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