1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * elf.c - ELF access library 4 * 5 * Adapted from kpatch (https://github.com/dynup/kpatch): 6 * Copyright (C) 2013-2015 Josh Poimboeuf <jpoimboe@redhat.com> 7 * Copyright (C) 2014 Seth Jennings <sjenning@redhat.com> 8 */ 9 10 #include <sys/types.h> 11 #include <sys/stat.h> 12 #include <sys/mman.h> 13 #include <fcntl.h> 14 #include <stdio.h> 15 #include <stdlib.h> 16 #include <string.h> 17 #include <unistd.h> 18 #include <errno.h> 19 #include <linux/interval_tree_generic.h> 20 #include <objtool/builtin.h> 21 22 #include <objtool/elf.h> 23 #include <objtool/warn.h> 24 25 #define MAX_NAME_LEN 128 26 27 static inline u32 str_hash(const char *str) 28 { 29 return jhash(str, strlen(str), 0); 30 } 31 32 #define __elf_table(name) (elf->name##_hash) 33 #define __elf_bits(name) (elf->name##_bits) 34 35 #define elf_hash_add(name, node, key) \ 36 hlist_add_head(node, &__elf_table(name)[hash_min(key, __elf_bits(name))]) 37 38 #define elf_hash_for_each_possible(name, obj, member, key) \ 39 hlist_for_each_entry(obj, &__elf_table(name)[hash_min(key, __elf_bits(name))], member) 40 41 #define elf_alloc_hash(name, size) \ 42 ({ \ 43 __elf_bits(name) = max(10, ilog2(size)); \ 44 __elf_table(name) = mmap(NULL, sizeof(struct hlist_head) << __elf_bits(name), \ 45 PROT_READ|PROT_WRITE, \ 46 MAP_PRIVATE|MAP_ANON, -1, 0); \ 47 if (__elf_table(name) == (void *)-1L) { \ 48 WARN("mmap fail " #name); \ 49 __elf_table(name) = NULL; \ 50 } \ 51 __elf_table(name); \ 52 }) 53 54 static inline unsigned long __sym_start(struct symbol *s) 55 { 56 return s->offset; 57 } 58 59 static inline unsigned long __sym_last(struct symbol *s) 60 { 61 return s->offset + s->len - 1; 62 } 63 64 INTERVAL_TREE_DEFINE(struct symbol, node, unsigned long, __subtree_last, 65 __sym_start, __sym_last, static, __sym) 66 67 #define __sym_for_each(_iter, _tree, _start, _end) \ 68 for (_iter = __sym_iter_first((_tree), (_start), (_end)); \ 69 _iter; _iter = __sym_iter_next(_iter, (_start), (_end))) 70 71 struct symbol_hole { 72 unsigned long key; 73 const struct symbol *sym; 74 }; 75 76 /* 77 * Find !section symbol where @offset is after it. 78 */ 79 static int symbol_hole_by_offset(const void *key, const struct rb_node *node) 80 { 81 const struct symbol *s = rb_entry(node, struct symbol, node); 82 struct symbol_hole *sh = (void *)key; 83 84 if (sh->key < s->offset) 85 return -1; 86 87 if (sh->key >= s->offset + s->len) { 88 if (s->type != STT_SECTION) 89 sh->sym = s; 90 return 1; 91 } 92 93 return 0; 94 } 95 96 struct section *find_section_by_name(const struct elf *elf, const char *name) 97 { 98 struct section *sec; 99 100 elf_hash_for_each_possible(section_name, sec, name_hash, str_hash(name)) { 101 if (!strcmp(sec->name, name)) 102 return sec; 103 } 104 105 return NULL; 106 } 107 108 static struct section *find_section_by_index(struct elf *elf, 109 unsigned int idx) 110 { 111 struct section *sec; 112 113 elf_hash_for_each_possible(section, sec, hash, idx) { 114 if (sec->idx == idx) 115 return sec; 116 } 117 118 return NULL; 119 } 120 121 static struct symbol *find_symbol_by_index(struct elf *elf, unsigned int idx) 122 { 123 struct symbol *sym; 124 125 elf_hash_for_each_possible(symbol, sym, hash, idx) { 126 if (sym->idx == idx) 127 return sym; 128 } 129 130 return NULL; 131 } 132 133 struct symbol *find_symbol_by_offset(struct section *sec, unsigned long offset) 134 { 135 struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree; 136 struct symbol *iter; 137 138 __sym_for_each(iter, tree, offset, offset) { 139 if (iter->offset == offset && iter->type != STT_SECTION) 140 return iter; 141 } 142 143 return NULL; 144 } 145 146 struct symbol *find_func_by_offset(struct section *sec, unsigned long offset) 147 { 148 struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree; 149 struct symbol *iter; 150 151 __sym_for_each(iter, tree, offset, offset) { 152 if (iter->offset == offset && iter->type == STT_FUNC) 153 return iter; 154 } 155 156 return NULL; 157 } 158 159 struct symbol *find_symbol_containing(const struct section *sec, unsigned long offset) 160 { 161 struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree; 162 struct symbol *iter; 163 164 __sym_for_each(iter, tree, offset, offset) { 165 if (iter->type != STT_SECTION) 166 return iter; 167 } 168 169 return NULL; 170 } 171 172 /* 173 * Returns size of hole starting at @offset. 174 */ 175 int find_symbol_hole_containing(const struct section *sec, unsigned long offset) 176 { 177 struct symbol_hole hole = { 178 .key = offset, 179 .sym = NULL, 180 }; 181 struct rb_node *n; 182 struct symbol *s; 183 184 /* 185 * Find the rightmost symbol for which @offset is after it. 186 */ 187 n = rb_find(&hole, &sec->symbol_tree.rb_root, symbol_hole_by_offset); 188 189 /* found a symbol that contains @offset */ 190 if (n) 191 return 0; /* not a hole */ 192 193 /* didn't find a symbol for which @offset is after it */ 194 if (!hole.sym) 195 return 0; /* not a hole */ 196 197 /* @offset >= sym->offset + sym->len, find symbol after it */ 198 n = rb_next(&hole.sym->node); 199 if (!n) 200 return -1; /* until end of address space */ 201 202 /* hole until start of next symbol */ 203 s = rb_entry(n, struct symbol, node); 204 return s->offset - offset; 205 } 206 207 struct symbol *find_func_containing(struct section *sec, unsigned long offset) 208 { 209 struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree; 210 struct symbol *iter; 211 212 __sym_for_each(iter, tree, offset, offset) { 213 if (iter->type == STT_FUNC) 214 return iter; 215 } 216 217 return NULL; 218 } 219 220 struct symbol *find_symbol_by_name(const struct elf *elf, const char *name) 221 { 222 struct symbol *sym; 223 224 elf_hash_for_each_possible(symbol_name, sym, name_hash, str_hash(name)) { 225 if (!strcmp(sym->name, name)) 226 return sym; 227 } 228 229 return NULL; 230 } 231 232 struct reloc *find_reloc_by_dest_range(const struct elf *elf, struct section *sec, 233 unsigned long offset, unsigned int len) 234 { 235 struct reloc *reloc, *r = NULL; 236 unsigned long o; 237 238 if (!sec->reloc) 239 return NULL; 240 241 sec = sec->reloc; 242 243 for_offset_range(o, offset, offset + len) { 244 elf_hash_for_each_possible(reloc, reloc, hash, 245 sec_offset_hash(sec, o)) { 246 if (reloc->sec != sec) 247 continue; 248 249 if (reloc->offset >= offset && reloc->offset < offset + len) { 250 if (!r || reloc->offset < r->offset) 251 r = reloc; 252 } 253 } 254 if (r) 255 return r; 256 } 257 258 return NULL; 259 } 260 261 struct reloc *find_reloc_by_dest(const struct elf *elf, struct section *sec, unsigned long offset) 262 { 263 return find_reloc_by_dest_range(elf, sec, offset, 1); 264 } 265 266 static int read_sections(struct elf *elf) 267 { 268 Elf_Scn *s = NULL; 269 struct section *sec; 270 size_t shstrndx, sections_nr; 271 int i; 272 273 if (elf_getshdrnum(elf->elf, §ions_nr)) { 274 WARN_ELF("elf_getshdrnum"); 275 return -1; 276 } 277 278 if (elf_getshdrstrndx(elf->elf, &shstrndx)) { 279 WARN_ELF("elf_getshdrstrndx"); 280 return -1; 281 } 282 283 if (!elf_alloc_hash(section, sections_nr) || 284 !elf_alloc_hash(section_name, sections_nr)) 285 return -1; 286 287 elf->section_data = calloc(sections_nr, sizeof(*sec)); 288 if (!elf->section_data) { 289 perror("calloc"); 290 return -1; 291 } 292 for (i = 0; i < sections_nr; i++) { 293 sec = &elf->section_data[i]; 294 295 INIT_LIST_HEAD(&sec->symbol_list); 296 INIT_LIST_HEAD(&sec->reloc_list); 297 298 s = elf_getscn(elf->elf, i); 299 if (!s) { 300 WARN_ELF("elf_getscn"); 301 return -1; 302 } 303 304 sec->idx = elf_ndxscn(s); 305 306 if (!gelf_getshdr(s, &sec->sh)) { 307 WARN_ELF("gelf_getshdr"); 308 return -1; 309 } 310 311 sec->name = elf_strptr(elf->elf, shstrndx, sec->sh.sh_name); 312 if (!sec->name) { 313 WARN_ELF("elf_strptr"); 314 return -1; 315 } 316 317 if (sec->sh.sh_size != 0) { 318 sec->data = elf_getdata(s, NULL); 319 if (!sec->data) { 320 WARN_ELF("elf_getdata"); 321 return -1; 322 } 323 if (sec->data->d_off != 0 || 324 sec->data->d_size != sec->sh.sh_size) { 325 WARN("unexpected data attributes for %s", 326 sec->name); 327 return -1; 328 } 329 } 330 331 if (sec->sh.sh_flags & SHF_EXECINSTR) 332 elf->text_size += sec->sh.sh_size; 333 334 list_add_tail(&sec->list, &elf->sections); 335 elf_hash_add(section, &sec->hash, sec->idx); 336 elf_hash_add(section_name, &sec->name_hash, str_hash(sec->name)); 337 } 338 339 if (opts.stats) { 340 printf("nr_sections: %lu\n", (unsigned long)sections_nr); 341 printf("section_bits: %d\n", elf->section_bits); 342 } 343 344 /* sanity check, one more call to elf_nextscn() should return NULL */ 345 if (elf_nextscn(elf->elf, s)) { 346 WARN("section entry mismatch"); 347 return -1; 348 } 349 350 return 0; 351 } 352 353 static void elf_add_symbol(struct elf *elf, struct symbol *sym) 354 { 355 struct list_head *entry; 356 struct rb_node *pnode; 357 struct symbol *iter; 358 359 INIT_LIST_HEAD(&sym->reloc_list); 360 INIT_LIST_HEAD(&sym->pv_target); 361 sym->alias = sym; 362 363 sym->type = GELF_ST_TYPE(sym->sym.st_info); 364 sym->bind = GELF_ST_BIND(sym->sym.st_info); 365 366 if (sym->type == STT_FILE) 367 elf->num_files++; 368 369 sym->offset = sym->sym.st_value; 370 sym->len = sym->sym.st_size; 371 372 __sym_for_each(iter, &sym->sec->symbol_tree, sym->offset, sym->offset) { 373 if (iter->offset == sym->offset && iter->type == sym->type) 374 iter->alias = sym; 375 } 376 377 __sym_insert(sym, &sym->sec->symbol_tree); 378 pnode = rb_prev(&sym->node); 379 if (pnode) 380 entry = &rb_entry(pnode, struct symbol, node)->list; 381 else 382 entry = &sym->sec->symbol_list; 383 list_add(&sym->list, entry); 384 elf_hash_add(symbol, &sym->hash, sym->idx); 385 elf_hash_add(symbol_name, &sym->name_hash, str_hash(sym->name)); 386 387 /* 388 * Don't store empty STT_NOTYPE symbols in the rbtree. They 389 * can exist within a function, confusing the sorting. 390 */ 391 if (!sym->len) 392 __sym_remove(sym, &sym->sec->symbol_tree); 393 } 394 395 static int read_symbols(struct elf *elf) 396 { 397 struct section *symtab, *symtab_shndx, *sec; 398 struct symbol *sym, *pfunc; 399 int symbols_nr, i; 400 char *coldstr; 401 Elf_Data *shndx_data = NULL; 402 Elf32_Word shndx; 403 404 symtab = find_section_by_name(elf, ".symtab"); 405 if (symtab) { 406 symtab_shndx = find_section_by_name(elf, ".symtab_shndx"); 407 if (symtab_shndx) 408 shndx_data = symtab_shndx->data; 409 410 symbols_nr = symtab->sh.sh_size / symtab->sh.sh_entsize; 411 } else { 412 /* 413 * A missing symbol table is actually possible if it's an empty 414 * .o file. This can happen for thunk_64.o. Make sure to at 415 * least allocate the symbol hash tables so we can do symbol 416 * lookups without crashing. 417 */ 418 symbols_nr = 0; 419 } 420 421 if (!elf_alloc_hash(symbol, symbols_nr) || 422 !elf_alloc_hash(symbol_name, symbols_nr)) 423 return -1; 424 425 elf->symbol_data = calloc(symbols_nr, sizeof(*sym)); 426 if (!elf->symbol_data) { 427 perror("calloc"); 428 return -1; 429 } 430 for (i = 0; i < symbols_nr; i++) { 431 sym = &elf->symbol_data[i]; 432 433 sym->idx = i; 434 435 if (!gelf_getsymshndx(symtab->data, shndx_data, i, &sym->sym, 436 &shndx)) { 437 WARN_ELF("gelf_getsymshndx"); 438 goto err; 439 } 440 441 sym->name = elf_strptr(elf->elf, symtab->sh.sh_link, 442 sym->sym.st_name); 443 if (!sym->name) { 444 WARN_ELF("elf_strptr"); 445 goto err; 446 } 447 448 if ((sym->sym.st_shndx > SHN_UNDEF && 449 sym->sym.st_shndx < SHN_LORESERVE) || 450 (shndx_data && sym->sym.st_shndx == SHN_XINDEX)) { 451 if (sym->sym.st_shndx != SHN_XINDEX) 452 shndx = sym->sym.st_shndx; 453 454 sym->sec = find_section_by_index(elf, shndx); 455 if (!sym->sec) { 456 WARN("couldn't find section for symbol %s", 457 sym->name); 458 goto err; 459 } 460 if (GELF_ST_TYPE(sym->sym.st_info) == STT_SECTION) { 461 sym->name = sym->sec->name; 462 sym->sec->sym = sym; 463 } 464 } else 465 sym->sec = find_section_by_index(elf, 0); 466 467 elf_add_symbol(elf, sym); 468 } 469 470 if (opts.stats) { 471 printf("nr_symbols: %lu\n", (unsigned long)symbols_nr); 472 printf("symbol_bits: %d\n", elf->symbol_bits); 473 } 474 475 /* Create parent/child links for any cold subfunctions */ 476 list_for_each_entry(sec, &elf->sections, list) { 477 sec_for_each_sym(sec, sym) { 478 char pname[MAX_NAME_LEN + 1]; 479 size_t pnamelen; 480 if (sym->type != STT_FUNC) 481 continue; 482 483 if (sym->pfunc == NULL) 484 sym->pfunc = sym; 485 486 if (sym->cfunc == NULL) 487 sym->cfunc = sym; 488 489 coldstr = strstr(sym->name, ".cold"); 490 if (!coldstr) 491 continue; 492 493 pnamelen = coldstr - sym->name; 494 if (pnamelen > MAX_NAME_LEN) { 495 WARN("%s(): parent function name exceeds maximum length of %d characters", 496 sym->name, MAX_NAME_LEN); 497 return -1; 498 } 499 500 strncpy(pname, sym->name, pnamelen); 501 pname[pnamelen] = '\0'; 502 pfunc = find_symbol_by_name(elf, pname); 503 504 if (!pfunc) { 505 WARN("%s(): can't find parent function", 506 sym->name); 507 return -1; 508 } 509 510 sym->pfunc = pfunc; 511 pfunc->cfunc = sym; 512 513 /* 514 * Unfortunately, -fnoreorder-functions puts the child 515 * inside the parent. Remove the overlap so we can 516 * have sane assumptions. 517 * 518 * Note that pfunc->len now no longer matches 519 * pfunc->sym.st_size. 520 */ 521 if (sym->sec == pfunc->sec && 522 sym->offset >= pfunc->offset && 523 sym->offset + sym->len == pfunc->offset + pfunc->len) { 524 pfunc->len -= sym->len; 525 } 526 } 527 } 528 529 return 0; 530 531 err: 532 free(sym); 533 return -1; 534 } 535 536 static struct section *elf_create_reloc_section(struct elf *elf, 537 struct section *base, 538 int reltype); 539 540 int elf_add_reloc(struct elf *elf, struct section *sec, unsigned long offset, 541 unsigned int type, struct symbol *sym, s64 addend) 542 { 543 struct reloc *reloc; 544 545 if (!sec->reloc && !elf_create_reloc_section(elf, sec, SHT_RELA)) 546 return -1; 547 548 reloc = malloc(sizeof(*reloc)); 549 if (!reloc) { 550 perror("malloc"); 551 return -1; 552 } 553 memset(reloc, 0, sizeof(*reloc)); 554 555 reloc->sec = sec->reloc; 556 reloc->offset = offset; 557 reloc->type = type; 558 reloc->sym = sym; 559 reloc->addend = addend; 560 561 list_add_tail(&reloc->sym_reloc_entry, &sym->reloc_list); 562 list_add_tail(&reloc->list, &sec->reloc->reloc_list); 563 elf_hash_add(reloc, &reloc->hash, reloc_hash(reloc)); 564 565 sec->reloc->sh.sh_size += sec->reloc->sh.sh_entsize; 566 sec->reloc->changed = true; 567 568 return 0; 569 } 570 571 /* 572 * Ensure that any reloc section containing references to @sym is marked 573 * changed such that it will get re-generated in elf_rebuild_reloc_sections() 574 * with the new symbol index. 575 */ 576 static void elf_dirty_reloc_sym(struct elf *elf, struct symbol *sym) 577 { 578 struct reloc *reloc; 579 580 list_for_each_entry(reloc, &sym->reloc_list, sym_reloc_entry) 581 reloc->sec->changed = true; 582 } 583 584 /* 585 * The libelf API is terrible; gelf_update_sym*() takes a data block relative 586 * index value, *NOT* the symbol index. As such, iterate the data blocks and 587 * adjust index until it fits. 588 * 589 * If no data block is found, allow adding a new data block provided the index 590 * is only one past the end. 591 */ 592 static int elf_update_symbol(struct elf *elf, struct section *symtab, 593 struct section *symtab_shndx, struct symbol *sym) 594 { 595 Elf32_Word shndx = sym->sec ? sym->sec->idx : SHN_UNDEF; 596 Elf_Data *symtab_data = NULL, *shndx_data = NULL; 597 Elf64_Xword entsize = symtab->sh.sh_entsize; 598 int max_idx, idx = sym->idx; 599 Elf_Scn *s, *t = NULL; 600 bool is_special_shndx = sym->sym.st_shndx >= SHN_LORESERVE && 601 sym->sym.st_shndx != SHN_XINDEX; 602 603 if (is_special_shndx) 604 shndx = sym->sym.st_shndx; 605 606 s = elf_getscn(elf->elf, symtab->idx); 607 if (!s) { 608 WARN_ELF("elf_getscn"); 609 return -1; 610 } 611 612 if (symtab_shndx) { 613 t = elf_getscn(elf->elf, symtab_shndx->idx); 614 if (!t) { 615 WARN_ELF("elf_getscn"); 616 return -1; 617 } 618 } 619 620 for (;;) { 621 /* get next data descriptor for the relevant sections */ 622 symtab_data = elf_getdata(s, symtab_data); 623 if (t) 624 shndx_data = elf_getdata(t, shndx_data); 625 626 /* end-of-list */ 627 if (!symtab_data) { 628 /* 629 * Over-allocate to avoid O(n^2) symbol creation 630 * behaviour. The down side is that libelf doesn't 631 * like this; see elf_truncate_section() for the fixup. 632 */ 633 int num = max(1U, sym->idx/3); 634 void *buf; 635 636 if (idx) { 637 /* we don't do holes in symbol tables */ 638 WARN("index out of range"); 639 return -1; 640 } 641 642 /* if @idx == 0, it's the next contiguous entry, create it */ 643 symtab_data = elf_newdata(s); 644 if (t) 645 shndx_data = elf_newdata(t); 646 647 buf = calloc(num, entsize); 648 if (!buf) { 649 WARN("malloc"); 650 return -1; 651 } 652 653 symtab_data->d_buf = buf; 654 symtab_data->d_size = num * entsize; 655 symtab_data->d_align = 1; 656 symtab_data->d_type = ELF_T_SYM; 657 658 symtab->changed = true; 659 symtab->truncate = true; 660 661 if (t) { 662 buf = calloc(num, sizeof(Elf32_Word)); 663 if (!buf) { 664 WARN("malloc"); 665 return -1; 666 } 667 668 shndx_data->d_buf = buf; 669 shndx_data->d_size = num * sizeof(Elf32_Word); 670 shndx_data->d_align = sizeof(Elf32_Word); 671 shndx_data->d_type = ELF_T_WORD; 672 673 symtab_shndx->changed = true; 674 symtab_shndx->truncate = true; 675 } 676 677 break; 678 } 679 680 /* empty blocks should not happen */ 681 if (!symtab_data->d_size) { 682 WARN("zero size data"); 683 return -1; 684 } 685 686 /* is this the right block? */ 687 max_idx = symtab_data->d_size / entsize; 688 if (idx < max_idx) 689 break; 690 691 /* adjust index and try again */ 692 idx -= max_idx; 693 } 694 695 /* something went side-ways */ 696 if (idx < 0) { 697 WARN("negative index"); 698 return -1; 699 } 700 701 /* setup extended section index magic and write the symbol */ 702 if ((shndx >= SHN_UNDEF && shndx < SHN_LORESERVE) || is_special_shndx) { 703 sym->sym.st_shndx = shndx; 704 if (!shndx_data) 705 shndx = 0; 706 } else { 707 sym->sym.st_shndx = SHN_XINDEX; 708 if (!shndx_data) { 709 WARN("no .symtab_shndx"); 710 return -1; 711 } 712 } 713 714 if (!gelf_update_symshndx(symtab_data, shndx_data, idx, &sym->sym, shndx)) { 715 WARN_ELF("gelf_update_symshndx"); 716 return -1; 717 } 718 719 return 0; 720 } 721 722 static struct symbol * 723 __elf_create_symbol(struct elf *elf, struct symbol *sym) 724 { 725 struct section *symtab, *symtab_shndx; 726 Elf32_Word first_non_local, new_idx; 727 struct symbol *old; 728 729 symtab = find_section_by_name(elf, ".symtab"); 730 if (symtab) { 731 symtab_shndx = find_section_by_name(elf, ".symtab_shndx"); 732 } else { 733 WARN("no .symtab"); 734 return NULL; 735 } 736 737 new_idx = symtab->sh.sh_size / symtab->sh.sh_entsize; 738 739 if (GELF_ST_BIND(sym->sym.st_info) != STB_LOCAL) 740 goto non_local; 741 742 /* 743 * Move the first global symbol, as per sh_info, into a new, higher 744 * symbol index. This fees up a spot for a new local symbol. 745 */ 746 first_non_local = symtab->sh.sh_info; 747 old = find_symbol_by_index(elf, first_non_local); 748 if (old) { 749 old->idx = new_idx; 750 751 hlist_del(&old->hash); 752 elf_hash_add(symbol, &old->hash, old->idx); 753 754 elf_dirty_reloc_sym(elf, old); 755 756 if (elf_update_symbol(elf, symtab, symtab_shndx, old)) { 757 WARN("elf_update_symbol move"); 758 return NULL; 759 } 760 761 new_idx = first_non_local; 762 } 763 764 /* 765 * Either way, we will add a LOCAL symbol. 766 */ 767 symtab->sh.sh_info += 1; 768 769 non_local: 770 sym->idx = new_idx; 771 if (elf_update_symbol(elf, symtab, symtab_shndx, sym)) { 772 WARN("elf_update_symbol"); 773 return NULL; 774 } 775 776 symtab->sh.sh_size += symtab->sh.sh_entsize; 777 symtab->changed = true; 778 779 if (symtab_shndx) { 780 symtab_shndx->sh.sh_size += sizeof(Elf32_Word); 781 symtab_shndx->changed = true; 782 } 783 784 return sym; 785 } 786 787 static struct symbol * 788 elf_create_section_symbol(struct elf *elf, struct section *sec) 789 { 790 struct symbol *sym = calloc(1, sizeof(*sym)); 791 792 if (!sym) { 793 perror("malloc"); 794 return NULL; 795 } 796 797 sym->name = sec->name; 798 sym->sec = sec; 799 800 // st_name 0 801 sym->sym.st_info = GELF_ST_INFO(STB_LOCAL, STT_SECTION); 802 // st_other 0 803 // st_value 0 804 // st_size 0 805 806 sym = __elf_create_symbol(elf, sym); 807 if (sym) 808 elf_add_symbol(elf, sym); 809 810 return sym; 811 } 812 813 static int elf_add_string(struct elf *elf, struct section *strtab, char *str); 814 815 struct symbol * 816 elf_create_prefix_symbol(struct elf *elf, struct symbol *orig, long size) 817 { 818 struct symbol *sym = calloc(1, sizeof(*sym)); 819 size_t namelen = strlen(orig->name) + sizeof("__pfx_"); 820 char *name = malloc(namelen); 821 822 if (!sym || !name) { 823 perror("malloc"); 824 return NULL; 825 } 826 827 snprintf(name, namelen, "__pfx_%s", orig->name); 828 829 sym->name = name; 830 sym->sec = orig->sec; 831 832 sym->sym.st_name = elf_add_string(elf, NULL, name); 833 sym->sym.st_info = orig->sym.st_info; 834 sym->sym.st_value = orig->sym.st_value - size; 835 sym->sym.st_size = size; 836 837 sym = __elf_create_symbol(elf, sym); 838 if (sym) 839 elf_add_symbol(elf, sym); 840 841 return sym; 842 } 843 844 int elf_add_reloc_to_insn(struct elf *elf, struct section *sec, 845 unsigned long offset, unsigned int type, 846 struct section *insn_sec, unsigned long insn_off) 847 { 848 struct symbol *sym = insn_sec->sym; 849 int addend = insn_off; 850 851 if (!sym) { 852 /* 853 * Due to how weak functions work, we must use section based 854 * relocations. Symbol based relocations would result in the 855 * weak and non-weak function annotations being overlaid on the 856 * non-weak function after linking. 857 */ 858 sym = elf_create_section_symbol(elf, insn_sec); 859 if (!sym) 860 return -1; 861 862 insn_sec->sym = sym; 863 } 864 865 return elf_add_reloc(elf, sec, offset, type, sym, addend); 866 } 867 868 static int read_rel_reloc(struct section *sec, int i, struct reloc *reloc, unsigned int *symndx) 869 { 870 if (!gelf_getrel(sec->data, i, &reloc->rel)) { 871 WARN_ELF("gelf_getrel"); 872 return -1; 873 } 874 reloc->type = GELF_R_TYPE(reloc->rel.r_info); 875 reloc->addend = 0; 876 reloc->offset = reloc->rel.r_offset; 877 *symndx = GELF_R_SYM(reloc->rel.r_info); 878 return 0; 879 } 880 881 static int read_rela_reloc(struct section *sec, int i, struct reloc *reloc, unsigned int *symndx) 882 { 883 if (!gelf_getrela(sec->data, i, &reloc->rela)) { 884 WARN_ELF("gelf_getrela"); 885 return -1; 886 } 887 reloc->type = GELF_R_TYPE(reloc->rela.r_info); 888 reloc->addend = reloc->rela.r_addend; 889 reloc->offset = reloc->rela.r_offset; 890 *symndx = GELF_R_SYM(reloc->rela.r_info); 891 return 0; 892 } 893 894 static int read_relocs(struct elf *elf) 895 { 896 unsigned long nr_reloc, max_reloc = 0, tot_reloc = 0; 897 struct section *sec; 898 struct reloc *reloc; 899 unsigned int symndx; 900 struct symbol *sym; 901 int i; 902 903 if (!elf_alloc_hash(reloc, elf->text_size / 16)) 904 return -1; 905 906 list_for_each_entry(sec, &elf->sections, list) { 907 if ((sec->sh.sh_type != SHT_RELA) && 908 (sec->sh.sh_type != SHT_REL)) 909 continue; 910 911 sec->base = find_section_by_index(elf, sec->sh.sh_info); 912 if (!sec->base) { 913 WARN("can't find base section for reloc section %s", 914 sec->name); 915 return -1; 916 } 917 918 sec->base->reloc = sec; 919 920 nr_reloc = 0; 921 sec->reloc_data = calloc(sec->sh.sh_size / sec->sh.sh_entsize, sizeof(*reloc)); 922 if (!sec->reloc_data) { 923 perror("calloc"); 924 return -1; 925 } 926 for (i = 0; i < sec->sh.sh_size / sec->sh.sh_entsize; i++) { 927 reloc = &sec->reloc_data[i]; 928 switch (sec->sh.sh_type) { 929 case SHT_REL: 930 if (read_rel_reloc(sec, i, reloc, &symndx)) 931 return -1; 932 break; 933 case SHT_RELA: 934 if (read_rela_reloc(sec, i, reloc, &symndx)) 935 return -1; 936 break; 937 default: return -1; 938 } 939 940 reloc->sec = sec; 941 reloc->idx = i; 942 reloc->sym = sym = find_symbol_by_index(elf, symndx); 943 if (!reloc->sym) { 944 WARN("can't find reloc entry symbol %d for %s", 945 symndx, sec->name); 946 return -1; 947 } 948 949 list_add_tail(&reloc->sym_reloc_entry, &sym->reloc_list); 950 list_add_tail(&reloc->list, &sec->reloc_list); 951 elf_hash_add(reloc, &reloc->hash, reloc_hash(reloc)); 952 953 nr_reloc++; 954 } 955 max_reloc = max(max_reloc, nr_reloc); 956 tot_reloc += nr_reloc; 957 } 958 959 if (opts.stats) { 960 printf("max_reloc: %lu\n", max_reloc); 961 printf("tot_reloc: %lu\n", tot_reloc); 962 printf("reloc_bits: %d\n", elf->reloc_bits); 963 } 964 965 return 0; 966 } 967 968 struct elf *elf_open_read(const char *name, int flags) 969 { 970 struct elf *elf; 971 Elf_Cmd cmd; 972 973 elf_version(EV_CURRENT); 974 975 elf = malloc(sizeof(*elf)); 976 if (!elf) { 977 perror("malloc"); 978 return NULL; 979 } 980 memset(elf, 0, offsetof(struct elf, sections)); 981 982 INIT_LIST_HEAD(&elf->sections); 983 984 elf->fd = open(name, flags); 985 if (elf->fd == -1) { 986 fprintf(stderr, "objtool: Can't open '%s': %s\n", 987 name, strerror(errno)); 988 goto err; 989 } 990 991 if ((flags & O_ACCMODE) == O_RDONLY) 992 cmd = ELF_C_READ_MMAP; 993 else if ((flags & O_ACCMODE) == O_RDWR) 994 cmd = ELF_C_RDWR; 995 else /* O_WRONLY */ 996 cmd = ELF_C_WRITE; 997 998 elf->elf = elf_begin(elf->fd, cmd, NULL); 999 if (!elf->elf) { 1000 WARN_ELF("elf_begin"); 1001 goto err; 1002 } 1003 1004 if (!gelf_getehdr(elf->elf, &elf->ehdr)) { 1005 WARN_ELF("gelf_getehdr"); 1006 goto err; 1007 } 1008 1009 if (read_sections(elf)) 1010 goto err; 1011 1012 if (read_symbols(elf)) 1013 goto err; 1014 1015 if (read_relocs(elf)) 1016 goto err; 1017 1018 return elf; 1019 1020 err: 1021 elf_close(elf); 1022 return NULL; 1023 } 1024 1025 static int elf_add_string(struct elf *elf, struct section *strtab, char *str) 1026 { 1027 Elf_Data *data; 1028 Elf_Scn *s; 1029 int len; 1030 1031 if (!strtab) 1032 strtab = find_section_by_name(elf, ".strtab"); 1033 if (!strtab) { 1034 WARN("can't find .strtab section"); 1035 return -1; 1036 } 1037 1038 s = elf_getscn(elf->elf, strtab->idx); 1039 if (!s) { 1040 WARN_ELF("elf_getscn"); 1041 return -1; 1042 } 1043 1044 data = elf_newdata(s); 1045 if (!data) { 1046 WARN_ELF("elf_newdata"); 1047 return -1; 1048 } 1049 1050 data->d_buf = str; 1051 data->d_size = strlen(str) + 1; 1052 data->d_align = 1; 1053 1054 len = strtab->sh.sh_size; 1055 strtab->sh.sh_size += data->d_size; 1056 strtab->changed = true; 1057 1058 return len; 1059 } 1060 1061 struct section *elf_create_section(struct elf *elf, const char *name, 1062 unsigned int sh_flags, size_t entsize, int nr) 1063 { 1064 struct section *sec, *shstrtab; 1065 size_t size = entsize * nr; 1066 Elf_Scn *s; 1067 1068 sec = malloc(sizeof(*sec)); 1069 if (!sec) { 1070 perror("malloc"); 1071 return NULL; 1072 } 1073 memset(sec, 0, sizeof(*sec)); 1074 1075 INIT_LIST_HEAD(&sec->symbol_list); 1076 INIT_LIST_HEAD(&sec->reloc_list); 1077 1078 s = elf_newscn(elf->elf); 1079 if (!s) { 1080 WARN_ELF("elf_newscn"); 1081 return NULL; 1082 } 1083 1084 sec->name = strdup(name); 1085 if (!sec->name) { 1086 perror("strdup"); 1087 return NULL; 1088 } 1089 1090 sec->idx = elf_ndxscn(s); 1091 sec->changed = true; 1092 1093 sec->data = elf_newdata(s); 1094 if (!sec->data) { 1095 WARN_ELF("elf_newdata"); 1096 return NULL; 1097 } 1098 1099 sec->data->d_size = size; 1100 sec->data->d_align = 1; 1101 1102 if (size) { 1103 sec->data->d_buf = malloc(size); 1104 if (!sec->data->d_buf) { 1105 perror("malloc"); 1106 return NULL; 1107 } 1108 memset(sec->data->d_buf, 0, size); 1109 } 1110 1111 if (!gelf_getshdr(s, &sec->sh)) { 1112 WARN_ELF("gelf_getshdr"); 1113 return NULL; 1114 } 1115 1116 sec->sh.sh_size = size; 1117 sec->sh.sh_entsize = entsize; 1118 sec->sh.sh_type = SHT_PROGBITS; 1119 sec->sh.sh_addralign = 1; 1120 sec->sh.sh_flags = SHF_ALLOC | sh_flags; 1121 1122 /* Add section name to .shstrtab (or .strtab for Clang) */ 1123 shstrtab = find_section_by_name(elf, ".shstrtab"); 1124 if (!shstrtab) 1125 shstrtab = find_section_by_name(elf, ".strtab"); 1126 if (!shstrtab) { 1127 WARN("can't find .shstrtab or .strtab section"); 1128 return NULL; 1129 } 1130 sec->sh.sh_name = elf_add_string(elf, shstrtab, sec->name); 1131 if (sec->sh.sh_name == -1) 1132 return NULL; 1133 1134 list_add_tail(&sec->list, &elf->sections); 1135 elf_hash_add(section, &sec->hash, sec->idx); 1136 elf_hash_add(section_name, &sec->name_hash, str_hash(sec->name)); 1137 1138 elf->changed = true; 1139 1140 return sec; 1141 } 1142 1143 static struct section *elf_create_rel_reloc_section(struct elf *elf, struct section *base) 1144 { 1145 char *relocname; 1146 struct section *sec; 1147 1148 relocname = malloc(strlen(base->name) + strlen(".rel") + 1); 1149 if (!relocname) { 1150 perror("malloc"); 1151 return NULL; 1152 } 1153 strcpy(relocname, ".rel"); 1154 strcat(relocname, base->name); 1155 1156 sec = elf_create_section(elf, relocname, 0, sizeof(GElf_Rel), 0); 1157 free(relocname); 1158 if (!sec) 1159 return NULL; 1160 1161 base->reloc = sec; 1162 sec->base = base; 1163 1164 sec->sh.sh_type = SHT_REL; 1165 sec->sh.sh_addralign = 8; 1166 sec->sh.sh_link = find_section_by_name(elf, ".symtab")->idx; 1167 sec->sh.sh_info = base->idx; 1168 sec->sh.sh_flags = SHF_INFO_LINK; 1169 1170 return sec; 1171 } 1172 1173 static struct section *elf_create_rela_reloc_section(struct elf *elf, struct section *base) 1174 { 1175 char *relocname; 1176 struct section *sec; 1177 int addrsize = elf_class_addrsize(elf); 1178 1179 relocname = malloc(strlen(base->name) + strlen(".rela") + 1); 1180 if (!relocname) { 1181 perror("malloc"); 1182 return NULL; 1183 } 1184 strcpy(relocname, ".rela"); 1185 strcat(relocname, base->name); 1186 1187 if (addrsize == sizeof(u32)) 1188 sec = elf_create_section(elf, relocname, 0, sizeof(Elf32_Rela), 0); 1189 else 1190 sec = elf_create_section(elf, relocname, 0, sizeof(GElf_Rela), 0); 1191 free(relocname); 1192 if (!sec) 1193 return NULL; 1194 1195 base->reloc = sec; 1196 sec->base = base; 1197 1198 sec->sh.sh_type = SHT_RELA; 1199 sec->sh.sh_addralign = addrsize; 1200 sec->sh.sh_link = find_section_by_name(elf, ".symtab")->idx; 1201 sec->sh.sh_info = base->idx; 1202 sec->sh.sh_flags = SHF_INFO_LINK; 1203 1204 return sec; 1205 } 1206 1207 static struct section *elf_create_reloc_section(struct elf *elf, 1208 struct section *base, 1209 int reltype) 1210 { 1211 switch (reltype) { 1212 case SHT_REL: return elf_create_rel_reloc_section(elf, base); 1213 case SHT_RELA: return elf_create_rela_reloc_section(elf, base); 1214 default: return NULL; 1215 } 1216 } 1217 1218 static int elf_rebuild_rel_reloc_section(struct section *sec) 1219 { 1220 struct reloc *reloc; 1221 int idx = 0; 1222 void *buf; 1223 1224 /* Allocate a buffer for relocations */ 1225 buf = malloc(sec->sh.sh_size); 1226 if (!buf) { 1227 perror("malloc"); 1228 return -1; 1229 } 1230 1231 sec->data->d_buf = buf; 1232 sec->data->d_size = sec->sh.sh_size; 1233 sec->data->d_type = ELF_T_REL; 1234 1235 idx = 0; 1236 list_for_each_entry(reloc, &sec->reloc_list, list) { 1237 reloc->rel.r_offset = reloc->offset; 1238 reloc->rel.r_info = GELF_R_INFO(reloc->sym->idx, reloc->type); 1239 if (!gelf_update_rel(sec->data, idx, &reloc->rel)) { 1240 WARN_ELF("gelf_update_rel"); 1241 return -1; 1242 } 1243 idx++; 1244 } 1245 1246 return 0; 1247 } 1248 1249 static int elf_rebuild_rela_reloc_section(struct section *sec) 1250 { 1251 struct reloc *reloc; 1252 int idx = 0; 1253 void *buf; 1254 1255 /* Allocate a buffer for relocations with addends */ 1256 buf = malloc(sec->sh.sh_size); 1257 if (!buf) { 1258 perror("malloc"); 1259 return -1; 1260 } 1261 1262 sec->data->d_buf = buf; 1263 sec->data->d_size = sec->sh.sh_size; 1264 sec->data->d_type = ELF_T_RELA; 1265 1266 idx = 0; 1267 list_for_each_entry(reloc, &sec->reloc_list, list) { 1268 reloc->rela.r_offset = reloc->offset; 1269 reloc->rela.r_addend = reloc->addend; 1270 reloc->rela.r_info = GELF_R_INFO(reloc->sym->idx, reloc->type); 1271 if (!gelf_update_rela(sec->data, idx, &reloc->rela)) { 1272 WARN_ELF("gelf_update_rela"); 1273 return -1; 1274 } 1275 idx++; 1276 } 1277 1278 return 0; 1279 } 1280 1281 static int elf_rebuild_reloc_section(struct elf *elf, struct section *sec) 1282 { 1283 switch (sec->sh.sh_type) { 1284 case SHT_REL: return elf_rebuild_rel_reloc_section(sec); 1285 case SHT_RELA: return elf_rebuild_rela_reloc_section(sec); 1286 default: return -1; 1287 } 1288 } 1289 1290 int elf_write_insn(struct elf *elf, struct section *sec, 1291 unsigned long offset, unsigned int len, 1292 const char *insn) 1293 { 1294 Elf_Data *data = sec->data; 1295 1296 if (data->d_type != ELF_T_BYTE || data->d_off) { 1297 WARN("write to unexpected data for section: %s", sec->name); 1298 return -1; 1299 } 1300 1301 memcpy(data->d_buf + offset, insn, len); 1302 elf_flagdata(data, ELF_C_SET, ELF_F_DIRTY); 1303 1304 elf->changed = true; 1305 1306 return 0; 1307 } 1308 1309 int elf_write_reloc(struct elf *elf, struct reloc *reloc) 1310 { 1311 struct section *sec = reloc->sec; 1312 1313 if (sec->sh.sh_type == SHT_REL) { 1314 reloc->rel.r_info = GELF_R_INFO(reloc->sym->idx, reloc->type); 1315 reloc->rel.r_offset = reloc->offset; 1316 1317 if (!gelf_update_rel(sec->data, reloc->idx, &reloc->rel)) { 1318 WARN_ELF("gelf_update_rel"); 1319 return -1; 1320 } 1321 } else { 1322 reloc->rela.r_info = GELF_R_INFO(reloc->sym->idx, reloc->type); 1323 reloc->rela.r_addend = reloc->addend; 1324 reloc->rela.r_offset = reloc->offset; 1325 1326 if (!gelf_update_rela(sec->data, reloc->idx, &reloc->rela)) { 1327 WARN_ELF("gelf_update_rela"); 1328 return -1; 1329 } 1330 } 1331 1332 elf->changed = true; 1333 1334 return 0; 1335 } 1336 1337 /* 1338 * When Elf_Scn::sh_size is smaller than the combined Elf_Data::d_size 1339 * do you: 1340 * 1341 * A) adhere to the section header and truncate the data, or 1342 * B) ignore the section header and write out all the data you've got? 1343 * 1344 * Yes, libelf sucks and we need to manually truncate if we over-allocate data. 1345 */ 1346 static int elf_truncate_section(struct elf *elf, struct section *sec) 1347 { 1348 u64 size = sec->sh.sh_size; 1349 bool truncated = false; 1350 Elf_Data *data = NULL; 1351 Elf_Scn *s; 1352 1353 s = elf_getscn(elf->elf, sec->idx); 1354 if (!s) { 1355 WARN_ELF("elf_getscn"); 1356 return -1; 1357 } 1358 1359 for (;;) { 1360 /* get next data descriptor for the relevant section */ 1361 data = elf_getdata(s, data); 1362 1363 if (!data) { 1364 if (size) { 1365 WARN("end of section data but non-zero size left\n"); 1366 return -1; 1367 } 1368 return 0; 1369 } 1370 1371 if (truncated) { 1372 /* when we remove symbols */ 1373 WARN("truncated; but more data\n"); 1374 return -1; 1375 } 1376 1377 if (!data->d_size) { 1378 WARN("zero size data"); 1379 return -1; 1380 } 1381 1382 if (data->d_size > size) { 1383 truncated = true; 1384 data->d_size = size; 1385 } 1386 1387 size -= data->d_size; 1388 } 1389 } 1390 1391 int elf_write(struct elf *elf) 1392 { 1393 struct section *sec; 1394 Elf_Scn *s; 1395 1396 if (opts.dryrun) 1397 return 0; 1398 1399 /* Update changed relocation sections and section headers: */ 1400 list_for_each_entry(sec, &elf->sections, list) { 1401 if (sec->truncate) 1402 elf_truncate_section(elf, sec); 1403 1404 if (sec->changed) { 1405 s = elf_getscn(elf->elf, sec->idx); 1406 if (!s) { 1407 WARN_ELF("elf_getscn"); 1408 return -1; 1409 } 1410 if (!gelf_update_shdr(s, &sec->sh)) { 1411 WARN_ELF("gelf_update_shdr"); 1412 return -1; 1413 } 1414 1415 if (sec->base && 1416 elf_rebuild_reloc_section(elf, sec)) { 1417 WARN("elf_rebuild_reloc_section"); 1418 return -1; 1419 } 1420 1421 sec->changed = false; 1422 elf->changed = true; 1423 } 1424 } 1425 1426 /* Make sure the new section header entries get updated properly. */ 1427 elf_flagelf(elf->elf, ELF_C_SET, ELF_F_DIRTY); 1428 1429 /* Write all changes to the file. */ 1430 if (elf_update(elf->elf, ELF_C_WRITE) < 0) { 1431 WARN_ELF("elf_update"); 1432 return -1; 1433 } 1434 1435 elf->changed = false; 1436 1437 return 0; 1438 } 1439 1440 void elf_close(struct elf *elf) 1441 { 1442 struct section *sec, *tmpsec; 1443 struct symbol *sym, *tmpsym; 1444 struct reloc *reloc, *tmpreloc; 1445 1446 if (elf->elf) 1447 elf_end(elf->elf); 1448 1449 if (elf->fd > 0) 1450 close(elf->fd); 1451 1452 list_for_each_entry_safe(sec, tmpsec, &elf->sections, list) { 1453 list_for_each_entry_safe(sym, tmpsym, &sec->symbol_list, list) { 1454 list_del(&sym->list); 1455 hash_del(&sym->hash); 1456 } 1457 list_for_each_entry_safe(reloc, tmpreloc, &sec->reloc_list, list) { 1458 list_del(&reloc->list); 1459 hash_del(&reloc->hash); 1460 } 1461 list_del(&sec->list); 1462 free(sec->reloc_data); 1463 } 1464 1465 free(elf->symbol_data); 1466 free(elf->section_data); 1467 free(elf); 1468 } 1469