1// SPDX-License-Identifier: GPL-2.0
2/*
3 * It is possible to use SO_REUSEPORT to open multiple sockets bound to
4 * equivalent local addresses using AF_INET and AF_INET6 at the same time.  If
5 * the AF_INET6 socket has IPV6_V6ONLY set, it's clear which socket should
6 * receive a given incoming packet.  However, when it is not set, incoming v4
7 * packets should prefer the AF_INET socket(s).  This behavior was defined with
8 * the original SO_REUSEPORT implementation, but broke with
9 * e32ea7e74727 ("soreuseport: fast reuseport UDP socket selection")
10 * This test creates these mixed AF_INET/AF_INET6 sockets and asserts the
11 * AF_INET preference for v4 packets.
12 */
13
14#define _GNU_SOURCE
15
16#include <arpa/inet.h>
17#include <errno.h>
18#include <error.h>
19#include <linux/in.h>
20#include <linux/unistd.h>
21#include <stdio.h>
22#include <stdlib.h>
23#include <string.h>
24#include <sys/epoll.h>
25#include <sys/types.h>
26#include <sys/socket.h>
27#include <unistd.h>
28
29static const int PORT = 8888;
30
31static void build_rcv_fd(int family, int proto, int *rcv_fds, int count)
32{
33	struct sockaddr_storage addr;
34	struct sockaddr_in  *addr4;
35	struct sockaddr_in6 *addr6;
36	int opt, i;
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 < count; ++i) {
56		rcv_fds[i] = socket(family, proto, 0);
57		if (rcv_fds[i] < 0)
58			error(1, errno, "failed to create receive socket");
59
60		opt = 1;
61		if (setsockopt(rcv_fds[i], SOL_SOCKET, SO_REUSEPORT, &opt,
62			       sizeof(opt)))
63			error(1, errno, "failed to set SO_REUSEPORT");
64
65		if (bind(rcv_fds[i], (struct sockaddr *)&addr, sizeof(addr)))
66			error(1, errno, "failed to bind receive socket");
67
68		if (proto == SOCK_STREAM && listen(rcv_fds[i], 10))
69			error(1, errno, "failed to listen on receive port");
70	}
71}
72
73static void send_from_v4(int proto)
74{
75	struct sockaddr_in  saddr, daddr;
76	int fd;
77
78	saddr.sin_family = AF_INET;
79	saddr.sin_addr.s_addr = htonl(INADDR_ANY);
80	saddr.sin_port = 0;
81
82	daddr.sin_family = AF_INET;
83	daddr.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
84	daddr.sin_port = htons(PORT);
85
86	fd = socket(AF_INET, proto, 0);
87	if (fd < 0)
88		error(1, errno, "failed to create send socket");
89
90	if (bind(fd, (struct sockaddr *)&saddr, sizeof(saddr)))
91		error(1, errno, "failed to bind send socket");
92
93	if (connect(fd, (struct sockaddr *)&daddr, sizeof(daddr)))
94		error(1, errno, "failed to connect send socket");
95
96	if (send(fd, "a", 1, 0) < 0)
97		error(1, errno, "failed to send message");
98
99	close(fd);
100}
101
102static int receive_once(int epfd, int proto)
103{
104	struct epoll_event ev;
105	int i, fd;
106	char buf[8];
107
108	i = epoll_wait(epfd, &ev, 1, -1);
109	if (i < 0)
110		error(1, errno, "epoll_wait failed");
111
112	if (proto == SOCK_STREAM) {
113		fd = accept(ev.data.fd, NULL, NULL);
114		if (fd < 0)
115			error(1, errno, "failed to accept");
116		i = recv(fd, buf, sizeof(buf), 0);
117		close(fd);
118	} else {
119		i = recv(ev.data.fd, buf, sizeof(buf), 0);
120	}
121
122	if (i < 0)
123		error(1, errno, "failed to recv");
124
125	return ev.data.fd;
126}
127
128static void test(int *rcv_fds, int count, int proto)
129{
130	struct epoll_event ev;
131	int epfd, i, test_fd;
132	int test_family;
133	socklen_t len;
134
135	epfd = epoll_create(1);
136	if (epfd < 0)
137		error(1, errno, "failed to create epoll");
138
139	ev.events = EPOLLIN;
140	for (i = 0; i < count; ++i) {
141		ev.data.fd = rcv_fds[i];
142		if (epoll_ctl(epfd, EPOLL_CTL_ADD, rcv_fds[i], &ev))
143			error(1, errno, "failed to register sock epoll");
144	}
145
146	send_from_v4(proto);
147
148	test_fd = receive_once(epfd, proto);
149	len = sizeof(test_family);
150	if (getsockopt(test_fd, SOL_SOCKET, SO_DOMAIN, &test_family, &len))
151		error(1, errno, "failed to read socket domain");
152	if (test_family != AF_INET)
153		error(1, 0, "expected to receive on v4 socket but got v6 (%d)",
154		      test_family);
155
156	close(epfd);
157}
158
159int main(void)
160{
161	int rcv_fds[32], i;
162
163	fprintf(stderr, "---- UDP IPv4 created before IPv6 ----\n");
164	build_rcv_fd(AF_INET, SOCK_DGRAM, rcv_fds, 5);
165	build_rcv_fd(AF_INET6, SOCK_DGRAM, &(rcv_fds[5]), 5);
166	test(rcv_fds, 10, SOCK_DGRAM);
167	for (i = 0; i < 10; ++i)
168		close(rcv_fds[i]);
169
170	fprintf(stderr, "---- UDP IPv6 created before IPv4 ----\n");
171	build_rcv_fd(AF_INET6, SOCK_DGRAM, rcv_fds, 5);
172	build_rcv_fd(AF_INET, SOCK_DGRAM, &(rcv_fds[5]), 5);
173	test(rcv_fds, 10, SOCK_DGRAM);
174	for (i = 0; i < 10; ++i)
175		close(rcv_fds[i]);
176
177	/* NOTE: UDP socket lookups traverse a different code path when there
178	 * are > 10 sockets in a group.
179	 */
180	fprintf(stderr, "---- UDP IPv4 created before IPv6 (large) ----\n");
181	build_rcv_fd(AF_INET, SOCK_DGRAM, rcv_fds, 16);
182	build_rcv_fd(AF_INET6, SOCK_DGRAM, &(rcv_fds[16]), 16);
183	test(rcv_fds, 32, SOCK_DGRAM);
184	for (i = 0; i < 32; ++i)
185		close(rcv_fds[i]);
186
187	fprintf(stderr, "---- UDP IPv6 created before IPv4 (large) ----\n");
188	build_rcv_fd(AF_INET6, SOCK_DGRAM, rcv_fds, 16);
189	build_rcv_fd(AF_INET, SOCK_DGRAM, &(rcv_fds[16]), 16);
190	test(rcv_fds, 32, SOCK_DGRAM);
191	for (i = 0; i < 32; ++i)
192		close(rcv_fds[i]);
193
194	fprintf(stderr, "---- TCP IPv4 created before IPv6 ----\n");
195	build_rcv_fd(AF_INET, SOCK_STREAM, rcv_fds, 5);
196	build_rcv_fd(AF_INET6, SOCK_STREAM, &(rcv_fds[5]), 5);
197	test(rcv_fds, 10, SOCK_STREAM);
198	for (i = 0; i < 10; ++i)
199		close(rcv_fds[i]);
200
201	fprintf(stderr, "---- TCP IPv6 created before IPv4 ----\n");
202	build_rcv_fd(AF_INET6, SOCK_STREAM, rcv_fds, 5);
203	build_rcv_fd(AF_INET, SOCK_STREAM, &(rcv_fds[5]), 5);
204	test(rcv_fds, 10, SOCK_STREAM);
205	for (i = 0; i < 10; ++i)
206		close(rcv_fds[i]);
207
208	fprintf(stderr, "SUCCESS\n");
209	return 0;
210}
211