xref: /linux/tools/testing/selftests/bpf/prog_tests/select_reuseport.c (revision 3a39d672e7f48b8d6b91a09afa4b55352773b4b5)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2018 Facebook */
3 
4 #include <stdlib.h>
5 #include <unistd.h>
6 #include <stdbool.h>
7 #include <string.h>
8 #include <errno.h>
9 #include <assert.h>
10 #include <fcntl.h>
11 #include <linux/bpf.h>
12 #include <linux/err.h>
13 #include <linux/types.h>
14 #include <linux/if_ether.h>
15 #include <sys/types.h>
16 #include <sys/epoll.h>
17 #include <sys/socket.h>
18 #include <netinet/in.h>
19 #include <bpf/bpf.h>
20 #include <bpf/libbpf.h>
21 #include "bpf_util.h"
22 
23 #include "test_progs.h"
24 #include "test_select_reuseport_common.h"
25 
26 #define MAX_TEST_NAME 80
27 #define MIN_TCPHDR_LEN 20
28 #define UDPHDR_LEN 8
29 
30 #define TCP_SYNCOOKIE_SYSCTL "/proc/sys/net/ipv4/tcp_syncookies"
31 #define TCP_FO_SYSCTL "/proc/sys/net/ipv4/tcp_fastopen"
32 #define REUSEPORT_ARRAY_SIZE 32
33 
34 static int result_map, tmp_index_ovr_map, linum_map, data_check_map;
35 static __u32 expected_results[NR_RESULTS];
36 static int sk_fds[REUSEPORT_ARRAY_SIZE];
37 static int reuseport_array = -1, outer_map = -1;
38 static enum bpf_map_type inner_map_type;
39 static int select_by_skb_data_prog;
40 static struct bpf_object *obj;
41 static __u32 index_zero;
42 static int epfd;
43 
44 static union sa46 {
45 	struct sockaddr_in6 v6;
46 	struct sockaddr_in v4;
47 	sa_family_t family;
48 } srv_sa;
49 
50 #define RET_IF(condition, tag, format...) ({				\
51 	if (CHECK_FAIL(condition)) {					\
52 		printf(tag " " format);					\
53 		return;							\
54 	}								\
55 })
56 
57 #define RET_ERR(condition, tag, format...) ({				\
58 	if (CHECK_FAIL(condition)) {					\
59 		printf(tag " " format);					\
60 		return -1;						\
61 	}								\
62 })
63 
create_maps(enum bpf_map_type inner_type)64 static int create_maps(enum bpf_map_type inner_type)
65 {
66 	LIBBPF_OPTS(bpf_map_create_opts, opts);
67 
68 	inner_map_type = inner_type;
69 
70 	/* Creating reuseport_array */
71 	reuseport_array = bpf_map_create(inner_type, "reuseport_array",
72 					 sizeof(__u32), sizeof(__u32), REUSEPORT_ARRAY_SIZE, NULL);
73 	RET_ERR(reuseport_array < 0, "creating reuseport_array",
74 		"reuseport_array:%d errno:%d\n", reuseport_array, errno);
75 
76 	/* Creating outer_map */
77 	opts.inner_map_fd = reuseport_array;
78 	outer_map = bpf_map_create(BPF_MAP_TYPE_ARRAY_OF_MAPS, "outer_map",
79 				   sizeof(__u32), sizeof(__u32), 1, &opts);
80 	RET_ERR(outer_map < 0, "creating outer_map",
81 		"outer_map:%d errno:%d\n", outer_map, errno);
82 
83 	return 0;
84 }
85 
prepare_bpf_obj(void)86 static int prepare_bpf_obj(void)
87 {
88 	struct bpf_program *prog;
89 	struct bpf_map *map;
90 	int err;
91 
92 	obj = bpf_object__open("test_select_reuseport_kern.bpf.o");
93 	err = libbpf_get_error(obj);
94 	RET_ERR(err, "open test_select_reuseport_kern.bpf.o",
95 		"obj:%p PTR_ERR(obj):%d\n", obj, err);
96 
97 	map = bpf_object__find_map_by_name(obj, "outer_map");
98 	RET_ERR(!map, "find outer_map", "!map\n");
99 	err = bpf_map__reuse_fd(map, outer_map);
100 	RET_ERR(err, "reuse outer_map", "err:%d\n", err);
101 
102 	err = bpf_object__load(obj);
103 	RET_ERR(err, "load bpf_object", "err:%d\n", err);
104 
105 	prog = bpf_object__next_program(obj, NULL);
106 	RET_ERR(!prog, "get first bpf_program", "!prog\n");
107 	select_by_skb_data_prog = bpf_program__fd(prog);
108 	RET_ERR(select_by_skb_data_prog < 0, "get prog fd",
109 		"select_by_skb_data_prog:%d\n", select_by_skb_data_prog);
110 
111 	map = bpf_object__find_map_by_name(obj, "result_map");
112 	RET_ERR(!map, "find result_map", "!map\n");
113 	result_map = bpf_map__fd(map);
114 	RET_ERR(result_map < 0, "get result_map fd",
115 		"result_map:%d\n", result_map);
116 
117 	map = bpf_object__find_map_by_name(obj, "tmp_index_ovr_map");
118 	RET_ERR(!map, "find tmp_index_ovr_map\n", "!map");
119 	tmp_index_ovr_map = bpf_map__fd(map);
120 	RET_ERR(tmp_index_ovr_map < 0, "get tmp_index_ovr_map fd",
121 		"tmp_index_ovr_map:%d\n", tmp_index_ovr_map);
122 
123 	map = bpf_object__find_map_by_name(obj, "linum_map");
124 	RET_ERR(!map, "find linum_map", "!map\n");
125 	linum_map = bpf_map__fd(map);
126 	RET_ERR(linum_map < 0, "get linum_map fd",
127 		"linum_map:%d\n", linum_map);
128 
129 	map = bpf_object__find_map_by_name(obj, "data_check_map");
130 	RET_ERR(!map, "find data_check_map", "!map\n");
131 	data_check_map = bpf_map__fd(map);
132 	RET_ERR(data_check_map < 0, "get data_check_map fd",
133 		"data_check_map:%d\n", data_check_map);
134 
135 	return 0;
136 }
137 
sa46_init_loopback(union sa46 * sa,sa_family_t family)138 static void sa46_init_loopback(union sa46 *sa, sa_family_t family)
139 {
140 	memset(sa, 0, sizeof(*sa));
141 	sa->family = family;
142 	if (sa->family == AF_INET6)
143 		sa->v6.sin6_addr = in6addr_loopback;
144 	else
145 		sa->v4.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
146 }
147 
sa46_init_inany(union sa46 * sa,sa_family_t family)148 static void sa46_init_inany(union sa46 *sa, sa_family_t family)
149 {
150 	memset(sa, 0, sizeof(*sa));
151 	sa->family = family;
152 	if (sa->family == AF_INET6)
153 		sa->v6.sin6_addr = in6addr_any;
154 	else
155 		sa->v4.sin_addr.s_addr = INADDR_ANY;
156 }
157 
read_int_sysctl(const char * sysctl)158 static int read_int_sysctl(const char *sysctl)
159 {
160 	char buf[16];
161 	int fd, ret;
162 
163 	fd = open(sysctl, 0);
164 	RET_ERR(fd == -1, "open(sysctl)",
165 		"sysctl:%s fd:%d errno:%d\n", sysctl, fd, errno);
166 
167 	ret = read(fd, buf, sizeof(buf));
168 	RET_ERR(ret <= 0, "read(sysctl)",
169 		"sysctl:%s ret:%d errno:%d\n", sysctl, ret, errno);
170 
171 	close(fd);
172 	return atoi(buf);
173 }
174 
write_int_sysctl(const char * sysctl,int v)175 static int write_int_sysctl(const char *sysctl, int v)
176 {
177 	int fd, ret, size;
178 	char buf[16];
179 
180 	fd = open(sysctl, O_RDWR);
181 	RET_ERR(fd == -1, "open(sysctl)",
182 		"sysctl:%s fd:%d errno:%d\n", sysctl, fd, errno);
183 
184 	size = snprintf(buf, sizeof(buf), "%d", v);
185 	ret = write(fd, buf, size);
186 	RET_ERR(ret != size, "write(sysctl)",
187 		"sysctl:%s ret:%d size:%d errno:%d\n",
188 		sysctl, ret, size, errno);
189 
190 	close(fd);
191 	return 0;
192 }
193 
enable_fastopen(void)194 static int enable_fastopen(void)
195 {
196 	int fo;
197 
198 	fo = read_int_sysctl(TCP_FO_SYSCTL);
199 	if (fo < 0)
200 		return -1;
201 
202 	return write_int_sysctl(TCP_FO_SYSCTL, fo | 7);
203 }
204 
enable_syncookie(void)205 static int enable_syncookie(void)
206 {
207 	return write_int_sysctl(TCP_SYNCOOKIE_SYSCTL, 2);
208 }
209 
disable_syncookie(void)210 static int disable_syncookie(void)
211 {
212 	return write_int_sysctl(TCP_SYNCOOKIE_SYSCTL, 0);
213 }
214 
get_linum(void)215 static long get_linum(void)
216 {
217 	__u32 linum;
218 	int err;
219 
220 	err = bpf_map_lookup_elem(linum_map, &index_zero, &linum);
221 	RET_ERR(err < 0, "lookup_elem(linum_map)", "err:%d errno:%d\n",
222 		err, errno);
223 
224 	return linum;
225 }
226 
check_data(int type,sa_family_t family,const struct cmd * cmd,int cli_fd)227 static void check_data(int type, sa_family_t family, const struct cmd *cmd,
228 		       int cli_fd)
229 {
230 	struct data_check expected = {}, result;
231 	union sa46 cli_sa;
232 	socklen_t addrlen;
233 	int err;
234 
235 	addrlen = sizeof(cli_sa);
236 	err = getsockname(cli_fd, (struct sockaddr *)&cli_sa,
237 			  &addrlen);
238 	RET_IF(err < 0, "getsockname(cli_fd)", "err:%d errno:%d\n",
239 	       err, errno);
240 
241 	err = bpf_map_lookup_elem(data_check_map, &index_zero, &result);
242 	RET_IF(err < 0, "lookup_elem(data_check_map)", "err:%d errno:%d\n",
243 	       err, errno);
244 
245 	if (type == SOCK_STREAM) {
246 		expected.len = MIN_TCPHDR_LEN;
247 		expected.ip_protocol = IPPROTO_TCP;
248 	} else {
249 		expected.len = UDPHDR_LEN;
250 		expected.ip_protocol = IPPROTO_UDP;
251 	}
252 
253 	if (family == AF_INET6) {
254 		expected.eth_protocol = htons(ETH_P_IPV6);
255 		expected.bind_inany = !srv_sa.v6.sin6_addr.s6_addr32[3] &&
256 			!srv_sa.v6.sin6_addr.s6_addr32[2] &&
257 			!srv_sa.v6.sin6_addr.s6_addr32[1] &&
258 			!srv_sa.v6.sin6_addr.s6_addr32[0];
259 
260 		memcpy(&expected.skb_addrs[0], cli_sa.v6.sin6_addr.s6_addr32,
261 		       sizeof(cli_sa.v6.sin6_addr));
262 		memcpy(&expected.skb_addrs[4], &in6addr_loopback,
263 		       sizeof(in6addr_loopback));
264 		expected.skb_ports[0] = cli_sa.v6.sin6_port;
265 		expected.skb_ports[1] = srv_sa.v6.sin6_port;
266 	} else {
267 		expected.eth_protocol = htons(ETH_P_IP);
268 		expected.bind_inany = !srv_sa.v4.sin_addr.s_addr;
269 
270 		expected.skb_addrs[0] = cli_sa.v4.sin_addr.s_addr;
271 		expected.skb_addrs[1] = htonl(INADDR_LOOPBACK);
272 		expected.skb_ports[0] = cli_sa.v4.sin_port;
273 		expected.skb_ports[1] = srv_sa.v4.sin_port;
274 	}
275 
276 	if (memcmp(&result, &expected, offsetof(struct data_check,
277 						equal_check_end))) {
278 		printf("unexpected data_check\n");
279 		printf("  result: (0x%x, %u, %u)\n",
280 		       result.eth_protocol, result.ip_protocol,
281 		       result.bind_inany);
282 		printf("expected: (0x%x, %u, %u)\n",
283 		       expected.eth_protocol, expected.ip_protocol,
284 		       expected.bind_inany);
285 		RET_IF(1, "data_check result != expected",
286 		       "bpf_prog_linum:%ld\n", get_linum());
287 	}
288 
289 	RET_IF(!result.hash, "data_check result.hash empty",
290 	       "result.hash:%u", result.hash);
291 
292 	expected.len += cmd ? sizeof(*cmd) : 0;
293 	if (type == SOCK_STREAM)
294 		RET_IF(expected.len > result.len, "expected.len > result.len",
295 		       "expected.len:%u result.len:%u bpf_prog_linum:%ld\n",
296 		       expected.len, result.len, get_linum());
297 	else
298 		RET_IF(expected.len != result.len, "expected.len != result.len",
299 		       "expected.len:%u result.len:%u bpf_prog_linum:%ld\n",
300 		       expected.len, result.len, get_linum());
301 }
302 
result_to_str(enum result res)303 static const char *result_to_str(enum result res)
304 {
305 	switch (res) {
306 	case DROP_ERR_INNER_MAP:
307 		return "DROP_ERR_INNER_MAP";
308 	case DROP_ERR_SKB_DATA:
309 		return "DROP_ERR_SKB_DATA";
310 	case DROP_ERR_SK_SELECT_REUSEPORT:
311 		return "DROP_ERR_SK_SELECT_REUSEPORT";
312 	case DROP_MISC:
313 		return "DROP_MISC";
314 	case PASS:
315 		return "PASS";
316 	case PASS_ERR_SK_SELECT_REUSEPORT:
317 		return "PASS_ERR_SK_SELECT_REUSEPORT";
318 	default:
319 		return "UNKNOWN";
320 	}
321 }
322 
check_results(void)323 static void check_results(void)
324 {
325 	__u32 results[NR_RESULTS];
326 	__u32 i, broken = 0;
327 	int err;
328 
329 	for (i = 0; i < NR_RESULTS; i++) {
330 		err = bpf_map_lookup_elem(result_map, &i, &results[i]);
331 		RET_IF(err < 0, "lookup_elem(result_map)",
332 		       "i:%u err:%d errno:%d\n", i, err, errno);
333 	}
334 
335 	for (i = 0; i < NR_RESULTS; i++) {
336 		if (results[i] != expected_results[i]) {
337 			broken = i;
338 			break;
339 		}
340 	}
341 
342 	if (i == NR_RESULTS)
343 		return;
344 
345 	printf("unexpected result\n");
346 	printf(" result: [");
347 	printf("%u", results[0]);
348 	for (i = 1; i < NR_RESULTS; i++)
349 		printf(", %u", results[i]);
350 	printf("]\n");
351 
352 	printf("expected: [");
353 	printf("%u", expected_results[0]);
354 	for (i = 1; i < NR_RESULTS; i++)
355 		printf(", %u", expected_results[i]);
356 	printf("]\n");
357 
358 	printf("mismatch on %s (bpf_prog_linum:%ld)\n", result_to_str(broken),
359 	       get_linum());
360 
361 	CHECK_FAIL(true);
362 }
363 
send_data(int type,sa_family_t family,void * data,size_t len,enum result expected)364 static int send_data(int type, sa_family_t family, void *data, size_t len,
365 		     enum result expected)
366 {
367 	union sa46 cli_sa;
368 	int fd, err;
369 
370 	fd = socket(family, type, 0);
371 	RET_ERR(fd == -1, "socket()", "fd:%d errno:%d\n", fd, errno);
372 
373 	sa46_init_loopback(&cli_sa, family);
374 	err = bind(fd, (struct sockaddr *)&cli_sa, sizeof(cli_sa));
375 	RET_ERR(fd == -1, "bind(cli_sa)", "err:%d errno:%d\n", err, errno);
376 
377 	err = sendto(fd, data, len, MSG_FASTOPEN, (struct sockaddr *)&srv_sa,
378 		     sizeof(srv_sa));
379 	RET_ERR(err != len && expected >= PASS,
380 		"sendto()", "family:%u err:%d errno:%d expected:%d\n",
381 		family, err, errno, expected);
382 
383 	return fd;
384 }
385 
do_test(int type,sa_family_t family,struct cmd * cmd,enum result expected)386 static void do_test(int type, sa_family_t family, struct cmd *cmd,
387 		    enum result expected)
388 {
389 	int nev, srv_fd, cli_fd;
390 	struct epoll_event ev;
391 	struct cmd rcv_cmd;
392 	ssize_t nread;
393 
394 	cli_fd = send_data(type, family, cmd, cmd ? sizeof(*cmd) : 0,
395 			   expected);
396 	if (cli_fd < 0)
397 		return;
398 	nev = epoll_wait(epfd, &ev, 1, expected >= PASS ? 5 : 0);
399 	RET_IF((nev <= 0 && expected >= PASS) ||
400 	       (nev > 0 && expected < PASS),
401 	       "nev <> expected",
402 	       "nev:%d expected:%d type:%d family:%d data:(%d, %d)\n",
403 	       nev, expected, type, family,
404 	       cmd ? cmd->reuseport_index : -1,
405 	       cmd ? cmd->pass_on_failure : -1);
406 	check_results();
407 	check_data(type, family, cmd, cli_fd);
408 
409 	if (expected < PASS)
410 		return;
411 
412 	RET_IF(expected != PASS_ERR_SK_SELECT_REUSEPORT &&
413 	       cmd->reuseport_index != ev.data.u32,
414 	       "check cmd->reuseport_index",
415 	       "cmd:(%u, %u) ev.data.u32:%u\n",
416 	       cmd->pass_on_failure, cmd->reuseport_index, ev.data.u32);
417 
418 	srv_fd = sk_fds[ev.data.u32];
419 	if (type == SOCK_STREAM) {
420 		int new_fd = accept(srv_fd, NULL, 0);
421 
422 		RET_IF(new_fd == -1, "accept(srv_fd)",
423 		       "ev.data.u32:%u new_fd:%d errno:%d\n",
424 		       ev.data.u32, new_fd, errno);
425 
426 		nread = recv(new_fd, &rcv_cmd, sizeof(rcv_cmd), MSG_DONTWAIT);
427 		RET_IF(nread != sizeof(rcv_cmd),
428 		       "recv(new_fd)",
429 		       "ev.data.u32:%u nread:%zd sizeof(rcv_cmd):%zu errno:%d\n",
430 		       ev.data.u32, nread, sizeof(rcv_cmd), errno);
431 
432 		close(new_fd);
433 	} else {
434 		nread = recv(srv_fd, &rcv_cmd, sizeof(rcv_cmd), MSG_DONTWAIT);
435 		RET_IF(nread != sizeof(rcv_cmd),
436 		       "recv(sk_fds)",
437 		       "ev.data.u32:%u nread:%zd sizeof(rcv_cmd):%zu errno:%d\n",
438 		       ev.data.u32, nread, sizeof(rcv_cmd), errno);
439 	}
440 
441 	close(cli_fd);
442 }
443 
test_err_inner_map(int type,sa_family_t family)444 static void test_err_inner_map(int type, sa_family_t family)
445 {
446 	struct cmd cmd = {
447 		.reuseport_index = 0,
448 		.pass_on_failure = 0,
449 	};
450 
451 	expected_results[DROP_ERR_INNER_MAP]++;
452 	do_test(type, family, &cmd, DROP_ERR_INNER_MAP);
453 }
454 
test_err_skb_data(int type,sa_family_t family)455 static void test_err_skb_data(int type, sa_family_t family)
456 {
457 	expected_results[DROP_ERR_SKB_DATA]++;
458 	do_test(type, family, NULL, DROP_ERR_SKB_DATA);
459 }
460 
test_err_sk_select_port(int type,sa_family_t family)461 static void test_err_sk_select_port(int type, sa_family_t family)
462 {
463 	struct cmd cmd = {
464 		.reuseport_index = REUSEPORT_ARRAY_SIZE,
465 		.pass_on_failure = 0,
466 	};
467 
468 	expected_results[DROP_ERR_SK_SELECT_REUSEPORT]++;
469 	do_test(type, family, &cmd, DROP_ERR_SK_SELECT_REUSEPORT);
470 }
471 
test_pass(int type,sa_family_t family)472 static void test_pass(int type, sa_family_t family)
473 {
474 	struct cmd cmd;
475 	int i;
476 
477 	cmd.pass_on_failure = 0;
478 	for (i = 0; i < REUSEPORT_ARRAY_SIZE; i++) {
479 		expected_results[PASS]++;
480 		cmd.reuseport_index = i;
481 		do_test(type, family, &cmd, PASS);
482 	}
483 }
484 
test_syncookie(int type,sa_family_t family)485 static void test_syncookie(int type, sa_family_t family)
486 {
487 	int err, tmp_index = 1;
488 	struct cmd cmd = {
489 		.reuseport_index = 0,
490 		.pass_on_failure = 0,
491 	};
492 
493 	/*
494 	 * +1 for TCP-SYN and
495 	 * +1 for the TCP-ACK (ack the syncookie)
496 	 */
497 	expected_results[PASS] += 2;
498 	enable_syncookie();
499 	/*
500 	 * Simulate TCP-SYN and TCP-ACK are handled by two different sk:
501 	 * TCP-SYN: select sk_fds[tmp_index = 1] tmp_index is from the
502 	 *          tmp_index_ovr_map
503 	 * TCP-ACK: select sk_fds[reuseport_index = 0] reuseport_index
504 	 *          is from the cmd.reuseport_index
505 	 */
506 	err = bpf_map_update_elem(tmp_index_ovr_map, &index_zero,
507 				  &tmp_index, BPF_ANY);
508 	RET_IF(err < 0, "update_elem(tmp_index_ovr_map, 0, 1)",
509 	       "err:%d errno:%d\n", err, errno);
510 	do_test(type, family, &cmd, PASS);
511 	err = bpf_map_lookup_elem(tmp_index_ovr_map, &index_zero,
512 				  &tmp_index);
513 	RET_IF(err < 0 || tmp_index >= 0,
514 	       "lookup_elem(tmp_index_ovr_map)",
515 	       "err:%d errno:%d tmp_index:%d\n",
516 	       err, errno, tmp_index);
517 	disable_syncookie();
518 }
519 
test_pass_on_err(int type,sa_family_t family)520 static void test_pass_on_err(int type, sa_family_t family)
521 {
522 	struct cmd cmd = {
523 		.reuseport_index = REUSEPORT_ARRAY_SIZE,
524 		.pass_on_failure = 1,
525 	};
526 
527 	expected_results[PASS_ERR_SK_SELECT_REUSEPORT] += 1;
528 	do_test(type, family, &cmd, PASS_ERR_SK_SELECT_REUSEPORT);
529 }
530 
test_detach_bpf(int type,sa_family_t family)531 static void test_detach_bpf(int type, sa_family_t family)
532 {
533 #ifdef SO_DETACH_REUSEPORT_BPF
534 	__u32 nr_run_before = 0, nr_run_after = 0, tmp, i;
535 	struct epoll_event ev;
536 	int cli_fd, err, nev;
537 	struct cmd cmd = {};
538 	int optvalue = 0;
539 
540 	err = setsockopt(sk_fds[0], SOL_SOCKET, SO_DETACH_REUSEPORT_BPF,
541 			 &optvalue, sizeof(optvalue));
542 	RET_IF(err == -1, "setsockopt(SO_DETACH_REUSEPORT_BPF)",
543 	       "err:%d errno:%d\n", err, errno);
544 
545 	err = setsockopt(sk_fds[1], SOL_SOCKET, SO_DETACH_REUSEPORT_BPF,
546 			 &optvalue, sizeof(optvalue));
547 	RET_IF(err == 0 || errno != ENOENT,
548 	       "setsockopt(SO_DETACH_REUSEPORT_BPF)",
549 	       "err:%d errno:%d\n", err, errno);
550 
551 	for (i = 0; i < NR_RESULTS; i++) {
552 		err = bpf_map_lookup_elem(result_map, &i, &tmp);
553 		RET_IF(err < 0, "lookup_elem(result_map)",
554 		       "i:%u err:%d errno:%d\n", i, err, errno);
555 		nr_run_before += tmp;
556 	}
557 
558 	cli_fd = send_data(type, family, &cmd, sizeof(cmd), PASS);
559 	if (cli_fd < 0)
560 		return;
561 	nev = epoll_wait(epfd, &ev, 1, 5);
562 	RET_IF(nev <= 0, "nev <= 0",
563 	       "nev:%d expected:1 type:%d family:%d data:(0, 0)\n",
564 	       nev,  type, family);
565 
566 	for (i = 0; i < NR_RESULTS; i++) {
567 		err = bpf_map_lookup_elem(result_map, &i, &tmp);
568 		RET_IF(err < 0, "lookup_elem(result_map)",
569 		       "i:%u err:%d errno:%d\n", i, err, errno);
570 		nr_run_after += tmp;
571 	}
572 
573 	RET_IF(nr_run_before != nr_run_after,
574 	       "nr_run_before != nr_run_after",
575 	       "nr_run_before:%u nr_run_after:%u\n",
576 	       nr_run_before, nr_run_after);
577 
578 	close(cli_fd);
579 #else
580 	test__skip();
581 #endif
582 }
583 
prepare_sk_fds(int type,sa_family_t family,bool inany)584 static void prepare_sk_fds(int type, sa_family_t family, bool inany)
585 {
586 	const int first = REUSEPORT_ARRAY_SIZE - 1;
587 	int i, err, optval = 1;
588 	struct epoll_event ev;
589 	socklen_t addrlen;
590 
591 	if (inany)
592 		sa46_init_inany(&srv_sa, family);
593 	else
594 		sa46_init_loopback(&srv_sa, family);
595 	addrlen = sizeof(srv_sa);
596 
597 	/*
598 	 * The sk_fds[] is filled from the back such that the order
599 	 * is exactly opposite to the (struct sock_reuseport *)reuse->socks[].
600 	 */
601 	for (i = first; i >= 0; i--) {
602 		sk_fds[i] = socket(family, type, 0);
603 		RET_IF(sk_fds[i] == -1, "socket()", "sk_fds[%d]:%d errno:%d\n",
604 		       i, sk_fds[i], errno);
605 		err = setsockopt(sk_fds[i], SOL_SOCKET, SO_REUSEPORT,
606 				 &optval, sizeof(optval));
607 		RET_IF(err == -1, "setsockopt(SO_REUSEPORT)",
608 		       "sk_fds[%d] err:%d errno:%d\n",
609 		       i, err, errno);
610 
611 		if (i == first) {
612 			err = setsockopt(sk_fds[i], SOL_SOCKET,
613 					 SO_ATTACH_REUSEPORT_EBPF,
614 					 &select_by_skb_data_prog,
615 					 sizeof(select_by_skb_data_prog));
616 			RET_IF(err < 0, "setsockopt(SO_ATTACH_REUEPORT_EBPF)",
617 			       "err:%d errno:%d\n", err, errno);
618 		}
619 
620 		err = bind(sk_fds[i], (struct sockaddr *)&srv_sa, addrlen);
621 		RET_IF(err < 0, "bind()", "sk_fds[%d] err:%d errno:%d\n",
622 		       i, err, errno);
623 
624 		if (type == SOCK_STREAM) {
625 			err = listen(sk_fds[i], 10);
626 			RET_IF(err < 0, "listen()",
627 			       "sk_fds[%d] err:%d errno:%d\n",
628 			       i, err, errno);
629 		}
630 
631 		err = bpf_map_update_elem(reuseport_array, &i, &sk_fds[i],
632 					  BPF_NOEXIST);
633 		RET_IF(err < 0, "update_elem(reuseport_array)",
634 		       "sk_fds[%d] err:%d errno:%d\n", i, err, errno);
635 
636 		if (i == first) {
637 			socklen_t addrlen = sizeof(srv_sa);
638 
639 			err = getsockname(sk_fds[i], (struct sockaddr *)&srv_sa,
640 					  &addrlen);
641 			RET_IF(err == -1, "getsockname()",
642 			       "sk_fds[%d] err:%d errno:%d\n", i, err, errno);
643 		}
644 	}
645 
646 	epfd = epoll_create(1);
647 	RET_IF(epfd == -1, "epoll_create(1)",
648 	       "epfd:%d errno:%d\n", epfd, errno);
649 
650 	ev.events = EPOLLIN;
651 	for (i = 0; i < REUSEPORT_ARRAY_SIZE; i++) {
652 		ev.data.u32 = i;
653 		err = epoll_ctl(epfd, EPOLL_CTL_ADD, sk_fds[i], &ev);
654 		RET_IF(err, "epoll_ctl(EPOLL_CTL_ADD)", "sk_fds[%d]\n", i);
655 	}
656 }
657 
setup_per_test(int type,sa_family_t family,bool inany,bool no_inner_map)658 static void setup_per_test(int type, sa_family_t family, bool inany,
659 			   bool no_inner_map)
660 {
661 	int ovr = -1, err;
662 
663 	prepare_sk_fds(type, family, inany);
664 	err = bpf_map_update_elem(tmp_index_ovr_map, &index_zero, &ovr,
665 				  BPF_ANY);
666 	RET_IF(err < 0, "update_elem(tmp_index_ovr_map, 0, -1)",
667 	       "err:%d errno:%d\n", err, errno);
668 
669 	/* Install reuseport_array to outer_map? */
670 	if (no_inner_map)
671 		return;
672 
673 	err = bpf_map_update_elem(outer_map, &index_zero, &reuseport_array,
674 				  BPF_ANY);
675 	RET_IF(err < 0, "update_elem(outer_map, 0, reuseport_array)",
676 	       "err:%d errno:%d\n", err, errno);
677 }
678 
cleanup_per_test(bool no_inner_map)679 static void cleanup_per_test(bool no_inner_map)
680 {
681 	int i, err, zero = 0;
682 
683 	memset(expected_results, 0, sizeof(expected_results));
684 
685 	for (i = 0; i < NR_RESULTS; i++) {
686 		err = bpf_map_update_elem(result_map, &i, &zero, BPF_ANY);
687 		RET_IF(err, "reset elem in result_map",
688 		       "i:%u err:%d errno:%d\n", i, err, errno);
689 	}
690 
691 	err = bpf_map_update_elem(linum_map, &zero, &zero, BPF_ANY);
692 	RET_IF(err, "reset line number in linum_map", "err:%d errno:%d\n",
693 	       err, errno);
694 
695 	for (i = 0; i < REUSEPORT_ARRAY_SIZE; i++)
696 		close(sk_fds[i]);
697 	close(epfd);
698 
699 	/* Delete reuseport_array from outer_map? */
700 	if (no_inner_map)
701 		return;
702 
703 	err = bpf_map_delete_elem(outer_map, &index_zero);
704 	RET_IF(err < 0, "delete_elem(outer_map)",
705 	       "err:%d errno:%d\n", err, errno);
706 }
707 
cleanup(void)708 static void cleanup(void)
709 {
710 	if (outer_map >= 0) {
711 		close(outer_map);
712 		outer_map = -1;
713 	}
714 
715 	if (reuseport_array >= 0) {
716 		close(reuseport_array);
717 		reuseport_array = -1;
718 	}
719 
720 	if (obj) {
721 		bpf_object__close(obj);
722 		obj = NULL;
723 	}
724 
725 	memset(expected_results, 0, sizeof(expected_results));
726 }
727 
maptype_str(enum bpf_map_type type)728 static const char *maptype_str(enum bpf_map_type type)
729 {
730 	switch (type) {
731 	case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
732 		return "reuseport_sockarray";
733 	case BPF_MAP_TYPE_SOCKMAP:
734 		return "sockmap";
735 	case BPF_MAP_TYPE_SOCKHASH:
736 		return "sockhash";
737 	default:
738 		return "unknown";
739 	}
740 }
741 
family_str(sa_family_t family)742 static const char *family_str(sa_family_t family)
743 {
744 	switch (family) {
745 	case AF_INET:
746 		return "IPv4";
747 	case AF_INET6:
748 		return "IPv6";
749 	default:
750 		return "unknown";
751 	}
752 }
753 
sotype_str(int sotype)754 static const char *sotype_str(int sotype)
755 {
756 	switch (sotype) {
757 	case SOCK_STREAM:
758 		return "TCP";
759 	case SOCK_DGRAM:
760 		return "UDP";
761 	default:
762 		return "unknown";
763 	}
764 }
765 
766 #define TEST_INIT(fn_, ...) { .fn = fn_, .name = #fn_, __VA_ARGS__ }
767 
test_config(int sotype,sa_family_t family,bool inany)768 static void test_config(int sotype, sa_family_t family, bool inany)
769 {
770 	const struct test {
771 		void (*fn)(int sotype, sa_family_t family);
772 		const char *name;
773 		bool no_inner_map;
774 		int need_sotype;
775 	} tests[] = {
776 		TEST_INIT(test_err_inner_map,
777 			  .no_inner_map = true),
778 		TEST_INIT(test_err_skb_data),
779 		TEST_INIT(test_err_sk_select_port),
780 		TEST_INIT(test_pass),
781 		TEST_INIT(test_syncookie,
782 			  .need_sotype = SOCK_STREAM),
783 		TEST_INIT(test_pass_on_err),
784 		TEST_INIT(test_detach_bpf),
785 	};
786 	struct netns_obj *netns;
787 	char s[MAX_TEST_NAME];
788 	const struct test *t;
789 
790 	for (t = tests; t < tests + ARRAY_SIZE(tests); t++) {
791 		if (t->need_sotype && t->need_sotype != sotype)
792 			continue; /* test not compatible with socket type */
793 
794 		snprintf(s, sizeof(s), "%s %s/%s %s %s",
795 			 maptype_str(inner_map_type),
796 			 family_str(family), sotype_str(sotype),
797 			 inany ? "INANY" : "LOOPBACK", t->name);
798 
799 		if (!test__start_subtest(s))
800 			continue;
801 
802 		netns = netns_new("select_reuseport", true);
803 		if (!ASSERT_OK_PTR(netns, "netns_new"))
804 			continue;
805 
806 		if (CHECK_FAIL(enable_fastopen()))
807 			goto out;
808 		if (CHECK_FAIL(disable_syncookie()))
809 			goto out;
810 
811 		setup_per_test(sotype, family, inany, t->no_inner_map);
812 		t->fn(sotype, family);
813 		cleanup_per_test(t->no_inner_map);
814 
815 out:
816 		netns_free(netns);
817 	}
818 }
819 
820 #define BIND_INANY true
821 
test_all(void)822 static void test_all(void)
823 {
824 	const struct config {
825 		int sotype;
826 		sa_family_t family;
827 		bool inany;
828 	} configs[] = {
829 		{ SOCK_STREAM, AF_INET },
830 		{ SOCK_STREAM, AF_INET, BIND_INANY },
831 		{ SOCK_STREAM, AF_INET6 },
832 		{ SOCK_STREAM, AF_INET6, BIND_INANY },
833 		{ SOCK_DGRAM, AF_INET },
834 		{ SOCK_DGRAM, AF_INET6 },
835 	};
836 	const struct config *c;
837 
838 	for (c = configs; c < configs + ARRAY_SIZE(configs); c++)
839 		test_config(c->sotype, c->family, c->inany);
840 }
841 
test_map_type(enum bpf_map_type mt)842 void test_map_type(enum bpf_map_type mt)
843 {
844 	if (create_maps(mt))
845 		goto out;
846 	if (prepare_bpf_obj())
847 		goto out;
848 
849 	test_all();
850 out:
851 	cleanup();
852 }
853 
serial_test_select_reuseport(void)854 void serial_test_select_reuseport(void)
855 {
856 	test_map_type(BPF_MAP_TYPE_REUSEPORT_SOCKARRAY);
857 	test_map_type(BPF_MAP_TYPE_SOCKMAP);
858 	test_map_type(BPF_MAP_TYPE_SOCKHASH);
859 }
860