1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved. 24 */ 25 26 #include <stdio.h> 27 #include <stdlib.h> 28 #include <unistd.h> 29 #include <stdarg.h> 30 #include <string.h> 31 #include <strings.h> 32 #include <errno.h> 33 #include <fcntl.h> 34 #include <libintl.h> 35 #include <locale.h> 36 #include <fcntl.h> 37 #include <ar.h> 38 #include <gelf.h> 39 #include "conv.h" 40 #include "libld.h" 41 #include "machdep.h" 42 #include "msg.h" 43 44 /* 45 * The following prevent us from having to include ctype.h which defines these 46 * functions as macros which reference the __ctype[] array. Go through .plt's 47 * to get to these functions in libc rather than have every invocation of ld 48 * have to suffer the R_SPARC_COPY overhead of the __ctype[] array. 49 */ 50 extern int isspace(int); 51 52 /* 53 * We examine ELF objects, and archives containing ELF objects, in order 54 * to determine the ELFCLASS of the resulting object and/or the linker to be 55 * used. We want to avoid the overhead of libelf for this, at least until 56 * we are certain that we need it, so we start by reading bytes from 57 * the beginning of the file. This type defines the buffer used to read 58 * these initial bytes. 59 * 60 * A plain ELF object will start with an ELF header, whereas an archive 61 * starts with a magic string (ARMAG) that is SARMAG bytes long. Any valid 62 * ELF file or archive will contain more bytes than this buffer, so any 63 * file shorter than this can be safely assummed not to be of interest. 64 * 65 * The ELF header for ELFCLASS32 and ELFCLASS64 are identical up through the 66 * the e_version field, and all the information we require is found in this 67 * common prefix. Furthermore, this cannot change, as the layout of an ELF 68 * header is fixed by the ELF ABI. Hence, the ehdr part of this union is 69 * not a full ELF header, but only the class-independent prefix that we need. 70 * 71 * As this is a raw (non-libelf) read, we are responsible for handling any 72 * byte order difference between the object and the system running this 73 * program when we read any datum larger than a byte (i.e. e_machine) from 74 * this header. 75 */ 76 typedef union { 77 struct { /* Must match start of ELFxx_Ehdr in <sys/elf.h> */ 78 uchar_t e_ident[EI_NIDENT]; /* ident bytes */ 79 Half e_type; /* file type */ 80 Half e_machine; /* target machine */ 81 } ehdr; 82 char armag[SARMAG]; 83 } FILE_HDR; 84 85 86 /* 87 * Print a message to stdout 88 */ 89 void 90 veprintf(Lm_list *lml, Error error, const char *format, va_list args) 91 { 92 static const char *strings[ERR_NUM]; 93 94 #if defined(lint) 95 /* 96 * The lml argument is only meaningful for diagnostics sent to ld.so.1. 97 * Supress the lint error by making a dummy assignment. 98 */ 99 lml = 0; 100 #endif 101 /* 102 * For error types we issue a prefix for, make sure the necessary 103 * string has been internationalized and is ready. 104 */ 105 switch (error) { 106 case ERR_WARNING_NF: 107 if (strings[ERR_WARNING_NF] == NULL) 108 strings[ERR_WARNING_NF] = MSG_INTL(MSG_ERR_WARNING); 109 break; 110 case ERR_WARNING: 111 if (strings[ERR_WARNING] == NULL) 112 strings[ERR_WARNING] = MSG_INTL(MSG_ERR_WARNING); 113 break; 114 case ERR_GUIDANCE: 115 if (strings[ERR_GUIDANCE] == NULL) 116 strings[ERR_GUIDANCE] = MSG_INTL(MSG_ERR_GUIDANCE); 117 break; 118 case ERR_FATAL: 119 if (strings[ERR_FATAL] == NULL) 120 strings[ERR_FATAL] = MSG_INTL(MSG_ERR_FATAL); 121 break; 122 case ERR_ELF: 123 if (strings[ERR_ELF] == NULL) 124 strings[ERR_ELF] = MSG_INTL(MSG_ERR_ELF); 125 } 126 127 /* If strings[] element for our error type is non-NULL, issue prefix */ 128 if (strings[error] != NULL) { 129 (void) fputs(MSG_ORIG(MSG_STR_LDDIAG), stderr); 130 (void) fputs(strings[error], stderr); 131 } 132 133 (void) vfprintf(stderr, format, args); 134 if (error == ERR_ELF) { 135 int elferr; 136 137 if ((elferr = elf_errno()) != 0) 138 (void) fprintf(stderr, MSG_ORIG(MSG_STR_ELFDIAG), 139 elf_errmsg(elferr)); 140 } 141 (void) fprintf(stderr, MSG_ORIG(MSG_STR_NL)); 142 (void) fflush(stderr); 143 } 144 145 146 /* 147 * Print a message to stdout 148 */ 149 /* VARARGS3 */ 150 void 151 eprintf(Lm_list *lml, Error error, const char *format, ...) 152 { 153 va_list args; 154 155 va_start(args, format); 156 veprintf(lml, error, format, args); 157 va_end(args); 158 } 159 160 161 /* 162 * Examine the first object in an archive to determine its ELFCLASS 163 * and machine type. 164 * 165 * entry: 166 * fd - Open file descriptor for file 167 * elf - libelf ELF descriptor 168 * class_ret, mach_ret - Address of variables to receive ELFCLASS 169 * and machine type. 170 * 171 * exit: 172 * On success, *class_ret and *mach_ret are filled in, and True (1) 173 * is returned. On failure, False (0) is returned. 174 */ 175 static int 176 archive(int fd, Elf *elf, uchar_t *class_ret, Half *mach_ret) 177 { 178 Elf_Cmd cmd = ELF_C_READ; 179 Elf_Arhdr *arhdr; 180 Elf *_elf = NULL; 181 int found = 0; 182 183 /* 184 * Process each item within the archive until we find the first 185 * ELF object, or alternatively another archive to recurse into. 186 * Stop after analyzing the first plain object found. 187 */ 188 while (!found && ((_elf = elf_begin(fd, cmd, elf)) != NULL)) { 189 if ((arhdr = elf_getarhdr(_elf)) == NULL) 190 return (0); 191 if (*arhdr->ar_name != '/') { 192 switch (elf_kind(_elf)) { 193 case ELF_K_AR: 194 found = archive(fd, _elf, class_ret, mach_ret); 195 break; 196 case ELF_K_ELF: 197 if (gelf_getclass(_elf) == ELFCLASS64) { 198 Elf64_Ehdr *ehdr; 199 200 if ((ehdr = elf64_getehdr(_elf)) == 201 NULL) 202 break; 203 *class_ret = ehdr->e_ident[EI_CLASS]; 204 *mach_ret = ehdr->e_machine; 205 } else { 206 Elf32_Ehdr *ehdr; 207 208 if ((ehdr = elf32_getehdr(_elf)) == 209 NULL) 210 break; 211 *class_ret = ehdr->e_ident[EI_CLASS]; 212 *mach_ret = ehdr->e_machine; 213 } 214 found = 1; 215 break; 216 } 217 } 218 219 cmd = elf_next(_elf); 220 (void) elf_end(_elf); 221 } 222 223 return (found); 224 } 225 226 /* 227 * Determine: 228 * - ELFCLASS of resulting object (class) 229 * - Whether user specified class of the linker (ldclass) 230 * - ELF machine type of resulting object (m_mach) 231 * 232 * In order of priority, we determine this information as follows: 233 * 234 * - Command line options (-32, -64, -z altexec64, -z target). 235 * - From the first plain object seen on the command line. (This is 236 * by far the most common case.) 237 * - From the first object contained within the first archive 238 * on the command line. 239 * - If all else fails, we assume a 32-bit object for the native machine. 240 * 241 * entry: 242 * argc, argv - Command line argument vector 243 * class_ret - Address of variable to receive ELFCLASS of output object 244 * ldclass_ret - Address of variable to receive ELFCLASS of 245 * linker to use. This will be ELFCLASS32/ELFCLASS64 if one 246 * is explicitly specified, and ELFCLASSNONE otherwise. 247 * ELFCLASSNONE therefore means that we should use the best 248 * link-editor that the system/kernel will allow. 249 */ 250 static int 251 process_args(int argc, char **argv, uchar_t *class_ret, uchar_t *ldclass_ret, 252 Half *mach) 253 { 254 uchar_t ldclass = ELFCLASSNONE, class = ELFCLASSNONE, ar_class; 255 Half mach32 = EM_NONE, mach64 = EM_NONE, ar_mach; 256 int c, ar_found = 0; 257 258 /* 259 * In general, libld.so is responsible for processing the 260 * command line options. The exception to this are those options 261 * that contain information about which linker to run and the 262 * class/machine of the output object. We examine the options 263 * here looking for the following: 264 * 265 * -32 Produce an ELFCLASS32 object. This is the default, so 266 * -32 is only needed when linking entirely from archives, 267 * and the first archive contains a mix of 32 and 64-bit 268 * objects, and the first object in that archive is 64-bit. 269 * We do not expect this option to get much use, but it 270 * ensures that the user can handle any situation. 271 * 272 * -64 Produce an ELFCLASS64 object. (Note that this will 273 * indirectly cause the use of the 64-bit linker if 274 * the system is 64-bit capable). The most common need 275 * for this option is when linking a filter object entirely 276 * from a mapfile. The less common case is when linking 277 * entirely from archives, and the first archive contains 278 * a mix of 32 and 64-bit objects, and the first object 279 * in that archive is 32-bit. 280 * 281 * -z altexec64 282 * Use the 64-bit linker regardless of the class 283 * of the output object. 284 * 285 * -z target=platform 286 * Produce output object for the specified platform. 287 * This option is needed when producing an object 288 * for a non-native target entirely from a mapfile, 289 * or when linking entirely from an archive containing 290 * objects for multiple targets, and the first object 291 * in the archive is not for the desired target. 292 * 293 * If we've already processed an object and we find -32/-64, and 294 * the object is of the wrong class, we have an error condition. 295 * We ignore it here, and let it fall through to libld, where the 296 * proper diagnosis and error message will occur. 297 */ 298 opterr = 0; 299 optind = 1; 300 getmore: 301 while ((c = ld_getopt(0, optind, argc, argv)) != -1) { 302 switch (c) { 303 case '3': 304 if (strncmp(optarg, MSG_ORIG(MSG_ARG_TWO), 305 MSG_ARG_TWO_SIZE) == 0) 306 class = ELFCLASS32; 307 break; 308 309 case '6': 310 if (strncmp(optarg, MSG_ORIG(MSG_ARG_FOUR), 311 MSG_ARG_FOUR_SIZE) == 0) 312 class = ELFCLASS64; 313 break; 314 315 case 'z': 316 #if !defined(_LP64) 317 /* -z altexec64 */ 318 if (strncmp(optarg, MSG_ORIG(MSG_ARG_ALTEXEC64), 319 MSG_ARG_ALTEXEC64_SIZE) == 0) { 320 ldclass = ELFCLASS64; 321 break; 322 } 323 #endif 324 /* -z target=platform */ 325 if (strncmp(optarg, MSG_ORIG(MSG_ARG_TARGET), 326 MSG_ARG_TARGET_SIZE) == 0) { 327 char *pstr = optarg + MSG_ARG_TARGET_SIZE; 328 329 if (strcasecmp(pstr, 330 MSG_ORIG(MSG_TARG_SPARC)) == 0) { 331 mach32 = EM_SPARC; 332 mach64 = EM_SPARCV9; 333 } else if (strcasecmp(pstr, 334 MSG_ORIG(MSG_TARG_X86)) == 0) { 335 mach32 = EM_386; 336 mach64 = EM_AMD64; 337 } else { 338 eprintf(0, ERR_FATAL, 339 MSG_INTL(MSG_ERR_BADTARG), pstr); 340 return (1); 341 } 342 } 343 break; 344 } 345 } 346 347 /* 348 * Continue to look for the first ELF object to determine the class of 349 * objects to operate on. At the same time, look for the first archive 350 * of ELF objects --- if no plain ELF object is specified, the type 351 * of the first ELF object in the first archive will be used. If 352 * there is no object, and no archive, then we fall back to a 32-bit 353 * object for the native machine. 354 */ 355 for (; optind < argc; optind++) { 356 int fd; 357 FILE_HDR hdr; 358 359 /* 360 * If we detect some more options return to getopt(). 361 * Checking argv[optind][1] against null prevents a forever 362 * loop if an unadorned `-' argument is passed to us. 363 */ 364 if (argv[optind][0] == '-') { 365 if (argv[optind][1] == '\0') 366 continue; 367 else 368 goto getmore; 369 } 370 371 /* 372 * If we've already determined the object class and 373 * machine type, continue to the next argument. Only 374 * the first object contributes to this decision, and 375 * there's no value to opening or examing the subsequent 376 * ones. We do need to keep going though, because there 377 * may be additional options that might affect our 378 * class/machine decision. 379 */ 380 if ((class != ELFCLASSNONE) && (mach32 != EM_NONE)) 381 continue; 382 383 /* 384 * Open the file and determine if it is an object. We are 385 * looking for ELF objects, or archives of ELF objects. 386 * 387 * Plain objects are simple, and are the common case, so 388 * we examine them directly and avoid the map-unmap-map 389 * that would occur if we used libelf. Archives are too 390 * complex to be worth accessing directly, so if we identify 391 * an archive, we use libelf on it and accept the cost. 392 */ 393 if ((fd = open(argv[optind], O_RDONLY)) == -1) { 394 int err = errno; 395 396 eprintf(0, ERR_FATAL, MSG_INTL(MSG_SYS_OPEN), 397 argv[optind], strerror(err)); 398 return (1); 399 } 400 401 if (pread(fd, &hdr, sizeof (hdr), 0) != sizeof (hdr)) { 402 (void) close(fd); 403 continue; 404 } 405 406 if ((hdr.ehdr.e_ident[EI_MAG0] == ELFMAG0) && 407 (hdr.ehdr.e_ident[EI_MAG1] == ELFMAG1) && 408 (hdr.ehdr.e_ident[EI_MAG2] == ELFMAG2) && 409 (hdr.ehdr.e_ident[EI_MAG3] == ELFMAG3)) { 410 if (class == ELFCLASSNONE) { 411 class = hdr.ehdr.e_ident[EI_CLASS]; 412 if ((class != ELFCLASS32) && 413 (class != ELFCLASS64)) 414 class = ELFCLASSNONE; 415 } 416 417 if (mach32 == EM_NONE) { 418 int one = 1; 419 uchar_t *one_p = (uchar_t *)&one; 420 int ld_elfdata; 421 422 ld_elfdata = (one_p[0] == 1) ? 423 ELFDATA2LSB : ELFDATA2MSB; 424 /* 425 * Both the 32 and 64-bit versions get the 426 * type from the object. If the user has 427 * asked for an inconsistant class/machine 428 * combination, libld will catch it. 429 */ 430 mach32 = mach64 = 431 (ld_elfdata == hdr.ehdr.e_ident[EI_DATA]) ? 432 hdr.ehdr.e_machine : 433 BSWAP_HALF(hdr.ehdr.e_machine); 434 } 435 } else if (!ar_found && 436 (memcmp(&hdr.armag, ARMAG, SARMAG) == 0)) { 437 Elf *elf; 438 439 (void) elf_version(EV_CURRENT); 440 if ((elf = elf_begin(fd, ELF_C_READ, NULL)) == NULL) { 441 (void) close(fd); 442 continue; 443 } 444 if (elf_kind(elf) == ELF_K_AR) 445 ar_found = 446 archive(fd, elf, &ar_class, &ar_mach); 447 (void) elf_end(elf); 448 } 449 450 (void) close(fd); 451 } 452 453 /* 454 * ELFCLASS of output object: If we did not establish a class from a 455 * command option, or from the first plain object, then use the class 456 * from the first archive, and failing that, default to 32-bit. 457 */ 458 if (class == ELFCLASSNONE) 459 class = ar_found ? ar_class : ELFCLASS32; 460 *class_ret = class; 461 462 /* ELFCLASS of link-editor to use */ 463 *ldclass_ret = ldclass; 464 465 /* 466 * Machine type of output object: If we did not establish a machine 467 * type from the command line, or from the first plain object, then 468 * use the machine established by the first archive, and failing that, 469 * use the native machine. 470 */ 471 *mach = (class == ELFCLASS64) ? mach64 : mach32; 472 if (*mach == EM_NONE) 473 if (ar_found) 474 *mach = ar_mach; 475 else 476 *mach = (class == ELFCLASS64) ? M_MACH_64 : M_MACH_32; 477 478 return (0); 479 } 480 481 /* 482 * Process an LD_OPTIONS environment string. This routine is first called to 483 * count the number of options, and second to initialize a new argument array 484 * with each option. 485 */ 486 static int 487 process_ldoptions(char *str, char **nargv) 488 { 489 int argc = 0; 490 char *arg = str; 491 492 /* 493 * Walk the environment string processing any arguments that are 494 * separated by white space. 495 */ 496 while (*str != '\0') { 497 if (isspace(*str)) { 498 /* 499 * If a new argument array has been provided, terminate 500 * the original environment string, and initialize the 501 * appropriate argument array entry. 502 */ 503 if (nargv) { 504 *str++ = '\0'; 505 nargv[argc] = arg; 506 } 507 508 argc++; 509 while (isspace(*str)) 510 str++; 511 arg = str; 512 } else 513 str++; 514 } 515 if (arg != str) { 516 /* 517 * If a new argument array has been provided, initialize the 518 * final argument array entry. 519 */ 520 if (nargv) 521 nargv[argc] = arg; 522 argc++; 523 } 524 525 return (argc); 526 } 527 528 /* 529 * Determine whether an LD_OPTIONS environment variable is set, and if so, 530 * prepend environment string as a series of options to the argv array. 531 */ 532 static int 533 prepend_ldoptions(int *argcp, char ***argvp) 534 { 535 int nargc; 536 char **nargv, *ld_options; 537 int err, count; 538 539 if ((ld_options = getenv(MSG_ORIG(MSG_LD_OPTIONS))) == NULL) 540 return (0); 541 542 /* 543 * Get rid of any leading white space, and make sure the environment 544 * string has size. 545 */ 546 while (isspace(*ld_options)) 547 ld_options++; 548 if (ld_options[0] == '\0') 549 return (0); 550 551 /* 552 * Prevent modification of actual environment strings. 553 */ 554 if ((ld_options = strdup(ld_options)) == NULL) { 555 err = errno; 556 eprintf(0, ERR_FATAL, MSG_INTL(MSG_SYS_ALLOC), strerror(err)); 557 return (1); 558 } 559 560 /* 561 * Determine the number of options provided. 562 */ 563 nargc = process_ldoptions(ld_options, NULL); 564 565 /* 566 * Allocate a new argv array big enough to hold the new options from 567 * the environment string and the old argv options. 568 */ 569 if ((nargv = malloc((nargc + *argcp + 1) * sizeof (char *))) == NULL) { 570 err = errno; 571 eprintf(0, ERR_FATAL, MSG_INTL(MSG_SYS_ALLOC), strerror(err)); 572 return (1); 573 } 574 575 /* 576 * Initialize first element of new argv array to be the first element 577 * of the old argv array (ie. calling programs name). Then add the new 578 * args obtained from the environment. 579 */ 580 nargc = 0; 581 nargv[nargc++] = (*argvp)[0]; 582 nargc += process_ldoptions(ld_options, &nargv[nargc]); 583 584 /* 585 * Now add the original argv array (skipping argv[0]) to the end of the 586 * new argv array, and re-vector argc and argv to reference this new 587 * array 588 */ 589 for (count = 1; count < *argcp; count++, nargc++) 590 nargv[nargc] = (*argvp)[count]; 591 592 nargv[nargc] = NULL; 593 594 *argcp = nargc; 595 *argvp = nargv; 596 597 return (0); 598 } 599 600 /* 601 * Check to see if there is a LD_ALTEXEC=<path to alternate ld> in the 602 * environment. If so, first null the environment variable out, and then 603 * exec() the binary pointed to by the environment variable, passing the same 604 * arguments as the originating process. This mechanism permits using 605 * alternate link-editors (debugging/developer copies) even in complex build 606 * environments. 607 */ 608 static int 609 ld_altexec(char **argv, char **envp) 610 { 611 char *execstr; 612 char **str; 613 int err; 614 615 for (str = envp; *str; str++) { 616 if (strncmp(*str, MSG_ORIG(MSG_LD_ALTEXEC), 617 MSG_LD_ALTEXEC_SIZE) == 0) { 618 break; 619 } 620 } 621 622 /* 623 * If LD_ALTEXEC isn't set, return to continue executing the present 624 * link-editor. 625 */ 626 if (*str == 0) 627 return (0); 628 629 /* 630 * Get a pointer to the actual string. If it's a null entry, return. 631 */ 632 execstr = strdup(*str + MSG_LD_ALTEXEC_SIZE); 633 if (*execstr == '\0') 634 return (0); 635 636 /* 637 * Null out the LD_ALTEXEC= environment entry. 638 */ 639 (*str)[MSG_LD_ALTEXEC_SIZE] = '\0'; 640 641 /* 642 * Set argv[0] to point to our new linker 643 */ 644 argv[0] = execstr; 645 646 /* 647 * And attempt to execute it. 648 */ 649 (void) execve(execstr, argv, envp); 650 651 /* 652 * If the exec() fails, return a failure indication. 653 */ 654 err = errno; 655 eprintf(0, ERR_FATAL, MSG_INTL(MSG_SYS_EXEC), execstr, 656 strerror(err)); 657 return (1); 658 } 659 660 int 661 main(int argc, char **argv, char **envp) 662 { 663 char **oargv = argv; 664 uchar_t class, ldclass, checkclass; 665 Half mach; 666 667 /* 668 * Establish locale. 669 */ 670 (void) setlocale(LC_MESSAGES, MSG_ORIG(MSG_STR_EMPTY)); 671 (void) textdomain(MSG_ORIG(MSG_SUNW_OST_SGS)); 672 673 /* 674 * Execute an alternate linker if the LD_ALTEXEC environment variable is 675 * set. If a specified alternative could not be found, bail. 676 */ 677 if (ld_altexec(argv, envp)) 678 return (1); 679 680 /* 681 * Check the LD_OPTIONS environment variable, and if present prepend 682 * the arguments specified to the command line argument list. 683 */ 684 if (prepend_ldoptions(&argc, &argv)) 685 return (1); 686 687 /* 688 * Examine the command arguments to determine: 689 * - object class 690 * - link-editor class 691 * - target machine 692 */ 693 if (process_args(argc, argv, &class, &ldclass, &mach)) 694 return (1); 695 696 /* 697 * Unless a 32-bit link-editor was explicitly requested, try 698 * to exec the 64-bit version. 699 */ 700 if (ldclass != ELFCLASS32) 701 checkclass = conv_check_native(oargv, envp); 702 703 /* 704 * If an attempt to exec the 64-bit link-editor fails: 705 * - Bail if the 64-bit linker was explicitly requested 706 * - Continue quietly if the 64-bit linker was not requested. 707 * This is undoubtedly due to hardware/kernel limitations, 708 * and therefore represents the best we can do. Note that 709 * the 32-bit linker is capable of linking anything the 710 * 64-bit version is, subject to a 4GB limit on memory, and 711 * 2GB object size. 712 */ 713 if ((ldclass == ELFCLASS64) && (checkclass != ELFCLASS64)) { 714 eprintf(0, ERR_FATAL, MSG_INTL(MSG_SYS_64)); 715 return (1); 716 } 717 718 /* Call the libld entry point for the specified ELFCLASS */ 719 if (class == ELFCLASS64) 720 return (ld64_main(argc, argv, mach)); 721 else 722 return (ld32_main(argc, argv, mach)); 723 } 724 725 /* 726 * We supply this function for the msg module 727 */ 728 const char * 729 _ld_msg(Msg mid) 730 { 731 return (gettext(MSG_ORIG(mid))); 732 } 733