Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 1 | /* Print the strings of printable characters in files. |
Ulrich Drepper | 82c345d | 2006-04-04 22:07:54 +0000 | [diff] [blame] | 2 | Copyright (C) 2005, 2006 Red Hat, Inc. |
Ulrich Drepper | 361df7d | 2006-04-04 21:38:57 +0000 | [diff] [blame] | 3 | This file is part of Red Hat elfutils. |
Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 4 | Written by Ulrich Drepper <drepper@redhat.com>, 2005. |
| 5 | |
Ulrich Drepper | 361df7d | 2006-04-04 21:38:57 +0000 | [diff] [blame] | 6 | Red Hat elfutils is free software; you can redistribute it and/or modify |
| 7 | it under the terms of the GNU General Public License as published by the |
| 8 | Free Software Foundation; version 2 of the License. |
Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 9 | |
Ulrich Drepper | 361df7d | 2006-04-04 21:38:57 +0000 | [diff] [blame] | 10 | Red Hat elfutils is distributed in the hope that it will be useful, but |
| 11 | WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | General Public License for more details. |
| 14 | |
| 15 | You should have received a copy of the GNU General Public License along |
| 16 | with Red Hat elfutils; if not, write to the Free Software Foundation, |
Ulrich Drepper | 1e9ef50 | 2006-04-04 22:29:06 +0000 | [diff] [blame] | 17 | Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301 USA. |
Ulrich Drepper | 361df7d | 2006-04-04 21:38:57 +0000 | [diff] [blame] | 18 | |
| 19 | Red Hat elfutils is an included package of the Open Invention Network. |
| 20 | An included package of the Open Invention Network is a package for which |
| 21 | Open Invention Network licensees cross-license their patents. No patent |
| 22 | license is granted, either expressly or impliedly, by designation as an |
| 23 | included package. Should you wish to participate in the Open Invention |
| 24 | Network licensing program, please visit www.openinventionnetwork.com |
| 25 | <http://www.openinventionnetwork.com>. */ |
Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 26 | |
| 27 | #ifdef HAVE_CONFIG_H |
| 28 | # include <config.h> |
| 29 | #endif |
| 30 | |
| 31 | #include <argp.h> |
| 32 | #include <assert.h> |
| 33 | #include <ctype.h> |
| 34 | #include <endian.h> |
| 35 | #include <errno.h> |
| 36 | #include <error.h> |
| 37 | #include <fcntl.h> |
| 38 | #include <gelf.h> |
| 39 | #include <inttypes.h> |
| 40 | #include <libintl.h> |
| 41 | #include <locale.h> |
| 42 | #include <stdbool.h> |
| 43 | #include <stdio.h> |
| 44 | #include <stdio_ext.h> |
| 45 | #include <stdlib.h> |
| 46 | #include <string.h> |
| 47 | #include <unistd.h> |
| 48 | #include <sys/mman.h> |
| 49 | #include <sys/param.h> |
| 50 | #include <sys/stat.h> |
| 51 | |
| 52 | #include <system.h> |
| 53 | |
| 54 | |
| 55 | /* Prototypes of local functions. */ |
| 56 | static int read_fd (int fd, const char *fname, off64_t fdlen); |
| 57 | static int read_elf (Elf *elf, int fd, const char *fname, off64_t fdlen); |
| 58 | |
| 59 | |
| 60 | /* Name and version of program. */ |
| 61 | static void print_version (FILE *stream, struct argp_state *state); |
| 62 | void (*argp_program_version_hook) (FILE *, struct argp_state *) = print_version; |
| 63 | |
| 64 | /* Bug report address. */ |
| 65 | const char *argp_program_bug_address = PACKAGE_BUGREPORT; |
| 66 | |
| 67 | /* Definitions of arguments for argp functions. */ |
| 68 | static const struct argp_option options[] = |
| 69 | { |
| 70 | { NULL, 0, NULL, 0, N_("Output Selection:"), 0 }, |
| 71 | { "all", 'a', NULL, 0, N_("Scan entire file, not only loaded sections"), 0 }, |
| 72 | { "bytes", 'n', "MIN-LEN", 0, |
| 73 | N_("Only NUL-terminated sequences of MIN-LEN characters or more are printed"), 0 }, |
| 74 | { "encoding", 'e', "SELECTOR", 0, N_("\ |
| 75 | Select character size and endianess: s = 7-bit, S = 8-bit, {b,l} = 16-bit, {B,L} = 32-bit"), |
| 76 | 0}, |
| 77 | { "print-file-name", 'f', NULL, 0, |
| 78 | N_("Print name of the file before each string."), 0 }, |
| 79 | { "radix", 't', "{o,d,x}", 0, |
| 80 | N_("Print location of the string in base 8, 10, or 16 respectively."), 0 }, |
| 81 | { NULL, 'o', NULL, 0, N_("Alias for --radix=o"), 0 }, |
| 82 | |
| 83 | { NULL, 0, NULL, 0, N_("Miscellaneous:"), 0 }, |
| 84 | { NULL, 0, NULL, 0, NULL, 0 } |
| 85 | }; |
| 86 | |
| 87 | /* Short description of program. */ |
| 88 | static const char doc[] = N_("\ |
| 89 | Print the strings of printable characters in files."); |
| 90 | |
| 91 | /* Strings for arguments in help texts. */ |
| 92 | static const char args_doc[] = N_("[FILE...]"); |
| 93 | |
| 94 | /* Prototype for option handler. */ |
| 95 | static error_t parse_opt (int key, char *arg, struct argp_state *state); |
| 96 | |
| 97 | /* Data structure to communicate with argp functions. */ |
| 98 | static struct argp argp = |
| 99 | { |
| 100 | options, parse_opt, args_doc, doc, NULL, NULL, NULL |
| 101 | }; |
| 102 | |
| 103 | |
| 104 | /* Global variables. */ |
| 105 | |
| 106 | /* True if whole file and not only loaded sections are looked at. */ |
| 107 | static bool entire_file; |
| 108 | |
| 109 | /* Minimum length of any sequence reported. */ |
| 110 | static size_t min_len = 4; |
| 111 | |
| 112 | /* Number of bytes per character. */ |
| 113 | static size_t bytes_per_char = 1; |
| 114 | |
| 115 | /* Minimum length of any sequence reported in bytes. */ |
| 116 | static size_t min_len_bytes; |
| 117 | |
| 118 | /* True if multibyte characters are in big-endian order. */ |
| 119 | static bool big_endian; |
| 120 | |
| 121 | /* True unless 7-bit ASCII are expected. */ |
| 122 | static bool char_7bit; |
| 123 | |
| 124 | /* True if file names should be printed before strings. */ |
| 125 | static bool print_file_name; |
| 126 | |
| 127 | /* Location print format string. */ |
| 128 | static const char *locfmt; |
| 129 | |
| 130 | /* Page size in use. */ |
| 131 | static size_t ps; |
| 132 | |
| 133 | |
| 134 | /* Mapped parts of the ELF file. */ |
| 135 | static unsigned char *elfmap; |
| 136 | static unsigned char *elfmap_base; |
| 137 | static size_t elfmap_size; |
| 138 | static off64_t elfmap_off; |
| 139 | |
| 140 | |
| 141 | int |
| 142 | main (int argc, char *argv[]) |
| 143 | { |
| 144 | /* We use no threads. */ |
| 145 | __fsetlocking (stdin, FSETLOCKING_BYCALLER); |
| 146 | __fsetlocking (stdout, FSETLOCKING_BYCALLER); |
| 147 | |
| 148 | /* Set locale. */ |
| 149 | (void) setlocale (LC_ALL, ""); |
| 150 | |
| 151 | /* Make sure the message catalog can be found. */ |
| 152 | (void) bindtextdomain (PACKAGE, LOCALEDIR); |
| 153 | |
| 154 | /* Initialize the message catalog. */ |
| 155 | (void) textdomain (PACKAGE); |
| 156 | |
| 157 | /* Parse and process arguments. */ |
| 158 | int remaining; |
| 159 | (void) argp_parse (&argp, argc, argv, 0, &remaining, NULL); |
| 160 | |
| 161 | /* Tell the library which version we are expecting. */ |
| 162 | elf_version (EV_CURRENT); |
| 163 | |
| 164 | /* Determine the page size. We will likely need it a couple of times. */ |
| 165 | ps = sysconf (_SC_PAGESIZE); |
| 166 | |
| 167 | struct stat64 st; |
| 168 | int result = 0; |
| 169 | if (remaining == argc) |
| 170 | /* We read from standard input. This we cannot do for a |
| 171 | structured file. */ |
| 172 | result = read_fd (STDOUT_FILENO, |
| 173 | print_file_name ? "{standard input}" : NULL, |
| 174 | fstat64 (STDOUT_FILENO, &st) == 0 |
| 175 | ? st.st_size : INT64_C (0x7fffffffffffffff)); |
| 176 | else |
| 177 | do |
| 178 | { |
| 179 | int fd = (strcmp (argv[remaining], "-") == 0 |
| 180 | ? STDIN_FILENO : open (argv[remaining], O_RDONLY)); |
| 181 | if (unlikely (fd == -1)) |
| 182 | { |
| 183 | error (0, errno, gettext ("cannot open '%s'"), argv[remaining]); |
| 184 | result = 1; |
| 185 | } |
| 186 | else |
| 187 | { |
| 188 | const char *fname = print_file_name ? argv[remaining] : NULL; |
| 189 | int fstat_fail = fstat64 (fd, &st); |
| 190 | off64_t fdlen = (fstat_fail |
| 191 | ? INT64_C (0x7fffffffffffffff) : st.st_size); |
| 192 | if (fdlen > (off64_t) min_len_bytes) |
| 193 | { |
| 194 | Elf *elf = NULL; |
| 195 | if (entire_file |
| 196 | || fstat_fail |
| 197 | || !S_ISREG (st.st_mode) |
| 198 | || (elf = elf_begin (fd, ELF_C_READ, NULL)) == NULL |
| 199 | || elf_kind (elf) != ELF_K_ELF) |
| 200 | result |= read_fd (fd, fname, fdlen); |
| 201 | else |
| 202 | result |= read_elf (elf, fd, fname, fdlen); |
| 203 | |
| 204 | /* This call will succeed even if ELF is NULL. */ |
| 205 | elf_end (elf); |
| 206 | } |
| 207 | |
| 208 | if (strcmp (argv[remaining], "-") != 0) |
| 209 | close (fd); |
| 210 | } |
| 211 | |
| 212 | if (elfmap != NULL && elfmap != MAP_FAILED) |
Ulrich Drepper | 1d93810 | 2005-09-02 19:51:21 +0000 | [diff] [blame] | 213 | munmap (elfmap, elfmap_size); |
| 214 | elfmap = NULL; |
Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 215 | } |
| 216 | while (++remaining < argc); |
| 217 | |
| 218 | return result; |
| 219 | } |
| 220 | |
| 221 | |
| 222 | /* Print the version information. */ |
| 223 | static void |
| 224 | print_version (FILE *stream, struct argp_state *state __attribute__ ((unused))) |
| 225 | { |
| 226 | fprintf (stream, "strings (%s) %s\n", PACKAGE_NAME, VERSION); |
| 227 | fprintf (stream, gettext ("\ |
| 228 | Copyright (C) %s Red Hat, Inc.\n\ |
| 229 | This is free software; see the source for copying conditions. There is NO\n\ |
| 230 | warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n\ |
Ulrich Drepper | 82c345d | 2006-04-04 22:07:54 +0000 | [diff] [blame] | 231 | "), "2006"); |
Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 232 | fprintf (stream, gettext ("Written by %s.\n"), "Ulrich Drepper"); |
| 233 | } |
| 234 | |
| 235 | |
| 236 | /* Handle program arguments. */ |
| 237 | static error_t |
| 238 | parse_opt (int key, char *arg, |
| 239 | struct argp_state *state __attribute__ ((unused))) |
| 240 | { |
| 241 | switch (key) |
| 242 | { |
| 243 | case 'a': |
| 244 | entire_file = true; |
| 245 | break; |
| 246 | |
| 247 | case 'e': |
| 248 | /* We expect a string of one character. */ |
| 249 | switch (arg[1] != '\0' ? '\0' : arg[0]) |
| 250 | { |
| 251 | case 's': |
| 252 | case 'S': |
| 253 | char_7bit = arg[0] == 's'; |
| 254 | bytes_per_char = 1; |
| 255 | break; |
| 256 | |
| 257 | case 'b': |
| 258 | case 'B': |
| 259 | big_endian = true; |
| 260 | /* FALLTHROUGH */ |
| 261 | |
| 262 | case 'l': |
| 263 | case 'L': |
| 264 | bytes_per_char = isupper (arg[0]) ? 4 : 2; |
| 265 | break; |
| 266 | |
| 267 | default: |
| 268 | error (0, 0, gettext ("invalid value '%s' for %s parameter"), |
| 269 | arg, "-e"); |
| 270 | argp_help (&argp, stderr, ARGP_HELP_SEE, "strings"); |
| 271 | return ARGP_ERR_UNKNOWN; |
| 272 | } |
| 273 | break; |
| 274 | |
| 275 | case 'f': |
| 276 | print_file_name = true; |
| 277 | break; |
| 278 | |
| 279 | case 'n': |
| 280 | min_len = atoi (arg); |
| 281 | break; |
| 282 | |
| 283 | case 'o': |
| 284 | goto octfmt; |
| 285 | |
| 286 | case 't': |
| 287 | switch (arg[0]) |
| 288 | { |
| 289 | case 'd': |
| 290 | locfmt = "%7" PRId64 " "; |
| 291 | break; |
| 292 | |
| 293 | case 'o': |
| 294 | octfmt: |
| 295 | locfmt = "%7" PRIo64 " "; |
| 296 | break; |
| 297 | |
| 298 | case 'x': |
| 299 | locfmt = "%7" PRIx64 " "; |
| 300 | break; |
| 301 | |
| 302 | default: |
| 303 | error (0, 0, gettext ("invalid value '%s' for %s parameter"), |
| 304 | arg, "-t"); |
| 305 | argp_help (&argp, stderr, ARGP_HELP_SEE, "strings"); |
| 306 | return ARGP_ERR_UNKNOWN; |
| 307 | } |
| 308 | break; |
| 309 | |
| 310 | case ARGP_KEY_FINI: |
| 311 | /* Compute the length in bytes of any match. */ |
| 312 | if (min_len <= 0 || min_len > INT_MAX / bytes_per_char) |
| 313 | error (EXIT_FAILURE, 0, |
| 314 | gettext ("invalid minimum length of matched string size")); |
| 315 | min_len_bytes = min_len * bytes_per_char; |
| 316 | break; |
| 317 | |
| 318 | default: |
| 319 | return ARGP_ERR_UNKNOWN; |
| 320 | } |
| 321 | return 0; |
| 322 | } |
| 323 | |
| 324 | |
| 325 | static void |
| 326 | process_chunk_mb (const char *fname, const unsigned char *buf, off64_t to, |
| 327 | size_t len, char **unprinted) |
| 328 | { |
| 329 | size_t curlen = *unprinted == NULL ? 0 : strlen (*unprinted); |
| 330 | const unsigned char *start = buf; |
| 331 | while (len >= bytes_per_char) |
| 332 | { |
| 333 | uint32_t ch; |
| 334 | |
| 335 | if (bytes_per_char == 2) |
| 336 | { |
| 337 | if (big_endian) |
| 338 | ch = buf[0] << 8 | buf[1]; |
| 339 | else |
| 340 | ch = buf[1] << 8 | buf[0]; |
| 341 | } |
| 342 | else |
| 343 | { |
| 344 | if (big_endian) |
| 345 | ch = buf[0] << 24 | buf[1] << 16 | buf[2] << 8 | buf[3]; |
| 346 | else |
| 347 | ch = buf[3] << 24 | buf[2] << 16 | buf[1] << 8 | buf[0]; |
| 348 | } |
| 349 | |
| 350 | if (ch <= 255 && (isprint (ch) || ch == '\t')) |
| 351 | { |
| 352 | ++buf; |
| 353 | ++curlen; |
| 354 | } |
| 355 | else |
| 356 | { |
| 357 | if (curlen >= min_len) |
| 358 | { |
| 359 | /* We found a match. */ |
| 360 | if (unlikely (fname != NULL)) |
| 361 | { |
| 362 | fputs_unlocked (fname, stdout); |
| 363 | fputs_unlocked (": ", stdout); |
| 364 | } |
| 365 | |
| 366 | if (unlikely (locfmt != NULL)) |
| 367 | printf (locfmt, (int64_t) to - len - (buf - start)); |
| 368 | |
| 369 | if (unlikely (*unprinted != NULL)) |
| 370 | { |
| 371 | fputs_unlocked (*unprinted, stdout); |
| 372 | free (*unprinted); |
| 373 | *unprinted = NULL; |
| 374 | } |
| 375 | |
| 376 | /* There is no sane way of printing the string. If we |
| 377 | assume the file data is encoded in UCS-2/UTF-16 or |
| 378 | UCS-4/UTF-32 respectively we could covert the string. |
| 379 | But there is no such guarantee. */ |
| 380 | fwrite_unlocked (start, 1, buf - start, stdout); |
| 381 | putc_unlocked ('\n', stdout); |
| 382 | } |
| 383 | |
| 384 | start = ++buf; |
| 385 | curlen = 0; |
| 386 | |
| 387 | if (len <= min_len) |
| 388 | break; |
| 389 | } |
| 390 | |
| 391 | --len; |
| 392 | } |
| 393 | |
| 394 | if (curlen != 0) |
| 395 | *unprinted = xstrndup ((const char *) start, curlen); |
| 396 | } |
| 397 | |
| 398 | |
| 399 | static void |
| 400 | process_chunk (const char *fname, const unsigned char *buf, off64_t to, |
| 401 | size_t len, char **unprinted) |
| 402 | { |
| 403 | /* We are not going to slow the check down for the 2- and 4-byte |
| 404 | encodings. Handle them special. */ |
| 405 | if (unlikely (bytes_per_char != 1)) |
| 406 | { |
| 407 | process_chunk_mb (fname, buf, to, len, unprinted); |
| 408 | return; |
| 409 | } |
| 410 | |
| 411 | size_t curlen = *unprinted == NULL ? 0 : strlen (*unprinted); |
| 412 | const unsigned char *start = buf; |
| 413 | while (len > 0) |
| 414 | { |
| 415 | if ((isprint (*buf) || *buf == '\t') && (! char_7bit || *buf <= 127)) |
| 416 | { |
| 417 | ++buf; |
| 418 | ++curlen; |
| 419 | } |
| 420 | else |
| 421 | { |
| 422 | if (curlen >= min_len) |
| 423 | { |
| 424 | /* We found a match. */ |
| 425 | if (unlikely (fname != NULL)) |
| 426 | { |
| 427 | fputs_unlocked (fname, stdout); |
| 428 | fputs_unlocked (": ", stdout); |
| 429 | } |
| 430 | |
| 431 | if (unlikely (locfmt != NULL)) |
| 432 | printf (locfmt, (int64_t) to - len - (buf - start)); |
| 433 | |
| 434 | if (unlikely (*unprinted != NULL)) |
| 435 | { |
| 436 | fputs_unlocked (*unprinted, stdout); |
| 437 | free (*unprinted); |
| 438 | *unprinted = NULL; |
| 439 | } |
| 440 | fwrite_unlocked (start, 1, buf - start, stdout); |
| 441 | putc_unlocked ('\n', stdout); |
| 442 | } |
| 443 | |
| 444 | start = ++buf; |
| 445 | curlen = 0; |
| 446 | |
| 447 | if (len <= min_len) |
| 448 | break; |
| 449 | } |
| 450 | |
| 451 | --len; |
| 452 | } |
| 453 | |
| 454 | if (curlen != 0) |
| 455 | *unprinted = xstrndup ((const char *) start, curlen); |
| 456 | } |
| 457 | |
| 458 | |
| 459 | /* Map a file in as large chunks as possible. */ |
| 460 | static void * |
| 461 | map_file (int fd, off64_t start_off, off64_t fdlen, size_t *map_sizep) |
| 462 | { |
Ulrich Drepper | 618795f | 2005-09-02 20:01:37 +0000 | [diff] [blame] | 463 | #if _MUDFLAP |
| 464 | (void) fd; |
| 465 | (void) start_off; |
| 466 | (void) fdlen; |
| 467 | (void) map_sizep; |
| 468 | return MAP_FAILED; |
| 469 | #else |
Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 470 | /* Maximum size we mmap. We use an #ifdef to avoid overflows on |
| 471 | 32-bit machines. 64-bit machines these days do not have usable |
| 472 | address spaces larger than about 43 bits. Not that any file |
| 473 | should be that large. */ |
Ulrich Drepper | 618795f | 2005-09-02 20:01:37 +0000 | [diff] [blame] | 474 | # if SIZE_MAX > 0xffffffff |
Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 475 | const size_t mmap_max = 0x4000000000lu; |
Ulrich Drepper | 618795f | 2005-09-02 20:01:37 +0000 | [diff] [blame] | 476 | # else |
Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 477 | const size_t mmap_max = 0x40000000lu; |
Ulrich Drepper | 618795f | 2005-09-02 20:01:37 +0000 | [diff] [blame] | 478 | # endif |
Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 479 | |
| 480 | /* Try to mmap the file. */ |
| 481 | size_t map_size = MIN ((off64_t) mmap_max, fdlen); |
| 482 | const size_t map_size_min = MAX (MAX (SIZE_MAX / 16, 2 * ps), |
| 483 | roundup (2 * min_len_bytes + 1, ps)); |
| 484 | void *mem; |
| 485 | while (1) |
| 486 | { |
| 487 | /* We map the memory for reading only here. Since we will |
| 488 | always look at every byte of the file it makes sense to |
| 489 | use MAP_POPULATE. */ |
| 490 | mem = mmap64 (NULL, map_size, PROT_READ, MAP_PRIVATE | MAP_POPULATE, |
| 491 | fd, start_off); |
| 492 | if (mem != MAP_FAILED) |
| 493 | { |
| 494 | /* We will go through the mapping sequentially. */ |
| 495 | (void) posix_madvise (mem, map_size, POSIX_MADV_SEQUENTIAL); |
| 496 | break; |
| 497 | } |
| 498 | if (errno != EINVAL && errno != ENOMEM) |
| 499 | /* This is an error other than the lack of address space. */ |
| 500 | break; |
| 501 | |
| 502 | /* Maybe the size of the mapping is too big. Try again. */ |
| 503 | map_size /= 2; |
| 504 | if (map_size < map_size_min) |
| 505 | /* That size should have fit. */ |
| 506 | break; |
| 507 | } |
| 508 | |
| 509 | *map_sizep = map_size; |
| 510 | return mem; |
Ulrich Drepper | 618795f | 2005-09-02 20:01:37 +0000 | [diff] [blame] | 511 | #endif |
Ulrich Drepper | fbe998a | 2005-08-29 16:27:10 +0000 | [diff] [blame] | 512 | } |
| 513 | |
| 514 | |
| 515 | /* Read the file without mapping. */ |
| 516 | static int |
| 517 | read_block_no_mmap (int fd, const char *fname, off64_t from, off64_t fdlen) |
| 518 | { |
| 519 | char *unprinted = NULL; |
| 520 | #define CHUNKSIZE 65536 |
| 521 | unsigned char *buf = xmalloc (CHUNKSIZE + min_len_bytes |
| 522 | + bytes_per_char - 1); |
| 523 | size_t ntrailer = 0; |
| 524 | int result = 0; |
| 525 | while (fdlen > 0) |
| 526 | { |
| 527 | ssize_t n = TEMP_FAILURE_RETRY (read (fd, buf + ntrailer, |
| 528 | MIN (fdlen, CHUNKSIZE))); |
| 529 | if (n == 0) |
| 530 | { |
| 531 | /* There are less than MIN_LEN+1 bytes left so there cannot be |
| 532 | another match. */ |
| 533 | assert (unprinted == NULL || ntrailer == 0); |
| 534 | break; |
| 535 | } |
| 536 | if (unlikely (n < 0)) |
| 537 | { |
| 538 | /* Something went wrong. */ |
| 539 | result = 1; |
| 540 | break; |
| 541 | } |
| 542 | |
| 543 | /* Account for the number of bytes read in this round. */ |
| 544 | fdlen -= n; |
| 545 | |
| 546 | /* Do not use the signed N value. Note that the addition cannot |
| 547 | overflow. */ |
| 548 | size_t nb = (size_t) n + ntrailer; |
| 549 | if (nb >= min_len_bytes) |
| 550 | { |
| 551 | /* We only use complete charactesr. */ |
| 552 | nb &= ~(bytes_per_char - 1); |
| 553 | |
| 554 | process_chunk (fname, buf, from + nb, nb, &unprinted); |
| 555 | |
| 556 | /* If the last bytes of the buffer (module the character |
| 557 | size) have been printed we are not copying them. */ |
| 558 | size_t to_keep = unprinted != NULL ? 0 : min_len_bytes; |
| 559 | |
| 560 | memmove (buf, buf + nb - to_keep, to_keep + nb); |
| 561 | ntrailer = to_keep + nb; |
| 562 | from += nb; |
| 563 | } |
| 564 | else |
| 565 | ntrailer = nb; |
| 566 | } |
| 567 | |
| 568 | free (buf); |
| 569 | |
| 570 | /* Don't print anything we collected so far. There is no |
| 571 | terminating NUL byte. */ |
| 572 | free (unprinted); |
| 573 | |
| 574 | return result; |
| 575 | } |
| 576 | |
| 577 | |
| 578 | static int |
| 579 | read_block (int fd, const char *fname, off64_t fdlen, off64_t from, off64_t to) |
| 580 | { |
| 581 | assert ((off64_t) min_len_bytes < fdlen); |
| 582 | |
| 583 | if (elfmap == NULL) |
| 584 | { |
| 585 | /* We need a completely new mapping. */ |
| 586 | elfmap_off = from & ~(ps - 1); |
| 587 | elfmap_base = elfmap = map_file (fd, elfmap_off, fdlen, &elfmap_size); |
| 588 | |
| 589 | if (unlikely (elfmap == MAP_FAILED)) |
| 590 | /* Let the kernel know we are going to read everything in sequence. */ |
| 591 | (void) posix_fadvise (fd, 0, 0, POSIX_FADV_SEQUENTIAL); |
| 592 | } |
| 593 | |
| 594 | if (unlikely (elfmap == MAP_FAILED)) |
| 595 | { |
| 596 | /* Read from the file descriptor. For this we must position the |
| 597 | read pointer. */ |
| 598 | // XXX Eventually add flag which avoids this if the position |
| 599 | // XXX is known to match. |
| 600 | if (lseek64 (fd, from, SEEK_SET) != from) |
| 601 | error (EXIT_FAILURE, errno, gettext ("lseek64 failed")); |
| 602 | |
| 603 | return read_block_no_mmap (fd, fname, from, to - from); |
| 604 | } |
| 605 | |
| 606 | if (to < (off64_t) elfmap_off || from > (off64_t) (elfmap_off + elfmap_size)) |
| 607 | { |
| 608 | /* The existing mapping cannot fit at all. Map the new area. |
| 609 | We always map the full range of ELFMAP_SIZE bytes even if |
| 610 | this extend beyond the end of the file. The Linux kernel |
| 611 | handles this OK if the access pages are not touched. */ |
| 612 | elfmap_off = from & ~(ps - 1); |
| 613 | if (mmap64 (elfmap, elfmap_size, PROT_READ, |
| 614 | MAP_PRIVATE | MAP_POPULATE | MAP_FIXED, fd, from) |
| 615 | == MAP_FAILED) |
| 616 | error (EXIT_FAILURE, errno, gettext ("re-mmap failed")); |
| 617 | elfmap_base = elfmap; |
| 618 | } |
| 619 | |
| 620 | char *unprinted = NULL; |
| 621 | |
| 622 | /* Use the existing mapping as much as possible. If necessary, map |
| 623 | new pages. */ |
| 624 | if (from >= (off64_t) elfmap_off |
| 625 | && from < (off64_t) (elfmap_off + elfmap_size)) |
| 626 | /* There are at least a few bytes in this mapping which we can |
| 627 | use. */ |
| 628 | process_chunk (fname, elfmap_base + (from - elfmap_off), |
| 629 | MIN (to, (off64_t) (elfmap_off + elfmap_size)), |
| 630 | MIN (to, (off64_t) (elfmap_off + elfmap_size)) - from, |
| 631 | &unprinted); |
| 632 | |
| 633 | if (to > (off64_t) (elfmap_off + elfmap_size)) |
| 634 | { |
| 635 | unsigned char *remap_base = elfmap_base; |
| 636 | size_t read_now = elfmap_size - (elfmap_base - elfmap); |
| 637 | |
| 638 | assert (from >= (off64_t) elfmap_off |
| 639 | && from < (off64_t) (elfmap_off + elfmap_size)); |
| 640 | off64_t handled_to = elfmap_off + elfmap_size; |
| 641 | assert (elfmap == elfmap_base |
| 642 | || (elfmap_base - elfmap |
| 643 | == (ptrdiff_t) ((min_len_bytes + ps - 1) & ~(ps - 1)))); |
| 644 | if (elfmap == elfmap_base) |
| 645 | { |
| 646 | size_t keep_area = (min_len_bytes + ps - 1) & ~(ps - 1); |
| 647 | assert (elfmap_size >= keep_area + ps); |
| 648 | /* The keep area is used for the content of the previous |
| 649 | buffer we have to keep. This means copying those bytes |
| 650 | and for this we have to make the data writable. */ |
| 651 | if (unlikely (mprotect (elfmap, keep_area, PROT_READ | PROT_WRITE) |
| 652 | != 0)) |
| 653 | error (EXIT_FAILURE, errno, gettext ("mprotect failed")); |
| 654 | |
| 655 | elfmap_base = elfmap + keep_area; |
| 656 | } |
| 657 | |
| 658 | while (1) |
| 659 | { |
| 660 | /* Map the rest of the file, eventually again in pieces. |
| 661 | We speed things up with a nice Linux feature. Note |
| 662 | that we have at least two pages mapped. */ |
| 663 | size_t to_keep = unprinted != NULL ? 0 : min_len_bytes; |
| 664 | |
| 665 | assert (read_now >= to_keep); |
| 666 | memmove (elfmap_base - to_keep, |
| 667 | remap_base + read_now - to_keep, to_keep); |
| 668 | remap_base = elfmap_base; |
| 669 | |
| 670 | assert ((elfmap_size - (elfmap_base - elfmap)) % bytes_per_char |
| 671 | == 0); |
| 672 | read_now = MIN (to - handled_to, |
| 673 | (ptrdiff_t) elfmap_size - (elfmap_base - elfmap)); |
| 674 | |
| 675 | assert (handled_to % ps == 0); |
| 676 | assert (handled_to % bytes_per_char == 0); |
| 677 | if (mmap64 (remap_base, read_now, PROT_READ, |
| 678 | MAP_PRIVATE | MAP_POPULATE | MAP_FIXED, fd, handled_to) |
| 679 | == MAP_FAILED) |
| 680 | error (EXIT_FAILURE, errno, gettext ("re=mmap failed")); |
| 681 | elfmap_off = handled_to; |
| 682 | |
| 683 | process_chunk (fname, remap_base - to_keep, |
| 684 | elfmap_off + (read_now & ~(bytes_per_char - 1)), |
| 685 | to_keep + (read_now & ~(bytes_per_char - 1)), |
| 686 | &unprinted); |
| 687 | handled_to += read_now; |
| 688 | if (handled_to >= to) |
| 689 | break; |
| 690 | } |
| 691 | } |
| 692 | |
| 693 | /* Don't print anything we collected so far. There is no |
| 694 | terminating NUL byte. */ |
| 695 | free (unprinted); |
| 696 | |
| 697 | return 0; |
| 698 | } |
| 699 | |
| 700 | |
| 701 | static int |
| 702 | read_fd (int fd, const char *fname, off64_t fdlen) |
| 703 | { |
| 704 | return read_block (fd, fname, fdlen, 0, fdlen); |
| 705 | } |
| 706 | |
| 707 | |
| 708 | static int |
| 709 | read_elf (Elf *elf, int fd, const char *fname, off64_t fdlen) |
| 710 | { |
| 711 | assert (fdlen >= 0); |
| 712 | |
| 713 | /* We will look at each section separately. The ELF file is not |
| 714 | mmapped. The libelf implementation will load the needed parts on |
| 715 | demand. Since we only interate over the section header table the |
| 716 | memory consumption at this stage is kept minimal. */ |
| 717 | Elf_Scn *scn = elf_nextscn (elf, NULL); |
| 718 | if (scn == NULL) |
| 719 | return read_fd (fd, fname, fdlen); |
| 720 | |
| 721 | int result = 0; |
| 722 | do |
| 723 | { |
| 724 | GElf_Shdr shdr_mem; |
| 725 | GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_mem); |
| 726 | |
| 727 | /* Only look in sections which are loaded at runtime and |
| 728 | actually have content. */ |
| 729 | if (shdr != NULL && shdr->sh_type != SHT_NOBITS |
| 730 | && (shdr->sh_flags & SHF_ALLOC) != 0) |
| 731 | result |= read_block (fd, fname, fdlen, shdr->sh_offset, |
| 732 | shdr->sh_offset + shdr->sh_size); |
| 733 | } |
| 734 | while ((scn = elf_nextscn (elf, scn)) != NULL); |
| 735 | |
| 736 | if (elfmap != NULL && elfmap != MAP_FAILED) |
| 737 | munmap (elfmap, elfmap_size); |
| 738 | elfmap = NULL; |
| 739 | |
| 740 | return result; |
| 741 | } |