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