ltrace-elf.c revision 3c636fb789a29cac0c8f7f0982fb17afeee489dc
1/* 2 * This file is part of ltrace. 3 * Copyright (C) 2006,2010,2011,2012,2013 Petr Machata, Red Hat Inc. 4 * Copyright (C) 2010 Zachary T Welch, CodeSourcery 5 * Copyright (C) 2010 Joe Damato 6 * Copyright (C) 1997,1998,2001,2004,2007,2008,2009 Juan Cespedes 7 * Copyright (C) 2006 Olaf Hering, SUSE Linux GmbH 8 * Copyright (C) 2006 Eric Vaitl, Cisco Systems, Inc. 9 * Copyright (C) 2006 Paul Gilliam, IBM Corporation 10 * Copyright (C) 2006 Ian Wienand 11 * 12 * This program is free software; you can redistribute it and/or 13 * modify it under the terms of the GNU General Public License as 14 * published by the Free Software Foundation; either version 2 of the 15 * License, or (at your option) any later version. 16 * 17 * This program is distributed in the hope that it will be useful, but 18 * WITHOUT ANY WARRANTY; without even the implied warranty of 19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 20 * General Public License for more details. 21 * 22 * You should have received a copy of the GNU General Public License 23 * along with this program; if not, write to the Free Software 24 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 25 * 02110-1301 USA 26 */ 27 28#include "config.h" 29 30#include <assert.h> 31#ifdef __linux__ 32#include <endian.h> 33#endif 34#include <errno.h> 35#include <fcntl.h> 36#include <gelf.h> 37#include <inttypes.h> 38#include <search.h> 39#include <stdint.h> 40#include <stdio.h> 41#include <stdlib.h> 42#include <string.h> 43#include <strings.h> 44#include <unistd.h> 45 46#include "backend.h" 47#include "filter.h" 48#include "library.h" 49#include "ltrace-elf.h" 50#include "proc.h" 51#include "debug.h" 52#include "options.h" 53 54#ifndef ARCH_HAVE_LTELF_DATA 55int 56arch_elf_init(struct ltelf *lte, struct library *lib) 57{ 58 return 0; 59} 60 61void 62arch_elf_destroy(struct ltelf *lte) 63{ 64} 65#endif 66 67int 68default_elf_add_plt_entry(struct process *proc, struct ltelf *lte, 69 const char *a_name, GElf_Rela *rela, size_t ndx, 70 struct library_symbol **ret) 71{ 72 char *name = strdup(a_name); 73 if (name == NULL) { 74 fail_message: 75 fprintf(stderr, "Couldn't create symbol for PLT entry: %s\n", 76 strerror(errno)); 77 fail: 78 free(name); 79 return -1; 80 } 81 82 GElf_Addr addr = arch_plt_sym_val(lte, ndx, rela); 83 84 struct library_symbol *libsym = malloc(sizeof(*libsym)); 85 if (libsym == NULL) 86 goto fail_message; 87 88 /* XXX The double cast should be removed when 89 * arch_addr_t becomes integral type. */ 90 arch_addr_t taddr = (arch_addr_t) 91 (uintptr_t)(addr + lte->bias); 92 93 if (library_symbol_init(libsym, taddr, name, 1, LS_TOPLT_EXEC) < 0) { 94 free(libsym); 95 goto fail; 96 } 97 98 libsym->next = *ret; 99 *ret = libsym; 100 return 0; 101} 102 103#ifndef ARCH_HAVE_ADD_PLT_ENTRY 104enum plt_status 105arch_elf_add_plt_entry(struct process *proc, struct ltelf *lte, 106 const char *a_name, GElf_Rela *rela, size_t ndx, 107 struct library_symbol **ret) 108{ 109 return PLT_DEFAULT; 110} 111#endif 112 113Elf_Data * 114elf_loaddata(Elf_Scn *scn, GElf_Shdr *shdr) 115{ 116 Elf_Data *data = elf_getdata(scn, NULL); 117 if (data == NULL || elf_getdata(scn, data) != NULL 118 || data->d_off || data->d_size != shdr->sh_size) 119 return NULL; 120 return data; 121} 122 123static int 124elf_get_section_if(struct ltelf *lte, Elf_Scn **tgt_sec, GElf_Shdr *tgt_shdr, 125 int (*predicate)(Elf_Scn *, GElf_Shdr *, void *data), 126 void *data) 127{ 128 int i; 129 for (i = 1; i < lte->ehdr.e_shnum; ++i) { 130 Elf_Scn *scn; 131 GElf_Shdr shdr; 132 133 scn = elf_getscn(lte->elf, i); 134 if (scn == NULL || gelf_getshdr(scn, &shdr) == NULL) { 135 debug(1, "Couldn't read section or header."); 136 return -1; 137 } 138 if (predicate(scn, &shdr, data)) { 139 *tgt_sec = scn; 140 *tgt_shdr = shdr; 141 return 0; 142 } 143 } 144 145 *tgt_sec = NULL; 146 return 0; 147} 148 149static int 150inside_p(Elf_Scn *scn, GElf_Shdr *shdr, void *data) 151{ 152 GElf_Addr addr = *(GElf_Addr *)data; 153 return addr >= shdr->sh_addr 154 && addr < shdr->sh_addr + shdr->sh_size; 155} 156 157int 158elf_get_section_covering(struct ltelf *lte, GElf_Addr addr, 159 Elf_Scn **tgt_sec, GElf_Shdr *tgt_shdr) 160{ 161 return elf_get_section_if(lte, tgt_sec, tgt_shdr, 162 &inside_p, &addr); 163} 164 165static int 166type_p(Elf_Scn *scn, GElf_Shdr *shdr, void *data) 167{ 168 GElf_Word type = *(GElf_Word *)data; 169 return shdr->sh_type == type; 170} 171 172int 173elf_get_section_type(struct ltelf *lte, GElf_Word type, 174 Elf_Scn **tgt_sec, GElf_Shdr *tgt_shdr) 175{ 176 return elf_get_section_if(lte, tgt_sec, tgt_shdr, 177 &type_p, &type); 178} 179 180struct section_named_data { 181 struct ltelf *lte; 182 const char *name; 183}; 184 185static int 186name_p(Elf_Scn *scn, GElf_Shdr *shdr, void *d) 187{ 188 struct section_named_data *data = d; 189 const char *name = elf_strptr(data->lte->elf, 190 data->lte->ehdr.e_shstrndx, 191 shdr->sh_name); 192 return strcmp(name, data->name) == 0; 193} 194 195int 196elf_get_section_named(struct ltelf *lte, const char *name, 197 Elf_Scn **tgt_sec, GElf_Shdr *tgt_shdr) 198{ 199 struct section_named_data data = { 200 .lte = lte, 201 .name = name, 202 }; 203 return elf_get_section_if(lte, tgt_sec, tgt_shdr, 204 &name_p, &data); 205} 206 207static int 208need_data(Elf_Data *data, GElf_Xword offset, GElf_Xword size) 209{ 210 assert(data != NULL); 211 if (data->d_size < size || offset > data->d_size - size) { 212 debug(1, "Not enough data to read %"PRId64"-byte value" 213 " at offset %"PRId64".", size, offset); 214 return -1; 215 } 216 return 0; 217} 218 219#define DEF_READER(NAME, SIZE) \ 220 int \ 221 NAME(Elf_Data *data, GElf_Xword offset, uint##SIZE##_t *retp) \ 222 { \ 223 if (need_data(data, offset, SIZE / 8) < 0) \ 224 return -1; \ 225 \ 226 if (data->d_buf == NULL) /* NODATA section */ { \ 227 *retp = 0; \ 228 return 0; \ 229 } \ 230 \ 231 union { \ 232 uint##SIZE##_t dst; \ 233 char buf[0]; \ 234 } u; \ 235 memcpy(u.buf, data->d_buf + offset, sizeof(u.dst)); \ 236 *retp = u.dst; \ 237 return 0; \ 238 } 239 240DEF_READER(elf_read_u8, 8) 241DEF_READER(elf_read_u16, 16) 242DEF_READER(elf_read_u32, 32) 243DEF_READER(elf_read_u64, 64) 244 245#undef DEF_READER 246 247int 248open_elf(struct ltelf *lte, const char *filename) 249{ 250 lte->fd = open(filename, O_RDONLY); 251 if (lte->fd == -1) 252 return 1; 253 254 elf_version(EV_CURRENT); 255 256#ifdef HAVE_ELF_C_READ_MMAP 257 lte->elf = elf_begin(lte->fd, ELF_C_READ_MMAP, NULL); 258#else 259 lte->elf = elf_begin(lte->fd, ELF_C_READ, NULL); 260#endif 261 262 if (lte->elf == NULL || elf_kind(lte->elf) != ELF_K_ELF) { 263 fprintf(stderr, "\"%s\" is not an ELF file\n", filename); 264 exit(EXIT_FAILURE); 265 } 266 267 if (gelf_getehdr(lte->elf, <e->ehdr) == NULL) { 268 fprintf(stderr, "can't read ELF header of \"%s\": %s\n", 269 filename, elf_errmsg(-1)); 270 exit(EXIT_FAILURE); 271 } 272 273 if (lte->ehdr.e_type != ET_EXEC && lte->ehdr.e_type != ET_DYN) { 274 fprintf(stderr, "\"%s\" is neither an ELF executable" 275 " nor a shared library\n", filename); 276 exit(EXIT_FAILURE); 277 } 278 279 if (1 280#ifdef LT_ELF_MACHINE 281 && (lte->ehdr.e_ident[EI_CLASS] != LT_ELFCLASS 282 || lte->ehdr.e_machine != LT_ELF_MACHINE) 283#endif 284#ifdef LT_ELF_MACHINE2 285 && (lte->ehdr.e_ident[EI_CLASS] != LT_ELFCLASS2 286 || lte->ehdr.e_machine != LT_ELF_MACHINE2) 287#endif 288#ifdef LT_ELF_MACHINE3 289 && (lte->ehdr.e_ident[EI_CLASS] != LT_ELFCLASS3 290 || lte->ehdr.e_machine != LT_ELF_MACHINE3) 291#endif 292 ) { 293 fprintf(stderr, 294 "\"%s\" is ELF from incompatible architecture\n", 295 filename); 296 exit(EXIT_FAILURE); 297 } 298 299 return 0; 300} 301 302static void 303read_symbol_table(struct ltelf *lte, const char *filename, 304 Elf_Scn *scn, GElf_Shdr *shdr, const char *name, 305 Elf_Data **datap, size_t *countp, const char **strsp) 306{ 307 *datap = elf_getdata(scn, NULL); 308 *countp = shdr->sh_size / shdr->sh_entsize; 309 if ((*datap == NULL || elf_getdata(scn, *datap) != NULL) 310 && options.static_filter != NULL) { 311 fprintf(stderr, "Couldn't get data of section" 312 " %s from \"%s\": %s\n", 313 name, filename, elf_errmsg(-1)); 314 exit(EXIT_FAILURE); 315 } 316 317 scn = elf_getscn(lte->elf, shdr->sh_link); 318 GElf_Shdr shdr2; 319 if (scn == NULL || gelf_getshdr(scn, &shdr2) == NULL) { 320 fprintf(stderr, "Couldn't get header of section" 321 " #%d from \"%s\": %s\n", 322 shdr->sh_link, filename, elf_errmsg(-1)); 323 exit(EXIT_FAILURE); 324 } 325 326 Elf_Data *data = elf_getdata(scn, NULL); 327 if (data == NULL || elf_getdata(scn, data) != NULL 328 || shdr2.sh_size != data->d_size || data->d_off) { 329 fprintf(stderr, "Couldn't get data of section" 330 " #%d from \"%s\": %s\n", 331 shdr2.sh_link, filename, elf_errmsg(-1)); 332 exit(EXIT_FAILURE); 333 } 334 335 *strsp = data->d_buf; 336} 337 338static int 339do_init_elf(struct ltelf *lte, const char *filename) 340{ 341 int i; 342 GElf_Addr relplt_addr = 0; 343 GElf_Addr soname_offset = 0; 344 345 debug(DEBUG_FUNCTION, "do_init_elf(filename=%s)", filename); 346 debug(1, "Reading ELF from %s...", filename); 347 348 for (i = 1; i < lte->ehdr.e_shnum; ++i) { 349 Elf_Scn *scn; 350 GElf_Shdr shdr; 351 const char *name; 352 353 scn = elf_getscn(lte->elf, i); 354 if (scn == NULL || gelf_getshdr(scn, &shdr) == NULL) { 355 fprintf(stderr, "Couldn't get section #%d from" 356 " \"%s\": %s\n", i, filename, elf_errmsg(-1)); 357 exit(EXIT_FAILURE); 358 } 359 360 name = elf_strptr(lte->elf, lte->ehdr.e_shstrndx, shdr.sh_name); 361 if (name == NULL) { 362 fprintf(stderr, "Couldn't get name of section #%d from" 363 " \"%s\": %s\n", i, filename, elf_errmsg(-1)); 364 exit(EXIT_FAILURE); 365 } 366 367 if (shdr.sh_type == SHT_SYMTAB) { 368 read_symbol_table(lte, filename, 369 scn, &shdr, name, <e->symtab, 370 <e->symtab_count, <e->strtab); 371 372 } else if (shdr.sh_type == SHT_DYNSYM) { 373 read_symbol_table(lte, filename, 374 scn, &shdr, name, <e->dynsym, 375 <e->dynsym_count, <e->dynstr); 376 377 } else if (shdr.sh_type == SHT_DYNAMIC) { 378 Elf_Data *data; 379 size_t j; 380 381 lte->dyn_addr = shdr.sh_addr + lte->bias; 382 lte->dyn_sz = shdr.sh_size; 383 384 data = elf_getdata(scn, NULL); 385 if (data == NULL || elf_getdata(scn, data) != NULL) { 386 fprintf(stderr, "Couldn't get .dynamic data" 387 " from \"%s\": %s\n", 388 filename, strerror(errno)); 389 exit(EXIT_FAILURE); 390 } 391 392 for (j = 0; j < shdr.sh_size / shdr.sh_entsize; ++j) { 393 GElf_Dyn dyn; 394 395 if (gelf_getdyn(data, j, &dyn) == NULL) { 396 fprintf(stderr, "Couldn't get .dynamic" 397 " data from \"%s\": %s\n", 398 filename, strerror(errno)); 399 exit(EXIT_FAILURE); 400 } 401 if (dyn.d_tag == DT_JMPREL) 402 relplt_addr = dyn.d_un.d_ptr; 403 else if (dyn.d_tag == DT_PLTRELSZ) 404 lte->relplt_size = dyn.d_un.d_val; 405 else if (dyn.d_tag == DT_SONAME) 406 soname_offset = dyn.d_un.d_val; 407 } 408 } else if (shdr.sh_type == SHT_PROGBITS 409 || shdr.sh_type == SHT_NOBITS) { 410 if (strcmp(name, ".plt") == 0) { 411 lte->plt_addr = shdr.sh_addr; 412 lte->plt_size = shdr.sh_size; 413 lte->plt_data = elf_loaddata(scn, &shdr); 414 if (lte->plt_data == NULL) 415 fprintf(stderr, 416 "Can't load .plt data\n"); 417 lte->plt_flags = shdr.sh_flags; 418 } 419#ifdef ARCH_SUPPORTS_OPD 420 else if (strcmp(name, ".opd") == 0) { 421 lte->opd_addr = (GElf_Addr *) (long) shdr.sh_addr; 422 lte->opd_size = shdr.sh_size; 423 lte->opd = elf_rawdata(scn, NULL); 424 } 425#endif 426 } 427 } 428 429 if (lte->dynsym == NULL || lte->dynstr == NULL) { 430 fprintf(stderr, "Couldn't find .dynsym or .dynstr in \"%s\"\n", 431 filename); 432 exit(EXIT_FAILURE); 433 } 434 435 if (!relplt_addr || !lte->plt_addr) { 436 debug(1, "%s has no PLT relocations", filename); 437 lte->relplt = NULL; 438 lte->relplt_count = 0; 439 } else if (lte->relplt_size == 0) { 440 debug(1, "%s has unknown PLT size", filename); 441 lte->relplt = NULL; 442 lte->relplt_count = 0; 443 } else { 444 445 for (i = 1; i < lte->ehdr.e_shnum; ++i) { 446 Elf_Scn *scn; 447 GElf_Shdr shdr; 448 449 scn = elf_getscn(lte->elf, i); 450 if (scn == NULL || gelf_getshdr(scn, &shdr) == NULL) { 451 fprintf(stderr, "Couldn't get section header" 452 " from \"%s\": %s\n", 453 filename, elf_errmsg(-1)); 454 exit(EXIT_FAILURE); 455 } 456 if (shdr.sh_addr == relplt_addr 457 && shdr.sh_size == lte->relplt_size) { 458 lte->relplt = elf_getdata(scn, NULL); 459 lte->relplt_count = 460 shdr.sh_size / shdr.sh_entsize; 461 if (lte->relplt == NULL 462 || elf_getdata(scn, lte->relplt) != NULL) { 463 fprintf(stderr, "Couldn't get .rel*.plt" 464 " data from \"%s\": %s\n", 465 filename, elf_errmsg(-1)); 466 exit(EXIT_FAILURE); 467 } 468 break; 469 } 470 } 471 472 if (i == lte->ehdr.e_shnum) { 473 fprintf(stderr, 474 "Couldn't find .rel*.plt section in \"%s\"\n", 475 filename); 476 exit(EXIT_FAILURE); 477 } 478 479 debug(1, "%s %zd PLT relocations", filename, lte->relplt_count); 480 } 481 482 if (soname_offset != 0) 483 lte->soname = lte->dynstr + soname_offset; 484 485 return 0; 486} 487 488void 489do_close_elf(struct ltelf *lte) 490{ 491 debug(DEBUG_FUNCTION, "do_close_elf()"); 492 arch_elf_destroy(lte); 493 elf_end(lte->elf); 494 close(lte->fd); 495} 496 497int 498elf_get_sym_info(struct ltelf *lte, const char *filename, 499 size_t sym_index, GElf_Rela *rela, GElf_Sym *sym) 500{ 501 int i = sym_index; 502 GElf_Rel rel; 503 void *ret; 504 505 if (lte->relplt->d_type == ELF_T_REL) { 506 ret = gelf_getrel(lte->relplt, i, &rel); 507 rela->r_offset = rel.r_offset; 508 rela->r_info = rel.r_info; 509 rela->r_addend = 0; 510 } else { 511 ret = gelf_getrela(lte->relplt, i, rela); 512 } 513 514 if (ret == NULL 515 || ELF64_R_SYM(rela->r_info) >= lte->dynsym_count 516 || gelf_getsym(lte->dynsym, ELF64_R_SYM(rela->r_info), 517 sym) == NULL) { 518 fprintf(stderr, 519 "Couldn't get relocation from \"%s\": %s\n", 520 filename, elf_errmsg(-1)); 521 exit(EXIT_FAILURE); 522 } 523 524 return 0; 525} 526 527#ifndef ARCH_HAVE_GET_SYMINFO 528int 529arch_get_sym_info(struct ltelf *lte, const char *filename, 530 size_t sym_index, GElf_Rela *rela, GElf_Sym *sym) 531{ 532 return elf_get_sym_info(lte, filename, sym_index, rela, sym); 533} 534#endif 535 536static void 537mark_chain_latent(struct library_symbol *libsym) 538{ 539 for (; libsym != NULL; libsym = libsym->next) { 540 debug(DEBUG_FUNCTION, "marking %s latent", libsym->name); 541 libsym->latent = 1; 542 } 543} 544 545static int 546populate_plt(struct process *proc, const char *filename, 547 struct ltelf *lte, struct library *lib, 548 int latent_plts) 549{ 550 size_t i; 551 for (i = 0; i < lte->relplt_count; ++i) { 552 GElf_Rela rela; 553 GElf_Sym sym; 554 555 if (arch_get_sym_info(lte, filename, i, &rela, &sym) < 0) 556 continue; /* Skip this entry. */ 557 558 char const *name = lte->dynstr + sym.st_name; 559 560 /* If the symbol wasn't matched, reject it, unless we 561 * need to keep latent PLT breakpoints for tracing 562 * exports. */ 563 int matched = filter_matches_symbol(options.plt_filter, 564 name, lib); 565 if (!matched && !latent_plts) 566 continue; 567 568 struct library_symbol *libsym = NULL; 569 switch (arch_elf_add_plt_entry(proc, lte, name, 570 &rela, i, &libsym)) { 571 case PLT_DEFAULT: 572 if (default_elf_add_plt_entry(proc, lte, name, 573 &rela, i, &libsym) < 0) 574 /* fall-through */ 575 case PLT_FAIL: 576 return -1; 577 /* fall-through */ 578 case PLT_OK: 579 if (libsym != NULL) { 580 /* If we are adding those symbols just 581 * for tracing exports, mark them all 582 * latent. */ 583 if (!matched) 584 mark_chain_latent(libsym); 585 library_add_symbol(lib, libsym); 586 } 587 } 588 } 589 return 0; 590} 591 592/* When -x rules result in request to trace several aliases, we only 593 * want to add such symbol once. The only way that those symbols 594 * differ in is their name, e.g. in glibc you have __GI___libc_free, 595 * __cfree, __free, __libc_free, cfree and free all defined on the 596 * same address. So instead we keep this unique symbol struct for 597 * each address, and replace name in libsym with a shorter variant if 598 * we find it. */ 599struct unique_symbol { 600 arch_addr_t addr; 601 struct library_symbol *libsym; 602}; 603 604static int 605unique_symbol_cmp(const void *key, const void *val) 606{ 607 const struct unique_symbol *sym_key = key; 608 const struct unique_symbol *sym_val = val; 609 return sym_key->addr != sym_val->addr; 610} 611 612static enum callback_status 613symbol_with_address(struct library_symbol *sym, void *addrptr) 614{ 615 return sym->enter_addr == *(arch_addr_t *)addrptr 616 ? CBS_STOP : CBS_CONT; 617} 618 619static int 620populate_this_symtab(struct process *proc, const char *filename, 621 struct ltelf *lte, struct library *lib, 622 Elf_Data *symtab, const char *strtab, size_t size, 623 struct library_exported_name **names) 624{ 625 /* If a valid NAMES is passed, we pass in *NAMES a list of 626 * symbol names that this library exports. */ 627 if (names != NULL) 628 *names = NULL; 629 630 /* Using sorted array would be arguably better, but this 631 * should be well enough for the number of symbols that we 632 * typically deal with. */ 633 size_t num_symbols = 0; 634 struct unique_symbol *symbols = malloc(sizeof(*symbols) * size); 635 if (symbols == NULL) { 636 fprintf(stderr, "couldn't insert symbols for -x: %s\n", 637 strerror(errno)); 638 return -1; 639 } 640 641 GElf_Word secflags[lte->ehdr.e_shnum]; 642 size_t i; 643 for (i = 1; i < lte->ehdr.e_shnum; ++i) { 644 Elf_Scn *scn = elf_getscn(lte->elf, i); 645 GElf_Shdr shdr; 646 if (scn == NULL || gelf_getshdr(scn, &shdr) == NULL) 647 secflags[i] = 0; 648 else 649 secflags[i] = shdr.sh_flags; 650 } 651 652 for (i = 0; i < size; ++i) { 653 GElf_Sym sym; 654 if (gelf_getsym(symtab, i, &sym) == NULL) { 655 fail: 656 fprintf(stderr, 657 "couldn't get symbol #%zd from %s: %s\n", 658 i, filename, elf_errmsg(-1)); 659 continue; 660 } 661 662 /* XXX support IFUNC as well. */ 663 if (GELF_ST_TYPE(sym.st_info) != STT_FUNC 664 || sym.st_value == 0 665 || sym.st_shndx == STN_UNDEF) 666 continue; 667 668 /* Find symbol name and snip version. */ 669 const char *orig_name = strtab + sym.st_name; 670 const char *version = strchr(orig_name, '@'); 671 size_t len = version != NULL ? (assert(version > orig_name), 672 (size_t)(version - orig_name)) 673 : strlen(orig_name); 674 char name[len + 1]; 675 memcpy(name, orig_name, len); 676 name[len] = 0; 677 678 /* If we are interested in exports, store this name. */ 679 char *name_copy = NULL; 680 if (names != NULL) { 681 struct library_exported_name *export = NULL; 682 name_copy = strdup(name); 683 684 if (name_copy == NULL 685 || (export = malloc(sizeof(*export))) == NULL) { 686 free(name_copy); 687 fprintf(stderr, "Couldn't store symbol %s. " 688 "Tracing may be incomplete.\n", name); 689 } else { 690 export->name = name_copy; 691 export->own_name = 1; 692 export->next = *names; 693 *names = export; 694 } 695 } 696 697 /* If the symbol is not matched, skip it. We already 698 * stored it to export list above. */ 699 if (!filter_matches_symbol(options.static_filter, name, lib)) 700 continue; 701 702 arch_addr_t addr = (arch_addr_t) 703 (uintptr_t)(sym.st_value + lte->bias); 704 arch_addr_t naddr; 705 706 /* On arches that support OPD, the value of typical 707 * function symbol will be a pointer to .opd, but some 708 * will point directly to .text. We don't want to 709 * translate those. */ 710 if (secflags[sym.st_shndx] & SHF_EXECINSTR) { 711 naddr = addr; 712 } else if (arch_translate_address(lte, addr, &naddr) < 0) { 713 fprintf(stderr, 714 "couldn't translate address of %s@%s: %s\n", 715 name, lib->soname, strerror(errno)); 716 continue; 717 } 718 719 char *full_name; 720 int own_full_name = 1; 721 if (name_copy == NULL) { 722 full_name = strdup(name); 723 if (full_name == NULL) 724 goto fail; 725 } else { 726 full_name = name_copy; 727 own_full_name = 0; 728 } 729 730 /* Look whether we already have a symbol for this 731 * address. If not, add this one. */ 732 struct unique_symbol key = { naddr, NULL }; 733 struct unique_symbol *unique 734 = lsearch(&key, symbols, &num_symbols, 735 sizeof(*symbols), &unique_symbol_cmp); 736 737 if (unique->libsym == NULL) { 738 struct library_symbol *libsym = malloc(sizeof(*libsym)); 739 if (libsym == NULL 740 || library_symbol_init(libsym, naddr, 741 full_name, own_full_name, 742 LS_TOPLT_NONE) < 0) { 743 --num_symbols; 744 goto fail; 745 } 746 unique->libsym = libsym; 747 unique->addr = naddr; 748 749 } else if (strlen(full_name) < strlen(unique->libsym->name)) { 750 library_symbol_set_name(unique->libsym, 751 full_name, own_full_name); 752 753 } else if (own_full_name) { 754 free(full_name); 755 } 756 } 757 758 /* Now we do the union of this set of unique symbols with 759 * what's already in the library. */ 760 for (i = 0; i < num_symbols; ++i) { 761 struct library_symbol *this_sym = symbols[i].libsym; 762 assert(this_sym != NULL); 763 struct library_symbol *other 764 = library_each_symbol(lib, NULL, symbol_with_address, 765 &this_sym->enter_addr); 766 if (other != NULL) { 767 library_symbol_destroy(this_sym); 768 free(this_sym); 769 symbols[i].libsym = NULL; 770 } 771 } 772 773 for (i = 0; i < num_symbols; ++i) 774 if (symbols[i].libsym != NULL) 775 library_add_symbol(lib, symbols[i].libsym); 776 777 free(symbols); 778 return 0; 779} 780 781static int 782populate_symtab(struct process *proc, const char *filename, 783 struct ltelf *lte, struct library *lib, 784 int symtabs, int exports) 785{ 786 int status; 787 if (symtabs && lte->symtab != NULL && lte->strtab != NULL 788 && (status = populate_this_symtab(proc, filename, lte, lib, 789 lte->symtab, lte->strtab, 790 lte->symtab_count, NULL)) < 0) 791 return status; 792 793 /* Check whether we want to trace symbols implemented by this 794 * library (-l). */ 795 struct library_exported_name **names = NULL; 796 if (exports) { 797 debug(DEBUG_FUNCTION, "-l matches %s", lib->soname); 798 names = &lib->exported_names; 799 } 800 801 return populate_this_symtab(proc, filename, lte, lib, 802 lte->dynsym, lte->dynstr, 803 lte->dynsym_count, names); 804} 805 806static int 807read_module(struct library *lib, struct process *proc, 808 const char *filename, GElf_Addr bias, int main) 809{ 810 struct ltelf lte = {}; 811 if (open_elf(<e, filename) < 0) 812 return -1; 813 814 /* XXX When we abstract ABI into a module, this should instead 815 * become something like 816 * 817 * proc->abi = arch_get_abi(lte.ehdr); 818 * 819 * The code in open_elf needs to be replaced by this logic. 820 * Be warned that libltrace.c calls open_elf as well to 821 * determine whether ABI is supported. This is to get 822 * reasonable error messages when trying to run 64-bit binary 823 * with 32-bit ltrace. It is desirable to preserve this. */ 824 proc->e_machine = lte.ehdr.e_machine; 825 proc->e_class = lte.ehdr.e_ident[EI_CLASS]; 826 get_arch_dep(proc); 827 828 /* Find out the base address. For PIE main binaries we look 829 * into auxv, otherwise we scan phdrs. */ 830 if (main && lte.ehdr.e_type == ET_DYN) { 831 arch_addr_t entry; 832 if (process_get_entry(proc, &entry, NULL) < 0) { 833 fprintf(stderr, "Couldn't find entry of PIE %s\n", 834 filename); 835 return -1; 836 } 837 /* XXX The double cast should be removed when 838 * arch_addr_t becomes integral type. */ 839 lte.entry_addr = (GElf_Addr)(uintptr_t)entry; 840 lte.bias = (GElf_Addr)(uintptr_t)entry - lte.ehdr.e_entry; 841 842 } else { 843 GElf_Phdr phdr; 844 size_t i; 845 for (i = 0; gelf_getphdr (lte.elf, i, &phdr) != NULL; ++i) { 846 if (phdr.p_type == PT_LOAD) { 847 lte.base_addr = phdr.p_vaddr + bias; 848 break; 849 } 850 } 851 852 lte.bias = bias; 853 lte.entry_addr = lte.ehdr.e_entry + lte.bias; 854 855 if (lte.base_addr == 0) { 856 fprintf(stderr, 857 "Couldn't determine base address of %s\n", 858 filename); 859 return -1; 860 } 861 } 862 863 if (do_init_elf(<e, filename) < 0) 864 return -1; 865 866 if (arch_elf_init(<e, lib) < 0) { 867 fprintf(stderr, "Backend initialization failed.\n"); 868 return -1; 869 } 870 871 int status = 0; 872 if (lib == NULL) 873 goto fail; 874 875 /* Note that we set soname and pathname as soon as they are 876 * allocated, so in case of further errors, this get released 877 * when LIB is released, which should happen in the caller 878 * when we return error. */ 879 880 if (lib->pathname == NULL) { 881 char *pathname = strdup(filename); 882 if (pathname == NULL) 883 goto fail; 884 library_set_pathname(lib, pathname, 1); 885 } 886 887 if (lte.soname != NULL) { 888 char *soname = strdup(lte.soname); 889 if (soname == NULL) 890 goto fail; 891 library_set_soname(lib, soname, 1); 892 } else { 893 const char *soname = rindex(lib->pathname, '/'); 894 if (soname != NULL) 895 soname += 1; 896 else 897 soname = lib->pathname; 898 library_set_soname(lib, soname, 0); 899 } 900 901 /* XXX The double cast should be removed when 902 * arch_addr_t becomes integral type. */ 903 arch_addr_t entry = (arch_addr_t)(uintptr_t)lte.entry_addr; 904 if (arch_translate_address(<e, entry, &entry) < 0) 905 goto fail; 906 907 /* XXX The double cast should be removed when 908 * arch_addr_t becomes integral type. */ 909 lib->base = (arch_addr_t)(uintptr_t)lte.base_addr; 910 lib->entry = entry; 911 /* XXX The double cast should be removed when 912 * arch_addr_t becomes integral type. */ 913 lib->dyn_addr = (arch_addr_t)(uintptr_t)lte.dyn_addr; 914 915 /* There are two reasons that we need to inspect symbol tables 916 * or populate PLT entries. Either the user requested 917 * corresponding tracing features (respectively -x and -e), or 918 * they requested tracing exported symbols (-l). 919 * 920 * In the latter case we need to keep even those PLT slots 921 * that are not requested by -e (but we keep them latent). We 922 * also need to inspect .dynsym to find what exports this 923 * library provide, to turn on existing latent PLT 924 * entries. */ 925 926 int plts = filter_matches_library(options.plt_filter, lib); 927 if ((plts || options.export_filter != NULL) 928 && populate_plt(proc, filename, <e, lib, 929 options.export_filter != NULL) < 0) 930 goto fail; 931 932 int exports = filter_matches_library(options.export_filter, lib); 933 int symtabs = filter_matches_library(options.static_filter, lib); 934 if ((symtabs || exports) 935 && populate_symtab(proc, filename, <e, lib, 936 symtabs, exports) < 0) 937 goto fail; 938 939done: 940 do_close_elf(<e); 941 return status; 942 943fail: 944 status = -1; 945 goto done; 946} 947 948int 949ltelf_read_library(struct library *lib, struct process *proc, 950 const char *filename, GElf_Addr bias) 951{ 952 return read_module(lib, proc, filename, bias, 0); 953} 954 955 956struct library * 957ltelf_read_main_binary(struct process *proc, const char *path) 958{ 959 struct library *lib = malloc(sizeof(*lib)); 960 if (lib == NULL) 961 return NULL; 962 library_init(lib, LT_LIBTYPE_MAIN); 963 library_set_pathname(lib, path, 0); 964 965 /* There is a race between running the process and reading its 966 * binary for internal consumption. So open the binary from 967 * the /proc filesystem. XXX Note that there is similar race 968 * for libraries, but there we don't have a nice answer like 969 * that. Presumably we could read the DSOs from the process 970 * memory image, but that's not currently done. */ 971 char *fname = pid2name(proc->pid); 972 if (fname == NULL 973 || read_module(lib, proc, fname, 0, 1) < 0) { 974 library_destroy(lib); 975 free(lib); 976 lib = NULL; 977 } 978 979 free(fname); 980 return lib; 981} 982