Loading...
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
34static int result_map, tmp_index_ovr_map, linum_map, data_check_map;
35static __u32 expected_results[NR_RESULTS];
36static int sk_fds[REUSEPORT_ARRAY_SIZE];
37static int reuseport_array = -1, outer_map = -1;
38static enum bpf_map_type inner_map_type;
39static int select_by_skb_data_prog;
40static struct bpf_object *obj;
41static __u32 index_zero;
42static int epfd;
43
44static 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
64static 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
86static 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
138static 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
148static 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
158static 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
175static 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
194static 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
205static int enable_syncookie(void)
206{
207 return write_int_sysctl(TCP_SYNCOOKIE_SYSCTL, 2);
208}
209
210static int disable_syncookie(void)
211{
212 return write_int_sysctl(TCP_SYNCOOKIE_SYSCTL, 0);
213}
214
215static 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
227static 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
303static 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
323static 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
364static 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
386static 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
444static 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
455static 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
461static 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
472static 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
485static 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
520static 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
531static 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
584static 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
658static 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
679static 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
708static 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
728static 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
742static 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
754static 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
768static 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
815out:
816 netns_free(netns);
817 }
818}
819
820#define BIND_INANY true
821
822static 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
842void 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();
850out:
851 cleanup();
852}
853
854void 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}
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_rlimit.h"
22#include "bpf_util.h"
23
24#include "test_progs.h"
25#include "test_select_reuseport_common.h"
26
27#define MAX_TEST_NAME 80
28#define MIN_TCPHDR_LEN 20
29#define UDPHDR_LEN 8
30
31#define TCP_SYNCOOKIE_SYSCTL "/proc/sys/net/ipv4/tcp_syncookies"
32#define TCP_FO_SYSCTL "/proc/sys/net/ipv4/tcp_fastopen"
33#define REUSEPORT_ARRAY_SIZE 32
34
35static int result_map, tmp_index_ovr_map, linum_map, data_check_map;
36static __u32 expected_results[NR_RESULTS];
37static int sk_fds[REUSEPORT_ARRAY_SIZE];
38static int reuseport_array = -1, outer_map = -1;
39static enum bpf_map_type inner_map_type;
40static int select_by_skb_data_prog;
41static int saved_tcp_syncookie = -1;
42static struct bpf_object *obj;
43static int saved_tcp_fo = -1;
44static __u32 index_zero;
45static int epfd;
46
47static union sa46 {
48 struct sockaddr_in6 v6;
49 struct sockaddr_in v4;
50 sa_family_t family;
51} srv_sa;
52
53#define RET_IF(condition, tag, format...) ({ \
54 if (CHECK_FAIL(condition)) { \
55 printf(tag " " format); \
56 return; \
57 } \
58})
59
60#define RET_ERR(condition, tag, format...) ({ \
61 if (CHECK_FAIL(condition)) { \
62 printf(tag " " format); \
63 return -1; \
64 } \
65})
66
67static int create_maps(enum bpf_map_type inner_type)
68{
69 struct bpf_create_map_attr attr = {};
70
71 inner_map_type = inner_type;
72
73 /* Creating reuseport_array */
74 attr.name = "reuseport_array";
75 attr.map_type = inner_type;
76 attr.key_size = sizeof(__u32);
77 attr.value_size = sizeof(__u32);
78 attr.max_entries = REUSEPORT_ARRAY_SIZE;
79
80 reuseport_array = bpf_create_map_xattr(&attr);
81 RET_ERR(reuseport_array == -1, "creating reuseport_array",
82 "reuseport_array:%d errno:%d\n", reuseport_array, errno);
83
84 /* Creating outer_map */
85 attr.name = "outer_map";
86 attr.map_type = BPF_MAP_TYPE_ARRAY_OF_MAPS;
87 attr.key_size = sizeof(__u32);
88 attr.value_size = sizeof(__u32);
89 attr.max_entries = 1;
90 attr.inner_map_fd = reuseport_array;
91 outer_map = bpf_create_map_xattr(&attr);
92 RET_ERR(outer_map == -1, "creating outer_map",
93 "outer_map:%d errno:%d\n", outer_map, errno);
94
95 return 0;
96}
97
98static int prepare_bpf_obj(void)
99{
100 struct bpf_program *prog;
101 struct bpf_map *map;
102 int err;
103
104 obj = bpf_object__open("test_select_reuseport_kern.o");
105 RET_ERR(IS_ERR_OR_NULL(obj), "open test_select_reuseport_kern.o",
106 "obj:%p PTR_ERR(obj):%ld\n", obj, PTR_ERR(obj));
107
108 map = bpf_object__find_map_by_name(obj, "outer_map");
109 RET_ERR(!map, "find outer_map", "!map\n");
110 err = bpf_map__reuse_fd(map, outer_map);
111 RET_ERR(err, "reuse outer_map", "err:%d\n", err);
112
113 err = bpf_object__load(obj);
114 RET_ERR(err, "load bpf_object", "err:%d\n", err);
115
116 prog = bpf_program__next(NULL, obj);
117 RET_ERR(!prog, "get first bpf_program", "!prog\n");
118 select_by_skb_data_prog = bpf_program__fd(prog);
119 RET_ERR(select_by_skb_data_prog == -1, "get prog fd",
120 "select_by_skb_data_prog:%d\n", select_by_skb_data_prog);
121
122 map = bpf_object__find_map_by_name(obj, "result_map");
123 RET_ERR(!map, "find result_map", "!map\n");
124 result_map = bpf_map__fd(map);
125 RET_ERR(result_map == -1, "get result_map fd",
126 "result_map:%d\n", result_map);
127
128 map = bpf_object__find_map_by_name(obj, "tmp_index_ovr_map");
129 RET_ERR(!map, "find tmp_index_ovr_map\n", "!map");
130 tmp_index_ovr_map = bpf_map__fd(map);
131 RET_ERR(tmp_index_ovr_map == -1, "get tmp_index_ovr_map fd",
132 "tmp_index_ovr_map:%d\n", tmp_index_ovr_map);
133
134 map = bpf_object__find_map_by_name(obj, "linum_map");
135 RET_ERR(!map, "find linum_map", "!map\n");
136 linum_map = bpf_map__fd(map);
137 RET_ERR(linum_map == -1, "get linum_map fd",
138 "linum_map:%d\n", linum_map);
139
140 map = bpf_object__find_map_by_name(obj, "data_check_map");
141 RET_ERR(!map, "find data_check_map", "!map\n");
142 data_check_map = bpf_map__fd(map);
143 RET_ERR(data_check_map == -1, "get data_check_map fd",
144 "data_check_map:%d\n", data_check_map);
145
146 return 0;
147}
148
149static void sa46_init_loopback(union sa46 *sa, sa_family_t family)
150{
151 memset(sa, 0, sizeof(*sa));
152 sa->family = family;
153 if (sa->family == AF_INET6)
154 sa->v6.sin6_addr = in6addr_loopback;
155 else
156 sa->v4.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
157}
158
159static void sa46_init_inany(union sa46 *sa, sa_family_t family)
160{
161 memset(sa, 0, sizeof(*sa));
162 sa->family = family;
163 if (sa->family == AF_INET6)
164 sa->v6.sin6_addr = in6addr_any;
165 else
166 sa->v4.sin_addr.s_addr = INADDR_ANY;
167}
168
169static int read_int_sysctl(const char *sysctl)
170{
171 char buf[16];
172 int fd, ret;
173
174 fd = open(sysctl, 0);
175 RET_ERR(fd == -1, "open(sysctl)",
176 "sysctl:%s fd:%d errno:%d\n", sysctl, fd, errno);
177
178 ret = read(fd, buf, sizeof(buf));
179 RET_ERR(ret <= 0, "read(sysctl)",
180 "sysctl:%s ret:%d errno:%d\n", sysctl, ret, errno);
181
182 close(fd);
183 return atoi(buf);
184}
185
186static int write_int_sysctl(const char *sysctl, int v)
187{
188 int fd, ret, size;
189 char buf[16];
190
191 fd = open(sysctl, O_RDWR);
192 RET_ERR(fd == -1, "open(sysctl)",
193 "sysctl:%s fd:%d errno:%d\n", sysctl, fd, errno);
194
195 size = snprintf(buf, sizeof(buf), "%d", v);
196 ret = write(fd, buf, size);
197 RET_ERR(ret != size, "write(sysctl)",
198 "sysctl:%s ret:%d size:%d errno:%d\n",
199 sysctl, ret, size, errno);
200
201 close(fd);
202 return 0;
203}
204
205static void restore_sysctls(void)
206{
207 if (saved_tcp_fo != -1)
208 write_int_sysctl(TCP_FO_SYSCTL, saved_tcp_fo);
209 if (saved_tcp_syncookie != -1)
210 write_int_sysctl(TCP_SYNCOOKIE_SYSCTL, saved_tcp_syncookie);
211}
212
213static int enable_fastopen(void)
214{
215 int fo;
216
217 fo = read_int_sysctl(TCP_FO_SYSCTL);
218 if (fo < 0)
219 return -1;
220
221 return write_int_sysctl(TCP_FO_SYSCTL, fo | 7);
222}
223
224static int enable_syncookie(void)
225{
226 return write_int_sysctl(TCP_SYNCOOKIE_SYSCTL, 2);
227}
228
229static int disable_syncookie(void)
230{
231 return write_int_sysctl(TCP_SYNCOOKIE_SYSCTL, 0);
232}
233
234static long get_linum(void)
235{
236 __u32 linum;
237 int err;
238
239 err = bpf_map_lookup_elem(linum_map, &index_zero, &linum);
240 RET_ERR(err == -1, "lookup_elem(linum_map)", "err:%d errno:%d\n",
241 err, errno);
242
243 return linum;
244}
245
246static void check_data(int type, sa_family_t family, const struct cmd *cmd,
247 int cli_fd)
248{
249 struct data_check expected = {}, result;
250 union sa46 cli_sa;
251 socklen_t addrlen;
252 int err;
253
254 addrlen = sizeof(cli_sa);
255 err = getsockname(cli_fd, (struct sockaddr *)&cli_sa,
256 &addrlen);
257 RET_IF(err == -1, "getsockname(cli_fd)", "err:%d errno:%d\n",
258 err, errno);
259
260 err = bpf_map_lookup_elem(data_check_map, &index_zero, &result);
261 RET_IF(err == -1, "lookup_elem(data_check_map)", "err:%d errno:%d\n",
262 err, errno);
263
264 if (type == SOCK_STREAM) {
265 expected.len = MIN_TCPHDR_LEN;
266 expected.ip_protocol = IPPROTO_TCP;
267 } else {
268 expected.len = UDPHDR_LEN;
269 expected.ip_protocol = IPPROTO_UDP;
270 }
271
272 if (family == AF_INET6) {
273 expected.eth_protocol = htons(ETH_P_IPV6);
274 expected.bind_inany = !srv_sa.v6.sin6_addr.s6_addr32[3] &&
275 !srv_sa.v6.sin6_addr.s6_addr32[2] &&
276 !srv_sa.v6.sin6_addr.s6_addr32[1] &&
277 !srv_sa.v6.sin6_addr.s6_addr32[0];
278
279 memcpy(&expected.skb_addrs[0], cli_sa.v6.sin6_addr.s6_addr32,
280 sizeof(cli_sa.v6.sin6_addr));
281 memcpy(&expected.skb_addrs[4], &in6addr_loopback,
282 sizeof(in6addr_loopback));
283 expected.skb_ports[0] = cli_sa.v6.sin6_port;
284 expected.skb_ports[1] = srv_sa.v6.sin6_port;
285 } else {
286 expected.eth_protocol = htons(ETH_P_IP);
287 expected.bind_inany = !srv_sa.v4.sin_addr.s_addr;
288
289 expected.skb_addrs[0] = cli_sa.v4.sin_addr.s_addr;
290 expected.skb_addrs[1] = htonl(INADDR_LOOPBACK);
291 expected.skb_ports[0] = cli_sa.v4.sin_port;
292 expected.skb_ports[1] = srv_sa.v4.sin_port;
293 }
294
295 if (memcmp(&result, &expected, offsetof(struct data_check,
296 equal_check_end))) {
297 printf("unexpected data_check\n");
298 printf(" result: (0x%x, %u, %u)\n",
299 result.eth_protocol, result.ip_protocol,
300 result.bind_inany);
301 printf("expected: (0x%x, %u, %u)\n",
302 expected.eth_protocol, expected.ip_protocol,
303 expected.bind_inany);
304 RET_IF(1, "data_check result != expected",
305 "bpf_prog_linum:%ld\n", get_linum());
306 }
307
308 RET_IF(!result.hash, "data_check result.hash empty",
309 "result.hash:%u", result.hash);
310
311 expected.len += cmd ? sizeof(*cmd) : 0;
312 if (type == SOCK_STREAM)
313 RET_IF(expected.len > result.len, "expected.len > result.len",
314 "expected.len:%u result.len:%u bpf_prog_linum:%ld\n",
315 expected.len, result.len, get_linum());
316 else
317 RET_IF(expected.len != result.len, "expected.len != result.len",
318 "expected.len:%u result.len:%u bpf_prog_linum:%ld\n",
319 expected.len, result.len, get_linum());
320}
321
322static const char *result_to_str(enum result res)
323{
324 switch (res) {
325 case DROP_ERR_INNER_MAP:
326 return "DROP_ERR_INNER_MAP";
327 case DROP_ERR_SKB_DATA:
328 return "DROP_ERR_SKB_DATA";
329 case DROP_ERR_SK_SELECT_REUSEPORT:
330 return "DROP_ERR_SK_SELECT_REUSEPORT";
331 case DROP_MISC:
332 return "DROP_MISC";
333 case PASS:
334 return "PASS";
335 case PASS_ERR_SK_SELECT_REUSEPORT:
336 return "PASS_ERR_SK_SELECT_REUSEPORT";
337 default:
338 return "UNKNOWN";
339 }
340}
341
342static void check_results(void)
343{
344 __u32 results[NR_RESULTS];
345 __u32 i, broken = 0;
346 int err;
347
348 for (i = 0; i < NR_RESULTS; i++) {
349 err = bpf_map_lookup_elem(result_map, &i, &results[i]);
350 RET_IF(err == -1, "lookup_elem(result_map)",
351 "i:%u err:%d errno:%d\n", i, err, errno);
352 }
353
354 for (i = 0; i < NR_RESULTS; i++) {
355 if (results[i] != expected_results[i]) {
356 broken = i;
357 break;
358 }
359 }
360
361 if (i == NR_RESULTS)
362 return;
363
364 printf("unexpected result\n");
365 printf(" result: [");
366 printf("%u", results[0]);
367 for (i = 1; i < NR_RESULTS; i++)
368 printf(", %u", results[i]);
369 printf("]\n");
370
371 printf("expected: [");
372 printf("%u", expected_results[0]);
373 for (i = 1; i < NR_RESULTS; i++)
374 printf(", %u", expected_results[i]);
375 printf("]\n");
376
377 printf("mismatch on %s (bpf_prog_linum:%ld)\n", result_to_str(broken),
378 get_linum());
379
380 CHECK_FAIL(true);
381}
382
383static int send_data(int type, sa_family_t family, void *data, size_t len,
384 enum result expected)
385{
386 union sa46 cli_sa;
387 int fd, err;
388
389 fd = socket(family, type, 0);
390 RET_ERR(fd == -1, "socket()", "fd:%d errno:%d\n", fd, errno);
391
392 sa46_init_loopback(&cli_sa, family);
393 err = bind(fd, (struct sockaddr *)&cli_sa, sizeof(cli_sa));
394 RET_ERR(fd == -1, "bind(cli_sa)", "err:%d errno:%d\n", err, errno);
395
396 err = sendto(fd, data, len, MSG_FASTOPEN, (struct sockaddr *)&srv_sa,
397 sizeof(srv_sa));
398 RET_ERR(err != len && expected >= PASS,
399 "sendto()", "family:%u err:%d errno:%d expected:%d\n",
400 family, err, errno, expected);
401
402 return fd;
403}
404
405static void do_test(int type, sa_family_t family, struct cmd *cmd,
406 enum result expected)
407{
408 int nev, srv_fd, cli_fd;
409 struct epoll_event ev;
410 struct cmd rcv_cmd;
411 ssize_t nread;
412
413 cli_fd = send_data(type, family, cmd, cmd ? sizeof(*cmd) : 0,
414 expected);
415 if (cli_fd < 0)
416 return;
417 nev = epoll_wait(epfd, &ev, 1, expected >= PASS ? 5 : 0);
418 RET_IF((nev <= 0 && expected >= PASS) ||
419 (nev > 0 && expected < PASS),
420 "nev <> expected",
421 "nev:%d expected:%d type:%d family:%d data:(%d, %d)\n",
422 nev, expected, type, family,
423 cmd ? cmd->reuseport_index : -1,
424 cmd ? cmd->pass_on_failure : -1);
425 check_results();
426 check_data(type, family, cmd, cli_fd);
427
428 if (expected < PASS)
429 return;
430
431 RET_IF(expected != PASS_ERR_SK_SELECT_REUSEPORT &&
432 cmd->reuseport_index != ev.data.u32,
433 "check cmd->reuseport_index",
434 "cmd:(%u, %u) ev.data.u32:%u\n",
435 cmd->pass_on_failure, cmd->reuseport_index, ev.data.u32);
436
437 srv_fd = sk_fds[ev.data.u32];
438 if (type == SOCK_STREAM) {
439 int new_fd = accept(srv_fd, NULL, 0);
440
441 RET_IF(new_fd == -1, "accept(srv_fd)",
442 "ev.data.u32:%u new_fd:%d errno:%d\n",
443 ev.data.u32, new_fd, errno);
444
445 nread = recv(new_fd, &rcv_cmd, sizeof(rcv_cmd), MSG_DONTWAIT);
446 RET_IF(nread != sizeof(rcv_cmd),
447 "recv(new_fd)",
448 "ev.data.u32:%u nread:%zd sizeof(rcv_cmd):%zu errno:%d\n",
449 ev.data.u32, nread, sizeof(rcv_cmd), errno);
450
451 close(new_fd);
452 } else {
453 nread = recv(srv_fd, &rcv_cmd, sizeof(rcv_cmd), MSG_DONTWAIT);
454 RET_IF(nread != sizeof(rcv_cmd),
455 "recv(sk_fds)",
456 "ev.data.u32:%u nread:%zd sizeof(rcv_cmd):%zu errno:%d\n",
457 ev.data.u32, nread, sizeof(rcv_cmd), errno);
458 }
459
460 close(cli_fd);
461}
462
463static void test_err_inner_map(int type, sa_family_t family)
464{
465 struct cmd cmd = {
466 .reuseport_index = 0,
467 .pass_on_failure = 0,
468 };
469
470 expected_results[DROP_ERR_INNER_MAP]++;
471 do_test(type, family, &cmd, DROP_ERR_INNER_MAP);
472}
473
474static void test_err_skb_data(int type, sa_family_t family)
475{
476 expected_results[DROP_ERR_SKB_DATA]++;
477 do_test(type, family, NULL, DROP_ERR_SKB_DATA);
478}
479
480static void test_err_sk_select_port(int type, sa_family_t family)
481{
482 struct cmd cmd = {
483 .reuseport_index = REUSEPORT_ARRAY_SIZE,
484 .pass_on_failure = 0,
485 };
486
487 expected_results[DROP_ERR_SK_SELECT_REUSEPORT]++;
488 do_test(type, family, &cmd, DROP_ERR_SK_SELECT_REUSEPORT);
489}
490
491static void test_pass(int type, sa_family_t family)
492{
493 struct cmd cmd;
494 int i;
495
496 cmd.pass_on_failure = 0;
497 for (i = 0; i < REUSEPORT_ARRAY_SIZE; i++) {
498 expected_results[PASS]++;
499 cmd.reuseport_index = i;
500 do_test(type, family, &cmd, PASS);
501 }
502}
503
504static void test_syncookie(int type, sa_family_t family)
505{
506 int err, tmp_index = 1;
507 struct cmd cmd = {
508 .reuseport_index = 0,
509 .pass_on_failure = 0,
510 };
511
512 /*
513 * +1 for TCP-SYN and
514 * +1 for the TCP-ACK (ack the syncookie)
515 */
516 expected_results[PASS] += 2;
517 enable_syncookie();
518 /*
519 * Simulate TCP-SYN and TCP-ACK are handled by two different sk:
520 * TCP-SYN: select sk_fds[tmp_index = 1] tmp_index is from the
521 * tmp_index_ovr_map
522 * TCP-ACK: select sk_fds[reuseport_index = 0] reuseport_index
523 * is from the cmd.reuseport_index
524 */
525 err = bpf_map_update_elem(tmp_index_ovr_map, &index_zero,
526 &tmp_index, BPF_ANY);
527 RET_IF(err == -1, "update_elem(tmp_index_ovr_map, 0, 1)",
528 "err:%d errno:%d\n", err, errno);
529 do_test(type, family, &cmd, PASS);
530 err = bpf_map_lookup_elem(tmp_index_ovr_map, &index_zero,
531 &tmp_index);
532 RET_IF(err == -1 || tmp_index != -1,
533 "lookup_elem(tmp_index_ovr_map)",
534 "err:%d errno:%d tmp_index:%d\n",
535 err, errno, tmp_index);
536 disable_syncookie();
537}
538
539static void test_pass_on_err(int type, sa_family_t family)
540{
541 struct cmd cmd = {
542 .reuseport_index = REUSEPORT_ARRAY_SIZE,
543 .pass_on_failure = 1,
544 };
545
546 expected_results[PASS_ERR_SK_SELECT_REUSEPORT] += 1;
547 do_test(type, family, &cmd, PASS_ERR_SK_SELECT_REUSEPORT);
548}
549
550static void test_detach_bpf(int type, sa_family_t family)
551{
552#ifdef SO_DETACH_REUSEPORT_BPF
553 __u32 nr_run_before = 0, nr_run_after = 0, tmp, i;
554 struct epoll_event ev;
555 int cli_fd, err, nev;
556 struct cmd cmd = {};
557 int optvalue = 0;
558
559 err = setsockopt(sk_fds[0], SOL_SOCKET, SO_DETACH_REUSEPORT_BPF,
560 &optvalue, sizeof(optvalue));
561 RET_IF(err == -1, "setsockopt(SO_DETACH_REUSEPORT_BPF)",
562 "err:%d errno:%d\n", err, errno);
563
564 err = setsockopt(sk_fds[1], SOL_SOCKET, SO_DETACH_REUSEPORT_BPF,
565 &optvalue, sizeof(optvalue));
566 RET_IF(err == 0 || errno != ENOENT,
567 "setsockopt(SO_DETACH_REUSEPORT_BPF)",
568 "err:%d errno:%d\n", err, errno);
569
570 for (i = 0; i < NR_RESULTS; i++) {
571 err = bpf_map_lookup_elem(result_map, &i, &tmp);
572 RET_IF(err == -1, "lookup_elem(result_map)",
573 "i:%u err:%d errno:%d\n", i, err, errno);
574 nr_run_before += tmp;
575 }
576
577 cli_fd = send_data(type, family, &cmd, sizeof(cmd), PASS);
578 if (cli_fd < 0)
579 return;
580 nev = epoll_wait(epfd, &ev, 1, 5);
581 RET_IF(nev <= 0, "nev <= 0",
582 "nev:%d expected:1 type:%d family:%d data:(0, 0)\n",
583 nev, type, family);
584
585 for (i = 0; i < NR_RESULTS; i++) {
586 err = bpf_map_lookup_elem(result_map, &i, &tmp);
587 RET_IF(err == -1, "lookup_elem(result_map)",
588 "i:%u err:%d errno:%d\n", i, err, errno);
589 nr_run_after += tmp;
590 }
591
592 RET_IF(nr_run_before != nr_run_after,
593 "nr_run_before != nr_run_after",
594 "nr_run_before:%u nr_run_after:%u\n",
595 nr_run_before, nr_run_after);
596
597 close(cli_fd);
598#else
599 test__skip();
600#endif
601}
602
603static void prepare_sk_fds(int type, sa_family_t family, bool inany)
604{
605 const int first = REUSEPORT_ARRAY_SIZE - 1;
606 int i, err, optval = 1;
607 struct epoll_event ev;
608 socklen_t addrlen;
609
610 if (inany)
611 sa46_init_inany(&srv_sa, family);
612 else
613 sa46_init_loopback(&srv_sa, family);
614 addrlen = sizeof(srv_sa);
615
616 /*
617 * The sk_fds[] is filled from the back such that the order
618 * is exactly opposite to the (struct sock_reuseport *)reuse->socks[].
619 */
620 for (i = first; i >= 0; i--) {
621 sk_fds[i] = socket(family, type, 0);
622 RET_IF(sk_fds[i] == -1, "socket()", "sk_fds[%d]:%d errno:%d\n",
623 i, sk_fds[i], errno);
624 err = setsockopt(sk_fds[i], SOL_SOCKET, SO_REUSEPORT,
625 &optval, sizeof(optval));
626 RET_IF(err == -1, "setsockopt(SO_REUSEPORT)",
627 "sk_fds[%d] err:%d errno:%d\n",
628 i, err, errno);
629
630 if (i == first) {
631 err = setsockopt(sk_fds[i], SOL_SOCKET,
632 SO_ATTACH_REUSEPORT_EBPF,
633 &select_by_skb_data_prog,
634 sizeof(select_by_skb_data_prog));
635 RET_IF(err == -1, "setsockopt(SO_ATTACH_REUEPORT_EBPF)",
636 "err:%d errno:%d\n", err, errno);
637 }
638
639 err = bind(sk_fds[i], (struct sockaddr *)&srv_sa, addrlen);
640 RET_IF(err == -1, "bind()", "sk_fds[%d] err:%d errno:%d\n",
641 i, err, errno);
642
643 if (type == SOCK_STREAM) {
644 err = listen(sk_fds[i], 10);
645 RET_IF(err == -1, "listen()",
646 "sk_fds[%d] err:%d errno:%d\n",
647 i, err, errno);
648 }
649
650 err = bpf_map_update_elem(reuseport_array, &i, &sk_fds[i],
651 BPF_NOEXIST);
652 RET_IF(err == -1, "update_elem(reuseport_array)",
653 "sk_fds[%d] err:%d errno:%d\n", i, err, errno);
654
655 if (i == first) {
656 socklen_t addrlen = sizeof(srv_sa);
657
658 err = getsockname(sk_fds[i], (struct sockaddr *)&srv_sa,
659 &addrlen);
660 RET_IF(err == -1, "getsockname()",
661 "sk_fds[%d] err:%d errno:%d\n", i, err, errno);
662 }
663 }
664
665 epfd = epoll_create(1);
666 RET_IF(epfd == -1, "epoll_create(1)",
667 "epfd:%d errno:%d\n", epfd, errno);
668
669 ev.events = EPOLLIN;
670 for (i = 0; i < REUSEPORT_ARRAY_SIZE; i++) {
671 ev.data.u32 = i;
672 err = epoll_ctl(epfd, EPOLL_CTL_ADD, sk_fds[i], &ev);
673 RET_IF(err, "epoll_ctl(EPOLL_CTL_ADD)", "sk_fds[%d]\n", i);
674 }
675}
676
677static void setup_per_test(int type, sa_family_t family, bool inany,
678 bool no_inner_map)
679{
680 int ovr = -1, err;
681
682 prepare_sk_fds(type, family, inany);
683 err = bpf_map_update_elem(tmp_index_ovr_map, &index_zero, &ovr,
684 BPF_ANY);
685 RET_IF(err == -1, "update_elem(tmp_index_ovr_map, 0, -1)",
686 "err:%d errno:%d\n", err, errno);
687
688 /* Install reuseport_array to outer_map? */
689 if (no_inner_map)
690 return;
691
692 err = bpf_map_update_elem(outer_map, &index_zero, &reuseport_array,
693 BPF_ANY);
694 RET_IF(err == -1, "update_elem(outer_map, 0, reuseport_array)",
695 "err:%d errno:%d\n", err, errno);
696}
697
698static void cleanup_per_test(bool no_inner_map)
699{
700 int i, err, zero = 0;
701
702 memset(expected_results, 0, sizeof(expected_results));
703
704 for (i = 0; i < NR_RESULTS; i++) {
705 err = bpf_map_update_elem(result_map, &i, &zero, BPF_ANY);
706 RET_IF(err, "reset elem in result_map",
707 "i:%u err:%d errno:%d\n", i, err, errno);
708 }
709
710 err = bpf_map_update_elem(linum_map, &zero, &zero, BPF_ANY);
711 RET_IF(err, "reset line number in linum_map", "err:%d errno:%d\n",
712 err, errno);
713
714 for (i = 0; i < REUSEPORT_ARRAY_SIZE; i++)
715 close(sk_fds[i]);
716 close(epfd);
717
718 /* Delete reuseport_array from outer_map? */
719 if (no_inner_map)
720 return;
721
722 err = bpf_map_delete_elem(outer_map, &index_zero);
723 RET_IF(err == -1, "delete_elem(outer_map)",
724 "err:%d errno:%d\n", err, errno);
725}
726
727static void cleanup(void)
728{
729 if (outer_map != -1) {
730 close(outer_map);
731 outer_map = -1;
732 }
733
734 if (reuseport_array != -1) {
735 close(reuseport_array);
736 reuseport_array = -1;
737 }
738
739 if (obj) {
740 bpf_object__close(obj);
741 obj = NULL;
742 }
743
744 memset(expected_results, 0, sizeof(expected_results));
745}
746
747static const char *maptype_str(enum bpf_map_type type)
748{
749 switch (type) {
750 case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
751 return "reuseport_sockarray";
752 case BPF_MAP_TYPE_SOCKMAP:
753 return "sockmap";
754 case BPF_MAP_TYPE_SOCKHASH:
755 return "sockhash";
756 default:
757 return "unknown";
758 }
759}
760
761static const char *family_str(sa_family_t family)
762{
763 switch (family) {
764 case AF_INET:
765 return "IPv4";
766 case AF_INET6:
767 return "IPv6";
768 default:
769 return "unknown";
770 }
771}
772
773static const char *sotype_str(int sotype)
774{
775 switch (sotype) {
776 case SOCK_STREAM:
777 return "TCP";
778 case SOCK_DGRAM:
779 return "UDP";
780 default:
781 return "unknown";
782 }
783}
784
785#define TEST_INIT(fn_, ...) { .fn = fn_, .name = #fn_, __VA_ARGS__ }
786
787static void test_config(int sotype, sa_family_t family, bool inany)
788{
789 const struct test {
790 void (*fn)(int sotype, sa_family_t family);
791 const char *name;
792 bool no_inner_map;
793 int need_sotype;
794 } tests[] = {
795 TEST_INIT(test_err_inner_map,
796 .no_inner_map = true),
797 TEST_INIT(test_err_skb_data),
798 TEST_INIT(test_err_sk_select_port),
799 TEST_INIT(test_pass),
800 TEST_INIT(test_syncookie,
801 .need_sotype = SOCK_STREAM),
802 TEST_INIT(test_pass_on_err),
803 TEST_INIT(test_detach_bpf),
804 };
805 char s[MAX_TEST_NAME];
806 const struct test *t;
807
808 for (t = tests; t < tests + ARRAY_SIZE(tests); t++) {
809 if (t->need_sotype && t->need_sotype != sotype)
810 continue; /* test not compatible with socket type */
811
812 snprintf(s, sizeof(s), "%s %s/%s %s %s",
813 maptype_str(inner_map_type),
814 family_str(family), sotype_str(sotype),
815 inany ? "INANY" : "LOOPBACK", t->name);
816
817 if (!test__start_subtest(s))
818 continue;
819
820 setup_per_test(sotype, family, inany, t->no_inner_map);
821 t->fn(sotype, family);
822 cleanup_per_test(t->no_inner_map);
823 }
824}
825
826#define BIND_INANY true
827
828static void test_all(void)
829{
830 const struct config {
831 int sotype;
832 sa_family_t family;
833 bool inany;
834 } configs[] = {
835 { SOCK_STREAM, AF_INET },
836 { SOCK_STREAM, AF_INET, BIND_INANY },
837 { SOCK_STREAM, AF_INET6 },
838 { SOCK_STREAM, AF_INET6, BIND_INANY },
839 { SOCK_DGRAM, AF_INET },
840 { SOCK_DGRAM, AF_INET6 },
841 };
842 const struct config *c;
843
844 for (c = configs; c < configs + ARRAY_SIZE(configs); c++)
845 test_config(c->sotype, c->family, c->inany);
846}
847
848void test_map_type(enum bpf_map_type mt)
849{
850 if (create_maps(mt))
851 goto out;
852 if (prepare_bpf_obj())
853 goto out;
854
855 test_all();
856out:
857 cleanup();
858}
859
860void test_select_reuseport(void)
861{
862 saved_tcp_fo = read_int_sysctl(TCP_FO_SYSCTL);
863 if (saved_tcp_fo < 0)
864 goto out;
865 saved_tcp_syncookie = read_int_sysctl(TCP_SYNCOOKIE_SYSCTL);
866 if (saved_tcp_syncookie < 0)
867 goto out;
868
869 if (enable_fastopen())
870 goto out;
871 if (disable_syncookie())
872 goto out;
873
874 test_map_type(BPF_MAP_TYPE_REUSEPORT_SOCKARRAY);
875 test_map_type(BPF_MAP_TYPE_SOCKMAP);
876 test_map_type(BPF_MAP_TYPE_SOCKHASH);
877out:
878 restore_sysctls();
879}