1 // SPDX-License-Identifier: GPL-2.0 2 #include <fcntl.h> 3 #include <stdio.h> 4 #include <errno.h> 5 #include <stdlib.h> 6 #include <string.h> 7 #include <unistd.h> 8 #include <inttypes.h> 9 10 #include "compress.h" 11 #include "dso.h" 12 #include "map.h" 13 #include "maps.h" 14 #include "symbol.h" 15 #include "symsrc.h" 16 #include "demangle-cxx.h" 17 #include "demangle-ocaml.h" 18 #include "demangle-java.h" 19 #include "demangle-rust.h" 20 #include "machine.h" 21 #include "vdso.h" 22 #include "debug.h" 23 #include "util/copyfile.h" 24 #include <linux/ctype.h> 25 #include <linux/kernel.h> 26 #include <linux/zalloc.h> 27 #include <linux/string.h> 28 #include <symbol/kallsyms.h> 29 #include <internal/lib.h> 30 31 #ifdef HAVE_LIBBFD_SUPPORT 32 #define PACKAGE 'perf' 33 #include <bfd.h> 34 #endif 35 36 #if defined(HAVE_LIBBFD_SUPPORT) || defined(HAVE_CPLUS_DEMANGLE_SUPPORT) 37 #ifndef DMGL_PARAMS 38 #define DMGL_PARAMS (1 << 0) /* Include function args */ 39 #define DMGL_ANSI (1 << 1) /* Include const, volatile, etc */ 40 #endif 41 #endif 42 43 #ifndef EM_AARCH64 44 #define EM_AARCH64 183 /* ARM 64 bit */ 45 #endif 46 47 #ifndef EM_LOONGARCH 48 #define EM_LOONGARCH 258 49 #endif 50 51 #ifndef ELF32_ST_VISIBILITY 52 #define ELF32_ST_VISIBILITY(o) ((o) & 0x03) 53 #endif 54 55 /* For ELF64 the definitions are the same. */ 56 #ifndef ELF64_ST_VISIBILITY 57 #define ELF64_ST_VISIBILITY(o) ELF32_ST_VISIBILITY (o) 58 #endif 59 60 /* How to extract information held in the st_other field. */ 61 #ifndef GELF_ST_VISIBILITY 62 #define GELF_ST_VISIBILITY(val) ELF64_ST_VISIBILITY (val) 63 #endif 64 65 typedef Elf64_Nhdr GElf_Nhdr; 66 67 68 #ifndef HAVE_ELF_GETPHDRNUM_SUPPORT 69 static int elf_getphdrnum(Elf *elf, size_t *dst) 70 { 71 GElf_Ehdr gehdr; 72 GElf_Ehdr *ehdr; 73 74 ehdr = gelf_getehdr(elf, &gehdr); 75 if (!ehdr) 76 return -1; 77 78 *dst = ehdr->e_phnum; 79 80 return 0; 81 } 82 #endif 83 84 #ifndef HAVE_ELF_GETSHDRSTRNDX_SUPPORT 85 static int elf_getshdrstrndx(Elf *elf __maybe_unused, size_t *dst __maybe_unused) 86 { 87 pr_err("%s: update your libelf to > 0.140, this one lacks elf_getshdrstrndx().\n", __func__); 88 return -1; 89 } 90 #endif 91 92 #ifndef NT_GNU_BUILD_ID 93 #define NT_GNU_BUILD_ID 3 94 #endif 95 96 /** 97 * elf_symtab__for_each_symbol - iterate thru all the symbols 98 * 99 * @syms: struct elf_symtab instance to iterate 100 * @idx: uint32_t idx 101 * @sym: GElf_Sym iterator 102 */ 103 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \ 104 for (idx = 0, gelf_getsym(syms, idx, &sym);\ 105 idx < nr_syms; \ 106 idx++, gelf_getsym(syms, idx, &sym)) 107 108 static inline uint8_t elf_sym__type(const GElf_Sym *sym) 109 { 110 return GELF_ST_TYPE(sym->st_info); 111 } 112 113 static inline uint8_t elf_sym__visibility(const GElf_Sym *sym) 114 { 115 return GELF_ST_VISIBILITY(sym->st_other); 116 } 117 118 #ifndef STT_GNU_IFUNC 119 #define STT_GNU_IFUNC 10 120 #endif 121 122 static inline int elf_sym__is_function(const GElf_Sym *sym) 123 { 124 return (elf_sym__type(sym) == STT_FUNC || 125 elf_sym__type(sym) == STT_GNU_IFUNC) && 126 sym->st_name != 0 && 127 sym->st_shndx != SHN_UNDEF; 128 } 129 130 static inline bool elf_sym__is_object(const GElf_Sym *sym) 131 { 132 return elf_sym__type(sym) == STT_OBJECT && 133 sym->st_name != 0 && 134 sym->st_shndx != SHN_UNDEF; 135 } 136 137 static inline int elf_sym__is_label(const GElf_Sym *sym) 138 { 139 return elf_sym__type(sym) == STT_NOTYPE && 140 sym->st_name != 0 && 141 sym->st_shndx != SHN_UNDEF && 142 sym->st_shndx != SHN_ABS && 143 elf_sym__visibility(sym) != STV_HIDDEN && 144 elf_sym__visibility(sym) != STV_INTERNAL; 145 } 146 147 static bool elf_sym__filter(GElf_Sym *sym) 148 { 149 return elf_sym__is_function(sym) || elf_sym__is_object(sym); 150 } 151 152 static inline const char *elf_sym__name(const GElf_Sym *sym, 153 const Elf_Data *symstrs) 154 { 155 return symstrs->d_buf + sym->st_name; 156 } 157 158 static inline const char *elf_sec__name(const GElf_Shdr *shdr, 159 const Elf_Data *secstrs) 160 { 161 return secstrs->d_buf + shdr->sh_name; 162 } 163 164 static inline int elf_sec__is_text(const GElf_Shdr *shdr, 165 const Elf_Data *secstrs) 166 { 167 return strstr(elf_sec__name(shdr, secstrs), "text") != NULL; 168 } 169 170 static inline bool elf_sec__is_data(const GElf_Shdr *shdr, 171 const Elf_Data *secstrs) 172 { 173 return strstr(elf_sec__name(shdr, secstrs), "data") != NULL; 174 } 175 176 static bool elf_sec__filter(GElf_Shdr *shdr, Elf_Data *secstrs) 177 { 178 return elf_sec__is_text(shdr, secstrs) || 179 elf_sec__is_data(shdr, secstrs); 180 } 181 182 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr) 183 { 184 Elf_Scn *sec = NULL; 185 GElf_Shdr shdr; 186 size_t cnt = 1; 187 188 while ((sec = elf_nextscn(elf, sec)) != NULL) { 189 gelf_getshdr(sec, &shdr); 190 191 if ((addr >= shdr.sh_addr) && 192 (addr < (shdr.sh_addr + shdr.sh_size))) 193 return cnt; 194 195 ++cnt; 196 } 197 198 return -1; 199 } 200 201 Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep, 202 GElf_Shdr *shp, const char *name, size_t *idx) 203 { 204 Elf_Scn *sec = NULL; 205 size_t cnt = 1; 206 207 /* ELF is corrupted/truncated, avoid calling elf_strptr. */ 208 if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) 209 return NULL; 210 211 while ((sec = elf_nextscn(elf, sec)) != NULL) { 212 char *str; 213 214 gelf_getshdr(sec, shp); 215 str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name); 216 if (str && !strcmp(name, str)) { 217 if (idx) 218 *idx = cnt; 219 return sec; 220 } 221 ++cnt; 222 } 223 224 return NULL; 225 } 226 227 bool filename__has_section(const char *filename, const char *sec) 228 { 229 int fd; 230 Elf *elf; 231 GElf_Ehdr ehdr; 232 GElf_Shdr shdr; 233 bool found = false; 234 235 fd = open(filename, O_RDONLY); 236 if (fd < 0) 237 return false; 238 239 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL); 240 if (elf == NULL) 241 goto out; 242 243 if (gelf_getehdr(elf, &ehdr) == NULL) 244 goto elf_out; 245 246 found = !!elf_section_by_name(elf, &ehdr, &shdr, sec, NULL); 247 248 elf_out: 249 elf_end(elf); 250 out: 251 close(fd); 252 return found; 253 } 254 255 static int elf_read_program_header(Elf *elf, u64 vaddr, GElf_Phdr *phdr) 256 { 257 size_t i, phdrnum; 258 u64 sz; 259 260 if (elf_getphdrnum(elf, &phdrnum)) 261 return -1; 262 263 for (i = 0; i < phdrnum; i++) { 264 if (gelf_getphdr(elf, i, phdr) == NULL) 265 return -1; 266 267 if (phdr->p_type != PT_LOAD) 268 continue; 269 270 sz = max(phdr->p_memsz, phdr->p_filesz); 271 if (!sz) 272 continue; 273 274 if (vaddr >= phdr->p_vaddr && (vaddr < phdr->p_vaddr + sz)) 275 return 0; 276 } 277 278 /* Not found any valid program header */ 279 return -1; 280 } 281 282 static bool want_demangle(bool is_kernel_sym) 283 { 284 return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle; 285 } 286 287 /* 288 * Demangle C++ function signature, typically replaced by demangle-cxx.cpp 289 * version. 290 */ 291 #ifndef HAVE_CXA_DEMANGLE_SUPPORT 292 char *cxx_demangle_sym(const char *str __maybe_unused, bool params __maybe_unused, 293 bool modifiers __maybe_unused) 294 { 295 #ifdef HAVE_LIBBFD_SUPPORT 296 int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0); 297 298 return bfd_demangle(NULL, str, flags); 299 #elif defined(HAVE_CPLUS_DEMANGLE_SUPPORT) 300 int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0); 301 302 return cplus_demangle(str, flags); 303 #else 304 return NULL; 305 #endif 306 } 307 #endif /* !HAVE_CXA_DEMANGLE_SUPPORT */ 308 309 static char *demangle_sym(struct dso *dso, int kmodule, const char *elf_name) 310 { 311 char *demangled = NULL; 312 313 /* 314 * We need to figure out if the object was created from C++ sources 315 * DWARF DW_compile_unit has this, but we don't always have access 316 * to it... 317 */ 318 if (!want_demangle(dso__kernel(dso) || kmodule)) 319 return demangled; 320 321 demangled = cxx_demangle_sym(elf_name, verbose > 0, verbose > 0); 322 if (demangled == NULL) { 323 demangled = ocaml_demangle_sym(elf_name); 324 if (demangled == NULL) { 325 demangled = java_demangle_sym(elf_name, JAVA_DEMANGLE_NORET); 326 } 327 } 328 else if (rust_is_mangled(demangled)) 329 /* 330 * Input to Rust demangling is the BFD-demangled 331 * name which it Rust-demangles in place. 332 */ 333 rust_demangle_sym(demangled); 334 335 return demangled; 336 } 337 338 struct rel_info { 339 u32 nr_entries; 340 u32 *sorted; 341 bool is_rela; 342 Elf_Data *reldata; 343 GElf_Rela rela; 344 GElf_Rel rel; 345 }; 346 347 static u32 get_rel_symidx(struct rel_info *ri, u32 idx) 348 { 349 idx = ri->sorted ? ri->sorted[idx] : idx; 350 if (ri->is_rela) { 351 gelf_getrela(ri->reldata, idx, &ri->rela); 352 return GELF_R_SYM(ri->rela.r_info); 353 } 354 gelf_getrel(ri->reldata, idx, &ri->rel); 355 return GELF_R_SYM(ri->rel.r_info); 356 } 357 358 static u64 get_rel_offset(struct rel_info *ri, u32 x) 359 { 360 if (ri->is_rela) { 361 GElf_Rela rela; 362 363 gelf_getrela(ri->reldata, x, &rela); 364 return rela.r_offset; 365 } else { 366 GElf_Rel rel; 367 368 gelf_getrel(ri->reldata, x, &rel); 369 return rel.r_offset; 370 } 371 } 372 373 static int rel_cmp(const void *a, const void *b, void *r) 374 { 375 struct rel_info *ri = r; 376 u64 a_offset = get_rel_offset(ri, *(const u32 *)a); 377 u64 b_offset = get_rel_offset(ri, *(const u32 *)b); 378 379 return a_offset < b_offset ? -1 : (a_offset > b_offset ? 1 : 0); 380 } 381 382 static int sort_rel(struct rel_info *ri) 383 { 384 size_t sz = sizeof(ri->sorted[0]); 385 u32 i; 386 387 ri->sorted = calloc(ri->nr_entries, sz); 388 if (!ri->sorted) 389 return -1; 390 for (i = 0; i < ri->nr_entries; i++) 391 ri->sorted[i] = i; 392 qsort_r(ri->sorted, ri->nr_entries, sz, rel_cmp, ri); 393 return 0; 394 } 395 396 /* 397 * For x86_64, the GNU linker is putting IFUNC information in the relocation 398 * addend. 399 */ 400 static bool addend_may_be_ifunc(GElf_Ehdr *ehdr, struct rel_info *ri) 401 { 402 return ehdr->e_machine == EM_X86_64 && ri->is_rela && 403 GELF_R_TYPE(ri->rela.r_info) == R_X86_64_IRELATIVE; 404 } 405 406 static bool get_ifunc_name(Elf *elf, struct dso *dso, GElf_Ehdr *ehdr, 407 struct rel_info *ri, char *buf, size_t buf_sz) 408 { 409 u64 addr = ri->rela.r_addend; 410 struct symbol *sym; 411 GElf_Phdr phdr; 412 413 if (!addend_may_be_ifunc(ehdr, ri)) 414 return false; 415 416 if (elf_read_program_header(elf, addr, &phdr)) 417 return false; 418 419 addr -= phdr.p_vaddr - phdr.p_offset; 420 421 sym = dso__find_symbol_nocache(dso, addr); 422 423 /* Expecting the address to be an IFUNC or IFUNC alias */ 424 if (!sym || sym->start != addr || (sym->type != STT_GNU_IFUNC && !sym->ifunc_alias)) 425 return false; 426 427 snprintf(buf, buf_sz, "%s@plt", sym->name); 428 429 return true; 430 } 431 432 static void exit_rel(struct rel_info *ri) 433 { 434 zfree(&ri->sorted); 435 } 436 437 static bool get_plt_sizes(struct dso *dso, GElf_Ehdr *ehdr, GElf_Shdr *shdr_plt, 438 u64 *plt_header_size, u64 *plt_entry_size) 439 { 440 switch (ehdr->e_machine) { 441 case EM_ARM: 442 *plt_header_size = 20; 443 *plt_entry_size = 12; 444 return true; 445 case EM_AARCH64: 446 *plt_header_size = 32; 447 *plt_entry_size = 16; 448 return true; 449 case EM_LOONGARCH: 450 *plt_header_size = 32; 451 *plt_entry_size = 16; 452 return true; 453 case EM_SPARC: 454 *plt_header_size = 48; 455 *plt_entry_size = 12; 456 return true; 457 case EM_SPARCV9: 458 *plt_header_size = 128; 459 *plt_entry_size = 32; 460 return true; 461 case EM_386: 462 case EM_X86_64: 463 *plt_entry_size = shdr_plt->sh_entsize; 464 /* Size is 8 or 16, if not, assume alignment indicates size */ 465 if (*plt_entry_size != 8 && *plt_entry_size != 16) 466 *plt_entry_size = shdr_plt->sh_addralign == 8 ? 8 : 16; 467 *plt_header_size = *plt_entry_size; 468 break; 469 default: /* FIXME: s390/alpha/mips/parisc/poperpc/sh/xtensa need to be checked */ 470 *plt_header_size = shdr_plt->sh_entsize; 471 *plt_entry_size = shdr_plt->sh_entsize; 472 break; 473 } 474 if (*plt_entry_size) 475 return true; 476 pr_debug("Missing PLT entry size for %s\n", dso__long_name(dso)); 477 return false; 478 } 479 480 static bool machine_is_x86(GElf_Half e_machine) 481 { 482 return e_machine == EM_386 || e_machine == EM_X86_64; 483 } 484 485 struct rela_dyn { 486 GElf_Addr offset; 487 u32 sym_idx; 488 }; 489 490 struct rela_dyn_info { 491 struct dso *dso; 492 Elf_Data *plt_got_data; 493 u32 nr_entries; 494 struct rela_dyn *sorted; 495 Elf_Data *dynsym_data; 496 Elf_Data *dynstr_data; 497 Elf_Data *rela_dyn_data; 498 }; 499 500 static void exit_rela_dyn(struct rela_dyn_info *di) 501 { 502 zfree(&di->sorted); 503 } 504 505 static int cmp_offset(const void *a, const void *b) 506 { 507 const struct rela_dyn *va = a; 508 const struct rela_dyn *vb = b; 509 510 return va->offset < vb->offset ? -1 : (va->offset > vb->offset ? 1 : 0); 511 } 512 513 static int sort_rela_dyn(struct rela_dyn_info *di) 514 { 515 u32 i, n; 516 517 di->sorted = calloc(di->nr_entries, sizeof(di->sorted[0])); 518 if (!di->sorted) 519 return -1; 520 521 /* Get data for sorting: the offset and symbol index */ 522 for (i = 0, n = 0; i < di->nr_entries; i++) { 523 GElf_Rela rela; 524 u32 sym_idx; 525 526 gelf_getrela(di->rela_dyn_data, i, &rela); 527 sym_idx = GELF_R_SYM(rela.r_info); 528 if (sym_idx) { 529 di->sorted[n].sym_idx = sym_idx; 530 di->sorted[n].offset = rela.r_offset; 531 n += 1; 532 } 533 } 534 535 /* Sort by offset */ 536 di->nr_entries = n; 537 qsort(di->sorted, n, sizeof(di->sorted[0]), cmp_offset); 538 539 return 0; 540 } 541 542 static void get_rela_dyn_info(Elf *elf, GElf_Ehdr *ehdr, struct rela_dyn_info *di, Elf_Scn *scn) 543 { 544 GElf_Shdr rela_dyn_shdr; 545 GElf_Shdr shdr; 546 547 di->plt_got_data = elf_getdata(scn, NULL); 548 549 scn = elf_section_by_name(elf, ehdr, &rela_dyn_shdr, ".rela.dyn", NULL); 550 if (!scn || !rela_dyn_shdr.sh_link || !rela_dyn_shdr.sh_entsize) 551 return; 552 553 di->nr_entries = rela_dyn_shdr.sh_size / rela_dyn_shdr.sh_entsize; 554 di->rela_dyn_data = elf_getdata(scn, NULL); 555 556 scn = elf_getscn(elf, rela_dyn_shdr.sh_link); 557 if (!scn || !gelf_getshdr(scn, &shdr) || !shdr.sh_link) 558 return; 559 560 di->dynsym_data = elf_getdata(scn, NULL); 561 di->dynstr_data = elf_getdata(elf_getscn(elf, shdr.sh_link), NULL); 562 563 if (!di->plt_got_data || !di->dynstr_data || !di->dynsym_data || !di->rela_dyn_data) 564 return; 565 566 /* Sort into offset order */ 567 sort_rela_dyn(di); 568 } 569 570 /* Get instruction displacement from a plt entry for x86_64 */ 571 static u32 get_x86_64_plt_disp(const u8 *p) 572 { 573 u8 endbr64[] = {0xf3, 0x0f, 0x1e, 0xfa}; 574 int n = 0; 575 576 /* Skip endbr64 */ 577 if (!memcmp(p, endbr64, sizeof(endbr64))) 578 n += sizeof(endbr64); 579 /* Skip bnd prefix */ 580 if (p[n] == 0xf2) 581 n += 1; 582 /* jmp with 4-byte displacement */ 583 if (p[n] == 0xff && p[n + 1] == 0x25) { 584 u32 disp; 585 586 n += 2; 587 /* Also add offset from start of entry to end of instruction */ 588 memcpy(&disp, p + n, sizeof(disp)); 589 return n + 4 + le32toh(disp); 590 } 591 return 0; 592 } 593 594 static bool get_plt_got_name(GElf_Shdr *shdr, size_t i, 595 struct rela_dyn_info *di, 596 char *buf, size_t buf_sz) 597 { 598 struct rela_dyn vi, *vr; 599 const char *sym_name; 600 char *demangled; 601 GElf_Sym sym; 602 bool result; 603 u32 disp; 604 605 if (!di->sorted) 606 return false; 607 608 disp = get_x86_64_plt_disp(di->plt_got_data->d_buf + i); 609 if (!disp) 610 return false; 611 612 /* Compute target offset of the .plt.got entry */ 613 vi.offset = shdr->sh_offset + di->plt_got_data->d_off + i + disp; 614 615 /* Find that offset in .rela.dyn (sorted by offset) */ 616 vr = bsearch(&vi, di->sorted, di->nr_entries, sizeof(di->sorted[0]), cmp_offset); 617 if (!vr) 618 return false; 619 620 /* Get the associated symbol */ 621 gelf_getsym(di->dynsym_data, vr->sym_idx, &sym); 622 sym_name = elf_sym__name(&sym, di->dynstr_data); 623 demangled = demangle_sym(di->dso, 0, sym_name); 624 if (demangled != NULL) 625 sym_name = demangled; 626 627 snprintf(buf, buf_sz, "%s@plt", sym_name); 628 629 result = *sym_name; 630 631 free(demangled); 632 633 return result; 634 } 635 636 static int dso__synthesize_plt_got_symbols(struct dso *dso, Elf *elf, 637 GElf_Ehdr *ehdr, 638 char *buf, size_t buf_sz) 639 { 640 struct rela_dyn_info di = { .dso = dso }; 641 struct symbol *sym; 642 GElf_Shdr shdr; 643 Elf_Scn *scn; 644 int err = -1; 645 size_t i; 646 647 scn = elf_section_by_name(elf, ehdr, &shdr, ".plt.got", NULL); 648 if (!scn || !shdr.sh_entsize) 649 return 0; 650 651 if (ehdr->e_machine == EM_X86_64) 652 get_rela_dyn_info(elf, ehdr, &di, scn); 653 654 for (i = 0; i < shdr.sh_size; i += shdr.sh_entsize) { 655 if (!get_plt_got_name(&shdr, i, &di, buf, buf_sz)) 656 snprintf(buf, buf_sz, "offset_%#" PRIx64 "@plt", (u64)shdr.sh_offset + i); 657 sym = symbol__new(shdr.sh_offset + i, shdr.sh_entsize, STB_GLOBAL, STT_FUNC, buf); 658 if (!sym) 659 goto out; 660 symbols__insert(dso__symbols(dso), sym); 661 } 662 err = 0; 663 out: 664 exit_rela_dyn(&di); 665 return err; 666 } 667 668 /* 669 * We need to check if we have a .dynsym, so that we can handle the 670 * .plt, synthesizing its symbols, that aren't on the symtabs (be it 671 * .dynsym or .symtab). 672 * And always look at the original dso, not at debuginfo packages, that 673 * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS). 674 */ 675 int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss) 676 { 677 uint32_t idx; 678 GElf_Sym sym; 679 u64 plt_offset, plt_header_size, plt_entry_size; 680 GElf_Shdr shdr_plt, plt_sec_shdr; 681 struct symbol *f, *plt_sym; 682 GElf_Shdr shdr_rel_plt, shdr_dynsym; 683 Elf_Data *syms, *symstrs; 684 Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym; 685 GElf_Ehdr ehdr; 686 char sympltname[1024]; 687 Elf *elf; 688 int nr = 0, err = -1; 689 struct rel_info ri = { .is_rela = false }; 690 bool lazy_plt; 691 692 elf = ss->elf; 693 ehdr = ss->ehdr; 694 695 if (!elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL)) 696 return 0; 697 698 /* 699 * A symbol from a previous section (e.g. .init) can have been expanded 700 * by symbols__fixup_end() to overlap .plt. Truncate it before adding 701 * a symbol for .plt header. 702 */ 703 f = dso__find_symbol_nocache(dso, shdr_plt.sh_offset); 704 if (f && f->start < shdr_plt.sh_offset && f->end > shdr_plt.sh_offset) 705 f->end = shdr_plt.sh_offset; 706 707 if (!get_plt_sizes(dso, &ehdr, &shdr_plt, &plt_header_size, &plt_entry_size)) 708 return 0; 709 710 /* Add a symbol for .plt header */ 711 plt_sym = symbol__new(shdr_plt.sh_offset, plt_header_size, STB_GLOBAL, STT_FUNC, ".plt"); 712 if (!plt_sym) 713 goto out_elf_end; 714 symbols__insert(dso__symbols(dso), plt_sym); 715 716 /* Only x86 has .plt.got */ 717 if (machine_is_x86(ehdr.e_machine) && 718 dso__synthesize_plt_got_symbols(dso, elf, &ehdr, sympltname, sizeof(sympltname))) 719 goto out_elf_end; 720 721 /* Only x86 has .plt.sec */ 722 if (machine_is_x86(ehdr.e_machine) && 723 elf_section_by_name(elf, &ehdr, &plt_sec_shdr, ".plt.sec", NULL)) { 724 if (!get_plt_sizes(dso, &ehdr, &plt_sec_shdr, &plt_header_size, &plt_entry_size)) 725 return 0; 726 /* Extend .plt symbol to entire .plt */ 727 plt_sym->end = plt_sym->start + shdr_plt.sh_size; 728 /* Use .plt.sec offset */ 729 plt_offset = plt_sec_shdr.sh_offset; 730 lazy_plt = false; 731 } else { 732 plt_offset = shdr_plt.sh_offset; 733 lazy_plt = true; 734 } 735 736 scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt, 737 ".rela.plt", NULL); 738 if (scn_plt_rel == NULL) { 739 scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt, 740 ".rel.plt", NULL); 741 if (scn_plt_rel == NULL) 742 return 0; 743 } 744 745 if (shdr_rel_plt.sh_type != SHT_RELA && 746 shdr_rel_plt.sh_type != SHT_REL) 747 return 0; 748 749 if (!shdr_rel_plt.sh_link) 750 return 0; 751 752 if (shdr_rel_plt.sh_link == ss->dynsym_idx) { 753 scn_dynsym = ss->dynsym; 754 shdr_dynsym = ss->dynshdr; 755 } else if (shdr_rel_plt.sh_link == ss->symtab_idx) { 756 /* 757 * A static executable can have a .plt due to IFUNCs, in which 758 * case .symtab is used not .dynsym. 759 */ 760 scn_dynsym = ss->symtab; 761 shdr_dynsym = ss->symshdr; 762 } else { 763 goto out_elf_end; 764 } 765 766 if (!scn_dynsym) 767 return 0; 768 769 /* 770 * Fetch the relocation section to find the idxes to the GOT 771 * and the symbols in the .dynsym they refer to. 772 */ 773 ri.reldata = elf_getdata(scn_plt_rel, NULL); 774 if (!ri.reldata) 775 goto out_elf_end; 776 777 syms = elf_getdata(scn_dynsym, NULL); 778 if (syms == NULL) 779 goto out_elf_end; 780 781 scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link); 782 if (scn_symstrs == NULL) 783 goto out_elf_end; 784 785 symstrs = elf_getdata(scn_symstrs, NULL); 786 if (symstrs == NULL) 787 goto out_elf_end; 788 789 if (symstrs->d_size == 0) 790 goto out_elf_end; 791 792 ri.nr_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize; 793 794 ri.is_rela = shdr_rel_plt.sh_type == SHT_RELA; 795 796 if (lazy_plt) { 797 /* 798 * Assume a .plt with the same number of entries as the number 799 * of relocation entries is not lazy and does not have a header. 800 */ 801 if (ri.nr_entries * plt_entry_size == shdr_plt.sh_size) 802 dso__delete_symbol(dso, plt_sym); 803 else 804 plt_offset += plt_header_size; 805 } 806 807 /* 808 * x86 doesn't insert IFUNC relocations in .plt order, so sort to get 809 * back in order. 810 */ 811 if (machine_is_x86(ehdr.e_machine) && sort_rel(&ri)) 812 goto out_elf_end; 813 814 for (idx = 0; idx < ri.nr_entries; idx++) { 815 const char *elf_name = NULL; 816 char *demangled = NULL; 817 818 gelf_getsym(syms, get_rel_symidx(&ri, idx), &sym); 819 820 elf_name = elf_sym__name(&sym, symstrs); 821 demangled = demangle_sym(dso, 0, elf_name); 822 if (demangled) 823 elf_name = demangled; 824 if (*elf_name) 825 snprintf(sympltname, sizeof(sympltname), "%s@plt", elf_name); 826 else if (!get_ifunc_name(elf, dso, &ehdr, &ri, sympltname, sizeof(sympltname))) 827 snprintf(sympltname, sizeof(sympltname), 828 "offset_%#" PRIx64 "@plt", plt_offset); 829 free(demangled); 830 831 f = symbol__new(plt_offset, plt_entry_size, STB_GLOBAL, STT_FUNC, sympltname); 832 if (!f) 833 goto out_elf_end; 834 835 plt_offset += plt_entry_size; 836 symbols__insert(dso__symbols(dso), f); 837 ++nr; 838 } 839 840 err = 0; 841 out_elf_end: 842 exit_rel(&ri); 843 if (err == 0) 844 return nr; 845 pr_debug("%s: problems reading %s PLT info.\n", 846 __func__, dso__long_name(dso)); 847 return 0; 848 } 849 850 char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name) 851 { 852 return demangle_sym(dso, kmodule, elf_name); 853 } 854 855 /* 856 * Align offset to 4 bytes as needed for note name and descriptor data. 857 */ 858 #define NOTE_ALIGN(n) (((n) + 3) & -4U) 859 860 static int elf_read_build_id(Elf *elf, void *bf, size_t size) 861 { 862 int err = -1; 863 GElf_Ehdr ehdr; 864 GElf_Shdr shdr; 865 Elf_Data *data; 866 Elf_Scn *sec; 867 Elf_Kind ek; 868 void *ptr; 869 870 if (size < BUILD_ID_SIZE) 871 goto out; 872 873 ek = elf_kind(elf); 874 if (ek != ELF_K_ELF) 875 goto out; 876 877 if (gelf_getehdr(elf, &ehdr) == NULL) { 878 pr_err("%s: cannot get elf header.\n", __func__); 879 goto out; 880 } 881 882 /* 883 * Check following sections for notes: 884 * '.note.gnu.build-id' 885 * '.notes' 886 * '.note' (VDSO specific) 887 */ 888 do { 889 sec = elf_section_by_name(elf, &ehdr, &shdr, 890 ".note.gnu.build-id", NULL); 891 if (sec) 892 break; 893 894 sec = elf_section_by_name(elf, &ehdr, &shdr, 895 ".notes", NULL); 896 if (sec) 897 break; 898 899 sec = elf_section_by_name(elf, &ehdr, &shdr, 900 ".note", NULL); 901 if (sec) 902 break; 903 904 return err; 905 906 } while (0); 907 908 data = elf_getdata(sec, NULL); 909 if (data == NULL) 910 goto out; 911 912 ptr = data->d_buf; 913 while (ptr < (data->d_buf + data->d_size)) { 914 GElf_Nhdr *nhdr = ptr; 915 size_t namesz = NOTE_ALIGN(nhdr->n_namesz), 916 descsz = NOTE_ALIGN(nhdr->n_descsz); 917 const char *name; 918 919 ptr += sizeof(*nhdr); 920 name = ptr; 921 ptr += namesz; 922 if (nhdr->n_type == NT_GNU_BUILD_ID && 923 nhdr->n_namesz == sizeof("GNU")) { 924 if (memcmp(name, "GNU", sizeof("GNU")) == 0) { 925 size_t sz = min(size, descsz); 926 memcpy(bf, ptr, sz); 927 memset(bf + sz, 0, size - sz); 928 err = sz; 929 break; 930 } 931 } 932 ptr += descsz; 933 } 934 935 out: 936 return err; 937 } 938 939 #ifdef HAVE_LIBBFD_BUILDID_SUPPORT 940 941 static int read_build_id(const char *filename, struct build_id *bid) 942 { 943 size_t size = sizeof(bid->data); 944 int err = -1; 945 bfd *abfd; 946 947 abfd = bfd_openr(filename, NULL); 948 if (!abfd) 949 return -1; 950 951 if (!bfd_check_format(abfd, bfd_object)) { 952 pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename); 953 goto out_close; 954 } 955 956 if (!abfd->build_id || abfd->build_id->size > size) 957 goto out_close; 958 959 memcpy(bid->data, abfd->build_id->data, abfd->build_id->size); 960 memset(bid->data + abfd->build_id->size, 0, size - abfd->build_id->size); 961 err = bid->size = abfd->build_id->size; 962 963 out_close: 964 bfd_close(abfd); 965 return err; 966 } 967 968 #else // HAVE_LIBBFD_BUILDID_SUPPORT 969 970 static int read_build_id(const char *filename, struct build_id *bid) 971 { 972 size_t size = sizeof(bid->data); 973 int fd, err = -1; 974 Elf *elf; 975 976 if (size < BUILD_ID_SIZE) 977 goto out; 978 979 fd = open(filename, O_RDONLY); 980 if (fd < 0) 981 goto out; 982 983 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL); 984 if (elf == NULL) { 985 pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename); 986 goto out_close; 987 } 988 989 err = elf_read_build_id(elf, bid->data, size); 990 if (err > 0) 991 bid->size = err; 992 993 elf_end(elf); 994 out_close: 995 close(fd); 996 out: 997 return err; 998 } 999 1000 #endif // HAVE_LIBBFD_BUILDID_SUPPORT 1001 1002 int filename__read_build_id(const char *filename, struct build_id *bid) 1003 { 1004 struct kmod_path m = { .name = NULL, }; 1005 char path[PATH_MAX]; 1006 int err; 1007 1008 if (!filename) 1009 return -EFAULT; 1010 1011 err = kmod_path__parse(&m, filename); 1012 if (err) 1013 return -1; 1014 1015 if (m.comp) { 1016 int error = 0, fd; 1017 1018 fd = filename__decompress(filename, path, sizeof(path), m.comp, &error); 1019 if (fd < 0) { 1020 pr_debug("Failed to decompress (error %d) %s\n", 1021 error, filename); 1022 return -1; 1023 } 1024 close(fd); 1025 filename = path; 1026 } 1027 1028 err = read_build_id(filename, bid); 1029 1030 if (m.comp) 1031 unlink(filename); 1032 return err; 1033 } 1034 1035 int sysfs__read_build_id(const char *filename, struct build_id *bid) 1036 { 1037 size_t size = sizeof(bid->data); 1038 int fd, err = -1; 1039 1040 fd = open(filename, O_RDONLY); 1041 if (fd < 0) 1042 goto out; 1043 1044 while (1) { 1045 char bf[BUFSIZ]; 1046 GElf_Nhdr nhdr; 1047 size_t namesz, descsz; 1048 1049 if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr)) 1050 break; 1051 1052 namesz = NOTE_ALIGN(nhdr.n_namesz); 1053 descsz = NOTE_ALIGN(nhdr.n_descsz); 1054 if (nhdr.n_type == NT_GNU_BUILD_ID && 1055 nhdr.n_namesz == sizeof("GNU")) { 1056 if (read(fd, bf, namesz) != (ssize_t)namesz) 1057 break; 1058 if (memcmp(bf, "GNU", sizeof("GNU")) == 0) { 1059 size_t sz = min(descsz, size); 1060 if (read(fd, bid->data, sz) == (ssize_t)sz) { 1061 memset(bid->data + sz, 0, size - sz); 1062 bid->size = sz; 1063 err = 0; 1064 break; 1065 } 1066 } else if (read(fd, bf, descsz) != (ssize_t)descsz) 1067 break; 1068 } else { 1069 int n = namesz + descsz; 1070 1071 if (n > (int)sizeof(bf)) { 1072 n = sizeof(bf); 1073 pr_debug("%s: truncating reading of build id in sysfs file %s: n_namesz=%u, n_descsz=%u.\n", 1074 __func__, filename, nhdr.n_namesz, nhdr.n_descsz); 1075 } 1076 if (read(fd, bf, n) != n) 1077 break; 1078 } 1079 } 1080 close(fd); 1081 out: 1082 return err; 1083 } 1084 1085 #ifdef HAVE_LIBBFD_SUPPORT 1086 1087 int filename__read_debuglink(const char *filename, char *debuglink, 1088 size_t size) 1089 { 1090 int err = -1; 1091 asection *section; 1092 bfd *abfd; 1093 1094 abfd = bfd_openr(filename, NULL); 1095 if (!abfd) 1096 return -1; 1097 1098 if (!bfd_check_format(abfd, bfd_object)) { 1099 pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename); 1100 goto out_close; 1101 } 1102 1103 section = bfd_get_section_by_name(abfd, ".gnu_debuglink"); 1104 if (!section) 1105 goto out_close; 1106 1107 if (section->size > size) 1108 goto out_close; 1109 1110 if (!bfd_get_section_contents(abfd, section, debuglink, 0, 1111 section->size)) 1112 goto out_close; 1113 1114 err = 0; 1115 1116 out_close: 1117 bfd_close(abfd); 1118 return err; 1119 } 1120 1121 #else 1122 1123 int filename__read_debuglink(const char *filename, char *debuglink, 1124 size_t size) 1125 { 1126 int fd, err = -1; 1127 Elf *elf; 1128 GElf_Ehdr ehdr; 1129 GElf_Shdr shdr; 1130 Elf_Data *data; 1131 Elf_Scn *sec; 1132 Elf_Kind ek; 1133 1134 fd = open(filename, O_RDONLY); 1135 if (fd < 0) 1136 goto out; 1137 1138 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL); 1139 if (elf == NULL) { 1140 pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename); 1141 goto out_close; 1142 } 1143 1144 ek = elf_kind(elf); 1145 if (ek != ELF_K_ELF) 1146 goto out_elf_end; 1147 1148 if (gelf_getehdr(elf, &ehdr) == NULL) { 1149 pr_err("%s: cannot get elf header.\n", __func__); 1150 goto out_elf_end; 1151 } 1152 1153 sec = elf_section_by_name(elf, &ehdr, &shdr, 1154 ".gnu_debuglink", NULL); 1155 if (sec == NULL) 1156 goto out_elf_end; 1157 1158 data = elf_getdata(sec, NULL); 1159 if (data == NULL) 1160 goto out_elf_end; 1161 1162 /* the start of this section is a zero-terminated string */ 1163 strncpy(debuglink, data->d_buf, size); 1164 1165 err = 0; 1166 1167 out_elf_end: 1168 elf_end(elf); 1169 out_close: 1170 close(fd); 1171 out: 1172 return err; 1173 } 1174 1175 #endif 1176 1177 static int dso__swap_init(struct dso *dso, unsigned char eidata) 1178 { 1179 static unsigned int const endian = 1; 1180 1181 dso__set_needs_swap(dso, DSO_SWAP__NO); 1182 1183 switch (eidata) { 1184 case ELFDATA2LSB: 1185 /* We are big endian, DSO is little endian. */ 1186 if (*(unsigned char const *)&endian != 1) 1187 dso__set_needs_swap(dso, DSO_SWAP__YES); 1188 break; 1189 1190 case ELFDATA2MSB: 1191 /* We are little endian, DSO is big endian. */ 1192 if (*(unsigned char const *)&endian != 0) 1193 dso__set_needs_swap(dso, DSO_SWAP__YES); 1194 break; 1195 1196 default: 1197 pr_err("unrecognized DSO data encoding %d\n", eidata); 1198 return -EINVAL; 1199 } 1200 1201 return 0; 1202 } 1203 1204 bool symsrc__possibly_runtime(struct symsrc *ss) 1205 { 1206 return ss->dynsym || ss->opdsec; 1207 } 1208 1209 bool symsrc__has_symtab(struct symsrc *ss) 1210 { 1211 return ss->symtab != NULL; 1212 } 1213 1214 void symsrc__destroy(struct symsrc *ss) 1215 { 1216 zfree(&ss->name); 1217 elf_end(ss->elf); 1218 close(ss->fd); 1219 } 1220 1221 bool elf__needs_adjust_symbols(GElf_Ehdr ehdr) 1222 { 1223 /* 1224 * Usually vmlinux is an ELF file with type ET_EXEC for most 1225 * architectures; except Arm64 kernel is linked with option 1226 * '-share', so need to check type ET_DYN. 1227 */ 1228 return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL || 1229 ehdr.e_type == ET_DYN; 1230 } 1231 1232 static Elf *read_gnu_debugdata(struct dso *dso, Elf *elf, const char *name, int *fd_ret) 1233 { 1234 Elf *elf_embedded; 1235 GElf_Ehdr ehdr; 1236 GElf_Shdr shdr; 1237 Elf_Scn *scn; 1238 Elf_Data *scn_data; 1239 FILE *wrapped; 1240 size_t shndx; 1241 char temp_filename[] = "/tmp/perf.gnu_debugdata.elf.XXXXXX"; 1242 int ret, temp_fd; 1243 1244 if (gelf_getehdr(elf, &ehdr) == NULL) { 1245 pr_debug("%s: cannot read %s ELF file.\n", __func__, name); 1246 *dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF; 1247 return NULL; 1248 } 1249 1250 scn = elf_section_by_name(elf, &ehdr, &shdr, ".gnu_debugdata", &shndx); 1251 if (!scn) { 1252 *dso__load_errno(dso) = -ENOENT; 1253 return NULL; 1254 } 1255 1256 if (shdr.sh_type == SHT_NOBITS) { 1257 pr_debug("%s: .gnu_debugdata of ELF file %s has no data.\n", __func__, name); 1258 *dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF; 1259 return NULL; 1260 } 1261 1262 scn_data = elf_rawdata(scn, NULL); 1263 if (!scn_data) { 1264 pr_debug("%s: error reading .gnu_debugdata of %s: %s\n", __func__, 1265 name, elf_errmsg(-1)); 1266 *dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF; 1267 return NULL; 1268 } 1269 1270 wrapped = fmemopen(scn_data->d_buf, scn_data->d_size, "r"); 1271 if (!wrapped) { 1272 pr_debug("%s: fmemopen: %s\n", __func__, strerror(errno)); 1273 *dso__load_errno(dso) = -errno; 1274 return NULL; 1275 } 1276 1277 temp_fd = mkstemp(temp_filename); 1278 if (temp_fd < 0) { 1279 pr_debug("%s: mkstemp: %s\n", __func__, strerror(errno)); 1280 *dso__load_errno(dso) = -errno; 1281 fclose(wrapped); 1282 return NULL; 1283 } 1284 unlink(temp_filename); 1285 1286 ret = lzma_decompress_stream_to_file(wrapped, temp_fd); 1287 fclose(wrapped); 1288 if (ret < 0) { 1289 *dso__load_errno(dso) = -errno; 1290 close(temp_fd); 1291 return NULL; 1292 } 1293 1294 elf_embedded = elf_begin(temp_fd, PERF_ELF_C_READ_MMAP, NULL); 1295 if (!elf_embedded) { 1296 pr_debug("%s: error reading .gnu_debugdata of %s: %s\n", __func__, 1297 name, elf_errmsg(-1)); 1298 *dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF; 1299 close(temp_fd); 1300 return NULL; 1301 } 1302 pr_debug("%s: using .gnu_debugdata of %s\n", __func__, name); 1303 *fd_ret = temp_fd; 1304 return elf_embedded; 1305 } 1306 1307 int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name, 1308 enum dso_binary_type type) 1309 { 1310 GElf_Ehdr ehdr; 1311 Elf *elf; 1312 int fd; 1313 1314 if (dso__needs_decompress(dso)) { 1315 fd = dso__decompress_kmodule_fd(dso, name); 1316 if (fd < 0) 1317 return -1; 1318 1319 type = dso__symtab_type(dso); 1320 } else { 1321 fd = open(name, O_RDONLY); 1322 if (fd < 0) { 1323 *dso__load_errno(dso) = errno; 1324 return -1; 1325 } 1326 } 1327 1328 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL); 1329 if (elf == NULL) { 1330 pr_debug("%s: cannot read %s ELF file.\n", __func__, name); 1331 *dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF; 1332 goto out_close; 1333 } 1334 1335 if (type == DSO_BINARY_TYPE__GNU_DEBUGDATA) { 1336 int new_fd; 1337 Elf *embedded = read_gnu_debugdata(dso, elf, name, &new_fd); 1338 1339 if (!embedded) 1340 goto out_close; 1341 1342 elf_end(elf); 1343 close(fd); 1344 fd = new_fd; 1345 elf = embedded; 1346 } 1347 1348 if (gelf_getehdr(elf, &ehdr) == NULL) { 1349 *dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF; 1350 pr_debug("%s: cannot get elf header.\n", __func__); 1351 goto out_elf_end; 1352 } 1353 1354 if (dso__swap_init(dso, ehdr.e_ident[EI_DATA])) { 1355 *dso__load_errno(dso) = DSO_LOAD_ERRNO__INTERNAL_ERROR; 1356 goto out_elf_end; 1357 } 1358 1359 /* Always reject images with a mismatched build-id: */ 1360 if (dso__has_build_id(dso) && !symbol_conf.ignore_vmlinux_buildid) { 1361 u8 build_id[BUILD_ID_SIZE]; 1362 struct build_id bid; 1363 int size; 1364 1365 size = elf_read_build_id(elf, build_id, BUILD_ID_SIZE); 1366 if (size <= 0) { 1367 *dso__load_errno(dso) = DSO_LOAD_ERRNO__CANNOT_READ_BUILDID; 1368 goto out_elf_end; 1369 } 1370 1371 build_id__init(&bid, build_id, size); 1372 if (!dso__build_id_equal(dso, &bid)) { 1373 pr_debug("%s: build id mismatch for %s.\n", __func__, name); 1374 *dso__load_errno(dso) = DSO_LOAD_ERRNO__MISMATCHING_BUILDID; 1375 goto out_elf_end; 1376 } 1377 } 1378 1379 ss->is_64_bit = (gelf_getclass(elf) == ELFCLASS64); 1380 1381 ss->symtab_idx = 0; 1382 ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab", 1383 &ss->symtab_idx); 1384 if (ss->symshdr.sh_type != SHT_SYMTAB) 1385 ss->symtab = NULL; 1386 1387 ss->dynsym_idx = 0; 1388 ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym", 1389 &ss->dynsym_idx); 1390 if (ss->dynshdr.sh_type != SHT_DYNSYM) 1391 ss->dynsym = NULL; 1392 1393 ss->opdidx = 0; 1394 ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd", 1395 &ss->opdidx); 1396 if (ss->opdshdr.sh_type != SHT_PROGBITS) 1397 ss->opdsec = NULL; 1398 1399 if (dso__kernel(dso) == DSO_SPACE__USER) 1400 ss->adjust_symbols = true; 1401 else 1402 ss->adjust_symbols = elf__needs_adjust_symbols(ehdr); 1403 1404 ss->name = strdup(name); 1405 if (!ss->name) { 1406 *dso__load_errno(dso) = errno; 1407 goto out_elf_end; 1408 } 1409 1410 ss->elf = elf; 1411 ss->fd = fd; 1412 ss->ehdr = ehdr; 1413 ss->type = type; 1414 1415 return 0; 1416 1417 out_elf_end: 1418 elf_end(elf); 1419 out_close: 1420 close(fd); 1421 return -1; 1422 } 1423 1424 static bool is_exe_text(int flags) 1425 { 1426 return (flags & (SHF_ALLOC | SHF_EXECINSTR)) == (SHF_ALLOC | SHF_EXECINSTR); 1427 } 1428 1429 /* 1430 * Some executable module sections like .noinstr.text might be laid out with 1431 * .text so they can use the same mapping (memory address to file offset). 1432 * Check if that is the case. Refer to kernel layout_sections(). Return the 1433 * maximum offset. 1434 */ 1435 static u64 max_text_section(Elf *elf, GElf_Ehdr *ehdr) 1436 { 1437 Elf_Scn *sec = NULL; 1438 GElf_Shdr shdr; 1439 u64 offs = 0; 1440 1441 /* Doesn't work for some arch */ 1442 if (ehdr->e_machine == EM_PARISC || 1443 ehdr->e_machine == EM_ALPHA) 1444 return 0; 1445 1446 /* ELF is corrupted/truncated, avoid calling elf_strptr. */ 1447 if (!elf_rawdata(elf_getscn(elf, ehdr->e_shstrndx), NULL)) 1448 return 0; 1449 1450 while ((sec = elf_nextscn(elf, sec)) != NULL) { 1451 char *sec_name; 1452 1453 if (!gelf_getshdr(sec, &shdr)) 1454 break; 1455 1456 if (!is_exe_text(shdr.sh_flags)) 1457 continue; 1458 1459 /* .init and .exit sections are not placed with .text */ 1460 sec_name = elf_strptr(elf, ehdr->e_shstrndx, shdr.sh_name); 1461 if (!sec_name || 1462 strstarts(sec_name, ".init") || 1463 strstarts(sec_name, ".exit")) 1464 break; 1465 1466 /* Must be next to previous, assumes .text is first */ 1467 if (offs && PERF_ALIGN(offs, shdr.sh_addralign ?: 1) != shdr.sh_offset) 1468 break; 1469 1470 offs = shdr.sh_offset + shdr.sh_size; 1471 } 1472 1473 return offs; 1474 } 1475 1476 /** 1477 * ref_reloc_sym_not_found - has kernel relocation symbol been found. 1478 * @kmap: kernel maps and relocation reference symbol 1479 * 1480 * This function returns %true if we are dealing with the kernel maps and the 1481 * relocation reference symbol has not yet been found. Otherwise %false is 1482 * returned. 1483 */ 1484 static bool ref_reloc_sym_not_found(struct kmap *kmap) 1485 { 1486 return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name && 1487 !kmap->ref_reloc_sym->unrelocated_addr; 1488 } 1489 1490 /** 1491 * ref_reloc - kernel relocation offset. 1492 * @kmap: kernel maps and relocation reference symbol 1493 * 1494 * This function returns the offset of kernel addresses as determined by using 1495 * the relocation reference symbol i.e. if the kernel has not been relocated 1496 * then the return value is zero. 1497 */ 1498 static u64 ref_reloc(struct kmap *kmap) 1499 { 1500 if (kmap && kmap->ref_reloc_sym && 1501 kmap->ref_reloc_sym->unrelocated_addr) 1502 return kmap->ref_reloc_sym->addr - 1503 kmap->ref_reloc_sym->unrelocated_addr; 1504 return 0; 1505 } 1506 1507 void __weak arch__sym_update(struct symbol *s __maybe_unused, 1508 GElf_Sym *sym __maybe_unused) { } 1509 1510 static int dso__process_kernel_symbol(struct dso *dso, struct map *map, 1511 GElf_Sym *sym, GElf_Shdr *shdr, 1512 struct maps *kmaps, struct kmap *kmap, 1513 struct dso **curr_dsop, 1514 const char *section_name, 1515 bool adjust_kernel_syms, bool kmodule, bool *remap_kernel, 1516 u64 max_text_sh_offset) 1517 { 1518 struct dso *curr_dso = *curr_dsop; 1519 struct map *curr_map; 1520 char dso_name[PATH_MAX]; 1521 1522 /* Adjust symbol to map to file offset */ 1523 if (adjust_kernel_syms) 1524 sym->st_value -= shdr->sh_addr - shdr->sh_offset; 1525 1526 if (strcmp(section_name, (dso__short_name(curr_dso) + dso__short_name_len(dso))) == 0) 1527 return 0; 1528 1529 if (strcmp(section_name, ".text") == 0) { 1530 /* 1531 * The initial kernel mapping is based on 1532 * kallsyms and identity maps. Overwrite it to 1533 * map to the kernel dso. 1534 */ 1535 if (*remap_kernel && dso__kernel(dso) && !kmodule) { 1536 *remap_kernel = false; 1537 map__set_start(map, shdr->sh_addr + ref_reloc(kmap)); 1538 map__set_end(map, map__start(map) + shdr->sh_size); 1539 map__set_pgoff(map, shdr->sh_offset); 1540 map__set_mapping_type(map, MAPPING_TYPE__DSO); 1541 /* Ensure maps are correctly ordered */ 1542 if (kmaps) { 1543 int err; 1544 struct map *tmp = map__get(map); 1545 1546 maps__remove(kmaps, map); 1547 err = maps__insert(kmaps, map); 1548 map__put(tmp); 1549 if (err) 1550 return err; 1551 } 1552 } 1553 1554 /* 1555 * The initial module mapping is based on 1556 * /proc/modules mapped to offset zero. 1557 * Overwrite it to map to the module dso. 1558 */ 1559 if (*remap_kernel && kmodule) { 1560 *remap_kernel = false; 1561 map__set_pgoff(map, shdr->sh_offset); 1562 } 1563 1564 dso__put(*curr_dsop); 1565 *curr_dsop = dso__get(dso); 1566 return 0; 1567 } 1568 1569 if (!kmap) 1570 return 0; 1571 1572 /* 1573 * perf does not record module section addresses except for .text, but 1574 * some sections can use the same mapping as .text. 1575 */ 1576 if (kmodule && adjust_kernel_syms && is_exe_text(shdr->sh_flags) && 1577 shdr->sh_offset <= max_text_sh_offset) { 1578 dso__put(*curr_dsop); 1579 *curr_dsop = dso__get(dso); 1580 return 0; 1581 } 1582 1583 snprintf(dso_name, sizeof(dso_name), "%s%s", dso__short_name(dso), section_name); 1584 1585 curr_map = maps__find_by_name(kmaps, dso_name); 1586 if (curr_map == NULL) { 1587 u64 start = sym->st_value; 1588 1589 if (kmodule) 1590 start += map__start(map) + shdr->sh_offset; 1591 1592 curr_dso = dso__new(dso_name); 1593 if (curr_dso == NULL) 1594 return -1; 1595 dso__set_kernel(curr_dso, dso__kernel(dso)); 1596 RC_CHK_ACCESS(curr_dso)->long_name = dso__long_name(dso); 1597 RC_CHK_ACCESS(curr_dso)->long_name_len = dso__long_name_len(dso); 1598 dso__set_binary_type(curr_dso, dso__binary_type(dso)); 1599 dso__set_adjust_symbols(curr_dso, dso__adjust_symbols(dso)); 1600 curr_map = map__new2(start, curr_dso); 1601 if (curr_map == NULL) { 1602 dso__put(curr_dso); 1603 return -1; 1604 } 1605 if (dso__kernel(curr_dso)) 1606 map__kmap(curr_map)->kmaps = kmaps; 1607 1608 if (adjust_kernel_syms) { 1609 map__set_start(curr_map, shdr->sh_addr + ref_reloc(kmap)); 1610 map__set_end(curr_map, map__start(curr_map) + shdr->sh_size); 1611 map__set_pgoff(curr_map, shdr->sh_offset); 1612 } else { 1613 map__set_mapping_type(curr_map, MAPPING_TYPE__IDENTITY); 1614 } 1615 dso__set_symtab_type(curr_dso, dso__symtab_type(dso)); 1616 if (maps__insert(kmaps, curr_map)) 1617 return -1; 1618 dsos__add(&maps__machine(kmaps)->dsos, curr_dso); 1619 dso__set_loaded(curr_dso); 1620 dso__put(*curr_dsop); 1621 *curr_dsop = curr_dso; 1622 } else { 1623 dso__put(*curr_dsop); 1624 *curr_dsop = dso__get(map__dso(curr_map)); 1625 } 1626 map__put(curr_map); 1627 1628 return 0; 1629 } 1630 1631 static int 1632 dso__load_sym_internal(struct dso *dso, struct map *map, struct symsrc *syms_ss, 1633 struct symsrc *runtime_ss, int kmodule, int dynsym) 1634 { 1635 struct kmap *kmap = dso__kernel(dso) ? map__kmap(map) : NULL; 1636 struct maps *kmaps = kmap ? map__kmaps(map) : NULL; 1637 struct dso *curr_dso = NULL; 1638 Elf_Data *symstrs, *secstrs, *secstrs_run, *secstrs_sym; 1639 uint32_t nr_syms; 1640 uint32_t idx; 1641 GElf_Ehdr ehdr; 1642 GElf_Shdr shdr; 1643 GElf_Shdr tshdr; 1644 Elf_Data *syms, *opddata = NULL; 1645 GElf_Sym sym; 1646 Elf_Scn *sec, *sec_strndx; 1647 Elf *elf; 1648 int nr = 0; 1649 bool remap_kernel = false, adjust_kernel_syms = false; 1650 u64 max_text_sh_offset = 0; 1651 1652 if (kmap && !kmaps) 1653 return -1; 1654 1655 elf = syms_ss->elf; 1656 ehdr = syms_ss->ehdr; 1657 if (dynsym) { 1658 sec = syms_ss->dynsym; 1659 shdr = syms_ss->dynshdr; 1660 } else { 1661 sec = syms_ss->symtab; 1662 shdr = syms_ss->symshdr; 1663 } 1664 1665 if (elf_section_by_name(runtime_ss->elf, &runtime_ss->ehdr, &tshdr, 1666 ".text", NULL)) { 1667 dso__set_text_offset(dso, tshdr.sh_addr - tshdr.sh_offset); 1668 dso__set_text_end(dso, tshdr.sh_offset + tshdr.sh_size); 1669 } 1670 1671 if (runtime_ss->opdsec) 1672 opddata = elf_rawdata(runtime_ss->opdsec, NULL); 1673 1674 syms = elf_getdata(sec, NULL); 1675 if (syms == NULL) 1676 goto out_elf_end; 1677 1678 sec = elf_getscn(elf, shdr.sh_link); 1679 if (sec == NULL) 1680 goto out_elf_end; 1681 1682 symstrs = elf_getdata(sec, NULL); 1683 if (symstrs == NULL) 1684 goto out_elf_end; 1685 1686 sec_strndx = elf_getscn(runtime_ss->elf, runtime_ss->ehdr.e_shstrndx); 1687 if (sec_strndx == NULL) 1688 goto out_elf_end; 1689 1690 secstrs_run = elf_getdata(sec_strndx, NULL); 1691 if (secstrs_run == NULL) 1692 goto out_elf_end; 1693 1694 sec_strndx = elf_getscn(elf, ehdr.e_shstrndx); 1695 if (sec_strndx == NULL) 1696 goto out_elf_end; 1697 1698 secstrs_sym = elf_getdata(sec_strndx, NULL); 1699 if (secstrs_sym == NULL) 1700 goto out_elf_end; 1701 1702 nr_syms = shdr.sh_size / shdr.sh_entsize; 1703 1704 memset(&sym, 0, sizeof(sym)); 1705 1706 /* 1707 * The kernel relocation symbol is needed in advance in order to adjust 1708 * kernel maps correctly. 1709 */ 1710 if (ref_reloc_sym_not_found(kmap)) { 1711 elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) { 1712 const char *elf_name = elf_sym__name(&sym, symstrs); 1713 1714 if (strcmp(elf_name, kmap->ref_reloc_sym->name)) 1715 continue; 1716 kmap->ref_reloc_sym->unrelocated_addr = sym.st_value; 1717 map__set_reloc(map, kmap->ref_reloc_sym->addr - kmap->ref_reloc_sym->unrelocated_addr); 1718 break; 1719 } 1720 } 1721 1722 /* 1723 * Handle any relocation of vdso necessary because older kernels 1724 * attempted to prelink vdso to its virtual address. 1725 */ 1726 if (dso__is_vdso(dso)) 1727 map__set_reloc(map, map__start(map) - dso__text_offset(dso)); 1728 1729 dso__set_adjust_symbols(dso, runtime_ss->adjust_symbols || ref_reloc(kmap)); 1730 /* 1731 * Initial kernel and module mappings do not map to the dso. 1732 * Flag the fixups. 1733 */ 1734 if (dso__kernel(dso)) { 1735 remap_kernel = true; 1736 adjust_kernel_syms = dso__adjust_symbols(dso); 1737 } 1738 1739 if (kmodule && adjust_kernel_syms) 1740 max_text_sh_offset = max_text_section(runtime_ss->elf, &runtime_ss->ehdr); 1741 1742 curr_dso = dso__get(dso); 1743 elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) { 1744 struct symbol *f; 1745 const char *elf_name = elf_sym__name(&sym, symstrs); 1746 char *demangled = NULL; 1747 int is_label = elf_sym__is_label(&sym); 1748 const char *section_name; 1749 bool used_opd = false; 1750 1751 if (!is_label && !elf_sym__filter(&sym)) 1752 continue; 1753 1754 /* Reject ARM ELF "mapping symbols": these aren't unique and 1755 * don't identify functions, so will confuse the profile 1756 * output: */ 1757 if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) { 1758 if (elf_name[0] == '$' && strchr("adtx", elf_name[1]) 1759 && (elf_name[2] == '\0' || elf_name[2] == '.')) 1760 continue; 1761 } 1762 1763 if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) { 1764 u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr; 1765 u64 *opd = opddata->d_buf + offset; 1766 sym.st_value = DSO__SWAP(dso, u64, *opd); 1767 sym.st_shndx = elf_addr_to_index(runtime_ss->elf, 1768 sym.st_value); 1769 used_opd = true; 1770 } 1771 1772 /* 1773 * When loading symbols in a data mapping, ABS symbols (which 1774 * has a value of SHN_ABS in its st_shndx) failed at 1775 * elf_getscn(). And it marks the loading as a failure so 1776 * already loaded symbols cannot be fixed up. 1777 * 1778 * I'm not sure what should be done. Just ignore them for now. 1779 * - Namhyung Kim 1780 */ 1781 if (sym.st_shndx == SHN_ABS) 1782 continue; 1783 1784 sec = elf_getscn(syms_ss->elf, sym.st_shndx); 1785 if (!sec) 1786 goto out_elf_end; 1787 1788 gelf_getshdr(sec, &shdr); 1789 1790 /* 1791 * If the attribute bit SHF_ALLOC is not set, the section 1792 * doesn't occupy memory during process execution. 1793 * E.g. ".gnu.warning.*" section is used by linker to generate 1794 * warnings when calling deprecated functions, the symbols in 1795 * the section aren't loaded to memory during process execution, 1796 * so skip them. 1797 */ 1798 if (!(shdr.sh_flags & SHF_ALLOC)) 1799 continue; 1800 1801 secstrs = secstrs_sym; 1802 1803 /* 1804 * We have to fallback to runtime when syms' section header has 1805 * NOBITS set. NOBITS results in file offset (sh_offset) not 1806 * being incremented. So sh_offset used below has different 1807 * values for syms (invalid) and runtime (valid). 1808 */ 1809 if (shdr.sh_type == SHT_NOBITS) { 1810 sec = elf_getscn(runtime_ss->elf, sym.st_shndx); 1811 if (!sec) 1812 goto out_elf_end; 1813 1814 gelf_getshdr(sec, &shdr); 1815 secstrs = secstrs_run; 1816 } 1817 1818 if (is_label && !elf_sec__filter(&shdr, secstrs)) 1819 continue; 1820 1821 section_name = elf_sec__name(&shdr, secstrs); 1822 1823 /* On ARM, symbols for thumb functions have 1 added to 1824 * the symbol address as a flag - remove it */ 1825 if ((ehdr.e_machine == EM_ARM) && 1826 (GELF_ST_TYPE(sym.st_info) == STT_FUNC) && 1827 (sym.st_value & 1)) 1828 --sym.st_value; 1829 1830 if (dso__kernel(dso)) { 1831 if (dso__process_kernel_symbol(dso, map, &sym, &shdr, 1832 kmaps, kmap, &curr_dso, 1833 section_name, 1834 adjust_kernel_syms, 1835 kmodule, 1836 &remap_kernel, 1837 max_text_sh_offset)) 1838 goto out_elf_end; 1839 } else if ((used_opd && runtime_ss->adjust_symbols) || 1840 (!used_opd && syms_ss->adjust_symbols)) { 1841 GElf_Phdr phdr; 1842 1843 if (elf_read_program_header(runtime_ss->elf, 1844 (u64)sym.st_value, &phdr)) { 1845 pr_debug4("%s: failed to find program header for " 1846 "symbol: %s st_value: %#" PRIx64 "\n", 1847 __func__, elf_name, (u64)sym.st_value); 1848 pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " " 1849 "sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n", 1850 __func__, (u64)sym.st_value, (u64)shdr.sh_addr, 1851 (u64)shdr.sh_offset); 1852 /* 1853 * Fail to find program header, let's rollback 1854 * to use shdr.sh_addr and shdr.sh_offset to 1855 * calibrate symbol's file address, though this 1856 * is not necessary for normal C ELF file, we 1857 * still need to handle java JIT symbols in this 1858 * case. 1859 */ 1860 sym.st_value -= shdr.sh_addr - shdr.sh_offset; 1861 } else { 1862 pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " " 1863 "p_vaddr: %#" PRIx64 " p_offset: %#" PRIx64 "\n", 1864 __func__, (u64)sym.st_value, (u64)phdr.p_vaddr, 1865 (u64)phdr.p_offset); 1866 sym.st_value -= phdr.p_vaddr - phdr.p_offset; 1867 } 1868 } 1869 1870 demangled = demangle_sym(dso, kmodule, elf_name); 1871 if (demangled != NULL) 1872 elf_name = demangled; 1873 1874 f = symbol__new(sym.st_value, sym.st_size, 1875 GELF_ST_BIND(sym.st_info), 1876 GELF_ST_TYPE(sym.st_info), elf_name); 1877 free(demangled); 1878 if (!f) 1879 goto out_elf_end; 1880 1881 arch__sym_update(f, &sym); 1882 1883 __symbols__insert(dso__symbols(curr_dso), f, dso__kernel(dso)); 1884 nr++; 1885 } 1886 dso__put(curr_dso); 1887 1888 /* 1889 * For misannotated, zeroed, ASM function sizes. 1890 */ 1891 if (nr > 0) { 1892 symbols__fixup_end(dso__symbols(dso), false); 1893 symbols__fixup_duplicate(dso__symbols(dso)); 1894 if (kmap) { 1895 /* 1896 * We need to fixup this here too because we create new 1897 * maps here, for things like vsyscall sections. 1898 */ 1899 maps__fixup_end(kmaps); 1900 } 1901 } 1902 return nr; 1903 out_elf_end: 1904 dso__put(curr_dso); 1905 return -1; 1906 } 1907 1908 int dso__load_sym(struct dso *dso, struct map *map, struct symsrc *syms_ss, 1909 struct symsrc *runtime_ss, int kmodule) 1910 { 1911 int nr = 0; 1912 int err = -1; 1913 1914 dso__set_symtab_type(dso, syms_ss->type); 1915 dso__set_is_64_bit(dso, syms_ss->is_64_bit); 1916 dso__set_rel(dso, syms_ss->ehdr.e_type == ET_REL); 1917 1918 /* 1919 * Modules may already have symbols from kallsyms, but those symbols 1920 * have the wrong values for the dso maps, so remove them. 1921 */ 1922 if (kmodule && syms_ss->symtab) 1923 symbols__delete(dso__symbols(dso)); 1924 1925 if (!syms_ss->symtab) { 1926 /* 1927 * If the vmlinux is stripped, fail so we will fall back 1928 * to using kallsyms. The vmlinux runtime symbols aren't 1929 * of much use. 1930 */ 1931 if (dso__kernel(dso)) 1932 return err; 1933 } else { 1934 err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss, 1935 kmodule, 0); 1936 if (err < 0) 1937 return err; 1938 nr = err; 1939 } 1940 1941 if (syms_ss->dynsym) { 1942 err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss, 1943 kmodule, 1); 1944 if (err < 0) 1945 return err; 1946 nr += err; 1947 } 1948 1949 /* 1950 * The .gnu_debugdata is a special situation: it contains a symbol 1951 * table, but the runtime file may also contain dynsym entries which are 1952 * not present there. We need to load both. 1953 */ 1954 if (syms_ss->type == DSO_BINARY_TYPE__GNU_DEBUGDATA && runtime_ss->dynsym) { 1955 err = dso__load_sym_internal(dso, map, runtime_ss, runtime_ss, 1956 kmodule, 1); 1957 if (err < 0) 1958 return err; 1959 nr += err; 1960 } 1961 1962 return nr; 1963 } 1964 1965 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data) 1966 { 1967 GElf_Phdr phdr; 1968 size_t i, phdrnum; 1969 int err; 1970 u64 sz; 1971 1972 if (elf_getphdrnum(elf, &phdrnum)) 1973 return -1; 1974 1975 for (i = 0; i < phdrnum; i++) { 1976 if (gelf_getphdr(elf, i, &phdr) == NULL) 1977 return -1; 1978 if (phdr.p_type != PT_LOAD) 1979 continue; 1980 if (exe) { 1981 if (!(phdr.p_flags & PF_X)) 1982 continue; 1983 } else { 1984 if (!(phdr.p_flags & PF_R)) 1985 continue; 1986 } 1987 sz = min(phdr.p_memsz, phdr.p_filesz); 1988 if (!sz) 1989 continue; 1990 err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data); 1991 if (err) 1992 return err; 1993 } 1994 return 0; 1995 } 1996 1997 int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data, 1998 bool *is_64_bit) 1999 { 2000 int err; 2001 Elf *elf; 2002 2003 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL); 2004 if (elf == NULL) 2005 return -1; 2006 2007 if (is_64_bit) 2008 *is_64_bit = (gelf_getclass(elf) == ELFCLASS64); 2009 2010 err = elf_read_maps(elf, exe, mapfn, data); 2011 2012 elf_end(elf); 2013 return err; 2014 } 2015 2016 enum dso_type dso__type_fd(int fd) 2017 { 2018 enum dso_type dso_type = DSO__TYPE_UNKNOWN; 2019 GElf_Ehdr ehdr; 2020 Elf_Kind ek; 2021 Elf *elf; 2022 2023 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL); 2024 if (elf == NULL) 2025 goto out; 2026 2027 ek = elf_kind(elf); 2028 if (ek != ELF_K_ELF) 2029 goto out_end; 2030 2031 if (gelf_getclass(elf) == ELFCLASS64) { 2032 dso_type = DSO__TYPE_64BIT; 2033 goto out_end; 2034 } 2035 2036 if (gelf_getehdr(elf, &ehdr) == NULL) 2037 goto out_end; 2038 2039 if (ehdr.e_machine == EM_X86_64) 2040 dso_type = DSO__TYPE_X32BIT; 2041 else 2042 dso_type = DSO__TYPE_32BIT; 2043 out_end: 2044 elf_end(elf); 2045 out: 2046 return dso_type; 2047 } 2048 2049 static int copy_bytes(int from, off_t from_offs, int to, off_t to_offs, u64 len) 2050 { 2051 ssize_t r; 2052 size_t n; 2053 int err = -1; 2054 char *buf = malloc(page_size); 2055 2056 if (buf == NULL) 2057 return -1; 2058 2059 if (lseek(to, to_offs, SEEK_SET) != to_offs) 2060 goto out; 2061 2062 if (lseek(from, from_offs, SEEK_SET) != from_offs) 2063 goto out; 2064 2065 while (len) { 2066 n = page_size; 2067 if (len < n) 2068 n = len; 2069 /* Use read because mmap won't work on proc files */ 2070 r = read(from, buf, n); 2071 if (r < 0) 2072 goto out; 2073 if (!r) 2074 break; 2075 n = r; 2076 r = write(to, buf, n); 2077 if (r < 0) 2078 goto out; 2079 if ((size_t)r != n) 2080 goto out; 2081 len -= n; 2082 } 2083 2084 err = 0; 2085 out: 2086 free(buf); 2087 return err; 2088 } 2089 2090 struct kcore { 2091 int fd; 2092 int elfclass; 2093 Elf *elf; 2094 GElf_Ehdr ehdr; 2095 }; 2096 2097 static int kcore__open(struct kcore *kcore, const char *filename) 2098 { 2099 GElf_Ehdr *ehdr; 2100 2101 kcore->fd = open(filename, O_RDONLY); 2102 if (kcore->fd == -1) 2103 return -1; 2104 2105 kcore->elf = elf_begin(kcore->fd, ELF_C_READ, NULL); 2106 if (!kcore->elf) 2107 goto out_close; 2108 2109 kcore->elfclass = gelf_getclass(kcore->elf); 2110 if (kcore->elfclass == ELFCLASSNONE) 2111 goto out_end; 2112 2113 ehdr = gelf_getehdr(kcore->elf, &kcore->ehdr); 2114 if (!ehdr) 2115 goto out_end; 2116 2117 return 0; 2118 2119 out_end: 2120 elf_end(kcore->elf); 2121 out_close: 2122 close(kcore->fd); 2123 return -1; 2124 } 2125 2126 static int kcore__init(struct kcore *kcore, char *filename, int elfclass, 2127 bool temp) 2128 { 2129 kcore->elfclass = elfclass; 2130 2131 if (temp) 2132 kcore->fd = mkstemp(filename); 2133 else 2134 kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400); 2135 if (kcore->fd == -1) 2136 return -1; 2137 2138 kcore->elf = elf_begin(kcore->fd, ELF_C_WRITE, NULL); 2139 if (!kcore->elf) 2140 goto out_close; 2141 2142 if (!gelf_newehdr(kcore->elf, elfclass)) 2143 goto out_end; 2144 2145 memset(&kcore->ehdr, 0, sizeof(GElf_Ehdr)); 2146 2147 return 0; 2148 2149 out_end: 2150 elf_end(kcore->elf); 2151 out_close: 2152 close(kcore->fd); 2153 unlink(filename); 2154 return -1; 2155 } 2156 2157 static void kcore__close(struct kcore *kcore) 2158 { 2159 elf_end(kcore->elf); 2160 close(kcore->fd); 2161 } 2162 2163 static int kcore__copy_hdr(struct kcore *from, struct kcore *to, size_t count) 2164 { 2165 GElf_Ehdr *ehdr = &to->ehdr; 2166 GElf_Ehdr *kehdr = &from->ehdr; 2167 2168 memcpy(ehdr->e_ident, kehdr->e_ident, EI_NIDENT); 2169 ehdr->e_type = kehdr->e_type; 2170 ehdr->e_machine = kehdr->e_machine; 2171 ehdr->e_version = kehdr->e_version; 2172 ehdr->e_entry = 0; 2173 ehdr->e_shoff = 0; 2174 ehdr->e_flags = kehdr->e_flags; 2175 ehdr->e_phnum = count; 2176 ehdr->e_shentsize = 0; 2177 ehdr->e_shnum = 0; 2178 ehdr->e_shstrndx = 0; 2179 2180 if (from->elfclass == ELFCLASS32) { 2181 ehdr->e_phoff = sizeof(Elf32_Ehdr); 2182 ehdr->e_ehsize = sizeof(Elf32_Ehdr); 2183 ehdr->e_phentsize = sizeof(Elf32_Phdr); 2184 } else { 2185 ehdr->e_phoff = sizeof(Elf64_Ehdr); 2186 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 2187 ehdr->e_phentsize = sizeof(Elf64_Phdr); 2188 } 2189 2190 if (!gelf_update_ehdr(to->elf, ehdr)) 2191 return -1; 2192 2193 if (!gelf_newphdr(to->elf, count)) 2194 return -1; 2195 2196 return 0; 2197 } 2198 2199 static int kcore__add_phdr(struct kcore *kcore, int idx, off_t offset, 2200 u64 addr, u64 len) 2201 { 2202 GElf_Phdr phdr = { 2203 .p_type = PT_LOAD, 2204 .p_flags = PF_R | PF_W | PF_X, 2205 .p_offset = offset, 2206 .p_vaddr = addr, 2207 .p_paddr = 0, 2208 .p_filesz = len, 2209 .p_memsz = len, 2210 .p_align = page_size, 2211 }; 2212 2213 if (!gelf_update_phdr(kcore->elf, idx, &phdr)) 2214 return -1; 2215 2216 return 0; 2217 } 2218 2219 static off_t kcore__write(struct kcore *kcore) 2220 { 2221 return elf_update(kcore->elf, ELF_C_WRITE); 2222 } 2223 2224 struct phdr_data { 2225 off_t offset; 2226 off_t rel; 2227 u64 addr; 2228 u64 len; 2229 struct list_head node; 2230 struct phdr_data *remaps; 2231 }; 2232 2233 struct sym_data { 2234 u64 addr; 2235 struct list_head node; 2236 }; 2237 2238 struct kcore_copy_info { 2239 u64 stext; 2240 u64 etext; 2241 u64 first_symbol; 2242 u64 last_symbol; 2243 u64 first_module; 2244 u64 first_module_symbol; 2245 u64 last_module_symbol; 2246 size_t phnum; 2247 struct list_head phdrs; 2248 struct list_head syms; 2249 }; 2250 2251 #define kcore_copy__for_each_phdr(k, p) \ 2252 list_for_each_entry((p), &(k)->phdrs, node) 2253 2254 static struct phdr_data *phdr_data__new(u64 addr, u64 len, off_t offset) 2255 { 2256 struct phdr_data *p = zalloc(sizeof(*p)); 2257 2258 if (p) { 2259 p->addr = addr; 2260 p->len = len; 2261 p->offset = offset; 2262 } 2263 2264 return p; 2265 } 2266 2267 static struct phdr_data *kcore_copy_info__addnew(struct kcore_copy_info *kci, 2268 u64 addr, u64 len, 2269 off_t offset) 2270 { 2271 struct phdr_data *p = phdr_data__new(addr, len, offset); 2272 2273 if (p) 2274 list_add_tail(&p->node, &kci->phdrs); 2275 2276 return p; 2277 } 2278 2279 static void kcore_copy__free_phdrs(struct kcore_copy_info *kci) 2280 { 2281 struct phdr_data *p, *tmp; 2282 2283 list_for_each_entry_safe(p, tmp, &kci->phdrs, node) { 2284 list_del_init(&p->node); 2285 free(p); 2286 } 2287 } 2288 2289 static struct sym_data *kcore_copy__new_sym(struct kcore_copy_info *kci, 2290 u64 addr) 2291 { 2292 struct sym_data *s = zalloc(sizeof(*s)); 2293 2294 if (s) { 2295 s->addr = addr; 2296 list_add_tail(&s->node, &kci->syms); 2297 } 2298 2299 return s; 2300 } 2301 2302 static void kcore_copy__free_syms(struct kcore_copy_info *kci) 2303 { 2304 struct sym_data *s, *tmp; 2305 2306 list_for_each_entry_safe(s, tmp, &kci->syms, node) { 2307 list_del_init(&s->node); 2308 free(s); 2309 } 2310 } 2311 2312 static int kcore_copy__process_kallsyms(void *arg, const char *name, char type, 2313 u64 start) 2314 { 2315 struct kcore_copy_info *kci = arg; 2316 2317 if (!kallsyms__is_function(type)) 2318 return 0; 2319 2320 if (strchr(name, '[')) { 2321 if (!kci->first_module_symbol || start < kci->first_module_symbol) 2322 kci->first_module_symbol = start; 2323 if (start > kci->last_module_symbol) 2324 kci->last_module_symbol = start; 2325 return 0; 2326 } 2327 2328 if (!kci->first_symbol || start < kci->first_symbol) 2329 kci->first_symbol = start; 2330 2331 if (!kci->last_symbol || start > kci->last_symbol) 2332 kci->last_symbol = start; 2333 2334 if (!strcmp(name, "_stext")) { 2335 kci->stext = start; 2336 return 0; 2337 } 2338 2339 if (!strcmp(name, "_etext")) { 2340 kci->etext = start; 2341 return 0; 2342 } 2343 2344 if (is_entry_trampoline(name) && !kcore_copy__new_sym(kci, start)) 2345 return -1; 2346 2347 return 0; 2348 } 2349 2350 static int kcore_copy__parse_kallsyms(struct kcore_copy_info *kci, 2351 const char *dir) 2352 { 2353 char kallsyms_filename[PATH_MAX]; 2354 2355 scnprintf(kallsyms_filename, PATH_MAX, "%s/kallsyms", dir); 2356 2357 if (symbol__restricted_filename(kallsyms_filename, "/proc/kallsyms")) 2358 return -1; 2359 2360 if (kallsyms__parse(kallsyms_filename, kci, 2361 kcore_copy__process_kallsyms) < 0) 2362 return -1; 2363 2364 return 0; 2365 } 2366 2367 static int kcore_copy__process_modules(void *arg, 2368 const char *name __maybe_unused, 2369 u64 start, u64 size __maybe_unused) 2370 { 2371 struct kcore_copy_info *kci = arg; 2372 2373 if (!kci->first_module || start < kci->first_module) 2374 kci->first_module = start; 2375 2376 return 0; 2377 } 2378 2379 static int kcore_copy__parse_modules(struct kcore_copy_info *kci, 2380 const char *dir) 2381 { 2382 char modules_filename[PATH_MAX]; 2383 2384 scnprintf(modules_filename, PATH_MAX, "%s/modules", dir); 2385 2386 if (symbol__restricted_filename(modules_filename, "/proc/modules")) 2387 return -1; 2388 2389 if (modules__parse(modules_filename, kci, 2390 kcore_copy__process_modules) < 0) 2391 return -1; 2392 2393 return 0; 2394 } 2395 2396 static int kcore_copy__map(struct kcore_copy_info *kci, u64 start, u64 end, 2397 u64 pgoff, u64 s, u64 e) 2398 { 2399 u64 len, offset; 2400 2401 if (s < start || s >= end) 2402 return 0; 2403 2404 offset = (s - start) + pgoff; 2405 len = e < end ? e - s : end - s; 2406 2407 return kcore_copy_info__addnew(kci, s, len, offset) ? 0 : -1; 2408 } 2409 2410 static int kcore_copy__read_map(u64 start, u64 len, u64 pgoff, void *data) 2411 { 2412 struct kcore_copy_info *kci = data; 2413 u64 end = start + len; 2414 struct sym_data *sdat; 2415 2416 if (kcore_copy__map(kci, start, end, pgoff, kci->stext, kci->etext)) 2417 return -1; 2418 2419 if (kcore_copy__map(kci, start, end, pgoff, kci->first_module, 2420 kci->last_module_symbol)) 2421 return -1; 2422 2423 list_for_each_entry(sdat, &kci->syms, node) { 2424 u64 s = round_down(sdat->addr, page_size); 2425 2426 if (kcore_copy__map(kci, start, end, pgoff, s, s + len)) 2427 return -1; 2428 } 2429 2430 return 0; 2431 } 2432 2433 static int kcore_copy__read_maps(struct kcore_copy_info *kci, Elf *elf) 2434 { 2435 if (elf_read_maps(elf, true, kcore_copy__read_map, kci) < 0) 2436 return -1; 2437 2438 return 0; 2439 } 2440 2441 static void kcore_copy__find_remaps(struct kcore_copy_info *kci) 2442 { 2443 struct phdr_data *p, *k = NULL; 2444 u64 kend; 2445 2446 if (!kci->stext) 2447 return; 2448 2449 /* Find phdr that corresponds to the kernel map (contains stext) */ 2450 kcore_copy__for_each_phdr(kci, p) { 2451 u64 pend = p->addr + p->len - 1; 2452 2453 if (p->addr <= kci->stext && pend >= kci->stext) { 2454 k = p; 2455 break; 2456 } 2457 } 2458 2459 if (!k) 2460 return; 2461 2462 kend = k->offset + k->len; 2463 2464 /* Find phdrs that remap the kernel */ 2465 kcore_copy__for_each_phdr(kci, p) { 2466 u64 pend = p->offset + p->len; 2467 2468 if (p == k) 2469 continue; 2470 2471 if (p->offset >= k->offset && pend <= kend) 2472 p->remaps = k; 2473 } 2474 } 2475 2476 static void kcore_copy__layout(struct kcore_copy_info *kci) 2477 { 2478 struct phdr_data *p; 2479 off_t rel = 0; 2480 2481 kcore_copy__find_remaps(kci); 2482 2483 kcore_copy__for_each_phdr(kci, p) { 2484 if (!p->remaps) { 2485 p->rel = rel; 2486 rel += p->len; 2487 } 2488 kci->phnum += 1; 2489 } 2490 2491 kcore_copy__for_each_phdr(kci, p) { 2492 struct phdr_data *k = p->remaps; 2493 2494 if (k) 2495 p->rel = p->offset - k->offset + k->rel; 2496 } 2497 } 2498 2499 static int kcore_copy__calc_maps(struct kcore_copy_info *kci, const char *dir, 2500 Elf *elf) 2501 { 2502 if (kcore_copy__parse_kallsyms(kci, dir)) 2503 return -1; 2504 2505 if (kcore_copy__parse_modules(kci, dir)) 2506 return -1; 2507 2508 if (kci->stext) 2509 kci->stext = round_down(kci->stext, page_size); 2510 else 2511 kci->stext = round_down(kci->first_symbol, page_size); 2512 2513 if (kci->etext) { 2514 kci->etext = round_up(kci->etext, page_size); 2515 } else if (kci->last_symbol) { 2516 kci->etext = round_up(kci->last_symbol, page_size); 2517 kci->etext += page_size; 2518 } 2519 2520 if (kci->first_module_symbol && 2521 (!kci->first_module || kci->first_module_symbol < kci->first_module)) 2522 kci->first_module = kci->first_module_symbol; 2523 2524 kci->first_module = round_down(kci->first_module, page_size); 2525 2526 if (kci->last_module_symbol) { 2527 kci->last_module_symbol = round_up(kci->last_module_symbol, 2528 page_size); 2529 kci->last_module_symbol += page_size; 2530 } 2531 2532 if (!kci->stext || !kci->etext) 2533 return -1; 2534 2535 if (kci->first_module && !kci->last_module_symbol) 2536 return -1; 2537 2538 if (kcore_copy__read_maps(kci, elf)) 2539 return -1; 2540 2541 kcore_copy__layout(kci); 2542 2543 return 0; 2544 } 2545 2546 static int kcore_copy__copy_file(const char *from_dir, const char *to_dir, 2547 const char *name) 2548 { 2549 char from_filename[PATH_MAX]; 2550 char to_filename[PATH_MAX]; 2551 2552 scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name); 2553 scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name); 2554 2555 return copyfile_mode(from_filename, to_filename, 0400); 2556 } 2557 2558 static int kcore_copy__unlink(const char *dir, const char *name) 2559 { 2560 char filename[PATH_MAX]; 2561 2562 scnprintf(filename, PATH_MAX, "%s/%s", dir, name); 2563 2564 return unlink(filename); 2565 } 2566 2567 static int kcore_copy__compare_fds(int from, int to) 2568 { 2569 char *buf_from; 2570 char *buf_to; 2571 ssize_t ret; 2572 size_t len; 2573 int err = -1; 2574 2575 buf_from = malloc(page_size); 2576 buf_to = malloc(page_size); 2577 if (!buf_from || !buf_to) 2578 goto out; 2579 2580 while (1) { 2581 /* Use read because mmap won't work on proc files */ 2582 ret = read(from, buf_from, page_size); 2583 if (ret < 0) 2584 goto out; 2585 2586 if (!ret) 2587 break; 2588 2589 len = ret; 2590 2591 if (readn(to, buf_to, len) != (int)len) 2592 goto out; 2593 2594 if (memcmp(buf_from, buf_to, len)) 2595 goto out; 2596 } 2597 2598 err = 0; 2599 out: 2600 free(buf_to); 2601 free(buf_from); 2602 return err; 2603 } 2604 2605 static int kcore_copy__compare_files(const char *from_filename, 2606 const char *to_filename) 2607 { 2608 int from, to, err = -1; 2609 2610 from = open(from_filename, O_RDONLY); 2611 if (from < 0) 2612 return -1; 2613 2614 to = open(to_filename, O_RDONLY); 2615 if (to < 0) 2616 goto out_close_from; 2617 2618 err = kcore_copy__compare_fds(from, to); 2619 2620 close(to); 2621 out_close_from: 2622 close(from); 2623 return err; 2624 } 2625 2626 static int kcore_copy__compare_file(const char *from_dir, const char *to_dir, 2627 const char *name) 2628 { 2629 char from_filename[PATH_MAX]; 2630 char to_filename[PATH_MAX]; 2631 2632 scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name); 2633 scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name); 2634 2635 return kcore_copy__compare_files(from_filename, to_filename); 2636 } 2637 2638 /** 2639 * kcore_copy - copy kallsyms, modules and kcore from one directory to another. 2640 * @from_dir: from directory 2641 * @to_dir: to directory 2642 * 2643 * This function copies kallsyms, modules and kcore files from one directory to 2644 * another. kallsyms and modules are copied entirely. Only code segments are 2645 * copied from kcore. It is assumed that two segments suffice: one for the 2646 * kernel proper and one for all the modules. The code segments are determined 2647 * from kallsyms and modules files. The kernel map starts at _stext or the 2648 * lowest function symbol, and ends at _etext or the highest function symbol. 2649 * The module map starts at the lowest module address and ends at the highest 2650 * module symbol. Start addresses are rounded down to the nearest page. End 2651 * addresses are rounded up to the nearest page. An extra page is added to the 2652 * highest kernel symbol and highest module symbol to, hopefully, encompass that 2653 * symbol too. Because it contains only code sections, the resulting kcore is 2654 * unusual. One significant peculiarity is that the mapping (start -> pgoff) 2655 * is not the same for the kernel map and the modules map. That happens because 2656 * the data is copied adjacently whereas the original kcore has gaps. Finally, 2657 * kallsyms file is compared with its copy to check that modules have not been 2658 * loaded or unloaded while the copies were taking place. 2659 * 2660 * Return: %0 on success, %-1 on failure. 2661 */ 2662 int kcore_copy(const char *from_dir, const char *to_dir) 2663 { 2664 struct kcore kcore; 2665 struct kcore extract; 2666 int idx = 0, err = -1; 2667 off_t offset, sz; 2668 struct kcore_copy_info kci = { .stext = 0, }; 2669 char kcore_filename[PATH_MAX]; 2670 char extract_filename[PATH_MAX]; 2671 struct phdr_data *p; 2672 2673 INIT_LIST_HEAD(&kci.phdrs); 2674 INIT_LIST_HEAD(&kci.syms); 2675 2676 if (kcore_copy__copy_file(from_dir, to_dir, "kallsyms")) 2677 return -1; 2678 2679 if (kcore_copy__copy_file(from_dir, to_dir, "modules")) 2680 goto out_unlink_kallsyms; 2681 2682 scnprintf(kcore_filename, PATH_MAX, "%s/kcore", from_dir); 2683 scnprintf(extract_filename, PATH_MAX, "%s/kcore", to_dir); 2684 2685 if (kcore__open(&kcore, kcore_filename)) 2686 goto out_unlink_modules; 2687 2688 if (kcore_copy__calc_maps(&kci, from_dir, kcore.elf)) 2689 goto out_kcore_close; 2690 2691 if (kcore__init(&extract, extract_filename, kcore.elfclass, false)) 2692 goto out_kcore_close; 2693 2694 if (kcore__copy_hdr(&kcore, &extract, kci.phnum)) 2695 goto out_extract_close; 2696 2697 offset = gelf_fsize(extract.elf, ELF_T_EHDR, 1, EV_CURRENT) + 2698 gelf_fsize(extract.elf, ELF_T_PHDR, kci.phnum, EV_CURRENT); 2699 offset = round_up(offset, page_size); 2700 2701 kcore_copy__for_each_phdr(&kci, p) { 2702 off_t offs = p->rel + offset; 2703 2704 if (kcore__add_phdr(&extract, idx++, offs, p->addr, p->len)) 2705 goto out_extract_close; 2706 } 2707 2708 sz = kcore__write(&extract); 2709 if (sz < 0 || sz > offset) 2710 goto out_extract_close; 2711 2712 kcore_copy__for_each_phdr(&kci, p) { 2713 off_t offs = p->rel + offset; 2714 2715 if (p->remaps) 2716 continue; 2717 if (copy_bytes(kcore.fd, p->offset, extract.fd, offs, p->len)) 2718 goto out_extract_close; 2719 } 2720 2721 if (kcore_copy__compare_file(from_dir, to_dir, "kallsyms")) 2722 goto out_extract_close; 2723 2724 err = 0; 2725 2726 out_extract_close: 2727 kcore__close(&extract); 2728 if (err) 2729 unlink(extract_filename); 2730 out_kcore_close: 2731 kcore__close(&kcore); 2732 out_unlink_modules: 2733 if (err) 2734 kcore_copy__unlink(to_dir, "modules"); 2735 out_unlink_kallsyms: 2736 if (err) 2737 kcore_copy__unlink(to_dir, "kallsyms"); 2738 2739 kcore_copy__free_phdrs(&kci); 2740 kcore_copy__free_syms(&kci); 2741 2742 return err; 2743 } 2744 2745 int kcore_extract__create(struct kcore_extract *kce) 2746 { 2747 struct kcore kcore; 2748 struct kcore extract; 2749 size_t count = 1; 2750 int idx = 0, err = -1; 2751 off_t offset = page_size, sz; 2752 2753 if (kcore__open(&kcore, kce->kcore_filename)) 2754 return -1; 2755 2756 strcpy(kce->extract_filename, PERF_KCORE_EXTRACT); 2757 if (kcore__init(&extract, kce->extract_filename, kcore.elfclass, true)) 2758 goto out_kcore_close; 2759 2760 if (kcore__copy_hdr(&kcore, &extract, count)) 2761 goto out_extract_close; 2762 2763 if (kcore__add_phdr(&extract, idx, offset, kce->addr, kce->len)) 2764 goto out_extract_close; 2765 2766 sz = kcore__write(&extract); 2767 if (sz < 0 || sz > offset) 2768 goto out_extract_close; 2769 2770 if (copy_bytes(kcore.fd, kce->offs, extract.fd, offset, kce->len)) 2771 goto out_extract_close; 2772 2773 err = 0; 2774 2775 out_extract_close: 2776 kcore__close(&extract); 2777 if (err) 2778 unlink(kce->extract_filename); 2779 out_kcore_close: 2780 kcore__close(&kcore); 2781 2782 return err; 2783 } 2784 2785 void kcore_extract__delete(struct kcore_extract *kce) 2786 { 2787 unlink(kce->extract_filename); 2788 } 2789 2790 #ifdef HAVE_GELF_GETNOTE_SUPPORT 2791 2792 static void sdt_adjust_loc(struct sdt_note *tmp, GElf_Addr base_off) 2793 { 2794 if (!base_off) 2795 return; 2796 2797 if (tmp->bit32) 2798 tmp->addr.a32[SDT_NOTE_IDX_LOC] = 2799 tmp->addr.a32[SDT_NOTE_IDX_LOC] + base_off - 2800 tmp->addr.a32[SDT_NOTE_IDX_BASE]; 2801 else 2802 tmp->addr.a64[SDT_NOTE_IDX_LOC] = 2803 tmp->addr.a64[SDT_NOTE_IDX_LOC] + base_off - 2804 tmp->addr.a64[SDT_NOTE_IDX_BASE]; 2805 } 2806 2807 static void sdt_adjust_refctr(struct sdt_note *tmp, GElf_Addr base_addr, 2808 GElf_Addr base_off) 2809 { 2810 if (!base_off) 2811 return; 2812 2813 if (tmp->bit32 && tmp->addr.a32[SDT_NOTE_IDX_REFCTR]) 2814 tmp->addr.a32[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off); 2815 else if (tmp->addr.a64[SDT_NOTE_IDX_REFCTR]) 2816 tmp->addr.a64[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off); 2817 } 2818 2819 /** 2820 * populate_sdt_note : Parse raw data and identify SDT note 2821 * @elf: elf of the opened file 2822 * @data: raw data of a section with description offset applied 2823 * @len: note description size 2824 * @type: type of the note 2825 * @sdt_notes: List to add the SDT note 2826 * 2827 * Responsible for parsing the @data in section .note.stapsdt in @elf and 2828 * if its an SDT note, it appends to @sdt_notes list. 2829 */ 2830 static int populate_sdt_note(Elf **elf, const char *data, size_t len, 2831 struct list_head *sdt_notes) 2832 { 2833 const char *provider, *name, *args; 2834 struct sdt_note *tmp = NULL; 2835 GElf_Ehdr ehdr; 2836 GElf_Shdr shdr; 2837 int ret = -EINVAL; 2838 2839 union { 2840 Elf64_Addr a64[NR_ADDR]; 2841 Elf32_Addr a32[NR_ADDR]; 2842 } buf; 2843 2844 Elf_Data dst = { 2845 .d_buf = &buf, .d_type = ELF_T_ADDR, .d_version = EV_CURRENT, 2846 .d_size = gelf_fsize((*elf), ELF_T_ADDR, NR_ADDR, EV_CURRENT), 2847 .d_off = 0, .d_align = 0 2848 }; 2849 Elf_Data src = { 2850 .d_buf = (void *) data, .d_type = ELF_T_ADDR, 2851 .d_version = EV_CURRENT, .d_size = dst.d_size, .d_off = 0, 2852 .d_align = 0 2853 }; 2854 2855 tmp = (struct sdt_note *)calloc(1, sizeof(struct sdt_note)); 2856 if (!tmp) { 2857 ret = -ENOMEM; 2858 goto out_err; 2859 } 2860 2861 INIT_LIST_HEAD(&tmp->note_list); 2862 2863 if (len < dst.d_size + 3) 2864 goto out_free_note; 2865 2866 /* Translation from file representation to memory representation */ 2867 if (gelf_xlatetom(*elf, &dst, &src, 2868 elf_getident(*elf, NULL)[EI_DATA]) == NULL) { 2869 pr_err("gelf_xlatetom : %s\n", elf_errmsg(-1)); 2870 goto out_free_note; 2871 } 2872 2873 /* Populate the fields of sdt_note */ 2874 provider = data + dst.d_size; 2875 2876 name = (const char *)memchr(provider, '\0', data + len - provider); 2877 if (name++ == NULL) 2878 goto out_free_note; 2879 2880 tmp->provider = strdup(provider); 2881 if (!tmp->provider) { 2882 ret = -ENOMEM; 2883 goto out_free_note; 2884 } 2885 tmp->name = strdup(name); 2886 if (!tmp->name) { 2887 ret = -ENOMEM; 2888 goto out_free_prov; 2889 } 2890 2891 args = memchr(name, '\0', data + len - name); 2892 2893 /* 2894 * There is no argument if: 2895 * - We reached the end of the note; 2896 * - There is not enough room to hold a potential string; 2897 * - The argument string is empty or just contains ':'. 2898 */ 2899 if (args == NULL || data + len - args < 2 || 2900 args[1] == ':' || args[1] == '\0') 2901 tmp->args = NULL; 2902 else { 2903 tmp->args = strdup(++args); 2904 if (!tmp->args) { 2905 ret = -ENOMEM; 2906 goto out_free_name; 2907 } 2908 } 2909 2910 if (gelf_getclass(*elf) == ELFCLASS32) { 2911 memcpy(&tmp->addr, &buf, 3 * sizeof(Elf32_Addr)); 2912 tmp->bit32 = true; 2913 } else { 2914 memcpy(&tmp->addr, &buf, 3 * sizeof(Elf64_Addr)); 2915 tmp->bit32 = false; 2916 } 2917 2918 if (!gelf_getehdr(*elf, &ehdr)) { 2919 pr_debug("%s : cannot get elf header.\n", __func__); 2920 ret = -EBADF; 2921 goto out_free_args; 2922 } 2923 2924 /* Adjust the prelink effect : 2925 * Find out the .stapsdt.base section. 2926 * This scn will help us to handle prelinking (if present). 2927 * Compare the retrieved file offset of the base section with the 2928 * base address in the description of the SDT note. If its different, 2929 * then accordingly, adjust the note location. 2930 */ 2931 if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_BASE_SCN, NULL)) 2932 sdt_adjust_loc(tmp, shdr.sh_offset); 2933 2934 /* Adjust reference counter offset */ 2935 if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_PROBES_SCN, NULL)) 2936 sdt_adjust_refctr(tmp, shdr.sh_addr, shdr.sh_offset); 2937 2938 list_add_tail(&tmp->note_list, sdt_notes); 2939 return 0; 2940 2941 out_free_args: 2942 zfree(&tmp->args); 2943 out_free_name: 2944 zfree(&tmp->name); 2945 out_free_prov: 2946 zfree(&tmp->provider); 2947 out_free_note: 2948 free(tmp); 2949 out_err: 2950 return ret; 2951 } 2952 2953 /** 2954 * construct_sdt_notes_list : constructs a list of SDT notes 2955 * @elf : elf to look into 2956 * @sdt_notes : empty list_head 2957 * 2958 * Scans the sections in 'elf' for the section 2959 * .note.stapsdt. It, then calls populate_sdt_note to find 2960 * out the SDT events and populates the 'sdt_notes'. 2961 */ 2962 static int construct_sdt_notes_list(Elf *elf, struct list_head *sdt_notes) 2963 { 2964 GElf_Ehdr ehdr; 2965 Elf_Scn *scn = NULL; 2966 Elf_Data *data; 2967 GElf_Shdr shdr; 2968 size_t shstrndx, next; 2969 GElf_Nhdr nhdr; 2970 size_t name_off, desc_off, offset; 2971 int ret = 0; 2972 2973 if (gelf_getehdr(elf, &ehdr) == NULL) { 2974 ret = -EBADF; 2975 goto out_ret; 2976 } 2977 if (elf_getshdrstrndx(elf, &shstrndx) != 0) { 2978 ret = -EBADF; 2979 goto out_ret; 2980 } 2981 2982 /* Look for the required section */ 2983 scn = elf_section_by_name(elf, &ehdr, &shdr, SDT_NOTE_SCN, NULL); 2984 if (!scn) { 2985 ret = -ENOENT; 2986 goto out_ret; 2987 } 2988 2989 if ((shdr.sh_type != SHT_NOTE) || (shdr.sh_flags & SHF_ALLOC)) { 2990 ret = -ENOENT; 2991 goto out_ret; 2992 } 2993 2994 data = elf_getdata(scn, NULL); 2995 2996 /* Get the SDT notes */ 2997 for (offset = 0; (next = gelf_getnote(data, offset, &nhdr, &name_off, 2998 &desc_off)) > 0; offset = next) { 2999 if (nhdr.n_namesz == sizeof(SDT_NOTE_NAME) && 3000 !memcmp(data->d_buf + name_off, SDT_NOTE_NAME, 3001 sizeof(SDT_NOTE_NAME))) { 3002 /* Check the type of the note */ 3003 if (nhdr.n_type != SDT_NOTE_TYPE) 3004 goto out_ret; 3005 3006 ret = populate_sdt_note(&elf, ((data->d_buf) + desc_off), 3007 nhdr.n_descsz, sdt_notes); 3008 if (ret < 0) 3009 goto out_ret; 3010 } 3011 } 3012 if (list_empty(sdt_notes)) 3013 ret = -ENOENT; 3014 3015 out_ret: 3016 return ret; 3017 } 3018 3019 /** 3020 * get_sdt_note_list : Wrapper to construct a list of sdt notes 3021 * @head : empty list_head 3022 * @target : file to find SDT notes from 3023 * 3024 * This opens the file, initializes 3025 * the ELF and then calls construct_sdt_notes_list. 3026 */ 3027 int get_sdt_note_list(struct list_head *head, const char *target) 3028 { 3029 Elf *elf; 3030 int fd, ret; 3031 3032 fd = open(target, O_RDONLY); 3033 if (fd < 0) 3034 return -EBADF; 3035 3036 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL); 3037 if (!elf) { 3038 ret = -EBADF; 3039 goto out_close; 3040 } 3041 ret = construct_sdt_notes_list(elf, head); 3042 elf_end(elf); 3043 out_close: 3044 close(fd); 3045 return ret; 3046 } 3047 3048 /** 3049 * cleanup_sdt_note_list : free the sdt notes' list 3050 * @sdt_notes: sdt notes' list 3051 * 3052 * Free up the SDT notes in @sdt_notes. 3053 * Returns the number of SDT notes free'd. 3054 */ 3055 int cleanup_sdt_note_list(struct list_head *sdt_notes) 3056 { 3057 struct sdt_note *tmp, *pos; 3058 int nr_free = 0; 3059 3060 list_for_each_entry_safe(pos, tmp, sdt_notes, note_list) { 3061 list_del_init(&pos->note_list); 3062 zfree(&pos->args); 3063 zfree(&pos->name); 3064 zfree(&pos->provider); 3065 free(pos); 3066 nr_free++; 3067 } 3068 return nr_free; 3069 } 3070 3071 /** 3072 * sdt_notes__get_count: Counts the number of sdt events 3073 * @start: list_head to sdt_notes list 3074 * 3075 * Returns the number of SDT notes in a list 3076 */ 3077 int sdt_notes__get_count(struct list_head *start) 3078 { 3079 struct sdt_note *sdt_ptr; 3080 int count = 0; 3081 3082 list_for_each_entry(sdt_ptr, start, note_list) 3083 count++; 3084 return count; 3085 } 3086 #endif 3087 3088 void symbol__elf_init(void) 3089 { 3090 elf_version(EV_CURRENT); 3091 } 3092