1 /* 2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 12 #include <stdarg.h> 13 #include <stdio.h> 14 #include <stdlib.h> 15 #include <string.h> 16 17 #include "vpx_config.h" 18 #include "vpx/vpx_integer.h" 19 20 typedef enum { 21 OUTPUT_FMT_PLAIN, 22 OUTPUT_FMT_RVDS, 23 OUTPUT_FMT_GAS, 24 OUTPUT_FMT_C_HEADER, 25 } output_fmt_t; 26 27 int log_msg(const char *fmt, ...) { 28 int res; 29 va_list ap; 30 va_start(ap, fmt); 31 res = vfprintf(stderr, fmt, ap); 32 va_end(ap); 33 return res; 34 } 35 36 #if defined(__GNUC__) && __GNUC__ 37 38 #if defined(FORCE_PARSE_ELF) 39 40 #if defined(__MACH__) 41 #undef __MACH__ 42 #endif 43 44 #if !defined(__ELF__) 45 #define __ELF__ 46 #endif 47 #endif 48 49 #if defined(__MACH__) 50 51 #include <mach-o/loader.h> 52 #include <mach-o/nlist.h> 53 54 int print_macho_equ(output_fmt_t mode, uint8_t* name, int val) { 55 switch (mode) { 56 case OUTPUT_FMT_RVDS: 57 printf("%-40s EQU %5d\n", name, val); 58 return 0; 59 case OUTPUT_FMT_GAS: 60 printf(".set %-40s, %5d\n", name, val); 61 return 0; 62 case OUTPUT_FMT_C_HEADER: 63 printf("#define %-40s %5d\n", name, val); 64 return 0; 65 default: 66 log_msg("Unsupported mode: %d", mode); 67 return 1; 68 } 69 } 70 71 int parse_macho(uint8_t *base_buf, size_t sz, output_fmt_t mode) { 72 int i, j; 73 struct mach_header header; 74 uint8_t *buf = base_buf; 75 int base_data_section = 0; 76 int bits = 0; 77 78 /* We can read in mach_header for 32 and 64 bit architectures 79 * because it's identical to mach_header_64 except for the last 80 * element (uint32_t reserved), which we don't use. Then, when 81 * we know which architecture we're looking at, increment buf 82 * appropriately. 83 */ 84 memcpy(&header, buf, sizeof(struct mach_header)); 85 86 if (header.magic == MH_MAGIC) { 87 if (header.cputype == CPU_TYPE_ARM 88 || header.cputype == CPU_TYPE_X86) { 89 bits = 32; 90 buf += sizeof(struct mach_header); 91 } else { 92 log_msg("Bad cputype for object file. Currently only tested for CPU_TYPE_[ARM|X86].\n"); 93 goto bail; 94 } 95 } else if (header.magic == MH_MAGIC_64) { 96 if (header.cputype == CPU_TYPE_X86_64) { 97 bits = 64; 98 buf += sizeof(struct mach_header_64); 99 } else { 100 log_msg("Bad cputype for object file. Currently only tested for CPU_TYPE_X86_64.\n"); 101 goto bail; 102 } 103 } else { 104 log_msg("Bad magic number for object file. 0x%x or 0x%x expected, 0x%x found.\n", 105 MH_MAGIC, MH_MAGIC_64, header.magic); 106 goto bail; 107 } 108 109 if (header.filetype != MH_OBJECT) { 110 log_msg("Bad filetype for object file. Currently only tested for MH_OBJECT.\n"); 111 goto bail; 112 } 113 114 for (i = 0; i < header.ncmds; i++) { 115 struct load_command lc; 116 117 memcpy(&lc, buf, sizeof(struct load_command)); 118 119 if (lc.cmd == LC_SEGMENT) { 120 uint8_t *seg_buf = buf; 121 struct section s; 122 struct segment_command seg_c; 123 124 memcpy(&seg_c, seg_buf, sizeof(struct segment_command)); 125 seg_buf += sizeof(struct segment_command); 126 127 /* Although each section is given it's own offset, nlist.n_value 128 * references the offset of the first section. This isn't 129 * apparent without debug information because the offset of the 130 * data section is the same as the first section. However, with 131 * debug sections mixed in, the offset of the debug section 132 * increases but n_value still references the first section. 133 */ 134 if (seg_c.nsects < 1) { 135 log_msg("Not enough sections\n"); 136 goto bail; 137 } 138 139 memcpy(&s, seg_buf, sizeof(struct section)); 140 base_data_section = s.offset; 141 } else if (lc.cmd == LC_SEGMENT_64) { 142 uint8_t *seg_buf = buf; 143 struct section_64 s; 144 struct segment_command_64 seg_c; 145 146 memcpy(&seg_c, seg_buf, sizeof(struct segment_command_64)); 147 seg_buf += sizeof(struct segment_command_64); 148 149 /* Explanation in LG_SEGMENT */ 150 if (seg_c.nsects < 1) { 151 log_msg("Not enough sections\n"); 152 goto bail; 153 } 154 155 memcpy(&s, seg_buf, sizeof(struct section_64)); 156 base_data_section = s.offset; 157 } else if (lc.cmd == LC_SYMTAB) { 158 if (base_data_section != 0) { 159 struct symtab_command sc; 160 uint8_t *sym_buf = base_buf; 161 uint8_t *str_buf = base_buf; 162 163 memcpy(&sc, buf, sizeof(struct symtab_command)); 164 165 if (sc.cmdsize != sizeof(struct symtab_command)) { 166 log_msg("Can't find symbol table!\n"); 167 goto bail; 168 } 169 170 sym_buf += sc.symoff; 171 str_buf += sc.stroff; 172 173 for (j = 0; j < sc.nsyms; j++) { 174 /* Location of string is cacluated each time from the 175 * start of the string buffer. On darwin the symbols 176 * are prefixed by "_", so we bump the pointer by 1. 177 * The target value is defined as an int in *_asm_*_offsets.c, 178 * which is 4 bytes on all targets we currently use. 179 */ 180 if (bits == 32) { 181 struct nlist nl; 182 int val; 183 184 memcpy(&nl, sym_buf, sizeof(struct nlist)); 185 sym_buf += sizeof(struct nlist); 186 187 memcpy(&val, base_buf + base_data_section + nl.n_value, 188 sizeof(val)); 189 print_macho_equ(mode, str_buf + nl.n_un.n_strx + 1, val); 190 } else { /* if (bits == 64) */ 191 struct nlist_64 nl; 192 int val; 193 194 memcpy(&nl, sym_buf, sizeof(struct nlist_64)); 195 sym_buf += sizeof(struct nlist_64); 196 197 memcpy(&val, base_buf + base_data_section + nl.n_value, 198 sizeof(val)); 199 print_macho_equ(mode, str_buf + nl.n_un.n_strx + 1, val); 200 } 201 } 202 } 203 } 204 205 buf += lc.cmdsize; 206 } 207 208 return 0; 209 bail: 210 return 1; 211 212 } 213 214 #elif defined(__ELF__) 215 #include "elf.h" 216 217 #define COPY_STRUCT(dst, buf, ofst, sz) do {\ 218 if(ofst + sizeof((*(dst))) > sz) goto bail;\ 219 memcpy(dst, buf+ofst, sizeof((*(dst))));\ 220 } while(0) 221 222 #define ENDIAN_ASSIGN(val, memb) do {\ 223 if(!elf->le_data) {log_msg("Big Endian data not supported yet!\n");goto bail;}\ 224 (val) = (memb);\ 225 } while(0) 226 227 #define ENDIAN_ASSIGN_IN_PLACE(memb) do {\ 228 ENDIAN_ASSIGN(memb, memb);\ 229 } while(0) 230 231 typedef struct { 232 uint8_t *buf; /* Buffer containing ELF data */ 233 size_t sz; /* Buffer size */ 234 int le_data; /* Data is little-endian */ 235 unsigned char e_ident[EI_NIDENT]; /* Magic number and other info */ 236 int bits; /* 32 or 64 */ 237 Elf32_Ehdr hdr32; 238 Elf64_Ehdr hdr64; 239 } elf_obj_t; 240 241 int parse_elf_header(elf_obj_t *elf) { 242 int res; 243 /* Verify ELF Magic numbers */ 244 COPY_STRUCT(&elf->e_ident, elf->buf, 0, elf->sz); 245 res = elf->e_ident[EI_MAG0] == ELFMAG0; 246 res &= elf->e_ident[EI_MAG1] == ELFMAG1; 247 res &= elf->e_ident[EI_MAG2] == ELFMAG2; 248 res &= elf->e_ident[EI_MAG3] == ELFMAG3; 249 res &= elf->e_ident[EI_CLASS] == ELFCLASS32 250 || elf->e_ident[EI_CLASS] == ELFCLASS64; 251 res &= elf->e_ident[EI_DATA] == ELFDATA2LSB; 252 253 if (!res) goto bail; 254 255 elf->le_data = elf->e_ident[EI_DATA] == ELFDATA2LSB; 256 257 /* Read in relevant values */ 258 if (elf->e_ident[EI_CLASS] == ELFCLASS32) { 259 elf->bits = 32; 260 COPY_STRUCT(&elf->hdr32, elf->buf, 0, elf->sz); 261 262 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_type); 263 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_machine); 264 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_version); 265 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_entry); 266 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phoff); 267 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shoff); 268 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_flags); 269 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_ehsize); 270 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phentsize); 271 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phnum); 272 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shentsize); 273 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shnum); 274 ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shstrndx); 275 } else { /* if (elf->e_ident[EI_CLASS] == ELFCLASS64) */ 276 elf->bits = 64; 277 COPY_STRUCT(&elf->hdr64, elf->buf, 0, elf->sz); 278 279 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_type); 280 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_machine); 281 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_version); 282 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_entry); 283 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phoff); 284 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shoff); 285 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_flags); 286 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_ehsize); 287 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phentsize); 288 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phnum); 289 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shentsize); 290 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shnum); 291 ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shstrndx); 292 } 293 294 return 0; 295 bail: 296 log_msg("Failed to parse ELF file header"); 297 return 1; 298 } 299 300 int parse_elf_section(elf_obj_t *elf, int idx, Elf32_Shdr *hdr32, Elf64_Shdr *hdr64) { 301 if (hdr32) { 302 if (idx >= elf->hdr32.e_shnum) 303 goto bail; 304 305 COPY_STRUCT(hdr32, elf->buf, elf->hdr32.e_shoff + idx * elf->hdr32.e_shentsize, 306 elf->sz); 307 ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_name); 308 ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_type); 309 ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_flags); 310 ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_addr); 311 ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_offset); 312 ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_size); 313 ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_link); 314 ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_info); 315 ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_addralign); 316 ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_entsize); 317 } else { /* if (hdr64) */ 318 if (idx >= elf->hdr64.e_shnum) 319 goto bail; 320 321 COPY_STRUCT(hdr64, elf->buf, elf->hdr64.e_shoff + idx * elf->hdr64.e_shentsize, 322 elf->sz); 323 ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_name); 324 ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_type); 325 ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_flags); 326 ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_addr); 327 ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_offset); 328 ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_size); 329 ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_link); 330 ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_info); 331 ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_addralign); 332 ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_entsize); 333 } 334 335 return 0; 336 bail: 337 return 1; 338 } 339 340 const char *parse_elf_string_table(elf_obj_t *elf, int s_idx, int idx) { 341 if (elf->bits == 32) { 342 Elf32_Shdr shdr; 343 344 if (parse_elf_section(elf, s_idx, &shdr, NULL)) { 345 log_msg("Failed to parse ELF string table: section %d, index %d\n", 346 s_idx, idx); 347 return ""; 348 } 349 350 return (char *)(elf->buf + shdr.sh_offset + idx); 351 } else { /* if (elf->bits == 64) */ 352 Elf64_Shdr shdr; 353 354 if (parse_elf_section(elf, s_idx, NULL, &shdr)) { 355 log_msg("Failed to parse ELF string table: section %d, index %d\n", 356 s_idx, idx); 357 return ""; 358 } 359 360 return (char *)(elf->buf + shdr.sh_offset + idx); 361 } 362 } 363 364 int parse_elf_symbol(elf_obj_t *elf, unsigned int ofst, Elf32_Sym *sym32, Elf64_Sym *sym64) { 365 if (sym32) { 366 COPY_STRUCT(sym32, elf->buf, ofst, elf->sz); 367 ENDIAN_ASSIGN_IN_PLACE(sym32->st_name); 368 ENDIAN_ASSIGN_IN_PLACE(sym32->st_value); 369 ENDIAN_ASSIGN_IN_PLACE(sym32->st_size); 370 ENDIAN_ASSIGN_IN_PLACE(sym32->st_info); 371 ENDIAN_ASSIGN_IN_PLACE(sym32->st_other); 372 ENDIAN_ASSIGN_IN_PLACE(sym32->st_shndx); 373 } else { /* if (sym64) */ 374 COPY_STRUCT(sym64, elf->buf, ofst, elf->sz); 375 ENDIAN_ASSIGN_IN_PLACE(sym64->st_name); 376 ENDIAN_ASSIGN_IN_PLACE(sym64->st_value); 377 ENDIAN_ASSIGN_IN_PLACE(sym64->st_size); 378 ENDIAN_ASSIGN_IN_PLACE(sym64->st_info); 379 ENDIAN_ASSIGN_IN_PLACE(sym64->st_other); 380 ENDIAN_ASSIGN_IN_PLACE(sym64->st_shndx); 381 } 382 return 0; 383 bail: 384 return 1; 385 } 386 387 int parse_elf(uint8_t *buf, size_t sz, output_fmt_t mode) { 388 elf_obj_t elf; 389 unsigned int ofst; 390 int i; 391 Elf32_Off strtab_off32; 392 Elf64_Off strtab_off64; /* save String Table offset for later use */ 393 394 memset(&elf, 0, sizeof(elf)); 395 elf.buf = buf; 396 elf.sz = sz; 397 398 /* Parse Header */ 399 if (parse_elf_header(&elf)) 400 goto bail; 401 402 if (elf.bits == 32) { 403 Elf32_Shdr shdr; 404 for (i = 0; i < elf.hdr32.e_shnum; i++) { 405 parse_elf_section(&elf, i, &shdr, NULL); 406 407 if (shdr.sh_type == SHT_STRTAB) { 408 char strtsb_name[128]; 409 410 strcpy(strtsb_name, (char *)(elf.buf + shdr.sh_offset + shdr.sh_name)); 411 412 if (!(strcmp(strtsb_name, ".shstrtab"))) { 413 /* log_msg("found section: %s\n", strtsb_name); */ 414 strtab_off32 = shdr.sh_offset; 415 break; 416 } 417 } 418 } 419 } else { /* if (elf.bits == 64) */ 420 Elf64_Shdr shdr; 421 for (i = 0; i < elf.hdr64.e_shnum; i++) { 422 parse_elf_section(&elf, i, NULL, &shdr); 423 424 if (shdr.sh_type == SHT_STRTAB) { 425 char strtsb_name[128]; 426 427 strcpy(strtsb_name, (char *)(elf.buf + shdr.sh_offset + shdr.sh_name)); 428 429 if (!(strcmp(strtsb_name, ".shstrtab"))) { 430 /* log_msg("found section: %s\n", strtsb_name); */ 431 strtab_off64 = shdr.sh_offset; 432 break; 433 } 434 } 435 } 436 } 437 438 /* Parse all Symbol Tables */ 439 if (elf.bits == 32) { 440 Elf32_Shdr shdr; 441 for (i = 0; i < elf.hdr32.e_shnum; i++) { 442 parse_elf_section(&elf, i, &shdr, NULL); 443 444 if (shdr.sh_type == SHT_SYMTAB) { 445 for (ofst = shdr.sh_offset; 446 ofst < shdr.sh_offset + shdr.sh_size; 447 ofst += shdr.sh_entsize) { 448 Elf32_Sym sym; 449 450 parse_elf_symbol(&elf, ofst, &sym, NULL); 451 452 /* For all OBJECTS (data objects), extract the value from the 453 * proper data segment. 454 */ 455 /* if (ELF32_ST_TYPE(sym.st_info) == STT_OBJECT && sym.st_name) 456 log_msg("found data object %s\n", 457 parse_elf_string_table(&elf, 458 shdr.sh_link, 459 sym.st_name)); 460 */ 461 462 if (ELF32_ST_TYPE(sym.st_info) == STT_OBJECT 463 && sym.st_size == 4) { 464 Elf32_Shdr dhdr; 465 int val = 0; 466 char section_name[128]; 467 468 parse_elf_section(&elf, sym.st_shndx, &dhdr, NULL); 469 470 /* For explanition - refer to _MSC_VER version of code */ 471 strcpy(section_name, (char *)(elf.buf + strtab_off32 + dhdr.sh_name)); 472 /* log_msg("Section_name: %s, Section_type: %d\n", section_name, dhdr.sh_type); */ 473 474 if (strcmp(section_name, ".bss")) { 475 if (sizeof(val) != sym.st_size) { 476 /* The target value is declared as an int in 477 * *_asm_*_offsets.c, which is 4 bytes on all 478 * targets we currently use. Complain loudly if 479 * this is not true. 480 */ 481 log_msg("Symbol size is wrong\n"); 482 goto bail; 483 } 484 485 memcpy(&val, 486 elf.buf + dhdr.sh_offset + sym.st_value, 487 sym.st_size); 488 } 489 490 if (!elf.le_data) { 491 log_msg("Big Endian data not supported yet!\n"); 492 goto bail; 493 } 494 495 switch (mode) { 496 case OUTPUT_FMT_RVDS: 497 printf("%-40s EQU %5d\n", 498 parse_elf_string_table(&elf, 499 shdr.sh_link, 500 sym.st_name), 501 val); 502 break; 503 case OUTPUT_FMT_GAS: 504 printf(".equ %-40s, %5d\n", 505 parse_elf_string_table(&elf, 506 shdr.sh_link, 507 sym.st_name), 508 val); 509 break; 510 case OUTPUT_FMT_C_HEADER: 511 printf("#define %-40s %5d\n", 512 parse_elf_string_table(&elf, 513 shdr.sh_link, 514 sym.st_name), 515 val); 516 break; 517 default: 518 printf("%s = %d\n", 519 parse_elf_string_table(&elf, 520 shdr.sh_link, 521 sym.st_name), 522 val); 523 } 524 } 525 } 526 } 527 } 528 } else { /* if (elf.bits == 64) */ 529 Elf64_Shdr shdr; 530 for (i = 0; i < elf.hdr64.e_shnum; i++) { 531 parse_elf_section(&elf, i, NULL, &shdr); 532 533 if (shdr.sh_type == SHT_SYMTAB) { 534 for (ofst = shdr.sh_offset; 535 ofst < shdr.sh_offset + shdr.sh_size; 536 ofst += shdr.sh_entsize) { 537 Elf64_Sym sym; 538 539 parse_elf_symbol(&elf, ofst, NULL, &sym); 540 541 /* For all OBJECTS (data objects), extract the value from the 542 * proper data segment. 543 */ 544 /* if (ELF64_ST_TYPE(sym.st_info) == STT_OBJECT && sym.st_name) 545 log_msg("found data object %s\n", 546 parse_elf_string_table(&elf, 547 shdr.sh_link, 548 sym.st_name)); 549 */ 550 551 if (ELF64_ST_TYPE(sym.st_info) == STT_OBJECT 552 && sym.st_size == 4) { 553 Elf64_Shdr dhdr; 554 int val = 0; 555 char section_name[128]; 556 557 parse_elf_section(&elf, sym.st_shndx, NULL, &dhdr); 558 559 /* For explanition - refer to _MSC_VER version of code */ 560 strcpy(section_name, (char *)(elf.buf + strtab_off64 + dhdr.sh_name)); 561 /* log_msg("Section_name: %s, Section_type: %d\n", section_name, dhdr.sh_type); */ 562 563 if ((strcmp(section_name, ".bss"))) { 564 if (sizeof(val) != sym.st_size) { 565 /* The target value is declared as an int in 566 * *_asm_*_offsets.c, which is 4 bytes on all 567 * targets we currently use. Complain loudly if 568 * this is not true. 569 */ 570 log_msg("Symbol size is wrong\n"); 571 goto bail; 572 } 573 574 memcpy(&val, 575 elf.buf + dhdr.sh_offset + sym.st_value, 576 sym.st_size); 577 } 578 579 if (!elf.le_data) { 580 log_msg("Big Endian data not supported yet!\n"); 581 goto bail; 582 } 583 584 switch (mode) { 585 case OUTPUT_FMT_RVDS: 586 printf("%-40s EQU %5d\n", 587 parse_elf_string_table(&elf, 588 shdr.sh_link, 589 sym.st_name), 590 val); 591 break; 592 case OUTPUT_FMT_GAS: 593 printf(".equ %-40s, %5d\n", 594 parse_elf_string_table(&elf, 595 shdr.sh_link, 596 sym.st_name), 597 val); 598 break; 599 default: 600 printf("%s = %d\n", 601 parse_elf_string_table(&elf, 602 shdr.sh_link, 603 sym.st_name), 604 val); 605 } 606 } 607 } 608 } 609 } 610 } 611 612 if (mode == OUTPUT_FMT_RVDS) 613 printf(" END\n"); 614 615 return 0; 616 bail: 617 log_msg("Parse error: File does not appear to be valid ELF32 or ELF64\n"); 618 return 1; 619 } 620 621 #endif 622 #endif /* defined(__GNUC__) && __GNUC__ */ 623 624 625 #if defined(_MSC_VER) || defined(__MINGW32__) || defined(__CYGWIN__) 626 /* See "Microsoft Portable Executable and Common Object File Format Specification" 627 for reference. 628 */ 629 #define get_le32(x) ((*(x)) | (*(x+1)) << 8 |(*(x+2)) << 16 | (*(x+3)) << 24 ) 630 #define get_le16(x) ((*(x)) | (*(x+1)) << 8) 631 632 int parse_coff(uint8_t *buf, size_t sz) { 633 unsigned int nsections, symtab_ptr, symtab_sz, strtab_ptr; 634 unsigned int sectionrawdata_ptr; 635 unsigned int i; 636 uint8_t *ptr; 637 uint32_t symoffset; 638 639 char **sectionlist; // this array holds all section names in their correct order. 640 // it is used to check if the symbol is in .bss or .rdata section. 641 642 nsections = get_le16(buf + 2); 643 symtab_ptr = get_le32(buf + 8); 644 symtab_sz = get_le32(buf + 12); 645 strtab_ptr = symtab_ptr + symtab_sz * 18; 646 647 if (nsections > 96) { 648 log_msg("Too many sections\n"); 649 return 1; 650 } 651 652 sectionlist = malloc(nsections * sizeof(sectionlist)); 653 654 if (sectionlist == NULL) { 655 log_msg("Allocating first level of section list failed\n"); 656 return 1; 657 } 658 659 // log_msg("COFF: Found %u symbols in %u sections.\n", symtab_sz, nsections); 660 661 /* 662 The size of optional header is always zero for an obj file. So, the section header 663 follows the file header immediately. 664 */ 665 666 ptr = buf + 20; // section header 667 668 for (i = 0; i < nsections; i++) { 669 char sectionname[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0}; 670 strncpy(sectionname, ptr, 8); 671 // log_msg("COFF: Parsing section %s\n",sectionname); 672 673 sectionlist[i] = malloc(strlen(sectionname) + 1); 674 675 if (sectionlist[i] == NULL) { 676 log_msg("Allocating storage for %s failed\n", sectionname); 677 goto bail; 678 } 679 strcpy(sectionlist[i], sectionname); 680 681 // check if it's .rdata and is not a COMDAT section. 682 if (!strcmp(sectionname, ".rdata") && 683 (get_le32(ptr + 36) & 0x1000) == 0) { 684 sectionrawdata_ptr = get_le32(ptr + 20); 685 } 686 687 ptr += 40; 688 } 689 690 // log_msg("COFF: Symbol table at offset %u\n", symtab_ptr); 691 // log_msg("COFF: raw data pointer ofset for section .rdata is %u\n", sectionrawdata_ptr); 692 693 /* The compiler puts the data with non-zero offset in .rdata section, but puts the data with 694 zero offset in .bss section. So, if the data in in .bss section, set offset=0. 695 Note from Wiki: In an object module compiled from C, the bss section contains 696 the local variables (but not functions) that were declared with the static keyword, 697 except for those with non-zero initial values. (In C, static variables are initialized 698 to zero by default.) It also contains the non-local (both extern and static) variables 699 that are also initialized to zero (either explicitly or by default). 700 */ 701 // move to symbol table 702 /* COFF symbol table: 703 offset field 704 0 Name(*) 705 8 Value 706 12 SectionNumber 707 14 Type 708 16 StorageClass 709 17 NumberOfAuxSymbols 710 */ 711 ptr = buf + symtab_ptr; 712 713 for (i = 0; i < symtab_sz; i++) { 714 int16_t section = get_le16(ptr + 12); // section number 715 716 if (section > 0 && ptr[16] == 2) { 717 // if(section > 0 && ptr[16] == 3 && get_le32(ptr+8)) { 718 719 if (get_le32(ptr)) { 720 char name[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0}; 721 strncpy(name, ptr, 8); 722 // log_msg("COFF: Parsing symbol %s\n",name); 723 /* The 64bit Windows compiler doesn't prefix with an _. 724 * Check what's there, and bump if necessary 725 */ 726 if (name[0] == '_') 727 printf("%-40s EQU ", name + 1); 728 else 729 printf("%-40s EQU ", name); 730 } else { 731 // log_msg("COFF: Parsing symbol %s\n", 732 // buf + strtab_ptr + get_le32(ptr+4)); 733 if ((buf + strtab_ptr + get_le32(ptr + 4))[0] == '_') 734 printf("%-40s EQU ", 735 buf + strtab_ptr + get_le32(ptr + 4) + 1); 736 else 737 printf("%-40s EQU ", buf + strtab_ptr + get_le32(ptr + 4)); 738 } 739 740 if (!(strcmp(sectionlist[section - 1], ".bss"))) { 741 symoffset = 0; 742 } else { 743 symoffset = get_le32(buf + sectionrawdata_ptr + get_le32(ptr + 8)); 744 } 745 746 // log_msg(" Section: %d\n",section); 747 // log_msg(" Class: %d\n",ptr[16]); 748 // log_msg(" Address: %u\n",get_le32(ptr+8)); 749 // log_msg(" Offset: %u\n", symoffset); 750 751 printf("%5d\n", symoffset); 752 } 753 754 ptr += 18; 755 } 756 757 printf(" END\n"); 758 759 for (i = 0; i < nsections; i++) { 760 free(sectionlist[i]); 761 } 762 763 free(sectionlist); 764 765 return 0; 766 bail: 767 768 for (i = 0; i < nsections; i++) { 769 free(sectionlist[i]); 770 } 771 772 free(sectionlist); 773 774 return 1; 775 } 776 #endif /* defined(_MSC_VER) || defined(__MINGW32__) || defined(__CYGWIN__) */ 777 778 int main(int argc, char **argv) { 779 output_fmt_t mode = OUTPUT_FMT_PLAIN; 780 const char *f; 781 uint8_t *file_buf; 782 int res; 783 FILE *fp; 784 long int file_size; 785 786 if (argc < 2 || argc > 3) { 787 fprintf(stderr, "Usage: %s [output format] <obj file>\n\n", argv[0]); 788 fprintf(stderr, " <obj file>\tobject file to parse\n"); 789 fprintf(stderr, "Output Formats:\n"); 790 fprintf(stderr, " gas - compatible with GNU assembler\n"); 791 fprintf(stderr, " rvds - compatible with armasm\n"); 792 fprintf(stderr, " cheader - c/c++ header file\n"); 793 goto bail; 794 } 795 796 f = argv[2]; 797 798 if (!strcmp(argv[1], "rvds")) 799 mode = OUTPUT_FMT_RVDS; 800 else if (!strcmp(argv[1], "gas")) 801 mode = OUTPUT_FMT_GAS; 802 else if (!strcmp(argv[1], "cheader")) 803 mode = OUTPUT_FMT_C_HEADER; 804 else 805 f = argv[1]; 806 807 fp = fopen(f, "rb"); 808 809 if (!fp) { 810 perror("Unable to open file"); 811 goto bail; 812 } 813 814 if (fseek(fp, 0, SEEK_END)) { 815 perror("stat"); 816 goto bail; 817 } 818 819 file_size = ftell(fp); 820 file_buf = malloc(file_size); 821 822 if (!file_buf) { 823 perror("malloc"); 824 goto bail; 825 } 826 827 rewind(fp); 828 829 if (fread(file_buf, sizeof(char), file_size, fp) != file_size) { 830 perror("read"); 831 goto bail; 832 } 833 834 if (fclose(fp)) { 835 perror("close"); 836 goto bail; 837 } 838 839 #if defined(__GNUC__) && __GNUC__ 840 #if defined(__MACH__) 841 res = parse_macho(file_buf, file_size, mode); 842 #elif defined(__ELF__) 843 res = parse_elf(file_buf, file_size, mode); 844 #endif 845 #endif 846 #if defined(_MSC_VER) || defined(__MINGW32__) || defined(__CYGWIN__) 847 res = parse_coff(file_buf, file_size); 848 #endif 849 850 free(file_buf); 851 852 if (!res) 853 return EXIT_SUCCESS; 854 855 bail: 856 return EXIT_FAILURE; 857 } 858