Brian Silverman | 8649792 | 2018-02-10 19:28:39 -0500 | [diff] [blame^] | 1 | /* Find debugging and symbol information for a module in libdwfl. |
| 2 | Copyright (C) 2005-2012, 2014, 2015 Red Hat, Inc. |
| 3 | This file is part of elfutils. |
| 4 | |
| 5 | This file is free software; you can redistribute it and/or modify |
| 6 | it under the terms of either |
| 7 | |
| 8 | * the GNU Lesser General Public License as published by the Free |
| 9 | Software Foundation; either version 3 of the License, or (at |
| 10 | your option) any later version |
| 11 | |
| 12 | or |
| 13 | |
| 14 | * the GNU General Public License as published by the Free |
| 15 | Software Foundation; either version 2 of the License, or (at |
| 16 | your option) any later version |
| 17 | |
| 18 | or both in parallel, as here. |
| 19 | |
| 20 | elfutils is distributed in the hope that it will be useful, but |
| 21 | WITHOUT ANY WARRANTY; without even the implied warranty of |
| 22 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 23 | General Public License for more details. |
| 24 | |
| 25 | You should have received copies of the GNU General Public License and |
| 26 | the GNU Lesser General Public License along with this program. If |
| 27 | not, see <http://www.gnu.org/licenses/>. */ |
| 28 | |
| 29 | #ifdef HAVE_CONFIG_H |
| 30 | # include <config.h> |
| 31 | #endif |
| 32 | |
| 33 | #include "libdwflP.h" |
| 34 | #include <inttypes.h> |
| 35 | #include <fcntl.h> |
| 36 | #include <string.h> |
| 37 | #include <unistd.h> |
| 38 | #include "../libdw/libdwP.h" /* DWARF_E_* values are here. */ |
| 39 | #include "../libelf/libelfP.h" |
| 40 | #include "system.h" |
| 41 | |
| 42 | static inline Dwfl_Error |
| 43 | open_elf_file (Elf **elf, int *fd, char **name) |
| 44 | { |
| 45 | if (*elf == NULL) |
| 46 | { |
| 47 | /* CBFAIL uses errno if it's set, so clear it first in case we don't |
| 48 | set it with an open failure below. */ |
| 49 | errno = 0; |
| 50 | |
| 51 | /* If there was a pre-primed file name left that the callback left |
| 52 | behind, try to open that file name. */ |
| 53 | if (*fd < 0 && *name != NULL) |
| 54 | *fd = TEMP_FAILURE_RETRY (open (*name, O_RDONLY)); |
| 55 | |
| 56 | if (*fd < 0) |
| 57 | return CBFAIL; |
| 58 | |
| 59 | return __libdw_open_file (fd, elf, true, false); |
| 60 | } |
| 61 | else if (unlikely (elf_kind (*elf) != ELF_K_ELF)) |
| 62 | { |
| 63 | elf_end (*elf); |
| 64 | *elf = NULL; |
| 65 | close (*fd); |
| 66 | *fd = -1; |
| 67 | return DWFL_E_BADELF; |
| 68 | } |
| 69 | |
| 70 | /* Elf file already open and looks fine. */ |
| 71 | return DWFL_E_NOERROR; |
| 72 | } |
| 73 | |
| 74 | /* Open libelf FILE->fd and compute the load base of ELF as loaded in MOD. |
| 75 | When we return success, FILE->elf and FILE->vaddr are set up. */ |
| 76 | static inline Dwfl_Error |
| 77 | open_elf (Dwfl_Module *mod, struct dwfl_file *file) |
| 78 | { |
| 79 | Dwfl_Error error = open_elf_file (&file->elf, &file->fd, &file->name); |
| 80 | if (error != DWFL_E_NOERROR) |
| 81 | return error; |
| 82 | |
| 83 | GElf_Ehdr ehdr_mem, *ehdr = gelf_getehdr (file->elf, &ehdr_mem); |
| 84 | if (ehdr == NULL) |
| 85 | { |
| 86 | elf_error: |
| 87 | elf_end (file->elf); |
| 88 | file->elf = NULL; |
| 89 | close (file->fd); |
| 90 | file->fd = -1; |
| 91 | return DWFL_E (LIBELF, elf_errno ()); |
| 92 | } |
| 93 | |
| 94 | if (ehdr->e_type != ET_REL) |
| 95 | { |
| 96 | /* In any non-ET_REL file, we compute the "synchronization address". |
| 97 | |
| 98 | We start with the address at the end of the first PT_LOAD |
| 99 | segment. When prelink converts REL to RELA in an ET_DYN |
| 100 | file, it expands the space between the beginning of the |
| 101 | segment and the actual code/data addresses. Since that |
| 102 | change wasn't made in the debug file, the distance from |
| 103 | p_vaddr to an address of interest (in an st_value or DWARF |
| 104 | data) now differs between the main and debug files. The |
| 105 | distance from address_sync to an address of interest remains |
| 106 | consistent. |
| 107 | |
| 108 | If there are no section headers at all (full stripping), then |
| 109 | the end of the first segment is a valid synchronization address. |
| 110 | This cannot happen in a prelinked file, since prelink itself |
| 111 | relies on section headers for prelinking and for undoing it. |
| 112 | (If you do full stripping on a prelinked file, then you get what |
| 113 | you deserve--you can neither undo the prelinking, nor expect to |
| 114 | line it up with a debug file separated before prelinking.) |
| 115 | |
| 116 | However, when prelink processes an ET_EXEC file, it can do |
| 117 | something different. There it juggles the "special" sections |
| 118 | (SHT_DYNSYM et al) to make space for the additional prelink |
| 119 | special sections. Sometimes it will do this by moving a special |
| 120 | section like .dynstr after the real program sections in the first |
| 121 | PT_LOAD segment--i.e. to the end. That changes the end address of |
| 122 | the segment, so it no longer lines up correctly and is not a valid |
| 123 | synchronization address to use. Because of this, we need to apply |
| 124 | a different prelink-savvy means to discover the synchronization |
| 125 | address when there is a separate debug file and a prelinked main |
| 126 | file. That is done in find_debuginfo, below. */ |
| 127 | |
| 128 | size_t phnum; |
| 129 | if (unlikely (elf_getphdrnum (file->elf, &phnum) != 0)) |
| 130 | goto elf_error; |
| 131 | |
| 132 | file->vaddr = file->address_sync = 0; |
| 133 | for (size_t i = 0; i < phnum; ++i) |
| 134 | { |
| 135 | GElf_Phdr ph_mem; |
| 136 | GElf_Phdr *ph = gelf_getphdr (file->elf, i, &ph_mem); |
| 137 | if (unlikely (ph == NULL)) |
| 138 | goto elf_error; |
| 139 | if (ph->p_type == PT_LOAD) |
| 140 | { |
| 141 | file->vaddr = ph->p_vaddr & -ph->p_align; |
| 142 | file->address_sync = ph->p_vaddr + ph->p_memsz; |
| 143 | break; |
| 144 | } |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | /* We only want to set the module e_type explictly once, derived from |
| 149 | the main ELF file. (It might be changed for the kernel, because |
| 150 | that is special - see below.) open_elf is always called first for |
| 151 | the main ELF file, because both find_dw and find_symtab call |
| 152 | __libdwfl_getelf first to open the main file. So don't let debug |
| 153 | or aux files override the module e_type. The kernel heuristic |
| 154 | below could otherwise trigger for non-kernel/non-main files, since |
| 155 | their phdrs might not match the actual load addresses. */ |
| 156 | if (file == &mod->main) |
| 157 | { |
| 158 | mod->e_type = ehdr->e_type; |
| 159 | |
| 160 | /* Relocatable Linux kernels are ET_EXEC but act like ET_DYN. */ |
| 161 | if (mod->e_type == ET_EXEC && file->vaddr != mod->low_addr) |
| 162 | mod->e_type = ET_DYN; |
| 163 | } |
| 164 | else |
| 165 | assert (mod->main.elf != NULL); |
| 166 | |
| 167 | return DWFL_E_NOERROR; |
| 168 | } |
| 169 | |
| 170 | /* We have an authoritative build ID for this module MOD, so don't use |
| 171 | a file by name that doesn't match that ID. */ |
| 172 | static void |
| 173 | mod_verify_build_id (Dwfl_Module *mod) |
| 174 | { |
| 175 | assert (mod->build_id_len > 0); |
| 176 | |
| 177 | switch (__builtin_expect (__libdwfl_find_build_id (mod, false, |
| 178 | mod->main.elf), 2)) |
| 179 | { |
| 180 | case 2: |
| 181 | /* Build ID matches as it should. */ |
| 182 | return; |
| 183 | |
| 184 | case -1: /* ELF error. */ |
| 185 | mod->elferr = INTUSE(dwfl_errno) (); |
| 186 | break; |
| 187 | |
| 188 | case 0: /* File has no build ID note. */ |
| 189 | case 1: /* FIle has a build ID that does not match. */ |
| 190 | mod->elferr = DWFL_E_WRONG_ID_ELF; |
| 191 | break; |
| 192 | |
| 193 | default: |
| 194 | abort (); |
| 195 | } |
| 196 | |
| 197 | /* We get here when it was the right ELF file. Clear it out. */ |
| 198 | elf_end (mod->main.elf); |
| 199 | mod->main.elf = NULL; |
| 200 | if (mod->main.fd >= 0) |
| 201 | { |
| 202 | close (mod->main.fd); |
| 203 | mod->main.fd = -1; |
| 204 | } |
| 205 | } |
| 206 | |
| 207 | /* Find the main ELF file for this module and open libelf on it. |
| 208 | When we return success, MOD->main.elf and MOD->main.bias are set up. */ |
| 209 | void |
| 210 | internal_function |
| 211 | __libdwfl_getelf (Dwfl_Module *mod) |
| 212 | { |
| 213 | if (mod->main.elf != NULL /* Already done. */ |
| 214 | || mod->elferr != DWFL_E_NOERROR) /* Cached failure. */ |
| 215 | return; |
| 216 | |
| 217 | mod->main.fd = (*mod->dwfl->callbacks->find_elf) (MODCB_ARGS (mod), |
| 218 | &mod->main.name, |
| 219 | &mod->main.elf); |
| 220 | const bool fallback = mod->main.elf == NULL && mod->main.fd < 0; |
| 221 | mod->elferr = open_elf (mod, &mod->main); |
| 222 | if (mod->elferr != DWFL_E_NOERROR) |
| 223 | return; |
| 224 | |
| 225 | if (!mod->main.valid) |
| 226 | { |
| 227 | /* Clear any explicitly reported build ID, just in case it was wrong. |
| 228 | We'll fetch it from the file when asked. */ |
| 229 | free (mod->build_id_bits); |
| 230 | mod->build_id_bits = NULL; |
| 231 | mod->build_id_len = 0; |
| 232 | } |
| 233 | else if (fallback) |
| 234 | mod_verify_build_id (mod); |
| 235 | |
| 236 | mod->main_bias = mod->e_type == ET_REL ? 0 : mod->low_addr - mod->main.vaddr; |
| 237 | } |
| 238 | |
| 239 | static inline void |
| 240 | consider_shdr (GElf_Addr interp, |
| 241 | GElf_Word sh_type, |
| 242 | GElf_Xword sh_flags, |
| 243 | GElf_Addr sh_addr, |
| 244 | GElf_Xword sh_size, |
| 245 | GElf_Addr *phighest) |
| 246 | { |
| 247 | if ((sh_flags & SHF_ALLOC) |
| 248 | && ((sh_type == SHT_PROGBITS && sh_addr != interp) |
| 249 | || sh_type == SHT_NOBITS)) |
| 250 | { |
| 251 | const GElf_Addr sh_end = sh_addr + sh_size; |
| 252 | if (sh_end > *phighest) |
| 253 | *phighest = sh_end; |
| 254 | } |
| 255 | } |
| 256 | |
| 257 | /* If the main file might have been prelinked, then we need to |
| 258 | discover the correct synchronization address between the main and |
| 259 | debug files. Because of prelink's section juggling, we cannot rely |
| 260 | on the address_sync computed from PT_LOAD segments (see open_elf). |
| 261 | |
| 262 | We will attempt to discover a synchronization address based on the |
| 263 | section headers instead. But finding a section address that is |
| 264 | safe to use requires identifying which sections are SHT_PROGBITS. |
| 265 | We can do that in the main file, but in the debug file all the |
| 266 | allocated sections have been transformed into SHT_NOBITS so we have |
| 267 | lost the means to match them up correctly. |
| 268 | |
| 269 | The only method left to us is to decode the .gnu.prelink_undo |
| 270 | section in the prelinked main file. This shows what the sections |
| 271 | looked like before prelink juggled them--when they still had a |
| 272 | direct correspondence to the debug file. */ |
| 273 | static Dwfl_Error |
| 274 | find_prelink_address_sync (Dwfl_Module *mod, struct dwfl_file *file) |
| 275 | { |
| 276 | /* The magic section is only identified by name. */ |
| 277 | size_t shstrndx; |
| 278 | if (elf_getshdrstrndx (mod->main.elf, &shstrndx) < 0) |
| 279 | return DWFL_E_LIBELF; |
| 280 | |
| 281 | Elf_Scn *scn = NULL; |
| 282 | while ((scn = elf_nextscn (mod->main.elf, scn)) != NULL) |
| 283 | { |
| 284 | GElf_Shdr shdr_mem; |
| 285 | GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_mem); |
| 286 | if (unlikely (shdr == NULL)) |
| 287 | return DWFL_E_LIBELF; |
| 288 | if (shdr->sh_type == SHT_PROGBITS |
| 289 | && !(shdr->sh_flags & SHF_ALLOC) |
| 290 | && shdr->sh_name != 0) |
| 291 | { |
| 292 | const char *secname = elf_strptr (mod->main.elf, shstrndx, |
| 293 | shdr->sh_name); |
| 294 | if (unlikely (secname == NULL)) |
| 295 | return DWFL_E_LIBELF; |
| 296 | if (!strcmp (secname, ".gnu.prelink_undo")) |
| 297 | break; |
| 298 | } |
| 299 | } |
| 300 | |
| 301 | if (scn == NULL) |
| 302 | /* There was no .gnu.prelink_undo section. */ |
| 303 | return DWFL_E_NOERROR; |
| 304 | |
| 305 | Elf_Data *undodata = elf_rawdata (scn, NULL); |
| 306 | if (unlikely (undodata == NULL)) |
| 307 | return DWFL_E_LIBELF; |
| 308 | |
| 309 | /* Decode the section. It consists of the original ehdr, phdrs, |
| 310 | and shdrs (but omits section 0). */ |
| 311 | |
| 312 | union |
| 313 | { |
| 314 | Elf32_Ehdr e32; |
| 315 | Elf64_Ehdr e64; |
| 316 | } ehdr; |
| 317 | Elf_Data dst = |
| 318 | { |
| 319 | .d_buf = &ehdr, |
| 320 | .d_size = sizeof ehdr, |
| 321 | .d_type = ELF_T_EHDR, |
| 322 | .d_version = EV_CURRENT |
| 323 | }; |
| 324 | Elf_Data src = *undodata; |
| 325 | src.d_size = gelf_fsize (mod->main.elf, ELF_T_EHDR, 1, EV_CURRENT); |
| 326 | src.d_type = ELF_T_EHDR; |
| 327 | if (unlikely (gelf_xlatetom (mod->main.elf, &dst, &src, |
| 328 | elf_getident (mod->main.elf, NULL)[EI_DATA]) |
| 329 | == NULL)) |
| 330 | return DWFL_E_LIBELF; |
| 331 | |
| 332 | size_t shentsize = gelf_fsize (mod->main.elf, ELF_T_SHDR, 1, EV_CURRENT); |
| 333 | size_t phentsize = gelf_fsize (mod->main.elf, ELF_T_PHDR, 1, EV_CURRENT); |
| 334 | |
| 335 | uint_fast16_t phnum; |
| 336 | uint_fast16_t shnum; |
| 337 | if (ehdr.e32.e_ident[EI_CLASS] == ELFCLASS32) |
| 338 | { |
| 339 | if (ehdr.e32.e_shentsize != shentsize |
| 340 | || ehdr.e32.e_phentsize != phentsize) |
| 341 | return DWFL_E_BAD_PRELINK; |
| 342 | phnum = ehdr.e32.e_phnum; |
| 343 | shnum = ehdr.e32.e_shnum; |
| 344 | } |
| 345 | else |
| 346 | { |
| 347 | if (ehdr.e64.e_shentsize != shentsize |
| 348 | || ehdr.e64.e_phentsize != phentsize) |
| 349 | return DWFL_E_BAD_PRELINK; |
| 350 | phnum = ehdr.e64.e_phnum; |
| 351 | shnum = ehdr.e64.e_shnum; |
| 352 | } |
| 353 | |
| 354 | /* Since prelink does not store the zeroth section header in the undo |
| 355 | section, it cannot support SHN_XINDEX encoding. */ |
| 356 | if (unlikely (shnum >= SHN_LORESERVE) || unlikely(shnum == 0) |
| 357 | || unlikely (undodata->d_size != (src.d_size |
| 358 | + phnum * phentsize |
| 359 | + (shnum - 1) * shentsize))) |
| 360 | return DWFL_E_BAD_PRELINK; |
| 361 | |
| 362 | --shnum; |
| 363 | |
| 364 | /* We look at the allocated SHT_PROGBITS (or SHT_NOBITS) sections. (Most |
| 365 | every file will have some SHT_PROGBITS sections, but it's possible to |
| 366 | have one with nothing but .bss, i.e. SHT_NOBITS.) The special sections |
| 367 | that can be moved around have different sh_type values--except for |
| 368 | .interp, the section that became the PT_INTERP segment. So we exclude |
| 369 | the SHT_PROGBITS section whose address matches the PT_INTERP p_vaddr. |
| 370 | For this reason, we must examine the phdrs first to find PT_INTERP. */ |
| 371 | |
| 372 | GElf_Addr main_interp = 0; |
| 373 | { |
| 374 | size_t main_phnum; |
| 375 | if (unlikely (elf_getphdrnum (mod->main.elf, &main_phnum))) |
| 376 | return DWFL_E_LIBELF; |
| 377 | for (size_t i = 0; i < main_phnum; ++i) |
| 378 | { |
| 379 | GElf_Phdr phdr; |
| 380 | if (unlikely (gelf_getphdr (mod->main.elf, i, &phdr) == NULL)) |
| 381 | return DWFL_E_LIBELF; |
| 382 | if (phdr.p_type == PT_INTERP) |
| 383 | { |
| 384 | main_interp = phdr.p_vaddr; |
| 385 | break; |
| 386 | } |
| 387 | } |
| 388 | } |
| 389 | |
| 390 | src.d_buf += src.d_size; |
| 391 | src.d_type = ELF_T_PHDR; |
| 392 | src.d_size = phnum * phentsize; |
| 393 | |
| 394 | GElf_Addr undo_interp = 0; |
| 395 | bool class32 = ehdr.e32.e_ident[EI_CLASS] == ELFCLASS32; |
| 396 | { |
| 397 | size_t phdr_size = class32 ? sizeof (Elf32_Phdr) : sizeof (Elf64_Phdr); |
| 398 | if (unlikely (phnum > SIZE_MAX / phdr_size)) |
| 399 | return DWFL_E_NOMEM; |
| 400 | const size_t phdrs_bytes = phnum * phdr_size; |
| 401 | void *phdrs = malloc (phdrs_bytes); |
| 402 | if (unlikely (phdrs == NULL)) |
| 403 | return DWFL_E_NOMEM; |
| 404 | dst.d_buf = phdrs; |
| 405 | dst.d_size = phdrs_bytes; |
| 406 | if (unlikely (gelf_xlatetom (mod->main.elf, &dst, &src, |
| 407 | ehdr.e32.e_ident[EI_DATA]) == NULL)) |
| 408 | { |
| 409 | free (phdrs); |
| 410 | return DWFL_E_LIBELF; |
| 411 | } |
| 412 | if (class32) |
| 413 | { |
| 414 | Elf32_Phdr (*p32)[phnum] = phdrs; |
| 415 | for (uint_fast16_t i = 0; i < phnum; ++i) |
| 416 | if ((*p32)[i].p_type == PT_INTERP) |
| 417 | { |
| 418 | undo_interp = (*p32)[i].p_vaddr; |
| 419 | break; |
| 420 | } |
| 421 | } |
| 422 | else |
| 423 | { |
| 424 | Elf64_Phdr (*p64)[phnum] = phdrs; |
| 425 | for (uint_fast16_t i = 0; i < phnum; ++i) |
| 426 | if ((*p64)[i].p_type == PT_INTERP) |
| 427 | { |
| 428 | undo_interp = (*p64)[i].p_vaddr; |
| 429 | break; |
| 430 | } |
| 431 | } |
| 432 | free (phdrs); |
| 433 | } |
| 434 | |
| 435 | if (unlikely ((main_interp == 0) != (undo_interp == 0))) |
| 436 | return DWFL_E_BAD_PRELINK; |
| 437 | |
| 438 | src.d_buf += src.d_size; |
| 439 | src.d_type = ELF_T_SHDR; |
| 440 | src.d_size = gelf_fsize (mod->main.elf, ELF_T_SHDR, shnum, EV_CURRENT); |
| 441 | |
| 442 | size_t shdr_size = class32 ? sizeof (Elf32_Shdr) : sizeof (Elf64_Shdr); |
| 443 | if (unlikely (shnum > SIZE_MAX / shdr_size)) |
| 444 | return DWFL_E_NOMEM; |
| 445 | const size_t shdrs_bytes = shnum * shdr_size; |
| 446 | void *shdrs = malloc (shdrs_bytes); |
| 447 | if (unlikely (shdrs == NULL)) |
| 448 | return DWFL_E_NOMEM; |
| 449 | dst.d_buf = shdrs; |
| 450 | dst.d_size = shdrs_bytes; |
| 451 | if (unlikely (gelf_xlatetom (mod->main.elf, &dst, &src, |
| 452 | ehdr.e32.e_ident[EI_DATA]) == NULL)) |
| 453 | { |
| 454 | free (shdrs); |
| 455 | return DWFL_E_LIBELF; |
| 456 | } |
| 457 | |
| 458 | /* Now we can look at the original section headers of the main file |
| 459 | before it was prelinked. First we'll apply our method to the main |
| 460 | file sections as they are after prelinking, to calculate the |
| 461 | synchronization address of the main file. Then we'll apply that |
| 462 | same method to the saved section headers, to calculate the matching |
| 463 | synchronization address of the debug file. |
| 464 | |
| 465 | The method is to consider SHF_ALLOC sections that are either |
| 466 | SHT_PROGBITS or SHT_NOBITS, excluding the section whose sh_addr |
| 467 | matches the PT_INTERP p_vaddr. The special sections that can be |
| 468 | moved by prelink have other types, except for .interp (which |
| 469 | becomes PT_INTERP). The "real" sections cannot move as such, but |
| 470 | .bss can be split into .dynbss and .bss, with the total memory |
| 471 | image remaining the same but being spread across the two sections. |
| 472 | So we consider the highest section end, which still matches up. */ |
| 473 | |
| 474 | GElf_Addr highest; |
| 475 | |
| 476 | highest = 0; |
| 477 | scn = NULL; |
| 478 | while ((scn = elf_nextscn (mod->main.elf, scn)) != NULL) |
| 479 | { |
| 480 | GElf_Shdr sh_mem; |
| 481 | GElf_Shdr *sh = gelf_getshdr (scn, &sh_mem); |
| 482 | if (unlikely (sh == NULL)) |
| 483 | { |
| 484 | free (shdrs); |
| 485 | return DWFL_E_LIBELF; |
| 486 | } |
| 487 | consider_shdr (main_interp, sh->sh_type, sh->sh_flags, |
| 488 | sh->sh_addr, sh->sh_size, &highest); |
| 489 | } |
| 490 | if (highest > mod->main.vaddr) |
| 491 | { |
| 492 | mod->main.address_sync = highest; |
| 493 | |
| 494 | highest = 0; |
| 495 | if (class32) |
| 496 | { |
| 497 | Elf32_Shdr (*s32)[shnum] = shdrs; |
| 498 | for (size_t i = 0; i < shnum; ++i) |
| 499 | consider_shdr (undo_interp, (*s32)[i].sh_type, |
| 500 | (*s32)[i].sh_flags, (*s32)[i].sh_addr, |
| 501 | (*s32)[i].sh_size, &highest); |
| 502 | } |
| 503 | else |
| 504 | { |
| 505 | Elf64_Shdr (*s64)[shnum] = shdrs; |
| 506 | for (size_t i = 0; i < shnum; ++i) |
| 507 | consider_shdr (undo_interp, (*s64)[i].sh_type, |
| 508 | (*s64)[i].sh_flags, (*s64)[i].sh_addr, |
| 509 | (*s64)[i].sh_size, &highest); |
| 510 | } |
| 511 | |
| 512 | if (highest > file->vaddr) |
| 513 | file->address_sync = highest; |
| 514 | else |
| 515 | { |
| 516 | free (shdrs); |
| 517 | return DWFL_E_BAD_PRELINK; |
| 518 | } |
| 519 | } |
| 520 | |
| 521 | free (shdrs); |
| 522 | |
| 523 | return DWFL_E_NOERROR; |
| 524 | } |
| 525 | |
| 526 | /* Find the separate debuginfo file for this module and open libelf on it. |
| 527 | When we return success, MOD->debug is set up. */ |
| 528 | static Dwfl_Error |
| 529 | find_debuginfo (Dwfl_Module *mod) |
| 530 | { |
| 531 | if (mod->debug.elf != NULL) |
| 532 | return DWFL_E_NOERROR; |
| 533 | |
| 534 | GElf_Word debuglink_crc = 0; |
| 535 | const char *debuglink_file; |
| 536 | debuglink_file = INTUSE(dwelf_elf_gnu_debuglink) (mod->main.elf, |
| 537 | &debuglink_crc); |
| 538 | |
| 539 | mod->debug.fd = (*mod->dwfl->callbacks->find_debuginfo) (MODCB_ARGS (mod), |
| 540 | mod->main.name, |
| 541 | debuglink_file, |
| 542 | debuglink_crc, |
| 543 | &mod->debug.name); |
| 544 | Dwfl_Error result = open_elf (mod, &mod->debug); |
| 545 | if (result == DWFL_E_NOERROR && mod->debug.address_sync != 0) |
| 546 | result = find_prelink_address_sync (mod, &mod->debug); |
| 547 | return result; |
| 548 | } |
| 549 | |
| 550 | /* Try to find the alternative debug link for the given DWARF and set |
| 551 | it if found. Only called when mod->dw is already setup but still |
| 552 | might need an alternative (dwz multi) debug file. filename is either |
| 553 | the main or debug name from which the Dwarf was created. */ |
| 554 | static void |
| 555 | find_debug_altlink (Dwfl_Module *mod, const char *filename) |
| 556 | { |
| 557 | assert (mod->dw != NULL); |
| 558 | |
| 559 | const char *altname; |
| 560 | const void *build_id; |
| 561 | ssize_t build_id_len = INTUSE(dwelf_dwarf_gnu_debugaltlink) (mod->dw, |
| 562 | &altname, |
| 563 | &build_id); |
| 564 | |
| 565 | if (build_id_len > 0) |
| 566 | { |
| 567 | /* We could store altfile in the module, but don't really need it. */ |
| 568 | char *altfile = NULL; |
| 569 | mod->alt_fd = (*mod->dwfl->callbacks->find_debuginfo) (MODCB_ARGS (mod), |
| 570 | filename, |
| 571 | altname, |
| 572 | 0, |
| 573 | &altfile); |
| 574 | |
| 575 | /* The (internal) callbacks might just set mod->alt_elf directly |
| 576 | because they open the Elf anyway for sanity checking. |
| 577 | Otherwise open either the given file name or use the fd |
| 578 | returned. */ |
| 579 | Dwfl_Error error = open_elf_file (&mod->alt_elf, &mod->alt_fd, |
| 580 | &altfile); |
| 581 | if (error == DWFL_E_NOERROR) |
| 582 | { |
| 583 | mod->alt = INTUSE(dwarf_begin_elf) (mod->alt_elf, |
| 584 | DWARF_C_READ, NULL); |
| 585 | if (mod->alt == NULL) |
| 586 | { |
| 587 | elf_end (mod->alt_elf); |
| 588 | mod->alt_elf = NULL; |
| 589 | close (mod->alt_fd); |
| 590 | mod->alt_fd = -1; |
| 591 | } |
| 592 | else |
| 593 | dwarf_setalt (mod->dw, mod->alt); |
| 594 | } |
| 595 | |
| 596 | free (altfile); /* See above, we don't really need it. */ |
| 597 | } |
| 598 | } |
| 599 | |
| 600 | /* Try to find a symbol table in FILE. |
| 601 | Returns DWFL_E_NOERROR if a proper one is found. |
| 602 | Returns DWFL_E_NO_SYMTAB if not, but still sets results for SHT_DYNSYM. */ |
| 603 | static Dwfl_Error |
| 604 | load_symtab (struct dwfl_file *file, struct dwfl_file **symfile, |
| 605 | Elf_Scn **symscn, Elf_Scn **xndxscn, |
| 606 | size_t *syments, int *first_global, GElf_Word *strshndx) |
| 607 | { |
| 608 | bool symtab = false; |
| 609 | Elf_Scn *scn = NULL; |
| 610 | while ((scn = elf_nextscn (file->elf, scn)) != NULL) |
| 611 | { |
| 612 | GElf_Shdr shdr_mem, *shdr = gelf_getshdr (scn, &shdr_mem); |
| 613 | if (shdr != NULL) |
| 614 | switch (shdr->sh_type) |
| 615 | { |
| 616 | case SHT_SYMTAB: |
| 617 | if (shdr->sh_entsize == 0) |
| 618 | break; |
| 619 | symtab = true; |
| 620 | *symscn = scn; |
| 621 | *symfile = file; |
| 622 | *strshndx = shdr->sh_link; |
| 623 | *syments = shdr->sh_size / shdr->sh_entsize; |
| 624 | *first_global = shdr->sh_info; |
| 625 | if (*xndxscn != NULL) |
| 626 | return DWFL_E_NOERROR; |
| 627 | break; |
| 628 | |
| 629 | case SHT_DYNSYM: |
| 630 | if (symtab) |
| 631 | break; |
| 632 | /* Use this if need be, but keep looking for SHT_SYMTAB. */ |
| 633 | if (shdr->sh_entsize == 0) |
| 634 | break; |
| 635 | *symscn = scn; |
| 636 | *symfile = file; |
| 637 | *strshndx = shdr->sh_link; |
| 638 | *syments = shdr->sh_size / shdr->sh_entsize; |
| 639 | *first_global = shdr->sh_info; |
| 640 | break; |
| 641 | |
| 642 | case SHT_SYMTAB_SHNDX: |
| 643 | *xndxscn = scn; |
| 644 | if (symtab) |
| 645 | return DWFL_E_NOERROR; |
| 646 | break; |
| 647 | |
| 648 | default: |
| 649 | break; |
| 650 | } |
| 651 | } |
| 652 | |
| 653 | if (symtab) |
| 654 | /* We found one, though no SHT_SYMTAB_SHNDX to go with it. */ |
| 655 | return DWFL_E_NOERROR; |
| 656 | |
| 657 | /* We found no SHT_SYMTAB, so any SHT_SYMTAB_SHNDX was bogus. |
| 658 | We might have found an SHT_DYNSYM and set *SYMSCN et al though. */ |
| 659 | *xndxscn = NULL; |
| 660 | return DWFL_E_NO_SYMTAB; |
| 661 | } |
| 662 | |
| 663 | |
| 664 | /* Translate addresses into file offsets. |
| 665 | OFFS[*] start out zero and remain zero if unresolved. */ |
| 666 | static void |
| 667 | find_offsets (Elf *elf, GElf_Addr main_bias, size_t phnum, size_t n, |
| 668 | GElf_Addr addrs[n], GElf_Off offs[n]) |
| 669 | { |
| 670 | size_t unsolved = n; |
| 671 | for (size_t i = 0; i < phnum; ++i) |
| 672 | { |
| 673 | GElf_Phdr phdr_mem; |
| 674 | GElf_Phdr *phdr = gelf_getphdr (elf, i, &phdr_mem); |
| 675 | if (phdr != NULL && phdr->p_type == PT_LOAD && phdr->p_memsz > 0) |
| 676 | for (size_t j = 0; j < n; ++j) |
| 677 | if (offs[j] == 0 |
| 678 | && addrs[j] >= phdr->p_vaddr + main_bias |
| 679 | && addrs[j] - (phdr->p_vaddr + main_bias) < phdr->p_filesz) |
| 680 | { |
| 681 | offs[j] = addrs[j] - (phdr->p_vaddr + main_bias) + phdr->p_offset; |
| 682 | if (--unsolved == 0) |
| 683 | break; |
| 684 | } |
| 685 | } |
| 686 | } |
| 687 | |
| 688 | /* Various addresses we might want to pull from the dynamic segment. */ |
| 689 | enum |
| 690 | { |
| 691 | i_symtab, |
| 692 | i_strtab, |
| 693 | i_hash, |
| 694 | i_gnu_hash, |
| 695 | i_max |
| 696 | }; |
| 697 | |
| 698 | /* Translate pointers into file offsets. ADJUST is either zero |
| 699 | in case the dynamic segment wasn't adjusted or mod->main_bias. |
| 700 | Will set mod->symfile if the translated offsets can be used as |
| 701 | symbol table. */ |
| 702 | static void |
| 703 | translate_offs (GElf_Addr adjust, |
| 704 | Dwfl_Module *mod, size_t phnum, |
| 705 | GElf_Addr addrs[i_max], GElf_Xword strsz, |
| 706 | GElf_Ehdr *ehdr) |
| 707 | { |
| 708 | GElf_Off offs[i_max] = { 0, }; |
| 709 | find_offsets (mod->main.elf, adjust, phnum, i_max, addrs, offs); |
| 710 | |
| 711 | /* Figure out the size of the symbol table. */ |
| 712 | if (offs[i_hash] != 0) |
| 713 | { |
| 714 | /* In the original format, .hash says the size of .dynsym. */ |
| 715 | |
| 716 | size_t entsz = SH_ENTSIZE_HASH (ehdr); |
| 717 | Elf_Data *data = elf_getdata_rawchunk (mod->main.elf, |
| 718 | offs[i_hash] + entsz, entsz, |
| 719 | (entsz == 4 |
| 720 | ? ELF_T_WORD : ELF_T_XWORD)); |
| 721 | if (data != NULL) |
| 722 | mod->syments = (entsz == 4 |
| 723 | ? *(const GElf_Word *) data->d_buf |
| 724 | : *(const GElf_Xword *) data->d_buf); |
| 725 | } |
| 726 | if (offs[i_gnu_hash] != 0 && mod->syments == 0) |
| 727 | { |
| 728 | /* In the new format, we can derive it with some work. */ |
| 729 | |
| 730 | const struct |
| 731 | { |
| 732 | Elf32_Word nbuckets; |
| 733 | Elf32_Word symndx; |
| 734 | Elf32_Word maskwords; |
| 735 | Elf32_Word shift2; |
| 736 | } *header; |
| 737 | |
| 738 | Elf_Data *data = elf_getdata_rawchunk (mod->main.elf, offs[i_gnu_hash], |
| 739 | sizeof *header, ELF_T_WORD); |
| 740 | if (data != NULL) |
| 741 | { |
| 742 | header = data->d_buf; |
| 743 | Elf32_Word nbuckets = header->nbuckets; |
| 744 | Elf32_Word symndx = header->symndx; |
| 745 | GElf_Off buckets_at = (offs[i_gnu_hash] + sizeof *header |
| 746 | + (gelf_getclass (mod->main.elf) |
| 747 | * sizeof (Elf32_Word) |
| 748 | * header->maskwords)); |
| 749 | |
| 750 | // elf_getdata_rawchunk takes a size_t, make sure it |
| 751 | // doesn't overflow. |
| 752 | #if SIZE_MAX <= UINT32_MAX |
| 753 | if (nbuckets > SIZE_MAX / sizeof (Elf32_Word)) |
| 754 | data = NULL; |
| 755 | else |
| 756 | #endif |
| 757 | data = elf_getdata_rawchunk (mod->main.elf, buckets_at, |
| 758 | nbuckets * sizeof (Elf32_Word), |
| 759 | ELF_T_WORD); |
| 760 | if (data != NULL && symndx < nbuckets) |
| 761 | { |
| 762 | const Elf32_Word *const buckets = data->d_buf; |
| 763 | Elf32_Word maxndx = symndx; |
| 764 | for (Elf32_Word bucket = 0; bucket < nbuckets; ++bucket) |
| 765 | if (buckets[bucket] > maxndx) |
| 766 | maxndx = buckets[bucket]; |
| 767 | |
| 768 | GElf_Off hasharr_at = (buckets_at |
| 769 | + nbuckets * sizeof (Elf32_Word)); |
| 770 | hasharr_at += (maxndx - symndx) * sizeof (Elf32_Word); |
| 771 | do |
| 772 | { |
| 773 | data = elf_getdata_rawchunk (mod->main.elf, |
| 774 | hasharr_at, |
| 775 | sizeof (Elf32_Word), |
| 776 | ELF_T_WORD); |
| 777 | if (data != NULL |
| 778 | && (*(const Elf32_Word *) data->d_buf & 1u)) |
| 779 | { |
| 780 | mod->syments = maxndx + 1; |
| 781 | break; |
| 782 | } |
| 783 | ++maxndx; |
| 784 | hasharr_at += sizeof (Elf32_Word); |
| 785 | } |
| 786 | while (data != NULL); |
| 787 | } |
| 788 | } |
| 789 | } |
| 790 | if (offs[i_strtab] > offs[i_symtab] && mod->syments == 0) |
| 791 | mod->syments = ((offs[i_strtab] - offs[i_symtab]) |
| 792 | / gelf_fsize (mod->main.elf, |
| 793 | ELF_T_SYM, 1, EV_CURRENT)); |
| 794 | |
| 795 | if (mod->syments > 0) |
| 796 | { |
| 797 | mod->symdata = elf_getdata_rawchunk (mod->main.elf, |
| 798 | offs[i_symtab], |
| 799 | gelf_fsize (mod->main.elf, |
| 800 | ELF_T_SYM, |
| 801 | mod->syments, |
| 802 | EV_CURRENT), |
| 803 | ELF_T_SYM); |
| 804 | if (mod->symdata != NULL) |
| 805 | { |
| 806 | mod->symstrdata = elf_getdata_rawchunk (mod->main.elf, |
| 807 | offs[i_strtab], |
| 808 | strsz, |
| 809 | ELF_T_BYTE); |
| 810 | if (mod->symstrdata == NULL) |
| 811 | mod->symdata = NULL; |
| 812 | } |
| 813 | if (mod->symdata == NULL) |
| 814 | mod->symerr = DWFL_E (LIBELF, elf_errno ()); |
| 815 | else |
| 816 | { |
| 817 | mod->symfile = &mod->main; |
| 818 | mod->symerr = DWFL_E_NOERROR; |
| 819 | } |
| 820 | } |
| 821 | } |
| 822 | |
| 823 | /* Try to find a dynamic symbol table via phdrs. */ |
| 824 | static void |
| 825 | find_dynsym (Dwfl_Module *mod) |
| 826 | { |
| 827 | GElf_Ehdr ehdr_mem; |
| 828 | GElf_Ehdr *ehdr = gelf_getehdr (mod->main.elf, &ehdr_mem); |
| 829 | |
| 830 | size_t phnum; |
| 831 | if (unlikely (elf_getphdrnum (mod->main.elf, &phnum) != 0)) |
| 832 | return; |
| 833 | |
| 834 | for (size_t i = 0; i < phnum; ++i) |
| 835 | { |
| 836 | GElf_Phdr phdr_mem; |
| 837 | GElf_Phdr *phdr = gelf_getphdr (mod->main.elf, i, &phdr_mem); |
| 838 | if (phdr == NULL) |
| 839 | break; |
| 840 | |
| 841 | if (phdr->p_type == PT_DYNAMIC) |
| 842 | { |
| 843 | /* Examine the dynamic section for the pointers we need. */ |
| 844 | |
| 845 | Elf_Data *data = elf_getdata_rawchunk (mod->main.elf, |
| 846 | phdr->p_offset, phdr->p_filesz, |
| 847 | ELF_T_DYN); |
| 848 | if (data == NULL) |
| 849 | continue; |
| 850 | |
| 851 | GElf_Addr addrs[i_max] = { 0, }; |
| 852 | GElf_Xword strsz = 0; |
| 853 | size_t n = data->d_size / gelf_fsize (mod->main.elf, |
| 854 | ELF_T_DYN, 1, EV_CURRENT); |
| 855 | for (size_t j = 0; j < n; ++j) |
| 856 | { |
| 857 | GElf_Dyn dyn_mem; |
| 858 | GElf_Dyn *dyn = gelf_getdyn (data, j, &dyn_mem); |
| 859 | if (dyn != NULL) |
| 860 | switch (dyn->d_tag) |
| 861 | { |
| 862 | case DT_SYMTAB: |
| 863 | addrs[i_symtab] = dyn->d_un.d_ptr; |
| 864 | continue; |
| 865 | |
| 866 | case DT_HASH: |
| 867 | addrs[i_hash] = dyn->d_un.d_ptr; |
| 868 | continue; |
| 869 | |
| 870 | case DT_GNU_HASH: |
| 871 | addrs[i_gnu_hash] = dyn->d_un.d_ptr; |
| 872 | continue; |
| 873 | |
| 874 | case DT_STRTAB: |
| 875 | addrs[i_strtab] = dyn->d_un.d_ptr; |
| 876 | continue; |
| 877 | |
| 878 | case DT_STRSZ: |
| 879 | strsz = dyn->d_un.d_val; |
| 880 | continue; |
| 881 | |
| 882 | default: |
| 883 | continue; |
| 884 | |
| 885 | case DT_NULL: |
| 886 | break; |
| 887 | } |
| 888 | break; |
| 889 | } |
| 890 | |
| 891 | /* First try unadjusted, like ELF files from disk, vdso. |
| 892 | Then try for already adjusted dynamic section, like ELF |
| 893 | from remote memory. */ |
| 894 | translate_offs (0, mod, phnum, addrs, strsz, ehdr); |
| 895 | if (mod->symfile == NULL) |
| 896 | translate_offs (mod->main_bias, mod, phnum, addrs, strsz, ehdr); |
| 897 | |
| 898 | return; |
| 899 | } |
| 900 | } |
| 901 | } |
| 902 | |
| 903 | |
| 904 | #if USE_LZMA |
| 905 | /* Try to find the offset between the main file and .gnu_debugdata. */ |
| 906 | static bool |
| 907 | find_aux_address_sync (Dwfl_Module *mod) |
| 908 | { |
| 909 | /* Don't trust the phdrs in the minisymtab elf file to be setup correctly. |
| 910 | The address_sync is equal to the main file it is embedded in at first. */ |
| 911 | mod->aux_sym.address_sync = mod->main.address_sync; |
| 912 | |
| 913 | /* Adjust address_sync for the difference in entry addresses, attempting to |
| 914 | account for ELF relocation changes after aux was split. */ |
| 915 | GElf_Ehdr ehdr_main, ehdr_aux; |
| 916 | if (unlikely (gelf_getehdr (mod->main.elf, &ehdr_main) == NULL) |
| 917 | || unlikely (gelf_getehdr (mod->aux_sym.elf, &ehdr_aux) == NULL)) |
| 918 | return false; |
| 919 | mod->aux_sym.address_sync += ehdr_aux.e_entry - ehdr_main.e_entry; |
| 920 | |
| 921 | /* The shdrs are setup OK to make find_prelink_address_sync () do the right |
| 922 | thing, which is possibly more reliable, but it needs .gnu.prelink_undo. */ |
| 923 | if (mod->aux_sym.address_sync != 0) |
| 924 | return find_prelink_address_sync (mod, &mod->aux_sym) == DWFL_E_NOERROR; |
| 925 | |
| 926 | return true; |
| 927 | } |
| 928 | #endif |
| 929 | |
| 930 | /* Try to find the auxiliary symbol table embedded in the main elf file |
| 931 | section .gnu_debugdata. Only matters if the symbol information comes |
| 932 | from the main file dynsym. No harm done if not found. */ |
| 933 | static void |
| 934 | find_aux_sym (Dwfl_Module *mod __attribute__ ((unused)), |
| 935 | Elf_Scn **aux_symscn __attribute__ ((unused)), |
| 936 | Elf_Scn **aux_xndxscn __attribute__ ((unused)), |
| 937 | GElf_Word *aux_strshndx __attribute__ ((unused))) |
| 938 | { |
| 939 | /* Since a .gnu_debugdata section is compressed using lzma don't do |
| 940 | anything unless we have support for that. */ |
| 941 | #if USE_LZMA |
| 942 | Elf *elf = mod->main.elf; |
| 943 | |
| 944 | size_t shstrndx; |
| 945 | if (elf_getshdrstrndx (elf, &shstrndx) < 0) |
| 946 | return; |
| 947 | |
| 948 | Elf_Scn *scn = NULL; |
| 949 | while ((scn = elf_nextscn (elf, scn)) != NULL) |
| 950 | { |
| 951 | GElf_Shdr shdr_mem; |
| 952 | GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_mem); |
| 953 | if (shdr == NULL) |
| 954 | return; |
| 955 | |
| 956 | const char *name = elf_strptr (elf, shstrndx, shdr->sh_name); |
| 957 | if (name == NULL) |
| 958 | return; |
| 959 | |
| 960 | if (!strcmp (name, ".gnu_debugdata")) |
| 961 | break; |
| 962 | } |
| 963 | |
| 964 | if (scn == NULL) |
| 965 | return; |
| 966 | |
| 967 | /* Found the .gnu_debugdata section. Uncompress the lzma image and |
| 968 | turn it into an ELF image. */ |
| 969 | Elf_Data *rawdata = elf_rawdata (scn, NULL); |
| 970 | if (rawdata == NULL) |
| 971 | return; |
| 972 | |
| 973 | Dwfl_Error error; |
| 974 | void *buffer = NULL; |
| 975 | size_t size = 0; |
| 976 | error = __libdw_unlzma (-1, 0, rawdata->d_buf, rawdata->d_size, |
| 977 | &buffer, &size); |
| 978 | if (error == DWFL_E_NOERROR) |
| 979 | { |
| 980 | if (unlikely (size == 0)) |
| 981 | free (buffer); |
| 982 | else |
| 983 | { |
| 984 | mod->aux_sym.elf = elf_memory (buffer, size); |
| 985 | if (mod->aux_sym.elf == NULL) |
| 986 | free (buffer); |
| 987 | else |
| 988 | { |
| 989 | mod->aux_sym.fd = -1; |
| 990 | mod->aux_sym.elf->flags |= ELF_F_MALLOCED; |
| 991 | if (open_elf (mod, &mod->aux_sym) != DWFL_E_NOERROR) |
| 992 | return; |
| 993 | if (! find_aux_address_sync (mod)) |
| 994 | { |
| 995 | elf_end (mod->aux_sym.elf); |
| 996 | mod->aux_sym.elf = NULL; |
| 997 | return; |
| 998 | } |
| 999 | |
| 1000 | /* So far, so good. Get minisymtab table data and cache it. */ |
| 1001 | bool minisymtab = false; |
| 1002 | scn = NULL; |
| 1003 | while ((scn = elf_nextscn (mod->aux_sym.elf, scn)) != NULL) |
| 1004 | { |
| 1005 | GElf_Shdr shdr_mem, *shdr = gelf_getshdr (scn, &shdr_mem); |
| 1006 | if (shdr != NULL) |
| 1007 | switch (shdr->sh_type) |
| 1008 | { |
| 1009 | case SHT_SYMTAB: |
| 1010 | minisymtab = true; |
| 1011 | *aux_symscn = scn; |
| 1012 | *aux_strshndx = shdr->sh_link; |
| 1013 | mod->aux_syments = shdr->sh_size / shdr->sh_entsize; |
| 1014 | mod->aux_first_global = shdr->sh_info; |
| 1015 | if (*aux_xndxscn != NULL) |
| 1016 | return; |
| 1017 | break; |
| 1018 | |
| 1019 | case SHT_SYMTAB_SHNDX: |
| 1020 | *aux_xndxscn = scn; |
| 1021 | if (minisymtab) |
| 1022 | return; |
| 1023 | break; |
| 1024 | |
| 1025 | default: |
| 1026 | break; |
| 1027 | } |
| 1028 | } |
| 1029 | |
| 1030 | if (minisymtab) |
| 1031 | /* We found one, though no SHT_SYMTAB_SHNDX to go with it. */ |
| 1032 | return; |
| 1033 | |
| 1034 | /* We found no SHT_SYMTAB, so everything else is bogus. */ |
| 1035 | *aux_xndxscn = NULL; |
| 1036 | *aux_strshndx = 0; |
| 1037 | mod->aux_syments = 0; |
| 1038 | elf_end (mod->aux_sym.elf); |
| 1039 | mod->aux_sym.elf = NULL; |
| 1040 | return; |
| 1041 | } |
| 1042 | } |
| 1043 | } |
| 1044 | else |
| 1045 | free (buffer); |
| 1046 | #endif |
| 1047 | } |
| 1048 | |
| 1049 | /* Try to find a symbol table in either MOD->main.elf or MOD->debug.elf. */ |
| 1050 | static void |
| 1051 | find_symtab (Dwfl_Module *mod) |
| 1052 | { |
| 1053 | if (mod->symdata != NULL || mod->aux_symdata != NULL /* Already done. */ |
| 1054 | || mod->symerr != DWFL_E_NOERROR) /* Cached previous failure. */ |
| 1055 | return; |
| 1056 | |
| 1057 | __libdwfl_getelf (mod); |
| 1058 | mod->symerr = mod->elferr; |
| 1059 | if (mod->symerr != DWFL_E_NOERROR) |
| 1060 | return; |
| 1061 | |
| 1062 | /* First see if the main ELF file has the debugging information. */ |
| 1063 | Elf_Scn *symscn = NULL, *xndxscn = NULL; |
| 1064 | Elf_Scn *aux_symscn = NULL, *aux_xndxscn = NULL; |
| 1065 | GElf_Word strshndx, aux_strshndx = 0; |
| 1066 | mod->symerr = load_symtab (&mod->main, &mod->symfile, &symscn, |
| 1067 | &xndxscn, &mod->syments, &mod->first_global, |
| 1068 | &strshndx); |
| 1069 | switch (mod->symerr) |
| 1070 | { |
| 1071 | default: |
| 1072 | return; |
| 1073 | |
| 1074 | case DWFL_E_NOERROR: |
| 1075 | break; |
| 1076 | |
| 1077 | case DWFL_E_NO_SYMTAB: |
| 1078 | /* Now we have to look for a separate debuginfo file. */ |
| 1079 | mod->symerr = find_debuginfo (mod); |
| 1080 | switch (mod->symerr) |
| 1081 | { |
| 1082 | default: |
| 1083 | return; |
| 1084 | |
| 1085 | case DWFL_E_NOERROR: |
| 1086 | mod->symerr = load_symtab (&mod->debug, &mod->symfile, &symscn, |
| 1087 | &xndxscn, &mod->syments, |
| 1088 | &mod->first_global, &strshndx); |
| 1089 | break; |
| 1090 | |
| 1091 | case DWFL_E_CB: /* The find_debuginfo hook failed. */ |
| 1092 | mod->symerr = DWFL_E_NO_SYMTAB; |
| 1093 | break; |
| 1094 | } |
| 1095 | |
| 1096 | switch (mod->symerr) |
| 1097 | { |
| 1098 | default: |
| 1099 | return; |
| 1100 | |
| 1101 | case DWFL_E_NOERROR: |
| 1102 | break; |
| 1103 | |
| 1104 | case DWFL_E_NO_SYMTAB: |
| 1105 | /* There might be an auxiliary table. */ |
| 1106 | find_aux_sym (mod, &aux_symscn, &aux_xndxscn, &aux_strshndx); |
| 1107 | |
| 1108 | if (symscn != NULL) |
| 1109 | { |
| 1110 | /* We still have the dynamic symbol table. */ |
| 1111 | mod->symerr = DWFL_E_NOERROR; |
| 1112 | break; |
| 1113 | } |
| 1114 | |
| 1115 | if (aux_symscn != NULL) |
| 1116 | { |
| 1117 | /* We still have the auxiliary symbol table. */ |
| 1118 | mod->symerr = DWFL_E_NOERROR; |
| 1119 | goto aux_cache; |
| 1120 | } |
| 1121 | |
| 1122 | /* Last ditch, look for dynamic symbols without section headers. */ |
| 1123 | find_dynsym (mod); |
| 1124 | return; |
| 1125 | } |
| 1126 | break; |
| 1127 | } |
| 1128 | |
| 1129 | /* This does some sanity checks on the string table section. */ |
| 1130 | if (elf_strptr (mod->symfile->elf, strshndx, 0) == NULL) |
| 1131 | { |
| 1132 | elferr: |
| 1133 | mod->symdata = NULL; |
| 1134 | mod->syments = 0; |
| 1135 | mod->first_global = 0; |
| 1136 | mod->symerr = DWFL_E (LIBELF, elf_errno ()); |
| 1137 | goto aux_cleanup; /* This cleans up some more and tries find_dynsym. */ |
| 1138 | } |
| 1139 | |
| 1140 | /* Cache the data; MOD->syments and MOD->first_global were set |
| 1141 | above. If any of the sections is compressed, uncompress it |
| 1142 | first. Only the string data setion could theoretically be |
| 1143 | compressed GNU style (as .zdebug_str). Everything else only ELF |
| 1144 | gabi style (SHF_COMPRESSED). */ |
| 1145 | |
| 1146 | Elf_Scn *symstrscn = elf_getscn (mod->symfile->elf, strshndx); |
| 1147 | if (symstrscn == NULL) |
| 1148 | goto elferr; |
| 1149 | |
| 1150 | GElf_Shdr shdr_mem; |
| 1151 | GElf_Shdr *shdr = gelf_getshdr (symstrscn, &shdr_mem); |
| 1152 | if (shdr == NULL) |
| 1153 | goto elferr; |
| 1154 | |
| 1155 | size_t shstrndx; |
| 1156 | if (elf_getshdrstrndx (mod->symfile->elf, &shstrndx) < 0) |
| 1157 | goto elferr; |
| 1158 | |
| 1159 | const char *sname = elf_strptr (mod->symfile->elf, shstrndx, shdr->sh_name); |
| 1160 | if (sname == NULL) |
| 1161 | goto elferr; |
| 1162 | |
| 1163 | if (strncmp (sname, ".zdebug", strlen (".zdebug")) == 0) |
| 1164 | /* Try to uncompress, but it might already have been, an error |
| 1165 | might just indicate, already uncompressed. */ |
| 1166 | elf_compress_gnu (symstrscn, 0, 0); |
| 1167 | |
| 1168 | if ((shdr->sh_flags & SHF_COMPRESSED) != 0) |
| 1169 | if (elf_compress (symstrscn, 0, 0) < 0) |
| 1170 | goto elferr; |
| 1171 | |
| 1172 | mod->symstrdata = elf_getdata (symstrscn, NULL); |
| 1173 | if (mod->symstrdata == NULL || mod->symstrdata->d_buf == NULL) |
| 1174 | goto elferr; |
| 1175 | |
| 1176 | if (xndxscn == NULL) |
| 1177 | mod->symxndxdata = NULL; |
| 1178 | else |
| 1179 | { |
| 1180 | shdr = gelf_getshdr (xndxscn, &shdr_mem); |
| 1181 | if (shdr == NULL) |
| 1182 | goto elferr; |
| 1183 | |
| 1184 | if ((shdr->sh_flags & SHF_COMPRESSED) != 0) |
| 1185 | if (elf_compress (xndxscn, 0, 0) < 0) |
| 1186 | goto elferr; |
| 1187 | |
| 1188 | mod->symxndxdata = elf_getdata (xndxscn, NULL); |
| 1189 | if (mod->symxndxdata == NULL || mod->symxndxdata->d_buf == NULL) |
| 1190 | goto elferr; |
| 1191 | } |
| 1192 | |
| 1193 | shdr = gelf_getshdr (symscn, &shdr_mem); |
| 1194 | if (shdr == NULL) |
| 1195 | goto elferr; |
| 1196 | |
| 1197 | if ((shdr->sh_flags & SHF_COMPRESSED) != 0) |
| 1198 | if (elf_compress (symscn, 0, 0) < 0) |
| 1199 | goto elferr; |
| 1200 | |
| 1201 | mod->symdata = elf_getdata (symscn, NULL); |
| 1202 | if (mod->symdata == NULL || mod->symdata->d_buf == NULL) |
| 1203 | goto elferr; |
| 1204 | |
| 1205 | // Sanity check number of symbols. |
| 1206 | shdr = gelf_getshdr (symscn, &shdr_mem); |
| 1207 | if (shdr == NULL || shdr->sh_entsize == 0 |
| 1208 | || mod->syments > mod->symdata->d_size / shdr->sh_entsize |
| 1209 | || (size_t) mod->first_global > mod->syments) |
| 1210 | goto elferr; |
| 1211 | |
| 1212 | /* Cache any auxiliary symbol info, when it fails, just ignore aux_sym. */ |
| 1213 | if (aux_symscn != NULL) |
| 1214 | { |
| 1215 | aux_cache: |
| 1216 | /* This does some sanity checks on the string table section. */ |
| 1217 | if (elf_strptr (mod->aux_sym.elf, aux_strshndx, 0) == NULL) |
| 1218 | { |
| 1219 | aux_cleanup: |
| 1220 | mod->aux_syments = 0; |
| 1221 | elf_end (mod->aux_sym.elf); |
| 1222 | mod->aux_sym.elf = NULL; |
| 1223 | /* We thought we had something through shdrs, but it failed... |
| 1224 | Last ditch, look for dynamic symbols without section headers. */ |
| 1225 | find_dynsym (mod); |
| 1226 | return; |
| 1227 | } |
| 1228 | |
| 1229 | Elf_Scn *aux_strscn = elf_getscn (mod->aux_sym.elf, aux_strshndx); |
| 1230 | if (aux_strscn == NULL) |
| 1231 | goto elferr; |
| 1232 | |
| 1233 | shdr = gelf_getshdr (aux_strscn, &shdr_mem); |
| 1234 | if (shdr == NULL) |
| 1235 | goto elferr; |
| 1236 | |
| 1237 | size_t aux_shstrndx; |
| 1238 | if (elf_getshdrstrndx (mod->aux_sym.elf, &aux_shstrndx) < 0) |
| 1239 | goto elferr; |
| 1240 | |
| 1241 | sname = elf_strptr (mod->aux_sym.elf, aux_shstrndx, |
| 1242 | shdr->sh_name); |
| 1243 | if (sname == NULL) |
| 1244 | goto elferr; |
| 1245 | |
| 1246 | if (strncmp (sname, ".zdebug", strlen (".zdebug")) == 0) |
| 1247 | /* Try to uncompress, but it might already have been, an error |
| 1248 | might just indicate, already uncompressed. */ |
| 1249 | elf_compress_gnu (aux_strscn, 0, 0); |
| 1250 | |
| 1251 | if ((shdr->sh_flags & SHF_COMPRESSED) != 0) |
| 1252 | if (elf_compress (aux_strscn, 0, 0) < 0) |
| 1253 | goto elferr; |
| 1254 | |
| 1255 | mod->aux_symstrdata = elf_getdata (aux_strscn, NULL); |
| 1256 | if (mod->aux_symstrdata == NULL || mod->aux_symstrdata->d_buf == NULL) |
| 1257 | goto aux_cleanup; |
| 1258 | |
| 1259 | if (aux_xndxscn == NULL) |
| 1260 | mod->aux_symxndxdata = NULL; |
| 1261 | else |
| 1262 | { |
| 1263 | shdr = gelf_getshdr (aux_xndxscn, &shdr_mem); |
| 1264 | if (shdr == NULL) |
| 1265 | goto elferr; |
| 1266 | |
| 1267 | if ((shdr->sh_flags & SHF_COMPRESSED) != 0) |
| 1268 | if (elf_compress (aux_xndxscn, 0, 0) < 0) |
| 1269 | goto elferr; |
| 1270 | |
| 1271 | mod->aux_symxndxdata = elf_getdata (aux_xndxscn, NULL); |
| 1272 | if (mod->aux_symxndxdata == NULL |
| 1273 | || mod->aux_symxndxdata->d_buf == NULL) |
| 1274 | goto aux_cleanup; |
| 1275 | } |
| 1276 | |
| 1277 | shdr = gelf_getshdr (aux_symscn, &shdr_mem); |
| 1278 | if (shdr == NULL) |
| 1279 | goto elferr; |
| 1280 | |
| 1281 | if ((shdr->sh_flags & SHF_COMPRESSED) != 0) |
| 1282 | if (elf_compress (aux_symscn, 0, 0) < 0) |
| 1283 | goto elferr; |
| 1284 | |
| 1285 | mod->aux_symdata = elf_getdata (aux_symscn, NULL); |
| 1286 | if (mod->aux_symdata == NULL || mod->aux_symdata->d_buf == NULL) |
| 1287 | goto aux_cleanup; |
| 1288 | |
| 1289 | // Sanity check number of aux symbols. |
| 1290 | shdr = gelf_getshdr (aux_symscn, &shdr_mem); |
| 1291 | if (mod->aux_syments > mod->aux_symdata->d_size / shdr->sh_entsize |
| 1292 | || (size_t) mod->aux_first_global > mod->aux_syments) |
| 1293 | goto aux_cleanup; |
| 1294 | } |
| 1295 | } |
| 1296 | |
| 1297 | |
| 1298 | /* Try to open a libebl backend for MOD. */ |
| 1299 | Dwfl_Error |
| 1300 | internal_function |
| 1301 | __libdwfl_module_getebl (Dwfl_Module *mod) |
| 1302 | { |
| 1303 | if (mod->ebl == NULL) |
| 1304 | { |
| 1305 | __libdwfl_getelf (mod); |
| 1306 | if (mod->elferr != DWFL_E_NOERROR) |
| 1307 | return mod->elferr; |
| 1308 | |
| 1309 | mod->ebl = ebl_openbackend (mod->main.elf); |
| 1310 | if (mod->ebl == NULL) |
| 1311 | return DWFL_E_LIBEBL; |
| 1312 | } |
| 1313 | return DWFL_E_NOERROR; |
| 1314 | } |
| 1315 | |
| 1316 | /* Try to start up libdw on DEBUGFILE. */ |
| 1317 | static Dwfl_Error |
| 1318 | load_dw (Dwfl_Module *mod, struct dwfl_file *debugfile) |
| 1319 | { |
| 1320 | if (mod->e_type == ET_REL && !debugfile->relocated) |
| 1321 | { |
| 1322 | const Dwfl_Callbacks *const cb = mod->dwfl->callbacks; |
| 1323 | |
| 1324 | /* The debugging sections have to be relocated. */ |
| 1325 | if (cb->section_address == NULL) |
| 1326 | return DWFL_E_NOREL; |
| 1327 | |
| 1328 | Dwfl_Error error = __libdwfl_module_getebl (mod); |
| 1329 | if (error != DWFL_E_NOERROR) |
| 1330 | return error; |
| 1331 | |
| 1332 | find_symtab (mod); |
| 1333 | Dwfl_Error result = mod->symerr; |
| 1334 | if (result == DWFL_E_NOERROR) |
| 1335 | result = __libdwfl_relocate (mod, debugfile->elf, true); |
| 1336 | if (result != DWFL_E_NOERROR) |
| 1337 | return result; |
| 1338 | |
| 1339 | /* Don't keep the file descriptors around. */ |
| 1340 | if (mod->main.fd != -1 && elf_cntl (mod->main.elf, ELF_C_FDREAD) == 0) |
| 1341 | { |
| 1342 | close (mod->main.fd); |
| 1343 | mod->main.fd = -1; |
| 1344 | } |
| 1345 | if (debugfile->fd != -1 && elf_cntl (debugfile->elf, ELF_C_FDREAD) == 0) |
| 1346 | { |
| 1347 | close (debugfile->fd); |
| 1348 | debugfile->fd = -1; |
| 1349 | } |
| 1350 | } |
| 1351 | |
| 1352 | mod->dw = INTUSE(dwarf_begin_elf) (debugfile->elf, DWARF_C_READ, NULL); |
| 1353 | if (mod->dw == NULL) |
| 1354 | { |
| 1355 | int err = INTUSE(dwarf_errno) (); |
| 1356 | return err == DWARF_E_NO_DWARF ? DWFL_E_NO_DWARF : DWFL_E (LIBDW, err); |
| 1357 | } |
| 1358 | |
| 1359 | /* Until we have iterated through all CU's, we might do lazy lookups. */ |
| 1360 | mod->lazycu = 1; |
| 1361 | |
| 1362 | return DWFL_E_NOERROR; |
| 1363 | } |
| 1364 | |
| 1365 | /* Try to start up libdw on either the main file or the debuginfo file. */ |
| 1366 | static void |
| 1367 | find_dw (Dwfl_Module *mod) |
| 1368 | { |
| 1369 | if (mod->dw != NULL /* Already done. */ |
| 1370 | || mod->dwerr != DWFL_E_NOERROR) /* Cached previous failure. */ |
| 1371 | return; |
| 1372 | |
| 1373 | __libdwfl_getelf (mod); |
| 1374 | mod->dwerr = mod->elferr; |
| 1375 | if (mod->dwerr != DWFL_E_NOERROR) |
| 1376 | return; |
| 1377 | |
| 1378 | /* First see if the main ELF file has the debugging information. */ |
| 1379 | mod->dwerr = load_dw (mod, &mod->main); |
| 1380 | switch (mod->dwerr) |
| 1381 | { |
| 1382 | case DWFL_E_NOERROR: |
| 1383 | mod->debug.elf = mod->main.elf; |
| 1384 | mod->debug.address_sync = mod->main.address_sync; |
| 1385 | |
| 1386 | /* The Dwarf might need an alt debug file, find that now after |
| 1387 | everything about the debug file has been setup (the |
| 1388 | find_debuginfo callback might need it). */ |
| 1389 | find_debug_altlink (mod, mod->main.name); |
| 1390 | return; |
| 1391 | |
| 1392 | case DWFL_E_NO_DWARF: |
| 1393 | break; |
| 1394 | |
| 1395 | default: |
| 1396 | goto canonicalize; |
| 1397 | } |
| 1398 | |
| 1399 | /* Now we have to look for a separate debuginfo file. */ |
| 1400 | mod->dwerr = find_debuginfo (mod); |
| 1401 | switch (mod->dwerr) |
| 1402 | { |
| 1403 | case DWFL_E_NOERROR: |
| 1404 | mod->dwerr = load_dw (mod, &mod->debug); |
| 1405 | if (mod->dwerr == DWFL_E_NOERROR) |
| 1406 | { |
| 1407 | /* The Dwarf might need an alt debug file, find that now after |
| 1408 | everything about the debug file has been setup (the |
| 1409 | find_debuginfo callback might need it). */ |
| 1410 | find_debug_altlink (mod, mod->debug.name); |
| 1411 | return; |
| 1412 | } |
| 1413 | |
| 1414 | break; |
| 1415 | |
| 1416 | case DWFL_E_CB: /* The find_debuginfo hook failed. */ |
| 1417 | mod->dwerr = DWFL_E_NO_DWARF; |
| 1418 | return; |
| 1419 | |
| 1420 | default: |
| 1421 | break; |
| 1422 | } |
| 1423 | |
| 1424 | canonicalize: |
| 1425 | mod->dwerr = __libdwfl_canon_error (mod->dwerr); |
| 1426 | } |
| 1427 | |
| 1428 | Dwarf * |
| 1429 | dwfl_module_getdwarf (Dwfl_Module *mod, Dwarf_Addr *bias) |
| 1430 | { |
| 1431 | if (mod == NULL) |
| 1432 | return NULL; |
| 1433 | |
| 1434 | find_dw (mod); |
| 1435 | if (mod->dwerr == DWFL_E_NOERROR) |
| 1436 | { |
| 1437 | /* If dwfl_module_getelf was used previously, then partial apply |
| 1438 | relocation to miscellaneous sections in the debug file too. */ |
| 1439 | if (mod->e_type == ET_REL |
| 1440 | && mod->main.relocated && ! mod->debug.relocated) |
| 1441 | { |
| 1442 | mod->debug.relocated = true; |
| 1443 | if (mod->debug.elf != mod->main.elf) |
| 1444 | (void) __libdwfl_relocate (mod, mod->debug.elf, false); |
| 1445 | } |
| 1446 | |
| 1447 | *bias = dwfl_adjusted_dwarf_addr (mod, 0); |
| 1448 | return mod->dw; |
| 1449 | } |
| 1450 | |
| 1451 | __libdwfl_seterrno (mod->dwerr); |
| 1452 | return NULL; |
| 1453 | } |
| 1454 | INTDEF (dwfl_module_getdwarf) |
| 1455 | |
| 1456 | int |
| 1457 | dwfl_module_getsymtab (Dwfl_Module *mod) |
| 1458 | { |
| 1459 | if (mod == NULL) |
| 1460 | return -1; |
| 1461 | |
| 1462 | find_symtab (mod); |
| 1463 | if (mod->symerr == DWFL_E_NOERROR) |
| 1464 | /* We will skip the auxiliary zero entry if there is another one. */ |
| 1465 | return (mod->syments + mod->aux_syments |
| 1466 | - (mod->syments > 0 && mod->aux_syments > 0 ? 1 : 0)); |
| 1467 | |
| 1468 | __libdwfl_seterrno (mod->symerr); |
| 1469 | return -1; |
| 1470 | } |
| 1471 | INTDEF (dwfl_module_getsymtab) |
| 1472 | |
| 1473 | int |
| 1474 | dwfl_module_getsymtab_first_global (Dwfl_Module *mod) |
| 1475 | { |
| 1476 | if (mod == NULL) |
| 1477 | return -1; |
| 1478 | |
| 1479 | find_symtab (mod); |
| 1480 | if (mod->symerr == DWFL_E_NOERROR) |
| 1481 | { |
| 1482 | /* All local symbols should come before all global symbols. If |
| 1483 | we have an auxiliary table make sure all the main locals come |
| 1484 | first, then all aux locals, then all main globals and finally all |
| 1485 | aux globals. And skip the auxiliary table zero undefined |
| 1486 | entry. */ |
| 1487 | int skip_aux_zero = (mod->syments > 0 && mod->aux_syments > 0) ? 1 : 0; |
| 1488 | return mod->first_global + mod->aux_first_global - skip_aux_zero; |
| 1489 | } |
| 1490 | |
| 1491 | __libdwfl_seterrno (mod->symerr); |
| 1492 | return -1; |
| 1493 | } |
| 1494 | INTDEF (dwfl_module_getsymtab_first_global) |