoll_wait(epfd, &ev, 1, -1);
168 	if (i < 0)
169 		error(1, errno, "epoll_wait failed");
170 
171 	if (proto == SOCK_STREAM) {
172 		fd = accept(ev.data.fd, NULL, NULL);
173 		if (fd < 0)
174 			error(1, errno, "failed to accept");
175 		i = recv(fd, buf, sizeof(buf), 0);
176 		close(fd);
177 	} else {
178 		i = recv(ev.data.fd, buf, sizeof(buf), 0);
179 	}
180 
181 	if (i < 0)
182 		error(1, errno, "failed to recv");
183 
184 	for (i = 0; i < len; ++i)
185 		if (ev.data.fd == rcv_fd[i])
186 			break;
187 	if (i == len)
188 		error(1, 0, "failed to find socket");
189 	fprintf(stderr, "send node %d, receive socket %d\n", node_id, i);
190 	if (node_id != i)
191 		error(1, 0, "node id/receive socket mismatch");
192 }
193 
194 static void test(int *rcv_fd, int len, int family, int proto)
195 {
196 	struct epoll_event ev;
197 	int epfd, node;
198 
199 	build_rcv_group(rcv_fd, len, family, proto);
200 	attach_bpf(rcv_fd[0]);
201 
202 	epfd = epoll_create(1);
203 	if (epfd < 0)
204 		error(1, errno, "failed to create epoll");
205 	for (node = 0; node < len; ++node) {
206 		ev.events = EPOLLIN;
207 		ev.data.fd = rcv_fd[node];
208 		if (epoll_ctl(epfd, EPOLL_CTL_ADD, rcv_fd[node], &ev))
209 			error(1, errno, "failed to register sock epoll");
210 	}
211 
212 	/* Forward iterate */
213 	for (node = 0; node < len; ++node) {
214 		if (!numa_bitmask_isbitset(numa_nodes_ptr, node))
215 			continue;
216 		send_from_node(node, family, proto);
217 		receive_on_node(rcv_fd, len, epfd, node, proto);
218 	}
219 
220 	/* Reverse iterate */
221 	for (node = len - 1; node >= 0; --node) {
222 		if (!numa_bitmask_isbitset(numa_nodes_ptr, node))
223 			continue;
224 		send_from_node(node, family, proto);
225 		receive_on_node(rcv_fd, len, epfd, node, proto);
226 	}
227 
228 	close(epfd);
229 	for (node = 0; node < len; ++node)
230 		close(rcv_fd[node]);
231 }
232 
233 int main(void)
234 {
235 	int *rcv_fd, nodes;
236 
237 	if (numa_available() < 0)
238 		ksft_exit_skip("no numa api support\n");
239 
240 	nodes = numa_max_node() + 1;
241 
242 	rcv_fd = calloc(nodes, sizeof(int));
243 	if (!rcv_fd)
244 		error(1, 0, "failed to allocate array");
245 
246 	fprintf(stderr, "---- IPv4 UDP ----\n");
247 	test(rcv_fd, nodes, AF_INET, SOCK_DGRAM);
248 
249 	fprintf(stderr, "---- IPv6 UDP ----\n");
250 	test(rcv_fd, nodes, AF_INET6, SOCK_DGRAM);
251 
252 	fprintf(stderr, "---- IPv4 TCP ----\n");
253 	test(rcv_fd, nodes, AF_INET, SOCK_STREAM);
254 
255 	fprintf(stderr, "---- IPv6 TCP ----\n");
256 	test(rcv_fd, nodes, AF_INET6, SOCK_STREAM);
257 
258 	free(rcv_fd);
259 
260 	fprintf(stderr, "SUCCESS\n");
261 	return 0;
262 }
263