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) 1988 AT&T
  24  *        All Rights Reserved
  25  *
  26  *
  27  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
  28  */
  29 
  30 /*
  31  * Symbol table management routines
  32  */
  33 
  34 #define ELF_TARGET_AMD64
  35 
  36 #include        <stdio.h>
  37 #include        <string.h>
  38 #include        <debug.h>
  39 #include        "msg.h"
  40 #include        "_libld.h"
  41 
  42 /*
  43  * AVL tree comparator function:
  44  *
  45  * The primary key is the symbol name hash with a secondary key of the symbol
  46  * name itself.
  47  */
  48 int
  49 ld_sym_avl_comp(const void *elem1, const void *elem2)
  50 {
  51         Sym_avlnode     *sav1 = (Sym_avlnode *)elem1;
  52         Sym_avlnode     *sav2 = (Sym_avlnode *)elem2;
  53         int             res;
  54 
  55         res = sav1->sav_hash - sav2->sav_hash;
  56 
  57         if (res < 0)
  58                 return (-1);
  59         if (res > 0)
  60                 return (1);
  61 
  62         /*
  63          * Hash is equal - now compare name
  64          */
  65         res = strcmp(sav1->sav_name, sav2->sav_name);
  66         if (res == 0)
  67                 return (0);
  68         if (res > 0)
  69                 return (1);
  70         return (-1);
  71 }
  72 
  73 /*
  74  * Focal point for verifying symbol names.
  75  */
  76 inline static const char *
  77 string(Ofl_desc *ofl, Ifl_desc *ifl, Sym *sym, const char *strs, size_t strsize,
  78     int symndx, Word shndx, Word symsecndx, const char *symsecname,
  79     const char *strsecname, sd_flag_t *flags)
  80 {
  81         Word    name = sym->st_name;
  82 
  83         if (name) {
  84                 if ((ifl->ifl_flags & FLG_IF_HSTRTAB) == 0) {
  85                         ld_eprintf(ofl, ERR_FATAL, MSG_INTL(MSG_FIL_NOSTRTABLE),
  86                             ifl->ifl_name, EC_WORD(symsecndx), symsecname,
  87                             symndx, EC_XWORD(name));
  88                         return (NULL);
  89                 }
  90                 if (name >= (Word)strsize) {
  91                         ld_eprintf(ofl, ERR_FATAL,
  92                             MSG_INTL(MSG_FIL_EXCSTRTABLE), ifl->ifl_name,
  93                             EC_WORD(symsecndx), symsecname, symndx,
  94                             EC_XWORD(name), strsecname, EC_XWORD(strsize));
  95                         return (NULL);
  96                 }
  97         }
  98 
  99         /*
 100          * Determine if we're dealing with a register and if so validate it.
 101          * If it's a scratch register, a fabricated name will be returned.
 102          */
 103         if (ld_targ.t_ms.ms_is_regsym != NULL) {
 104                 const char *regname = (*ld_targ.t_ms.ms_is_regsym)(ofl, ifl,
 105                     sym, strs, symndx, shndx, symsecname, flags);
 106 
 107                 if (regname == (const char *)S_ERROR) {
 108                         return (NULL);
 109                 }
 110                 if (regname)
 111                         return (regname);
 112         }
 113 
 114         /*
 115          * If this isn't a register, but we have a global symbol with a null
 116          * name, we're not going to be able to hash this, search for it, or
 117          * do anything interesting.  However, we've been accepting a symbol of
 118          * this kind for ages now, so give the user a warning (rather than a
 119          * fatal error), just in case this instance exists somewhere in the
 120          * world and hasn't, as yet, been a problem.
 121          */
 122         if ((name == 0) && (ELF_ST_BIND(sym->st_info) != STB_LOCAL)) {
 123                 ld_eprintf(ofl, ERR_WARNING, MSG_INTL(MSG_FIL_NONAMESYM),
 124                     ifl->ifl_name, EC_WORD(symsecndx), symsecname, symndx,
 125                     EC_XWORD(name));
 126         }
 127         return (strs + name);
 128 }
 129 
 130 /*
 131  * For producing symbol names strings to use in error messages.
 132  * If the symbol has a non-null name, then the string returned by
 133  * this function is the output from demangle(), surrounded by
 134  * single quotes. For null names, a descriptive string giving
 135  * the symbol section and index is generated.
 136  *
 137  * This function uses an internal static buffer to hold the resulting
 138  * string. The value returned is usable by the caller until the next
 139  * call, at which point it is overwritten.
 140  */
 141 static const char *
 142 demangle_symname(const char *name, const char *symtab_name, Word symndx)
 143 {
 144 #define INIT_BUFSIZE 256
 145 
 146         static char     *buf;
 147         static size_t   bufsize = 0;
 148         size_t          len;
 149         int             use_name;
 150 
 151         use_name = (name != NULL) && (*name != '\0');
 152 
 153         if (use_name) {
 154                 name = demangle(name);
 155                 len = strlen(name) + 2;   /* Include room for quotes */
 156         } else {
 157                 name = MSG_ORIG(MSG_STR_EMPTY);
 158                 len = strlen(symtab_name) + 2 + CONV_INV_BUFSIZE;
 159         }
 160         len++;                  /* Null termination */
 161 
 162         /* If our buffer is too small, double it until it is big enough */
 163         if (len > bufsize) {
 164                 size_t  new_bufsize = bufsize;
 165                 char    *new_buf;
 166 
 167                 if (new_bufsize == 0)
 168                         new_bufsize = INIT_BUFSIZE;
 169                 while (len > new_bufsize)
 170                         new_bufsize *= 2;
 171                 if ((new_buf = libld_malloc(new_bufsize)) == NULL)
 172                         return (name);
 173                 buf = new_buf;
 174                 bufsize = new_bufsize;
 175         }
 176 
 177         if (use_name) {
 178                 (void) snprintf(buf, bufsize, MSG_ORIG(MSG_FMT_SYMNAM), name);
 179         } else {
 180                 (void) snprintf(buf, bufsize, MSG_ORIG(MSG_FMT_NULLSYMNAM),
 181                     symtab_name, EC_WORD(symndx));
 182         }
 183 
 184         return (buf);
 185 
 186 #undef INIT_BUFSIZE
 187 }
 188 
 189 /*
 190  * Shared objects can be built that define specific symbols that can not be
 191  * directly bound to.  These objects have a syminfo section (and an associated
 192  * DF_1_NODIRECT dynamic flags entry).  Scan this table looking for symbols
 193  * that can't be bound to directly, and if this files symbol is presently
 194  * referenced, mark it so that we don't directly bind to it.
 195  */
 196 uintptr_t
 197 ld_sym_nodirect(Is_desc *isp, Ifl_desc *ifl, Ofl_desc *ofl)
 198 {
 199         Shdr            *sifshdr, *symshdr;
 200         Syminfo         *sifdata;
 201         Sym             *symdata;
 202         char            *strdata;
 203         ulong_t         cnt, _cnt;
 204 
 205         /*
 206          * Get the syminfo data, and determine the number of entries.
 207          */
 208         sifshdr = isp->is_shdr;
 209         sifdata = (Syminfo *)isp->is_indata->d_buf;
 210         cnt =  sifshdr->sh_size / sifshdr->sh_entsize;
 211 
 212         /*
 213          * Get the associated symbol table.
 214          */
 215         if ((sifshdr->sh_link == 0) || (sifshdr->sh_link >= ifl->ifl_shnum)) {
 216                 /*
 217                  * Broken input file
 218                  */
 219                 ld_eprintf(ofl, ERR_FATAL, MSG_INTL(MSG_FIL_INVSHINFO),
 220                     ifl->ifl_name, isp->is_name, EC_XWORD(sifshdr->sh_link));
 221                 return (0);
 222         }
 223         symshdr = ifl->ifl_isdesc[sifshdr->sh_link]->is_shdr;
 224         symdata = ifl->ifl_isdesc[sifshdr->sh_link]->is_indata->d_buf;
 225 
 226         /*
 227          * Get the string table associated with the symbol table.
 228          */
 229         strdata = ifl->ifl_isdesc[symshdr->sh_link]->is_indata->d_buf;
 230 
 231         /*
 232          * Traverse the syminfo data for symbols that can't be directly
 233          * bound to.
 234          */
 235         for (_cnt = 1, sifdata++; _cnt < cnt; _cnt++, sifdata++) {
 236                 Sym             *sym;
 237                 char            *str;
 238                 Sym_desc        *sdp;
 239 
 240                 if ((sifdata->si_flags & SYMINFO_FLG_NOEXTDIRECT) == 0)
 241                         continue;
 242 
 243                 sym = (Sym *)(symdata + _cnt);
 244                 str = (char *)(strdata + sym->st_name);
 245 
 246                 if ((sdp = ld_sym_find(str, SYM_NOHASH, NULL, ofl)) != NULL) {
 247                         if (ifl != sdp->sd_file)
 248                                 continue;
 249 
 250                         sdp->sd_flags &= ~FLG_SY_DIR;
 251                         sdp->sd_flags |= FLG_SY_NDIR;
 252                 }
 253         }
 254         return (0);
 255 }
 256 
 257 /*
 258  * If, during symbol processing, it is necessary to update a local symbols
 259  * contents before we have generated the symbol tables in the output image,
 260  * create a new symbol structure and copy the original symbol contents.  While
 261  * we are processing the input files, their local symbols are part of the
 262  * read-only mapped image.  Commonly, these symbols are copied to the new output
 263  * file image and then updated to reflect their new address and any change in
 264  * attributes.  However, sometimes during relocation counting, it is necessary
 265  * to adjust the symbols information.  This routine provides for the generation
 266  * of a new symbol image so that this update can be performed.
 267  * All global symbols are copied to an internal symbol table to improve locality
 268  * of reference and hence performance, and thus this copying is not necessary.
 269  */
 270 uintptr_t
 271 ld_sym_copy(Sym_desc *sdp)
 272 {
 273         Sym     *nsym;
 274 
 275         if (sdp->sd_flags & FLG_SY_CLEAN) {
 276                 if ((nsym = libld_malloc(sizeof (Sym))) == NULL)
 277                         return (S_ERROR);
 278                 *nsym = *(sdp->sd_sym);
 279                 sdp->sd_sym = nsym;
 280                 sdp->sd_flags &= ~FLG_SY_CLEAN;
 281         }
 282         return (1);
 283 }
 284 
 285 /*
 286  * Finds a given name in the link editors internal symbol table.  If no
 287  * hash value is specified it is calculated.  A pointer to the located
 288  * Sym_desc entry is returned, or NULL if the symbol is not found.
 289  */
 290 Sym_desc *
 291 ld_sym_find(const char *name, Word hash, avl_index_t *where, Ofl_desc *ofl)
 292 {
 293         Sym_avlnode     qsav, *sav;
 294 
 295         if (hash == SYM_NOHASH)
 296                 /* LINTED */
 297                 hash = (Word)elf_hash((const char *)name);
 298         qsav.sav_hash = hash;
 299         qsav.sav_name = name;
 300 
 301         /*
 302          * Perform search for symbol in AVL tree.  Note that the 'where' field
 303          * is passed in from the caller.  If a 'where' is present, it can be
 304          * used in subsequent 'ld_sym_enter()' calls if required.
 305          */
 306         sav = avl_find(&ofl->ofl_symavl, &qsav, where);
 307 
 308         /*
 309          * If symbol was not found in the avl tree, return null to show that.
 310          */
 311         if (sav == NULL)
 312                 return (NULL);
 313 
 314         /*
 315          * Return symbol found.
 316          */
 317         return (sav->sav_sdp);
 318 }
 319 
 320 /*
 321  * Enter a new symbol into the link editors internal symbol table.
 322  * If the symbol is from an input file, information regarding the input file
 323  * and input section is also recorded.  Otherwise (file == NULL) the symbol
 324  * has been internally generated (ie. _etext, _edata, etc.).
 325  */
 326 Sym_desc *
 327 ld_sym_enter(const char *name, Sym *osym, Word hash, Ifl_desc *ifl,
 328     Ofl_desc *ofl, Word ndx, Word shndx, sd_flag_t sdflags, avl_index_t *where)
 329 {
 330         Sym_desc        *sdp;
 331         Sym_aux         *sap;
 332         Sym_avlnode     *savl;
 333         char            *_name;
 334         Sym             *nsym;
 335         Half            etype;
 336         uchar_t         vis;
 337         avl_index_t     _where;
 338 
 339         /*
 340          * Establish the file type.
 341          */
 342         if (ifl)
 343                 etype = ifl->ifl_ehdr->e_type;
 344         else
 345                 etype = ET_NONE;
 346 
 347         ofl->ofl_entercnt++;
 348 
 349         /*
 350          * Allocate a Sym Descriptor, Auxiliary Descriptor, and a Sym AVLNode -
 351          * contiguously.
 352          */
 353         if ((savl = libld_calloc(S_DROUND(sizeof (Sym_avlnode)) +
 354             S_DROUND(sizeof (Sym_desc)) +
 355             S_DROUND(sizeof (Sym_aux)), 1)) == NULL)
 356                 return ((Sym_desc *)S_ERROR);
 357         sdp = (Sym_desc *)((uintptr_t)savl +
 358             S_DROUND(sizeof (Sym_avlnode)));
 359         sap = (Sym_aux *)((uintptr_t)sdp +
 360             S_DROUND(sizeof (Sym_desc)));
 361 
 362         savl->sav_sdp = sdp;
 363         sdp->sd_file = ifl;
 364         sdp->sd_aux = sap;
 365         savl->sav_hash = sap->sa_hash = hash;
 366 
 367         /*
 368          * Copy the symbol table entry from the input file into the internal
 369          * entry and have the symbol descriptor use it.
 370          */
 371         sdp->sd_sym = nsym = &sap->sa_sym;
 372         *nsym = *osym;
 373         sdp->sd_shndx = shndx;
 374         sdp->sd_flags |= sdflags;
 375 
 376         if ((_name = libld_malloc(strlen(name) + 1)) == NULL)
 377                 return ((Sym_desc *)S_ERROR);
 378         savl->sav_name = sdp->sd_name = (const char *)strcpy(_name, name);
 379 
 380         /*
 381          * Enter Symbol in AVL tree.
 382          */
 383         if (where == 0) {
 384                 /* LINTED */
 385                 Sym_avlnode     *_savl;
 386                 /*
 387                  * If a previous ld_sym_find() hasn't initialized 'where' do it
 388                  * now.
 389                  */
 390                 where = &_where;
 391                 _savl = avl_find(&ofl->ofl_symavl, savl, where);
 392                 assert(_savl == NULL);
 393         }
 394         avl_insert(&ofl->ofl_symavl, savl, *where);
 395 
 396         /*
 397          * Record the section index.  This is possible because the
 398          * `ifl_isdesc' table is filled before we start symbol processing.
 399          */
 400         if ((sdflags & FLG_SY_SPECSEC) || (nsym->st_shndx == SHN_UNDEF))
 401                 sdp->sd_isc = NULL;
 402         else {
 403                 sdp->sd_isc = ifl->ifl_isdesc[shndx];
 404 
 405                 /*
 406                  * If this symbol is from a relocatable object, make sure that
 407                  * it is still associated with a section.  For example, an
 408                  * unknown section type (SHT_NULL) would have been rejected on
 409                  * input with a warning.  Here, we make the use of the symbol
 410                  * fatal.  A symbol descriptor is still returned, so that the
 411                  * caller can continue processing all symbols, and hence flush
 412                  * out as many error conditions as possible.
 413                  */
 414                 if ((etype == ET_REL) && (sdp->sd_isc == NULL)) {
 415                         ld_eprintf(ofl, ERR_FATAL, MSG_INTL(MSG_SYM_INVSEC),
 416                             name, ifl->ifl_name, EC_XWORD(shndx));
 417                         return (sdp);
 418                 }
 419         }
 420 
 421         /*
 422          * Mark any COMMON symbols as 'tentative'.
 423          */
 424         if (sdflags & FLG_SY_SPECSEC) {
 425                 if (nsym->st_shndx == SHN_COMMON)
 426                         sdp->sd_flags |= FLG_SY_TENTSYM;
 427 #if     defined(_ELF64)
 428                 else if ((ld_targ.t_m.m_mach == EM_AMD64) &&
 429                     (nsym->st_shndx == SHN_X86_64_LCOMMON))
 430                         sdp->sd_flags |= FLG_SY_TENTSYM;
 431 #endif
 432         }
 433 
 434         /*
 435          * Establish the symbols visibility and reference.
 436          */
 437         vis = ELF_ST_VISIBILITY(nsym->st_other);
 438 
 439         if ((etype == ET_NONE) || (etype == ET_REL)) {
 440                 switch (vis) {
 441                 case STV_DEFAULT:
 442                         sdp->sd_flags |= FLG_SY_DEFAULT;
 443                         break;
 444                 case STV_INTERNAL:
 445                 case STV_HIDDEN:
 446                         sdp->sd_flags |= FLG_SY_HIDDEN;
 447                         break;
 448                 case STV_PROTECTED:
 449                         sdp->sd_flags |= FLG_SY_PROTECT;
 450                         break;
 451                 case STV_EXPORTED:
 452                         sdp->sd_flags |= FLG_SY_EXPORT;
 453                         break;
 454                 case STV_SINGLETON:
 455                         sdp->sd_flags |= (FLG_SY_SINGLE | FLG_SY_NDIR);
 456                         ofl->ofl_flags1 |= (FLG_OF1_NDIRECT | FLG_OF1_NGLBDIR);
 457                         break;
 458                 case STV_ELIMINATE:
 459                         sdp->sd_flags |= (FLG_SY_HIDDEN | FLG_SY_ELIM);
 460                         break;
 461                 default:
 462                         assert(vis <= STV_ELIMINATE);
 463                 }
 464 
 465                 sdp->sd_ref = REF_REL_NEED;
 466 
 467                 /*
 468                  * Under -Bnodirect, all exported interfaces that have not
 469                  * explicitly been defined protected or directly bound to, are
 470                  * tagged to prevent direct binding.
 471                  */
 472                 if ((ofl->ofl_flags1 & FLG_OF1_ALNODIR) &&
 473                     ((sdp->sd_flags & (FLG_SY_PROTECT | FLG_SY_DIR)) == 0) &&
 474                     (nsym->st_shndx != SHN_UNDEF)) {
 475                         sdp->sd_flags |= FLG_SY_NDIR;
 476                 }
 477         } else {
 478                 sdp->sd_ref = REF_DYN_SEEN;
 479 
 480                 /*
 481                  * If this is a protected symbol, remember this.  Note, this
 482                  * state is different from the FLG_SY_PROTECT used to establish
 483                  * a symbol definitions visibility.  This state is used to warn
 484                  * against possible copy relocations against this referenced
 485                  * symbol.
 486                  */
 487                 if (vis == STV_PROTECTED)
 488                         sdp->sd_flags |= FLG_SY_PROT;
 489 
 490                 /*
 491                  * If this is a SINGLETON definition, then indicate the symbol
 492                  * can not be directly bound to, and retain the visibility.
 493                  * This visibility will be inherited by any references made to
 494                  * this symbol.
 495                  */
 496                 if ((vis == STV_SINGLETON) && (nsym->st_shndx != SHN_UNDEF))
 497                         sdp->sd_flags |= (FLG_SY_SINGLE | FLG_SY_NDIR);
 498 
 499                 /*
 500                  * If the new symbol is from a shared library and is associated
 501                  * with a SHT_NOBITS section then this symbol originated from a
 502                  * tentative symbol.
 503                  */
 504                 if (sdp->sd_isc &&
 505                     (sdp->sd_isc->is_shdr->sh_type == SHT_NOBITS))
 506                         sdp->sd_flags |= FLG_SY_TENTSYM;
 507         }
 508 
 509         /*
 510          * Reclassify any SHN_SUNW_IGNORE symbols to SHN_UNDEF so as to
 511          * simplify future processing.
 512          */
 513         if (nsym->st_shndx == SHN_SUNW_IGNORE) {
 514                 sdp->sd_shndx = shndx = SHN_UNDEF;
 515                 sdp->sd_flags |= (FLG_SY_REDUCED |
 516                     FLG_SY_HIDDEN | FLG_SY_IGNORE | FLG_SY_ELIM);
 517         }
 518 
 519         /*
 520          * If this is an undefined, or common symbol from a relocatable object
 521          * determine whether it is a global or weak reference (see build_osym(),
 522          * where REF_DYN_NEED definitions are returned back to undefines).
 523          */
 524         if ((etype == ET_REL) &&
 525             (ELF_ST_BIND(nsym->st_info) == STB_GLOBAL) &&
 526             ((nsym->st_shndx == SHN_UNDEF) || ((sdflags & FLG_SY_SPECSEC) &&
 527 #if     defined(_ELF64)
 528             ((nsym->st_shndx == SHN_COMMON) ||
 529             ((ld_targ.t_m.m_mach == EM_AMD64) &&
 530             (nsym->st_shndx == SHN_X86_64_LCOMMON))))))
 531 #else
 532         /* BEGIN CSTYLED */
 533             (nsym->st_shndx == SHN_COMMON))))
 534         /* END CSTYLED */
 535 #endif
 536                 sdp->sd_flags |= FLG_SY_GLOBREF;
 537 
 538         /*
 539          * Record the input filename on the referenced or defined files list
 540          * for possible later diagnostics.  The `sa_rfile' pointer contains the
 541          * name of the file that first referenced this symbol and is used to
 542          * generate undefined symbol diagnostics (refer to sym_undef_entry()).
 543          * Note that this entry can be overridden if a reference from a
 544          * relocatable object is found after a reference from a shared object
 545          * (refer to sym_override()).
 546          * The `sa_dfiles' list is used to maintain the list of files that
 547          * define the same symbol.  This list can be used for two reasons:
 548          *
 549          *  -   To save the first definition of a symbol that is not available
 550          *      for this link-edit.
 551          *
 552          *  -   To save all definitions of a symbol when the -m option is in
 553          *      effect.  This is optional as it is used to list multiple
 554          *      (interposed) definitions of a symbol (refer to ldmap_out()),
 555          *      and can be quite expensive.
 556          */
 557         if (nsym->st_shndx == SHN_UNDEF) {
 558                 sap->sa_rfile = ifl->ifl_name;
 559         } else {
 560                 if (sdp->sd_ref == REF_DYN_SEEN) {
 561                         /*
 562                          * A symbol is determined to be unavailable if it
 563                          * belongs to a version of a shared object that this
 564                          * user does not wish to use, or if it belongs to an
 565                          * implicit shared object.
 566                          */
 567                         if (ifl->ifl_vercnt) {
 568                                 Ver_index       *vip;
 569                                 Half            vndx = ifl->ifl_versym[ndx];
 570 
 571                                 sap->sa_dverndx = vndx;
 572                                 vip = &ifl->ifl_verndx[vndx];
 573                                 if (!(vip->vi_flags & FLG_VER_AVAIL)) {
 574                                         sdp->sd_flags |= FLG_SY_NOTAVAIL;
 575                                         sap->sa_vfile = ifl->ifl_name;
 576                                 }
 577                         }
 578                         if (!(ifl->ifl_flags & FLG_IF_NEEDED))
 579                                 sdp->sd_flags |= FLG_SY_NOTAVAIL;
 580 
 581                 } else if (etype == ET_REL) {
 582                         /*
 583                          * If this symbol has been obtained from a versioned
 584                          * input relocatable object then the new symbol must be
 585                          * promoted to the versioning of the output file.
 586                          */
 587                         if (ifl->ifl_versym)
 588                                 ld_vers_promote(sdp, ndx, ifl, ofl);
 589                 }
 590 
 591                 if ((ofl->ofl_flags & FLG_OF_GENMAP) &&
 592                     ((sdflags & FLG_SY_SPECSEC) == 0))
 593                         if (aplist_append(&sap->sa_dfiles, ifl->ifl_name,
 594                             AL_CNT_SDP_DFILES) == NULL)
 595                                 return ((Sym_desc *)S_ERROR);
 596         }
 597 
 598         /*
 599          * Provided we're not processing a mapfile, diagnose the entered symbol.
 600          * Mapfile processing requires the symbol to be updated with additional
 601          * information, therefore the diagnosing of the symbol is deferred until
 602          * later (see Dbg_map_symbol()).
 603          */
 604         if ((ifl == NULL) || ((ifl->ifl_flags & FLG_IF_MAPFILE) == 0))
 605                 DBG_CALL(Dbg_syms_entered(ofl, nsym, sdp));
 606 
 607         return (sdp);
 608 }
 609 
 610 /*
 611  * Add a special symbol to the symbol table.  Takes special symbol name with
 612  * and without underscores.  This routine is called, after all other symbol
 613  * resolution has completed, to generate a reserved absolute symbol (the
 614  * underscore version).  Special symbols are updated with the appropriate
 615  * values in update_osym().  If the user has already defined this symbol
 616  * issue a warning and leave the symbol as is.  If the non-underscore symbol
 617  * is referenced then turn it into a weak alias of the underscored symbol.
 618  *
 619  * The bits in sdflags_u are OR'd into the flags field of the symbol for the
 620  * underscored symbol.
 621  *
 622  * If this is a global symbol, and it hasn't explicitly been defined as being
 623  * directly bound to, indicate that it can't be directly bound to.
 624  * Historically, most special symbols only have meaning to the object in which
 625  * they exist, however, they've always been global.  To ensure compatibility
 626  * with any unexpected use presently in effect, ensure these symbols don't get
 627  * directly bound to.  Note, that establishing this state here isn't sufficient
 628  * to create a syminfo table, only if a syminfo table is being created by some
 629  * other symbol directives will the nodirect binding be recorded.  This ensures
 630  * we don't create syminfo sections for all objects we create, as this might add
 631  * unnecessary bloat to users who haven't explicitly requested extra symbol
 632  * information.
 633  */
 634 static uintptr_t
 635 sym_add_spec(const char *name, const char *uname, Word sdaux_id,
 636     sd_flag_t sdflags_u, sd_flag_t sdflags, Ofl_desc *ofl)
 637 {
 638         Sym_desc        *sdp;
 639         Sym_desc        *usdp;
 640         Sym             *sym;
 641         Word            hash;
 642         avl_index_t     where;
 643 
 644         /* LINTED */
 645         hash = (Word)elf_hash(uname);
 646         if (usdp = ld_sym_find(uname, hash, &where, ofl)) {
 647                 /*
 648                  * If the underscore symbol exists and is undefined, or was
 649                  * defined in a shared library, convert it to a local symbol.
 650                  * Otherwise leave it as is and warn the user.
 651                  */
 652                 if ((usdp->sd_shndx == SHN_UNDEF) ||
 653                     (usdp->sd_ref != REF_REL_NEED)) {
 654                         usdp->sd_ref = REF_REL_NEED;
 655                         usdp->sd_shndx = usdp->sd_sym->st_shndx = SHN_ABS;
 656                         usdp->sd_flags |= FLG_SY_SPECSEC | sdflags_u;
 657                         usdp->sd_sym->st_info =
 658                             ELF_ST_INFO(STB_GLOBAL, STT_OBJECT);
 659                         usdp->sd_isc = NULL;
 660                         usdp->sd_sym->st_size = 0;
 661                         usdp->sd_sym->st_value = 0;
 662                         /* LINTED */
 663                         usdp->sd_aux->sa_symspec = (Half)sdaux_id;
 664 
 665                         /*
 666                          * If a user hasn't specifically indicated that the
 667                          * scope of this symbol be made local, then leave it
 668                          * as global (ie. prevent automatic scoping).  The GOT
 669                          * should be defined protected, whereas all other
 670                          * special symbols are tagged as no-direct.
 671                          */
 672                         if (!SYM_IS_HIDDEN(usdp) &&
 673                             (sdflags & FLG_SY_DEFAULT)) {
 674                                 usdp->sd_aux->sa_overndx = VER_NDX_GLOBAL;
 675                                 if (sdaux_id == SDAUX_ID_GOT) {
 676                                         usdp->sd_flags &= ~FLG_SY_NDIR;
 677                                         usdp->sd_flags |= FLG_SY_PROTECT;
 678                                         usdp->sd_sym->st_other = STV_PROTECTED;
 679                                 } else if (
 680                                     ((usdp->sd_flags & FLG_SY_DIR) == 0) &&
 681                                     ((ofl->ofl_flags & FLG_OF_SYMBOLIC) == 0)) {
 682                                         usdp->sd_flags |= FLG_SY_NDIR;
 683                                 }
 684                         }
 685                         usdp->sd_flags |= sdflags;
 686 
 687                         /*
 688                          * If the reference originated from a mapfile ensure
 689                          * we mark the symbol as used.
 690                          */
 691                         if (usdp->sd_flags & FLG_SY_MAPREF)
 692                                 usdp->sd_flags |= FLG_SY_MAPUSED;
 693 
 694                         DBG_CALL(Dbg_syms_updated(ofl, usdp, uname));
 695                 } else {
 696                         ld_eprintf(ofl, ERR_WARNING, MSG_INTL(MSG_SYM_RESERVE),
 697                             uname, usdp->sd_file->ifl_name);
 698                 }
 699         } else {
 700                 /*
 701                  * If the symbol does not exist create it.
 702                  */
 703                 if ((sym = libld_calloc(sizeof (Sym), 1)) == NULL)
 704                         return (S_ERROR);
 705                 sym->st_shndx = SHN_ABS;
 706                 sym->st_info = ELF_ST_INFO(STB_GLOBAL, STT_OBJECT);
 707                 sym->st_size = 0;
 708                 sym->st_value = 0;
 709                 DBG_CALL(Dbg_syms_created(ofl->ofl_lml, uname));
 710                 if ((usdp = ld_sym_enter(uname, sym, hash, (Ifl_desc *)NULL,
 711                     ofl, 0, SHN_ABS, (FLG_SY_SPECSEC | sdflags_u), &where)) ==
 712                     (Sym_desc *)S_ERROR)
 713                         return (S_ERROR);
 714                 usdp->sd_ref = REF_REL_NEED;
 715                 /* LINTED */
 716                 usdp->sd_aux->sa_symspec = (Half)sdaux_id;
 717 
 718                 usdp->sd_aux->sa_overndx = VER_NDX_GLOBAL;
 719 
 720                 if (sdaux_id == SDAUX_ID_GOT) {
 721                         usdp->sd_flags |= FLG_SY_PROTECT;
 722                         usdp->sd_sym->st_other = STV_PROTECTED;
 723                 } else if ((sdflags & FLG_SY_DEFAULT) &&
 724                     ((ofl->ofl_flags & FLG_OF_SYMBOLIC) == 0)) {
 725                         usdp->sd_flags |= FLG_SY_NDIR;
 726                 }
 727                 usdp->sd_flags |= sdflags;
 728         }
 729 
 730         if (name && (sdp = ld_sym_find(name, SYM_NOHASH, NULL, ofl)) &&
 731             (sdp->sd_sym->st_shndx == SHN_UNDEF)) {
 732                 uchar_t bind;
 733 
 734                 /*
 735                  * If the non-underscore symbol exists and is undefined
 736                  * convert it to be a local.  If the underscore has
 737                  * sa_symspec set (ie. it was created above) then simulate this
 738                  * as a weak alias.
 739                  */
 740                 sdp->sd_ref = REF_REL_NEED;
 741                 sdp->sd_shndx = sdp->sd_sym->st_shndx = SHN_ABS;
 742                 sdp->sd_flags |= FLG_SY_SPECSEC;
 743                 sdp->sd_isc = NULL;
 744                 sdp->sd_sym->st_size = 0;
 745                 sdp->sd_sym->st_value = 0;
 746                 /* LINTED */
 747                 sdp->sd_aux->sa_symspec = (Half)sdaux_id;
 748                 if (usdp->sd_aux->sa_symspec) {
 749                         usdp->sd_aux->sa_linkndx = 0;
 750                         sdp->sd_aux->sa_linkndx = 0;
 751                         bind = STB_WEAK;
 752                 } else
 753                         bind = STB_GLOBAL;
 754                 sdp->sd_sym->st_info = ELF_ST_INFO(bind, STT_OBJECT);
 755 
 756                 /*
 757                  * If a user hasn't specifically indicated the scope of this
 758                  * symbol be made local then leave it as global (ie. prevent
 759                  * automatic scoping).  The GOT should be defined protected,
 760                  * whereas all other special symbols are tagged as no-direct.
 761                  */
 762                 if (!SYM_IS_HIDDEN(sdp) &&
 763                     (sdflags & FLG_SY_DEFAULT)) {
 764                         sdp->sd_aux->sa_overndx = VER_NDX_GLOBAL;
 765                         if (sdaux_id == SDAUX_ID_GOT) {
 766                                 sdp->sd_flags &= ~FLG_SY_NDIR;
 767                                 sdp->sd_flags |= FLG_SY_PROTECT;
 768                                 sdp->sd_sym->st_other = STV_PROTECTED;
 769                         } else if (((sdp->sd_flags & FLG_SY_DIR) == 0) &&
 770                             ((ofl->ofl_flags & FLG_OF_SYMBOLIC) == 0)) {
 771                                 sdp->sd_flags |= FLG_SY_NDIR;
 772                         }
 773                 }
 774                 sdp->sd_flags |= sdflags;
 775 
 776                 /*
 777                  * If the reference originated from a mapfile ensure
 778                  * we mark the symbol as used.
 779                  */
 780                 if (sdp->sd_flags & FLG_SY_MAPREF)
 781                         sdp->sd_flags |= FLG_SY_MAPUSED;
 782 
 783                 DBG_CALL(Dbg_syms_updated(ofl, sdp, name));
 784         }
 785         return (1);
 786 }
 787 
 788 
 789 /*
 790  * Undefined symbols can fall into one of four types:
 791  *
 792  *  -   the symbol is really undefined (SHN_UNDEF).
 793  *
 794  *  -   versioning has been enabled, however this symbol has not been assigned
 795  *      to one of the defined versions.
 796  *
 797  *  -   the symbol has been defined by an implicitly supplied library, ie. one
 798  *      which was encounted because it was NEEDED by another library, rather
 799  *      than from a command line supplied library which would become the only
 800  *      dependency of the output file being produced.
 801  *
 802  *  -   the symbol has been defined by a version of a shared object that is
 803  *      not permitted for this link-edit.
 804  *
 805  * In all cases the file who made the first reference to this symbol will have
 806  * been recorded via the `sa_rfile' pointer.
 807  */
 808 typedef enum {
 809         UNDEF,          NOVERSION,      IMPLICIT,       NOTAVAIL,
 810         BNDLOCAL
 811 } Type;
 812 
 813 static const Msg format[] = {
 814         MSG_SYM_UND_UNDEF,              /* MSG_INTL(MSG_SYM_UND_UNDEF) */
 815         MSG_SYM_UND_NOVER,              /* MSG_INTL(MSG_SYM_UND_NOVER) */
 816         MSG_SYM_UND_IMPL,               /* MSG_INTL(MSG_SYM_UND_IMPL) */
 817         MSG_SYM_UND_NOTA,               /* MSG_INTL(MSG_SYM_UND_NOTA) */
 818         MSG_SYM_UND_BNDLOCAL            /* MSG_INTL(MSG_SYM_UND_BNDLOCAL) */
 819 };
 820 
 821 /*
 822  * Issue an undefined symbol message for the given symbol.
 823  *
 824  * entry:
 825  *      ofl - Output descriptor
 826  *      sdp - Undefined symbol to report
 827  *      type - Type of undefined symbol
 828  *      ofl_flag - One of 0, FLG_OF_FATAL, or FLG_OF_WARN.
 829  *      undef_state - Address of variable to be initialized to 0
 830  *              before the first call to sym_undef_entry, and passed
 831  *              to each subsequent call. A non-zero value for *undef_state
 832  *              indicates that this is not the first call in the series.
 833  *
 834  * exit:
 835  *      If *undef_state is 0, a title is issued.
 836  *
 837  *      A message for the undefined symbol is issued.
 838  *
 839  *      If ofl_flag is non-zero, its value is OR'd into *undef_state. Otherwise,
 840  *      all bits other than FLG_OF_FATAL and FLG_OF_WARN are set, in order to
 841  *      provide *undef_state with a non-zero value. These other bits have
 842  *      no meaning beyond that, and serve to ensure that *undef_state is
 843  *      non-zero if sym_undef_entry() has been called.
 844  */
 845 static void
 846 sym_undef_entry(Ofl_desc *ofl, Sym_desc *sdp, Type type, ofl_flag_t ofl_flag,
 847     ofl_flag_t *undef_state)
 848 {
 849         const char      *name1, *name2, *name3;
 850         Ifl_desc        *ifl = sdp->sd_file;
 851         Sym_aux         *sap = sdp->sd_aux;
 852 
 853         if (*undef_state == 0)
 854                 ld_eprintf(ofl, ERR_NONE, MSG_INTL(MSG_SYM_FMT_UNDEF),
 855                     MSG_INTL(MSG_SYM_UNDEF_ITM_11),
 856                     MSG_INTL(MSG_SYM_UNDEF_ITM_21),
 857                     MSG_INTL(MSG_SYM_UNDEF_ITM_12),
 858                     MSG_INTL(MSG_SYM_UNDEF_ITM_22));
 859 
 860         ofl->ofl_flags |= ofl_flag;
 861         *undef_state |= ofl_flag ? ofl_flag : ~(FLG_OF_FATAL | FLG_OF_WARN);
 862 
 863         switch (type) {
 864         case UNDEF:
 865         case BNDLOCAL:
 866                 name1 = sap->sa_rfile;
 867                 break;
 868         case NOVERSION:
 869                 name1 = ifl->ifl_name;
 870                 break;
 871         case IMPLICIT:
 872                 name1 = sap->sa_rfile;
 873                 name2 = ifl->ifl_name;
 874                 break;
 875         case NOTAVAIL:
 876                 name1 = sap->sa_rfile;
 877                 name2 = sap->sa_vfile;
 878                 name3 = ifl->ifl_verndx[sap->sa_dverndx].vi_name;
 879                 break;
 880         default:
 881                 return;
 882         }
 883 
 884         ld_eprintf(ofl, ERR_NONE, MSG_INTL(format[type]),
 885             demangle(sdp->sd_name), name1, name2, name3);
 886 }
 887 
 888 /*
 889  * If an undef symbol exists naming a bound for the output section,
 890  * turn it into a defined symbol with the correct value.
 891  *
 892  * We set an arbitrary 1KB limit on the resulting symbol names.
 893  */
 894 static void
 895 sym_add_bounds(Ofl_desc *ofl, Os_desc *osp, Word bound)
 896 {
 897         Sym_desc *bsdp;
 898         char symn[1024];
 899         size_t nsz;
 900 
 901         switch (bound) {
 902         case SDAUX_ID_SECBOUND_START:
 903                 nsz = snprintf(symn, sizeof (symn), "%s%s",
 904                     MSG_ORIG(MSG_SYM_SECBOUND_START), osp->os_name + 1);
 905                 if (nsz >= sizeof (symn))
 906                         return;
 907                 break;
 908         case SDAUX_ID_SECBOUND_STOP:
 909                 nsz = snprintf(symn, sizeof (symn), "%s%s",
 910                     MSG_ORIG(MSG_SYM_SECBOUND_STOP), osp->os_name + 1);
 911                 if (nsz >= sizeof (symn))
 912                         return;
 913                 break;
 914         default:
 915                 assert(0);
 916         }
 917 
 918         if ((bsdp = ld_sym_find(symn, SYM_NOHASH, NULL, ofl)) != NULL) {
 919                 if ((bsdp->sd_shndx != SHN_UNDEF) &&
 920                     (bsdp->sd_ref == REF_REL_NEED)) {
 921                         ld_eprintf(ofl, ERR_WARNING, MSG_INTL(MSG_SYM_RESERVE),
 922                             symn, bsdp->sd_file->ifl_name);
 923                         return;
 924                 }
 925 
 926                 DBG_CALL(Dbg_syms_updated(ofl, bsdp, symn));
 927 
 928                 bsdp->sd_aux->sa_symspec = bound;
 929                 bsdp->sd_aux->sa_boundsec = osp;
 930                 bsdp->sd_flags |= FLG_SY_SPECSEC;
 931                 bsdp->sd_ref = REF_REL_NEED;
 932                 bsdp->sd_sym->st_info = ELF_ST_INFO(STB_GLOBAL, STT_NOTYPE);
 933                 bsdp->sd_sym->st_other = STV_PROTECTED;
 934                 bsdp->sd_isc = NULL;
 935                 bsdp->sd_sym->st_size = 0;
 936                 bsdp->sd_sym->st_value = 0;
 937                 bsdp->sd_shndx = bsdp->sd_sym->st_shndx = SHN_ABS;
 938         }
 939 }
 940 
 941 /*
 942  * At this point all symbol input processing has been completed, therefore
 943  * complete the symbol table entries by generating any necessary internal
 944  * symbols.
 945  */
 946 uintptr_t
 947 ld_sym_spec(Ofl_desc *ofl)
 948 {
 949         Sym_desc        *sdp;
 950         Sg_desc         *sgp;
 951         Aliste          idx1;
 952 
 953         if (ofl->ofl_flags & FLG_OF_RELOBJ)
 954                 return (1);
 955 
 956         DBG_CALL(Dbg_syms_spec_title(ofl->ofl_lml));
 957 
 958         /*
 959          * For each section in the output file, look for symbols named for the
 960          * __start/__stop patterns.  If references exist, flesh the symbols to
 961          * be defined.
 962          *
 963          * The symbols are given values at the same time as the other special
 964          * symbols.
 965          */
 966         for (APLIST_TRAVERSE(ofl->ofl_segs, idx1, sgp)) {
 967                 Os_desc *osp;
 968                 Aliste idx2;
 969 
 970                 for (APLIST_TRAVERSE(sgp->sg_osdescs, idx2, osp)) {
 971                         sym_add_bounds(ofl, osp, SDAUX_ID_SECBOUND_START);
 972                         sym_add_bounds(ofl, osp, SDAUX_ID_SECBOUND_STOP);
 973                 }
 974         }
 975 
 976         if (sym_add_spec(MSG_ORIG(MSG_SYM_ETEXT), MSG_ORIG(MSG_SYM_ETEXT_U),
 977             SDAUX_ID_ETEXT, 0, (FLG_SY_DEFAULT | FLG_SY_EXPDEF),
 978             ofl) == S_ERROR)
 979                 return (S_ERROR);
 980         if (sym_add_spec(MSG_ORIG(MSG_SYM_EDATA), MSG_ORIG(MSG_SYM_EDATA_U),
 981             SDAUX_ID_EDATA, 0, (FLG_SY_DEFAULT | FLG_SY_EXPDEF),
 982             ofl) == S_ERROR)
 983                 return (S_ERROR);
 984         if (sym_add_spec(MSG_ORIG(MSG_SYM_END), MSG_ORIG(MSG_SYM_END_U),
 985             SDAUX_ID_END, FLG_SY_DYNSORT, (FLG_SY_DEFAULT | FLG_SY_EXPDEF),
 986             ofl) == S_ERROR)
 987                 return (S_ERROR);
 988         if (sym_add_spec(MSG_ORIG(MSG_SYM_L_END), MSG_ORIG(MSG_SYM_L_END_U),
 989             SDAUX_ID_END, 0, FLG_SY_HIDDEN, ofl) == S_ERROR)
 990                 return (S_ERROR);
 991         if (sym_add_spec(MSG_ORIG(MSG_SYM_L_START), MSG_ORIG(MSG_SYM_L_START_U),
 992             SDAUX_ID_START, 0, FLG_SY_HIDDEN, ofl) == S_ERROR)
 993                 return (S_ERROR);
 994 
 995         /*
 996          * Historically we've always produced a _DYNAMIC symbol, even for
 997          * static executables (in which case its value will be 0).
 998          */
 999         if (sym_add_spec(MSG_ORIG(MSG_SYM_DYNAMIC), MSG_ORIG(MSG_SYM_DYNAMIC_U),
1000             SDAUX_ID_DYN, FLG_SY_DYNSORT, (FLG_SY_DEFAULT | FLG_SY_EXPDEF),
1001             ofl) == S_ERROR)
1002                 return (S_ERROR);
1003 
1004         if (OFL_ALLOW_DYNSYM(ofl))
1005                 if (sym_add_spec(MSG_ORIG(MSG_SYM_PLKTBL),
1006                     MSG_ORIG(MSG_SYM_PLKTBL_U), SDAUX_ID_PLT,
1007                     FLG_SY_DYNSORT, (FLG_SY_DEFAULT | FLG_SY_EXPDEF),
1008                     ofl) == S_ERROR)
1009                         return (S_ERROR);
1010 
1011         /*
1012          * A GOT reference will be accompanied by the associated GOT symbol.
1013          * Make sure it gets assigned the appropriate special attributes.
1014          */
1015         if (((sdp = ld_sym_find(MSG_ORIG(MSG_SYM_GOFTBL_U),
1016             SYM_NOHASH, NULL, ofl)) != NULL) && (sdp->sd_ref != REF_DYN_SEEN)) {
1017                 if (sym_add_spec(MSG_ORIG(MSG_SYM_GOFTBL),
1018                     MSG_ORIG(MSG_SYM_GOFTBL_U), SDAUX_ID_GOT, FLG_SY_DYNSORT,
1019                     (FLG_SY_DEFAULT | FLG_SY_EXPDEF), ofl) == S_ERROR)
1020                         return (S_ERROR);
1021         }
1022 
1023         return (1);
1024 }
1025 
1026 /*
1027  * Determine a potential capability symbol's visibility.
1028  *
1029  * The -z symbolcap option transforms an object capabilities relocatable object
1030  * into a symbol capabilities relocatable object.  Any global function symbols,
1031  * or initialized global data symbols are candidates for transforming into local
1032  * symbol capabilities definitions.  However, if a user indicates that a symbol
1033  * should be demoted to local using a mapfile, then there is no need to
1034  * transform the associated global symbol.
1035  *
1036  * Normally, a symbol's visibility is determined after the symbol resolution
1037  * process, after all symbol state has been gathered and resolved.  However,
1038  * for -z symbolcap, this determination is too late.  When a global symbol is
1039  * read from an input file we need to determine it's visibility so as to decide
1040  * whether to create a local or not.
1041  *
1042  * If a user has explicitly defined this symbol as having local scope within a
1043  * mapfile, then a symbol of the same name already exists.  However, explicit
1044  * local definitions are uncommon, as most mapfiles define the global symbol
1045  * requirements together with an auto-reduction directive '*'.  If this state
1046  * has been defined, then we must make sure that the new symbol isn't a type
1047  * that can not be demoted to local.
1048  */
1049 static int
1050 sym_cap_vis(const char *name, Word hash, Sym *sym, Ofl_desc *ofl)
1051 {
1052         Sym_desc        *sdp;
1053         uchar_t         vis;
1054         avl_index_t     where;
1055         sd_flag_t       sdflags = 0;
1056 
1057         /*
1058          * Determine the visibility of the new symbol.
1059          */
1060         vis = ELF_ST_VISIBILITY(sym->st_other);
1061         switch (vis) {
1062         case STV_EXPORTED:
1063                 sdflags |= FLG_SY_EXPORT;
1064                 break;
1065         case STV_SINGLETON:
1066                 sdflags |= FLG_SY_SINGLE;
1067                 break;
1068         }
1069 
1070         /*
1071          * Determine whether a symbol definition already exists, and if so
1072          * obtain the visibility.
1073          */
1074         if ((sdp = ld_sym_find(name, hash, &where, ofl)) != NULL)
1075                 sdflags |= sdp->sd_flags;
1076 
1077         /*
1078          * Determine whether the symbol flags indicate this symbol should be
1079          * hidden.
1080          */
1081         if ((ofl->ofl_flags & (FLG_OF_AUTOLCL | FLG_OF_AUTOELM)) &&
1082             ((sdflags & MSK_SY_NOAUTO) == 0))
1083                 sdflags |= FLG_SY_HIDDEN;
1084 
1085         return ((sdflags & FLG_SY_HIDDEN) == 0);
1086 }
1087 
1088 /*
1089  * This routine checks to see if a symbols visibility needs to be reduced to
1090  * either SYMBOLIC or LOCAL.  This routine can be called from either
1091  * reloc_init() or sym_validate().
1092  */
1093 void
1094 ld_sym_adjust_vis(Sym_desc *sdp, Ofl_desc *ofl)
1095 {
1096         ofl_flag_t      oflags = ofl->ofl_flags;
1097         Sym             *sym = sdp->sd_sym;
1098 
1099         if ((sdp->sd_ref == REF_REL_NEED) &&
1100             (sdp->sd_sym->st_shndx != SHN_UNDEF)) {
1101                 /*
1102                  * If auto-reduction/elimination is enabled, reduce any
1103                  * non-versioned, and non-local capabilities global symbols.
1104                  * A symbol is a candidate for auto-reduction/elimination if:
1105                  *
1106                  *  -   the symbol wasn't explicitly defined within a mapfile
1107                  *      (in which case all the necessary state has been applied
1108                  *      to the symbol), or
1109                  *  -   the symbol isn't one of the family of reserved
1110                  *      special symbols (ie. _end, _etext, etc.), or
1111                  *  -   the symbol isn't a SINGLETON, or
1112                  *  -   the symbol wasn't explicitly defined within a version
1113                  *      definition associated with an input relocatable object.
1114                  *
1115                  * Indicate that the symbol has been reduced as it may be
1116                  * necessary to print these symbols later.
1117                  */
1118                 if ((oflags & (FLG_OF_AUTOLCL | FLG_OF_AUTOELM)) &&
1119                     ((sdp->sd_flags & MSK_SY_NOAUTO) == 0)) {
1120                         if ((sdp->sd_flags & FLG_SY_HIDDEN) == 0) {
1121                                 sdp->sd_flags |=
1122                                     (FLG_SY_REDUCED | FLG_SY_HIDDEN);
1123                         }
1124 
1125                         if (oflags & (FLG_OF_REDLSYM | FLG_OF_AUTOELM)) {
1126                                 sdp->sd_flags |= FLG_SY_ELIM;
1127                                 sym->st_other = STV_ELIMINATE |
1128                                     (sym->st_other & ~MSK_SYM_VISIBILITY);
1129                         } else if (ELF_ST_VISIBILITY(sym->st_other) !=
1130                             STV_INTERNAL)
1131                                 sym->st_other = STV_HIDDEN |
1132                                     (sym->st_other & ~MSK_SYM_VISIBILITY);
1133                 }
1134 
1135                 /*
1136                  * If -Bsymbolic is in effect, and the symbol hasn't explicitly
1137                  * been defined nodirect (via a mapfile), then bind the global
1138                  * symbol symbolically and assign the STV_PROTECTED visibility
1139                  * attribute.
1140                  */
1141                 if ((oflags & FLG_OF_SYMBOLIC) &&
1142                     ((sdp->sd_flags & (FLG_SY_HIDDEN | FLG_SY_NDIR)) == 0)) {
1143                         sdp->sd_flags |= FLG_SY_PROTECT;
1144                         if (ELF_ST_VISIBILITY(sym->st_other) == STV_DEFAULT)
1145                                 sym->st_other = STV_PROTECTED |
1146                                     (sym->st_other & ~MSK_SYM_VISIBILITY);
1147                 }
1148         }
1149 
1150         /*
1151          * Indicate that this symbol has had it's visibility checked so that
1152          * we don't need to do this investigation again.
1153          */
1154         sdp->sd_flags |= FLG_SY_VISIBLE;
1155 }
1156 
1157 /*
1158  * Make sure a symbol definition is local to the object being built.
1159  */
1160 inline static int
1161 ensure_sym_local(Ofl_desc *ofl, Sym_desc *sdp, const char *str)
1162 {
1163         if (sdp->sd_sym->st_shndx == SHN_UNDEF) {
1164                 if (str) {
1165                         ld_eprintf(ofl, ERR_FATAL, MSG_INTL(MSG_SYM_UNDEF),
1166                             str, demangle((char *)sdp->sd_name));
1167                 }
1168                 return (1);
1169         }
1170         if (sdp->sd_ref != REF_REL_NEED) {
1171                 if (str) {
1172                         ld_eprintf(ofl, ERR_FATAL, MSG_INTL(MSG_SYM_EXTERN),
1173                             str, demangle((char *)sdp->sd_name),
1174                             sdp->sd_file->ifl_name);
1175                 }
1176                 return (1);
1177         }
1178 
1179         sdp->sd_flags |= FLG_SY_UPREQD;
1180         if (sdp->sd_isc) {
1181                 sdp->sd_isc->is_flags |= FLG_IS_SECTREF;
1182                 sdp->sd_isc->is_file->ifl_flags |= FLG_IF_FILEREF;
1183         }
1184         return (0);
1185 }
1186 
1187 /*
1188  * Make sure all the symbol definitions required for initarray, finiarray, or
1189  * preinitarray's are local to the object being built.
1190  */
1191 static int
1192 ensure_array_local(Ofl_desc *ofl, APlist *apl, const char *str)
1193 {
1194         Aliste          idx;
1195         Sym_desc        *sdp;
1196         int             ret = 0;
1197 
1198         for (APLIST_TRAVERSE(apl, idx, sdp))
1199                 ret += ensure_sym_local(ofl, sdp, str);
1200 
1201         return (ret);
1202 }
1203 
1204 /*
1205  * After all symbol table input processing has been finished, and all relocation
1206  * counting has been carried out (ie. no more symbols will be read, generated,
1207  * or modified), validate and count the relevant entries:
1208  *
1209  *  -   check and print any undefined symbols remaining.  Note that if a symbol
1210  *      has been defined by virtue of the inclusion of  an implicit shared
1211  *      library, it is still classed as undefined.
1212  *
1213  *  -   count the number of global needed symbols together with the size of
1214  *      their associated name strings (if scoping has been indicated these
1215  *      symbols may be reduced to locals).
1216  *
1217  *  -   establish the size and alignment requirements for the global .bss
1218  *      section (the alignment of this section is based on the first symbol
1219  *      that it will contain).
1220  */
1221 uintptr_t
1222 ld_sym_validate(Ofl_desc *ofl)
1223 {
1224         Sym_avlnode     *sav;
1225         Sym_desc        *sdp;
1226         Sym             *sym;
1227         ofl_flag_t      oflags = ofl->ofl_flags;
1228         ofl_flag_t      undef = 0, needed = 0, verdesc = 0;
1229         Xword           bssalign = 0, tlsalign = 0;
1230         Boolean         need_bss, need_tlsbss;
1231         Xword           bsssize = 0, tlssize = 0;
1232 #if     defined(_ELF64)
1233         Xword           lbssalign = 0, lbsssize = 0;
1234         Boolean         need_lbss;
1235 #endif
1236         int             ret, allow_ldynsym;
1237         uchar_t         type;
1238         ofl_flag_t      undef_state = 0;
1239 
1240         DBG_CALL(Dbg_basic_validate(ofl->ofl_lml));
1241 
1242         /*
1243          * The need_XXX booleans are used to determine whether we need to
1244          * create each type of bss section. We used to create these sections
1245          * if the sum of the required sizes for each type were non-zero.
1246          * However, it is possible for a compiler to generate COMMON variables
1247          * of zero-length and this tricks that logic --- even zero-length
1248          * symbols need an output section.
1249          */
1250         need_bss = need_tlsbss = FALSE;
1251 #if     defined(_ELF64)
1252         need_lbss = FALSE;
1253 #endif
1254 
1255         /*
1256          * Determine how undefined symbols are handled:
1257          *
1258          * fatal:
1259          *      If this link-edit calls for no undefined symbols to remain
1260          *      (this is the default case when generating an executable but
1261          *      can be enforced for any object using -z defs), a fatal error
1262          *      condition will be indicated.
1263          *
1264          * warning:
1265          *      If we're creating a shared object, and either the -Bsymbolic
1266          *      flag is set, or the user has turned on the -z guidance feature,
1267          *      then a non-fatal warning is issued for each symbol.
1268          *
1269          * ignore:
1270          *      In all other cases, undefined symbols are quietly allowed.
1271          */
1272         if (oflags & FLG_OF_NOUNDEF) {
1273                 undef = FLG_OF_FATAL;
1274         } else if (oflags & FLG_OF_SHAROBJ) {
1275                 if ((oflags & FLG_OF_SYMBOLIC) ||
1276                     OFL_GUIDANCE(ofl, FLG_OFG_NO_DEFS))
1277                         undef = FLG_OF_WARN;
1278         }
1279 
1280         /*
1281          * If the symbol is referenced from an implicitly included shared object
1282          * (ie. it's not on the NEEDED list) then the symbol is also classified
1283          * as undefined and a fatal error condition will be indicated.
1284          */
1285         if ((oflags & FLG_OF_NOUNDEF) || !(oflags & FLG_OF_SHAROBJ))
1286                 needed = FLG_OF_FATAL;
1287         else if ((oflags & FLG_OF_SHAROBJ) &&
1288             OFL_GUIDANCE(ofl, FLG_OFG_NO_DEFS))
1289                 needed = FLG_OF_WARN;
1290 
1291         /*
1292          * If the output image is being versioned, then all symbol definitions
1293          * must be associated with a version.  Any symbol that isn't associated
1294          * with a version is classified as undefined, and a fatal error
1295          * condition is indicated.
1296          */
1297         if ((oflags & FLG_OF_VERDEF) && (ofl->ofl_vercnt > VER_NDX_GLOBAL))
1298                 verdesc = FLG_OF_FATAL;
1299 
1300         allow_ldynsym = OFL_ALLOW_LDYNSYM(ofl);
1301 
1302         if (allow_ldynsym) {
1303                 /*
1304                  * Normally, we disallow symbols with 0 size from appearing
1305                  * in a dyn[sym|tls]sort section. However, there are some
1306                  * symbols that serve special purposes that we want to exempt
1307                  * from this rule. Look them up, and set their
1308                  * FLG_SY_DYNSORT flag.
1309                  */
1310                 static const char *special[] = {
1311                         MSG_ORIG(MSG_SYM_INIT_U),       /* _init */
1312                         MSG_ORIG(MSG_SYM_FINI_U),       /* _fini */
1313                         MSG_ORIG(MSG_SYM_START),        /* _start */
1314                         NULL
1315                 };
1316                 int i;
1317 
1318                 for (i = 0; special[i] != NULL; i++) {
1319                         if (((sdp = ld_sym_find(special[i],
1320                             SYM_NOHASH, NULL, ofl)) != NULL) &&
1321                             (sdp->sd_sym->st_size == 0)) {
1322                                 if (ld_sym_copy(sdp) == S_ERROR)
1323                                         return (S_ERROR);
1324                                 sdp->sd_flags |= FLG_SY_DYNSORT;
1325                         }
1326                 }
1327         }
1328 
1329         /*
1330          * Collect and validate the globals from the internal symbol table.
1331          */
1332         for (sav = avl_first(&ofl->ofl_symavl); sav;
1333             sav = AVL_NEXT(&ofl->ofl_symavl, sav)) {
1334                 Is_desc         *isp;
1335                 int             undeferr = 0;
1336                 uchar_t         vis;
1337 
1338                 sdp = sav->sav_sdp;
1339 
1340                 /*
1341                  * If undefined symbols are allowed, and we're not being
1342                  * asked to supply guidance, ignore any symbols that are
1343                  * not needed.
1344                  */
1345                 if (!(oflags & FLG_OF_NOUNDEF) &&
1346                     !OFL_GUIDANCE(ofl, FLG_OFG_NO_DEFS) &&
1347                     (sdp->sd_ref == REF_DYN_SEEN))
1348                         continue;
1349 
1350                 /*
1351                  * If the symbol originates from an external or parent mapfile
1352                  * reference and hasn't been matched to a reference from a
1353                  * relocatable object, ignore it.
1354                  */
1355                 if ((sdp->sd_flags & (FLG_SY_EXTERN | FLG_SY_PARENT)) &&
1356                     ((sdp->sd_flags & FLG_SY_MAPUSED) == 0)) {
1357                         sdp->sd_flags |= FLG_SY_INVALID;
1358                         continue;
1359                 }
1360 
1361                 sym = sdp->sd_sym;
1362                 type = ELF_ST_TYPE(sym->st_info);
1363 
1364                 /*
1365                  * Sanity check TLS.
1366                  */
1367                 if ((type == STT_TLS) && (sym->st_size != 0) &&
1368                     (sym->st_shndx != SHN_UNDEF) &&
1369                     (sym->st_shndx != SHN_COMMON)) {
1370                         Is_desc         *isp = sdp->sd_isc;
1371                         Ifl_desc        *ifl = sdp->sd_file;
1372 
1373                         if ((isp == NULL) || (isp->is_shdr == NULL) ||
1374                             ((isp->is_shdr->sh_flags & SHF_TLS) == 0)) {
1375                                 ld_eprintf(ofl, ERR_FATAL,
1376                                     MSG_INTL(MSG_SYM_TLS),
1377                                     demangle(sdp->sd_name), ifl->ifl_name);
1378                                 continue;
1379                         }
1380                 }
1381 
1382                 if ((sdp->sd_flags & FLG_SY_VISIBLE) == 0)
1383                         ld_sym_adjust_vis(sdp, ofl);
1384 
1385                 if ((sdp->sd_flags & FLG_SY_REDUCED) &&
1386                     (oflags & FLG_OF_PROCRED)) {
1387                         DBG_CALL(Dbg_syms_reduce(ofl, DBG_SYM_REDUCE_GLOBAL,
1388                             sdp, 0, 0));
1389                 }
1390 
1391                 /*
1392                  * Record any STV_SINGLETON existence.
1393                  */
1394                 if ((vis = ELF_ST_VISIBILITY(sym->st_other)) == STV_SINGLETON)
1395                         ofl->ofl_dtflags_1 |= DF_1_SINGLETON;
1396 
1397                 /*
1398                  * If building a shared object or executable, and this is a
1399                  * non-weak UNDEF symbol with reduced visibility (STV_*), then
1400                  * give a fatal error.
1401                  */
1402                 if (((oflags & FLG_OF_RELOBJ) == 0) &&
1403                     (sym->st_shndx == SHN_UNDEF) &&
1404                     (ELF_ST_BIND(sym->st_info) != STB_WEAK)) {
1405                         if (vis && (vis != STV_SINGLETON)) {
1406                                 sym_undef_entry(ofl, sdp, BNDLOCAL,
1407                                     FLG_OF_FATAL, &undef_state);
1408                                 continue;
1409                         }
1410                 }
1411 
1412                 /*
1413                  * If this symbol is defined in a non-allocatable section,
1414                  * reduce it to local symbol.
1415                  */
1416                 if (((isp = sdp->sd_isc) != 0) && isp->is_shdr &&
1417                     ((isp->is_shdr->sh_flags & SHF_ALLOC) == 0)) {
1418                         sdp->sd_flags |= (FLG_SY_REDUCED | FLG_SY_HIDDEN);
1419                 }
1420 
1421                 /*
1422                  * If this symbol originated as a SHN_SUNW_IGNORE, it will have
1423                  * been processed as an SHN_UNDEF.  Return the symbol to its
1424                  * original index for validation, and propagation to the output
1425                  * file.
1426                  */
1427                 if (sdp->sd_flags & FLG_SY_IGNORE)
1428                         sdp->sd_shndx = SHN_SUNW_IGNORE;
1429 
1430                 if (undef) {
1431                         /*
1432                          * If a non-weak reference remains undefined, or if a
1433                          * mapfile reference is not bound to the relocatable
1434                          * objects that make up the object being built, we have
1435                          * a fatal error.
1436                          *
1437                          * The exceptions are symbols which are defined to be
1438                          * found in the parent (FLG_SY_PARENT), which is really
1439                          * only meaningful for direct binding, or are defined
1440                          * external (FLG_SY_EXTERN) so as to suppress -zdefs
1441                          * errors.
1442                          *
1443                          * Register symbols are always allowed to be UNDEF.
1444                          *
1445                          * Note that we don't include references created via -u
1446                          * in the same shared object binding test.  This is for
1447                          * backward compatibility, in that a number of archive
1448                          * makefile rules used -u to cause archive extraction.
1449                          * These same rules have been cut and pasted to apply
1450                          * to shared objects, and thus although the -u reference
1451                          * is redundant, flagging it as fatal could cause some
1452                          * build to fail.  Also we have documented the use of
1453                          * -u as a mechanism to cause binding to weak version
1454                          * definitions, thus giving users an error condition
1455                          * would be incorrect.
1456                          */
1457                         if (!(sdp->sd_flags & FLG_SY_REGSYM) &&
1458                             ((sym->st_shndx == SHN_UNDEF) &&
1459                             ((ELF_ST_BIND(sym->st_info) != STB_WEAK) &&
1460                             ((sdp->sd_flags &
1461                             (FLG_SY_PARENT | FLG_SY_EXTERN)) == 0)) ||
1462                             ((sdp->sd_flags &
1463                             (FLG_SY_MAPREF | FLG_SY_MAPUSED | FLG_SY_HIDDEN |
1464                             FLG_SY_PROTECT)) == FLG_SY_MAPREF))) {
1465                                 sym_undef_entry(ofl, sdp, UNDEF, undef,
1466                                     &undef_state);
1467                                 undeferr = 1;
1468                         }
1469 
1470                 } else {
1471                         /*
1472                          * For building things like shared objects (or anything
1473                          * -znodefs), undefined symbols are allowed.
1474                          *
1475                          * If a mapfile reference remains undefined the user
1476                          * would probably like a warning at least (they've
1477                          * usually mis-spelt the reference).  Refer to the above
1478                          * comments for discussion on -u references, which
1479                          * are not tested for in the same manner.
1480                          */
1481                         if ((sdp->sd_flags &
1482                             (FLG_SY_MAPREF | FLG_SY_MAPUSED)) ==
1483                             FLG_SY_MAPREF) {
1484                                 sym_undef_entry(ofl, sdp, UNDEF, FLG_OF_WARN,
1485                                     &undef_state);
1486                                 undeferr = 1;
1487                         }
1488                 }
1489 
1490                 /*
1491                  * If this symbol comes from a dependency mark the dependency
1492                  * as required (-z ignore can result in unused dependencies
1493                  * being dropped).  If we need to record dependency versioning
1494                  * information indicate what version of the needed shared object
1495                  * this symbol is part of.  Flag the symbol as undefined if it
1496                  * has not been made available to us.
1497                  */
1498                 if ((sdp->sd_ref == REF_DYN_NEED) &&
1499                     (!(sdp->sd_flags & FLG_SY_REFRSD))) {
1500                         sdp->sd_file->ifl_flags |= FLG_IF_DEPREQD;
1501 
1502                         /*
1503                          * Capture that we've bound to a symbol that doesn't
1504                          * allow being directly bound to.
1505                          */
1506                         if (sdp->sd_flags & FLG_SY_NDIR)
1507                                 ofl->ofl_flags1 |= FLG_OF1_NGLBDIR;
1508 
1509                         if (sdp->sd_file->ifl_vercnt) {
1510                                 int             vndx;
1511                                 Ver_index       *vip;
1512 
1513                                 vndx = sdp->sd_aux->sa_dverndx;
1514                                 vip = &sdp->sd_file->ifl_verndx[vndx];
1515                                 if (vip->vi_flags & FLG_VER_AVAIL) {
1516                                         vip->vi_flags |= FLG_VER_REFER;
1517                                 } else {
1518                                         sym_undef_entry(ofl, sdp, NOTAVAIL,
1519                                             FLG_OF_FATAL, &undef_state);
1520                                         continue;
1521                                 }
1522                         }
1523                 }
1524 
1525                 /*
1526                  * Test that we do not bind to symbol supplied from an implicit
1527                  * shared object.  If a binding is from a weak reference it can
1528                  * be ignored.
1529                  */
1530                 if (needed && !undeferr && (sdp->sd_flags & FLG_SY_GLOBREF) &&
1531                     (sdp->sd_ref == REF_DYN_NEED) &&
1532                     (sdp->sd_flags & FLG_SY_NOTAVAIL)) {
1533                         sym_undef_entry(ofl, sdp, IMPLICIT, needed,
1534                             &undef_state);
1535                         if (needed == FLG_OF_FATAL)
1536                                 continue;
1537                 }
1538 
1539                 /*
1540                  * Test that a symbol isn't going to be reduced to local scope
1541                  * which actually wants to bind to a shared object - if so it's
1542                  * a fatal error.
1543                  */
1544                 if ((sdp->sd_ref == REF_DYN_NEED) &&
1545                     (sdp->sd_flags & (FLG_SY_HIDDEN | FLG_SY_PROTECT))) {
1546                         sym_undef_entry(ofl, sdp, BNDLOCAL, FLG_OF_FATAL,
1547                             &undef_state);
1548                         continue;
1549                 }
1550 
1551                 /*
1552                  * If the output image is to be versioned then all symbol
1553                  * definitions must be associated with a version.  Remove any
1554                  * versioning that might be left associated with an undefined
1555                  * symbol.
1556                  */
1557                 if (verdesc && (sdp->sd_ref == REF_REL_NEED)) {
1558                         if (sym->st_shndx == SHN_UNDEF) {
1559                                 if (sdp->sd_aux && sdp->sd_aux->sa_overndx)
1560                                         sdp->sd_aux->sa_overndx = 0;
1561                         } else {
1562                                 if (!SYM_IS_HIDDEN(sdp) && sdp->sd_aux &&
1563                                     (sdp->sd_aux->sa_overndx == 0)) {
1564                                         sym_undef_entry(ofl, sdp, NOVERSION,
1565                                             verdesc, &undef_state);
1566                                         continue;
1567                                 }
1568                         }
1569                 }
1570 
1571                 /*
1572                  * If we don't need the symbol there's no need to process it
1573                  * any further.
1574                  */
1575                 if (sdp->sd_ref == REF_DYN_SEEN)
1576                         continue;
1577 
1578                 /*
1579                  * Calculate the size and alignment requirements for the global
1580                  * .bss and .tls sections.  If we're building a relocatable
1581                  * object only account for scoped COMMON symbols (these will
1582                  * be converted to .bss references).
1583                  *
1584                  * When -z nopartial is in effect, partially initialized
1585                  * symbols are directed to the special .data section
1586                  * created for that purpose (ofl->ofl_isparexpn).
1587                  * Otherwise, partially initialized symbols go to .bss.
1588                  *
1589                  * Also refer to make_mvsections() in sunwmove.c
1590                  */
1591                 if ((sym->st_shndx == SHN_COMMON) &&
1592                     (((oflags & FLG_OF_RELOBJ) == 0) ||
1593                     (SYM_IS_HIDDEN(sdp) && (oflags & FLG_OF_PROCRED)))) {
1594                         if ((sdp->sd_move == NULL) ||
1595                             ((sdp->sd_flags & FLG_SY_PAREXPN) == 0)) {
1596                                 if (type != STT_TLS) {
1597                                         need_bss = TRUE;
1598                                         bsssize = (Xword)S_ROUND(bsssize,
1599                                             sym->st_value) + sym->st_size;
1600                                         if (sym->st_value > bssalign)
1601                                                 bssalign = sym->st_value;
1602                                 } else {
1603                                         need_tlsbss = TRUE;
1604                                         tlssize = (Xword)S_ROUND(tlssize,
1605                                             sym->st_value) + sym->st_size;
1606                                         if (sym->st_value > tlsalign)
1607                                                 tlsalign = sym->st_value;
1608                                 }
1609                         }
1610                 }
1611 
1612 #if     defined(_ELF64)
1613                 /*
1614                  * Calculate the size and alignment requirement for the global
1615                  * .lbss. TLS or partially initialized symbols do not need to be
1616                  * considered yet.
1617                  */
1618                 if ((ld_targ.t_m.m_mach == EM_AMD64) &&
1619                     (sym->st_shndx == SHN_X86_64_LCOMMON)) {
1620                         need_lbss = TRUE;
1621                         lbsssize = (Xword)S_ROUND(lbsssize, sym->st_value) +
1622                             sym->st_size;
1623                         if (sym->st_value > lbssalign)
1624                                 lbssalign = sym->st_value;
1625                 }
1626 #endif
1627                 /*
1628                  * If a symbol was referenced via the command line
1629                  * (ld -u <>, ...), then this counts as a reference against the
1630                  * symbol. Mark any section that symbol is defined in.
1631                  */
1632                 if (((isp = sdp->sd_isc) != 0) &&
1633                     (sdp->sd_flags & FLG_SY_CMDREF)) {
1634                         isp->is_flags |= FLG_IS_SECTREF;
1635                         isp->is_file->ifl_flags |= FLG_IF_FILEREF;
1636                 }
1637 
1638                 /*
1639                  * Update the symbol count and the associated name string size.
1640                  * Note, a capabilities symbol must remain as visible as a
1641                  * global symbol.  However, the runtime linker recognizes the
1642                  * hidden requirement and ensures the symbol isn't made globally
1643                  * available at runtime.
1644                  */
1645                 if (SYM_IS_HIDDEN(sdp) && (oflags & FLG_OF_PROCRED)) {
1646                         /*
1647                          * If any reductions are being processed, keep a count
1648                          * of eliminated symbols, and if the symbol is being
1649                          * reduced to local, count it's size for the .symtab.
1650                          */
1651                         if (sdp->sd_flags & FLG_SY_ELIM) {
1652                                 ofl->ofl_elimcnt++;
1653                         } else {
1654                                 ofl->ofl_scopecnt++;
1655                                 if ((((sdp->sd_flags & FLG_SY_REGSYM) == 0) ||
1656                                     sym->st_name) && (st_insert(ofl->ofl_strtab,
1657                                     sdp->sd_name) == -1))
1658                                         return (S_ERROR);
1659                                 if (allow_ldynsym && sym->st_name &&
1660                                     ldynsym_symtype[type]) {
1661                                         ofl->ofl_dynscopecnt++;
1662                                         if (st_insert(ofl->ofl_dynstrtab,
1663                                             sdp->sd_name) == -1)
1664                                                 return (S_ERROR);
1665                                         /* Include it in sort section? */
1666                                         DYNSORT_COUNT(sdp, sym, type, ++);
1667                                 }
1668                         }
1669                 } else {
1670                         ofl->ofl_globcnt++;
1671 
1672                         /*
1673                          * Check to see if this global variable should go into
1674                          * a sort section. Sort sections require a
1675                          * .SUNW_ldynsym section, so, don't check unless a
1676                          * .SUNW_ldynsym is allowed.
1677                          */
1678                         if (allow_ldynsym)
1679                                 DYNSORT_COUNT(sdp, sym, type, ++);
1680 
1681                         /*
1682                          * If global direct bindings are in effect, or this
1683                          * symbol has bound to a dependency which was specified
1684                          * as requiring direct bindings, and it hasn't
1685                          * explicitly been defined as a non-direct binding
1686                          * symbol, mark it.
1687                          */
1688                         if (((ofl->ofl_dtflags_1 & DF_1_DIRECT) || (isp &&
1689                             (isp->is_file->ifl_flags & FLG_IF_DIRECT))) &&
1690                             ((sdp->sd_flags & FLG_SY_NDIR) == 0))
1691                                 sdp->sd_flags |= FLG_SY_DIR;
1692 
1693                         /*
1694                          * Insert the symbol name.
1695                          */
1696                         if (((sdp->sd_flags & FLG_SY_REGSYM) == 0) ||
1697                             sym->st_name) {
1698                                 if (st_insert(ofl->ofl_strtab,
1699                                     sdp->sd_name) == -1)
1700                                         return (S_ERROR);
1701 
1702                                 if (!(ofl->ofl_flags & FLG_OF_RELOBJ) &&
1703                                     (st_insert(ofl->ofl_dynstrtab,
1704                                     sdp->sd_name) == -1))
1705                                         return (S_ERROR);
1706                         }
1707 
1708                         /*
1709                          * If this section offers a global symbol - record that
1710                          * fact.
1711                          */
1712                         if (isp) {
1713                                 isp->is_flags |= FLG_IS_SECTREF;
1714                                 isp->is_file->ifl_flags |= FLG_IF_FILEREF;
1715                         }
1716                 }
1717         }
1718 
1719         /*
1720          * Guidance: Use -z defs|nodefs when building shared objects.
1721          *
1722          * Our caller issues this, unless we mask it out here. So we mask it
1723          * out unless we've issued at least one warnings or fatal error.
1724          */
1725         if (!((oflags & FLG_OF_SHAROBJ) && OFL_GUIDANCE(ofl, FLG_OFG_NO_DEFS) &&
1726             (undef_state & (FLG_OF_FATAL | FLG_OF_WARN))))
1727                 ofl->ofl_guideflags |= FLG_OFG_NO_DEFS;
1728 
1729         /*
1730          * If we've encountered a fatal error during symbol validation then
1731          * return now.
1732          */
1733         if (ofl->ofl_flags & FLG_OF_FATAL)
1734                 return (1);
1735 
1736         /*
1737          * Now that symbol resolution is completed, scan any register symbols.
1738          * From now on, we're only interested in those that contribute to the
1739          * output file.
1740          */
1741         if (ofl->ofl_regsyms) {
1742                 int     ndx;
1743 
1744                 for (ndx = 0; ndx < ofl->ofl_regsymsno; ndx++) {
1745                         if ((sdp = ofl->ofl_regsyms[ndx]) == NULL)
1746                                 continue;
1747                         if (sdp->sd_ref != REF_REL_NEED) {
1748                                 ofl->ofl_regsyms[ndx] = NULL;
1749                                 continue;
1750                         }
1751 
1752                         ofl->ofl_regsymcnt++;
1753                         if (sdp->sd_sym->st_name == 0)
1754                                 sdp->sd_name = MSG_ORIG(MSG_STR_EMPTY);
1755 
1756                         if (SYM_IS_HIDDEN(sdp) ||
1757                             (ELF_ST_BIND(sdp->sd_sym->st_info) == STB_LOCAL))
1758                                 ofl->ofl_lregsymcnt++;
1759                 }
1760         }
1761 
1762         /*
1763          * Generate the .bss section now that we know its size and alignment.
1764          */
1765         if (need_bss) {
1766                 if (ld_make_bss(ofl, bsssize, bssalign,
1767                     ld_targ.t_id.id_bss) == S_ERROR)
1768                         return (S_ERROR);
1769         }
1770         if (need_tlsbss) {
1771                 if (ld_make_bss(ofl, tlssize, tlsalign,
1772                     ld_targ.t_id.id_tlsbss) == S_ERROR)
1773                         return (S_ERROR);
1774         }
1775 #if     defined(_ELF64)
1776         if ((ld_targ.t_m.m_mach == EM_AMD64) &&
1777             need_lbss && !(oflags & FLG_OF_RELOBJ)) {
1778                 if (ld_make_bss(ofl, lbsssize, lbssalign,
1779                     ld_targ.t_id.id_lbss) == S_ERROR)
1780                         return (S_ERROR);
1781         }
1782 #endif
1783         /*
1784          * Determine what entry point symbol we need, and if found save its
1785          * symbol descriptor so that we can update the ELF header entry with the
1786          * symbols value later (see update_oehdr).  Make sure the symbol is
1787          * tagged to ensure its update in case -s is in effect.  Use any -e
1788          * option first, or the default entry points `_start' and `main'.
1789          */
1790         ret = 0;
1791         if (ofl->ofl_entry) {
1792                 if ((sdp = ld_sym_find(ofl->ofl_entry, SYM_NOHASH,
1793                     NULL, ofl)) == NULL) {
1794                         ld_eprintf(ofl, ERR_FATAL, MSG_INTL(MSG_ARG_NOENTRY),
1795                             ofl->ofl_entry);
1796                         ret++;
1797                 } else if (ensure_sym_local(ofl, sdp,
1798                     MSG_INTL(MSG_SYM_ENTRY)) != 0) {
1799                         ret++;
1800                 } else {
1801                         ofl->ofl_entry = (void *)sdp;
1802                 }
1803         } else if (((sdp = ld_sym_find(MSG_ORIG(MSG_SYM_START),
1804             SYM_NOHASH, NULL, ofl)) != NULL) && (ensure_sym_local(ofl,
1805             sdp, 0) == 0)) {
1806                 ofl->ofl_entry = (void *)sdp;
1807 
1808         } else if (((sdp = ld_sym_find(MSG_ORIG(MSG_SYM_MAIN),
1809             SYM_NOHASH, NULL, ofl)) != NULL) && (ensure_sym_local(ofl,
1810             sdp, 0) == 0)) {
1811                 ofl->ofl_entry = (void *)sdp;
1812         }
1813 
1814         /*
1815          * If ld -zdtrace=<sym> was given, then validate that the symbol is
1816          * defined within the current object being built.
1817          */
1818         if ((sdp = ofl->ofl_dtracesym) != 0)
1819                 ret += ensure_sym_local(ofl, sdp, MSG_ORIG(MSG_STR_DTRACE));
1820 
1821         /*
1822          * If any initarray, finiarray or preinitarray functions have been
1823          * requested, make sure they are defined within the current object
1824          * being built.
1825          */
1826         if (ofl->ofl_initarray) {
1827                 ret += ensure_array_local(ofl, ofl->ofl_initarray,
1828                     MSG_ORIG(MSG_SYM_INITARRAY));
1829         }
1830         if (ofl->ofl_finiarray) {
1831                 ret += ensure_array_local(ofl, ofl->ofl_finiarray,
1832                     MSG_ORIG(MSG_SYM_FINIARRAY));
1833         }
1834         if (ofl->ofl_preiarray) {
1835                 ret += ensure_array_local(ofl, ofl->ofl_preiarray,
1836                     MSG_ORIG(MSG_SYM_PREINITARRAY));
1837         }
1838 
1839         if (ret)
1840                 return (S_ERROR);
1841 
1842         /*
1843          * If we're required to record any needed dependencies versioning
1844          * information calculate it now that all symbols have been validated.
1845          */
1846         if ((oflags & (FLG_OF_VERNEED | FLG_OF_NOVERSEC)) == FLG_OF_VERNEED)
1847                 return (ld_vers_check_need(ofl));
1848         else
1849                 return (1);
1850 }
1851 
1852 /*
1853  * qsort(3c) comparison function.  As an optimization for associating weak
1854  * symbols to their strong counterparts sort global symbols according to their
1855  * section index, address and binding.
1856  */
1857 static int
1858 compare(const void *sdpp1, const void *sdpp2)
1859 {
1860         Sym_desc        *sdp1 = *((Sym_desc **)sdpp1);
1861         Sym_desc        *sdp2 = *((Sym_desc **)sdpp2);
1862         Sym             *sym1, *sym2;
1863         uchar_t         bind1, bind2;
1864 
1865         /*
1866          * Symbol descriptors may be zero, move these to the front of the
1867          * sorted array.
1868          */
1869         if (sdp1 == NULL)
1870                 return (-1);
1871         if (sdp2 == NULL)
1872                 return (1);
1873 
1874         sym1 = sdp1->sd_sym;
1875         sym2 = sdp2->sd_sym;
1876 
1877         /*
1878          * Compare the symbols section index.  This is important when sorting
1879          * the symbol tables of relocatable objects.  In this case, a symbols
1880          * value is the offset within the associated section, and thus many
1881          * symbols can have the same value, but are effectively different
1882          * addresses.
1883          */
1884         if (sym1->st_shndx > sym2->st_shndx)
1885                 return (1);
1886         if (sym1->st_shndx < sym2->st_shndx)
1887                 return (-1);
1888 
1889         /*
1890          * Compare the symbols value (address).
1891          */
1892         if (sym1->st_value > sym2->st_value)
1893                 return (1);
1894         if (sym1->st_value < sym2->st_value)
1895                 return (-1);
1896 
1897         bind1 = ELF_ST_BIND(sym1->st_info);
1898         bind2 = ELF_ST_BIND(sym2->st_info);
1899 
1900         /*
1901          * If two symbols have the same address place the weak symbol before
1902          * any strong counterpart.
1903          */
1904         if (bind1 > bind2)
1905                 return (-1);
1906         if (bind1 < bind2)
1907                 return (1);
1908 
1909         return (0);
1910 }
1911 
1912 /*
1913  * Issue a MSG_SYM_BADADDR error from ld_sym_process(). This error
1914  * is issued when a symbol address/size is not contained by the
1915  * target section.
1916  *
1917  * Such objects are at least partially corrupt, and the user would
1918  * be well advised to be skeptical of them, and to ask their compiler
1919  * supplier to fix the problem. However, a distinction needs to be
1920  * made between symbols that reference readonly text, and those that
1921  * access writable data. Other than throwing off profiling results,
1922  * the readonly section case is less serious. We have encountered
1923  * such objects in the field. In order to allow existing objects
1924  * to continue working, we issue a warning rather than a fatal error
1925  * if the symbol is against readonly text. Other cases are fatal.
1926  */
1927 static void
1928 issue_badaddr_msg(Ifl_desc *ifl, Ofl_desc *ofl, Sym_desc *sdp,
1929     Sym *sym, Word shndx)
1930 {
1931         Error           err;
1932         const char      *msg;
1933 
1934         if ((sdp->sd_isc->is_shdr->sh_flags & (SHF_WRITE | SHF_ALLOC)) ==
1935             SHF_ALLOC) {
1936                 msg = MSG_INTL(MSG_SYM_BADADDR_ROTXT);
1937                 err = ERR_WARNING;
1938         } else {
1939                 msg = MSG_INTL(MSG_SYM_BADADDR);
1940                 err = ERR_FATAL;
1941         }
1942 
1943         ld_eprintf(ofl, err, msg, demangle(sdp->sd_name),
1944             ifl->ifl_name, shndx, sdp->sd_isc->is_name,
1945             EC_XWORD(sdp->sd_isc->is_shdr->sh_size),
1946             EC_XWORD(sym->st_value), EC_XWORD(sym->st_size));
1947 }
1948 
1949 /*
1950  * Global symbols that are candidates for translation to local capability
1951  * symbols under -z symbolcap, are maintained on a local symbol list.  Once
1952  * all symbols of a file are processed, this list is traversed to cull any
1953  * unnecessary weak symbol aliases.
1954  */
1955 typedef struct {
1956         Sym_desc        *c_nsdp;        /* new lead symbol */
1957         Sym_desc        *c_osdp;        /* original symbol */
1958         Cap_group       *c_group;       /* symbol capability group */
1959         Word            c_ndx;          /* symbol index */
1960 } Cap_pair;
1961 
1962 /*
1963  * Process the symbol table for the specified input file.  At this point all
1964  * input sections from this input file have been assigned an input section
1965  * descriptor which is saved in the `ifl_isdesc' array.
1966  *
1967  *  -   local symbols are saved (as is) if the input file is a  relocatable
1968  *      object
1969  *
1970  *  -   global symbols are added to the linkers internal symbol table if they
1971  *      are not already present, otherwise a symbol resolution function is
1972  *      called upon to resolve the conflict.
1973  */
1974 uintptr_t
1975 ld_sym_process(Is_desc *isc, Ifl_desc *ifl, Ofl_desc *ofl)
1976 {
1977         /*
1978          * This macro tests the given symbol to see if it is out of
1979          * range relative to the section it references.
1980          *
1981          * entry:
1982          *      - ifl is a relative object (ET_REL)
1983          *      _sdp - Symbol descriptor
1984          *      _sym - Symbol
1985          *      _type - Symbol type
1986          *
1987          * The following are tested:
1988          *      - Symbol length is non-zero
1989          *      - Symbol type is a type that references code or data
1990          *      - Referenced section is not 0 (indicates an UNDEF symbol)
1991          *        and is not in the range of special values above SHN_LORESERVE
1992          *        (excluding SHN_XINDEX, which is OK).
1993          *      - We have a valid section header for the target section
1994          *
1995          * If the above are all true, and the symbol position is not
1996          * contained by the target section, this macro evaluates to
1997          * True (1). Otherwise, False(0).
1998          */
1999 #define SYM_LOC_BADADDR(_sdp, _sym, _type) \
2000         (_sym->st_size && dynsymsort_symtype[_type] && \
2001         (_sym->st_shndx != SHN_UNDEF) && \
2002         ((_sym->st_shndx < SHN_LORESERVE) || \
2003                 (_sym->st_shndx == SHN_XINDEX)) && \
2004         _sdp->sd_isc && _sdp->sd_isc->is_shdr && \
2005         ((_sym->st_value + _sym->st_size) > _sdp->sd_isc->is_shdr->sh_size))
2006 
2007         Conv_inv_buf_t  inv_buf;
2008         Sym             *sym = (Sym *)isc->is_indata->d_buf;
2009         Word            *symshndx = NULL;
2010         Shdr            *shdr = isc->is_shdr;
2011         Sym_desc        *sdp;
2012         size_t          strsize;
2013         char            *strs;
2014         uchar_t         type, bind;
2015         Word            ndx, hash, local, total;
2016         uchar_t         osabi = ifl->ifl_ehdr->e_ident[EI_OSABI];
2017         Half            mach = ifl->ifl_ehdr->e_machine;
2018         Half            etype = ifl->ifl_ehdr->e_type;
2019         int             etype_rel;
2020         const char      *symsecname, *strsecname;
2021         Word            symsecndx;
2022         avl_index_t     where;
2023         int             test_gnu_hidden_bit, weak;
2024         Cap_desc        *cdp = NULL;
2025         Alist           *cappairs = NULL;
2026 
2027         /*
2028          * Its possible that a file may contain more that one symbol table,
2029          * ie. .dynsym and .symtab in a shared library.  Only process the first
2030          * table (here, we assume .dynsym comes before .symtab).
2031          */
2032         if (ifl->ifl_symscnt)
2033                 return (1);
2034 
2035         if (isc->is_symshndx)
2036                 symshndx = isc->is_symshndx->is_indata->d_buf;
2037 
2038         DBG_CALL(Dbg_syms_process(ofl->ofl_lml, ifl));
2039 
2040         symsecndx = isc->is_scnndx;
2041         if (isc->is_name)
2042                 symsecname = isc->is_name;
2043         else
2044                 symsecname = MSG_ORIG(MSG_STR_EMPTY);
2045 
2046         /*
2047          * From the symbol tables section header information determine which
2048          * strtab table is needed to locate the actual symbol names.
2049          */
2050         if (ifl->ifl_flags & FLG_IF_HSTRTAB) {
2051                 ndx = shdr->sh_link;
2052                 if ((ndx == 0) || (ndx >= ifl->ifl_shnum)) {
2053                         ld_eprintf(ofl, ERR_FATAL,
2054                             MSG_INTL(MSG_FIL_INVSHLINK), ifl->ifl_name,
2055                             EC_WORD(symsecndx), symsecname, EC_XWORD(ndx));
2056                         return (S_ERROR);
2057                 }
2058                 strsize = ifl->ifl_isdesc[ndx]->is_shdr->sh_size;
2059                 strs = ifl->ifl_isdesc[ndx]->is_indata->d_buf;
2060                 if (ifl->ifl_isdesc[ndx]->is_name)
2061                         strsecname = ifl->ifl_isdesc[ndx]->is_name;
2062                 else
2063                         strsecname = MSG_ORIG(MSG_STR_EMPTY);
2064         } else {
2065                 /*
2066                  * There is no string table section in this input file
2067                  * although there are symbols in this symbol table section.
2068                  * This means that these symbols do not have names.
2069                  * Currently, only scratch register symbols are allowed
2070                  * not to have names.
2071                  */
2072                 strsize = 0;
2073                 strs = (char *)MSG_ORIG(MSG_STR_EMPTY);
2074                 strsecname = MSG_ORIG(MSG_STR_EMPTY);
2075         }
2076 
2077         /*
2078          * Determine the number of local symbols together with the total
2079          * number we have to process.
2080          */
2081         total = (Word)(shdr->sh_size / shdr->sh_entsize);
2082         local = shdr->sh_info;
2083 
2084         /*
2085          * Allocate a symbol table index array and a local symbol array
2086          * (global symbols are processed and added to the ofl->ofl_symbkt[]
2087          * array).  If we are dealing with a relocatable object, allocate the
2088          * local symbol descriptors.  If this isn't a relocatable object we
2089          * still have to process any shared object locals to determine if any
2090          * register symbols exist.  Although these aren't added to the output
2091          * image, they are used as part of symbol resolution.
2092          */
2093         if ((ifl->ifl_oldndx = libld_malloc((size_t)(total *
2094             sizeof (Sym_desc *)))) == NULL)
2095                 return (S_ERROR);
2096         etype_rel = (etype == ET_REL);
2097         if (etype_rel && local) {
2098                 if ((ifl->ifl_locs =
2099                     libld_calloc(sizeof (Sym_desc), local)) == NULL)
2100                         return (S_ERROR);
2101                 /* LINTED */
2102                 ifl->ifl_locscnt = (Word)local;
2103         }
2104         ifl->ifl_symscnt = total;
2105 
2106         /*
2107          * If there are local symbols to save add them to the symbol table
2108          * index array.
2109          */
2110         if (local) {
2111                 int             allow_ldynsym = OFL_ALLOW_LDYNSYM(ofl);
2112                 Sym_desc        *last_file_sdp = NULL;
2113                 int             last_file_ndx = 0;
2114 
2115                 for (sym++, ndx = 1; ndx < local; sym++, ndx++) {
2116                         sd_flag_t       sdflags = FLG_SY_CLEAN;
2117                         Word            shndx;
2118                         const char      *name;
2119                         Sym_desc        *rsdp;
2120                         int             shndx_bad = 0;
2121                         int             symtab_enter = 1;
2122 
2123                         /*
2124                          * Determine and validate the associated section index.
2125                          */
2126                         if (symshndx && (sym->st_shndx == SHN_XINDEX)) {
2127                                 shndx = symshndx[ndx];
2128                         } else if ((shndx = sym->st_shndx) >= SHN_LORESERVE) {
2129                                 sdflags |= FLG_SY_SPECSEC;
2130                         } else if (shndx > ifl->ifl_shnum) {
2131                                 /* We need the name before we can issue error */
2132                                 shndx_bad = 1;
2133                         }
2134 
2135                         /*
2136                          * Check if st_name has a valid value or not.
2137                          */
2138                         if ((name = string(ofl, ifl, sym, strs, strsize, ndx,
2139                             shndx, symsecndx, symsecname, strsecname,
2140                             &sdflags)) == NULL)
2141                                 continue;
2142 
2143                         /*
2144                          * Now that we have the name, if the section index
2145                          * was bad, report it.
2146                          */
2147                         if (shndx_bad) {
2148                                 ld_eprintf(ofl, ERR_WARNING,
2149                                     MSG_INTL(MSG_SYM_INVSHNDX),
2150                                     demangle_symname(name, symsecname, ndx),
2151                                     ifl->ifl_name,
2152                                     conv_sym_shndx(osabi, mach, sym->st_shndx,
2153                                     CONV_FMT_DECIMAL, &inv_buf));
2154                                 continue;
2155                         }
2156 
2157                         /*
2158                          * If this local symbol table originates from a shared
2159                          * object, then we're only interested in recording
2160                          * register symbols.  As local symbol descriptors aren't
2161                          * allocated for shared objects, one will be allocated
2162                          * to associated with the register symbol.  This symbol
2163                          * won't become part of the output image, but we must
2164                          * process it to test for register conflicts.
2165                          */
2166                         rsdp = sdp = NULL;
2167                         if (sdflags & FLG_SY_REGSYM) {
2168                                 /*
2169                                  * The presence of FLG_SY_REGSYM means that
2170                                  * the pointers in ld_targ.t_ms are non-NULL.
2171                                  */
2172                                 rsdp = (*ld_targ.t_ms.ms_reg_find)(sym, ofl);
2173                                 if (rsdp != 0) {
2174                                         /*
2175                                          * The fact that another register def-
2176                                          * inition has been found is fatal.
2177                                          * Call the verification routine to get
2178                                          * the error message and move on.
2179                                          */
2180                                         (void) (*ld_targ.t_ms.ms_reg_check)
2181                                             (rsdp, sym, name, ifl, ofl);
2182                                         continue;
2183                                 }
2184 
2185                                 if (etype == ET_DYN) {
2186                                         if ((sdp = libld_calloc(
2187                                             sizeof (Sym_desc), 1)) == NULL)
2188                                                 return (S_ERROR);
2189                                         sdp->sd_ref = REF_DYN_SEEN;
2190 
2191                                         /* Will not appear in output object */
2192                                         symtab_enter = 0;
2193                                 }
2194                         } else if (etype == ET_DYN)
2195                                 continue;
2196 
2197                         /*
2198                          * Fill in the remaining symbol descriptor information.
2199                          */
2200                         if (sdp == NULL) {
2201                                 sdp = &(ifl->ifl_locs[ndx]);
2202                                 sdp->sd_ref = REF_REL_NEED;
2203                                 sdp->sd_symndx = ndx;
2204                         }
2205                         if (rsdp == NULL) {
2206                                 sdp->sd_name = name;
2207                                 sdp->sd_sym = sym;
2208                                 sdp->sd_shndx = shndx;
2209                                 sdp->sd_flags = sdflags;
2210                                 sdp->sd_file = ifl;
2211                                 ifl->ifl_oldndx[ndx] = sdp;
2212                         }
2213 
2214                         DBG_CALL(Dbg_syms_entry(ofl->ofl_lml, ndx, sdp));
2215 
2216                         /*
2217                          * Reclassify any SHN_SUNW_IGNORE symbols to SHN_UNDEF
2218                          * so as to simplify future processing.
2219                          */
2220                         if (sym->st_shndx == SHN_SUNW_IGNORE) {
2221                                 sdp->sd_shndx = shndx = SHN_UNDEF;
2222                                 sdp->sd_flags |= (FLG_SY_IGNORE | FLG_SY_ELIM);
2223                         }
2224 
2225                         /*
2226                          * Process any register symbols.
2227                          */
2228                         if (sdp->sd_flags & FLG_SY_REGSYM) {
2229                                 /*
2230                                  * Add a diagnostic to indicate we've caught a
2231                                  * register symbol, as this can be useful if a
2232                                  * register conflict is later discovered.
2233                                  */
2234                                 DBG_CALL(Dbg_syms_entered(ofl, sym, sdp));
2235 
2236                                 /*
2237                                  * If this register symbol hasn't already been
2238                                  * recorded, enter it now.
2239                                  *
2240                                  * The presence of FLG_SY_REGSYM means that
2241                                  * the pointers in ld_targ.t_ms are non-NULL.
2242                                  */
2243                                 if ((rsdp == NULL) &&
2244                                     ((*ld_targ.t_ms.ms_reg_enter)(sdp, ofl) ==
2245                                     0))
2246                                         return (S_ERROR);
2247                         }
2248 
2249                         /*
2250                          * Assign an input section.
2251                          */
2252                         if ((sym->st_shndx != SHN_UNDEF) &&
2253                             ((sdp->sd_flags & FLG_SY_SPECSEC) == 0))
2254                                 sdp->sd_isc = ifl->ifl_isdesc[shndx];
2255 
2256                         /*
2257                          * If this symbol falls within the range of a section
2258                          * being discarded, then discard the symbol itself.
2259                          * There is no reason to keep this local symbol.
2260                          */
2261                         if (sdp->sd_isc &&
2262                             (sdp->sd_isc->is_flags & FLG_IS_DISCARD)) {
2263                                 sdp->sd_flags |= FLG_SY_ISDISC;
2264                                 DBG_CALL(Dbg_syms_discarded(ofl->ofl_lml, sdp));
2265                                 continue;
2266                         }
2267 
2268                         /*
2269                          * Skip any section symbols as new versions of these
2270                          * will be created.
2271                          */
2272                         if ((type = ELF_ST_TYPE(sym->st_info)) == STT_SECTION) {
2273                                 if (sym->st_shndx == SHN_UNDEF) {
2274                                         ld_eprintf(ofl, ERR_WARNING,
2275                                             MSG_INTL(MSG_SYM_INVSHNDX),
2276                                             demangle_symname(name, symsecname,
2277                                             ndx), ifl->ifl_name,
2278                                             conv_sym_shndx(osabi, mach,
2279                                             sym->st_shndx, CONV_FMT_DECIMAL,
2280                                             &inv_buf));
2281                                 }
2282                                 continue;
2283                         }
2284 
2285                         /*
2286                          * For a relocatable object, if this symbol is defined
2287                          * and has non-zero length and references an address
2288                          * within an associated section, then check its extents
2289                          * to make sure the section boundaries encompass it.
2290                          * If they don't, the ELF file is corrupt.
2291                          */
2292                         if (etype_rel) {
2293                                 if (SYM_LOC_BADADDR(sdp, sym, type)) {
2294                                         issue_badaddr_msg(ifl, ofl, sdp,
2295                                             sym, shndx);
2296                                         if (ofl->ofl_flags & FLG_OF_FATAL)
2297                                                 continue;
2298                                 }
2299 
2300                                 /*
2301                                  * We have observed relocatable objects
2302                                  * containing identical adjacent STT_FILE
2303                                  * symbols. Discard any other than the first,
2304                                  * as they are all equivalent and the extras
2305                                  * do not add information.
2306                                  *
2307                                  * For the purpose of this test, we assume
2308                                  * that only the symbol type and the string
2309                                  * table offset (st_name) matter.
2310                                  */
2311                                 if (type == STT_FILE) {
2312                                         int toss = (last_file_sdp != NULL) &&
2313                                             ((ndx - 1) == last_file_ndx) &&
2314                                             (sym->st_name ==
2315                                             last_file_sdp->sd_sym->st_name);
2316 
2317                                         last_file_sdp = sdp;
2318                                         last_file_ndx = ndx;
2319                                         if (toss) {
2320                                                 sdp->sd_flags |= FLG_SY_INVALID;
2321                                                 DBG_CALL(Dbg_syms_dup_discarded(
2322                                                     ofl->ofl_lml, ndx, sdp));
2323                                                 continue;
2324                                         }
2325                                 }
2326                         }
2327 
2328 
2329                         /*
2330                          * Sanity check for TLS
2331                          */
2332                         if ((sym->st_size != 0) && ((type == STT_TLS) &&
2333                             (sym->st_shndx != SHN_COMMON))) {
2334                                 Is_desc *isp = sdp->sd_isc;
2335 
2336                                 if ((isp == NULL) || (isp->is_shdr == NULL) ||
2337                                     ((isp->is_shdr->sh_flags & SHF_TLS) == 0)) {
2338                                         ld_eprintf(ofl, ERR_FATAL,
2339                                             MSG_INTL(MSG_SYM_TLS),
2340                                             demangle(sdp->sd_name),
2341                                             ifl->ifl_name);
2342                                         continue;
2343                                 }
2344                         }
2345 
2346                         /*
2347                          * Carry our some basic sanity checks (these are just
2348                          * some of the erroneous symbol entries we've come
2349                          * across, there's probably a lot more).  The symbol
2350                          * will not be carried forward to the output file, which
2351                          * won't be a problem unless a relocation is required
2352                          * against it.
2353                          */
2354                         if (((sdp->sd_flags & FLG_SY_SPECSEC) &&
2355                             ((sym->st_shndx == SHN_COMMON)) ||
2356                             ((type == STT_FILE) &&
2357                             (sym->st_shndx != SHN_ABS))) ||
2358                             (sdp->sd_isc && (sdp->sd_isc->is_osdesc == NULL))) {
2359                                 ld_eprintf(ofl, ERR_WARNING,
2360                                     MSG_INTL(MSG_SYM_INVSHNDX),
2361                                     demangle_symname(name, symsecname, ndx),
2362                                     ifl->ifl_name,
2363                                     conv_sym_shndx(osabi, mach, sym->st_shndx,
2364                                     CONV_FMT_DECIMAL, &inv_buf));
2365                                 sdp->sd_isc = NULL;
2366                                 sdp->sd_flags |= FLG_SY_INVALID;
2367                                 continue;
2368                         }
2369 
2370                         /*
2371                          * As these local symbols will become part of the output
2372                          * image, record their number and name string size.
2373                          * Globals are counted after all input file processing
2374                          * (and hence symbol resolution) is complete during
2375                          * sym_validate().
2376                          */
2377                         if (!(ofl->ofl_flags & FLG_OF_REDLSYM) &&
2378                             symtab_enter) {
2379                                 ofl->ofl_locscnt++;
2380 
2381                                 if ((((sdp->sd_flags & FLG_SY_REGSYM) == 0) ||
2382                                     sym->st_name) && (st_insert(ofl->ofl_strtab,
2383                                     sdp->sd_name) == -1))
2384                                         return (S_ERROR);
2385 
2386                                 if (allow_ldynsym && sym->st_name &&
2387                                     ldynsym_symtype[type]) {
2388                                         ofl->ofl_dynlocscnt++;
2389                                         if (st_insert(ofl->ofl_dynstrtab,
2390                                             sdp->sd_name) == -1)
2391                                                 return (S_ERROR);
2392                                         /* Include it in sort section? */
2393                                         DYNSORT_COUNT(sdp, sym, type, ++);
2394                                 }
2395                         }
2396                 }
2397         }
2398 
2399         /*
2400          * The GNU ld interprets the top bit of the 16-bit Versym value
2401          * (0x8000) as the "hidden" bit. If this bit is set, the linker
2402          * is supposed to act as if that symbol does not exist. The Solaris
2403          * linker does not support this mechanism, or the model of interface
2404          * evolution that it allows, but we honor it in GNU ld produced
2405          * objects in order to interoperate with them.
2406          *
2407          * Determine if we should honor the GNU hidden bit for this file.
2408          */
2409         test_gnu_hidden_bit = ((ifl->ifl_flags & FLG_IF_GNUVER) != 0) &&
2410             (ifl->ifl_versym != NULL);
2411 
2412         /*
2413          * Determine whether object capabilities for this file are being
2414          * converted into symbol capabilities.  If so, global function symbols,
2415          * and initialized global data symbols, need special translation and
2416          * processing.
2417          */
2418         if ((etype == ET_REL) && (ifl->ifl_flags & FLG_IF_OTOSCAP))
2419                 cdp = ifl->ifl_caps;
2420 
2421         /*
2422          * Now scan the global symbols entering them in the internal symbol
2423          * table or resolving them as necessary.
2424          */
2425         sym = (Sym *)isc->is_indata->d_buf;
2426         sym += local;
2427         weak = 0;
2428         /* LINTED */
2429         for (ndx = (int)local; ndx < total; sym++, ndx++) {
2430                 const char      *name;
2431                 sd_flag_t       sdflags = 0;
2432                 Word            shndx;
2433                 int             shndx_bad = 0;
2434                 Sym             *nsym = sym;
2435                 Cap_pair        *cpp = NULL;
2436                 uchar_t         ntype;
2437 
2438                 /*
2439                  * Determine and validate the associated section index.
2440                  */
2441                 if (symshndx && (nsym->st_shndx == SHN_XINDEX)) {
2442                         shndx = symshndx[ndx];
2443                 } else if ((shndx = nsym->st_shndx) >= SHN_LORESERVE) {
2444                         sdflags |= FLG_SY_SPECSEC;
2445                 } else if (shndx > ifl->ifl_shnum) {
2446                         /* We need the name before we can issue error */
2447                         shndx_bad = 1;
2448                 }
2449 
2450                 /*
2451                  * Check if st_name has a valid value or not.
2452                  */
2453                 if ((name = string(ofl, ifl, nsym, strs, strsize, ndx, shndx,
2454                     symsecndx, symsecname, strsecname, &sdflags)) == NULL)
2455                         continue;
2456 
2457                 /*
2458                  * Now that we have the name, report an erroneous section index.
2459                  */
2460                 if (shndx_bad) {
2461                         ld_eprintf(ofl, ERR_WARNING, MSG_INTL(MSG_SYM_INVSHNDX),
2462                             demangle_symname(name, symsecname, ndx),
2463                             ifl->ifl_name,
2464                             conv_sym_shndx(osabi, mach, nsym->st_shndx,
2465                             CONV_FMT_DECIMAL, &inv_buf));
2466                         continue;
2467                 }
2468 
2469                 /*
2470                  * Test for the GNU hidden bit, and ignore symbols that
2471                  * have it set.
2472                  */
2473                 if (test_gnu_hidden_bit &&
2474                     ((ifl->ifl_versym[ndx] & 0x8000) != 0))
2475                         continue;
2476 
2477                 /*
2478                  * The linker itself will generate symbols for _end, _etext,
2479                  * _edata, _DYNAMIC and _PROCEDURE_LINKAGE_TABLE_, so don't
2480                  * bother entering these symbols from shared objects.  This
2481                  * results in some wasted resolution processing, which is hard
2482                  * to feel, but if nothing else, pollutes diagnostic relocation
2483                  * output.
2484                  */
2485                 if (name[0] && (etype == ET_DYN) && (nsym->st_size == 0) &&
2486                     (ELF_ST_TYPE(nsym->st_info) == STT_OBJECT) &&
2487                     (name[0] == '_') && ((name[1] == 'e') ||
2488                     (name[1] == 'D') || (name[1] == 'P')) &&
2489                     ((strcmp(name, MSG_ORIG(MSG_SYM_ETEXT_U)) == 0) ||
2490                     (strcmp(name, MSG_ORIG(MSG_SYM_EDATA_U)) == 0) ||
2491                     (strcmp(name, MSG_ORIG(MSG_SYM_END_U)) == 0) ||
2492                     (strcmp(name, MSG_ORIG(MSG_SYM_DYNAMIC_U)) == 0) ||
2493                     (strcmp(name, MSG_ORIG(MSG_SYM_PLKTBL_U)) == 0))) {
2494                         ifl->ifl_oldndx[ndx] = 0;
2495                         continue;
2496                 }
2497 
2498                 /*
2499                  * The '-z wrap=XXX' option emulates the GNU ld --wrap=XXX
2500                  * option. When XXX is the symbol to be wrapped:
2501                  *
2502                  *  -   An undefined reference to XXX is converted to __wrap_XXX
2503                  *  -   An undefined reference to __real_XXX is converted to XXX
2504                  *
2505                  * The idea is that the user can supply a wrapper function
2506                  * __wrap_XXX that does some work, and then uses the name
2507                  * __real_XXX to pass the call on to the real function. The
2508                  * wrapper objects are linked with the original unmodified
2509                  * objects to produce a wrapped version of the output object.
2510                  */
2511                 if (ofl->ofl_wrap && name[0] && (shndx == SHN_UNDEF)) {
2512                         WrapSymNode wsn, *wsnp;
2513 
2514                         /*
2515                          * If this is the __real_XXX form, advance the
2516                          * pointer to reference the wrapped name.
2517                          */
2518                         wsn.wsn_name = name;
2519                         if ((*name == '_') &&
2520                             (strncmp(name, MSG_ORIG(MSG_STR_UU_REAL_U),
2521                             MSG_STR_UU_REAL_U_SIZE) == 0))
2522                                 wsn.wsn_name += MSG_STR_UU_REAL_U_SIZE;
2523 
2524                         /*
2525                          * Is this symbol in the wrap AVL tree? If so, map
2526                          * XXX to __wrap_XXX, and __real_XXX to XXX. Note that
2527                          * wsn.wsn_name will equal the current value of name
2528                          * if the __real_ prefix is not present.
2529                          */
2530                         if ((wsnp = avl_find(ofl->ofl_wrap, &wsn, 0)) != NULL) {
2531                                 const char *old_name = name;
2532 
2533                                 name = (wsn.wsn_name == name) ?
2534                                     wsnp->wsn_wrapname : wsn.wsn_name;
2535                                 DBG_CALL(Dbg_syms_wrap(ofl->ofl_lml, ndx,
2536                                     old_name, name));
2537                         }
2538                 }
2539 
2540                 /*
2541                  * Determine and validate the symbols binding.
2542                  */
2543                 bind = ELF_ST_BIND(nsym->st_info);
2544                 if ((bind != STB_GLOBAL) && (bind != STB_WEAK)) {
2545                         ld_eprintf(ofl, ERR_WARNING, MSG_INTL(MSG_SYM_NONGLOB),
2546                             demangle_symname(name, symsecname, ndx),
2547                             ifl->ifl_name,
2548                             conv_sym_info_bind(bind, 0, &inv_buf));
2549                         continue;
2550                 }
2551                 if (bind == STB_WEAK)
2552                         weak++;
2553 
2554                 /*
2555                  * If this symbol falls within the range of a section being
2556                  * discarded, then discard the symbol itself.
2557                  */
2558                 if (((sdflags & FLG_SY_SPECSEC) == 0) &&
2559                     (nsym->st_shndx != SHN_UNDEF)) {
2560                         Is_desc *isp;
2561 
2562                         if (shndx >= ifl->ifl_shnum) {
2563                                 /*
2564                                  * Carry our some basic sanity checks
2565                                  * The symbol will not be carried forward to
2566                                  * the output file, which won't be a problem
2567                                  * unless a relocation is required against it.
2568                                  */
2569                                 ld_eprintf(ofl, ERR_WARNING,
2570                                     MSG_INTL(MSG_SYM_INVSHNDX),
2571                                     demangle_symname(name, symsecname, ndx),
2572                                     ifl->ifl_name,
2573                                     conv_sym_shndx(osabi, mach, nsym->st_shndx,
2574                                     CONV_FMT_DECIMAL, &inv_buf));
2575                                 continue;
2576                         }
2577 
2578                         isp = ifl->ifl_isdesc[shndx];
2579                         if (isp && (isp->is_flags & FLG_IS_DISCARD)) {
2580                                 if ((sdp =
2581                                     libld_calloc(sizeof (Sym_desc), 1)) == NULL)
2582                                         return (S_ERROR);
2583 
2584                                 /*
2585                                  * Create a dummy symbol entry so that if we
2586                                  * find any references to this discarded symbol
2587                                  * we can compensate.
2588                                  */
2589                                 sdp->sd_name = name;
2590                                 sdp->sd_sym = nsym;
2591                                 sdp->sd_file = ifl;
2592                                 sdp->sd_isc = isp;
2593                                 sdp->sd_flags = FLG_SY_ISDISC;
2594                                 ifl->ifl_oldndx[ndx] = sdp;
2595 
2596                                 DBG_CALL(Dbg_syms_discarded(ofl->ofl_lml, sdp));
2597                                 continue;
2598                         }
2599                 }
2600 
2601                 /*
2602                  * If object capabilities for this file are being converted
2603                  * into symbol capabilities, then:
2604                  *
2605                  *  -   Any global function, or initialized global data symbol
2606                  *      definitions (ie., those that are not associated with
2607                  *      special symbol types, ie., ABS, COMMON, etc.), and which
2608                  *      have not been reduced to locals, are converted to symbol
2609                  *      references (UNDEF).  This ensures that any reference to
2610                  *      the original symbol, for example from a relocation, get
2611                  *      associated to a capabilities family lead symbol, ie., a
2612                  *      generic instance.
2613                  *
2614                  *  -   For each global function, or object symbol definition,
2615                  *      a new local symbol is created.  The function or object
2616                  *      is renamed using the capabilities CA_SUNW_ID definition
2617                  *      (which might have been fabricated for this purpose -
2618                  *      see get_cap_group()).  The new symbol name is:
2619                  *
2620                  *          <original name>%<capability group identifier>
2621                  *
2622                  *      This symbol is associated to the same location, and
2623                  *      becomes a capabilities family member.
2624                  */
2625                 /* LINTED */
2626                 hash = (Word)elf_hash(name);
2627 
2628                 ntype = ELF_ST_TYPE(nsym->st_info);
2629                 if (cdp && (nsym->st_shndx != SHN_UNDEF) &&
2630                     ((sdflags & FLG_SY_SPECSEC) == 0) &&
2631                     ((ntype == STT_FUNC) || (ntype == STT_OBJECT))) {
2632                         /*
2633                          * Determine this symbol's visibility.  If a mapfile has
2634                          * indicated this symbol should be local, then there's
2635                          * no point in transforming this global symbol to a
2636                          * capabilities symbol.  Otherwise, create a symbol
2637                          * capability pair descriptor to record this symbol as
2638                          * a candidate for translation.
2639                          */
2640                         if (sym_cap_vis(name, hash, sym, ofl) &&
2641                             ((cpp = alist_append(&cappairs, NULL,
2642                             sizeof (Cap_pair), AL_CNT_CAP_PAIRS)) == NULL))
2643                                 return (S_ERROR);
2644                 }
2645 
2646                 if (cpp) {
2647                         Sym     *rsym;
2648 
2649                         DBG_CALL(Dbg_syms_cap_convert(ofl, ndx, name, nsym));
2650 
2651                         /*
2652                          * Allocate a new symbol descriptor to represent the
2653                          * transformed global symbol.  The descriptor points
2654                          * to the original symbol information (which might
2655                          * indicate a global or weak visibility).  The symbol
2656                          * information will be transformed into a local symbol
2657                          * later, after any weak aliases are culled.
2658                          */
2659                         if ((cpp->c_osdp =
2660                             libld_malloc(sizeof (Sym_desc))) == NULL)
2661                                 return (S_ERROR);
2662 
2663                         cpp->c_osdp->sd_name = name;
2664                         cpp->c_osdp->sd_sym = nsym;
2665                         cpp->c_osdp->sd_shndx = shndx;
2666                         cpp->c_osdp->sd_file = ifl;
2667                         cpp->c_osdp->sd_isc = ifl->ifl_isdesc[shndx];
2668                         cpp->c_osdp->sd_ref = REF_REL_NEED;
2669 
2670                         /*
2671                          * Save the capabilities group this symbol belongs to,
2672                          * and the original symbol index.
2673                          */
2674                         cpp->c_group = cdp->ca_groups->apl_data[0];
2675                         cpp->c_ndx = ndx;
2676 
2677                         /*
2678                          * Replace the original symbol definition with a symbol
2679                          * reference.  Make sure this reference isn't left as a
2680                          * weak.
2681                          */
2682                         if ((rsym = libld_malloc(sizeof (Sym))) == NULL)
2683                                 return (S_ERROR);
2684 
2685                         *rsym = *nsym;
2686 
2687                         rsym->st_info = ELF_ST_INFO(STB_GLOBAL, ntype);
2688                         rsym->st_shndx = shndx = SHN_UNDEF;
2689                         rsym->st_value = 0;
2690                         rsym->st_size = 0;
2691 
2692                         sdflags |= FLG_SY_CAP;
2693 
2694                         nsym = rsym;
2695                 }
2696 
2697                 /*
2698                  * If the symbol does not already exist in the internal symbol
2699                  * table add it, otherwise resolve the conflict.  If the symbol
2700                  * from this file is kept, retain its symbol table index for
2701                  * possible use in associating a global alias.
2702                  */
2703                 if ((sdp = ld_sym_find(name, hash, &where, ofl)) == NULL) {
2704                         DBG_CALL(Dbg_syms_global(ofl->ofl_lml, ndx, name));
2705                         if ((sdp = ld_sym_enter(name, nsym, hash, ifl, ofl, ndx,
2706                             shndx, sdflags, &where)) == (Sym_desc *)S_ERROR)
2707                                 return (S_ERROR);
2708 
2709                 } else if (ld_sym_resolve(sdp, nsym, ifl, ofl, ndx, shndx,
2710                     sdflags) == S_ERROR)
2711                         return (S_ERROR);
2712 
2713                 /*
2714                  * Now that we have a symbol descriptor, retain the descriptor
2715                  * for later use by symbol capabilities processing.
2716                  */
2717                 if (cpp)
2718                         cpp->c_nsdp = sdp;
2719 
2720                 /*
2721                  * After we've compared a defined symbol in one shared
2722                  * object, flag the symbol so we don't compare it again.
2723                  */
2724                 if ((etype == ET_DYN) && (nsym->st_shndx != SHN_UNDEF) &&
2725                     ((sdp->sd_flags & FLG_SY_SOFOUND) == 0))
2726                         sdp->sd_flags |= FLG_SY_SOFOUND;
2727 
2728                 /*
2729                  * If the symbol is accepted from this file retain the symbol
2730                  * index for possible use in aliasing.
2731                  */
2732                 if (sdp->sd_file == ifl)
2733                         sdp->sd_symndx = ndx;
2734 
2735                 ifl->ifl_oldndx[ndx] = sdp;
2736 
2737                 /*
2738                  * If we've accepted a register symbol, continue to validate
2739                  * it.
2740                  */
2741                 if (sdp->sd_flags & FLG_SY_REGSYM) {
2742                         Sym_desc        *rsdp;
2743 
2744                         /*
2745                          * The presence of FLG_SY_REGSYM means that
2746                          * the pointers in ld_targ.t_ms are non-NULL.
2747                          */
2748                         rsdp = (*ld_targ.t_ms.ms_reg_find)(sdp->sd_sym, ofl);
2749                         if (rsdp == NULL) {
2750                                 if ((*ld_targ.t_ms.ms_reg_enter)(sdp, ofl) == 0)
2751                                         return (S_ERROR);
2752                         } else if (rsdp != sdp) {
2753                                 (void) (*ld_targ.t_ms.ms_reg_check)(rsdp,
2754                                     sdp->sd_sym, sdp->sd_name, ifl, ofl);
2755                         }
2756                 }
2757 
2758                 /*
2759                  * For a relocatable object, if this symbol is defined
2760                  * and has non-zero length and references an address
2761                  * within an associated section, then check its extents
2762                  * to make sure the section boundaries encompass it.
2763                  * If they don't, the ELF file is corrupt. Note that this
2764                  * global symbol may have come from another file to satisfy
2765                  * an UNDEF symbol of the same name from this one. In that
2766                  * case, we don't check it, because it was already checked
2767                  * as part of its own file.
2768                  */
2769                 if (etype_rel && (sdp->sd_file == ifl)) {
2770                         Sym *tsym = sdp->sd_sym;
2771 
2772                         if (SYM_LOC_BADADDR(sdp, tsym,
2773                             ELF_ST_TYPE(tsym->st_info))) {
2774                                 issue_badaddr_msg(ifl, ofl, sdp,
2775                                     tsym, tsym->st_shndx);
2776                                 continue;
2777                         }
2778                 }
2779         }
2780         DBG_CALL(Dbg_util_nl(ofl->ofl_lml, DBG_NL_STD));
2781 
2782         /*
2783          * Associate weak (alias) symbols to their non-weak counterparts by
2784          * scanning the global symbols one more time.
2785          *
2786          * This association is needed when processing the symbols from a shared
2787          * object dependency when a a weak definition satisfies a reference:
2788          *
2789          *  -   When building a dynamic executable, if a referenced symbol is a
2790          *      data item, the symbol data is copied to the executables address
2791          *      space.  In this copy-relocation case, we must also reassociate
2792          *      the alias symbol with its new location in the executable.
2793          *
2794          *  -   If the referenced symbol is a function then we may need to
2795          *      promote the symbols binding from undefined weak to undefined,
2796          *      otherwise the run-time linker will not generate the correct
2797          *      relocation error should the symbol not be found.
2798          *
2799          * Weak alias association is also required when a local dynsym table
2800          * is being created.  This table should only contain one instance of a
2801          * symbol that is associated to a given address.
2802          *
2803          * The true association between a weak/strong symbol pair is that both
2804          * symbol entries are identical, thus first we create a sorted symbol
2805          * list keyed off of the symbols section index and value.  If the symbol
2806          * belongs to the same section and has the same value, then the chances
2807          * are that the rest of the symbols data is the same.  This list is then
2808          * scanned for weak symbols, and if one is found then any strong
2809          * association will exist in the entries that follow.  Thus we just have
2810          * to scan one (typically a single alias) or more (in the uncommon
2811          * instance of multiple weak to strong associations) entries to
2812          * determine if a match exists.
2813          */
2814         if (weak && (OFL_ALLOW_LDYNSYM(ofl) || (etype == ET_DYN)) &&
2815             (total > local)) {
2816                 static Sym_desc **sort;
2817                 static size_t   osize = 0;
2818                 size_t          nsize = (total - local) * sizeof (Sym_desc *);
2819 
2820                 /*
2821                  * As we might be processing many input files, and many symbols,
2822                  * try and reuse a static sort buffer.  Note, presently we're
2823                  * playing the game of never freeing any buffers as there's a
2824                  * belief this wastes time.
2825                  */
2826                 if ((osize == 0) || (nsize > osize)) {
2827                         if ((sort = libld_malloc(nsize)) == NULL)
2828                                 return (S_ERROR);
2829                         osize = nsize;
2830                 }
2831                 (void) memcpy((void *)sort, &ifl->ifl_oldndx[local], nsize);
2832 
2833                 qsort(sort, (total - local), sizeof (Sym_desc *), compare);
2834 
2835                 for (ndx = 0; ndx < (total - local); ndx++) {
2836                         Sym_desc        *wsdp = sort[ndx];
2837                         Sym             *wsym;
2838                         int             sndx;
2839 
2840                         /*
2841                          * Ignore any empty symbol descriptor, or the case where
2842                          * the symbol has been resolved to a different file.
2843                          */
2844                         if ((wsdp == NULL) || (wsdp->sd_file != ifl))
2845                                 continue;
2846 
2847                         wsym = wsdp->sd_sym;
2848 
2849                         if ((wsym->st_shndx == SHN_UNDEF) ||
2850                             (wsdp->sd_flags & FLG_SY_SPECSEC) ||
2851                             (ELF_ST_BIND(wsym->st_info) != STB_WEAK))
2852                                 continue;
2853 
2854                         /*
2855                          * We have a weak symbol, if it has a strong alias it
2856                          * will have been sorted to one of the following sort
2857                          * table entries.  Note that we could have multiple weak
2858                          * symbols aliased to one strong (if this occurs then
2859                          * the strong symbol only maintains one alias back to
2860                          * the last weak).
2861                          */
2862                         for (sndx = ndx + 1; sndx < (total - local); sndx++) {
2863                                 Sym_desc        *ssdp = sort[sndx];
2864                                 Sym             *ssym;
2865                                 sd_flag_t       w_dynbits, s_dynbits;
2866 
2867                                 /*
2868                                  * Ignore any empty symbol descriptor, or the
2869                                  * case where the symbol has been resolved to a
2870                                  * different file.
2871                                  */
2872                                 if ((ssdp == NULL) || (ssdp->sd_file != ifl))
2873                                         continue;
2874 
2875                                 ssym = ssdp->sd_sym;
2876 
2877                                 if (ssym->st_shndx == SHN_UNDEF)
2878                                         continue;
2879 
2880                                 if ((ssym->st_shndx != wsym->st_shndx) ||
2881                                     (ssym->st_value != wsym->st_value))
2882                                         break;
2883 
2884                                 if ((ssym->st_size != wsym->st_size) ||
2885                                     (ssdp->sd_flags & FLG_SY_SPECSEC) ||
2886                                     (ELF_ST_BIND(ssym->st_info) == STB_WEAK))
2887                                         continue;
2888 
2889                                 /*
2890                                  * If a sharable object, set link fields so
2891                                  * that they reference each other.`
2892                                  */
2893                                 if (etype == ET_DYN) {
2894                                         ssdp->sd_aux->sa_linkndx =
2895                                             (Word)wsdp->sd_symndx;
2896                                         wsdp->sd_aux->sa_linkndx =
2897                                             (Word)ssdp->sd_symndx;
2898                                 }
2899 
2900                                 /*
2901                                  * Determine which of these two symbols go into
2902                                  * the sort section.  If a mapfile has made
2903                                  * explicit settings of the FLG_SY_*DYNSORT
2904                                  * flags for both symbols, then we do what they
2905                                  * say.  If one has the DYNSORT flags set, we
2906                                  * set the NODYNSORT bit in the other.  And if
2907                                  * neither has an explicit setting, then we
2908                                  * favor the weak symbol because they usually
2909                                  * lack the leading underscore.
2910                                  */
2911                                 w_dynbits = wsdp->sd_flags &
2912                                     (FLG_SY_DYNSORT | FLG_SY_NODYNSORT);
2913                                 s_dynbits = ssdp->sd_flags &
2914                                     (FLG_SY_DYNSORT | FLG_SY_NODYNSORT);
2915                                 if (!(w_dynbits && s_dynbits)) {
2916                                         if (s_dynbits) {
2917                                                 if (s_dynbits == FLG_SY_DYNSORT)
2918                                                         wsdp->sd_flags |=
2919                                                             FLG_SY_NODYNSORT;
2920                                         } else if (w_dynbits !=
2921                                             FLG_SY_NODYNSORT) {
2922                                                 ssdp->sd_flags |=
2923                                                     FLG_SY_NODYNSORT;
2924                                         }
2925                                 }
2926                                 break;
2927                         }
2928                 }
2929         }
2930 
2931         /*
2932          * Having processed all symbols, under -z symbolcap, reprocess any
2933          * symbols that are being translated from global to locals.  The symbol
2934          * pair that has been collected defines the original symbol (c_osdp),
2935          * which will become a local, and the new symbol (c_nsdp), which will
2936          * become a reference (UNDEF) for the original.
2937          *
2938          * Scan these symbol pairs looking for weak symbols, which have non-weak
2939          * aliases.  There is no need to translate both of these symbols to
2940          * locals, only the global is necessary.
2941          */
2942         if (cappairs) {
2943                 Aliste          idx1;
2944                 Cap_pair        *cpp1;
2945 
2946                 for (ALIST_TRAVERSE(cappairs, idx1, cpp1)) {
2947                         Sym_desc        *sdp1 = cpp1->c_osdp;
2948                         Sym             *sym1 = sdp1->sd_sym;
2949                         uchar_t         bind1 = ELF_ST_BIND(sym1->st_info);
2950                         Aliste          idx2;
2951                         Cap_pair        *cpp2;
2952 
2953                         /*
2954                          * If this symbol isn't weak, it's capability member is
2955                          * retained for the creation of a local symbol.
2956                          */
2957                         if (bind1 != STB_WEAK)
2958                                 continue;
2959 
2960                         /*
2961                          * If this is a weak symbol, traverse the capabilities
2962                          * list again to determine if a corresponding non-weak
2963                          * symbol exists.
2964                          */
2965                         for (ALIST_TRAVERSE(cappairs, idx2, cpp2)) {
2966                                 Sym_desc        *sdp2 = cpp2->c_osdp;
2967                                 Sym             *sym2 = sdp2->sd_sym;
2968                                 uchar_t         bind2 =
2969                                     ELF_ST_BIND(sym2->st_info);
2970 
2971                                 if ((cpp1 == cpp2) ||
2972                                     (cpp1->c_group != cpp2->c_group) ||
2973                                     (sym1->st_value != sym2->st_value) ||
2974                                     (bind2 == STB_WEAK))
2975                                         continue;
2976 
2977                                 /*
2978                                  * The weak symbol (sym1) has a non-weak (sym2)
2979                                  * counterpart.  There's no point in translating
2980                                  * both of these equivalent symbols to locals.
2981                                  * Add this symbol capability alias to the
2982                                  * capabilities family information, and remove
2983                                  * the weak symbol.
2984                                  */
2985                                 if (ld_cap_add_family(ofl, cpp2->c_nsdp,
2986                                     cpp1->c_nsdp, NULL, NULL) == S_ERROR)
2987                                         return (S_ERROR);
2988 
2989                                 free((void *)cpp1->c_osdp);
2990                                 (void) alist_delete(cappairs, &idx1);
2991                         }
2992                 }
2993 
2994                 DBG_CALL(Dbg_util_nl(ofl->ofl_lml, DBG_NL_STD));
2995 
2996                 /*
2997                  * The capability pairs information now represents all the
2998                  * global symbols that need transforming to locals.  These
2999                  * local symbols are renamed using their group identifiers.
3000                  */
3001                 for (ALIST_TRAVERSE(cappairs, idx1, cpp1)) {
3002                         Sym_desc        *osdp = cpp1->c_osdp;
3003                         Objcapset       *capset;
3004                         size_t          nsize, tsize;
3005                         const char      *oname;
3006                         char            *cname, *idstr;
3007                         Sym             *csym;
3008 
3009                         /*
3010                          * If the local symbol has not yet been translated
3011                          * convert it to a local symbol with a name.
3012                          */
3013                         if ((osdp->sd_flags & FLG_SY_CAP) != 0)
3014                                 continue;
3015 
3016                         /*
3017                          * As we're converting object capabilities to symbol
3018                          * capabilities, obtain the capabilities set for this
3019                          * object, so as to retrieve the CA_SUNW_ID value.
3020                          */
3021                         capset = &cpp1->c_group->cg_set;
3022 
3023                         /*
3024                          * Create a new name from the existing symbol and the
3025                          * capabilities group identifier.  Note, the delimiter
3026                          * between the symbol name and identifier name is hard-
3027                          * coded here (%), so that we establish a convention
3028                          * for transformed symbol names.
3029                          */
3030                         oname = osdp->sd_name;
3031 
3032                         idstr = capset->oc_id.cs_str;
3033                         nsize = strlen(oname);
3034                         tsize = nsize + 1 + strlen(idstr) + 1;
3035                         if ((cname = libld_malloc(tsize)) == 0)
3036                                 return (S_ERROR);
3037 
3038                         (void) strcpy(cname, oname);
3039                         cname[nsize++] = '%';
3040                         (void) strcpy(&cname[nsize], idstr);
3041 
3042                         /*
3043                          * Allocate a new symbol table entry, transform this
3044                          * symbol to a local, and assign the new name.
3045                          */
3046                         if ((csym = libld_malloc(sizeof (Sym))) == NULL)
3047                                 return (S_ERROR);
3048 
3049                         *csym = *osdp->sd_sym;
3050                         csym->st_info = ELF_ST_INFO(STB_LOCAL,
3051                             ELF_ST_TYPE(osdp->sd_sym->st_info));
3052 
3053                         osdp->sd_name = cname;
3054                         osdp->sd_sym = csym;
3055                         osdp->sd_flags = FLG_SY_CAP;
3056 
3057                         /*
3058                          * Keep track of this new local symbol.  As -z symbolcap
3059                          * can only be used to create a relocatable object, a
3060                          * dynamic symbol table can't exist.  Ensure there is
3061                          * space reserved in the string table.
3062                          */
3063                         ofl->ofl_caploclcnt++;
3064                         if (st_insert(ofl->ofl_strtab, cname) == -1)
3065                                 return (S_ERROR);
3066 
3067                         DBG_CALL(Dbg_syms_cap_local(ofl, cpp1->c_ndx,
3068                             cname, csym, osdp));
3069 
3070                         /*
3071                          * Establish this capability pair as a family.
3072                          */
3073                         if (ld_cap_add_family(ofl, cpp1->c_nsdp, osdp,
3074                             cpp1->c_group, &ifl->ifl_caps->ca_syms) == S_ERROR)
3075                                 return (S_ERROR);
3076                 }
3077         }
3078 
3079         return (1);
3080 
3081 #undef SYM_LOC_BADADDR
3082 }
3083 
3084 /*
3085  * Add an undefined symbol to the symbol table.  The reference originates from
3086  * the location identified by the message id (mid).  These references can
3087  * originate from command line options such as -e, -u, -initarray, etc.
3088  * (identified with MSG_INTL(MSG_STR_COMMAND)), or from internally generated
3089  * TLS relocation references (identified with MSG_INTL(MSG_STR_TLSREL)).
3090  */
3091 Sym_desc *
3092 ld_sym_add_u(const char *name, Ofl_desc *ofl, Msg mid)
3093 {
3094         Sym             *sym;
3095         Ifl_desc        *ifl = NULL, *_ifl;
3096         Sym_desc        *sdp;
3097         Word            hash;
3098         Aliste          idx;
3099         avl_index_t     where;
3100         const char      *reference = MSG_INTL(mid);
3101 
3102         /*
3103          * As an optimization, determine whether we've already generated this
3104          * reference.  If the symbol doesn't already exist we'll create it.
3105          * Or if the symbol does exist from a different source, we'll resolve
3106          * the conflict.
3107          */
3108         /* LINTED */
3109         hash = (Word)elf_hash(name);
3110         if ((sdp = ld_sym_find(name, hash, &where, ofl)) != NULL) {
3111                 if ((sdp->sd_sym->st_shndx == SHN_UNDEF) &&
3112                     (sdp->sd_file->ifl_name == reference))
3113                         return (sdp);
3114         }
3115 
3116         /*
3117          * Determine whether a pseudo input file descriptor exists to represent
3118          * the command line, as any global symbol needs an input file descriptor
3119          * during any symbol resolution (refer to map_ifl() which provides a
3120          * similar method for adding symbols from mapfiles).
3121          */
3122         for (APLIST_TRAVERSE(ofl->ofl_objs, idx, _ifl))
3123                 if (strcmp(_ifl->ifl_name, reference) == 0) {
3124                         ifl = _ifl;
3125                         break;
3126                 }
3127 
3128         /*
3129          * If no descriptor exists create one.
3130          */
3131         if (ifl == NULL) {
3132                 if ((ifl = libld_calloc(sizeof (Ifl_desc), 1)) == NULL)
3133                         return ((Sym_desc *)S_ERROR);
3134                 ifl->ifl_name = reference;
3135                 ifl->ifl_flags = FLG_IF_NEEDED | FLG_IF_FILEREF;
3136                 if ((ifl->ifl_ehdr = libld_calloc(sizeof (Ehdr), 1)) == NULL)
3137                         return ((Sym_desc *)S_ERROR);
3138                 ifl->ifl_ehdr->e_type = ET_REL;
3139 
3140                 if (aplist_append(&ofl->ofl_objs, ifl, AL_CNT_OFL_OBJS) == NULL)
3141                         return ((Sym_desc *)S_ERROR);
3142         }
3143 
3144         /*
3145          * Allocate a symbol structure and add it to the global symbol table.
3146          */
3147         if ((sym = libld_calloc(sizeof (Sym), 1)) == NULL)
3148                 return ((Sym_desc *)S_ERROR);
3149         sym->st_info = ELF_ST_INFO(STB_GLOBAL, STT_NOTYPE);
3150         sym->st_shndx = SHN_UNDEF;
3151 
3152         DBG_CALL(Dbg_syms_process(ofl->ofl_lml, ifl));
3153         if (sdp == NULL) {
3154                 DBG_CALL(Dbg_syms_global(ofl->ofl_lml, 0, name));
3155                 if ((sdp = ld_sym_enter(name, sym, hash, ifl, ofl, 0, SHN_UNDEF,
3156                     0, &where)) == (Sym_desc *)S_ERROR)
3157                         return ((Sym_desc *)S_ERROR);
3158         } else if (ld_sym_resolve(sdp, sym, ifl, ofl, 0,
3159             SHN_UNDEF, 0) == S_ERROR)
3160                 return ((Sym_desc *)S_ERROR);
3161 
3162         sdp->sd_flags &= ~FLG_SY_CLEAN;
3163         sdp->sd_flags |= FLG_SY_CMDREF;
3164 
3165         return (sdp);
3166 }
3167 
3168 /*
3169  * STT_SECTION symbols have their st_name field set to NULL, and consequently
3170  * have no name. Generate a name suitable for diagnostic use for such a symbol
3171  * and store it in the input section descriptor. The resulting name will be
3172  * of the form:
3173  *
3174  *      "XXX (section)"
3175  *
3176  * where XXX is the name of the section.
3177  *
3178  * entry:
3179  *      isc - Input section associated with the symbol.
3180  *      fmt - NULL, or format string to use.
3181  *
3182  * exit:
3183  *      Sets isp->is_sym_name to the allocated string. Returns the
3184  *      string pointer, or NULL on allocation failure.
3185  */
3186 const char *
3187 ld_stt_section_sym_name(Is_desc *isp)
3188 {
3189         const char      *fmt;
3190         char            *str;
3191         size_t          len;
3192 
3193         if ((isp == NULL) || (isp->is_name == NULL))
3194                 return (NULL);
3195 
3196         if (isp->is_sym_name == NULL) {
3197                 fmt = (isp->is_flags & FLG_IS_GNSTRMRG) ?
3198                     MSG_INTL(MSG_STR_SECTION_MSTR) : MSG_INTL(MSG_STR_SECTION);
3199 
3200                 len = strlen(fmt) + strlen(isp->is_name) + 1;
3201 
3202                 if ((str = libld_malloc(len)) == NULL)
3203                         return (NULL);
3204                 (void) snprintf(str, len, fmt, isp->is_name);
3205                 isp->is_sym_name = str;
3206         }
3207 
3208         return (isp->is_sym_name);
3209 }