1 // SPDX-License-Identifier: GPL-2.0 2 // Copyright (c) 2017-2018 Covalent IO, Inc. http://covalent.io 3 #include <stdio.h> 4 #include <stdlib.h> 5 #include <sys/socket.h> 6 #include <sys/ioctl.h> 7 #include <sys/select.h> 8 #include <netinet/in.h> 9 #include <arpa/inet.h> 10 #include <unistd.h> 11 #include <string.h> 12 #include <errno.h> 13 #include <stdbool.h> 14 #include <signal.h> 15 #include <fcntl.h> 16 #include <sys/wait.h> 17 #include <time.h> 18 #include <sched.h> 19 20 #include <sys/time.h> 21 #include <sys/types.h> 22 #include <sys/sendfile.h> 23 24 #include <linux/netlink.h> 25 #include <linux/socket.h> 26 #include <linux/sock_diag.h> 27 #include <linux/bpf.h> 28 #include <linux/if_link.h> 29 #include <linux/tls.h> 30 #include <assert.h> 31 #include <libgen.h> 32 33 #include <getopt.h> 34 35 #include <bpf/bpf.h> 36 #include <bpf/libbpf.h> 37 38 #include "bpf_util.h" 39 #include "cgroup_helpers.h" 40 41 int running; 42 static void running_handler(int a); 43 44 #ifndef TCP_ULP 45 # define TCP_ULP 31 46 #endif 47 #ifndef SOL_TLS 48 # define SOL_TLS 282 49 #endif 50 51 /* randomly selected ports for testing on lo */ 52 #define S1_PORT 10000 53 #define S2_PORT 10001 54 55 #define BPF_SOCKMAP_FILENAME "test_sockmap_kern.bpf.o" 56 #define BPF_SOCKHASH_FILENAME "test_sockhash_kern.bpf.o" 57 #define CG_PATH "/sockmap" 58 59 /* global sockets */ 60 int s1, s2, c1, c2, p1, p2; 61 int test_cnt; 62 int passed; 63 int failed; 64 int map_fd[9]; 65 struct bpf_map *maps[9]; 66 int prog_fd[9]; 67 68 int txmsg_pass; 69 int txmsg_redir; 70 int txmsg_drop; 71 int txmsg_apply; 72 int txmsg_cork; 73 int txmsg_start; 74 int txmsg_end; 75 int txmsg_start_push; 76 int txmsg_end_push; 77 int txmsg_start_pop; 78 int txmsg_pop; 79 int txmsg_ingress; 80 int txmsg_redir_skb; 81 int txmsg_ktls_skb; 82 int txmsg_ktls_skb_drop; 83 int txmsg_ktls_skb_redir; 84 int ktls; 85 int peek_flag; 86 int skb_use_parser; 87 int txmsg_omit_skb_parser; 88 89 static const struct option long_options[] = { 90 {"help", no_argument, NULL, 'h' }, 91 {"cgroup", required_argument, NULL, 'c' }, 92 {"rate", required_argument, NULL, 'r' }, 93 {"verbose", optional_argument, NULL, 'v' }, 94 {"iov_count", required_argument, NULL, 'i' }, 95 {"length", required_argument, NULL, 'l' }, 96 {"test", required_argument, NULL, 't' }, 97 {"data_test", no_argument, NULL, 'd' }, 98 {"txmsg", no_argument, &txmsg_pass, 1 }, 99 {"txmsg_redir", no_argument, &txmsg_redir, 1 }, 100 {"txmsg_drop", no_argument, &txmsg_drop, 1 }, 101 {"txmsg_apply", required_argument, NULL, 'a'}, 102 {"txmsg_cork", required_argument, NULL, 'k'}, 103 {"txmsg_start", required_argument, NULL, 's'}, 104 {"txmsg_end", required_argument, NULL, 'e'}, 105 {"txmsg_start_push", required_argument, NULL, 'p'}, 106 {"txmsg_end_push", required_argument, NULL, 'q'}, 107 {"txmsg_start_pop", required_argument, NULL, 'w'}, 108 {"txmsg_pop", required_argument, NULL, 'x'}, 109 {"txmsg_ingress", no_argument, &txmsg_ingress, 1 }, 110 {"txmsg_redir_skb", no_argument, &txmsg_redir_skb, 1 }, 111 {"ktls", no_argument, &ktls, 1 }, 112 {"peek", no_argument, &peek_flag, 1 }, 113 {"txmsg_omit_skb_parser", no_argument, &txmsg_omit_skb_parser, 1}, 114 {"whitelist", required_argument, NULL, 'n' }, 115 {"blacklist", required_argument, NULL, 'b' }, 116 {0, 0, NULL, 0 } 117 }; 118 119 struct test_env { 120 const char *type; 121 const char *subtest; 122 const char *prepend; 123 124 int test_num; 125 int subtest_num; 126 127 int succ_cnt; 128 int fail_cnt; 129 int fail_last; 130 }; 131 132 struct test_env env; 133 134 struct sockmap_options { 135 int verbose; 136 bool base; 137 bool sendpage; 138 bool data_test; 139 bool drop_expected; 140 bool check_recved_len; 141 bool tx_wait_mem; 142 int iov_count; 143 int iov_length; 144 int rate; 145 char *map; 146 char *whitelist; 147 char *blacklist; 148 char *prepend; 149 }; 150 151 struct _test { 152 char *title; 153 void (*tester)(int cg_fd, struct sockmap_options *opt); 154 }; 155 156 static void test_start(void) 157 { 158 env.subtest_num++; 159 } 160 161 static void test_fail(void) 162 { 163 env.fail_cnt++; 164 } 165 166 static void test_pass(void) 167 { 168 env.succ_cnt++; 169 } 170 171 static void test_reset(void) 172 { 173 txmsg_start = txmsg_end = 0; 174 txmsg_start_pop = txmsg_pop = 0; 175 txmsg_start_push = txmsg_end_push = 0; 176 txmsg_pass = txmsg_drop = txmsg_redir = 0; 177 txmsg_apply = txmsg_cork = 0; 178 txmsg_ingress = txmsg_redir_skb = 0; 179 txmsg_ktls_skb = txmsg_ktls_skb_drop = txmsg_ktls_skb_redir = 0; 180 txmsg_omit_skb_parser = 0; 181 skb_use_parser = 0; 182 } 183 184 static int test_start_subtest(const struct _test *t, struct sockmap_options *o) 185 { 186 env.type = o->map; 187 env.subtest = t->title; 188 env.prepend = o->prepend; 189 env.test_num++; 190 env.subtest_num = 0; 191 env.fail_last = env.fail_cnt; 192 test_reset(); 193 return 0; 194 } 195 196 static void test_end_subtest(void) 197 { 198 int error = env.fail_cnt - env.fail_last; 199 int type = strcmp(env.type, BPF_SOCKMAP_FILENAME); 200 201 if (!error) 202 test_pass(); 203 204 fprintf(stdout, "#%2d/%2d %8s:%s:%s:%s\n", 205 env.test_num, env.subtest_num, 206 !type ? "sockmap" : "sockhash", 207 env.prepend ? : "", 208 env.subtest, error ? "FAIL" : "OK"); 209 } 210 211 static void test_print_results(void) 212 { 213 fprintf(stdout, "Pass: %d Fail: %d\n", 214 env.succ_cnt, env.fail_cnt); 215 } 216 217 static void usage(char *argv[]) 218 { 219 int i; 220 221 printf(" Usage: %s --cgroup <cgroup_path>\n", argv[0]); 222 printf(" options:\n"); 223 for (i = 0; long_options[i].name != 0; i++) { 224 printf(" --%-12s", long_options[i].name); 225 if (long_options[i].flag != NULL) 226 printf(" flag (internal value:%d)\n", 227 *long_options[i].flag); 228 else 229 printf(" -%c\n", long_options[i].val); 230 } 231 printf("\n"); 232 } 233 234 char *sock_to_string(int s) 235 { 236 if (s == c1) 237 return "client1"; 238 else if (s == c2) 239 return "client2"; 240 else if (s == s1) 241 return "server1"; 242 else if (s == s2) 243 return "server2"; 244 else if (s == p1) 245 return "peer1"; 246 else if (s == p2) 247 return "peer2"; 248 else 249 return "unknown"; 250 } 251 252 static int sockmap_init_ktls(int verbose, int s) 253 { 254 struct tls12_crypto_info_aes_gcm_128 tls_tx = { 255 .info = { 256 .version = TLS_1_2_VERSION, 257 .cipher_type = TLS_CIPHER_AES_GCM_128, 258 }, 259 }; 260 struct tls12_crypto_info_aes_gcm_128 tls_rx = { 261 .info = { 262 .version = TLS_1_2_VERSION, 263 .cipher_type = TLS_CIPHER_AES_GCM_128, 264 }, 265 }; 266 int so_buf = 6553500; 267 int err; 268 269 err = setsockopt(s, 6, TCP_ULP, "tls", sizeof("tls")); 270 if (err) { 271 fprintf(stderr, "setsockopt: TCP_ULP(%s) failed with error %i\n", sock_to_string(s), err); 272 return -EINVAL; 273 } 274 err = setsockopt(s, SOL_TLS, TLS_TX, (void *)&tls_tx, sizeof(tls_tx)); 275 if (err) { 276 fprintf(stderr, "setsockopt: TLS_TX(%s) failed with error %i\n", sock_to_string(s), err); 277 return -EINVAL; 278 } 279 err = setsockopt(s, SOL_TLS, TLS_RX, (void *)&tls_rx, sizeof(tls_rx)); 280 if (err) { 281 fprintf(stderr, "setsockopt: TLS_RX(%s) failed with error %i\n", sock_to_string(s), err); 282 return -EINVAL; 283 } 284 err = setsockopt(s, SOL_SOCKET, SO_SNDBUF, &so_buf, sizeof(so_buf)); 285 if (err) { 286 fprintf(stderr, "setsockopt: (%s) failed sndbuf with error %i\n", sock_to_string(s), err); 287 return -EINVAL; 288 } 289 err = setsockopt(s, SOL_SOCKET, SO_RCVBUF, &so_buf, sizeof(so_buf)); 290 if (err) { 291 fprintf(stderr, "setsockopt: (%s) failed rcvbuf with error %i\n", sock_to_string(s), err); 292 return -EINVAL; 293 } 294 295 if (verbose) 296 fprintf(stdout, "socket(%s) kTLS enabled\n", sock_to_string(s)); 297 return 0; 298 } 299 static int sockmap_init_sockets(int verbose) 300 { 301 int i, err, one = 1; 302 struct sockaddr_in addr; 303 int *fds[4] = {&s1, &s2, &c1, &c2}; 304 305 s1 = s2 = p1 = p2 = c1 = c2 = 0; 306 307 /* Init sockets */ 308 for (i = 0; i < 4; i++) { 309 *fds[i] = socket(AF_INET, SOCK_STREAM, 0); 310 if (*fds[i] < 0) { 311 perror("socket s1 failed()"); 312 return errno; 313 } 314 } 315 316 /* Allow reuse */ 317 for (i = 0; i < 2; i++) { 318 err = setsockopt(*fds[i], SOL_SOCKET, SO_REUSEADDR, 319 (char *)&one, sizeof(one)); 320 if (err) { 321 perror("setsockopt failed()"); 322 return errno; 323 } 324 } 325 326 /* Non-blocking sockets */ 327 for (i = 0; i < 2; i++) { 328 err = ioctl(*fds[i], FIONBIO, (char *)&one); 329 if (err < 0) { 330 perror("ioctl s1 failed()"); 331 return errno; 332 } 333 } 334 335 /* Bind server sockets */ 336 memset(&addr, 0, sizeof(struct sockaddr_in)); 337 addr.sin_family = AF_INET; 338 addr.sin_addr.s_addr = inet_addr("127.0.0.1"); 339 340 addr.sin_port = htons(S1_PORT); 341 err = bind(s1, (struct sockaddr *)&addr, sizeof(addr)); 342 if (err < 0) { 343 perror("bind s1 failed()"); 344 return errno; 345 } 346 347 addr.sin_port = htons(S2_PORT); 348 err = bind(s2, (struct sockaddr *)&addr, sizeof(addr)); 349 if (err < 0) { 350 perror("bind s2 failed()"); 351 return errno; 352 } 353 354 /* Listen server sockets */ 355 addr.sin_port = htons(S1_PORT); 356 err = listen(s1, 32); 357 if (err < 0) { 358 perror("listen s1 failed()"); 359 return errno; 360 } 361 362 addr.sin_port = htons(S2_PORT); 363 err = listen(s2, 32); 364 if (err < 0) { 365 perror("listen s1 failed()"); 366 return errno; 367 } 368 369 /* Initiate Connect */ 370 addr.sin_port = htons(S1_PORT); 371 err = connect(c1, (struct sockaddr *)&addr, sizeof(addr)); 372 if (err < 0 && errno != EINPROGRESS) { 373 perror("connect c1 failed()"); 374 return errno; 375 } 376 377 addr.sin_port = htons(S2_PORT); 378 err = connect(c2, (struct sockaddr *)&addr, sizeof(addr)); 379 if (err < 0 && errno != EINPROGRESS) { 380 perror("connect c2 failed()"); 381 return errno; 382 } else if (err < 0) { 383 err = 0; 384 } 385 386 /* Accept Connecrtions */ 387 p1 = accept(s1, NULL, NULL); 388 if (p1 < 0) { 389 perror("accept s1 failed()"); 390 return errno; 391 } 392 393 p2 = accept(s2, NULL, NULL); 394 if (p2 < 0) { 395 perror("accept s1 failed()"); 396 return errno; 397 } 398 399 if (verbose > 1) { 400 printf("connected sockets: c1 <-> p1, c2 <-> p2\n"); 401 printf("cgroups binding: c1(%i) <-> s1(%i) - - - c2(%i) <-> s2(%i)\n", 402 c1, s1, c2, s2); 403 } 404 return 0; 405 } 406 407 struct msg_stats { 408 size_t bytes_sent; 409 size_t bytes_recvd; 410 struct timespec start; 411 struct timespec end; 412 }; 413 414 static int msg_loop_sendpage(int fd, int iov_length, int cnt, 415 struct msg_stats *s, 416 struct sockmap_options *opt) 417 { 418 bool drop = opt->drop_expected; 419 unsigned char k = 0; 420 FILE *file; 421 int i, fp; 422 423 file = tmpfile(); 424 if (!file) { 425 perror("create file for sendpage"); 426 return 1; 427 } 428 for (i = 0; i < iov_length * cnt; i++, k++) 429 fwrite(&k, sizeof(char), 1, file); 430 fflush(file); 431 fseek(file, 0, SEEK_SET); 432 433 fp = fileno(file); 434 435 clock_gettime(CLOCK_MONOTONIC, &s->start); 436 for (i = 0; i < cnt; i++) { 437 int sent; 438 439 errno = 0; 440 sent = sendfile(fd, fp, NULL, iov_length); 441 442 if (!drop && sent < 0) { 443 perror("sendpage loop error"); 444 fclose(file); 445 return sent; 446 } else if (drop && sent >= 0) { 447 printf("sendpage loop error expected: %i errno %i\n", 448 sent, errno); 449 fclose(file); 450 return -EIO; 451 } 452 453 if (sent > 0) 454 s->bytes_sent += sent; 455 } 456 clock_gettime(CLOCK_MONOTONIC, &s->end); 457 fclose(file); 458 return 0; 459 } 460 461 static void msg_free_iov(struct msghdr *msg) 462 { 463 int i; 464 465 for (i = 0; i < msg->msg_iovlen; i++) 466 free(msg->msg_iov[i].iov_base); 467 free(msg->msg_iov); 468 msg->msg_iov = NULL; 469 msg->msg_iovlen = 0; 470 } 471 472 static int msg_alloc_iov(struct msghdr *msg, 473 int iov_count, int iov_length, 474 bool data, bool xmit) 475 { 476 unsigned char k = 0; 477 struct iovec *iov; 478 int i; 479 480 iov = calloc(iov_count, sizeof(struct iovec)); 481 if (!iov) 482 return errno; 483 484 for (i = 0; i < iov_count; i++) { 485 unsigned char *d = calloc(iov_length, sizeof(char)); 486 487 if (!d) { 488 fprintf(stderr, "iov_count %i/%i OOM\n", i, iov_count); 489 goto unwind_iov; 490 } 491 iov[i].iov_base = d; 492 iov[i].iov_len = iov_length; 493 494 if (data && xmit) { 495 int j; 496 497 for (j = 0; j < iov_length; j++) 498 d[j] = k++; 499 } 500 } 501 502 msg->msg_iov = iov; 503 msg->msg_iovlen = iov_count; 504 505 return 0; 506 unwind_iov: 507 for (i--; i >= 0 ; i--) 508 free(msg->msg_iov[i].iov_base); 509 return -ENOMEM; 510 } 511 512 static int msg_verify_data(struct msghdr *msg, int size, int chunk_sz) 513 { 514 int i, j = 0, bytes_cnt = 0; 515 unsigned char k = 0; 516 517 for (i = 0; i < msg->msg_iovlen; i++) { 518 unsigned char *d = msg->msg_iov[i].iov_base; 519 520 /* Special case test for skb ingress + ktls */ 521 if (i == 0 && txmsg_ktls_skb) { 522 if (msg->msg_iov[i].iov_len < 4) 523 return -EIO; 524 if (memcmp(d, "PASS", 4) != 0) { 525 fprintf(stderr, 526 "detected skb data error with skb ingress update @iov[%i]:%i \"%02x %02x %02x %02x\" != \"PASS\"\n", 527 i, 0, d[0], d[1], d[2], d[3]); 528 return -EIO; 529 } 530 j = 4; /* advance index past PASS header */ 531 } 532 533 for (; j < msg->msg_iov[i].iov_len && size; j++) { 534 if (d[j] != k++) { 535 fprintf(stderr, 536 "detected data corruption @iov[%i]:%i %02x != %02x, %02x ?= %02x\n", 537 i, j, d[j], k - 1, d[j+1], k); 538 return -EIO; 539 } 540 bytes_cnt++; 541 if (bytes_cnt == chunk_sz) { 542 k = 0; 543 bytes_cnt = 0; 544 } 545 size--; 546 } 547 } 548 return 0; 549 } 550 551 static int msg_loop(int fd, int iov_count, int iov_length, int cnt, 552 struct msg_stats *s, bool tx, 553 struct sockmap_options *opt) 554 { 555 struct msghdr msg = {0}, msg_peek = {0}; 556 int err, i, flags = MSG_NOSIGNAL; 557 bool drop = opt->drop_expected; 558 bool data = opt->data_test; 559 int iov_alloc_length = iov_length; 560 561 if (!tx && opt->check_recved_len) 562 iov_alloc_length *= 2; 563 564 err = msg_alloc_iov(&msg, iov_count, iov_alloc_length, data, tx); 565 if (err) 566 goto out_errno; 567 if (peek_flag) { 568 err = msg_alloc_iov(&msg_peek, iov_count, iov_length, data, tx); 569 if (err) 570 goto out_errno; 571 } 572 573 if (tx) { 574 clock_gettime(CLOCK_MONOTONIC, &s->start); 575 for (i = 0; i < cnt; i++) { 576 int sent; 577 578 errno = 0; 579 sent = sendmsg(fd, &msg, flags); 580 581 if (!drop && sent < 0) { 582 if (opt->tx_wait_mem && errno == EACCES) { 583 errno = 0; 584 goto out_errno; 585 } 586 perror("sendmsg loop error"); 587 goto out_errno; 588 } else if (drop && sent >= 0) { 589 fprintf(stderr, 590 "sendmsg loop error expected: %i errno %i\n", 591 sent, errno); 592 errno = -EIO; 593 goto out_errno; 594 } 595 if (sent > 0) 596 s->bytes_sent += sent; 597 } 598 clock_gettime(CLOCK_MONOTONIC, &s->end); 599 } else { 600 int slct, recvp = 0, recv, max_fd = fd; 601 float total_bytes, txmsg_pop_total; 602 int fd_flags = O_NONBLOCK; 603 struct timeval timeout; 604 fd_set w; 605 606 fcntl(fd, fd_flags); 607 /* Account for pop bytes noting each iteration of apply will 608 * call msg_pop_data helper so we need to account for this 609 * by calculating the number of apply iterations. Note user 610 * of the tool can create cases where no data is sent by 611 * manipulating pop/push/pull/etc. For example txmsg_apply 1 612 * with txmsg_pop 1 will try to apply 1B at a time but each 613 * iteration will then pop 1B so no data will ever be sent. 614 * This is really only useful for testing edge cases in code 615 * paths. 616 */ 617 total_bytes = (float)iov_count * (float)iov_length * (float)cnt; 618 if (txmsg_apply) 619 txmsg_pop_total = txmsg_pop * (total_bytes / txmsg_apply); 620 else 621 txmsg_pop_total = txmsg_pop * cnt; 622 total_bytes -= txmsg_pop_total; 623 err = clock_gettime(CLOCK_MONOTONIC, &s->start); 624 if (err < 0) 625 perror("recv start time"); 626 while (s->bytes_recvd < total_bytes) { 627 if (txmsg_cork) { 628 timeout.tv_sec = 0; 629 timeout.tv_usec = 300000; 630 } else { 631 timeout.tv_sec = 3; 632 timeout.tv_usec = 0; 633 } 634 635 /* FD sets */ 636 FD_ZERO(&w); 637 FD_SET(fd, &w); 638 639 slct = select(max_fd + 1, &w, NULL, NULL, &timeout); 640 if (slct == -1) { 641 perror("select()"); 642 clock_gettime(CLOCK_MONOTONIC, &s->end); 643 goto out_errno; 644 } else if (!slct) { 645 if (opt->verbose) 646 fprintf(stderr, "unexpected timeout: recved %zu/%f pop_total %f\n", s->bytes_recvd, total_bytes, txmsg_pop_total); 647 errno = -EIO; 648 clock_gettime(CLOCK_MONOTONIC, &s->end); 649 goto out_errno; 650 } 651 652 if (opt->tx_wait_mem) { 653 FD_ZERO(&w); 654 FD_SET(fd, &w); 655 slct = select(max_fd + 1, NULL, NULL, &w, &timeout); 656 errno = 0; 657 close(fd); 658 goto out_errno; 659 } 660 661 errno = 0; 662 if (peek_flag) { 663 flags |= MSG_PEEK; 664 recvp = recvmsg(fd, &msg_peek, flags); 665 if (recvp < 0) { 666 if (errno != EWOULDBLOCK) { 667 clock_gettime(CLOCK_MONOTONIC, &s->end); 668 goto out_errno; 669 } 670 } 671 flags = 0; 672 } 673 674 recv = recvmsg(fd, &msg, flags); 675 if (recv < 0) { 676 if (errno != EWOULDBLOCK) { 677 clock_gettime(CLOCK_MONOTONIC, &s->end); 678 perror("recv failed()"); 679 goto out_errno; 680 } 681 } 682 683 if (recv > 0) 684 s->bytes_recvd += recv; 685 686 if (opt->check_recved_len && s->bytes_recvd > total_bytes) { 687 errno = EMSGSIZE; 688 fprintf(stderr, "recv failed(), bytes_recvd:%zd, total_bytes:%f\n", 689 s->bytes_recvd, total_bytes); 690 goto out_errno; 691 } 692 693 if (data) { 694 int chunk_sz = opt->sendpage ? 695 iov_length * cnt : 696 iov_length * iov_count; 697 698 errno = msg_verify_data(&msg, recv, chunk_sz); 699 if (errno) { 700 perror("data verify msg failed"); 701 goto out_errno; 702 } 703 if (recvp) { 704 errno = msg_verify_data(&msg_peek, 705 recvp, 706 chunk_sz); 707 if (errno) { 708 perror("data verify msg_peek failed"); 709 goto out_errno; 710 } 711 } 712 } 713 } 714 clock_gettime(CLOCK_MONOTONIC, &s->end); 715 } 716 717 msg_free_iov(&msg); 718 msg_free_iov(&msg_peek); 719 return err; 720 out_errno: 721 msg_free_iov(&msg); 722 msg_free_iov(&msg_peek); 723 return errno; 724 } 725 726 static float giga = 1000000000; 727 728 static inline float sentBps(struct msg_stats s) 729 { 730 return s.bytes_sent / (s.end.tv_sec - s.start.tv_sec); 731 } 732 733 static inline float recvdBps(struct msg_stats s) 734 { 735 return s.bytes_recvd / (s.end.tv_sec - s.start.tv_sec); 736 } 737 738 static int sendmsg_test(struct sockmap_options *opt) 739 { 740 float sent_Bps = 0, recvd_Bps = 0; 741 int rx_fd, txpid, rxpid, err = 0; 742 struct msg_stats s = {0}; 743 int iov_count = opt->iov_count; 744 int iov_buf = opt->iov_length; 745 int rx_status, tx_status; 746 int cnt = opt->rate; 747 748 errno = 0; 749 750 if (opt->base) 751 rx_fd = p1; 752 else 753 rx_fd = p2; 754 755 if (ktls) { 756 /* Redirecting into non-TLS socket which sends into a TLS 757 * socket is not a valid test. So in this case lets not 758 * enable kTLS but still run the test. 759 */ 760 if (!txmsg_redir || txmsg_ingress) { 761 err = sockmap_init_ktls(opt->verbose, rx_fd); 762 if (err) 763 return err; 764 } 765 err = sockmap_init_ktls(opt->verbose, c1); 766 if (err) 767 return err; 768 } 769 770 if (opt->tx_wait_mem) { 771 struct timeval timeout; 772 int rxtx_buf_len = 1024; 773 774 timeout.tv_sec = 3; 775 timeout.tv_usec = 0; 776 777 err = setsockopt(c2, SOL_SOCKET, SO_SNDTIMEO, &timeout, sizeof(struct timeval)); 778 err |= setsockopt(c2, SOL_SOCKET, SO_SNDBUFFORCE, &rxtx_buf_len, sizeof(int)); 779 err |= setsockopt(p2, SOL_SOCKET, SO_RCVBUFFORCE, &rxtx_buf_len, sizeof(int)); 780 if (err) { 781 perror("setsockopt failed()"); 782 return errno; 783 } 784 } 785 786 rxpid = fork(); 787 if (rxpid == 0) { 788 if (txmsg_pop || txmsg_start_pop) 789 iov_buf -= (txmsg_pop - txmsg_start_pop + 1); 790 if (opt->drop_expected || txmsg_ktls_skb_drop) 791 _exit(0); 792 793 if (!iov_buf) /* zero bytes sent case */ 794 _exit(0); 795 796 if (opt->sendpage) 797 iov_count = 1; 798 err = msg_loop(rx_fd, iov_count, iov_buf, 799 cnt, &s, false, opt); 800 if (opt->verbose > 1) 801 fprintf(stderr, 802 "msg_loop_rx: iov_count %i iov_buf %i cnt %i err %i\n", 803 iov_count, iov_buf, cnt, err); 804 if (s.end.tv_sec - s.start.tv_sec) { 805 sent_Bps = sentBps(s); 806 recvd_Bps = recvdBps(s); 807 } 808 if (opt->verbose > 1) 809 fprintf(stdout, 810 "rx_sendmsg: TX: %zuB %fB/s %fGB/s RX: %zuB %fB/s %fGB/s %s\n", 811 s.bytes_sent, sent_Bps, sent_Bps/giga, 812 s.bytes_recvd, recvd_Bps, recvd_Bps/giga, 813 peek_flag ? "(peek_msg)" : ""); 814 if (err && txmsg_cork) 815 err = 0; 816 exit(err ? 1 : 0); 817 } else if (rxpid == -1) { 818 perror("msg_loop_rx"); 819 return errno; 820 } 821 822 if (opt->tx_wait_mem) 823 close(c2); 824 825 txpid = fork(); 826 if (txpid == 0) { 827 if (opt->sendpage) 828 err = msg_loop_sendpage(c1, iov_buf, cnt, &s, opt); 829 else 830 err = msg_loop(c1, iov_count, iov_buf, 831 cnt, &s, true, opt); 832 833 if (err) 834 fprintf(stderr, 835 "msg_loop_tx: iov_count %i iov_buf %i cnt %i err %i\n", 836 iov_count, iov_buf, cnt, err); 837 if (s.end.tv_sec - s.start.tv_sec) { 838 sent_Bps = sentBps(s); 839 recvd_Bps = recvdBps(s); 840 } 841 if (opt->verbose > 1) 842 fprintf(stdout, 843 "tx_sendmsg: TX: %zuB %fB/s %f GB/s RX: %zuB %fB/s %fGB/s\n", 844 s.bytes_sent, sent_Bps, sent_Bps/giga, 845 s.bytes_recvd, recvd_Bps, recvd_Bps/giga); 846 exit(err ? 1 : 0); 847 } else if (txpid == -1) { 848 perror("msg_loop_tx"); 849 return errno; 850 } 851 852 assert(waitpid(rxpid, &rx_status, 0) == rxpid); 853 assert(waitpid(txpid, &tx_status, 0) == txpid); 854 if (WIFEXITED(rx_status)) { 855 err = WEXITSTATUS(rx_status); 856 if (err) { 857 fprintf(stderr, "rx thread exited with err %d.\n", err); 858 goto out; 859 } 860 } 861 if (WIFEXITED(tx_status)) { 862 err = WEXITSTATUS(tx_status); 863 if (err) 864 fprintf(stderr, "tx thread exited with err %d.\n", err); 865 } 866 out: 867 return err; 868 } 869 870 static int forever_ping_pong(int rate, struct sockmap_options *opt) 871 { 872 struct timeval timeout; 873 char buf[1024] = {0}; 874 int sc; 875 876 timeout.tv_sec = 10; 877 timeout.tv_usec = 0; 878 879 /* Ping/Pong data from client to server */ 880 sc = send(c1, buf, sizeof(buf), 0); 881 if (sc < 0) { 882 perror("send failed()"); 883 return sc; 884 } 885 886 do { 887 int s, rc, i, max_fd = p2; 888 fd_set w; 889 890 /* FD sets */ 891 FD_ZERO(&w); 892 FD_SET(c1, &w); 893 FD_SET(c2, &w); 894 FD_SET(p1, &w); 895 FD_SET(p2, &w); 896 897 s = select(max_fd + 1, &w, NULL, NULL, &timeout); 898 if (s == -1) { 899 perror("select()"); 900 break; 901 } else if (!s) { 902 fprintf(stderr, "unexpected timeout\n"); 903 break; 904 } 905 906 for (i = 0; i <= max_fd && s > 0; ++i) { 907 if (!FD_ISSET(i, &w)) 908 continue; 909 910 s--; 911 912 rc = recv(i, buf, sizeof(buf), 0); 913 if (rc < 0) { 914 if (errno != EWOULDBLOCK) { 915 perror("recv failed()"); 916 return rc; 917 } 918 } 919 920 if (rc == 0) { 921 close(i); 922 break; 923 } 924 925 sc = send(i, buf, rc, 0); 926 if (sc < 0) { 927 perror("send failed()"); 928 return sc; 929 } 930 } 931 932 if (rate) 933 sleep(rate); 934 935 if (opt->verbose) { 936 printf("."); 937 fflush(stdout); 938 939 } 940 } while (running); 941 942 return 0; 943 } 944 945 enum { 946 SELFTESTS, 947 PING_PONG, 948 SENDMSG, 949 BASE, 950 BASE_SENDPAGE, 951 SENDPAGE, 952 }; 953 954 static int run_options(struct sockmap_options *options, int cg_fd, int test) 955 { 956 int i, key, next_key, err, tx_prog_fd = -1, zero = 0; 957 958 /* If base test skip BPF setup */ 959 if (test == BASE || test == BASE_SENDPAGE) 960 goto run; 961 962 /* Attach programs to sockmap */ 963 if (!txmsg_omit_skb_parser) { 964 err = bpf_prog_attach(prog_fd[0], map_fd[0], 965 BPF_SK_SKB_STREAM_PARSER, 0); 966 if (err) { 967 fprintf(stderr, 968 "ERROR: bpf_prog_attach (sockmap %i->%i): %d (%s)\n", 969 prog_fd[0], map_fd[0], err, strerror(errno)); 970 return err; 971 } 972 } 973 974 err = bpf_prog_attach(prog_fd[1], map_fd[0], 975 BPF_SK_SKB_STREAM_VERDICT, 0); 976 if (err) { 977 fprintf(stderr, "ERROR: bpf_prog_attach (sockmap): %d (%s)\n", 978 err, strerror(errno)); 979 return err; 980 } 981 982 /* Attach programs to TLS sockmap */ 983 if (txmsg_ktls_skb) { 984 if (!txmsg_omit_skb_parser) { 985 err = bpf_prog_attach(prog_fd[0], map_fd[8], 986 BPF_SK_SKB_STREAM_PARSER, 0); 987 if (err) { 988 fprintf(stderr, 989 "ERROR: bpf_prog_attach (TLS sockmap %i->%i): %d (%s)\n", 990 prog_fd[0], map_fd[8], err, strerror(errno)); 991 return err; 992 } 993 } 994 995 err = bpf_prog_attach(prog_fd[2], map_fd[8], 996 BPF_SK_SKB_STREAM_VERDICT, 0); 997 if (err) { 998 fprintf(stderr, "ERROR: bpf_prog_attach (TLS sockmap): %d (%s)\n", 999 err, strerror(errno)); 1000 return err; 1001 } 1002 } 1003 1004 /* Attach to cgroups */ 1005 err = bpf_prog_attach(prog_fd[3], cg_fd, BPF_CGROUP_SOCK_OPS, 0); 1006 if (err) { 1007 fprintf(stderr, "ERROR: bpf_prog_attach (groups): %d (%s)\n", 1008 err, strerror(errno)); 1009 return err; 1010 } 1011 1012 run: 1013 err = sockmap_init_sockets(options->verbose); 1014 if (err) { 1015 fprintf(stderr, "ERROR: test socket failed: %d\n", err); 1016 goto out; 1017 } 1018 1019 /* Attach txmsg program to sockmap */ 1020 if (txmsg_pass) 1021 tx_prog_fd = prog_fd[4]; 1022 else if (txmsg_redir) 1023 tx_prog_fd = prog_fd[5]; 1024 else if (txmsg_apply) 1025 tx_prog_fd = prog_fd[6]; 1026 else if (txmsg_cork) 1027 tx_prog_fd = prog_fd[7]; 1028 else if (txmsg_drop) 1029 tx_prog_fd = prog_fd[8]; 1030 else 1031 tx_prog_fd = 0; 1032 1033 if (tx_prog_fd) { 1034 int redir_fd, i = 0; 1035 1036 err = bpf_prog_attach(tx_prog_fd, 1037 map_fd[1], BPF_SK_MSG_VERDICT, 0); 1038 if (err) { 1039 fprintf(stderr, 1040 "ERROR: bpf_prog_attach (txmsg): %d (%s)\n", 1041 err, strerror(errno)); 1042 goto out; 1043 } 1044 1045 err = bpf_map_update_elem(map_fd[1], &i, &c1, BPF_ANY); 1046 if (err) { 1047 fprintf(stderr, 1048 "ERROR: bpf_map_update_elem (txmsg): %d (%s\n", 1049 err, strerror(errno)); 1050 goto out; 1051 } 1052 1053 if (txmsg_redir) 1054 redir_fd = c2; 1055 else 1056 redir_fd = c1; 1057 1058 err = bpf_map_update_elem(map_fd[2], &i, &redir_fd, BPF_ANY); 1059 if (err) { 1060 fprintf(stderr, 1061 "ERROR: bpf_map_update_elem (txmsg): %d (%s\n", 1062 err, strerror(errno)); 1063 goto out; 1064 } 1065 1066 if (txmsg_apply) { 1067 err = bpf_map_update_elem(map_fd[3], 1068 &i, &txmsg_apply, BPF_ANY); 1069 if (err) { 1070 fprintf(stderr, 1071 "ERROR: bpf_map_update_elem (apply_bytes): %d (%s\n", 1072 err, strerror(errno)); 1073 goto out; 1074 } 1075 } 1076 1077 if (txmsg_cork) { 1078 err = bpf_map_update_elem(map_fd[4], 1079 &i, &txmsg_cork, BPF_ANY); 1080 if (err) { 1081 fprintf(stderr, 1082 "ERROR: bpf_map_update_elem (cork_bytes): %d (%s\n", 1083 err, strerror(errno)); 1084 goto out; 1085 } 1086 } 1087 1088 if (txmsg_start) { 1089 err = bpf_map_update_elem(map_fd[5], 1090 &i, &txmsg_start, BPF_ANY); 1091 if (err) { 1092 fprintf(stderr, 1093 "ERROR: bpf_map_update_elem (txmsg_start): %d (%s)\n", 1094 err, strerror(errno)); 1095 goto out; 1096 } 1097 } 1098 1099 if (txmsg_end) { 1100 i = 1; 1101 err = bpf_map_update_elem(map_fd[5], 1102 &i, &txmsg_end, BPF_ANY); 1103 if (err) { 1104 fprintf(stderr, 1105 "ERROR: bpf_map_update_elem (txmsg_end): %d (%s)\n", 1106 err, strerror(errno)); 1107 goto out; 1108 } 1109 } 1110 1111 if (txmsg_start_push) { 1112 i = 2; 1113 err = bpf_map_update_elem(map_fd[5], 1114 &i, &txmsg_start_push, BPF_ANY); 1115 if (err) { 1116 fprintf(stderr, 1117 "ERROR: bpf_map_update_elem (txmsg_start_push): %d (%s)\n", 1118 err, strerror(errno)); 1119 goto out; 1120 } 1121 } 1122 1123 if (txmsg_end_push) { 1124 i = 3; 1125 err = bpf_map_update_elem(map_fd[5], 1126 &i, &txmsg_end_push, BPF_ANY); 1127 if (err) { 1128 fprintf(stderr, 1129 "ERROR: bpf_map_update_elem %i@%i (txmsg_end_push): %d (%s)\n", 1130 txmsg_end_push, i, err, strerror(errno)); 1131 goto out; 1132 } 1133 } 1134 1135 if (txmsg_start_pop) { 1136 i = 4; 1137 err = bpf_map_update_elem(map_fd[5], 1138 &i, &txmsg_start_pop, BPF_ANY); 1139 if (err) { 1140 fprintf(stderr, 1141 "ERROR: bpf_map_update_elem %i@%i (txmsg_start_pop): %d (%s)\n", 1142 txmsg_start_pop, i, err, strerror(errno)); 1143 goto out; 1144 } 1145 } else { 1146 i = 4; 1147 bpf_map_update_elem(map_fd[5], 1148 &i, &txmsg_start_pop, BPF_ANY); 1149 } 1150 1151 if (txmsg_pop) { 1152 i = 5; 1153 err = bpf_map_update_elem(map_fd[5], 1154 &i, &txmsg_pop, BPF_ANY); 1155 if (err) { 1156 fprintf(stderr, 1157 "ERROR: bpf_map_update_elem %i@%i (txmsg_pop): %d (%s)\n", 1158 txmsg_pop, i, err, strerror(errno)); 1159 goto out; 1160 } 1161 } else { 1162 i = 5; 1163 bpf_map_update_elem(map_fd[5], 1164 &i, &txmsg_pop, BPF_ANY); 1165 1166 } 1167 1168 if (txmsg_ingress) { 1169 int in = BPF_F_INGRESS; 1170 1171 i = 0; 1172 err = bpf_map_update_elem(map_fd[6], &i, &in, BPF_ANY); 1173 if (err) { 1174 fprintf(stderr, 1175 "ERROR: bpf_map_update_elem (txmsg_ingress): %d (%s)\n", 1176 err, strerror(errno)); 1177 } 1178 i = 1; 1179 err = bpf_map_update_elem(map_fd[1], &i, &p1, BPF_ANY); 1180 if (err) { 1181 fprintf(stderr, 1182 "ERROR: bpf_map_update_elem (p1 txmsg): %d (%s)\n", 1183 err, strerror(errno)); 1184 } 1185 err = bpf_map_update_elem(map_fd[2], &i, &p1, BPF_ANY); 1186 if (err) { 1187 fprintf(stderr, 1188 "ERROR: bpf_map_update_elem (p1 redir): %d (%s)\n", 1189 err, strerror(errno)); 1190 } 1191 1192 i = 2; 1193 err = bpf_map_update_elem(map_fd[2], &i, &p2, BPF_ANY); 1194 if (err) { 1195 fprintf(stderr, 1196 "ERROR: bpf_map_update_elem (p2 txmsg): %d (%s)\n", 1197 err, strerror(errno)); 1198 } 1199 } 1200 1201 if (txmsg_ktls_skb) { 1202 int ingress = BPF_F_INGRESS; 1203 1204 i = 0; 1205 err = bpf_map_update_elem(map_fd[8], &i, &p2, BPF_ANY); 1206 if (err) { 1207 fprintf(stderr, 1208 "ERROR: bpf_map_update_elem (c1 sockmap): %d (%s)\n", 1209 err, strerror(errno)); 1210 } 1211 1212 if (txmsg_ktls_skb_redir) { 1213 i = 1; 1214 err = bpf_map_update_elem(map_fd[7], 1215 &i, &ingress, BPF_ANY); 1216 if (err) { 1217 fprintf(stderr, 1218 "ERROR: bpf_map_update_elem (txmsg_ingress): %d (%s)\n", 1219 err, strerror(errno)); 1220 } 1221 } 1222 1223 if (txmsg_ktls_skb_drop) { 1224 i = 1; 1225 err = bpf_map_update_elem(map_fd[7], &i, &i, BPF_ANY); 1226 } 1227 } 1228 1229 if (txmsg_redir_skb) { 1230 int skb_fd = (test == SENDMSG || test == SENDPAGE) ? 1231 p2 : p1; 1232 int ingress = BPF_F_INGRESS; 1233 1234 i = 0; 1235 err = bpf_map_update_elem(map_fd[7], 1236 &i, &ingress, BPF_ANY); 1237 if (err) { 1238 fprintf(stderr, 1239 "ERROR: bpf_map_update_elem (txmsg_ingress): %d (%s)\n", 1240 err, strerror(errno)); 1241 } 1242 1243 i = 3; 1244 err = bpf_map_update_elem(map_fd[0], &i, &skb_fd, BPF_ANY); 1245 if (err) { 1246 fprintf(stderr, 1247 "ERROR: bpf_map_update_elem (c1 sockmap): %d (%s)\n", 1248 err, strerror(errno)); 1249 } 1250 } 1251 } 1252 1253 if (skb_use_parser) { 1254 i = 2; 1255 err = bpf_map_update_elem(map_fd[7], &i, &skb_use_parser, BPF_ANY); 1256 } 1257 1258 if (txmsg_drop) 1259 options->drop_expected = true; 1260 1261 if (test == PING_PONG) 1262 err = forever_ping_pong(options->rate, options); 1263 else if (test == SENDMSG) { 1264 options->base = false; 1265 options->sendpage = false; 1266 err = sendmsg_test(options); 1267 } else if (test == SENDPAGE) { 1268 options->base = false; 1269 options->sendpage = true; 1270 err = sendmsg_test(options); 1271 } else if (test == BASE) { 1272 options->base = true; 1273 options->sendpage = false; 1274 err = sendmsg_test(options); 1275 } else if (test == BASE_SENDPAGE) { 1276 options->base = true; 1277 options->sendpage = true; 1278 err = sendmsg_test(options); 1279 } else 1280 fprintf(stderr, "unknown test\n"); 1281 out: 1282 /* Detatch and zero all the maps */ 1283 bpf_prog_detach2(prog_fd[3], cg_fd, BPF_CGROUP_SOCK_OPS); 1284 bpf_prog_detach2(prog_fd[0], map_fd[0], BPF_SK_SKB_STREAM_PARSER); 1285 bpf_prog_detach2(prog_fd[1], map_fd[0], BPF_SK_SKB_STREAM_VERDICT); 1286 bpf_prog_detach2(prog_fd[0], map_fd[8], BPF_SK_SKB_STREAM_PARSER); 1287 bpf_prog_detach2(prog_fd[2], map_fd[8], BPF_SK_SKB_STREAM_VERDICT); 1288 1289 if (tx_prog_fd >= 0) 1290 bpf_prog_detach2(tx_prog_fd, map_fd[1], BPF_SK_MSG_VERDICT); 1291 1292 for (i = 0; i < 8; i++) { 1293 key = next_key = 0; 1294 bpf_map_update_elem(map_fd[i], &key, &zero, BPF_ANY); 1295 while (bpf_map_get_next_key(map_fd[i], &key, &next_key) == 0) { 1296 bpf_map_update_elem(map_fd[i], &key, &zero, BPF_ANY); 1297 key = next_key; 1298 } 1299 } 1300 1301 close(s1); 1302 close(s2); 1303 close(p1); 1304 close(p2); 1305 close(c1); 1306 close(c2); 1307 return err; 1308 } 1309 1310 static char *test_to_str(int test) 1311 { 1312 switch (test) { 1313 case SENDMSG: 1314 return "sendmsg"; 1315 case SENDPAGE: 1316 return "sendpage"; 1317 } 1318 return "unknown"; 1319 } 1320 1321 static void append_str(char *dst, const char *src, size_t dst_cap) 1322 { 1323 size_t avail = dst_cap - strlen(dst); 1324 1325 if (avail <= 1) /* just zero byte could be written */ 1326 return; 1327 1328 strncat(dst, src, avail - 1); /* strncat() adds + 1 for zero byte */ 1329 } 1330 1331 #define OPTSTRING 60 1332 static void test_options(char *options) 1333 { 1334 char tstr[OPTSTRING]; 1335 1336 memset(options, 0, OPTSTRING); 1337 1338 if (txmsg_pass) 1339 append_str(options, "pass,", OPTSTRING); 1340 if (txmsg_redir) 1341 append_str(options, "redir,", OPTSTRING); 1342 if (txmsg_drop) 1343 append_str(options, "drop,", OPTSTRING); 1344 if (txmsg_apply) { 1345 snprintf(tstr, OPTSTRING, "apply %d,", txmsg_apply); 1346 append_str(options, tstr, OPTSTRING); 1347 } 1348 if (txmsg_cork) { 1349 snprintf(tstr, OPTSTRING, "cork %d,", txmsg_cork); 1350 append_str(options, tstr, OPTSTRING); 1351 } 1352 if (txmsg_start) { 1353 snprintf(tstr, OPTSTRING, "start %d,", txmsg_start); 1354 append_str(options, tstr, OPTSTRING); 1355 } 1356 if (txmsg_end) { 1357 snprintf(tstr, OPTSTRING, "end %d,", txmsg_end); 1358 append_str(options, tstr, OPTSTRING); 1359 } 1360 if (txmsg_start_pop) { 1361 snprintf(tstr, OPTSTRING, "pop (%d,%d),", 1362 txmsg_start_pop, txmsg_start_pop + txmsg_pop); 1363 append_str(options, tstr, OPTSTRING); 1364 } 1365 if (txmsg_ingress) 1366 append_str(options, "ingress,", OPTSTRING); 1367 if (txmsg_redir_skb) 1368 append_str(options, "redir_skb,", OPTSTRING); 1369 if (txmsg_ktls_skb) 1370 append_str(options, "ktls_skb,", OPTSTRING); 1371 if (ktls) 1372 append_str(options, "ktls,", OPTSTRING); 1373 if (peek_flag) 1374 append_str(options, "peek,", OPTSTRING); 1375 } 1376 1377 static int __test_exec(int cgrp, int test, struct sockmap_options *opt) 1378 { 1379 char *options = calloc(OPTSTRING, sizeof(char)); 1380 int err; 1381 1382 if (test == SENDPAGE) 1383 opt->sendpage = true; 1384 else 1385 opt->sendpage = false; 1386 1387 if (txmsg_drop) 1388 opt->drop_expected = true; 1389 else 1390 opt->drop_expected = false; 1391 1392 test_options(options); 1393 1394 if (opt->verbose) { 1395 fprintf(stdout, 1396 " [TEST %i]: (%i, %i, %i, %s, %s): ", 1397 test_cnt, opt->rate, opt->iov_count, opt->iov_length, 1398 test_to_str(test), options); 1399 fflush(stdout); 1400 } 1401 err = run_options(opt, cgrp, test); 1402 if (opt->verbose) 1403 fprintf(stdout, " %s\n", !err ? "PASS" : "FAILED"); 1404 test_cnt++; 1405 !err ? passed++ : failed++; 1406 free(options); 1407 return err; 1408 } 1409 1410 static void test_exec(int cgrp, struct sockmap_options *opt) 1411 { 1412 int type = strcmp(opt->map, BPF_SOCKMAP_FILENAME); 1413 int err; 1414 1415 if (type == 0) { 1416 test_start(); 1417 err = __test_exec(cgrp, SENDMSG, opt); 1418 if (err) 1419 test_fail(); 1420 } else { 1421 test_start(); 1422 err = __test_exec(cgrp, SENDPAGE, opt); 1423 if (err) 1424 test_fail(); 1425 } 1426 } 1427 1428 static void test_send_one(struct sockmap_options *opt, int cgrp) 1429 { 1430 opt->iov_length = 1; 1431 opt->iov_count = 1; 1432 opt->rate = 1; 1433 test_exec(cgrp, opt); 1434 1435 opt->iov_length = 1; 1436 opt->iov_count = 1024; 1437 opt->rate = 1; 1438 test_exec(cgrp, opt); 1439 1440 opt->iov_length = 1024; 1441 opt->iov_count = 1; 1442 opt->rate = 1; 1443 test_exec(cgrp, opt); 1444 1445 } 1446 1447 static void test_send_many(struct sockmap_options *opt, int cgrp) 1448 { 1449 opt->iov_length = 3; 1450 opt->iov_count = 1; 1451 opt->rate = 512; 1452 test_exec(cgrp, opt); 1453 1454 opt->rate = 100; 1455 opt->iov_count = 1; 1456 opt->iov_length = 5; 1457 test_exec(cgrp, opt); 1458 } 1459 1460 static void test_send_large(struct sockmap_options *opt, int cgrp) 1461 { 1462 opt->iov_length = 256; 1463 opt->iov_count = 1024; 1464 opt->rate = 2; 1465 test_exec(cgrp, opt); 1466 } 1467 1468 static void test_send(struct sockmap_options *opt, int cgrp) 1469 { 1470 test_send_one(opt, cgrp); 1471 test_send_many(opt, cgrp); 1472 test_send_large(opt, cgrp); 1473 sched_yield(); 1474 } 1475 1476 static void test_txmsg_pass(int cgrp, struct sockmap_options *opt) 1477 { 1478 /* Test small and large iov_count values with pass/redir/apply/cork */ 1479 txmsg_pass = 1; 1480 test_send(opt, cgrp); 1481 } 1482 1483 static void test_txmsg_redir(int cgrp, struct sockmap_options *opt) 1484 { 1485 txmsg_redir = 1; 1486 test_send(opt, cgrp); 1487 } 1488 1489 static void test_txmsg_redir_wait_sndmem(int cgrp, struct sockmap_options *opt) 1490 { 1491 txmsg_redir = 1; 1492 opt->tx_wait_mem = true; 1493 test_send_large(opt, cgrp); 1494 opt->tx_wait_mem = false; 1495 } 1496 1497 static void test_txmsg_drop(int cgrp, struct sockmap_options *opt) 1498 { 1499 txmsg_drop = 1; 1500 test_send(opt, cgrp); 1501 } 1502 1503 static void test_txmsg_ingress_redir(int cgrp, struct sockmap_options *opt) 1504 { 1505 txmsg_pass = txmsg_drop = 0; 1506 txmsg_ingress = txmsg_redir = 1; 1507 test_send(opt, cgrp); 1508 } 1509 1510 static void test_txmsg_skb(int cgrp, struct sockmap_options *opt) 1511 { 1512 bool data = opt->data_test; 1513 int k = ktls; 1514 1515 opt->data_test = true; 1516 ktls = 1; 1517 1518 txmsg_pass = txmsg_drop = 0; 1519 txmsg_ingress = txmsg_redir = 0; 1520 txmsg_ktls_skb = 1; 1521 txmsg_pass = 1; 1522 1523 /* Using data verification so ensure iov layout is 1524 * expected from test receiver side. e.g. has enough 1525 * bytes to write test code. 1526 */ 1527 opt->iov_length = 100; 1528 opt->iov_count = 1; 1529 opt->rate = 1; 1530 test_exec(cgrp, opt); 1531 1532 txmsg_ktls_skb_drop = 1; 1533 test_exec(cgrp, opt); 1534 1535 txmsg_ktls_skb_drop = 0; 1536 txmsg_ktls_skb_redir = 1; 1537 test_exec(cgrp, opt); 1538 txmsg_ktls_skb_redir = 0; 1539 1540 /* Tests that omit skb_parser */ 1541 txmsg_omit_skb_parser = 1; 1542 ktls = 0; 1543 txmsg_ktls_skb = 0; 1544 test_exec(cgrp, opt); 1545 1546 txmsg_ktls_skb_drop = 1; 1547 test_exec(cgrp, opt); 1548 txmsg_ktls_skb_drop = 0; 1549 1550 txmsg_ktls_skb_redir = 1; 1551 test_exec(cgrp, opt); 1552 1553 ktls = 1; 1554 test_exec(cgrp, opt); 1555 txmsg_omit_skb_parser = 0; 1556 1557 opt->data_test = data; 1558 ktls = k; 1559 } 1560 1561 /* Test cork with hung data. This tests poor usage patterns where 1562 * cork can leave data on the ring if user program is buggy and 1563 * doesn't flush them somehow. They do take some time however 1564 * because they wait for a timeout. Test pass, redir and cork with 1565 * apply logic. Use cork size of 4097 with send_large to avoid 1566 * aligning cork size with send size. 1567 */ 1568 static void test_txmsg_cork_hangs(int cgrp, struct sockmap_options *opt) 1569 { 1570 txmsg_pass = 1; 1571 txmsg_redir = 0; 1572 txmsg_cork = 4097; 1573 txmsg_apply = 4097; 1574 test_send_large(opt, cgrp); 1575 1576 txmsg_pass = 0; 1577 txmsg_redir = 1; 1578 txmsg_apply = 0; 1579 txmsg_cork = 4097; 1580 test_send_large(opt, cgrp); 1581 1582 txmsg_pass = 0; 1583 txmsg_redir = 1; 1584 txmsg_apply = 4097; 1585 txmsg_cork = 4097; 1586 test_send_large(opt, cgrp); 1587 } 1588 1589 static void test_txmsg_pull(int cgrp, struct sockmap_options *opt) 1590 { 1591 /* Test basic start/end */ 1592 txmsg_start = 1; 1593 txmsg_end = 2; 1594 test_send(opt, cgrp); 1595 1596 /* Test >4k pull */ 1597 txmsg_start = 4096; 1598 txmsg_end = 9182; 1599 test_send_large(opt, cgrp); 1600 1601 /* Test pull + redirect */ 1602 txmsg_redir = 0; 1603 txmsg_start = 1; 1604 txmsg_end = 2; 1605 test_send(opt, cgrp); 1606 1607 /* Test pull + cork */ 1608 txmsg_redir = 0; 1609 txmsg_cork = 512; 1610 txmsg_start = 1; 1611 txmsg_end = 2; 1612 test_send_many(opt, cgrp); 1613 1614 /* Test pull + cork + redirect */ 1615 txmsg_redir = 1; 1616 txmsg_cork = 512; 1617 txmsg_start = 1; 1618 txmsg_end = 2; 1619 test_send_many(opt, cgrp); 1620 } 1621 1622 static void test_txmsg_pop(int cgrp, struct sockmap_options *opt) 1623 { 1624 /* Test basic pop */ 1625 txmsg_start_pop = 1; 1626 txmsg_pop = 2; 1627 test_send_many(opt, cgrp); 1628 1629 /* Test pop with >4k */ 1630 txmsg_start_pop = 4096; 1631 txmsg_pop = 4096; 1632 test_send_large(opt, cgrp); 1633 1634 /* Test pop + redirect */ 1635 txmsg_redir = 1; 1636 txmsg_start_pop = 1; 1637 txmsg_pop = 2; 1638 test_send_many(opt, cgrp); 1639 1640 /* Test pop + cork */ 1641 txmsg_redir = 0; 1642 txmsg_cork = 512; 1643 txmsg_start_pop = 1; 1644 txmsg_pop = 2; 1645 test_send_many(opt, cgrp); 1646 1647 /* Test pop + redirect + cork */ 1648 txmsg_redir = 1; 1649 txmsg_cork = 4; 1650 txmsg_start_pop = 1; 1651 txmsg_pop = 2; 1652 test_send_many(opt, cgrp); 1653 } 1654 1655 static void test_txmsg_push(int cgrp, struct sockmap_options *opt) 1656 { 1657 /* Test basic push */ 1658 txmsg_start_push = 1; 1659 txmsg_end_push = 1; 1660 test_send(opt, cgrp); 1661 1662 /* Test push 4kB >4k */ 1663 txmsg_start_push = 4096; 1664 txmsg_end_push = 4096; 1665 test_send_large(opt, cgrp); 1666 1667 /* Test push + redirect */ 1668 txmsg_redir = 1; 1669 txmsg_start_push = 1; 1670 txmsg_end_push = 2; 1671 test_send_many(opt, cgrp); 1672 1673 /* Test push + cork */ 1674 txmsg_redir = 0; 1675 txmsg_cork = 512; 1676 txmsg_start_push = 1; 1677 txmsg_end_push = 2; 1678 test_send_many(opt, cgrp); 1679 } 1680 1681 static void test_txmsg_push_pop(int cgrp, struct sockmap_options *opt) 1682 { 1683 txmsg_start_push = 1; 1684 txmsg_end_push = 10; 1685 txmsg_start_pop = 5; 1686 txmsg_pop = 4; 1687 test_send_large(opt, cgrp); 1688 } 1689 1690 static void test_txmsg_apply(int cgrp, struct sockmap_options *opt) 1691 { 1692 txmsg_pass = 1; 1693 txmsg_redir = 0; 1694 txmsg_ingress = 0; 1695 txmsg_apply = 1; 1696 txmsg_cork = 0; 1697 test_send_one(opt, cgrp); 1698 1699 txmsg_pass = 0; 1700 txmsg_redir = 1; 1701 txmsg_ingress = 0; 1702 txmsg_apply = 1; 1703 txmsg_cork = 0; 1704 test_send_one(opt, cgrp); 1705 1706 txmsg_pass = 0; 1707 txmsg_redir = 1; 1708 txmsg_ingress = 1; 1709 txmsg_apply = 1; 1710 txmsg_cork = 0; 1711 test_send_one(opt, cgrp); 1712 1713 txmsg_pass = 1; 1714 txmsg_redir = 0; 1715 txmsg_ingress = 0; 1716 txmsg_apply = 1024; 1717 txmsg_cork = 0; 1718 test_send_large(opt, cgrp); 1719 1720 txmsg_pass = 0; 1721 txmsg_redir = 1; 1722 txmsg_ingress = 0; 1723 txmsg_apply = 1024; 1724 txmsg_cork = 0; 1725 test_send_large(opt, cgrp); 1726 1727 txmsg_pass = 0; 1728 txmsg_redir = 1; 1729 txmsg_ingress = 1; 1730 txmsg_apply = 1024; 1731 txmsg_cork = 0; 1732 test_send_large(opt, cgrp); 1733 } 1734 1735 static void test_txmsg_cork(int cgrp, struct sockmap_options *opt) 1736 { 1737 txmsg_pass = 1; 1738 txmsg_redir = 0; 1739 txmsg_apply = 0; 1740 txmsg_cork = 1; 1741 test_send(opt, cgrp); 1742 1743 txmsg_pass = 1; 1744 txmsg_redir = 0; 1745 txmsg_apply = 1; 1746 txmsg_cork = 1; 1747 test_send(opt, cgrp); 1748 } 1749 1750 static void test_txmsg_ingress_parser(int cgrp, struct sockmap_options *opt) 1751 { 1752 txmsg_pass = 1; 1753 skb_use_parser = 512; 1754 if (ktls == 1) 1755 skb_use_parser = 570; 1756 opt->iov_length = 256; 1757 opt->iov_count = 1; 1758 opt->rate = 2; 1759 test_exec(cgrp, opt); 1760 } 1761 1762 static void test_txmsg_ingress_parser2(int cgrp, struct sockmap_options *opt) 1763 { 1764 if (ktls == 1) 1765 return; 1766 skb_use_parser = 10; 1767 opt->iov_length = 20; 1768 opt->iov_count = 1; 1769 opt->rate = 1; 1770 opt->check_recved_len = true; 1771 test_exec(cgrp, opt); 1772 opt->check_recved_len = false; 1773 } 1774 1775 char *map_names[] = { 1776 "sock_map", 1777 "sock_map_txmsg", 1778 "sock_map_redir", 1779 "sock_apply_bytes", 1780 "sock_cork_bytes", 1781 "sock_bytes", 1782 "sock_redir_flags", 1783 "sock_skb_opts", 1784 "tls_sock_map", 1785 }; 1786 1787 int prog_attach_type[] = { 1788 BPF_SK_SKB_STREAM_PARSER, 1789 BPF_SK_SKB_STREAM_VERDICT, 1790 BPF_SK_SKB_STREAM_VERDICT, 1791 BPF_CGROUP_SOCK_OPS, 1792 BPF_SK_MSG_VERDICT, 1793 BPF_SK_MSG_VERDICT, 1794 BPF_SK_MSG_VERDICT, 1795 BPF_SK_MSG_VERDICT, 1796 BPF_SK_MSG_VERDICT, 1797 }; 1798 1799 int prog_type[] = { 1800 BPF_PROG_TYPE_SK_SKB, 1801 BPF_PROG_TYPE_SK_SKB, 1802 BPF_PROG_TYPE_SK_SKB, 1803 BPF_PROG_TYPE_SOCK_OPS, 1804 BPF_PROG_TYPE_SK_MSG, 1805 BPF_PROG_TYPE_SK_MSG, 1806 BPF_PROG_TYPE_SK_MSG, 1807 BPF_PROG_TYPE_SK_MSG, 1808 BPF_PROG_TYPE_SK_MSG, 1809 }; 1810 1811 static int populate_progs(char *bpf_file) 1812 { 1813 struct bpf_program *prog; 1814 struct bpf_object *obj; 1815 int i = 0; 1816 long err; 1817 1818 obj = bpf_object__open(bpf_file); 1819 err = libbpf_get_error(obj); 1820 if (err) { 1821 char err_buf[256]; 1822 1823 libbpf_strerror(err, err_buf, sizeof(err_buf)); 1824 printf("Unable to load eBPF objects in file '%s' : %s\n", 1825 bpf_file, err_buf); 1826 return -1; 1827 } 1828 1829 bpf_object__for_each_program(prog, obj) { 1830 bpf_program__set_type(prog, prog_type[i]); 1831 bpf_program__set_expected_attach_type(prog, 1832 prog_attach_type[i]); 1833 i++; 1834 } 1835 1836 i = bpf_object__load(obj); 1837 i = 0; 1838 bpf_object__for_each_program(prog, obj) { 1839 prog_fd[i] = bpf_program__fd(prog); 1840 i++; 1841 } 1842 1843 for (i = 0; i < ARRAY_SIZE(map_fd); i++) { 1844 maps[i] = bpf_object__find_map_by_name(obj, map_names[i]); 1845 map_fd[i] = bpf_map__fd(maps[i]); 1846 if (map_fd[i] < 0) { 1847 fprintf(stderr, "load_bpf_file: (%i) %s\n", 1848 map_fd[i], strerror(errno)); 1849 return -1; 1850 } 1851 } 1852 1853 return 0; 1854 } 1855 1856 struct _test test[] = { 1857 {"txmsg test passthrough", test_txmsg_pass}, 1858 {"txmsg test redirect", test_txmsg_redir}, 1859 {"txmsg test redirect wait send mem", test_txmsg_redir_wait_sndmem}, 1860 {"txmsg test drop", test_txmsg_drop}, 1861 {"txmsg test ingress redirect", test_txmsg_ingress_redir}, 1862 {"txmsg test skb", test_txmsg_skb}, 1863 {"txmsg test apply", test_txmsg_apply}, 1864 {"txmsg test cork", test_txmsg_cork}, 1865 {"txmsg test hanging corks", test_txmsg_cork_hangs}, 1866 {"txmsg test push_data", test_txmsg_push}, 1867 {"txmsg test pull-data", test_txmsg_pull}, 1868 {"txmsg test pop-data", test_txmsg_pop}, 1869 {"txmsg test push/pop data", test_txmsg_push_pop}, 1870 {"txmsg test ingress parser", test_txmsg_ingress_parser}, 1871 {"txmsg test ingress parser2", test_txmsg_ingress_parser2}, 1872 }; 1873 1874 static int check_whitelist(struct _test *t, struct sockmap_options *opt) 1875 { 1876 char *entry, *ptr; 1877 1878 if (!opt->whitelist) 1879 return 0; 1880 ptr = strdup(opt->whitelist); 1881 if (!ptr) 1882 return -ENOMEM; 1883 entry = strtok(ptr, ","); 1884 while (entry) { 1885 if ((opt->prepend && strstr(opt->prepend, entry) != 0) || 1886 strstr(opt->map, entry) != 0 || 1887 strstr(t->title, entry) != 0) 1888 return 0; 1889 entry = strtok(NULL, ","); 1890 } 1891 return -EINVAL; 1892 } 1893 1894 static int check_blacklist(struct _test *t, struct sockmap_options *opt) 1895 { 1896 char *entry, *ptr; 1897 1898 if (!opt->blacklist) 1899 return -EINVAL; 1900 ptr = strdup(opt->blacklist); 1901 if (!ptr) 1902 return -ENOMEM; 1903 entry = strtok(ptr, ","); 1904 while (entry) { 1905 if ((opt->prepend && strstr(opt->prepend, entry) != 0) || 1906 strstr(opt->map, entry) != 0 || 1907 strstr(t->title, entry) != 0) 1908 return 0; 1909 entry = strtok(NULL, ","); 1910 } 1911 return -EINVAL; 1912 } 1913 1914 static int __test_selftests(int cg_fd, struct sockmap_options *opt) 1915 { 1916 int i, err; 1917 1918 err = populate_progs(opt->map); 1919 if (err < 0) { 1920 fprintf(stderr, "ERROR: (%i) load bpf failed\n", err); 1921 return err; 1922 } 1923 1924 /* Tests basic commands and APIs */ 1925 for (i = 0; i < ARRAY_SIZE(test); i++) { 1926 struct _test t = test[i]; 1927 1928 if (check_whitelist(&t, opt) != 0) 1929 continue; 1930 if (check_blacklist(&t, opt) == 0) 1931 continue; 1932 1933 test_start_subtest(&t, opt); 1934 t.tester(cg_fd, opt); 1935 test_end_subtest(); 1936 } 1937 1938 return err; 1939 } 1940 1941 static void test_selftests_sockmap(int cg_fd, struct sockmap_options *opt) 1942 { 1943 opt->map = BPF_SOCKMAP_FILENAME; 1944 __test_selftests(cg_fd, opt); 1945 } 1946 1947 static void test_selftests_sockhash(int cg_fd, struct sockmap_options *opt) 1948 { 1949 opt->map = BPF_SOCKHASH_FILENAME; 1950 __test_selftests(cg_fd, opt); 1951 } 1952 1953 static void test_selftests_ktls(int cg_fd, struct sockmap_options *opt) 1954 { 1955 opt->map = BPF_SOCKHASH_FILENAME; 1956 opt->prepend = "ktls"; 1957 ktls = 1; 1958 __test_selftests(cg_fd, opt); 1959 ktls = 0; 1960 } 1961 1962 static int test_selftest(int cg_fd, struct sockmap_options *opt) 1963 { 1964 1965 test_selftests_sockmap(cg_fd, opt); 1966 test_selftests_sockhash(cg_fd, opt); 1967 test_selftests_ktls(cg_fd, opt); 1968 test_print_results(); 1969 return 0; 1970 } 1971 1972 int main(int argc, char **argv) 1973 { 1974 int iov_count = 1, length = 1024, rate = 1; 1975 struct sockmap_options options = {0}; 1976 int opt, longindex, err, cg_fd = 0; 1977 char *bpf_file = BPF_SOCKMAP_FILENAME; 1978 int test = SELFTESTS; 1979 bool cg_created = 0; 1980 1981 while ((opt = getopt_long(argc, argv, ":dhv:c:r:i:l:t:p:q:n:b:", 1982 long_options, &longindex)) != -1) { 1983 switch (opt) { 1984 case 's': 1985 txmsg_start = atoi(optarg); 1986 break; 1987 case 'e': 1988 txmsg_end = atoi(optarg); 1989 break; 1990 case 'p': 1991 txmsg_start_push = atoi(optarg); 1992 break; 1993 case 'q': 1994 txmsg_end_push = atoi(optarg); 1995 break; 1996 case 'w': 1997 txmsg_start_pop = atoi(optarg); 1998 break; 1999 case 'x': 2000 txmsg_pop = atoi(optarg); 2001 break; 2002 case 'a': 2003 txmsg_apply = atoi(optarg); 2004 break; 2005 case 'k': 2006 txmsg_cork = atoi(optarg); 2007 break; 2008 case 'c': 2009 cg_fd = open(optarg, O_DIRECTORY, O_RDONLY); 2010 if (cg_fd < 0) { 2011 fprintf(stderr, 2012 "ERROR: (%i) open cg path failed: %s\n", 2013 cg_fd, optarg); 2014 return cg_fd; 2015 } 2016 break; 2017 case 'r': 2018 rate = atoi(optarg); 2019 break; 2020 case 'v': 2021 options.verbose = 1; 2022 if (optarg) 2023 options.verbose = atoi(optarg); 2024 break; 2025 case 'i': 2026 iov_count = atoi(optarg); 2027 break; 2028 case 'l': 2029 length = atoi(optarg); 2030 break; 2031 case 'd': 2032 options.data_test = true; 2033 break; 2034 case 't': 2035 if (strcmp(optarg, "ping") == 0) { 2036 test = PING_PONG; 2037 } else if (strcmp(optarg, "sendmsg") == 0) { 2038 test = SENDMSG; 2039 } else if (strcmp(optarg, "base") == 0) { 2040 test = BASE; 2041 } else if (strcmp(optarg, "base_sendpage") == 0) { 2042 test = BASE_SENDPAGE; 2043 } else if (strcmp(optarg, "sendpage") == 0) { 2044 test = SENDPAGE; 2045 } else { 2046 usage(argv); 2047 return -1; 2048 } 2049 break; 2050 case 'n': 2051 options.whitelist = strdup(optarg); 2052 if (!options.whitelist) 2053 return -ENOMEM; 2054 break; 2055 case 'b': 2056 options.blacklist = strdup(optarg); 2057 if (!options.blacklist) 2058 return -ENOMEM; 2059 case 0: 2060 break; 2061 case 'h': 2062 default: 2063 usage(argv); 2064 return -1; 2065 } 2066 } 2067 2068 if (!cg_fd) { 2069 cg_fd = cgroup_setup_and_join(CG_PATH); 2070 if (cg_fd < 0) 2071 return cg_fd; 2072 cg_created = 1; 2073 } 2074 2075 /* Use libbpf 1.0 API mode */ 2076 libbpf_set_strict_mode(LIBBPF_STRICT_ALL); 2077 2078 if (test == SELFTESTS) { 2079 err = test_selftest(cg_fd, &options); 2080 goto out; 2081 } 2082 2083 err = populate_progs(bpf_file); 2084 if (err) { 2085 fprintf(stderr, "populate program: (%s) %s\n", 2086 bpf_file, strerror(errno)); 2087 return 1; 2088 } 2089 running = 1; 2090 2091 /* catch SIGINT */ 2092 signal(SIGINT, running_handler); 2093 2094 options.iov_count = iov_count; 2095 options.iov_length = length; 2096 options.rate = rate; 2097 2098 err = run_options(&options, cg_fd, test); 2099 out: 2100 if (options.whitelist) 2101 free(options.whitelist); 2102 if (options.blacklist) 2103 free(options.blacklist); 2104 close(cg_fd); 2105 if (cg_created) 2106 cleanup_cgroup_environment(); 2107 return err; 2108 } 2109 2110 void running_handler(int a) 2111 { 2112 running = 0; 2113 } 2114