1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) 2 3 /* 4 * resolve_btfids scans Elf object for .BTF_ids section and resolves 5 * its symbols with BTF ID values. 6 * 7 * Each symbol points to 4 bytes data and is expected to have 8 * following name syntax: 9 * 10 * __BTF_ID__<type>__<symbol>[__<id>] 11 * 12 * type is: 13 * 14 * func - lookup BTF_KIND_FUNC symbol with <symbol> name 15 * and store its ID into the data: 16 * 17 * __BTF_ID__func__vfs_close__1: 18 * .zero 4 19 * 20 * struct - lookup BTF_KIND_STRUCT symbol with <symbol> name 21 * and store its ID into the data: 22 * 23 * __BTF_ID__struct__sk_buff__1: 24 * .zero 4 25 * 26 * union - lookup BTF_KIND_UNION symbol with <symbol> name 27 * and store its ID into the data: 28 * 29 * __BTF_ID__union__thread_union__1: 30 * .zero 4 31 * 32 * typedef - lookup BTF_KIND_TYPEDEF symbol with <symbol> name 33 * and store its ID into the data: 34 * 35 * __BTF_ID__typedef__pid_t__1: 36 * .zero 4 37 * 38 * set - store symbol size into first 4 bytes and sort following 39 * ID list 40 * 41 * __BTF_ID__set__list: 42 * .zero 4 43 * list: 44 * __BTF_ID__func__vfs_getattr__3: 45 * .zero 4 46 * __BTF_ID__func__vfs_fallocate__4: 47 * .zero 4 48 * 49 * set8 - store symbol size into first 4 bytes and sort following 50 * ID list 51 * 52 * __BTF_ID__set8__list: 53 * .zero 8 54 * list: 55 * __BTF_ID__func__vfs_getattr__3: 56 * .zero 4 57 * .word (1 << 0) | (1 << 2) 58 * __BTF_ID__func__vfs_fallocate__5: 59 * .zero 4 60 * .word (1 << 3) | (1 << 1) | (1 << 2) 61 */ 62 63 #define _GNU_SOURCE 64 #include <stdio.h> 65 #include <string.h> 66 #include <unistd.h> 67 #include <stdlib.h> 68 #include <libelf.h> 69 #include <gelf.h> 70 #include <sys/stat.h> 71 #include <fcntl.h> 72 #include <errno.h> 73 #include <linux/rbtree.h> 74 #include <linux/zalloc.h> 75 #include <linux/err.h> 76 #include <bpf/btf.h> 77 #include <bpf/libbpf.h> 78 #include <parse-options.h> 79 80 #define BTF_IDS_SECTION ".BTF_ids" 81 #define BTF_ID "__BTF_ID__" 82 83 #define BTF_STRUCT "struct" 84 #define BTF_UNION "union" 85 #define BTF_TYPEDEF "typedef" 86 #define BTF_FUNC "func" 87 #define BTF_SET "set" 88 #define BTF_SET8 "set8" 89 90 #define ADDR_CNT 100 91 92 struct btf_id { 93 struct rb_node rb_node; 94 char *name; 95 union { 96 int id; 97 int cnt; 98 }; 99 int addr_cnt; 100 bool is_set; 101 bool is_set8; 102 Elf64_Addr addr[ADDR_CNT]; 103 }; 104 105 struct object { 106 const char *path; 107 const char *btf; 108 const char *base_btf_path; 109 110 struct { 111 int fd; 112 Elf *elf; 113 Elf_Data *symbols; 114 Elf_Data *idlist; 115 int symbols_shndx; 116 int idlist_shndx; 117 size_t strtabidx; 118 unsigned long idlist_addr; 119 } efile; 120 121 struct rb_root sets; 122 struct rb_root structs; 123 struct rb_root unions; 124 struct rb_root typedefs; 125 struct rb_root funcs; 126 127 int nr_funcs; 128 int nr_structs; 129 int nr_unions; 130 int nr_typedefs; 131 }; 132 133 static int verbose; 134 135 static int eprintf(int level, int var, const char *fmt, ...) 136 { 137 va_list args; 138 int ret = 0; 139 140 if (var >= level) { 141 va_start(args, fmt); 142 ret = vfprintf(stderr, fmt, args); 143 va_end(args); 144 } 145 return ret; 146 } 147 148 #ifndef pr_fmt 149 #define pr_fmt(fmt) fmt 150 #endif 151 152 #define pr_debug(fmt, ...) \ 153 eprintf(1, verbose, pr_fmt(fmt), ##__VA_ARGS__) 154 #define pr_debugN(n, fmt, ...) \ 155 eprintf(n, verbose, pr_fmt(fmt), ##__VA_ARGS__) 156 #define pr_debug2(fmt, ...) pr_debugN(2, pr_fmt(fmt), ##__VA_ARGS__) 157 #define pr_err(fmt, ...) \ 158 eprintf(0, verbose, pr_fmt(fmt), ##__VA_ARGS__) 159 #define pr_info(fmt, ...) \ 160 eprintf(0, verbose, pr_fmt(fmt), ##__VA_ARGS__) 161 162 static bool is_btf_id(const char *name) 163 { 164 return name && !strncmp(name, BTF_ID, sizeof(BTF_ID) - 1); 165 } 166 167 static struct btf_id *btf_id__find(struct rb_root *root, const char *name) 168 { 169 struct rb_node *p = root->rb_node; 170 struct btf_id *id; 171 int cmp; 172 173 while (p) { 174 id = rb_entry(p, struct btf_id, rb_node); 175 cmp = strcmp(id->name, name); 176 if (cmp < 0) 177 p = p->rb_left; 178 else if (cmp > 0) 179 p = p->rb_right; 180 else 181 return id; 182 } 183 return NULL; 184 } 185 186 static struct btf_id * 187 btf_id__add(struct rb_root *root, char *name, bool unique) 188 { 189 struct rb_node **p = &root->rb_node; 190 struct rb_node *parent = NULL; 191 struct btf_id *id; 192 int cmp; 193 194 while (*p != NULL) { 195 parent = *p; 196 id = rb_entry(parent, struct btf_id, rb_node); 197 cmp = strcmp(id->name, name); 198 if (cmp < 0) 199 p = &(*p)->rb_left; 200 else if (cmp > 0) 201 p = &(*p)->rb_right; 202 else 203 return unique ? NULL : id; 204 } 205 206 id = zalloc(sizeof(*id)); 207 if (id) { 208 pr_debug("adding symbol %s\n", name); 209 id->name = name; 210 rb_link_node(&id->rb_node, parent, p); 211 rb_insert_color(&id->rb_node, root); 212 } 213 return id; 214 } 215 216 static char *get_id(const char *prefix_end) 217 { 218 /* 219 * __BTF_ID__func__vfs_truncate__0 220 * prefix_end = ^ 221 * pos = ^ 222 */ 223 int len = strlen(prefix_end); 224 int pos = sizeof("__") - 1; 225 char *p, *id; 226 227 if (pos >= len) 228 return NULL; 229 230 id = strdup(prefix_end + pos); 231 if (id) { 232 /* 233 * __BTF_ID__func__vfs_truncate__0 234 * id = ^ 235 * 236 * cut the unique id part 237 */ 238 p = strrchr(id, '_'); 239 p--; 240 if (*p != '_') { 241 free(id); 242 return NULL; 243 } 244 *p = '\0'; 245 } 246 return id; 247 } 248 249 static struct btf_id *add_set(struct object *obj, char *name, bool is_set8) 250 { 251 /* 252 * __BTF_ID__set__name 253 * name = ^ 254 * id = ^ 255 */ 256 char *id = name + (is_set8 ? sizeof(BTF_SET8 "__") : sizeof(BTF_SET "__")) - 1; 257 int len = strlen(name); 258 259 if (id >= name + len) { 260 pr_err("FAILED to parse set name: %s\n", name); 261 return NULL; 262 } 263 264 return btf_id__add(&obj->sets, id, true); 265 } 266 267 static struct btf_id *add_symbol(struct rb_root *root, char *name, size_t size) 268 { 269 char *id; 270 271 id = get_id(name + size); 272 if (!id) { 273 pr_err("FAILED to parse symbol name: %s\n", name); 274 return NULL; 275 } 276 277 return btf_id__add(root, id, false); 278 } 279 280 /* Older libelf.h and glibc elf.h might not yet define the ELF compression types. */ 281 #ifndef SHF_COMPRESSED 282 #define SHF_COMPRESSED (1 << 11) /* Section with compressed data. */ 283 #endif 284 285 /* 286 * The data of compressed section should be aligned to 4 287 * (for 32bit) or 8 (for 64 bit) bytes. The binutils ld 288 * sets sh_addralign to 1, which makes libelf fail with 289 * misaligned section error during the update: 290 * FAILED elf_update(WRITE): invalid section alignment 291 * 292 * While waiting for ld fix, we fix the compressed sections 293 * sh_addralign value manualy. 294 */ 295 static int compressed_section_fix(Elf *elf, Elf_Scn *scn, GElf_Shdr *sh) 296 { 297 int expected = gelf_getclass(elf) == ELFCLASS32 ? 4 : 8; 298 299 if (!(sh->sh_flags & SHF_COMPRESSED)) 300 return 0; 301 302 if (sh->sh_addralign == expected) 303 return 0; 304 305 pr_debug2(" - fixing wrong alignment sh_addralign %u, expected %u\n", 306 sh->sh_addralign, expected); 307 308 sh->sh_addralign = expected; 309 310 if (gelf_update_shdr(scn, sh) == 0) { 311 pr_err("FAILED cannot update section header: %s\n", 312 elf_errmsg(-1)); 313 return -1; 314 } 315 return 0; 316 } 317 318 static int elf_collect(struct object *obj) 319 { 320 Elf_Scn *scn = NULL; 321 size_t shdrstrndx; 322 int idx = 0; 323 Elf *elf; 324 int fd; 325 326 fd = open(obj->path, O_RDWR, 0666); 327 if (fd == -1) { 328 pr_err("FAILED cannot open %s: %s\n", 329 obj->path, strerror(errno)); 330 return -1; 331 } 332 333 elf_version(EV_CURRENT); 334 335 elf = elf_begin(fd, ELF_C_RDWR_MMAP, NULL); 336 if (!elf) { 337 close(fd); 338 pr_err("FAILED cannot create ELF descriptor: %s\n", 339 elf_errmsg(-1)); 340 return -1; 341 } 342 343 obj->efile.fd = fd; 344 obj->efile.elf = elf; 345 346 elf_flagelf(elf, ELF_C_SET, ELF_F_LAYOUT); 347 348 if (elf_getshdrstrndx(elf, &shdrstrndx) != 0) { 349 pr_err("FAILED cannot get shdr str ndx\n"); 350 return -1; 351 } 352 353 /* 354 * Scan all the elf sections and look for save data 355 * from .BTF_ids section and symbols. 356 */ 357 while ((scn = elf_nextscn(elf, scn)) != NULL) { 358 Elf_Data *data; 359 GElf_Shdr sh; 360 char *name; 361 362 idx++; 363 if (gelf_getshdr(scn, &sh) != &sh) { 364 pr_err("FAILED get section(%d) header\n", idx); 365 return -1; 366 } 367 368 name = elf_strptr(elf, shdrstrndx, sh.sh_name); 369 if (!name) { 370 pr_err("FAILED get section(%d) name\n", idx); 371 return -1; 372 } 373 374 data = elf_getdata(scn, 0); 375 if (!data) { 376 pr_err("FAILED to get section(%d) data from %s\n", 377 idx, name); 378 return -1; 379 } 380 381 pr_debug2("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n", 382 idx, name, (unsigned long) data->d_size, 383 (int) sh.sh_link, (unsigned long) sh.sh_flags, 384 (int) sh.sh_type); 385 386 if (sh.sh_type == SHT_SYMTAB) { 387 obj->efile.symbols = data; 388 obj->efile.symbols_shndx = idx; 389 obj->efile.strtabidx = sh.sh_link; 390 } else if (!strcmp(name, BTF_IDS_SECTION)) { 391 obj->efile.idlist = data; 392 obj->efile.idlist_shndx = idx; 393 obj->efile.idlist_addr = sh.sh_addr; 394 } 395 396 if (compressed_section_fix(elf, scn, &sh)) 397 return -1; 398 } 399 400 return 0; 401 } 402 403 static int symbols_collect(struct object *obj) 404 { 405 Elf_Scn *scn = NULL; 406 int n, i; 407 GElf_Shdr sh; 408 char *name; 409 410 scn = elf_getscn(obj->efile.elf, obj->efile.symbols_shndx); 411 if (!scn) 412 return -1; 413 414 if (gelf_getshdr(scn, &sh) != &sh) 415 return -1; 416 417 n = sh.sh_size / sh.sh_entsize; 418 419 /* 420 * Scan symbols and look for the ones starting with 421 * __BTF_ID__* over .BTF_ids section. 422 */ 423 for (i = 0; i < n; i++) { 424 char *prefix; 425 struct btf_id *id; 426 GElf_Sym sym; 427 428 if (!gelf_getsym(obj->efile.symbols, i, &sym)) 429 return -1; 430 431 if (sym.st_shndx != obj->efile.idlist_shndx) 432 continue; 433 434 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, 435 sym.st_name); 436 437 if (!is_btf_id(name)) 438 continue; 439 440 /* 441 * __BTF_ID__TYPE__vfs_truncate__0 442 * prefix = ^ 443 */ 444 prefix = name + sizeof(BTF_ID) - 1; 445 446 /* struct */ 447 if (!strncmp(prefix, BTF_STRUCT, sizeof(BTF_STRUCT) - 1)) { 448 obj->nr_structs++; 449 id = add_symbol(&obj->structs, prefix, sizeof(BTF_STRUCT) - 1); 450 /* union */ 451 } else if (!strncmp(prefix, BTF_UNION, sizeof(BTF_UNION) - 1)) { 452 obj->nr_unions++; 453 id = add_symbol(&obj->unions, prefix, sizeof(BTF_UNION) - 1); 454 /* typedef */ 455 } else if (!strncmp(prefix, BTF_TYPEDEF, sizeof(BTF_TYPEDEF) - 1)) { 456 obj->nr_typedefs++; 457 id = add_symbol(&obj->typedefs, prefix, sizeof(BTF_TYPEDEF) - 1); 458 /* func */ 459 } else if (!strncmp(prefix, BTF_FUNC, sizeof(BTF_FUNC) - 1)) { 460 obj->nr_funcs++; 461 id = add_symbol(&obj->funcs, prefix, sizeof(BTF_FUNC) - 1); 462 /* set8 */ 463 } else if (!strncmp(prefix, BTF_SET8, sizeof(BTF_SET8) - 1)) { 464 id = add_set(obj, prefix, true); 465 /* 466 * SET8 objects store list's count, which is encoded 467 * in symbol's size, together with 'cnt' field hence 468 * that - 1. 469 */ 470 if (id) { 471 id->cnt = sym.st_size / sizeof(uint64_t) - 1; 472 id->is_set8 = true; 473 } 474 /* set */ 475 } else if (!strncmp(prefix, BTF_SET, sizeof(BTF_SET) - 1)) { 476 id = add_set(obj, prefix, false); 477 /* 478 * SET objects store list's count, which is encoded 479 * in symbol's size, together with 'cnt' field hence 480 * that - 1. 481 */ 482 if (id) { 483 id->cnt = sym.st_size / sizeof(int) - 1; 484 id->is_set = true; 485 } 486 } else { 487 pr_err("FAILED unsupported prefix %s\n", prefix); 488 return -1; 489 } 490 491 if (!id) 492 return -ENOMEM; 493 494 if (id->addr_cnt >= ADDR_CNT) { 495 pr_err("FAILED symbol %s crossed the number of allowed lists\n", 496 id->name); 497 return -1; 498 } 499 id->addr[id->addr_cnt++] = sym.st_value; 500 } 501 502 return 0; 503 } 504 505 static int symbols_resolve(struct object *obj) 506 { 507 int nr_typedefs = obj->nr_typedefs; 508 int nr_structs = obj->nr_structs; 509 int nr_unions = obj->nr_unions; 510 int nr_funcs = obj->nr_funcs; 511 struct btf *base_btf = NULL; 512 int err, type_id; 513 struct btf *btf; 514 __u32 nr_types; 515 516 if (obj->base_btf_path) { 517 base_btf = btf__parse(obj->base_btf_path, NULL); 518 err = libbpf_get_error(base_btf); 519 if (err) { 520 pr_err("FAILED: load base BTF from %s: %s\n", 521 obj->base_btf_path, strerror(-err)); 522 return -1; 523 } 524 } 525 526 btf = btf__parse_split(obj->btf ?: obj->path, base_btf); 527 err = libbpf_get_error(btf); 528 if (err) { 529 pr_err("FAILED: load BTF from %s: %s\n", 530 obj->btf ?: obj->path, strerror(-err)); 531 goto out; 532 } 533 534 err = -1; 535 nr_types = btf__type_cnt(btf); 536 537 /* 538 * Iterate all the BTF types and search for collected symbol IDs. 539 */ 540 for (type_id = 1; type_id < nr_types; type_id++) { 541 const struct btf_type *type; 542 struct rb_root *root; 543 struct btf_id *id; 544 const char *str; 545 int *nr; 546 547 type = btf__type_by_id(btf, type_id); 548 if (!type) { 549 pr_err("FAILED: malformed BTF, can't resolve type for ID %d\n", 550 type_id); 551 goto out; 552 } 553 554 if (btf_is_func(type) && nr_funcs) { 555 nr = &nr_funcs; 556 root = &obj->funcs; 557 } else if (btf_is_struct(type) && nr_structs) { 558 nr = &nr_structs; 559 root = &obj->structs; 560 } else if (btf_is_union(type) && nr_unions) { 561 nr = &nr_unions; 562 root = &obj->unions; 563 } else if (btf_is_typedef(type) && nr_typedefs) { 564 nr = &nr_typedefs; 565 root = &obj->typedefs; 566 } else 567 continue; 568 569 str = btf__name_by_offset(btf, type->name_off); 570 if (!str) { 571 pr_err("FAILED: malformed BTF, can't resolve name for ID %d\n", 572 type_id); 573 goto out; 574 } 575 576 id = btf_id__find(root, str); 577 if (id) { 578 if (id->id) { 579 pr_info("WARN: multiple IDs found for '%s': %d, %d - using %d\n", 580 str, id->id, type_id, id->id); 581 } else { 582 id->id = type_id; 583 (*nr)--; 584 } 585 } 586 } 587 588 err = 0; 589 out: 590 btf__free(base_btf); 591 btf__free(btf); 592 return err; 593 } 594 595 static int id_patch(struct object *obj, struct btf_id *id) 596 { 597 Elf_Data *data = obj->efile.idlist; 598 int *ptr = data->d_buf; 599 int i; 600 601 /* For set, set8, id->id may be 0 */ 602 if (!id->id && !id->is_set && !id->is_set8) 603 pr_err("WARN: resolve_btfids: unresolved symbol %s\n", id->name); 604 605 for (i = 0; i < id->addr_cnt; i++) { 606 unsigned long addr = id->addr[i]; 607 unsigned long idx = addr - obj->efile.idlist_addr; 608 609 pr_debug("patching addr %5lu: ID %7d [%s]\n", 610 idx, id->id, id->name); 611 612 if (idx >= data->d_size) { 613 pr_err("FAILED patching index %lu out of bounds %lu\n", 614 idx, data->d_size); 615 return -1; 616 } 617 618 idx = idx / sizeof(int); 619 ptr[idx] = id->id; 620 } 621 622 return 0; 623 } 624 625 static int __symbols_patch(struct object *obj, struct rb_root *root) 626 { 627 struct rb_node *next; 628 struct btf_id *id; 629 630 next = rb_first(root); 631 while (next) { 632 id = rb_entry(next, struct btf_id, rb_node); 633 634 if (id_patch(obj, id)) 635 return -1; 636 637 next = rb_next(next); 638 } 639 return 0; 640 } 641 642 static int cmp_id(const void *pa, const void *pb) 643 { 644 const int *a = pa, *b = pb; 645 646 return *a - *b; 647 } 648 649 static int sets_patch(struct object *obj) 650 { 651 Elf_Data *data = obj->efile.idlist; 652 int *ptr = data->d_buf; 653 struct rb_node *next; 654 655 next = rb_first(&obj->sets); 656 while (next) { 657 unsigned long addr, idx; 658 struct btf_id *id; 659 int *base; 660 int cnt; 661 662 id = rb_entry(next, struct btf_id, rb_node); 663 addr = id->addr[0]; 664 idx = addr - obj->efile.idlist_addr; 665 666 /* sets are unique */ 667 if (id->addr_cnt != 1) { 668 pr_err("FAILED malformed data for set '%s'\n", 669 id->name); 670 return -1; 671 } 672 673 idx = idx / sizeof(int); 674 base = &ptr[idx] + (id->is_set8 ? 2 : 1); 675 cnt = ptr[idx]; 676 677 pr_debug("sorting addr %5lu: cnt %6d [%s]\n", 678 (idx + 1) * sizeof(int), cnt, id->name); 679 680 qsort(base, cnt, id->is_set8 ? sizeof(uint64_t) : sizeof(int), cmp_id); 681 682 next = rb_next(next); 683 } 684 return 0; 685 } 686 687 static int symbols_patch(struct object *obj) 688 { 689 int err; 690 691 if (__symbols_patch(obj, &obj->structs) || 692 __symbols_patch(obj, &obj->unions) || 693 __symbols_patch(obj, &obj->typedefs) || 694 __symbols_patch(obj, &obj->funcs) || 695 __symbols_patch(obj, &obj->sets)) 696 return -1; 697 698 if (sets_patch(obj)) 699 return -1; 700 701 /* Set type to ensure endian translation occurs. */ 702 obj->efile.idlist->d_type = ELF_T_WORD; 703 704 elf_flagdata(obj->efile.idlist, ELF_C_SET, ELF_F_DIRTY); 705 706 err = elf_update(obj->efile.elf, ELF_C_WRITE); 707 if (err < 0) { 708 pr_err("FAILED elf_update(WRITE): %s\n", 709 elf_errmsg(-1)); 710 } 711 712 pr_debug("update %s for %s\n", 713 err >= 0 ? "ok" : "failed", obj->path); 714 return err < 0 ? -1 : 0; 715 } 716 717 static const char * const resolve_btfids_usage[] = { 718 "resolve_btfids [<options>] <ELF object>", 719 NULL 720 }; 721 722 int main(int argc, const char **argv) 723 { 724 struct object obj = { 725 .efile = { 726 .idlist_shndx = -1, 727 .symbols_shndx = -1, 728 }, 729 .structs = RB_ROOT, 730 .unions = RB_ROOT, 731 .typedefs = RB_ROOT, 732 .funcs = RB_ROOT, 733 .sets = RB_ROOT, 734 }; 735 struct option btfid_options[] = { 736 OPT_INCR('v', "verbose", &verbose, 737 "be more verbose (show errors, etc)"), 738 OPT_STRING(0, "btf", &obj.btf, "BTF data", 739 "BTF data"), 740 OPT_STRING('b', "btf_base", &obj.base_btf_path, "file", 741 "path of file providing base BTF"), 742 OPT_END() 743 }; 744 int err = -1; 745 746 argc = parse_options(argc, argv, btfid_options, resolve_btfids_usage, 747 PARSE_OPT_STOP_AT_NON_OPTION); 748 if (argc != 1) 749 usage_with_options(resolve_btfids_usage, btfid_options); 750 751 obj.path = argv[0]; 752 753 if (elf_collect(&obj)) 754 goto out; 755 756 /* 757 * We did not find .BTF_ids section or symbols section, 758 * nothing to do.. 759 */ 760 if (obj.efile.idlist_shndx == -1 || 761 obj.efile.symbols_shndx == -1) { 762 pr_debug("Cannot find .BTF_ids or symbols sections, nothing to do\n"); 763 err = 0; 764 goto out; 765 } 766 767 if (symbols_collect(&obj)) 768 goto out; 769 770 if (symbols_resolve(&obj)) 771 goto out; 772 773 if (symbols_patch(&obj)) 774 goto out; 775 776 err = 0; 777 out: 778 if (obj.efile.elf) { 779 elf_end(obj.efile.elf); 780 close(obj.efile.fd); 781 } 782 return err; 783 } 784