1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) 2 /* Copyright (C) 2017-2018 Netronome Systems, Inc. */ 3 4 #ifndef _GNU_SOURCE 5 #define _GNU_SOURCE 6 #endif 7 #include <errno.h> 8 #include <fcntl.h> 9 #include <signal.h> 10 #include <stdarg.h> 11 #include <stdio.h> 12 #include <stdlib.h> 13 #include <string.h> 14 #include <time.h> 15 #include <unistd.h> 16 #include <net/if.h> 17 #include <sys/ioctl.h> 18 #include <sys/types.h> 19 #include <sys/stat.h> 20 #include <sys/syscall.h> 21 #include <dirent.h> 22 23 #include <linux/err.h> 24 #include <linux/perf_event.h> 25 #include <linux/sizes.h> 26 #include <linux/keyctl.h> 27 28 #include <bpf/bpf.h> 29 #include <bpf/btf.h> 30 #include <bpf/hashmap.h> 31 #include <bpf/libbpf.h> 32 #include <bpf/libbpf_internal.h> 33 #include <bpf/skel_internal.h> 34 35 #include "cfg.h" 36 #include "main.h" 37 #include "xlated_dumper.h" 38 39 #define BPF_METADATA_PREFIX "bpf_metadata_" 40 #define BPF_METADATA_PREFIX_LEN (sizeof(BPF_METADATA_PREFIX) - 1) 41 42 enum dump_mode { 43 DUMP_JITED, 44 DUMP_XLATED, 45 }; 46 47 static const bool attach_types[] = { 48 [BPF_SK_SKB_STREAM_PARSER] = true, 49 [BPF_SK_SKB_STREAM_VERDICT] = true, 50 [BPF_SK_SKB_VERDICT] = true, 51 [BPF_SK_MSG_VERDICT] = true, 52 [BPF_FLOW_DISSECTOR] = true, 53 [__MAX_BPF_ATTACH_TYPE] = false, 54 }; 55 56 /* Textual representations traditionally used by the program and kept around 57 * for the sake of backwards compatibility. 58 */ 59 static const char * const attach_type_strings[] = { 60 [BPF_SK_SKB_STREAM_PARSER] = "stream_parser", 61 [BPF_SK_SKB_STREAM_VERDICT] = "stream_verdict", 62 [BPF_SK_SKB_VERDICT] = "skb_verdict", 63 [BPF_SK_MSG_VERDICT] = "msg_verdict", 64 [__MAX_BPF_ATTACH_TYPE] = NULL, 65 }; 66 67 static struct hashmap *prog_table; 68 69 static enum bpf_attach_type parse_attach_type(const char *str) 70 { 71 enum bpf_attach_type type; 72 73 for (type = 0; type < __MAX_BPF_ATTACH_TYPE; type++) { 74 if (attach_types[type]) { 75 const char *attach_type_str; 76 77 attach_type_str = libbpf_bpf_attach_type_str(type); 78 if (!strcmp(str, attach_type_str)) 79 return type; 80 } 81 82 if (attach_type_strings[type] && 83 is_prefix(str, attach_type_strings[type])) 84 return type; 85 } 86 87 return __MAX_BPF_ATTACH_TYPE; 88 } 89 90 static int prep_prog_info(struct bpf_prog_info *const info, enum dump_mode mode, 91 void **info_data, size_t *const info_data_sz) 92 { 93 struct bpf_prog_info holder = {}; 94 size_t needed = 0; 95 void *ptr; 96 97 if (mode == DUMP_JITED) { 98 holder.jited_prog_len = info->jited_prog_len; 99 needed += info->jited_prog_len; 100 } else { 101 holder.xlated_prog_len = info->xlated_prog_len; 102 needed += info->xlated_prog_len; 103 } 104 105 holder.nr_jited_ksyms = info->nr_jited_ksyms; 106 needed += info->nr_jited_ksyms * sizeof(__u64); 107 108 holder.nr_jited_func_lens = info->nr_jited_func_lens; 109 needed += info->nr_jited_func_lens * sizeof(__u32); 110 111 holder.nr_func_info = info->nr_func_info; 112 holder.func_info_rec_size = info->func_info_rec_size; 113 needed += info->nr_func_info * info->func_info_rec_size; 114 115 holder.nr_line_info = info->nr_line_info; 116 holder.line_info_rec_size = info->line_info_rec_size; 117 needed += info->nr_line_info * info->line_info_rec_size; 118 119 holder.nr_jited_line_info = info->nr_jited_line_info; 120 holder.jited_line_info_rec_size = info->jited_line_info_rec_size; 121 needed += info->nr_jited_line_info * info->jited_line_info_rec_size; 122 123 if (needed > *info_data_sz) { 124 ptr = realloc(*info_data, needed); 125 if (!ptr) 126 return -1; 127 128 *info_data = ptr; 129 *info_data_sz = needed; 130 } 131 ptr = *info_data; 132 133 if (mode == DUMP_JITED) { 134 holder.jited_prog_insns = ptr_to_u64(ptr); 135 ptr += holder.jited_prog_len; 136 } else { 137 holder.xlated_prog_insns = ptr_to_u64(ptr); 138 ptr += holder.xlated_prog_len; 139 } 140 141 holder.jited_ksyms = ptr_to_u64(ptr); 142 ptr += holder.nr_jited_ksyms * sizeof(__u64); 143 144 holder.jited_func_lens = ptr_to_u64(ptr); 145 ptr += holder.nr_jited_func_lens * sizeof(__u32); 146 147 holder.func_info = ptr_to_u64(ptr); 148 ptr += holder.nr_func_info * holder.func_info_rec_size; 149 150 holder.line_info = ptr_to_u64(ptr); 151 ptr += holder.nr_line_info * holder.line_info_rec_size; 152 153 holder.jited_line_info = ptr_to_u64(ptr); 154 ptr += holder.nr_jited_line_info * holder.jited_line_info_rec_size; 155 156 *info = holder; 157 return 0; 158 } 159 160 static void print_boot_time(__u64 nsecs, char *buf, unsigned int size) 161 { 162 struct timespec real_time_ts, boot_time_ts; 163 time_t wallclock_secs; 164 struct tm load_tm; 165 166 buf[--size] = '\0'; 167 168 if (clock_gettime(CLOCK_REALTIME, &real_time_ts) || 169 clock_gettime(CLOCK_BOOTTIME, &boot_time_ts)) { 170 perror("Can't read clocks"); 171 snprintf(buf, size, "%llu", nsecs / 1000000000); 172 return; 173 } 174 175 wallclock_secs = (real_time_ts.tv_sec - boot_time_ts.tv_sec) + 176 (real_time_ts.tv_nsec - boot_time_ts.tv_nsec + nsecs) / 177 1000000000; 178 179 180 if (!localtime_r(&wallclock_secs, &load_tm)) { 181 snprintf(buf, size, "%llu", nsecs / 1000000000); 182 return; 183 } 184 185 if (json_output) 186 strftime(buf, size, "%s", &load_tm); 187 else 188 strftime(buf, size, "%FT%T%z", &load_tm); 189 } 190 191 static void show_prog_maps(int fd, __u32 num_maps) 192 { 193 struct bpf_prog_info info = {}; 194 __u32 len = sizeof(info); 195 __u32 map_ids[num_maps]; 196 unsigned int i; 197 int err; 198 199 info.nr_map_ids = num_maps; 200 info.map_ids = ptr_to_u64(map_ids); 201 202 err = bpf_prog_get_info_by_fd(fd, &info, &len); 203 if (err || !info.nr_map_ids) 204 return; 205 206 if (json_output) { 207 jsonw_name(json_wtr, "map_ids"); 208 jsonw_start_array(json_wtr); 209 for (i = 0; i < info.nr_map_ids; i++) 210 jsonw_uint(json_wtr, map_ids[i]); 211 jsonw_end_array(json_wtr); 212 } else { 213 printf(" map_ids "); 214 for (i = 0; i < info.nr_map_ids; i++) 215 printf("%u%s", map_ids[i], 216 i == info.nr_map_ids - 1 ? "" : ","); 217 } 218 } 219 220 static void *find_metadata(int prog_fd, struct bpf_map_info *map_info) 221 { 222 struct bpf_prog_info prog_info; 223 __u32 prog_info_len; 224 __u32 map_info_len; 225 void *value = NULL; 226 __u32 *map_ids; 227 int nr_maps; 228 int key = 0; 229 int map_fd; 230 int ret; 231 __u32 i; 232 233 memset(&prog_info, 0, sizeof(prog_info)); 234 prog_info_len = sizeof(prog_info); 235 ret = bpf_prog_get_info_by_fd(prog_fd, &prog_info, &prog_info_len); 236 if (ret) 237 return NULL; 238 239 if (!prog_info.nr_map_ids) 240 return NULL; 241 242 map_ids = calloc(prog_info.nr_map_ids, sizeof(__u32)); 243 if (!map_ids) 244 return NULL; 245 246 nr_maps = prog_info.nr_map_ids; 247 memset(&prog_info, 0, sizeof(prog_info)); 248 prog_info.nr_map_ids = nr_maps; 249 prog_info.map_ids = ptr_to_u64(map_ids); 250 prog_info_len = sizeof(prog_info); 251 252 ret = bpf_prog_get_info_by_fd(prog_fd, &prog_info, &prog_info_len); 253 if (ret) 254 goto free_map_ids; 255 256 for (i = 0; i < prog_info.nr_map_ids; i++) { 257 map_fd = bpf_map_get_fd_by_id(map_ids[i]); 258 if (map_fd < 0) 259 goto free_map_ids; 260 261 memset(map_info, 0, sizeof(*map_info)); 262 map_info_len = sizeof(*map_info); 263 ret = bpf_map_get_info_by_fd(map_fd, map_info, &map_info_len); 264 if (ret < 0) { 265 close(map_fd); 266 goto free_map_ids; 267 } 268 269 if (map_info->type != BPF_MAP_TYPE_ARRAY || 270 map_info->key_size != sizeof(int) || 271 map_info->max_entries != 1 || 272 !map_info->btf_value_type_id || 273 !strstr(map_info->name, ".rodata")) { 274 close(map_fd); 275 continue; 276 } 277 278 value = malloc(map_info->value_size); 279 if (!value) { 280 close(map_fd); 281 goto free_map_ids; 282 } 283 284 if (bpf_map_lookup_elem(map_fd, &key, value)) { 285 close(map_fd); 286 free(value); 287 value = NULL; 288 goto free_map_ids; 289 } 290 291 close(map_fd); 292 break; 293 } 294 295 free_map_ids: 296 free(map_ids); 297 return value; 298 } 299 300 static bool has_metadata_prefix(const char *s) 301 { 302 return strncmp(s, BPF_METADATA_PREFIX, BPF_METADATA_PREFIX_LEN) == 0; 303 } 304 305 static void show_prog_metadata(int fd, __u32 num_maps) 306 { 307 const struct btf_type *t_datasec, *t_var; 308 struct bpf_map_info map_info; 309 struct btf_var_secinfo *vsi; 310 bool printed_header = false; 311 unsigned int i, vlen; 312 void *value = NULL; 313 const char *name; 314 struct btf *btf; 315 int err; 316 317 if (!num_maps) 318 return; 319 320 memset(&map_info, 0, sizeof(map_info)); 321 value = find_metadata(fd, &map_info); 322 if (!value) 323 return; 324 325 btf = btf__load_from_kernel_by_id(map_info.btf_id); 326 if (!btf) 327 goto out_free; 328 329 t_datasec = btf__type_by_id(btf, map_info.btf_value_type_id); 330 if (!btf_is_datasec(t_datasec)) 331 goto out_free; 332 333 vlen = btf_vlen(t_datasec); 334 vsi = btf_var_secinfos(t_datasec); 335 336 /* We don't proceed to check the kinds of the elements of the DATASEC. 337 * The verifier enforces them to be BTF_KIND_VAR. 338 */ 339 340 if (json_output) { 341 struct btf_dumper d = { 342 .btf = btf, 343 .jw = json_wtr, 344 .is_plain_text = false, 345 }; 346 347 for (i = 0; i < vlen; i++, vsi++) { 348 t_var = btf__type_by_id(btf, vsi->type); 349 name = btf__name_by_offset(btf, t_var->name_off); 350 351 if (!has_metadata_prefix(name)) 352 continue; 353 354 if (!printed_header) { 355 jsonw_name(json_wtr, "metadata"); 356 jsonw_start_object(json_wtr); 357 printed_header = true; 358 } 359 360 jsonw_name(json_wtr, name + BPF_METADATA_PREFIX_LEN); 361 err = btf_dumper_type(&d, t_var->type, value + vsi->offset); 362 if (err) { 363 p_err("btf dump failed: %d", err); 364 break; 365 } 366 } 367 if (printed_header) 368 jsonw_end_object(json_wtr); 369 } else { 370 json_writer_t *btf_wtr; 371 struct btf_dumper d = { 372 .btf = btf, 373 .is_plain_text = true, 374 }; 375 376 for (i = 0; i < vlen; i++, vsi++) { 377 t_var = btf__type_by_id(btf, vsi->type); 378 name = btf__name_by_offset(btf, t_var->name_off); 379 380 if (!has_metadata_prefix(name)) 381 continue; 382 383 if (!printed_header) { 384 printf("\tmetadata:"); 385 386 btf_wtr = jsonw_new(stdout); 387 if (!btf_wtr) { 388 p_err("jsonw alloc failed"); 389 goto out_free; 390 } 391 d.jw = btf_wtr, 392 393 printed_header = true; 394 } 395 396 printf("\n\t\t%s = ", name + BPF_METADATA_PREFIX_LEN); 397 398 jsonw_reset(btf_wtr); 399 err = btf_dumper_type(&d, t_var->type, value + vsi->offset); 400 if (err) { 401 p_err("btf dump failed: %d", err); 402 break; 403 } 404 } 405 if (printed_header) 406 jsonw_destroy(&btf_wtr); 407 } 408 409 out_free: 410 btf__free(btf); 411 free(value); 412 } 413 414 static void print_prog_header_json(struct bpf_prog_info *info, int fd) 415 { 416 const char *prog_type_str; 417 char prog_name[MAX_PROG_FULL_NAME]; 418 419 jsonw_uint_field(json_wtr, "id", info->id); 420 prog_type_str = libbpf_bpf_prog_type_str(info->type); 421 422 if (prog_type_str) 423 jsonw_string_field(json_wtr, "type", prog_type_str); 424 else 425 jsonw_uint_field(json_wtr, "type", info->type); 426 427 if (*info->name) { 428 get_prog_full_name(info, fd, prog_name, sizeof(prog_name)); 429 jsonw_string_field(json_wtr, "name", prog_name); 430 } 431 432 jsonw_name(json_wtr, "tag"); 433 jsonw_printf(json_wtr, "\"" BPF_TAG_FMT "\"", 434 info->tag[0], info->tag[1], info->tag[2], info->tag[3], 435 info->tag[4], info->tag[5], info->tag[6], info->tag[7]); 436 437 jsonw_bool_field(json_wtr, "gpl_compatible", info->gpl_compatible); 438 if (info->run_time_ns) { 439 jsonw_uint_field(json_wtr, "run_time_ns", info->run_time_ns); 440 jsonw_uint_field(json_wtr, "run_cnt", info->run_cnt); 441 } 442 if (info->recursion_misses) 443 jsonw_uint_field(json_wtr, "recursion_misses", info->recursion_misses); 444 } 445 446 static void print_prog_json(struct bpf_prog_info *info, int fd, bool orphaned) 447 { 448 char *memlock; 449 450 jsonw_start_object(json_wtr); 451 print_prog_header_json(info, fd); 452 print_dev_json(info->ifindex, info->netns_dev, info->netns_ino); 453 454 if (info->load_time) { 455 char buf[32]; 456 457 print_boot_time(info->load_time, buf, sizeof(buf)); 458 459 /* Piggy back on load_time, since 0 uid is a valid one */ 460 jsonw_name(json_wtr, "loaded_at"); 461 jsonw_printf(json_wtr, "%s", buf); 462 jsonw_uint_field(json_wtr, "uid", info->created_by_uid); 463 } 464 465 jsonw_bool_field(json_wtr, "orphaned", orphaned); 466 jsonw_uint_field(json_wtr, "bytes_xlated", info->xlated_prog_len); 467 468 if (info->jited_prog_len) { 469 jsonw_bool_field(json_wtr, "jited", true); 470 jsonw_uint_field(json_wtr, "bytes_jited", info->jited_prog_len); 471 } else { 472 jsonw_bool_field(json_wtr, "jited", false); 473 } 474 475 memlock = get_fdinfo(fd, "memlock"); 476 if (memlock) 477 jsonw_int_field(json_wtr, "bytes_memlock", atoll(memlock)); 478 free(memlock); 479 480 if (info->nr_map_ids) 481 show_prog_maps(fd, info->nr_map_ids); 482 483 if (info->btf_id) 484 jsonw_int_field(json_wtr, "btf_id", info->btf_id); 485 486 if (!hashmap__empty(prog_table)) { 487 struct hashmap_entry *entry; 488 489 jsonw_name(json_wtr, "pinned"); 490 jsonw_start_array(json_wtr); 491 hashmap__for_each_key_entry(prog_table, entry, info->id) 492 jsonw_string(json_wtr, entry->pvalue); 493 jsonw_end_array(json_wtr); 494 } 495 496 emit_obj_refs_json(refs_table, info->id, json_wtr); 497 498 show_prog_metadata(fd, info->nr_map_ids); 499 500 jsonw_end_object(json_wtr); 501 } 502 503 static void print_prog_header_plain(struct bpf_prog_info *info, int fd) 504 { 505 const char *prog_type_str; 506 char prog_name[MAX_PROG_FULL_NAME]; 507 508 printf("%u: ", info->id); 509 prog_type_str = libbpf_bpf_prog_type_str(info->type); 510 if (prog_type_str) 511 printf("%s ", prog_type_str); 512 else 513 printf("type %u ", info->type); 514 515 if (*info->name) { 516 get_prog_full_name(info, fd, prog_name, sizeof(prog_name)); 517 printf("name %s ", prog_name); 518 } 519 520 printf("tag "); 521 fprint_hex(stdout, info->tag, BPF_TAG_SIZE, ""); 522 print_dev_plain(info->ifindex, info->netns_dev, info->netns_ino); 523 printf("%s", info->gpl_compatible ? " gpl" : ""); 524 if (info->run_time_ns) 525 printf(" run_time_ns %llu run_cnt %llu", 526 info->run_time_ns, info->run_cnt); 527 if (info->recursion_misses) 528 printf(" recursion_misses %llu", info->recursion_misses); 529 printf("\n"); 530 } 531 532 static void print_prog_plain(struct bpf_prog_info *info, int fd, bool orphaned) 533 { 534 char *memlock; 535 536 print_prog_header_plain(info, fd); 537 538 if (info->load_time) { 539 char buf[32]; 540 541 print_boot_time(info->load_time, buf, sizeof(buf)); 542 543 /* Piggy back on load_time, since 0 uid is a valid one */ 544 printf("\tloaded_at %s uid %u\n", buf, info->created_by_uid); 545 } 546 547 printf("\txlated %uB", info->xlated_prog_len); 548 549 if (info->jited_prog_len) 550 printf(" jited %uB", info->jited_prog_len); 551 else 552 printf(" not jited"); 553 554 memlock = get_fdinfo(fd, "memlock"); 555 if (memlock) 556 printf(" memlock %sB", memlock); 557 free(memlock); 558 559 if (orphaned) 560 printf(" orphaned"); 561 562 if (info->nr_map_ids) 563 show_prog_maps(fd, info->nr_map_ids); 564 565 if (!hashmap__empty(prog_table)) { 566 struct hashmap_entry *entry; 567 568 hashmap__for_each_key_entry(prog_table, entry, info->id) 569 printf("\n\tpinned %s", (char *)entry->pvalue); 570 } 571 572 if (info->btf_id) 573 printf("\n\tbtf_id %u", info->btf_id); 574 575 emit_obj_refs_plain(refs_table, info->id, "\n\tpids "); 576 577 printf("\n"); 578 579 show_prog_metadata(fd, info->nr_map_ids); 580 } 581 582 static int show_prog(int fd) 583 { 584 struct bpf_prog_info info = {}; 585 __u32 len = sizeof(info); 586 int err; 587 588 err = bpf_prog_get_info_by_fd(fd, &info, &len); 589 if (err && err != -ENODEV) { 590 p_err("can't get prog info: %s", strerror(errno)); 591 return -1; 592 } 593 594 if (json_output) 595 print_prog_json(&info, fd, err == -ENODEV); 596 else 597 print_prog_plain(&info, fd, err == -ENODEV); 598 599 return 0; 600 } 601 602 static int do_show_subset(int argc, char **argv) 603 { 604 int *fds = NULL; 605 int nb_fds, i; 606 int err = -1; 607 608 fds = malloc(sizeof(int)); 609 if (!fds) { 610 p_err("mem alloc failed"); 611 return -1; 612 } 613 nb_fds = prog_parse_fds(&argc, &argv, &fds); 614 if (nb_fds < 1) 615 goto exit_free; 616 617 if (json_output && nb_fds > 1) 618 jsonw_start_array(json_wtr); /* root array */ 619 for (i = 0; i < nb_fds; i++) { 620 err = show_prog(fds[i]); 621 if (err) { 622 for (; i < nb_fds; i++) 623 close(fds[i]); 624 break; 625 } 626 close(fds[i]); 627 } 628 if (json_output && nb_fds > 1) 629 jsonw_end_array(json_wtr); /* root array */ 630 631 exit_free: 632 free(fds); 633 return err; 634 } 635 636 static int do_show(int argc, char **argv) 637 { 638 __u32 id = 0; 639 int err; 640 int fd; 641 642 if (show_pinned) { 643 prog_table = hashmap__new(hash_fn_for_key_as_id, 644 equal_fn_for_key_as_id, NULL); 645 if (IS_ERR(prog_table)) { 646 p_err("failed to create hashmap for pinned paths"); 647 return -1; 648 } 649 build_pinned_obj_table(prog_table, BPF_OBJ_PROG); 650 } 651 build_obj_refs_table(&refs_table, BPF_OBJ_PROG); 652 653 if (argc == 2) 654 return do_show_subset(argc, argv); 655 656 if (argc) 657 return BAD_ARG(); 658 659 if (json_output) 660 jsonw_start_array(json_wtr); 661 while (true) { 662 err = bpf_prog_get_next_id(id, &id); 663 if (err) { 664 if (errno == ENOENT) { 665 err = 0; 666 break; 667 } 668 p_err("can't get next program: %s%s", strerror(errno), 669 errno == EINVAL ? " -- kernel too old?" : ""); 670 err = -1; 671 break; 672 } 673 674 fd = bpf_prog_get_fd_by_id(id); 675 if (fd < 0) { 676 if (errno == ENOENT) 677 continue; 678 p_err("can't get prog by id (%u): %s", 679 id, strerror(errno)); 680 err = -1; 681 break; 682 } 683 684 err = show_prog(fd); 685 close(fd); 686 if (err) 687 break; 688 } 689 690 if (json_output) 691 jsonw_end_array(json_wtr); 692 693 delete_obj_refs_table(refs_table); 694 695 if (show_pinned) 696 delete_pinned_obj_table(prog_table); 697 698 return err; 699 } 700 701 static int 702 prog_dump(struct bpf_prog_info *info, enum dump_mode mode, 703 char *filepath, bool opcodes, bool visual, bool linum) 704 { 705 struct bpf_prog_linfo *prog_linfo = NULL; 706 const char *disasm_opt = NULL; 707 struct dump_data dd = {}; 708 void *func_info = NULL; 709 struct btf *btf = NULL; 710 char func_sig[1024]; 711 unsigned char *buf; 712 __u32 member_len; 713 int fd, err = -1; 714 ssize_t n; 715 716 if (mode == DUMP_JITED) { 717 if (info->jited_prog_len == 0 || !info->jited_prog_insns) { 718 p_err("error retrieving jit dump: no instructions returned or kernel.kptr_restrict set?"); 719 return -1; 720 } 721 buf = u64_to_ptr(info->jited_prog_insns); 722 member_len = info->jited_prog_len; 723 } else { /* DUMP_XLATED */ 724 if (info->xlated_prog_len == 0 || !info->xlated_prog_insns) { 725 p_err("error retrieving insn dump: kernel.kptr_restrict set?"); 726 return -1; 727 } 728 buf = u64_to_ptr(info->xlated_prog_insns); 729 member_len = info->xlated_prog_len; 730 } 731 732 if (info->btf_id) { 733 btf = btf__load_from_kernel_by_id(info->btf_id); 734 if (!btf) { 735 p_err("failed to get btf"); 736 return -1; 737 } 738 } 739 740 func_info = u64_to_ptr(info->func_info); 741 742 if (info->nr_line_info) { 743 prog_linfo = bpf_prog_linfo__new(info); 744 if (!prog_linfo) 745 p_info("error in processing bpf_line_info. continue without it."); 746 } 747 748 if (filepath) { 749 fd = open(filepath, O_WRONLY | O_CREAT | O_TRUNC, 0600); 750 if (fd < 0) { 751 p_err("can't open file %s: %s", filepath, 752 strerror(errno)); 753 goto exit_free; 754 } 755 756 n = write(fd, buf, member_len); 757 close(fd); 758 if (n != (ssize_t)member_len) { 759 p_err("error writing output file: %s", 760 n < 0 ? strerror(errno) : "short write"); 761 goto exit_free; 762 } 763 764 if (json_output) 765 jsonw_null(json_wtr); 766 } else if (mode == DUMP_JITED) { 767 const char *name = NULL; 768 769 if (info->ifindex) { 770 name = ifindex_to_arch(info->ifindex, info->netns_dev, 771 info->netns_ino, &disasm_opt); 772 if (!name) 773 goto exit_free; 774 } 775 776 if (info->nr_jited_func_lens && info->jited_func_lens) { 777 struct kernel_sym *sym = NULL; 778 struct bpf_func_info *record; 779 char sym_name[SYM_MAX_NAME]; 780 unsigned char *img = buf; 781 __u64 *ksyms = NULL; 782 __u32 *lens; 783 __u32 i; 784 if (info->nr_jited_ksyms) { 785 kernel_syms_load(&dd); 786 ksyms = u64_to_ptr(info->jited_ksyms); 787 } 788 789 if (json_output) 790 jsonw_start_array(json_wtr); 791 792 lens = u64_to_ptr(info->jited_func_lens); 793 for (i = 0; i < info->nr_jited_func_lens; i++) { 794 if (ksyms) { 795 sym = kernel_syms_search(&dd, ksyms[i]); 796 if (sym) 797 sprintf(sym_name, "%s", sym->name); 798 else 799 sprintf(sym_name, "0x%016llx", ksyms[i]); 800 } else { 801 strcpy(sym_name, "unknown"); 802 } 803 804 if (func_info) { 805 record = func_info + i * info->func_info_rec_size; 806 btf_dumper_type_only(btf, record->type_id, 807 func_sig, 808 sizeof(func_sig)); 809 } 810 811 if (json_output) { 812 jsonw_start_object(json_wtr); 813 if (func_info && func_sig[0] != '\0') { 814 jsonw_name(json_wtr, "proto"); 815 jsonw_string(json_wtr, func_sig); 816 } 817 jsonw_name(json_wtr, "name"); 818 jsonw_string(json_wtr, sym_name); 819 jsonw_name(json_wtr, "insns"); 820 } else { 821 if (func_info && func_sig[0] != '\0') 822 printf("%s:\n", func_sig); 823 printf("%s:\n", sym_name); 824 } 825 826 if (ksyms) { 827 if (disasm_print_insn(img, lens[i], opcodes, 828 name, disasm_opt, btf, 829 prog_linfo, ksyms[i], i, 830 linum)) 831 goto exit_free; 832 } else { 833 if (disasm_print_insn(img, lens[i], opcodes, 834 name, disasm_opt, btf, 835 NULL, 0, 0, false)) 836 goto exit_free; 837 } 838 839 img += lens[i]; 840 841 if (json_output) 842 jsonw_end_object(json_wtr); 843 else 844 printf("\n"); 845 } 846 847 if (json_output) 848 jsonw_end_array(json_wtr); 849 } else { 850 if (disasm_print_insn(buf, member_len, opcodes, name, 851 disasm_opt, btf, NULL, 0, 0, 852 false)) 853 goto exit_free; 854 } 855 } else { 856 kernel_syms_load(&dd); 857 dd.nr_jited_ksyms = info->nr_jited_ksyms; 858 dd.jited_ksyms = u64_to_ptr(info->jited_ksyms); 859 dd.btf = btf; 860 dd.func_info = func_info; 861 dd.finfo_rec_size = info->func_info_rec_size; 862 dd.prog_linfo = prog_linfo; 863 864 if (json_output) 865 dump_xlated_json(&dd, buf, member_len, opcodes, linum); 866 else if (visual) 867 dump_xlated_cfg(&dd, buf, member_len, opcodes, linum); 868 else 869 dump_xlated_plain(&dd, buf, member_len, opcodes, linum); 870 kernel_syms_destroy(&dd); 871 } 872 873 err = 0; 874 875 exit_free: 876 btf__free(btf); 877 bpf_prog_linfo__free(prog_linfo); 878 return err; 879 } 880 881 static int do_dump(int argc, char **argv) 882 { 883 struct bpf_prog_info info; 884 __u32 info_len = sizeof(info); 885 size_t info_data_sz = 0; 886 void *info_data = NULL; 887 char *filepath = NULL; 888 bool opcodes = false; 889 bool visual = false; 890 enum dump_mode mode; 891 bool linum = false; 892 int nb_fds, i = 0; 893 int *fds = NULL; 894 int err = -1; 895 896 if (is_prefix(*argv, "jited")) { 897 if (disasm_init()) 898 return -1; 899 mode = DUMP_JITED; 900 } else if (is_prefix(*argv, "xlated")) { 901 mode = DUMP_XLATED; 902 } else { 903 p_err("expected 'xlated' or 'jited', got: %s", *argv); 904 return -1; 905 } 906 NEXT_ARG(); 907 908 if (argc < 2) 909 usage(); 910 911 fds = malloc(sizeof(int)); 912 if (!fds) { 913 p_err("mem alloc failed"); 914 return -1; 915 } 916 nb_fds = prog_parse_fds(&argc, &argv, &fds); 917 if (nb_fds < 1) 918 goto exit_free; 919 920 while (argc) { 921 if (is_prefix(*argv, "file")) { 922 NEXT_ARG(); 923 if (!argc) { 924 p_err("expected file path"); 925 goto exit_close; 926 } 927 if (nb_fds > 1) { 928 p_err("several programs matched"); 929 goto exit_close; 930 } 931 932 filepath = *argv; 933 NEXT_ARG(); 934 } else if (is_prefix(*argv, "opcodes")) { 935 opcodes = true; 936 NEXT_ARG(); 937 } else if (is_prefix(*argv, "visual")) { 938 if (nb_fds > 1) { 939 p_err("several programs matched"); 940 goto exit_close; 941 } 942 943 visual = true; 944 NEXT_ARG(); 945 } else if (is_prefix(*argv, "linum")) { 946 linum = true; 947 NEXT_ARG(); 948 } else { 949 usage(); 950 goto exit_close; 951 } 952 } 953 954 if (filepath && (opcodes || visual || linum)) { 955 p_err("'file' is not compatible with 'opcodes', 'visual', or 'linum'"); 956 goto exit_close; 957 } 958 if (json_output && visual) { 959 p_err("'visual' is not compatible with JSON output"); 960 goto exit_close; 961 } 962 963 if (json_output && nb_fds > 1) 964 jsonw_start_array(json_wtr); /* root array */ 965 for (i = 0; i < nb_fds; i++) { 966 memset(&info, 0, sizeof(info)); 967 968 err = bpf_prog_get_info_by_fd(fds[i], &info, &info_len); 969 if (err) { 970 p_err("can't get prog info: %s", strerror(errno)); 971 break; 972 } 973 974 err = prep_prog_info(&info, mode, &info_data, &info_data_sz); 975 if (err) { 976 p_err("can't grow prog info_data"); 977 break; 978 } 979 980 err = bpf_prog_get_info_by_fd(fds[i], &info, &info_len); 981 if (err) { 982 p_err("can't get prog info: %s", strerror(errno)); 983 break; 984 } 985 986 if (json_output && nb_fds > 1) { 987 jsonw_start_object(json_wtr); /* prog object */ 988 print_prog_header_json(&info, fds[i]); 989 jsonw_name(json_wtr, "insns"); 990 } else if (nb_fds > 1) { 991 print_prog_header_plain(&info, fds[i]); 992 } 993 994 err = prog_dump(&info, mode, filepath, opcodes, visual, linum); 995 996 if (json_output && nb_fds > 1) 997 jsonw_end_object(json_wtr); /* prog object */ 998 else if (i != nb_fds - 1 && nb_fds > 1) 999 printf("\n"); 1000 1001 if (err) 1002 break; 1003 close(fds[i]); 1004 } 1005 if (json_output && nb_fds > 1) 1006 jsonw_end_array(json_wtr); /* root array */ 1007 1008 exit_close: 1009 for (; i < nb_fds; i++) 1010 close(fds[i]); 1011 exit_free: 1012 free(info_data); 1013 free(fds); 1014 return err; 1015 } 1016 1017 static int do_pin(int argc, char **argv) 1018 { 1019 int err; 1020 1021 err = do_pin_any(argc, argv, prog_parse_fd); 1022 if (!err && json_output) 1023 jsonw_null(json_wtr); 1024 return err; 1025 } 1026 1027 struct map_replace { 1028 int idx; 1029 int fd; 1030 char *name; 1031 }; 1032 1033 static int map_replace_compar(const void *p1, const void *p2) 1034 { 1035 const struct map_replace *a = p1, *b = p2; 1036 1037 return a->idx - b->idx; 1038 } 1039 1040 static int parse_attach_detach_args(int argc, char **argv, int *progfd, 1041 enum bpf_attach_type *attach_type, 1042 int *mapfd) 1043 { 1044 if (!REQ_ARGS(3)) 1045 return -EINVAL; 1046 1047 *progfd = prog_parse_fd(&argc, &argv); 1048 if (*progfd < 0) 1049 return *progfd; 1050 1051 *attach_type = parse_attach_type(*argv); 1052 if (*attach_type == __MAX_BPF_ATTACH_TYPE) { 1053 p_err("invalid attach/detach type"); 1054 return -EINVAL; 1055 } 1056 1057 if (*attach_type == BPF_FLOW_DISSECTOR) { 1058 *mapfd = 0; 1059 return 0; 1060 } 1061 1062 NEXT_ARG(); 1063 if (!REQ_ARGS(2)) 1064 return -EINVAL; 1065 1066 *mapfd = map_parse_fd(&argc, &argv, 0); 1067 if (*mapfd < 0) 1068 return *mapfd; 1069 1070 return 0; 1071 } 1072 1073 static int do_attach(int argc, char **argv) 1074 { 1075 enum bpf_attach_type attach_type; 1076 int err, progfd; 1077 int mapfd; 1078 1079 err = parse_attach_detach_args(argc, argv, 1080 &progfd, &attach_type, &mapfd); 1081 if (err) 1082 return err; 1083 1084 err = bpf_prog_attach(progfd, mapfd, attach_type, 0); 1085 if (err) { 1086 p_err("failed prog attach to map"); 1087 return -EINVAL; 1088 } 1089 1090 if (json_output) 1091 jsonw_null(json_wtr); 1092 return 0; 1093 } 1094 1095 static int do_detach(int argc, char **argv) 1096 { 1097 enum bpf_attach_type attach_type; 1098 int err, progfd; 1099 int mapfd; 1100 1101 err = parse_attach_detach_args(argc, argv, 1102 &progfd, &attach_type, &mapfd); 1103 if (err) 1104 return err; 1105 1106 err = bpf_prog_detach2(progfd, mapfd, attach_type); 1107 if (err) { 1108 p_err("failed prog detach from map"); 1109 return -EINVAL; 1110 } 1111 1112 if (json_output) 1113 jsonw_null(json_wtr); 1114 return 0; 1115 } 1116 1117 enum prog_tracelog_mode { 1118 TRACE_STDOUT, 1119 TRACE_STDERR, 1120 }; 1121 1122 static int 1123 prog_tracelog_stream(int prog_fd, enum prog_tracelog_mode mode) 1124 { 1125 FILE *file = mode == TRACE_STDOUT ? stdout : stderr; 1126 int stream_id = mode == TRACE_STDOUT ? 1 : 2; 1127 char buf[512]; 1128 int ret; 1129 1130 ret = 0; 1131 do { 1132 ret = bpf_prog_stream_read(prog_fd, stream_id, buf, sizeof(buf), NULL); 1133 if (ret > 0) 1134 fwrite(buf, sizeof(buf[0]), ret, file); 1135 } while (ret > 0); 1136 1137 fflush(file); 1138 return ret ? -1 : 0; 1139 } 1140 1141 static int do_tracelog_any(int argc, char **argv) 1142 { 1143 enum prog_tracelog_mode mode; 1144 int fd; 1145 1146 if (argc == 0) 1147 return do_tracelog(argc, argv); 1148 if (!is_prefix(*argv, "stdout") && !is_prefix(*argv, "stderr")) 1149 usage(); 1150 mode = is_prefix(*argv, "stdout") ? TRACE_STDOUT : TRACE_STDERR; 1151 NEXT_ARG(); 1152 1153 if (!REQ_ARGS(2)) 1154 return -1; 1155 1156 fd = prog_parse_fd(&argc, &argv); 1157 if (fd < 0) 1158 return -1; 1159 1160 return prog_tracelog_stream(fd, mode); 1161 } 1162 1163 static int check_single_stdin(char *file_data_in, char *file_ctx_in) 1164 { 1165 if (file_data_in && file_ctx_in && 1166 !strcmp(file_data_in, "-") && !strcmp(file_ctx_in, "-")) { 1167 p_err("cannot use standard input for both data_in and ctx_in"); 1168 return -1; 1169 } 1170 1171 return 0; 1172 } 1173 1174 static int get_run_data(const char *fname, void **data_ptr, unsigned int *size) 1175 { 1176 size_t block_size = 256; 1177 size_t buf_size = block_size; 1178 size_t nb_read = 0; 1179 void *tmp; 1180 FILE *f; 1181 1182 if (!fname) { 1183 *data_ptr = NULL; 1184 *size = 0; 1185 return 0; 1186 } 1187 1188 if (!strcmp(fname, "-")) 1189 f = stdin; 1190 else 1191 f = fopen(fname, "r"); 1192 if (!f) { 1193 p_err("failed to open %s: %s", fname, strerror(errno)); 1194 return -1; 1195 } 1196 1197 *data_ptr = malloc(block_size); 1198 if (!*data_ptr) { 1199 p_err("failed to allocate memory for data_in/ctx_in: %s", 1200 strerror(errno)); 1201 goto err_fclose; 1202 } 1203 1204 while ((nb_read += fread(*data_ptr + nb_read, 1, block_size, f))) { 1205 if (feof(f)) 1206 break; 1207 if (ferror(f)) { 1208 p_err("failed to read data_in/ctx_in from %s: %s", 1209 fname, strerror(errno)); 1210 goto err_free; 1211 } 1212 if (nb_read > buf_size - block_size) { 1213 if (buf_size == UINT32_MAX) { 1214 p_err("data_in/ctx_in is too long (max: %u)", 1215 UINT32_MAX); 1216 goto err_free; 1217 } 1218 /* No space for fread()-ing next chunk; realloc() */ 1219 buf_size *= 2; 1220 tmp = realloc(*data_ptr, buf_size); 1221 if (!tmp) { 1222 p_err("failed to reallocate data_in/ctx_in: %s", 1223 strerror(errno)); 1224 goto err_free; 1225 } 1226 *data_ptr = tmp; 1227 } 1228 } 1229 if (f != stdin) 1230 fclose(f); 1231 1232 *size = nb_read; 1233 return 0; 1234 1235 err_free: 1236 free(*data_ptr); 1237 *data_ptr = NULL; 1238 err_fclose: 1239 if (f != stdin) 1240 fclose(f); 1241 return -1; 1242 } 1243 1244 static void hex_print(void *data, unsigned int size, FILE *f) 1245 { 1246 size_t i, j; 1247 char c; 1248 1249 for (i = 0; i < size; i += 16) { 1250 /* Row offset */ 1251 fprintf(f, "%07zx\t", i); 1252 1253 /* Hexadecimal values */ 1254 for (j = i; j < i + 16 && j < size; j++) 1255 fprintf(f, "%02x%s", *(uint8_t *)(data + j), 1256 j % 2 ? " " : ""); 1257 for (; j < i + 16; j++) 1258 fprintf(f, " %s", j % 2 ? " " : ""); 1259 1260 /* ASCII values (if relevant), '.' otherwise */ 1261 fprintf(f, "| "); 1262 for (j = i; j < i + 16 && j < size; j++) { 1263 c = *(char *)(data + j); 1264 if (c < ' ' || c > '~') 1265 c = '.'; 1266 fprintf(f, "%c%s", c, j == i + 7 ? " " : ""); 1267 } 1268 1269 fprintf(f, "\n"); 1270 } 1271 } 1272 1273 static int 1274 print_run_output(void *data, unsigned int size, const char *fname, 1275 const char *json_key) 1276 { 1277 size_t nb_written; 1278 FILE *f; 1279 1280 if (!fname) 1281 return 0; 1282 1283 if (!strcmp(fname, "-")) { 1284 f = stdout; 1285 if (json_output) { 1286 jsonw_name(json_wtr, json_key); 1287 print_data_json(data, size); 1288 } else { 1289 hex_print(data, size, f); 1290 } 1291 return 0; 1292 } 1293 1294 f = fopen(fname, "w"); 1295 if (!f) { 1296 p_err("failed to open %s: %s", fname, strerror(errno)); 1297 return -1; 1298 } 1299 1300 nb_written = fwrite(data, 1, size, f); 1301 fclose(f); 1302 if (nb_written != size) { 1303 p_err("failed to write output data/ctx: %s", strerror(errno)); 1304 return -1; 1305 } 1306 1307 return 0; 1308 } 1309 1310 static int alloc_run_data(void **data_ptr, unsigned int size_out) 1311 { 1312 *data_ptr = calloc(size_out, 1); 1313 if (!*data_ptr) { 1314 p_err("failed to allocate memory for output data/ctx: %s", 1315 strerror(errno)); 1316 return -1; 1317 } 1318 1319 return 0; 1320 } 1321 1322 static int do_run(int argc, char **argv) 1323 { 1324 char *data_fname_in = NULL, *data_fname_out = NULL; 1325 char *ctx_fname_in = NULL, *ctx_fname_out = NULL; 1326 const unsigned int default_size = SZ_32K; 1327 void *data_in = NULL, *data_out = NULL; 1328 void *ctx_in = NULL, *ctx_out = NULL; 1329 unsigned int repeat = 1; 1330 int fd, err; 1331 LIBBPF_OPTS(bpf_test_run_opts, test_attr); 1332 1333 if (!REQ_ARGS(4)) 1334 return -1; 1335 1336 fd = prog_parse_fd(&argc, &argv); 1337 if (fd < 0) 1338 return -1; 1339 1340 while (argc) { 1341 if (detect_common_prefix(*argv, "data_in", "data_out", 1342 "data_size_out", NULL)) 1343 return -1; 1344 if (detect_common_prefix(*argv, "ctx_in", "ctx_out", 1345 "ctx_size_out", NULL)) 1346 return -1; 1347 1348 if (is_prefix(*argv, "data_in")) { 1349 NEXT_ARG(); 1350 if (!REQ_ARGS(1)) 1351 return -1; 1352 1353 data_fname_in = GET_ARG(); 1354 if (check_single_stdin(data_fname_in, ctx_fname_in)) 1355 return -1; 1356 } else if (is_prefix(*argv, "data_out")) { 1357 NEXT_ARG(); 1358 if (!REQ_ARGS(1)) 1359 return -1; 1360 1361 data_fname_out = GET_ARG(); 1362 } else if (is_prefix(*argv, "data_size_out")) { 1363 char *endptr; 1364 1365 NEXT_ARG(); 1366 if (!REQ_ARGS(1)) 1367 return -1; 1368 1369 test_attr.data_size_out = strtoul(*argv, &endptr, 0); 1370 if (*endptr) { 1371 p_err("can't parse %s as output data size", 1372 *argv); 1373 return -1; 1374 } 1375 NEXT_ARG(); 1376 } else if (is_prefix(*argv, "ctx_in")) { 1377 NEXT_ARG(); 1378 if (!REQ_ARGS(1)) 1379 return -1; 1380 1381 ctx_fname_in = GET_ARG(); 1382 if (check_single_stdin(data_fname_in, ctx_fname_in)) 1383 return -1; 1384 } else if (is_prefix(*argv, "ctx_out")) { 1385 NEXT_ARG(); 1386 if (!REQ_ARGS(1)) 1387 return -1; 1388 1389 ctx_fname_out = GET_ARG(); 1390 } else if (is_prefix(*argv, "ctx_size_out")) { 1391 char *endptr; 1392 1393 NEXT_ARG(); 1394 if (!REQ_ARGS(1)) 1395 return -1; 1396 1397 test_attr.ctx_size_out = strtoul(*argv, &endptr, 0); 1398 if (*endptr) { 1399 p_err("can't parse %s as output context size", 1400 *argv); 1401 return -1; 1402 } 1403 NEXT_ARG(); 1404 } else if (is_prefix(*argv, "repeat")) { 1405 char *endptr; 1406 1407 NEXT_ARG(); 1408 if (!REQ_ARGS(1)) 1409 return -1; 1410 1411 repeat = strtoul(*argv, &endptr, 0); 1412 if (*endptr) { 1413 p_err("can't parse %s as repeat number", 1414 *argv); 1415 return -1; 1416 } 1417 NEXT_ARG(); 1418 } else { 1419 p_err("expected no more arguments, 'data_in', 'data_out', 'data_size_out', 'ctx_in', 'ctx_out', 'ctx_size_out' or 'repeat', got: '%s'?", 1420 *argv); 1421 return -1; 1422 } 1423 } 1424 1425 err = get_run_data(data_fname_in, &data_in, &test_attr.data_size_in); 1426 if (err) 1427 return -1; 1428 1429 if (data_in) { 1430 if (!test_attr.data_size_out) 1431 test_attr.data_size_out = default_size; 1432 err = alloc_run_data(&data_out, test_attr.data_size_out); 1433 if (err) 1434 goto free_data_in; 1435 } 1436 1437 err = get_run_data(ctx_fname_in, &ctx_in, &test_attr.ctx_size_in); 1438 if (err) 1439 goto free_data_out; 1440 1441 if (ctx_in) { 1442 if (!test_attr.ctx_size_out) 1443 test_attr.ctx_size_out = default_size; 1444 err = alloc_run_data(&ctx_out, test_attr.ctx_size_out); 1445 if (err) 1446 goto free_ctx_in; 1447 } 1448 1449 test_attr.repeat = repeat; 1450 test_attr.data_in = data_in; 1451 test_attr.data_out = data_out; 1452 test_attr.ctx_in = ctx_in; 1453 test_attr.ctx_out = ctx_out; 1454 1455 err = bpf_prog_test_run_opts(fd, &test_attr); 1456 if (err) { 1457 p_err("failed to run program: %s", strerror(errno)); 1458 goto free_ctx_out; 1459 } 1460 1461 err = 0; 1462 1463 if (json_output) 1464 jsonw_start_object(json_wtr); /* root */ 1465 1466 /* Do not exit on errors occurring when printing output data/context, 1467 * we still want to print return value and duration for program run. 1468 */ 1469 if (test_attr.data_size_out) 1470 err += print_run_output(test_attr.data_out, 1471 test_attr.data_size_out, 1472 data_fname_out, "data_out"); 1473 if (test_attr.ctx_size_out) 1474 err += print_run_output(test_attr.ctx_out, 1475 test_attr.ctx_size_out, 1476 ctx_fname_out, "ctx_out"); 1477 1478 if (json_output) { 1479 jsonw_uint_field(json_wtr, "retval", test_attr.retval); 1480 jsonw_uint_field(json_wtr, "duration", test_attr.duration); 1481 jsonw_end_object(json_wtr); /* root */ 1482 } else { 1483 fprintf(stdout, "Return value: %u, duration%s: %uns\n", 1484 test_attr.retval, 1485 repeat > 1 ? " (average)" : "", test_attr.duration); 1486 } 1487 1488 free_ctx_out: 1489 free(ctx_out); 1490 free_ctx_in: 1491 free(ctx_in); 1492 free_data_out: 1493 free(data_out); 1494 free_data_in: 1495 free(data_in); 1496 1497 return err; 1498 } 1499 1500 static int 1501 get_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type, 1502 enum bpf_attach_type *expected_attach_type) 1503 { 1504 libbpf_print_fn_t print_backup; 1505 int ret; 1506 1507 ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type); 1508 if (!ret) 1509 return ret; 1510 1511 /* libbpf_prog_type_by_name() failed, let's re-run with debug level */ 1512 print_backup = libbpf_set_print(print_all_levels); 1513 ret = libbpf_prog_type_by_name(name, prog_type, expected_attach_type); 1514 libbpf_set_print(print_backup); 1515 1516 return ret; 1517 } 1518 1519 static int 1520 auto_attach_program(struct bpf_program *prog, const char *path) 1521 { 1522 struct bpf_link *link; 1523 int err; 1524 1525 link = bpf_program__attach(prog); 1526 if (!link) { 1527 p_info("Program %s does not support autoattach, falling back to pinning", 1528 bpf_program__name(prog)); 1529 return bpf_obj_pin(bpf_program__fd(prog), path); 1530 } 1531 1532 err = bpf_link__pin(link, path); 1533 bpf_link__destroy(link); 1534 return err; 1535 } 1536 1537 static int 1538 auto_attach_programs(struct bpf_object *obj, const char *path) 1539 { 1540 struct bpf_program *prog; 1541 char buf[PATH_MAX]; 1542 int err; 1543 1544 bpf_object__for_each_program(prog, obj) { 1545 err = pathname_concat(buf, sizeof(buf), path, bpf_program__name(prog)); 1546 if (err) 1547 goto err_unpin_programs; 1548 1549 err = auto_attach_program(prog, buf); 1550 if (err) 1551 goto err_unpin_programs; 1552 } 1553 1554 return 0; 1555 1556 err_unpin_programs: 1557 while ((prog = bpf_object__prev_program(obj, prog))) { 1558 if (pathname_concat(buf, sizeof(buf), path, bpf_program__name(prog))) 1559 continue; 1560 1561 bpf_program__unpin(prog, buf); 1562 } 1563 1564 return err; 1565 } 1566 1567 static int load_with_options(int argc, char **argv, bool first_prog_only) 1568 { 1569 enum bpf_prog_type common_prog_type = BPF_PROG_TYPE_UNSPEC; 1570 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts, 1571 .relaxed_maps = relaxed_maps, 1572 ); 1573 enum bpf_attach_type expected_attach_type; 1574 struct map_replace *map_replace = NULL; 1575 struct bpf_program *prog = NULL, *pos; 1576 unsigned int old_map_fds = 0; 1577 const char *pinmaps = NULL; 1578 __u32 xdpmeta_ifindex = 0; 1579 __u32 offload_ifindex = 0; 1580 bool auto_attach = false; 1581 struct bpf_object *obj; 1582 struct bpf_map *map; 1583 const char *pinfile; 1584 unsigned int i, j; 1585 const char *file; 1586 int idx, err; 1587 1588 1589 if (!REQ_ARGS(2)) 1590 return -1; 1591 file = GET_ARG(); 1592 pinfile = GET_ARG(); 1593 1594 while (argc) { 1595 if (is_prefix(*argv, "type")) { 1596 NEXT_ARG(); 1597 1598 if (common_prog_type != BPF_PROG_TYPE_UNSPEC) { 1599 p_err("program type already specified"); 1600 goto err_free_reuse_maps; 1601 } 1602 if (!REQ_ARGS(1)) 1603 goto err_free_reuse_maps; 1604 1605 err = libbpf_prog_type_by_name(*argv, &common_prog_type, 1606 &expected_attach_type); 1607 if (err < 0) { 1608 /* Put a '/' at the end of type to appease libbpf */ 1609 char *type = malloc(strlen(*argv) + 2); 1610 1611 if (!type) { 1612 p_err("mem alloc failed"); 1613 goto err_free_reuse_maps; 1614 } 1615 *type = 0; 1616 strcat(type, *argv); 1617 strcat(type, "/"); 1618 1619 err = get_prog_type_by_name(type, &common_prog_type, 1620 &expected_attach_type); 1621 free(type); 1622 if (err < 0) 1623 goto err_free_reuse_maps; 1624 } 1625 1626 NEXT_ARG(); 1627 } else if (is_prefix(*argv, "map")) { 1628 void *new_map_replace; 1629 char *endptr, *name; 1630 int fd; 1631 1632 NEXT_ARG(); 1633 1634 if (!REQ_ARGS(4)) 1635 goto err_free_reuse_maps; 1636 1637 if (is_prefix(*argv, "idx")) { 1638 NEXT_ARG(); 1639 1640 idx = strtoul(*argv, &endptr, 0); 1641 if (*endptr) { 1642 p_err("can't parse %s as IDX", *argv); 1643 goto err_free_reuse_maps; 1644 } 1645 name = NULL; 1646 } else if (is_prefix(*argv, "name")) { 1647 NEXT_ARG(); 1648 1649 name = *argv; 1650 idx = -1; 1651 } else { 1652 p_err("expected 'idx' or 'name', got: '%s'?", 1653 *argv); 1654 goto err_free_reuse_maps; 1655 } 1656 NEXT_ARG(); 1657 1658 fd = map_parse_fd(&argc, &argv, 0); 1659 if (fd < 0) 1660 goto err_free_reuse_maps; 1661 1662 new_map_replace = libbpf_reallocarray(map_replace, 1663 old_map_fds + 1, 1664 sizeof(*map_replace)); 1665 if (!new_map_replace) { 1666 p_err("mem alloc failed"); 1667 goto err_free_reuse_maps; 1668 } 1669 map_replace = new_map_replace; 1670 1671 map_replace[old_map_fds].idx = idx; 1672 map_replace[old_map_fds].name = name; 1673 map_replace[old_map_fds].fd = fd; 1674 old_map_fds++; 1675 } else if (is_prefix(*argv, "dev")) { 1676 p_info("Warning: 'bpftool prog load [...] dev <ifname>' syntax is deprecated.\n" 1677 "Going further, please use 'offload_dev <ifname>' to offload program to device.\n" 1678 "For applications using XDP hints only, use 'xdpmeta_dev <ifname>'."); 1679 goto offload_dev; 1680 } else if (is_prefix(*argv, "offload_dev")) { 1681 offload_dev: 1682 NEXT_ARG(); 1683 1684 if (offload_ifindex) { 1685 p_err("offload_dev already specified"); 1686 goto err_free_reuse_maps; 1687 } else if (xdpmeta_ifindex) { 1688 p_err("xdpmeta_dev and offload_dev are mutually exclusive"); 1689 goto err_free_reuse_maps; 1690 } 1691 if (!REQ_ARGS(1)) 1692 goto err_free_reuse_maps; 1693 1694 offload_ifindex = if_nametoindex(*argv); 1695 if (!offload_ifindex) { 1696 p_err("unrecognized netdevice '%s': %s", 1697 *argv, strerror(errno)); 1698 goto err_free_reuse_maps; 1699 } 1700 NEXT_ARG(); 1701 } else if (is_prefix(*argv, "xdpmeta_dev")) { 1702 NEXT_ARG(); 1703 1704 if (xdpmeta_ifindex) { 1705 p_err("xdpmeta_dev already specified"); 1706 goto err_free_reuse_maps; 1707 } else if (offload_ifindex) { 1708 p_err("xdpmeta_dev and offload_dev are mutually exclusive"); 1709 goto err_free_reuse_maps; 1710 } 1711 if (!REQ_ARGS(1)) 1712 goto err_free_reuse_maps; 1713 1714 xdpmeta_ifindex = if_nametoindex(*argv); 1715 if (!xdpmeta_ifindex) { 1716 p_err("unrecognized netdevice '%s': %s", 1717 *argv, strerror(errno)); 1718 goto err_free_reuse_maps; 1719 } 1720 NEXT_ARG(); 1721 } else if (is_prefix(*argv, "pinmaps")) { 1722 NEXT_ARG(); 1723 1724 if (!REQ_ARGS(1)) 1725 goto err_free_reuse_maps; 1726 1727 pinmaps = GET_ARG(); 1728 } else if (is_prefix(*argv, "autoattach")) { 1729 auto_attach = true; 1730 NEXT_ARG(); 1731 } else if (is_prefix(*argv, "kernel_btf")) { 1732 NEXT_ARG(); 1733 1734 if (!REQ_ARGS(1)) 1735 goto err_free_reuse_maps; 1736 1737 open_opts.btf_custom_path = GET_ARG(); 1738 } else { 1739 p_err("expected no more arguments, " 1740 "'type', 'map', 'offload_dev', 'xdpmeta_dev', 'pinmaps', " 1741 "'autoattach', or 'kernel_btf', got: '%s'?", 1742 *argv); 1743 goto err_free_reuse_maps; 1744 } 1745 } 1746 1747 set_max_rlimit(); 1748 1749 if (verifier_logs) 1750 /* log_level1 + log_level2 + stats, but not stable UAPI */ 1751 open_opts.kernel_log_level = 1 + 2 + 4; 1752 1753 obj = bpf_object__open_file(file, &open_opts); 1754 if (!obj) { 1755 p_err("failed to open object file"); 1756 goto err_free_reuse_maps; 1757 } 1758 1759 bpf_object__for_each_program(pos, obj) { 1760 enum bpf_prog_type prog_type = common_prog_type; 1761 1762 if (prog_type == BPF_PROG_TYPE_UNSPEC) { 1763 const char *sec_name = bpf_program__section_name(pos); 1764 1765 err = get_prog_type_by_name(sec_name, &prog_type, 1766 &expected_attach_type); 1767 if (err < 0) 1768 goto err_close_obj; 1769 } 1770 1771 if (prog_type == BPF_PROG_TYPE_XDP && xdpmeta_ifindex) { 1772 bpf_program__set_flags(pos, BPF_F_XDP_DEV_BOUND_ONLY); 1773 bpf_program__set_ifindex(pos, xdpmeta_ifindex); 1774 } else { 1775 bpf_program__set_ifindex(pos, offload_ifindex); 1776 } 1777 if (bpf_program__type(pos) != prog_type) 1778 bpf_program__set_type(pos, prog_type); 1779 bpf_program__set_expected_attach_type(pos, expected_attach_type); 1780 } 1781 1782 qsort(map_replace, old_map_fds, sizeof(*map_replace), 1783 map_replace_compar); 1784 1785 /* After the sort maps by name will be first on the list, because they 1786 * have idx == -1. Resolve them. 1787 */ 1788 j = 0; 1789 while (j < old_map_fds && map_replace[j].name) { 1790 i = 0; 1791 bpf_object__for_each_map(map, obj) { 1792 if (!strcmp(bpf_map__name(map), map_replace[j].name)) { 1793 map_replace[j].idx = i; 1794 break; 1795 } 1796 i++; 1797 } 1798 if (map_replace[j].idx == -1) { 1799 p_err("unable to find map '%s'", map_replace[j].name); 1800 goto err_close_obj; 1801 } 1802 j++; 1803 } 1804 /* Resort if any names were resolved */ 1805 if (j) 1806 qsort(map_replace, old_map_fds, sizeof(*map_replace), 1807 map_replace_compar); 1808 1809 /* Set ifindex and name reuse */ 1810 j = 0; 1811 idx = 0; 1812 bpf_object__for_each_map(map, obj) { 1813 if (bpf_map__type(map) != BPF_MAP_TYPE_PERF_EVENT_ARRAY) 1814 bpf_map__set_ifindex(map, offload_ifindex); 1815 1816 if (j < old_map_fds && idx == map_replace[j].idx) { 1817 err = bpf_map__reuse_fd(map, map_replace[j++].fd); 1818 if (err) { 1819 p_err("unable to set up map reuse: %d", err); 1820 goto err_close_obj; 1821 } 1822 1823 /* Next reuse wants to apply to the same map */ 1824 if (j < old_map_fds && map_replace[j].idx == idx) { 1825 p_err("replacement for map idx %d specified more than once", 1826 idx); 1827 goto err_close_obj; 1828 } 1829 } 1830 1831 idx++; 1832 } 1833 if (j < old_map_fds) { 1834 p_err("map idx '%d' not used", map_replace[j].idx); 1835 goto err_close_obj; 1836 } 1837 1838 err = bpf_object__load(obj); 1839 if (err) { 1840 p_err("failed to load object file"); 1841 goto err_close_obj; 1842 } 1843 1844 if (first_prog_only) 1845 err = mount_bpffs_for_file(pinfile); 1846 else 1847 err = create_and_mount_bpffs_dir(pinfile); 1848 if (err) 1849 goto err_close_obj; 1850 1851 if (first_prog_only) { 1852 prog = bpf_object__next_program(obj, NULL); 1853 if (!prog) { 1854 p_err("object file doesn't contain any bpf program"); 1855 goto err_close_obj; 1856 } 1857 1858 if (auto_attach) 1859 err = auto_attach_program(prog, pinfile); 1860 else 1861 err = bpf_obj_pin(bpf_program__fd(prog), pinfile); 1862 if (err) { 1863 p_err("failed to pin program %s", 1864 bpf_program__section_name(prog)); 1865 goto err_close_obj; 1866 } 1867 } else { 1868 if (auto_attach) 1869 err = auto_attach_programs(obj, pinfile); 1870 else 1871 err = bpf_object__pin_programs(obj, pinfile); 1872 if (err) { 1873 p_err("failed to pin all programs"); 1874 goto err_close_obj; 1875 } 1876 } 1877 1878 if (pinmaps) { 1879 err = create_and_mount_bpffs_dir(pinmaps); 1880 if (err) 1881 goto err_unpin; 1882 1883 err = bpf_object__pin_maps(obj, pinmaps); 1884 if (err) { 1885 p_err("failed to pin all maps"); 1886 goto err_unpin; 1887 } 1888 } 1889 1890 if (json_output) 1891 jsonw_null(json_wtr); 1892 1893 bpf_object__close(obj); 1894 for (i = 0; i < old_map_fds; i++) 1895 close(map_replace[i].fd); 1896 free(map_replace); 1897 1898 return 0; 1899 1900 err_unpin: 1901 if (first_prog_only) 1902 unlink(pinfile); 1903 else 1904 bpf_object__unpin_programs(obj, pinfile); 1905 err_close_obj: 1906 bpf_object__close(obj); 1907 err_free_reuse_maps: 1908 for (i = 0; i < old_map_fds; i++) 1909 close(map_replace[i].fd); 1910 free(map_replace); 1911 return -1; 1912 } 1913 1914 static int count_open_fds(void) 1915 { 1916 DIR *dp = opendir("/proc/self/fd"); 1917 struct dirent *de; 1918 int cnt = -3; 1919 1920 if (!dp) 1921 return -1; 1922 1923 while ((de = readdir(dp))) 1924 cnt++; 1925 1926 closedir(dp); 1927 return cnt; 1928 } 1929 1930 static int try_loader(struct gen_loader_opts *gen) 1931 { 1932 struct bpf_load_and_run_opts opts = {}; 1933 struct bpf_loader_ctx *ctx; 1934 char sig_buf[MAX_SIG_SIZE]; 1935 __u8 prog_sha[SHA256_DIGEST_LENGTH]; 1936 int ctx_sz = sizeof(*ctx) + 64 * max(sizeof(struct bpf_map_desc), 1937 sizeof(struct bpf_prog_desc)); 1938 int log_buf_sz = (1u << 24) - 1; 1939 int err, fds_before, fd_delta; 1940 char *log_buf = NULL; 1941 1942 ctx = alloca(ctx_sz); 1943 memset(ctx, 0, ctx_sz); 1944 ctx->sz = ctx_sz; 1945 if (verifier_logs) { 1946 ctx->log_level = 1 + 2 + 4; 1947 ctx->log_size = log_buf_sz; 1948 log_buf = malloc(log_buf_sz); 1949 if (!log_buf) 1950 return -ENOMEM; 1951 ctx->log_buf = (long) log_buf; 1952 } 1953 opts.ctx = ctx; 1954 opts.data = gen->data; 1955 opts.data_sz = gen->data_sz; 1956 opts.insns = gen->insns; 1957 opts.insns_sz = gen->insns_sz; 1958 fds_before = count_open_fds(); 1959 1960 if (sign_progs) { 1961 opts.excl_prog_hash = prog_sha; 1962 opts.excl_prog_hash_sz = sizeof(prog_sha); 1963 opts.signature = sig_buf; 1964 opts.signature_sz = MAX_SIG_SIZE; 1965 opts.keyring_id = KEY_SPEC_SESSION_KEYRING; 1966 1967 err = bpftool_prog_sign(&opts); 1968 if (err < 0) { 1969 p_err("failed to sign program"); 1970 goto out; 1971 } 1972 1973 err = register_session_key(cert_path); 1974 if (err < 0) { 1975 p_err("failed to add session key"); 1976 goto out; 1977 } 1978 } 1979 err = bpf_load_and_run(&opts); 1980 fd_delta = count_open_fds() - fds_before; 1981 if (err < 0 || verifier_logs) { 1982 fprintf(stderr, "err %d\n%s\n%s", err, opts.errstr, log_buf); 1983 if (fd_delta && err < 0) 1984 fprintf(stderr, "loader prog leaked %d FDs\n", 1985 fd_delta); 1986 } 1987 out: 1988 free(log_buf); 1989 return err; 1990 } 1991 1992 static int do_loader(int argc, char **argv) 1993 { 1994 DECLARE_LIBBPF_OPTS(bpf_object_open_opts, open_opts); 1995 DECLARE_LIBBPF_OPTS(gen_loader_opts, gen); 1996 struct bpf_object *obj; 1997 const char *file; 1998 int err = 0; 1999 2000 if (!REQ_ARGS(1)) 2001 return -1; 2002 file = GET_ARG(); 2003 2004 if (verifier_logs) 2005 /* log_level1 + log_level2 + stats, but not stable UAPI */ 2006 open_opts.kernel_log_level = 1 + 2 + 4; 2007 2008 obj = bpf_object__open_file(file, &open_opts); 2009 if (!obj) { 2010 err = -1; 2011 p_err("failed to open object file"); 2012 goto err_close_obj; 2013 } 2014 2015 if (sign_progs) 2016 gen.gen_hash = true; 2017 2018 err = bpf_object__gen_loader(obj, &gen); 2019 if (err) 2020 goto err_close_obj; 2021 2022 err = bpf_object__load(obj); 2023 if (err) { 2024 p_err("failed to load object file"); 2025 goto err_close_obj; 2026 } 2027 2028 if (verifier_logs) { 2029 struct dump_data dd = {}; 2030 2031 kernel_syms_load(&dd); 2032 dump_xlated_plain(&dd, (void *)gen.insns, gen.insns_sz, false, false); 2033 kernel_syms_destroy(&dd); 2034 } 2035 err = try_loader(&gen); 2036 err_close_obj: 2037 bpf_object__close(obj); 2038 return err; 2039 } 2040 2041 static int do_load(int argc, char **argv) 2042 { 2043 if (use_loader) 2044 return do_loader(argc, argv); 2045 return load_with_options(argc, argv, true); 2046 } 2047 2048 static int do_loadall(int argc, char **argv) 2049 { 2050 return load_with_options(argc, argv, false); 2051 } 2052 2053 #ifdef BPFTOOL_WITHOUT_SKELETONS 2054 2055 static int do_profile(int argc, char **argv) 2056 { 2057 p_err("bpftool prog profile command is not supported. Please build bpftool with clang >= 10.0.0"); 2058 return 0; 2059 } 2060 2061 #else /* BPFTOOL_WITHOUT_SKELETONS */ 2062 2063 #include "profiler.skel.h" 2064 2065 struct profile_metric { 2066 const char *name; 2067 struct bpf_perf_event_value val; 2068 struct perf_event_attr attr; 2069 bool selected; 2070 2071 /* calculate ratios like instructions per cycle */ 2072 const int ratio_metric; /* 0 for N/A, 1 for index 0 (cycles) */ 2073 const char *ratio_desc; 2074 const float ratio_mul; 2075 } metrics[] = { 2076 { 2077 .name = "cycles", 2078 .attr = { 2079 .type = PERF_TYPE_HARDWARE, 2080 .config = PERF_COUNT_HW_CPU_CYCLES, 2081 .exclude_user = 1, 2082 }, 2083 }, 2084 { 2085 .name = "instructions", 2086 .attr = { 2087 .type = PERF_TYPE_HARDWARE, 2088 .config = PERF_COUNT_HW_INSTRUCTIONS, 2089 .exclude_user = 1, 2090 }, 2091 .ratio_metric = 1, 2092 .ratio_desc = "insns per cycle", 2093 .ratio_mul = 1.0, 2094 }, 2095 { 2096 .name = "l1d_loads", 2097 .attr = { 2098 .type = PERF_TYPE_HW_CACHE, 2099 .config = 2100 PERF_COUNT_HW_CACHE_L1D | 2101 (PERF_COUNT_HW_CACHE_OP_READ << 8) | 2102 (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16), 2103 .exclude_user = 1, 2104 }, 2105 }, 2106 { 2107 .name = "llc_misses", 2108 .attr = { 2109 .type = PERF_TYPE_HW_CACHE, 2110 .config = 2111 PERF_COUNT_HW_CACHE_LL | 2112 (PERF_COUNT_HW_CACHE_OP_READ << 8) | 2113 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16), 2114 .exclude_user = 1 2115 }, 2116 .ratio_metric = 2, 2117 .ratio_desc = "LLC misses per million insns", 2118 .ratio_mul = 1e6, 2119 }, 2120 { 2121 .name = "itlb_misses", 2122 .attr = { 2123 .type = PERF_TYPE_HW_CACHE, 2124 .config = 2125 PERF_COUNT_HW_CACHE_ITLB | 2126 (PERF_COUNT_HW_CACHE_OP_READ << 8) | 2127 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16), 2128 .exclude_user = 1 2129 }, 2130 .ratio_metric = 2, 2131 .ratio_desc = "itlb misses per million insns", 2132 .ratio_mul = 1e6, 2133 }, 2134 { 2135 .name = "dtlb_misses", 2136 .attr = { 2137 .type = PERF_TYPE_HW_CACHE, 2138 .config = 2139 PERF_COUNT_HW_CACHE_DTLB | 2140 (PERF_COUNT_HW_CACHE_OP_READ << 8) | 2141 (PERF_COUNT_HW_CACHE_RESULT_MISS << 16), 2142 .exclude_user = 1 2143 }, 2144 .ratio_metric = 2, 2145 .ratio_desc = "dtlb misses per million insns", 2146 .ratio_mul = 1e6, 2147 }, 2148 }; 2149 2150 static __u64 profile_total_count; 2151 2152 #define MAX_NUM_PROFILE_METRICS 4 2153 2154 static int profile_parse_metrics(int argc, char **argv) 2155 { 2156 unsigned int metric_cnt; 2157 int selected_cnt = 0; 2158 unsigned int i; 2159 2160 metric_cnt = ARRAY_SIZE(metrics); 2161 2162 while (argc > 0) { 2163 for (i = 0; i < metric_cnt; i++) { 2164 if (is_prefix(argv[0], metrics[i].name)) { 2165 if (!metrics[i].selected) 2166 selected_cnt++; 2167 metrics[i].selected = true; 2168 break; 2169 } 2170 } 2171 if (i == metric_cnt) { 2172 p_err("unknown metric %s", argv[0]); 2173 return -1; 2174 } 2175 NEXT_ARG(); 2176 } 2177 if (selected_cnt > MAX_NUM_PROFILE_METRICS) { 2178 p_err("too many (%d) metrics, please specify no more than %d metrics at a time", 2179 selected_cnt, MAX_NUM_PROFILE_METRICS); 2180 return -1; 2181 } 2182 return selected_cnt; 2183 } 2184 2185 static void profile_read_values(struct profiler_bpf *obj) 2186 { 2187 __u32 m, cpu, num_cpu = obj->rodata->num_cpu; 2188 int reading_map_fd, count_map_fd; 2189 __u64 counts[num_cpu]; 2190 __u32 key = 0; 2191 int err; 2192 2193 reading_map_fd = bpf_map__fd(obj->maps.accum_readings); 2194 count_map_fd = bpf_map__fd(obj->maps.counts); 2195 if (reading_map_fd < 0 || count_map_fd < 0) { 2196 p_err("failed to get fd for map"); 2197 return; 2198 } 2199 2200 err = bpf_map_lookup_elem(count_map_fd, &key, counts); 2201 if (err) { 2202 p_err("failed to read count_map: %s", strerror(errno)); 2203 return; 2204 } 2205 2206 profile_total_count = 0; 2207 for (cpu = 0; cpu < num_cpu; cpu++) 2208 profile_total_count += counts[cpu]; 2209 2210 for (m = 0; m < ARRAY_SIZE(metrics); m++) { 2211 struct bpf_perf_event_value values[num_cpu]; 2212 2213 if (!metrics[m].selected) 2214 continue; 2215 2216 err = bpf_map_lookup_elem(reading_map_fd, &key, values); 2217 if (err) { 2218 p_err("failed to read reading_map: %s", 2219 strerror(errno)); 2220 return; 2221 } 2222 for (cpu = 0; cpu < num_cpu; cpu++) { 2223 metrics[m].val.counter += values[cpu].counter; 2224 metrics[m].val.enabled += values[cpu].enabled; 2225 metrics[m].val.running += values[cpu].running; 2226 } 2227 key++; 2228 } 2229 } 2230 2231 static void profile_print_readings_json(void) 2232 { 2233 __u32 m; 2234 2235 jsonw_start_array(json_wtr); 2236 for (m = 0; m < ARRAY_SIZE(metrics); m++) { 2237 if (!metrics[m].selected) 2238 continue; 2239 jsonw_start_object(json_wtr); 2240 jsonw_string_field(json_wtr, "metric", metrics[m].name); 2241 jsonw_lluint_field(json_wtr, "run_cnt", profile_total_count); 2242 jsonw_lluint_field(json_wtr, "value", metrics[m].val.counter); 2243 jsonw_lluint_field(json_wtr, "enabled", metrics[m].val.enabled); 2244 jsonw_lluint_field(json_wtr, "running", metrics[m].val.running); 2245 2246 jsonw_end_object(json_wtr); 2247 } 2248 jsonw_end_array(json_wtr); 2249 } 2250 2251 static void profile_print_readings_plain(void) 2252 { 2253 __u32 m; 2254 2255 printf("\n%18llu %-20s\n", profile_total_count, "run_cnt"); 2256 for (m = 0; m < ARRAY_SIZE(metrics); m++) { 2257 struct bpf_perf_event_value *val = &metrics[m].val; 2258 int r; 2259 2260 if (!metrics[m].selected) 2261 continue; 2262 printf("%18llu %-20s", val->counter, metrics[m].name); 2263 2264 r = metrics[m].ratio_metric - 1; 2265 if (r >= 0 && metrics[r].selected && 2266 metrics[r].val.counter > 0) { 2267 printf("# %8.2f %-30s", 2268 val->counter * metrics[m].ratio_mul / 2269 metrics[r].val.counter, 2270 metrics[m].ratio_desc); 2271 } else { 2272 printf("%-41s", ""); 2273 } 2274 2275 if (val->enabled > val->running) 2276 printf("(%4.2f%%)", 2277 val->running * 100.0 / val->enabled); 2278 printf("\n"); 2279 } 2280 } 2281 2282 static void profile_print_readings(void) 2283 { 2284 if (json_output) 2285 profile_print_readings_json(); 2286 else 2287 profile_print_readings_plain(); 2288 } 2289 2290 static char *profile_target_name(int tgt_fd) 2291 { 2292 struct bpf_func_info func_info = {}; 2293 struct bpf_prog_info info = {}; 2294 __u32 info_len = sizeof(info); 2295 const struct btf_type *t; 2296 __u32 func_info_rec_size; 2297 struct btf *btf = NULL; 2298 char *name = NULL; 2299 int err; 2300 2301 err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len); 2302 if (err) { 2303 p_err("failed to get info for prog FD %d", tgt_fd); 2304 goto out; 2305 } 2306 2307 if (info.btf_id == 0) { 2308 p_err("prog FD %d doesn't have valid btf", tgt_fd); 2309 goto out; 2310 } 2311 2312 func_info_rec_size = info.func_info_rec_size; 2313 if (info.nr_func_info == 0) { 2314 p_err("found 0 func_info for prog FD %d", tgt_fd); 2315 goto out; 2316 } 2317 2318 memset(&info, 0, sizeof(info)); 2319 info.nr_func_info = 1; 2320 info.func_info_rec_size = func_info_rec_size; 2321 info.func_info = ptr_to_u64(&func_info); 2322 2323 err = bpf_prog_get_info_by_fd(tgt_fd, &info, &info_len); 2324 if (err) { 2325 p_err("failed to get func_info for prog FD %d", tgt_fd); 2326 goto out; 2327 } 2328 2329 btf = btf__load_from_kernel_by_id(info.btf_id); 2330 if (!btf) { 2331 p_err("failed to load btf for prog FD %d", tgt_fd); 2332 goto out; 2333 } 2334 2335 t = btf__type_by_id(btf, func_info.type_id); 2336 if (!t) { 2337 p_err("btf %u doesn't have type %u", 2338 info.btf_id, func_info.type_id); 2339 goto out; 2340 } 2341 name = strdup(btf__name_by_offset(btf, t->name_off)); 2342 out: 2343 btf__free(btf); 2344 return name; 2345 } 2346 2347 static struct profiler_bpf *profile_obj; 2348 static int profile_tgt_fd = -1; 2349 static char *profile_tgt_name; 2350 static int *profile_perf_events; 2351 static int profile_perf_event_cnt; 2352 2353 static void profile_close_perf_events(struct profiler_bpf *obj) 2354 { 2355 int i; 2356 2357 for (i = profile_perf_event_cnt - 1; i >= 0; i--) 2358 close(profile_perf_events[i]); 2359 2360 free(profile_perf_events); 2361 profile_perf_event_cnt = 0; 2362 } 2363 2364 static int profile_open_perf_event(int mid, int cpu, int map_fd) 2365 { 2366 int pmu_fd; 2367 2368 pmu_fd = syscall(__NR_perf_event_open, &metrics[mid].attr, 2369 -1 /*pid*/, cpu, -1 /*group_fd*/, 0); 2370 if (pmu_fd < 0) { 2371 if (errno == ENODEV) { 2372 p_info("cpu %d may be offline, skip %s profiling.", 2373 cpu, metrics[mid].name); 2374 profile_perf_event_cnt++; 2375 return 0; 2376 } 2377 return -1; 2378 } 2379 2380 if (bpf_map_update_elem(map_fd, 2381 &profile_perf_event_cnt, 2382 &pmu_fd, BPF_ANY) || 2383 ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0)) { 2384 close(pmu_fd); 2385 return -1; 2386 } 2387 2388 profile_perf_events[profile_perf_event_cnt++] = pmu_fd; 2389 return 0; 2390 } 2391 2392 static int profile_open_perf_events(struct profiler_bpf *obj) 2393 { 2394 unsigned int cpu, m; 2395 int map_fd; 2396 2397 profile_perf_events = calloc( 2398 obj->rodata->num_cpu * obj->rodata->num_metric, sizeof(int)); 2399 if (!profile_perf_events) { 2400 p_err("failed to allocate memory for perf_event array: %s", 2401 strerror(errno)); 2402 return -1; 2403 } 2404 map_fd = bpf_map__fd(obj->maps.events); 2405 if (map_fd < 0) { 2406 p_err("failed to get fd for events map"); 2407 return -1; 2408 } 2409 2410 for (m = 0; m < ARRAY_SIZE(metrics); m++) { 2411 if (!metrics[m].selected) 2412 continue; 2413 for (cpu = 0; cpu < obj->rodata->num_cpu; cpu++) { 2414 if (profile_open_perf_event(m, cpu, map_fd)) { 2415 p_err("failed to create event %s on cpu %u", 2416 metrics[m].name, cpu); 2417 return -1; 2418 } 2419 } 2420 } 2421 return 0; 2422 } 2423 2424 static void profile_print_and_cleanup(void) 2425 { 2426 profile_close_perf_events(profile_obj); 2427 profile_read_values(profile_obj); 2428 profile_print_readings(); 2429 profiler_bpf__destroy(profile_obj); 2430 2431 close(profile_tgt_fd); 2432 free(profile_tgt_name); 2433 } 2434 2435 static void int_exit(int signo) 2436 { 2437 profile_print_and_cleanup(); 2438 exit(0); 2439 } 2440 2441 static int do_profile(int argc, char **argv) 2442 { 2443 int num_metric, num_cpu, err = -1; 2444 struct bpf_program *prog; 2445 unsigned long duration; 2446 char *endptr; 2447 2448 /* we at least need two args for the prog and one metric */ 2449 if (!REQ_ARGS(3)) 2450 return -EINVAL; 2451 2452 /* parse target fd */ 2453 profile_tgt_fd = prog_parse_fd(&argc, &argv); 2454 if (profile_tgt_fd < 0) { 2455 p_err("failed to parse fd"); 2456 return -1; 2457 } 2458 2459 /* parse profiling optional duration */ 2460 if (argc > 2 && is_prefix(argv[0], "duration")) { 2461 NEXT_ARG(); 2462 duration = strtoul(*argv, &endptr, 0); 2463 if (*endptr) 2464 usage(); 2465 NEXT_ARG(); 2466 } else { 2467 duration = UINT_MAX; 2468 } 2469 2470 num_metric = profile_parse_metrics(argc, argv); 2471 if (num_metric <= 0) 2472 goto out; 2473 2474 num_cpu = libbpf_num_possible_cpus(); 2475 if (num_cpu <= 0) { 2476 p_err("failed to identify number of CPUs"); 2477 goto out; 2478 } 2479 2480 profile_obj = profiler_bpf__open(); 2481 if (!profile_obj) { 2482 p_err("failed to open and/or load BPF object"); 2483 goto out; 2484 } 2485 2486 profile_obj->rodata->num_cpu = num_cpu; 2487 profile_obj->rodata->num_metric = num_metric; 2488 2489 /* adjust map sizes */ 2490 bpf_map__set_max_entries(profile_obj->maps.events, num_metric * num_cpu); 2491 bpf_map__set_max_entries(profile_obj->maps.fentry_readings, num_metric); 2492 bpf_map__set_max_entries(profile_obj->maps.accum_readings, num_metric); 2493 bpf_map__set_max_entries(profile_obj->maps.counts, 1); 2494 2495 /* change target name */ 2496 profile_tgt_name = profile_target_name(profile_tgt_fd); 2497 if (!profile_tgt_name) 2498 goto out; 2499 2500 bpf_object__for_each_program(prog, profile_obj->obj) { 2501 err = bpf_program__set_attach_target(prog, profile_tgt_fd, 2502 profile_tgt_name); 2503 if (err) { 2504 p_err("failed to set attach target\n"); 2505 goto out; 2506 } 2507 } 2508 2509 set_max_rlimit(); 2510 err = profiler_bpf__load(profile_obj); 2511 if (err) { 2512 p_err("failed to load profile_obj"); 2513 goto out; 2514 } 2515 2516 err = profile_open_perf_events(profile_obj); 2517 if (err) 2518 goto out; 2519 2520 err = profiler_bpf__attach(profile_obj); 2521 if (err) { 2522 p_err("failed to attach profile_obj"); 2523 goto out; 2524 } 2525 signal(SIGINT, int_exit); 2526 2527 sleep(duration); 2528 profile_print_and_cleanup(); 2529 return 0; 2530 2531 out: 2532 profile_close_perf_events(profile_obj); 2533 if (profile_obj) 2534 profiler_bpf__destroy(profile_obj); 2535 close(profile_tgt_fd); 2536 free(profile_tgt_name); 2537 return err; 2538 } 2539 2540 #endif /* BPFTOOL_WITHOUT_SKELETONS */ 2541 2542 static int do_help(int argc, char **argv) 2543 { 2544 if (json_output) { 2545 jsonw_null(json_wtr); 2546 return 0; 2547 } 2548 2549 fprintf(stderr, 2550 "Usage: %1$s %2$s { show | list } [PROG]\n" 2551 " %1$s %2$s dump xlated PROG [{ file FILE | [opcodes] [linum] [visual] }]\n" 2552 " %1$s %2$s dump jited PROG [{ file FILE | [opcodes] [linum] }]\n" 2553 " %1$s %2$s pin PROG FILE\n" 2554 " %1$s %2$s { load | loadall } OBJ PATH \\\n" 2555 " [type TYPE] [{ offload_dev | xdpmeta_dev } NAME] \\\n" 2556 " [map { idx IDX | name NAME } MAP]\\\n" 2557 " [pinmaps MAP_DIR]\n" 2558 " [autoattach]\n" 2559 " [kernel_btf BTF_FILE]\n" 2560 " %1$s %2$s attach PROG ATTACH_TYPE [MAP]\n" 2561 " %1$s %2$s detach PROG ATTACH_TYPE [MAP]\n" 2562 " %1$s %2$s run PROG \\\n" 2563 " data_in FILE \\\n" 2564 " [data_out FILE [data_size_out L]] \\\n" 2565 " [ctx_in FILE [ctx_out FILE [ctx_size_out M]]] \\\n" 2566 " [repeat N]\n" 2567 " %1$s %2$s profile PROG [duration DURATION] METRICs\n" 2568 " %1$s %2$s tracelog\n" 2569 " %1$s %2$s tracelog { stdout | stderr } PROG\n" 2570 " %1$s %2$s help\n" 2571 "\n" 2572 " " HELP_SPEC_MAP "\n" 2573 " " HELP_SPEC_PROGRAM "\n" 2574 " TYPE := { socket | kprobe | kretprobe | classifier | action |\n" 2575 " tracepoint | raw_tracepoint | xdp | perf_event | cgroup/skb |\n" 2576 " cgroup/sock | cgroup/dev | lwt_in | lwt_out | lwt_xmit |\n" 2577 " lwt_seg6local | sockops | sk_skb | sk_msg | lirc_mode2 |\n" 2578 " sk_reuseport | flow_dissector | cgroup/sysctl |\n" 2579 " cgroup/bind4 | cgroup/bind6 | cgroup/post_bind4 |\n" 2580 " cgroup/post_bind6 | cgroup/connect4 | cgroup/connect6 |\n" 2581 " cgroup/connect_unix | cgroup/getpeername4 | cgroup/getpeername6 |\n" 2582 " cgroup/getpeername_unix | cgroup/getsockname4 | cgroup/getsockname6 |\n" 2583 " cgroup/getsockname_unix | cgroup/sendmsg4 | cgroup/sendmsg6 |\n" 2584 " cgroup/sendmsg_unix | cgroup/recvmsg4 | cgroup/recvmsg6 | cgroup/recvmsg_unix |\n" 2585 " cgroup/getsockopt | cgroup/setsockopt | cgroup/sock_release |\n" 2586 " struct_ops | fentry | fexit | freplace | sk_lookup }\n" 2587 " ATTACH_TYPE := { sk_msg_verdict | sk_skb_verdict | sk_skb_stream_verdict |\n" 2588 " sk_skb_stream_parser | flow_dissector }\n" 2589 " METRIC := { cycles | instructions | l1d_loads | llc_misses | itlb_misses | dtlb_misses }\n" 2590 " " HELP_SPEC_OPTIONS " |\n" 2591 " {-f|--bpffs} | {-m|--mapcompat} | {-n|--nomount} |\n" 2592 " {-L|--use-loader} | [ {-S|--sign } {-k} <private_key.pem> {-i} <certificate.x509> ] \n" 2593 "", 2594 bin_name, argv[-2]); 2595 2596 return 0; 2597 } 2598 2599 static const struct cmd cmds[] = { 2600 { "show", do_show }, 2601 { "list", do_show }, 2602 { "help", do_help }, 2603 { "dump", do_dump }, 2604 { "pin", do_pin }, 2605 { "load", do_load }, 2606 { "loadall", do_loadall }, 2607 { "attach", do_attach }, 2608 { "detach", do_detach }, 2609 { "tracelog", do_tracelog_any }, 2610 { "run", do_run }, 2611 { "profile", do_profile }, 2612 { 0 } 2613 }; 2614 2615 int do_prog(int argc, char **argv) 2616 { 2617 return cmd_select(cmds, argc, argv, do_help); 2618 } 2619