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);
 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);
 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 static Boolean
 942 is_cname(const char *name)
 943 {
 944         if (strlen(name) == strspn(name,
 945             "abcdefghijklmnopqrstuvwxyz"
 946             "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
 947             "0123456789"
 948             "_"))
 949                 return (TRUE);
 950         else
 951                 return (FALSE);
 952 }
 953 
 954 /*
 955  * At this point all symbol input processing has been completed, therefore
 956  * complete the symbol table entries by generating any necessary internal
 957  * symbols.
 958  */
 959 uintptr_t
 960 ld_sym_spec(Ofl_desc *ofl)
 961 {
 962         Sym_desc        *sdp;
 963         Sg_desc         *sgp;
 964         Aliste          idx1;
 965 
 966         DBG_CALL(Dbg_syms_spec_title(ofl->ofl_lml));
 967 
 968         /*
 969          * For each section in the output file, look for symbols named for the
 970          * __start/__stop patterns.  If references exist, flesh the symbols to
 971          * be defined.
 972          *
 973          * The symbols are given values at the same time as the other special
 974          * symbols.
 975          */
 976         if (!(ofl->ofl_flags & FLG_OF_RELOBJ) ||
 977             (ofl->ofl_flags & FLG_OF_KMOD)) {
 978                 for (APLIST_TRAVERSE(ofl->ofl_segs, idx1, sgp)) {
 979                         Os_desc *osp;
 980                         Aliste idx2;
 981 
 982                         for (APLIST_TRAVERSE(sgp->sg_osdescs, idx2, osp)) {
 983                                 if (is_cname(osp->os_name)) {
 984                                         sym_add_bounds(ofl, osp,
 985                                             SDAUX_ID_SECBOUND_START);
 986                                         sym_add_bounds(ofl, osp,
 987                                             SDAUX_ID_SECBOUND_STOP);
 988                                 }
 989                         }
 990                 }
 991         }
 992 
 993         if (ofl->ofl_flags & FLG_OF_RELOBJ)
 994                 return (1);
 995 
 996         if (sym_add_spec(MSG_ORIG(MSG_SYM_ETEXT), MSG_ORIG(MSG_SYM_ETEXT_U),
 997             SDAUX_ID_ETEXT, 0, (FLG_SY_DEFAULT | FLG_SY_EXPDEF),
 998             ofl) == S_ERROR)
 999                 return (S_ERROR);
1000         if (sym_add_spec(MSG_ORIG(MSG_SYM_EDATA), MSG_ORIG(MSG_SYM_EDATA_U),
1001             SDAUX_ID_EDATA, 0, (FLG_SY_DEFAULT | FLG_SY_EXPDEF),
1002             ofl) == S_ERROR)
1003                 return (S_ERROR);
1004         if (sym_add_spec(MSG_ORIG(MSG_SYM_END), MSG_ORIG(MSG_SYM_END_U),
1005             SDAUX_ID_END, FLG_SY_DYNSORT, (FLG_SY_DEFAULT | FLG_SY_EXPDEF),
1006             ofl) == S_ERROR)
1007                 return (S_ERROR);
1008         if (sym_add_spec(MSG_ORIG(MSG_SYM_L_END), MSG_ORIG(MSG_SYM_L_END_U),
1009             SDAUX_ID_END, 0, FLG_SY_HIDDEN, ofl) == S_ERROR)
1010                 return (S_ERROR);
1011         if (sym_add_spec(MSG_ORIG(MSG_SYM_L_START), MSG_ORIG(MSG_SYM_L_START_U),
1012             SDAUX_ID_START, 0, FLG_SY_HIDDEN, ofl) == S_ERROR)
1013                 return (S_ERROR);
1014 
1015         /*
1016          * Historically we've always produced a _DYNAMIC symbol, even for
1017          * static executables (in which case its value will be 0).
1018          */
1019         if (sym_add_spec(MSG_ORIG(MSG_SYM_DYNAMIC), MSG_ORIG(MSG_SYM_DYNAMIC_U),
1020             SDAUX_ID_DYN, FLG_SY_DYNSORT, (FLG_SY_DEFAULT | FLG_SY_EXPDEF),
1021             ofl) == S_ERROR)
1022                 return (S_ERROR);
1023 
1024         if (OFL_ALLOW_DYNSYM(ofl))
1025                 if (sym_add_spec(MSG_ORIG(MSG_SYM_PLKTBL),
1026                     MSG_ORIG(MSG_SYM_PLKTBL_U), SDAUX_ID_PLT,
1027                     FLG_SY_DYNSORT, (FLG_SY_DEFAULT | FLG_SY_EXPDEF),
1028                     ofl) == S_ERROR)
1029                         return (S_ERROR);
1030 
1031         /*
1032          * A GOT reference will be accompanied by the associated GOT symbol.
1033          * Make sure it gets assigned the appropriate special attributes.
1034          */
1035         if (((sdp = ld_sym_find(MSG_ORIG(MSG_SYM_GOFTBL_U),
1036             SYM_NOHASH, NULL, ofl)) != NULL) && (sdp->sd_ref != REF_DYN_SEEN)) {
1037                 if (sym_add_spec(MSG_ORIG(MSG_SYM_GOFTBL),
1038                     MSG_ORIG(MSG_SYM_GOFTBL_U), SDAUX_ID_GOT, FLG_SY_DYNSORT,
1039                     (FLG_SY_DEFAULT | FLG_SY_EXPDEF), ofl) == S_ERROR)
1040                         return (S_ERROR);
1041         }
1042 
1043         return (1);
1044 }
1045 
1046 /*
1047  * Determine a potential capability symbol's visibility.
1048  *
1049  * The -z symbolcap option transforms an object capabilities relocatable object
1050  * into a symbol capabilities relocatable object.  Any global function symbols,
1051  * or initialized global data symbols are candidates for transforming into local
1052  * symbol capabilities definitions.  However, if a user indicates that a symbol
1053  * should be demoted to local using a mapfile, then there is no need to
1054  * transform the associated global symbol.
1055  *
1056  * Normally, a symbol's visibility is determined after the symbol resolution
1057  * process, after all symbol state has been gathered and resolved.  However,
1058  * for -z symbolcap, this determination is too late.  When a global symbol is
1059  * read from an input file we need to determine it's visibility so as to decide
1060  * whether to create a local or not.
1061  *
1062  * If a user has explicitly defined this symbol as having local scope within a
1063  * mapfile, then a symbol of the same name already exists.  However, explicit
1064  * local definitions are uncommon, as most mapfiles define the global symbol
1065  * requirements together with an auto-reduction directive '*'.  If this state
1066  * has been defined, then we must make sure that the new symbol isn't a type
1067  * that can not be demoted to local.
1068  */
1069 static int
1070 sym_cap_vis(const char *name, Word hash, Sym *sym, Ofl_desc *ofl)
1071 {
1072         Sym_desc        *sdp;
1073         uchar_t         vis;
1074         avl_index_t     where;
1075         sd_flag_t       sdflags = 0;
1076 
1077         /*
1078          * Determine the visibility of the new symbol.
1079          */
1080         vis = ELF_ST_VISIBILITY(sym->st_other);
1081         switch (vis) {
1082         case STV_EXPORTED:
1083                 sdflags |= FLG_SY_EXPORT;
1084                 break;
1085         case STV_SINGLETON:
1086                 sdflags |= FLG_SY_SINGLE;
1087                 break;
1088         }
1089 
1090         /*
1091          * Determine whether a symbol definition already exists, and if so
1092          * obtain the visibility.
1093          */
1094         if ((sdp = ld_sym_find(name, hash, &where, ofl)) != NULL)
1095                 sdflags |= sdp->sd_flags;
1096 
1097         /*
1098          * Determine whether the symbol flags indicate this symbol should be
1099          * hidden.
1100          */
1101         if ((ofl->ofl_flags & (FLG_OF_AUTOLCL | FLG_OF_AUTOELM)) &&
1102             ((sdflags & MSK_SY_NOAUTO) == 0))
1103                 sdflags |= FLG_SY_HIDDEN;
1104 
1105         return ((sdflags & FLG_SY_HIDDEN) == 0);
1106 }
1107 
1108 /*
1109  * This routine checks to see if a symbols visibility needs to be reduced to
1110  * either SYMBOLIC or LOCAL.  This routine can be called from either
1111  * reloc_init() or sym_validate().
1112  */
1113 void
1114 ld_sym_adjust_vis(Sym_desc *sdp, Ofl_desc *ofl)
1115 {
1116         ofl_flag_t      oflags = ofl->ofl_flags;
1117         Sym             *sym = sdp->sd_sym;
1118 
1119         if ((sdp->sd_ref == REF_REL_NEED) &&
1120             (sdp->sd_sym->st_shndx != SHN_UNDEF)) {
1121                 /*
1122                  * If auto-reduction/elimination is enabled, reduce any
1123                  * non-versioned, and non-local capabilities global symbols.
1124                  * A symbol is a candidate for auto-reduction/elimination if:
1125                  *
1126                  *  -   the symbol wasn't explicitly defined within a mapfile
1127                  *      (in which case all the necessary state has been applied
1128                  *      to the symbol), or
1129                  *  -   the symbol isn't one of the family of reserved
1130                  *      special symbols (ie. _end, _etext, etc.), or
1131                  *  -   the symbol isn't a SINGLETON, or
1132                  *  -   the symbol wasn't explicitly defined within a version
1133                  *      definition associated with an input relocatable object.
1134                  *
1135                  * Indicate that the symbol has been reduced as it may be
1136                  * necessary to print these symbols later.
1137                  */
1138                 if ((oflags & (FLG_OF_AUTOLCL | FLG_OF_AUTOELM)) &&
1139                     ((sdp->sd_flags & MSK_SY_NOAUTO) == 0)) {
1140                         if ((sdp->sd_flags & FLG_SY_HIDDEN) == 0) {
1141                                 sdp->sd_flags |=
1142                                     (FLG_SY_REDUCED | FLG_SY_HIDDEN);
1143                         }
1144 
1145                         if (oflags & (FLG_OF_REDLSYM | FLG_OF_AUTOELM)) {
1146                                 sdp->sd_flags |= FLG_SY_ELIM;
1147                                 sym->st_other = STV_ELIMINATE |
1148                                     (sym->st_other & ~MSK_SYM_VISIBILITY);
1149                         } else if (ELF_ST_VISIBILITY(sym->st_other) !=
1150                             STV_INTERNAL)
1151                                 sym->st_other = STV_HIDDEN |
1152                                     (sym->st_other & ~MSK_SYM_VISIBILITY);
1153                 }
1154 
1155                 /*
1156                  * If -Bsymbolic is in effect, and the symbol hasn't explicitly
1157                  * been defined nodirect (via a mapfile), then bind the global
1158                  * symbol symbolically and assign the STV_PROTECTED visibility
1159                  * attribute.
1160                  */
1161                 if ((oflags & FLG_OF_SYMBOLIC) &&
1162                     ((sdp->sd_flags & (FLG_SY_HIDDEN | FLG_SY_NDIR)) == 0)) {
1163                         sdp->sd_flags |= FLG_SY_PROTECT;
1164                         if (ELF_ST_VISIBILITY(sym->st_other) == STV_DEFAULT)
1165                                 sym->st_other = STV_PROTECTED |
1166                                     (sym->st_other & ~MSK_SYM_VISIBILITY);
1167                 }
1168         }
1169 
1170         /*
1171          * Indicate that this symbol has had it's visibility checked so that
1172          * we don't need to do this investigation again.
1173          */
1174         sdp->sd_flags |= FLG_SY_VISIBLE;
1175 }
1176 
1177 /*
1178  * Make sure a symbol definition is local to the object being built.
1179  */
1180 inline static int
1181 ensure_sym_local(Ofl_desc *ofl, Sym_desc *sdp, const char *str)
1182 {
1183         if (sdp->sd_sym->st_shndx == SHN_UNDEF) {
1184                 if (str) {
1185                         ld_eprintf(ofl, ERR_FATAL, MSG_INTL(MSG_SYM_UNDEF),
1186                             str, demangle((char *)sdp->sd_name));
1187                 }
1188                 return (1);
1189         }
1190         if (sdp->sd_ref != REF_REL_NEED) {
1191                 if (str) {
1192                         ld_eprintf(ofl, ERR_FATAL, MSG_INTL(MSG_SYM_EXTERN),
1193                             str, demangle((char *)sdp->sd_name),
1194                             sdp->sd_file->ifl_name);
1195                 }
1196                 return (1);
1197         }
1198 
1199         sdp->sd_flags |= FLG_SY_UPREQD;
1200         if (sdp->sd_isc) {
1201                 sdp->sd_isc->is_flags |= FLG_IS_SECTREF;
1202                 sdp->sd_isc->is_file->ifl_flags |= FLG_IF_FILEREF;
1203         }
1204         return (0);
1205 }
1206 
1207 /*
1208  * Make sure all the symbol definitions required for initarray, finiarray, or
1209  * preinitarray's are local to the object being built.
1210  */
1211 static int
1212 ensure_array_local(Ofl_desc *ofl, APlist *apl, const char *str)
1213 {
1214         Aliste          idx;
1215         Sym_desc        *sdp;
1216         int             ret = 0;
1217 
1218         for (APLIST_TRAVERSE(apl, idx, sdp))
1219                 ret += ensure_sym_local(ofl, sdp, str);
1220 
1221         return (ret);
1222 }
1223 
1224 /*
1225  * After all symbol table input processing has been finished, and all relocation
1226  * counting has been carried out (ie. no more symbols will be read, generated,
1227  * or modified), validate and count the relevant entries:
1228  *
1229  *  -   check and print any undefined symbols remaining.  Note that if a symbol
1230  *      has been defined by virtue of the inclusion of  an implicit shared
1231  *      library, it is still classed as undefined.
1232  *
1233  *  -   count the number of global needed symbols together with the size of
1234  *      their associated name strings (if scoping has been indicated these
1235  *      symbols may be reduced to locals).
1236  *
1237  *  -   establish the size and alignment requirements for the global .bss
1238  *      section (the alignment of this section is based on the first symbol
1239  *      that it will contain).
1240  */
1241 uintptr_t
1242 ld_sym_validate(Ofl_desc *ofl)
1243 {
1244         Sym_avlnode     *sav;
1245         Sym_desc        *sdp;
1246         Sym             *sym;
1247         ofl_flag_t      oflags = ofl->ofl_flags;
1248         ofl_flag_t      undef = 0, needed = 0, verdesc = 0;
1249         Xword           bssalign = 0, tlsalign = 0;
1250         Boolean         need_bss, need_tlsbss;
1251         Xword           bsssize = 0, tlssize = 0;
1252 #if     defined(_ELF64)
1253         Xword           lbssalign = 0, lbsssize = 0;
1254         Boolean         need_lbss;
1255 #endif
1256         int             ret, allow_ldynsym;
1257         uchar_t         type;
1258         ofl_flag_t      undef_state = 0;
1259 
1260         DBG_CALL(Dbg_basic_validate(ofl->ofl_lml));
1261 
1262         /*
1263          * The need_XXX booleans are used to determine whether we need to
1264          * create each type of bss section. We used to create these sections
1265          * if the sum of the required sizes for each type were non-zero.
1266          * However, it is possible for a compiler to generate COMMON variables
1267          * of zero-length and this tricks that logic --- even zero-length
1268          * symbols need an output section.
1269          */
1270         need_bss = need_tlsbss = FALSE;
1271 #if     defined(_ELF64)
1272         need_lbss = FALSE;
1273 #endif
1274 
1275         /*
1276          * Determine how undefined symbols are handled:
1277          *
1278          * fatal:
1279          *      If this link-edit calls for no undefined symbols to remain
1280          *      (this is the default case when generating an executable but
1281          *      can be enforced for any object using -z defs), a fatal error
1282          *      condition will be indicated.
1283          *
1284          * warning:
1285          *      If we're creating a shared object, and either the -Bsymbolic
1286          *      flag is set, or the user has turned on the -z guidance feature,
1287          *      then a non-fatal warning is issued for each symbol.
1288          *
1289          * ignore:
1290          *      In all other cases, undefined symbols are quietly allowed.
1291          */
1292         if (oflags & FLG_OF_NOUNDEF) {
1293                 undef = FLG_OF_FATAL;
1294         } else if (oflags & FLG_OF_SHAROBJ) {
1295                 if ((oflags & FLG_OF_SYMBOLIC) ||
1296                     OFL_GUIDANCE(ofl, FLG_OFG_NO_DEFS))
1297                         undef = FLG_OF_WARN;
1298         }
1299 
1300         /*
1301          * If the symbol is referenced from an implicitly included shared object
1302          * (ie. it's not on the NEEDED list) then the symbol is also classified
1303          * as undefined and a fatal error condition will be indicated.
1304          */
1305         if ((oflags & FLG_OF_NOUNDEF) || !(oflags & FLG_OF_SHAROBJ))
1306                 needed = FLG_OF_FATAL;
1307         else if ((oflags & FLG_OF_SHAROBJ) &&
1308             OFL_GUIDANCE(ofl, FLG_OFG_NO_DEFS))
1309                 needed = FLG_OF_WARN;
1310 
1311         /*
1312          * If the output image is being versioned, then all symbol definitions
1313          * must be associated with a version.  Any symbol that isn't associated
1314          * with a version is classified as undefined, and a fatal error
1315          * condition is indicated.
1316          */
1317         if ((oflags & FLG_OF_VERDEF) && (ofl->ofl_vercnt > VER_NDX_GLOBAL))
1318                 verdesc = FLG_OF_FATAL;
1319 
1320         allow_ldynsym = OFL_ALLOW_LDYNSYM(ofl);
1321 
1322         if (allow_ldynsym) {
1323                 /*
1324                  * Normally, we disallow symbols with 0 size from appearing
1325                  * in a dyn[sym|tls]sort section. However, there are some
1326                  * symbols that serve special purposes that we want to exempt
1327                  * from this rule. Look them up, and set their
1328                  * FLG_SY_DYNSORT flag.
1329                  */
1330                 static const char *special[] = {
1331                         MSG_ORIG(MSG_SYM_INIT_U),       /* _init */
1332                         MSG_ORIG(MSG_SYM_FINI_U),       /* _fini */
1333                         MSG_ORIG(MSG_SYM_START),        /* _start */
1334                         NULL
1335                 };
1336                 int i;
1337 
1338                 for (i = 0; special[i] != NULL; i++) {
1339                         if (((sdp = ld_sym_find(special[i],
1340                             SYM_NOHASH, NULL, ofl)) != NULL) &&
1341                             (sdp->sd_sym->st_size == 0)) {
1342                                 if (ld_sym_copy(sdp) == S_ERROR)
1343                                         return (S_ERROR);
1344                                 sdp->sd_flags |= FLG_SY_DYNSORT;
1345                         }
1346                 }
1347         }
1348 
1349         /*
1350          * Collect and validate the globals from the internal symbol table.
1351          */
1352         for (sav = avl_first(&ofl->ofl_symavl); sav;
1353             sav = AVL_NEXT(&ofl->ofl_symavl, sav)) {
1354                 Is_desc         *isp;
1355                 int             undeferr = 0;
1356                 uchar_t         vis;
1357 
1358                 sdp = sav->sav_sdp;
1359 
1360                 /*
1361                  * If undefined symbols are allowed, and we're not being
1362                  * asked to supply guidance, ignore any symbols that are
1363                  * not needed.
1364                  */
1365                 if (!(oflags & FLG_OF_NOUNDEF) &&
1366                     !OFL_GUIDANCE(ofl, FLG_OFG_NO_DEFS) &&
1367                     (sdp->sd_ref == REF_DYN_SEEN))
1368                         continue;
1369 
1370                 /*
1371                  * If the symbol originates from an external or parent mapfile
1372                  * reference and hasn't been matched to a reference from a
1373                  * relocatable object, ignore it.
1374                  */
1375                 if ((sdp->sd_flags & (FLG_SY_EXTERN | FLG_SY_PARENT)) &&
1376                     ((sdp->sd_flags & FLG_SY_MAPUSED) == 0)) {
1377                         sdp->sd_flags |= FLG_SY_INVALID;
1378                         continue;
1379                 }
1380 
1381                 sym = sdp->sd_sym;
1382                 type = ELF_ST_TYPE(sym->st_info);
1383 
1384                 /*
1385                  * Sanity check TLS.
1386                  */
1387                 if ((type == STT_TLS) && (sym->st_size != 0) &&
1388                     (sym->st_shndx != SHN_UNDEF) &&
1389                     (sym->st_shndx != SHN_COMMON)) {
1390                         Is_desc         *isp = sdp->sd_isc;
1391                         Ifl_desc        *ifl = sdp->sd_file;
1392 
1393                         if ((isp == NULL) || (isp->is_shdr == NULL) ||
1394                             ((isp->is_shdr->sh_flags & SHF_TLS) == 0)) {
1395                                 ld_eprintf(ofl, ERR_FATAL,
1396                                     MSG_INTL(MSG_SYM_TLS),
1397                                     demangle(sdp->sd_name), ifl->ifl_name);
1398                                 continue;
1399                         }
1400                 }
1401 
1402                 if ((sdp->sd_flags & FLG_SY_VISIBLE) == 0)
1403                         ld_sym_adjust_vis(sdp, ofl);
1404 
1405                 if ((sdp->sd_flags & FLG_SY_REDUCED) &&
1406                     (oflags & FLG_OF_PROCRED)) {
1407                         DBG_CALL(Dbg_syms_reduce(ofl, DBG_SYM_REDUCE_GLOBAL,
1408                             sdp, 0, 0));
1409                 }
1410 
1411                 /*
1412                  * Record any STV_SINGLETON existence.
1413                  */
1414                 if ((vis = ELF_ST_VISIBILITY(sym->st_other)) == STV_SINGLETON)
1415                         ofl->ofl_dtflags_1 |= DF_1_SINGLETON;
1416 
1417                 /*
1418                  * If building a shared object or executable, and this is a
1419                  * non-weak UNDEF symbol with reduced visibility (STV_*), then
1420                  * give a fatal error.
1421                  */
1422                 if (((oflags & FLG_OF_RELOBJ) == 0) &&
1423                     (sym->st_shndx == SHN_UNDEF) &&
1424                     (ELF_ST_BIND(sym->st_info) != STB_WEAK)) {
1425                         if (vis && (vis != STV_SINGLETON)) {
1426                                 sym_undef_entry(ofl, sdp, BNDLOCAL,
1427                                     FLG_OF_FATAL, &undef_state);
1428                                 continue;
1429                         }
1430                 }
1431 
1432                 /*
1433                  * If this symbol is defined in a non-allocatable section,
1434                  * reduce it to local symbol.
1435                  */
1436                 if (((isp = sdp->sd_isc) != 0) && isp->is_shdr &&
1437                     ((isp->is_shdr->sh_flags & SHF_ALLOC) == 0)) {
1438                         sdp->sd_flags |= (FLG_SY_REDUCED | FLG_SY_HIDDEN);
1439                 }
1440 
1441                 /*
1442                  * If this symbol originated as a SHN_SUNW_IGNORE, it will have
1443                  * been processed as an SHN_UNDEF.  Return the symbol to its
1444                  * original index for validation, and propagation to the output
1445                  * file.
1446                  */
1447                 if (sdp->sd_flags & FLG_SY_IGNORE)
1448                         sdp->sd_shndx = SHN_SUNW_IGNORE;
1449 
1450                 if (undef) {
1451                         /*
1452                          * If a non-weak reference remains undefined, or if a
1453                          * mapfile reference is not bound to the relocatable
1454                          * objects that make up the object being built, we have
1455                          * a fatal error.
1456                          *
1457                          * The exceptions are symbols which are defined to be
1458                          * found in the parent (FLG_SY_PARENT), which is really
1459                          * only meaningful for direct binding, or are defined
1460                          * external (FLG_SY_EXTERN) so as to suppress -zdefs
1461                          * errors.
1462                          *
1463                          * Register symbols are always allowed to be UNDEF.
1464                          *
1465                          * Note that we don't include references created via -u
1466                          * in the same shared object binding test.  This is for
1467                          * backward compatibility, in that a number of archive
1468                          * makefile rules used -u to cause archive extraction.
1469                          * These same rules have been cut and pasted to apply
1470                          * to shared objects, and thus although the -u reference
1471                          * is redundant, flagging it as fatal could cause some
1472                          * build to fail.  Also we have documented the use of
1473                          * -u as a mechanism to cause binding to weak version
1474                          * definitions, thus giving users an error condition
1475                          * would be incorrect.
1476                          */
1477                         if (!(sdp->sd_flags & FLG_SY_REGSYM) &&
1478                             ((sym->st_shndx == SHN_UNDEF) &&
1479                             ((ELF_ST_BIND(sym->st_info) != STB_WEAK) &&
1480                             ((sdp->sd_flags &
1481                             (FLG_SY_PARENT | FLG_SY_EXTERN)) == 0)) ||
1482                             ((sdp->sd_flags &
1483                             (FLG_SY_MAPREF | FLG_SY_MAPUSED | FLG_SY_HIDDEN |
1484                             FLG_SY_PROTECT)) == FLG_SY_MAPREF))) {
1485                                 sym_undef_entry(ofl, sdp, UNDEF, undef,
1486                                     &undef_state);
1487                                 undeferr = 1;
1488                         }
1489 
1490                 } else {
1491                         /*
1492                          * For building things like shared objects (or anything
1493                          * -znodefs), undefined symbols are allowed.
1494                          *
1495                          * If a mapfile reference remains undefined the user
1496                          * would probably like a warning at least (they've
1497                          * usually mis-spelt the reference).  Refer to the above
1498                          * comments for discussion on -u references, which
1499                          * are not tested for in the same manner.
1500                          */
1501                         if ((sdp->sd_flags &
1502                             (FLG_SY_MAPREF | FLG_SY_MAPUSED)) ==
1503                             FLG_SY_MAPREF) {
1504                                 sym_undef_entry(ofl, sdp, UNDEF, FLG_OF_WARN,
1505                                     &undef_state);
1506                                 undeferr = 1;
1507                         }
1508                 }
1509 
1510                 /*
1511                  * If this symbol comes from a dependency mark the dependency
1512                  * as required (-z ignore can result in unused dependencies
1513                  * being dropped).  If we need to record dependency versioning
1514                  * information indicate what version of the needed shared object
1515                  * this symbol is part of.  Flag the symbol as undefined if it
1516                  * has not been made available to us.
1517                  */
1518                 if ((sdp->sd_ref == REF_DYN_NEED) &&
1519                     (!(sdp->sd_flags & FLG_SY_REFRSD))) {
1520                         sdp->sd_file->ifl_flags |= FLG_IF_DEPREQD;
1521 
1522                         /*
1523                          * Capture that we've bound to a symbol that doesn't
1524                          * allow being directly bound to.
1525                          */
1526                         if (sdp->sd_flags & FLG_SY_NDIR)
1527                                 ofl->ofl_flags1 |= FLG_OF1_NGLBDIR;
1528 
1529                         if (sdp->sd_file->ifl_vercnt) {
1530                                 int             vndx;
1531                                 Ver_index       *vip;
1532 
1533                                 vndx = sdp->sd_aux->sa_dverndx;
1534                                 vip = &sdp->sd_file->ifl_verndx[vndx];
1535                                 if (vip->vi_flags & FLG_VER_AVAIL) {
1536                                         vip->vi_flags |= FLG_VER_REFER;
1537                                 } else {
1538                                         sym_undef_entry(ofl, sdp, NOTAVAIL,
1539                                             FLG_OF_FATAL, &undef_state);
1540                                         continue;
1541                                 }
1542                         }
1543                 }
1544 
1545                 /*
1546                  * Test that we do not bind to symbol supplied from an implicit
1547                  * shared object.  If a binding is from a weak reference it can
1548                  * be ignored.
1549                  */
1550                 if (needed && !undeferr && (sdp->sd_flags & FLG_SY_GLOBREF) &&
1551                     (sdp->sd_ref == REF_DYN_NEED) &&
1552                     (sdp->sd_flags & FLG_SY_NOTAVAIL)) {
1553                         sym_undef_entry(ofl, sdp, IMPLICIT, needed,
1554                             &undef_state);
1555                         if (needed == FLG_OF_FATAL)
1556                                 continue;
1557                 }
1558 
1559                 /*
1560                  * Test that a symbol isn't going to be reduced to local scope
1561                  * which actually wants to bind to a shared object - if so it's
1562                  * a fatal error.
1563                  */
1564                 if ((sdp->sd_ref == REF_DYN_NEED) &&
1565                     (sdp->sd_flags & (FLG_SY_HIDDEN | FLG_SY_PROTECT))) {
1566                         sym_undef_entry(ofl, sdp, BNDLOCAL, FLG_OF_FATAL,
1567                             &undef_state);
1568                         continue;
1569                 }
1570 
1571                 /*
1572                  * If the output image is to be versioned then all symbol
1573                  * definitions must be associated with a version.  Remove any
1574                  * versioning that might be left associated with an undefined
1575                  * symbol.
1576                  */
1577                 if (verdesc && (sdp->sd_ref == REF_REL_NEED)) {
1578                         if (sym->st_shndx == SHN_UNDEF) {
1579                                 if (sdp->sd_aux && sdp->sd_aux->sa_overndx)
1580                                         sdp->sd_aux->sa_overndx = 0;
1581                         } else {
1582                                 if (!SYM_IS_HIDDEN(sdp) && sdp->sd_aux &&
1583                                     (sdp->sd_aux->sa_overndx == 0)) {
1584                                         sym_undef_entry(ofl, sdp, NOVERSION,
1585                                             verdesc, &undef_state);
1586                                         continue;
1587                                 }
1588                         }
1589                 }
1590 
1591                 /*
1592                  * If we don't need the symbol there's no need to process it
1593                  * any further.
1594                  */
1595                 if (sdp->sd_ref == REF_DYN_SEEN)
1596                         continue;
1597 
1598                 /*
1599                  * Calculate the size and alignment requirements for the global
1600                  * .bss and .tls sections.  If we're building a relocatable
1601                  * object only account for scoped COMMON symbols (these will
1602                  * be converted to .bss references).
1603                  *
1604                  * When -z nopartial is in effect, partially initialized
1605                  * symbols are directed to the special .data section
1606                  * created for that purpose (ofl->ofl_isparexpn).
1607                  * Otherwise, partially initialized symbols go to .bss.
1608                  *
1609                  * Also refer to make_mvsections() in sunwmove.c
1610                  */
1611                 if ((sym->st_shndx == SHN_COMMON) &&
1612                     (((oflags & FLG_OF_RELOBJ) == 0) ||
1613                     (SYM_IS_HIDDEN(sdp) && (oflags & FLG_OF_PROCRED)))) {
1614                         if ((sdp->sd_move == NULL) ||
1615                             ((sdp->sd_flags & FLG_SY_PAREXPN) == 0)) {
1616                                 if (type != STT_TLS) {
1617                                         need_bss = TRUE;
1618                                         bsssize = (Xword)S_ROUND(bsssize,
1619                                             sym->st_value) + sym->st_size;
1620                                         if (sym->st_value > bssalign)
1621                                                 bssalign = sym->st_value;
1622                                 } else {
1623                                         need_tlsbss = TRUE;
1624                                         tlssize = (Xword)S_ROUND(tlssize,
1625                                             sym->st_value) + sym->st_size;
1626                                         if (sym->st_value > tlsalign)
1627                                                 tlsalign = sym->st_value;
1628                                 }
1629                         }
1630                 }
1631 
1632 #if     defined(_ELF64)
1633                 /*
1634                  * Calculate the size and alignment requirement for the global
1635                  * .lbss. TLS or partially initialized symbols do not need to be
1636                  * considered yet.
1637                  */
1638                 if ((ld_targ.t_m.m_mach == EM_AMD64) &&
1639                     (sym->st_shndx == SHN_X86_64_LCOMMON)) {
1640                         need_lbss = TRUE;
1641                         lbsssize = (Xword)S_ROUND(lbsssize, sym->st_value) +
1642                             sym->st_size;
1643                         if (sym->st_value > lbssalign)
1644                                 lbssalign = sym->st_value;
1645                 }
1646 #endif
1647                 /*
1648                  * If a symbol was referenced via the command line
1649                  * (ld -u <>, ...), then this counts as a reference against the
1650                  * symbol. Mark any section that symbol is defined in.
1651                  */
1652                 if (((isp = sdp->sd_isc) != 0) &&
1653                     (sdp->sd_flags & FLG_SY_CMDREF)) {
1654                         isp->is_flags |= FLG_IS_SECTREF;
1655                         isp->is_file->ifl_flags |= FLG_IF_FILEREF;
1656                 }
1657 
1658                 /*
1659                  * Update the symbol count and the associated name string size.
1660                  * Note, a capabilities symbol must remain as visible as a
1661                  * global symbol.  However, the runtime linker recognizes the
1662                  * hidden requirement and ensures the symbol isn't made globally
1663                  * available at runtime.
1664                  */
1665                 if (SYM_IS_HIDDEN(sdp) && (oflags & FLG_OF_PROCRED)) {
1666                         /*
1667                          * If any reductions are being processed, keep a count
1668                          * of eliminated symbols, and if the symbol is being
1669                          * reduced to local, count it's size for the .symtab.
1670                          */
1671                         if (sdp->sd_flags & FLG_SY_ELIM) {
1672                                 ofl->ofl_elimcnt++;
1673                         } else {
1674                                 ofl->ofl_scopecnt++;
1675                                 if ((((sdp->sd_flags & FLG_SY_REGSYM) == 0) ||
1676                                     sym->st_name) && (st_insert(ofl->ofl_strtab,
1677                                     sdp->sd_name) == -1))
1678                                         return (S_ERROR);
1679                                 if (allow_ldynsym && sym->st_name &&
1680                                     ldynsym_symtype[type]) {
1681                                         ofl->ofl_dynscopecnt++;
1682                                         if (st_insert(ofl->ofl_dynstrtab,
1683                                             sdp->sd_name) == -1)
1684                                                 return (S_ERROR);
1685                                         /* Include it in sort section? */
1686                                         DYNSORT_COUNT(sdp, sym, type, ++);
1687                                 }
1688                         }
1689                 } else {
1690                         ofl->ofl_globcnt++;
1691 
1692                         /*
1693                          * Check to see if this global variable should go into
1694                          * a sort section. Sort sections require a
1695                          * .SUNW_ldynsym section, so, don't check unless a
1696                          * .SUNW_ldynsym is allowed.
1697                          */
1698                         if (allow_ldynsym)
1699                                 DYNSORT_COUNT(sdp, sym, type, ++);
1700 
1701                         /*
1702                          * If global direct bindings are in effect, or this
1703                          * symbol has bound to a dependency which was specified
1704                          * as requiring direct bindings, and it hasn't
1705                          * explicitly been defined as a non-direct binding
1706                          * symbol, mark it.
1707                          */
1708                         if (((ofl->ofl_dtflags_1 & DF_1_DIRECT) || (isp &&
1709                             (isp->is_file->ifl_flags & FLG_IF_DIRECT))) &&
1710                             ((sdp->sd_flags & FLG_SY_NDIR) == 0))
1711                                 sdp->sd_flags |= FLG_SY_DIR;
1712 
1713                         /*
1714                          * Insert the symbol name.
1715                          */
1716                         if (((sdp->sd_flags & FLG_SY_REGSYM) == 0) ||
1717                             sym->st_name) {
1718                                 if (st_insert(ofl->ofl_strtab,
1719                                     sdp->sd_name) == -1)
1720                                         return (S_ERROR);
1721 
1722                                 if (!(ofl->ofl_flags & FLG_OF_RELOBJ) &&
1723                                     (st_insert(ofl->ofl_dynstrtab,
1724                                     sdp->sd_name) == -1))
1725                                         return (S_ERROR);
1726                         }
1727 
1728                         /*
1729                          * If this section offers a global symbol - record that
1730                          * fact.
1731                          */
1732                         if (isp) {
1733                                 isp->is_flags |= FLG_IS_SECTREF;
1734                                 isp->is_file->ifl_flags |= FLG_IF_FILEREF;
1735                         }
1736                 }
1737         }
1738 
1739         /*
1740          * Guidance: Use -z defs|nodefs when building shared objects.
1741          *
1742          * Our caller issues this, unless we mask it out here. So we mask it
1743          * out unless we've issued at least one warnings or fatal error.
1744          */
1745         if (!((oflags & FLG_OF_SHAROBJ) && OFL_GUIDANCE(ofl, FLG_OFG_NO_DEFS) &&
1746             (undef_state & (FLG_OF_FATAL | FLG_OF_WARN))))
1747                 ofl->ofl_guideflags |= FLG_OFG_NO_DEFS;
1748 
1749         /*
1750          * If we've encountered a fatal error during symbol validation then
1751          * return now.
1752          */
1753         if (ofl->ofl_flags & FLG_OF_FATAL)
1754                 return (1);
1755 
1756         /*
1757          * Now that symbol resolution is completed, scan any register symbols.
1758          * From now on, we're only interested in those that contribute to the
1759          * output file.
1760          */
1761         if (ofl->ofl_regsyms) {
1762                 int     ndx;
1763 
1764                 for (ndx = 0; ndx < ofl->ofl_regsymsno; ndx++) {
1765                         if ((sdp = ofl->ofl_regsyms[ndx]) == NULL)
1766                                 continue;
1767                         if (sdp->sd_ref != REF_REL_NEED) {
1768                                 ofl->ofl_regsyms[ndx] = NULL;
1769                                 continue;
1770                         }
1771 
1772                         ofl->ofl_regsymcnt++;
1773                         if (sdp->sd_sym->st_name == 0)
1774                                 sdp->sd_name = MSG_ORIG(MSG_STR_EMPTY);
1775 
1776                         if (SYM_IS_HIDDEN(sdp) ||
1777                             (ELF_ST_BIND(sdp->sd_sym->st_info) == STB_LOCAL))
1778                                 ofl->ofl_lregsymcnt++;
1779                 }
1780         }
1781 
1782         /*
1783          * Generate the .bss section now that we know its size and alignment.
1784          */
1785         if (need_bss) {
1786                 if (ld_make_bss(ofl, bsssize, bssalign,
1787                     ld_targ.t_id.id_bss) == S_ERROR)
1788                         return (S_ERROR);
1789         }
1790         if (need_tlsbss) {
1791                 if (ld_make_bss(ofl, tlssize, tlsalign,
1792                     ld_targ.t_id.id_tlsbss) == S_ERROR)
1793                         return (S_ERROR);
1794         }
1795 #if     defined(_ELF64)
1796         if ((ld_targ.t_m.m_mach == EM_AMD64) &&
1797             need_lbss && !(oflags & FLG_OF_RELOBJ)) {
1798                 if (ld_make_bss(ofl, lbsssize, lbssalign,
1799                     ld_targ.t_id.id_lbss) == S_ERROR)
1800                         return (S_ERROR);
1801         }
1802 #endif
1803         /*
1804          * Determine what entry point symbol we need, and if found save its
1805          * symbol descriptor so that we can update the ELF header entry with the
1806          * symbols value later (see update_oehdr).  Make sure the symbol is
1807          * tagged to ensure its update in case -s is in effect.  Use any -e
1808          * option first, or the default entry points `_start' and `main'.
1809          */
1810         ret = 0;
1811         if (ofl->ofl_entry) {
1812                 if ((sdp = ld_sym_find(ofl->ofl_entry, SYM_NOHASH,
1813                     NULL, ofl)) == NULL) {
1814                         ld_eprintf(ofl, ERR_FATAL, MSG_INTL(MSG_ARG_NOENTRY),
1815                             ofl->ofl_entry);
1816                         ret++;
1817                 } else if (ensure_sym_local(ofl, sdp,
1818                     MSG_INTL(MSG_SYM_ENTRY)) != 0) {
1819                         ret++;
1820                 } else {
1821                         ofl->ofl_entry = (void *)sdp;
1822                 }
1823         } else if (((sdp = ld_sym_find(MSG_ORIG(MSG_SYM_START),
1824             SYM_NOHASH, NULL, ofl)) != NULL) && (ensure_sym_local(ofl,
1825             sdp, 0) == 0)) {
1826                 ofl->ofl_entry = (void *)sdp;
1827 
1828         } else if (((sdp = ld_sym_find(MSG_ORIG(MSG_SYM_MAIN),
1829             SYM_NOHASH, NULL, ofl)) != NULL) && (ensure_sym_local(ofl,
1830             sdp, 0) == 0)) {
1831                 ofl->ofl_entry = (void *)sdp;
1832         }
1833 
1834         /*
1835          * If ld -zdtrace=<sym> was given, then validate that the symbol is
1836          * defined within the current object being built.
1837          */
1838         if ((sdp = ofl->ofl_dtracesym) != 0)
1839                 ret += ensure_sym_local(ofl, sdp, MSG_ORIG(MSG_STR_DTRACE));
1840 
1841         /*
1842          * If any initarray, finiarray or preinitarray functions have been
1843          * requested, make sure they are defined within the current object
1844          * being built.
1845          */
1846         if (ofl->ofl_initarray) {
1847                 ret += ensure_array_local(ofl, ofl->ofl_initarray,
1848                     MSG_ORIG(MSG_SYM_INITARRAY));
1849         }
1850         if (ofl->ofl_finiarray) {
1851                 ret += ensure_array_local(ofl, ofl->ofl_finiarray,
1852                     MSG_ORIG(MSG_SYM_FINIARRAY));
1853         }
1854         if (ofl->ofl_preiarray) {
1855                 ret += ensure_array_local(ofl, ofl->ofl_preiarray,
1856                     MSG_ORIG(MSG_SYM_PREINITARRAY));
1857         }
1858 
1859         if (ret)
1860                 return (S_ERROR);
1861 
1862         /*
1863          * If we're required to record any needed dependencies versioning
1864          * information calculate it now that all symbols have been validated.
1865          */
1866         if ((oflags & (FLG_OF_VERNEED | FLG_OF_NOVERSEC)) == FLG_OF_VERNEED)
1867                 return (ld_vers_check_need(ofl));
1868         else
1869                 return (1);
1870 }
1871 
1872 /*
1873  * qsort(3c) comparison function.  As an optimization for associating weak
1874  * symbols to their strong counterparts sort global symbols according to their
1875  * section index, address and binding.
1876  */
1877 static int
1878 compare(const void *sdpp1, const void *sdpp2)
1879 {
1880         Sym_desc        *sdp1 = *((Sym_desc **)sdpp1);
1881         Sym_desc        *sdp2 = *((Sym_desc **)sdpp2);
1882         Sym             *sym1, *sym2;
1883         uchar_t         bind1, bind2;
1884 
1885         /*
1886          * Symbol descriptors may be zero, move these to the front of the
1887          * sorted array.
1888          */
1889         if (sdp1 == NULL)
1890                 return (-1);
1891         if (sdp2 == NULL)
1892                 return (1);
1893 
1894         sym1 = sdp1->sd_sym;
1895         sym2 = sdp2->sd_sym;
1896 
1897         /*
1898          * Compare the symbols section index.  This is important when sorting
1899          * the symbol tables of relocatable objects.  In this case, a symbols
1900          * value is the offset within the associated section, and thus many
1901          * symbols can have the same value, but are effectively different
1902          * addresses.
1903          */
1904         if (sym1->st_shndx > sym2->st_shndx)
1905                 return (1);
1906         if (sym1->st_shndx < sym2->st_shndx)
1907                 return (-1);
1908 
1909         /*
1910          * Compare the symbols value (address).
1911          */
1912         if (sym1->st_value > sym2->st_value)
1913                 return (1);
1914         if (sym1->st_value < sym2->st_value)
1915                 return (-1);
1916 
1917         bind1 = ELF_ST_BIND(sym1->st_info);
1918         bind2 = ELF_ST_BIND(sym2->st_info);
1919 
1920         /*
1921          * If two symbols have the same address place the weak symbol before
1922          * any strong counterpart.
1923          */
1924         if (bind1 > bind2)
1925                 return (-1);
1926         if (bind1 < bind2)
1927                 return (1);
1928 
1929         return (0);
1930 }
1931 
1932 /*
1933  * Issue a MSG_SYM_BADADDR error from ld_sym_process(). This error
1934  * is issued when a symbol address/size is not contained by the
1935  * target section.
1936  *
1937  * Such objects are at least partially corrupt, and the user would
1938  * be well advised to be skeptical of them, and to ask their compiler
1939  * supplier to fix the problem. However, a distinction needs to be
1940  * made between symbols that reference readonly text, and those that
1941  * access writable data. Other than throwing off profiling results,
1942  * the readonly section case is less serious. We have encountered
1943  * such objects in the field. In order to allow existing objects
1944  * to continue working, we issue a warning rather than a fatal error
1945  * if the symbol is against readonly text. Other cases are fatal.
1946  */
1947 static void
1948 issue_badaddr_msg(Ifl_desc *ifl, Ofl_desc *ofl, Sym_desc *sdp,
1949     Sym *sym, Word shndx)
1950 {
1951         Error           err;
1952         const char      *msg;
1953 
1954         if ((sdp->sd_isc->is_shdr->sh_flags & (SHF_WRITE | SHF_ALLOC)) ==
1955             SHF_ALLOC) {
1956                 msg = MSG_INTL(MSG_SYM_BADADDR_ROTXT);
1957                 err = ERR_WARNING;
1958         } else {
1959                 msg = MSG_INTL(MSG_SYM_BADADDR);
1960                 err = ERR_FATAL;
1961         }
1962 
1963         ld_eprintf(ofl, err, msg, demangle(sdp->sd_name),
1964             ifl->ifl_name, shndx, sdp->sd_isc->is_name,
1965             EC_XWORD(sdp->sd_isc->is_shdr->sh_size),
1966             EC_XWORD(sym->st_value), EC_XWORD(sym->st_size));
1967 }
1968 
1969 /*
1970  * Global symbols that are candidates for translation to local capability
1971  * symbols under -z symbolcap, are maintained on a local symbol list.  Once
1972  * all symbols of a file are processed, this list is traversed to cull any
1973  * unnecessary weak symbol aliases.
1974  */
1975 typedef struct {
1976         Sym_desc        *c_nsdp;        /* new lead symbol */
1977         Sym_desc        *c_osdp;        /* original symbol */
1978         Cap_group       *c_group;       /* symbol capability group */
1979         Word            c_ndx;          /* symbol index */
1980 } Cap_pair;
1981 
1982 /*
1983  * Process the symbol table for the specified input file.  At this point all
1984  * input sections from this input file have been assigned an input section
1985  * descriptor which is saved in the `ifl_isdesc' array.
1986  *
1987  *  -   local symbols are saved (as is) if the input file is a  relocatable
1988  *      object
1989  *
1990  *  -   global symbols are added to the linkers internal symbol table if they
1991  *      are not already present, otherwise a symbol resolution function is
1992  *      called upon to resolve the conflict.
1993  */
1994 uintptr_t
1995 ld_sym_process(Is_desc *isc, Ifl_desc *ifl, Ofl_desc *ofl)
1996 {
1997         /*
1998          * This macro tests the given symbol to see if it is out of
1999          * range relative to the section it references.
2000          *
2001          * entry:
2002          *      - ifl is a relative object (ET_REL)
2003          *      _sdp - Symbol descriptor
2004          *      _sym - Symbol
2005          *      _type - Symbol type
2006          *
2007          * The following are tested:
2008          *      - Symbol length is non-zero
2009          *      - Symbol type is a type that references code or data
2010          *      - Referenced section is not 0 (indicates an UNDEF symbol)
2011          *        and is not in the range of special values above SHN_LORESERVE
2012          *        (excluding SHN_XINDEX, which is OK).
2013          *      - We have a valid section header for the target section
2014          *
2015          * If the above are all true, and the symbol position is not
2016          * contained by the target section, this macro evaluates to
2017          * True (1). Otherwise, False(0).
2018          */
2019 #define SYM_LOC_BADADDR(_sdp, _sym, _type) \
2020         (_sym->st_size && dynsymsort_symtype[_type] && \
2021         (_sym->st_shndx != SHN_UNDEF) && \
2022         ((_sym->st_shndx < SHN_LORESERVE) || \
2023                 (_sym->st_shndx == SHN_XINDEX)) && \
2024         _sdp->sd_isc && _sdp->sd_isc->is_shdr && \
2025         ((_sym->st_value + _sym->st_size) > _sdp->sd_isc->is_shdr->sh_size))
2026 
2027         Conv_inv_buf_t  inv_buf;
2028         Sym             *sym = (Sym *)isc->is_indata->d_buf;
2029         Word            *symshndx = NULL;
2030         Shdr            *shdr = isc->is_shdr;
2031         Sym_desc        *sdp;
2032         size_t          strsize;
2033         char            *strs;
2034         uchar_t         type, bind;
2035         Word            ndx, hash, local, total;
2036         uchar_t         osabi = ifl->ifl_ehdr->e_ident[EI_OSABI];
2037         Half            mach = ifl->ifl_ehdr->e_machine;
2038         Half            etype = ifl->ifl_ehdr->e_type;
2039         int             etype_rel;
2040         const char      *symsecname, *strsecname;
2041         Word            symsecndx;
2042         avl_index_t     where;
2043         int             test_gnu_hidden_bit, weak;
2044         Cap_desc        *cdp = NULL;
2045         Alist           *cappairs = NULL;
2046 
2047         /*
2048          * Its possible that a file may contain more that one symbol table,
2049          * ie. .dynsym and .symtab in a shared library.  Only process the first
2050          * table (here, we assume .dynsym comes before .symtab).
2051          */
2052         if (ifl->ifl_symscnt)
2053                 return (1);
2054 
2055         if (isc->is_symshndx)
2056                 symshndx = isc->is_symshndx->is_indata->d_buf;
2057 
2058         DBG_CALL(Dbg_syms_process(ofl->ofl_lml, ifl));
2059 
2060         symsecndx = isc->is_scnndx;
2061         if (isc->is_name)
2062                 symsecname = isc->is_name;
2063         else
2064                 symsecname = MSG_ORIG(MSG_STR_EMPTY);
2065 
2066         /*
2067          * From the symbol tables section header information determine which
2068          * strtab table is needed to locate the actual symbol names.
2069          */
2070         if (ifl->ifl_flags & FLG_IF_HSTRTAB) {
2071                 ndx = shdr->sh_link;
2072                 if ((ndx == 0) || (ndx >= ifl->ifl_shnum)) {
2073                         ld_eprintf(ofl, ERR_FATAL,
2074                             MSG_INTL(MSG_FIL_INVSHLINK), ifl->ifl_name,
2075                             EC_WORD(symsecndx), symsecname, EC_XWORD(ndx));
2076                         return (S_ERROR);
2077                 }
2078                 strsize = ifl->ifl_isdesc[ndx]->is_shdr->sh_size;
2079                 strs = ifl->ifl_isdesc[ndx]->is_indata->d_buf;
2080                 if (ifl->ifl_isdesc[ndx]->is_name)
2081                         strsecname = ifl->ifl_isdesc[ndx]->is_name;
2082                 else
2083                         strsecname = MSG_ORIG(MSG_STR_EMPTY);
2084         } else {
2085                 /*
2086                  * There is no string table section in this input file
2087                  * although there are symbols in this symbol table section.
2088                  * This means that these symbols do not have names.
2089                  * Currently, only scratch register symbols are allowed
2090                  * not to have names.
2091                  */
2092                 strsize = 0;
2093                 strs = (char *)MSG_ORIG(MSG_STR_EMPTY);
2094                 strsecname = MSG_ORIG(MSG_STR_EMPTY);
2095         }
2096 
2097         /*
2098          * Determine the number of local symbols together with the total
2099          * number we have to process.
2100          */
2101         total = (Word)(shdr->sh_size / shdr->sh_entsize);
2102         local = shdr->sh_info;
2103 
2104         /*
2105          * Allocate a symbol table index array and a local symbol array
2106          * (global symbols are processed and added to the ofl->ofl_symbkt[]
2107          * array).  If we are dealing with a relocatable object, allocate the
2108          * local symbol descriptors.  If this isn't a relocatable object we
2109          * still have to process any shared object locals to determine if any
2110          * register symbols exist.  Although these aren't added to the output
2111          * image, they are used as part of symbol resolution.
2112          */
2113         if ((ifl->ifl_oldndx = libld_malloc((size_t)(total *
2114             sizeof (Sym_desc *)))) == NULL)
2115                 return (S_ERROR);
2116         etype_rel = (etype == ET_REL);
2117         if (etype_rel && local) {
2118                 if ((ifl->ifl_locs =
2119                     libld_calloc(sizeof (Sym_desc), local)) == NULL)
2120                         return (S_ERROR);
2121                 /* LINTED */
2122                 ifl->ifl_locscnt = (Word)local;
2123         }
2124         ifl->ifl_symscnt = total;
2125 
2126         /*
2127          * If there are local symbols to save add them to the symbol table
2128          * index array.
2129          */
2130         if (local) {
2131                 int             allow_ldynsym = OFL_ALLOW_LDYNSYM(ofl);
2132                 Sym_desc        *last_file_sdp = NULL;
2133                 int             last_file_ndx = 0;
2134 
2135                 for (sym++, ndx = 1; ndx < local; sym++, ndx++) {
2136                         sd_flag_t       sdflags = FLG_SY_CLEAN;
2137                         Word            shndx;
2138                         const char      *name;
2139                         Sym_desc        *rsdp;
2140                         int             shndx_bad = 0;
2141                         int             symtab_enter = 1;
2142 
2143                         /*
2144                          * Determine and validate the associated section index.
2145                          */
2146                         if (symshndx && (sym->st_shndx == SHN_XINDEX)) {
2147                                 shndx = symshndx[ndx];
2148                         } else if ((shndx = sym->st_shndx) >= SHN_LORESERVE) {
2149                                 sdflags |= FLG_SY_SPECSEC;
2150                         } else if (shndx > ifl->ifl_shnum) {
2151                                 /* We need the name before we can issue error */
2152                                 shndx_bad = 1;
2153                         }
2154 
2155                         /*
2156                          * Check if st_name has a valid value or not.
2157                          */
2158                         if ((name = string(ofl, ifl, sym, strs, strsize, ndx,
2159                             shndx, symsecndx, symsecname, strsecname,
2160                             &sdflags)) == NULL)
2161                                 continue;
2162 
2163                         /*
2164                          * Now that we have the name, if the section index
2165                          * was bad, report it.
2166                          */
2167                         if (shndx_bad) {
2168                                 ld_eprintf(ofl, ERR_WARNING,
2169                                     MSG_INTL(MSG_SYM_INVSHNDX),
2170                                     demangle_symname(name, symsecname, ndx),
2171                                     ifl->ifl_name,
2172                                     conv_sym_shndx(osabi, mach, sym->st_shndx,
2173                                     CONV_FMT_DECIMAL, &inv_buf));
2174                                 continue;
2175                         }
2176 
2177                         /*
2178                          * If this local symbol table originates from a shared
2179                          * object, then we're only interested in recording
2180                          * register symbols.  As local symbol descriptors aren't
2181                          * allocated for shared objects, one will be allocated
2182                          * to associated with the register symbol.  This symbol
2183                          * won't become part of the output image, but we must
2184                          * process it to test for register conflicts.
2185                          */
2186                         rsdp = sdp = NULL;
2187                         if (sdflags & FLG_SY_REGSYM) {
2188                                 /*
2189                                  * The presence of FLG_SY_REGSYM means that
2190                                  * the pointers in ld_targ.t_ms are non-NULL.
2191                                  */
2192                                 rsdp = (*ld_targ.t_ms.ms_reg_find)(sym, ofl);
2193                                 if (rsdp != 0) {
2194                                         /*
2195                                          * The fact that another register def-
2196                                          * inition has been found is fatal.
2197                                          * Call the verification routine to get
2198                                          * the error message and move on.
2199                                          */
2200                                         (void) (*ld_targ.t_ms.ms_reg_check)
2201                                             (rsdp, sym, name, ifl, ofl);
2202                                         continue;
2203                                 }
2204 
2205                                 if (etype == ET_DYN) {
2206                                         if ((sdp = libld_calloc(
2207                                             sizeof (Sym_desc), 1)) == NULL)
2208                                                 return (S_ERROR);
2209                                         sdp->sd_ref = REF_DYN_SEEN;
2210 
2211                                         /* Will not appear in output object */
2212                                         symtab_enter = 0;
2213                                 }
2214                         } else if (etype == ET_DYN)
2215                                 continue;
2216 
2217                         /*
2218                          * Fill in the remaining symbol descriptor information.
2219                          */
2220                         if (sdp == NULL) {
2221                                 sdp = &(ifl->ifl_locs[ndx]);
2222                                 sdp->sd_ref = REF_REL_NEED;
2223                                 sdp->sd_symndx = ndx;
2224                         }
2225                         if (rsdp == NULL) {
2226                                 sdp->sd_name = name;
2227                                 sdp->sd_sym = sym;
2228                                 sdp->sd_shndx = shndx;
2229                                 sdp->sd_flags = sdflags;
2230                                 sdp->sd_file = ifl;
2231                                 ifl->ifl_oldndx[ndx] = sdp;
2232                         }
2233 
2234                         DBG_CALL(Dbg_syms_entry(ofl->ofl_lml, ndx, sdp));
2235 
2236                         /*
2237                          * Reclassify any SHN_SUNW_IGNORE symbols to SHN_UNDEF
2238                          * so as to simplify future processing.
2239                          */
2240                         if (sym->st_shndx == SHN_SUNW_IGNORE) {
2241                                 sdp->sd_shndx = shndx = SHN_UNDEF;
2242                                 sdp->sd_flags |= (FLG_SY_IGNORE | FLG_SY_ELIM);
2243                         }
2244 
2245                         /*
2246                          * Process any register symbols.
2247                          */
2248                         if (sdp->sd_flags & FLG_SY_REGSYM) {
2249                                 /*
2250                                  * Add a diagnostic to indicate we've caught a
2251                                  * register symbol, as this can be useful if a
2252                                  * register conflict is later discovered.
2253                                  */
2254                                 DBG_CALL(Dbg_syms_entered(ofl, sym, sdp));
2255 
2256                                 /*
2257                                  * If this register symbol hasn't already been
2258                                  * recorded, enter it now.
2259                                  *
2260                                  * The presence of FLG_SY_REGSYM means that
2261                                  * the pointers in ld_targ.t_ms are non-NULL.
2262                                  */
2263                                 if ((rsdp == NULL) &&
2264                                     ((*ld_targ.t_ms.ms_reg_enter)(sdp, ofl) ==
2265                                     0))
2266                                         return (S_ERROR);
2267                         }
2268 
2269                         /*
2270                          * Assign an input section.
2271                          */
2272                         if ((sym->st_shndx != SHN_UNDEF) &&
2273                             ((sdp->sd_flags & FLG_SY_SPECSEC) == 0))
2274                                 sdp->sd_isc = ifl->ifl_isdesc[shndx];
2275 
2276                         /*
2277                          * If this symbol falls within the range of a section
2278                          * being discarded, then discard the symbol itself.
2279                          * There is no reason to keep this local symbol.
2280                          */
2281                         if (sdp->sd_isc &&
2282                             (sdp->sd_isc->is_flags & FLG_IS_DISCARD)) {
2283                                 sdp->sd_flags |= FLG_SY_ISDISC;
2284                                 DBG_CALL(Dbg_syms_discarded(ofl->ofl_lml, sdp));
2285                                 continue;
2286                         }
2287 
2288                         /*
2289                          * Skip any section symbols as new versions of these
2290                          * will be created.
2291                          */
2292                         if ((type = ELF_ST_TYPE(sym->st_info)) == STT_SECTION) {
2293                                 if (sym->st_shndx == SHN_UNDEF) {
2294                                         ld_eprintf(ofl, ERR_WARNING,
2295                                             MSG_INTL(MSG_SYM_INVSHNDX),
2296                                             demangle_symname(name, symsecname,
2297                                             ndx), ifl->ifl_name,
2298                                             conv_sym_shndx(osabi, mach,
2299                                             sym->st_shndx, CONV_FMT_DECIMAL,
2300                                             &inv_buf));
2301                                 }
2302                                 continue;
2303                         }
2304 
2305                         /*
2306                          * For a relocatable object, if this symbol is defined
2307                          * and has non-zero length and references an address
2308                          * within an associated section, then check its extents
2309                          * to make sure the section boundaries encompass it.
2310                          * If they don't, the ELF file is corrupt.
2311                          */
2312                         if (etype_rel) {
2313                                 if (SYM_LOC_BADADDR(sdp, sym, type)) {
2314                                         issue_badaddr_msg(ifl, ofl, sdp,
2315                                             sym, shndx);
2316                                         if (ofl->ofl_flags & FLG_OF_FATAL)
2317                                                 continue;
2318                                 }
2319 
2320                                 /*
2321                                  * We have observed relocatable objects
2322                                  * containing identical adjacent STT_FILE
2323                                  * symbols. Discard any other than the first,
2324                                  * as they are all equivalent and the extras
2325                                  * do not add information.
2326                                  *
2327                                  * For the purpose of this test, we assume
2328                                  * that only the symbol type and the string
2329                                  * table offset (st_name) matter.
2330                                  */
2331                                 if (type == STT_FILE) {
2332                                         int toss = (last_file_sdp != NULL) &&
2333                                             ((ndx - 1) == last_file_ndx) &&
2334                                             (sym->st_name ==
2335                                             last_file_sdp->sd_sym->st_name);
2336 
2337                                         last_file_sdp = sdp;
2338                                         last_file_ndx = ndx;
2339                                         if (toss) {
2340                                                 sdp->sd_flags |= FLG_SY_INVALID;
2341                                                 DBG_CALL(Dbg_syms_dup_discarded(
2342                                                     ofl->ofl_lml, ndx, sdp));
2343                                                 continue;
2344                                         }
2345                                 }
2346                         }
2347 
2348 
2349                         /*
2350                          * Sanity check for TLS
2351                          */
2352                         if ((sym->st_size != 0) && ((type == STT_TLS) &&
2353                             (sym->st_shndx != SHN_COMMON))) {
2354                                 Is_desc *isp = sdp->sd_isc;
2355 
2356                                 if ((isp == NULL) || (isp->is_shdr == NULL) ||
2357                                     ((isp->is_shdr->sh_flags & SHF_TLS) == 0)) {
2358                                         ld_eprintf(ofl, ERR_FATAL,
2359                                             MSG_INTL(MSG_SYM_TLS),
2360                                             demangle(sdp->sd_name),
2361                                             ifl->ifl_name);
2362                                         continue;
2363                                 }
2364                         }
2365 
2366                         /*
2367                          * Carry our some basic sanity checks (these are just
2368                          * some of the erroneous symbol entries we've come
2369                          * across, there's probably a lot more).  The symbol
2370                          * will not be carried forward to the output file, which
2371                          * won't be a problem unless a relocation is required
2372                          * against it.
2373                          */
2374                         if (((sdp->sd_flags & FLG_SY_SPECSEC) &&
2375                             ((sym->st_shndx == SHN_COMMON)) ||
2376                             ((type == STT_FILE) &&
2377                             (sym->st_shndx != SHN_ABS))) ||
2378                             (sdp->sd_isc && (sdp->sd_isc->is_osdesc == NULL))) {
2379                                 ld_eprintf(ofl, ERR_WARNING,
2380                                     MSG_INTL(MSG_SYM_INVSHNDX),
2381                                     demangle_symname(name, symsecname, ndx),
2382                                     ifl->ifl_name,
2383                                     conv_sym_shndx(osabi, mach, sym->st_shndx,
2384                                     CONV_FMT_DECIMAL, &inv_buf));
2385                                 sdp->sd_isc = NULL;
2386                                 sdp->sd_flags |= FLG_SY_INVALID;
2387                                 continue;
2388                         }
2389 
2390                         /*
2391                          * As these local symbols will become part of the output
2392                          * image, record their number and name string size.
2393                          * Globals are counted after all input file processing
2394                          * (and hence symbol resolution) is complete during
2395                          * sym_validate().
2396                          */
2397                         if (!(ofl->ofl_flags & FLG_OF_REDLSYM) &&
2398                             symtab_enter) {
2399                                 ofl->ofl_locscnt++;
2400 
2401                                 if ((((sdp->sd_flags & FLG_SY_REGSYM) == 0) ||
2402                                     sym->st_name) && (st_insert(ofl->ofl_strtab,
2403                                     sdp->sd_name) == -1))
2404                                         return (S_ERROR);
2405 
2406                                 if (allow_ldynsym && sym->st_name &&
2407                                     ldynsym_symtype[type]) {
2408                                         ofl->ofl_dynlocscnt++;
2409                                         if (st_insert(ofl->ofl_dynstrtab,
2410                                             sdp->sd_name) == -1)
2411                                                 return (S_ERROR);
2412                                         /* Include it in sort section? */
2413                                         DYNSORT_COUNT(sdp, sym, type, ++);
2414                                 }
2415                         }
2416                 }
2417         }
2418 
2419         /*
2420          * The GNU ld interprets the top bit of the 16-bit Versym value
2421          * (0x8000) as the "hidden" bit. If this bit is set, the linker
2422          * is supposed to act as if that symbol does not exist. The Solaris
2423          * linker does not support this mechanism, or the model of interface
2424          * evolution that it allows, but we honor it in GNU ld produced
2425          * objects in order to interoperate with them.
2426          *
2427          * Determine if we should honor the GNU hidden bit for this file.
2428          */
2429         test_gnu_hidden_bit = ((ifl->ifl_flags & FLG_IF_GNUVER) != 0) &&
2430             (ifl->ifl_versym != NULL);
2431 
2432         /*
2433          * Determine whether object capabilities for this file are being
2434          * converted into symbol capabilities.  If so, global function symbols,
2435          * and initialized global data symbols, need special translation and
2436          * processing.
2437          */
2438         if ((etype == ET_REL) && (ifl->ifl_flags & FLG_IF_OTOSCAP))
2439                 cdp = ifl->ifl_caps;
2440 
2441         /*
2442          * Now scan the global symbols entering them in the internal symbol
2443          * table or resolving them as necessary.
2444          */
2445         sym = (Sym *)isc->is_indata->d_buf;
2446         sym += local;
2447         weak = 0;
2448         /* LINTED */
2449         for (ndx = (int)local; ndx < total; sym++, ndx++) {
2450                 const char      *name;
2451                 sd_flag_t       sdflags = 0;
2452                 Word            shndx;
2453                 int             shndx_bad = 0;
2454                 Sym             *nsym = sym;
2455                 Cap_pair        *cpp = NULL;
2456                 uchar_t         ntype;
2457 
2458                 /*
2459                  * Determine and validate the associated section index.
2460                  */
2461                 if (symshndx && (nsym->st_shndx == SHN_XINDEX)) {
2462                         shndx = symshndx[ndx];
2463                 } else if ((shndx = nsym->st_shndx) >= SHN_LORESERVE) {
2464                         sdflags |= FLG_SY_SPECSEC;
2465                 } else if (shndx > ifl->ifl_shnum) {
2466                         /* We need the name before we can issue error */
2467                         shndx_bad = 1;
2468                 }
2469 
2470                 /*
2471                  * Check if st_name has a valid value or not.
2472                  */
2473                 if ((name = string(ofl, ifl, nsym, strs, strsize, ndx, shndx,
2474                     symsecndx, symsecname, strsecname, &sdflags)) == NULL)
2475                         continue;
2476 
2477                 /*
2478                  * Now that we have the name, report an erroneous section index.
2479                  */
2480                 if (shndx_bad) {
2481                         ld_eprintf(ofl, ERR_WARNING, MSG_INTL(MSG_SYM_INVSHNDX),
2482                             demangle_symname(name, symsecname, ndx),
2483                             ifl->ifl_name,
2484                             conv_sym_shndx(osabi, mach, nsym->st_shndx,
2485                             CONV_FMT_DECIMAL, &inv_buf));
2486                         continue;
2487                 }
2488 
2489                 /*
2490                  * Test for the GNU hidden bit, and ignore symbols that
2491                  * have it set.
2492                  */
2493                 if (test_gnu_hidden_bit &&
2494                     ((ifl->ifl_versym[ndx] & 0x8000) != 0))
2495                         continue;
2496 
2497                 /*
2498                  * The linker itself will generate symbols for _end, _etext,
2499                  * _edata, _DYNAMIC and _PROCEDURE_LINKAGE_TABLE_, so don't
2500                  * bother entering these symbols from shared objects.  This
2501                  * results in some wasted resolution processing, which is hard
2502                  * to feel, but if nothing else, pollutes diagnostic relocation
2503                  * output.
2504                  */
2505                 if (name[0] && (etype == ET_DYN) && (nsym->st_size == 0) &&
2506                     (ELF_ST_TYPE(nsym->st_info) == STT_OBJECT) &&
2507                     (name[0] == '_') && ((name[1] == 'e') ||
2508                     (name[1] == 'D') || (name[1] == 'P')) &&
2509                     ((strcmp(name, MSG_ORIG(MSG_SYM_ETEXT_U)) == 0) ||
2510                     (strcmp(name, MSG_ORIG(MSG_SYM_EDATA_U)) == 0) ||
2511                     (strcmp(name, MSG_ORIG(MSG_SYM_END_U)) == 0) ||
2512                     (strcmp(name, MSG_ORIG(MSG_SYM_DYNAMIC_U)) == 0) ||
2513                     (strcmp(name, MSG_ORIG(MSG_SYM_PLKTBL_U)) == 0))) {
2514                         ifl->ifl_oldndx[ndx] = 0;
2515                         continue;
2516                 }
2517 
2518                 /*
2519                  * The '-z wrap=XXX' option emulates the GNU ld --wrap=XXX
2520                  * option. When XXX is the symbol to be wrapped:
2521                  *
2522                  *  -   An undefined reference to XXX is converted to __wrap_XXX
2523                  *  -   An undefined reference to __real_XXX is converted to XXX
2524                  *
2525                  * The idea is that the user can supply a wrapper function
2526                  * __wrap_XXX that does some work, and then uses the name
2527                  * __real_XXX to pass the call on to the real function. The
2528                  * wrapper objects are linked with the original unmodified
2529                  * objects to produce a wrapped version of the output object.
2530                  */
2531                 if (ofl->ofl_wrap && name[0] && (shndx == SHN_UNDEF)) {
2532                         WrapSymNode wsn, *wsnp;
2533 
2534                         /*
2535                          * If this is the __real_XXX form, advance the
2536                          * pointer to reference the wrapped name.
2537                          */
2538                         wsn.wsn_name = name;
2539                         if ((*name == '_') &&
2540                             (strncmp(name, MSG_ORIG(MSG_STR_UU_REAL_U),
2541                             MSG_STR_UU_REAL_U_SIZE) == 0))
2542                                 wsn.wsn_name += MSG_STR_UU_REAL_U_SIZE;
2543 
2544                         /*
2545                          * Is this symbol in the wrap AVL tree? If so, map
2546                          * XXX to __wrap_XXX, and __real_XXX to XXX. Note that
2547                          * wsn.wsn_name will equal the current value of name
2548                          * if the __real_ prefix is not present.
2549                          */
2550                         if ((wsnp = avl_find(ofl->ofl_wrap, &wsn, 0)) != NULL) {
2551                                 const char *old_name = name;
2552 
2553                                 name = (wsn.wsn_name == name) ?
2554                                     wsnp->wsn_wrapname : wsn.wsn_name;
2555                                 DBG_CALL(Dbg_syms_wrap(ofl->ofl_lml, ndx,
2556                                     old_name, name));
2557                         }
2558                 }
2559 
2560                 /*
2561                  * Determine and validate the symbols binding.
2562                  */
2563                 bind = ELF_ST_BIND(nsym->st_info);
2564                 if ((bind != STB_GLOBAL) && (bind != STB_WEAK)) {
2565                         ld_eprintf(ofl, ERR_WARNING, MSG_INTL(MSG_SYM_NONGLOB),
2566                             demangle_symname(name, symsecname, ndx),
2567                             ifl->ifl_name,
2568                             conv_sym_info_bind(bind, 0, &inv_buf));
2569                         continue;
2570                 }
2571                 if (bind == STB_WEAK)
2572                         weak++;
2573 
2574                 /*
2575                  * If this symbol falls within the range of a section being
2576                  * discarded, then discard the symbol itself.
2577                  */
2578                 if (((sdflags & FLG_SY_SPECSEC) == 0) &&
2579                     (nsym->st_shndx != SHN_UNDEF)) {
2580                         Is_desc *isp;
2581 
2582                         if (shndx >= ifl->ifl_shnum) {
2583                                 /*
2584                                  * Carry our some basic sanity checks
2585                                  * The symbol will not be carried forward to
2586                                  * the output file, which won't be a problem
2587                                  * unless a relocation is required against it.
2588                                  */
2589                                 ld_eprintf(ofl, ERR_WARNING,
2590                                     MSG_INTL(MSG_SYM_INVSHNDX),
2591                                     demangle_symname(name, symsecname, ndx),
2592                                     ifl->ifl_name,
2593                                     conv_sym_shndx(osabi, mach, nsym->st_shndx,
2594                                     CONV_FMT_DECIMAL, &inv_buf));
2595                                 continue;
2596                         }
2597 
2598                         isp = ifl->ifl_isdesc[shndx];
2599                         if (isp && (isp->is_flags & FLG_IS_DISCARD)) {
2600                                 if ((sdp =
2601                                     libld_calloc(sizeof (Sym_desc), 1)) == NULL)
2602                                         return (S_ERROR);
2603 
2604                                 /*
2605                                  * Create a dummy symbol entry so that if we
2606                                  * find any references to this discarded symbol
2607                                  * we can compensate.
2608                                  */
2609                                 sdp->sd_name = name;
2610                                 sdp->sd_sym = nsym;
2611                                 sdp->sd_file = ifl;
2612                                 sdp->sd_isc = isp;
2613                                 sdp->sd_flags = FLG_SY_ISDISC;
2614                                 ifl->ifl_oldndx[ndx] = sdp;
2615 
2616                                 DBG_CALL(Dbg_syms_discarded(ofl->ofl_lml, sdp));
2617                                 continue;
2618                         }
2619                 }
2620 
2621                 /*
2622                  * If object capabilities for this file are being converted
2623                  * into symbol capabilities, then:
2624                  *
2625                  *  -   Any global function, or initialized global data symbol
2626                  *      definitions (ie., those that are not associated with
2627                  *      special symbol types, ie., ABS, COMMON, etc.), and which
2628                  *      have not been reduced to locals, are converted to symbol
2629                  *      references (UNDEF).  This ensures that any reference to
2630                  *      the original symbol, for example from a relocation, get
2631                  *      associated to a capabilities family lead symbol, ie., a
2632                  *      generic instance.
2633                  *
2634                  *  -   For each global function, or object symbol definition,
2635                  *      a new local symbol is created.  The function or object
2636                  *      is renamed using the capabilities CA_SUNW_ID definition
2637                  *      (which might have been fabricated for this purpose -
2638                  *      see get_cap_group()).  The new symbol name is:
2639                  *
2640                  *          <original name>%<capability group identifier>
2641                  *
2642                  *      This symbol is associated to the same location, and
2643                  *      becomes a capabilities family member.
2644                  */
2645                 /* LINTED */
2646                 hash = (Word)elf_hash(name);
2647 
2648                 ntype = ELF_ST_TYPE(nsym->st_info);
2649                 if (cdp && (nsym->st_shndx != SHN_UNDEF) &&
2650                     ((sdflags & FLG_SY_SPECSEC) == 0) &&
2651                     ((ntype == STT_FUNC) || (ntype == STT_OBJECT))) {
2652                         /*
2653                          * Determine this symbol's visibility.  If a mapfile has
2654                          * indicated this symbol should be local, then there's
2655                          * no point in transforming this global symbol to a
2656                          * capabilities symbol.  Otherwise, create a symbol
2657                          * capability pair descriptor to record this symbol as
2658                          * a candidate for translation.
2659                          */
2660                         if (sym_cap_vis(name, hash, sym, ofl) &&
2661                             ((cpp = alist_append(&cappairs, NULL,
2662                             sizeof (Cap_pair), AL_CNT_CAP_PAIRS)) == NULL))
2663                                 return (S_ERROR);
2664                 }
2665 
2666                 if (cpp) {
2667                         Sym     *rsym;
2668 
2669                         DBG_CALL(Dbg_syms_cap_convert(ofl, ndx, name, nsym));
2670 
2671                         /*
2672                          * Allocate a new symbol descriptor to represent the
2673                          * transformed global symbol.  The descriptor points
2674                          * to the original symbol information (which might
2675                          * indicate a global or weak visibility).  The symbol
2676                          * information will be transformed into a local symbol
2677                          * later, after any weak aliases are culled.
2678                          */
2679                         if ((cpp->c_osdp =
2680                             libld_malloc(sizeof (Sym_desc))) == NULL)
2681                                 return (S_ERROR);
2682 
2683                         cpp->c_osdp->sd_name = name;
2684                         cpp->c_osdp->sd_sym = nsym;
2685                         cpp->c_osdp->sd_shndx = shndx;
2686                         cpp->c_osdp->sd_file = ifl;
2687                         cpp->c_osdp->sd_isc = ifl->ifl_isdesc[shndx];
2688                         cpp->c_osdp->sd_ref = REF_REL_NEED;
2689 
2690                         /*
2691                          * Save the capabilities group this symbol belongs to,
2692                          * and the original symbol index.
2693                          */
2694                         cpp->c_group = cdp->ca_groups->apl_data[0];
2695                         cpp->c_ndx = ndx;
2696 
2697                         /*
2698                          * Replace the original symbol definition with a symbol
2699                          * reference.  Make sure this reference isn't left as a
2700                          * weak.
2701                          */
2702                         if ((rsym = libld_malloc(sizeof (Sym))) == NULL)
2703                                 return (S_ERROR);
2704 
2705                         *rsym = *nsym;
2706 
2707                         rsym->st_info = ELF_ST_INFO(STB_GLOBAL, ntype);
2708                         rsym->st_shndx = shndx = SHN_UNDEF;
2709                         rsym->st_value = 0;
2710                         rsym->st_size = 0;
2711 
2712                         sdflags |= FLG_SY_CAP;
2713 
2714                         nsym = rsym;
2715                 }
2716 
2717                 /*
2718                  * If the symbol does not already exist in the internal symbol
2719                  * table add it, otherwise resolve the conflict.  If the symbol
2720                  * from this file is kept, retain its symbol table index for
2721                  * possible use in associating a global alias.
2722                  */
2723                 if ((sdp = ld_sym_find(name, hash, &where, ofl)) == NULL) {
2724                         DBG_CALL(Dbg_syms_global(ofl->ofl_lml, ndx, name));
2725                         if ((sdp = ld_sym_enter(name, nsym, hash, ifl, ofl, ndx,
2726                             shndx, sdflags, &where)) == (Sym_desc *)S_ERROR)
2727                                 return (S_ERROR);
2728 
2729                 } else if (ld_sym_resolve(sdp, nsym, ifl, ofl, ndx, shndx,
2730                     sdflags) == S_ERROR)
2731                         return (S_ERROR);
2732 
2733                 /*
2734                  * Now that we have a symbol descriptor, retain the descriptor
2735                  * for later use by symbol capabilities processing.
2736                  */
2737                 if (cpp)
2738                         cpp->c_nsdp = sdp;
2739 
2740                 /*
2741                  * After we've compared a defined symbol in one shared
2742                  * object, flag the symbol so we don't compare it again.
2743                  */
2744                 if ((etype == ET_DYN) && (nsym->st_shndx != SHN_UNDEF) &&
2745                     ((sdp->sd_flags & FLG_SY_SOFOUND) == 0))
2746                         sdp->sd_flags |= FLG_SY_SOFOUND;
2747 
2748                 /*
2749                  * If the symbol is accepted from this file retain the symbol
2750                  * index for possible use in aliasing.
2751                  */
2752                 if (sdp->sd_file == ifl)
2753                         sdp->sd_symndx = ndx;
2754 
2755                 ifl->ifl_oldndx[ndx] = sdp;
2756 
2757                 /*
2758                  * If we've accepted a register symbol, continue to validate
2759                  * it.
2760                  */
2761                 if (sdp->sd_flags & FLG_SY_REGSYM) {
2762                         Sym_desc        *rsdp;
2763 
2764                         /*
2765                          * The presence of FLG_SY_REGSYM means that
2766                          * the pointers in ld_targ.t_ms are non-NULL.
2767                          */
2768                         rsdp = (*ld_targ.t_ms.ms_reg_find)(sdp->sd_sym, ofl);
2769                         if (rsdp == NULL) {
2770                                 if ((*ld_targ.t_ms.ms_reg_enter)(sdp, ofl) == 0)
2771                                         return (S_ERROR);
2772                         } else if (rsdp != sdp) {
2773                                 (void) (*ld_targ.t_ms.ms_reg_check)(rsdp,
2774                                     sdp->sd_sym, sdp->sd_name, ifl, ofl);
2775                         }
2776                 }
2777 
2778                 /*
2779                  * For a relocatable object, if this symbol is defined
2780                  * and has non-zero length and references an address
2781                  * within an associated section, then check its extents
2782                  * to make sure the section boundaries encompass it.
2783                  * If they don't, the ELF file is corrupt. Note that this
2784                  * global symbol may have come from another file to satisfy
2785                  * an UNDEF symbol of the same name from this one. In that
2786                  * case, we don't check it, because it was already checked
2787                  * as part of its own file.
2788                  */
2789                 if (etype_rel && (sdp->sd_file == ifl)) {
2790                         Sym *tsym = sdp->sd_sym;
2791 
2792                         if (SYM_LOC_BADADDR(sdp, tsym,
2793                             ELF_ST_TYPE(tsym->st_info))) {
2794                                 issue_badaddr_msg(ifl, ofl, sdp,
2795                                     tsym, tsym->st_shndx);
2796                                 continue;
2797                         }
2798                 }
2799         }
2800         DBG_CALL(Dbg_util_nl(ofl->ofl_lml, DBG_NL_STD));
2801 
2802         /*
2803          * Associate weak (alias) symbols to their non-weak counterparts by
2804          * scanning the global symbols one more time.
2805          *
2806          * This association is needed when processing the symbols from a shared
2807          * object dependency when a a weak definition satisfies a reference:
2808          *
2809          *  -   When building a dynamic executable, if a referenced symbol is a
2810          *      data item, the symbol data is copied to the executables address
2811          *      space.  In this copy-relocation case, we must also reassociate
2812          *      the alias symbol with its new location in the executable.
2813          *
2814          *  -   If the referenced symbol is a function then we may need to
2815          *      promote the symbols binding from undefined weak to undefined,
2816          *      otherwise the run-time linker will not generate the correct
2817          *      relocation error should the symbol not be found.
2818          *
2819          * Weak alias association is also required when a local dynsym table
2820          * is being created.  This table should only contain one instance of a
2821          * symbol that is associated to a given address.
2822          *
2823          * The true association between a weak/strong symbol pair is that both
2824          * symbol entries are identical, thus first we create a sorted symbol
2825          * list keyed off of the symbols section index and value.  If the symbol
2826          * belongs to the same section and has the same value, then the chances
2827          * are that the rest of the symbols data is the same.  This list is then
2828          * scanned for weak symbols, and if one is found then any strong
2829          * association will exist in the entries that follow.  Thus we just have
2830          * to scan one (typically a single alias) or more (in the uncommon
2831          * instance of multiple weak to strong associations) entries to
2832          * determine if a match exists.
2833          */
2834         if (weak && (OFL_ALLOW_LDYNSYM(ofl) || (etype == ET_DYN)) &&
2835             (total > local)) {
2836                 static Sym_desc **sort;
2837                 static size_t   osize = 0;
2838                 size_t          nsize = (total - local) * sizeof (Sym_desc *);
2839 
2840                 /*
2841                  * As we might be processing many input files, and many symbols,
2842                  * try and reuse a static sort buffer.  Note, presently we're
2843                  * playing the game of never freeing any buffers as there's a
2844                  * belief this wastes time.
2845                  */
2846                 if ((osize == 0) || (nsize > osize)) {
2847                         if ((sort = libld_malloc(nsize)) == NULL)
2848                                 return (S_ERROR);
2849                         osize = nsize;
2850                 }
2851                 (void) memcpy((void *)sort, &ifl->ifl_oldndx[local], nsize);
2852 
2853                 qsort(sort, (total - local), sizeof (Sym_desc *), compare);
2854 
2855                 for (ndx = 0; ndx < (total - local); ndx++) {
2856                         Sym_desc        *wsdp = sort[ndx];
2857                         Sym             *wsym;
2858                         int             sndx;
2859 
2860                         /*
2861                          * Ignore any empty symbol descriptor, or the case where
2862                          * the symbol has been resolved to a different file.
2863                          */
2864                         if ((wsdp == NULL) || (wsdp->sd_file != ifl))
2865                                 continue;
2866 
2867                         wsym = wsdp->sd_sym;
2868 
2869                         if ((wsym->st_shndx == SHN_UNDEF) ||
2870                             (wsdp->sd_flags & FLG_SY_SPECSEC) ||
2871                             (ELF_ST_BIND(wsym->st_info) != STB_WEAK))
2872                                 continue;
2873 
2874                         /*
2875                          * We have a weak symbol, if it has a strong alias it
2876                          * will have been sorted to one of the following sort
2877                          * table entries.  Note that we could have multiple weak
2878                          * symbols aliased to one strong (if this occurs then
2879                          * the strong symbol only maintains one alias back to
2880                          * the last weak).
2881                          */
2882                         for (sndx = ndx + 1; sndx < (total - local); sndx++) {
2883                                 Sym_desc        *ssdp = sort[sndx];
2884                                 Sym             *ssym;
2885                                 sd_flag_t       w_dynbits, s_dynbits;
2886 
2887                                 /*
2888                                  * Ignore any empty symbol descriptor, or the
2889                                  * case where the symbol has been resolved to a
2890                                  * different file.
2891                                  */
2892                                 if ((ssdp == NULL) || (ssdp->sd_file != ifl))
2893                                         continue;
2894 
2895                                 ssym = ssdp->sd_sym;
2896 
2897                                 if (ssym->st_shndx == SHN_UNDEF)
2898                                         continue;
2899 
2900                                 if ((ssym->st_shndx != wsym->st_shndx) ||
2901                                     (ssym->st_value != wsym->st_value))
2902                                         break;
2903 
2904                                 if ((ssym->st_size != wsym->st_size) ||
2905                                     (ssdp->sd_flags & FLG_SY_SPECSEC) ||
2906                                     (ELF_ST_BIND(ssym->st_info) == STB_WEAK))
2907                                         continue;
2908 
2909                                 /*
2910                                  * If a sharable object, set link fields so
2911                                  * that they reference each other.`
2912                                  */
2913                                 if (etype == ET_DYN) {
2914                                         ssdp->sd_aux->sa_linkndx =
2915                                             (Word)wsdp->sd_symndx;
2916                                         wsdp->sd_aux->sa_linkndx =
2917                                             (Word)ssdp->sd_symndx;
2918                                 }
2919 
2920                                 /*
2921                                  * Determine which of these two symbols go into
2922                                  * the sort section.  If a mapfile has made
2923                                  * explicit settings of the FLG_SY_*DYNSORT
2924                                  * flags for both symbols, then we do what they
2925                                  * say.  If one has the DYNSORT flags set, we
2926                                  * set the NODYNSORT bit in the other.  And if
2927                                  * neither has an explicit setting, then we
2928                                  * favor the weak symbol because they usually
2929                                  * lack the leading underscore.
2930                                  */
2931                                 w_dynbits = wsdp->sd_flags &
2932                                     (FLG_SY_DYNSORT | FLG_SY_NODYNSORT);
2933                                 s_dynbits = ssdp->sd_flags &
2934                                     (FLG_SY_DYNSORT | FLG_SY_NODYNSORT);
2935                                 if (!(w_dynbits && s_dynbits)) {
2936                                         if (s_dynbits) {
2937                                                 if (s_dynbits == FLG_SY_DYNSORT)
2938                                                         wsdp->sd_flags |=
2939                                                             FLG_SY_NODYNSORT;
2940                                         } else if (w_dynbits !=
2941                                             FLG_SY_NODYNSORT) {
2942                                                 ssdp->sd_flags |=
2943                                                     FLG_SY_NODYNSORT;
2944                                         }
2945                                 }
2946                                 break;
2947                         }
2948                 }
2949         }
2950 
2951         /*
2952          * Having processed all symbols, under -z symbolcap, reprocess any
2953          * symbols that are being translated from global to locals.  The symbol
2954          * pair that has been collected defines the original symbol (c_osdp),
2955          * which will become a local, and the new symbol (c_nsdp), which will
2956          * become a reference (UNDEF) for the original.
2957          *
2958          * Scan these symbol pairs looking for weak symbols, which have non-weak
2959          * aliases.  There is no need to translate both of these symbols to
2960          * locals, only the global is necessary.
2961          */
2962         if (cappairs) {
2963                 Aliste          idx1;
2964                 Cap_pair        *cpp1;
2965 
2966                 for (ALIST_TRAVERSE(cappairs, idx1, cpp1)) {
2967                         Sym_desc        *sdp1 = cpp1->c_osdp;
2968                         Sym             *sym1 = sdp1->sd_sym;
2969                         uchar_t         bind1 = ELF_ST_BIND(sym1->st_info);
2970                         Aliste          idx2;
2971                         Cap_pair        *cpp2;
2972 
2973                         /*
2974                          * If this symbol isn't weak, it's capability member is
2975                          * retained for the creation of a local symbol.
2976                          */
2977                         if (bind1 != STB_WEAK)
2978                                 continue;
2979 
2980                         /*
2981                          * If this is a weak symbol, traverse the capabilities
2982                          * list again to determine if a corresponding non-weak
2983                          * symbol exists.
2984                          */
2985                         for (ALIST_TRAVERSE(cappairs, idx2, cpp2)) {
2986                                 Sym_desc        *sdp2 = cpp2->c_osdp;
2987                                 Sym             *sym2 = sdp2->sd_sym;
2988                                 uchar_t         bind2 =
2989                                     ELF_ST_BIND(sym2->st_info);
2990 
2991                                 if ((cpp1 == cpp2) ||
2992                                     (cpp1->c_group != cpp2->c_group) ||
2993                                     (sym1->st_value != sym2->st_value) ||
2994                                     (bind2 == STB_WEAK))
2995                                         continue;
2996 
2997                                 /*
2998                                  * The weak symbol (sym1) has a non-weak (sym2)
2999                                  * counterpart.  There's no point in translating
3000                                  * both of these equivalent symbols to locals.
3001                                  * Add this symbol capability alias to the
3002                                  * capabilities family information, and remove
3003                                  * the weak symbol.
3004                                  */
3005                                 if (ld_cap_add_family(ofl, cpp2->c_nsdp,
3006                                     cpp1->c_nsdp, NULL, NULL) == S_ERROR)
3007                                         return (S_ERROR);
3008 
3009                                 free((void *)cpp1->c_osdp);
3010                                 (void) alist_delete(cappairs, &idx1);
3011                         }
3012                 }
3013 
3014                 DBG_CALL(Dbg_util_nl(ofl->ofl_lml, DBG_NL_STD));
3015 
3016                 /*
3017                  * The capability pairs information now represents all the
3018                  * global symbols that need transforming to locals.  These
3019                  * local symbols are renamed using their group identifiers.
3020                  */
3021                 for (ALIST_TRAVERSE(cappairs, idx1, cpp1)) {
3022                         Sym_desc        *osdp = cpp1->c_osdp;
3023                         Objcapset       *capset;
3024                         size_t          nsize, tsize;
3025                         const char      *oname;
3026                         char            *cname, *idstr;
3027                         Sym             *csym;
3028 
3029                         /*
3030                          * If the local symbol has not yet been translated
3031                          * convert it to a local symbol with a name.
3032                          */
3033                         if ((osdp->sd_flags & FLG_SY_CAP) != 0)
3034                                 continue;
3035 
3036                         /*
3037                          * As we're converting object capabilities to symbol
3038                          * capabilities, obtain the capabilities set for this
3039                          * object, so as to retrieve the CA_SUNW_ID value.
3040                          */
3041                         capset = &cpp1->c_group->cg_set;
3042 
3043                         /*
3044                          * Create a new name from the existing symbol and the
3045                          * capabilities group identifier.  Note, the delimiter
3046                          * between the symbol name and identifier name is hard-
3047                          * coded here (%), so that we establish a convention
3048                          * for transformed symbol names.
3049                          */
3050                         oname = osdp->sd_name;
3051 
3052                         idstr = capset->oc_id.cs_str;
3053                         nsize = strlen(oname);
3054                         tsize = nsize + 1 + strlen(idstr) + 1;
3055                         if ((cname = libld_malloc(tsize)) == 0)
3056                                 return (S_ERROR);
3057 
3058                         (void) strcpy(cname, oname);
3059                         cname[nsize++] = '%';
3060                         (void) strcpy(&cname[nsize], idstr);
3061 
3062                         /*
3063                          * Allocate a new symbol table entry, transform this
3064                          * symbol to a local, and assign the new name.
3065                          */
3066                         if ((csym = libld_malloc(sizeof (Sym))) == NULL)
3067                                 return (S_ERROR);
3068 
3069                         *csym = *osdp->sd_sym;
3070                         csym->st_info = ELF_ST_INFO(STB_LOCAL,
3071                             ELF_ST_TYPE(osdp->sd_sym->st_info));
3072 
3073                         osdp->sd_name = cname;
3074                         osdp->sd_sym = csym;
3075                         osdp->sd_flags = FLG_SY_CAP;
3076 
3077                         /*
3078                          * Keep track of this new local symbol.  As -z symbolcap
3079                          * can only be used to create a relocatable object, a
3080                          * dynamic symbol table can't exist.  Ensure there is
3081                          * space reserved in the string table.
3082                          */
3083                         ofl->ofl_caploclcnt++;
3084                         if (st_insert(ofl->ofl_strtab, cname) == -1)
3085                                 return (S_ERROR);
3086 
3087                         DBG_CALL(Dbg_syms_cap_local(ofl, cpp1->c_ndx,
3088                             cname, csym, osdp));
3089 
3090                         /*
3091                          * Establish this capability pair as a family.
3092                          */
3093                         if (ld_cap_add_family(ofl, cpp1->c_nsdp, osdp,
3094                             cpp1->c_group, &ifl->ifl_caps->ca_syms) == S_ERROR)
3095                                 return (S_ERROR);
3096                 }
3097         }
3098 
3099         return (1);
3100 
3101 #undef SYM_LOC_BADADDR
3102 }
3103 
3104 /*
3105  * Add an undefined symbol to the symbol table.  The reference originates from
3106  * the location identified by the message id (mid).  These references can
3107  * originate from command line options such as -e, -u, -initarray, etc.
3108  * (identified with MSG_INTL(MSG_STR_COMMAND)), or from internally generated
3109  * TLS relocation references (identified with MSG_INTL(MSG_STR_TLSREL)).
3110  */
3111 Sym_desc *
3112 ld_sym_add_u(const char *name, Ofl_desc *ofl, Msg mid)
3113 {
3114         Sym             *sym;
3115         Ifl_desc        *ifl = NULL, *_ifl;
3116         Sym_desc        *sdp;
3117         Word            hash;
3118         Aliste          idx;
3119         avl_index_t     where;
3120         const char      *reference = MSG_INTL(mid);
3121 
3122         /*
3123          * As an optimization, determine whether we've already generated this
3124          * reference.  If the symbol doesn't already exist we'll create it.
3125          * Or if the symbol does exist from a different source, we'll resolve
3126          * the conflict.
3127          */
3128         /* LINTED */
3129         hash = (Word)elf_hash(name);
3130         if ((sdp = ld_sym_find(name, hash, &where, ofl)) != NULL) {
3131                 if ((sdp->sd_sym->st_shndx == SHN_UNDEF) &&
3132                     (sdp->sd_file->ifl_name == reference))
3133                         return (sdp);
3134         }
3135 
3136         /*
3137          * Determine whether a pseudo input file descriptor exists to represent
3138          * the command line, as any global symbol needs an input file descriptor
3139          * during any symbol resolution (refer to map_ifl() which provides a
3140          * similar method for adding symbols from mapfiles).
3141          */
3142         for (APLIST_TRAVERSE(ofl->ofl_objs, idx, _ifl))
3143                 if (strcmp(_ifl->ifl_name, reference) == 0) {
3144                         ifl = _ifl;
3145                         break;
3146                 }
3147 
3148         /*
3149          * If no descriptor exists create one.
3150          */
3151         if (ifl == NULL) {
3152                 if ((ifl = libld_calloc(sizeof (Ifl_desc), 1)) == NULL)
3153                         return ((Sym_desc *)S_ERROR);
3154                 ifl->ifl_name = reference;
3155                 ifl->ifl_flags = FLG_IF_NEEDED | FLG_IF_FILEREF;
3156                 if ((ifl->ifl_ehdr = libld_calloc(sizeof (Ehdr), 1)) == NULL)
3157                         return ((Sym_desc *)S_ERROR);
3158                 ifl->ifl_ehdr->e_type = ET_REL;
3159 
3160                 if (aplist_append(&ofl->ofl_objs, ifl, AL_CNT_OFL_OBJS) == NULL)
3161                         return ((Sym_desc *)S_ERROR);
3162         }
3163 
3164         /*
3165          * Allocate a symbol structure and add it to the global symbol table.
3166          */
3167         if ((sym = libld_calloc(sizeof (Sym), 1)) == NULL)
3168                 return ((Sym_desc *)S_ERROR);
3169         sym->st_info = ELF_ST_INFO(STB_GLOBAL, STT_NOTYPE);
3170         sym->st_shndx = SHN_UNDEF;
3171 
3172         DBG_CALL(Dbg_syms_process(ofl->ofl_lml, ifl));
3173         if (sdp == NULL) {
3174                 DBG_CALL(Dbg_syms_global(ofl->ofl_lml, 0, name));
3175                 if ((sdp = ld_sym_enter(name, sym, hash, ifl, ofl, 0, SHN_UNDEF,
3176                     0, &where)) == (Sym_desc *)S_ERROR)
3177                         return ((Sym_desc *)S_ERROR);
3178         } else if (ld_sym_resolve(sdp, sym, ifl, ofl, 0,
3179             SHN_UNDEF, 0) == S_ERROR)
3180                 return ((Sym_desc *)S_ERROR);
3181 
3182         sdp->sd_flags &= ~FLG_SY_CLEAN;
3183         sdp->sd_flags |= FLG_SY_CMDREF;
3184 
3185         return (sdp);
3186 }
3187 
3188 /*
3189  * STT_SECTION symbols have their st_name field set to NULL, and consequently
3190  * have no name. Generate a name suitable for diagnostic use for such a symbol
3191  * and store it in the input section descriptor. The resulting name will be
3192  * of the form:
3193  *
3194  *      "XXX (section)"
3195  *
3196  * where XXX is the name of the section.
3197  *
3198  * entry:
3199  *      isc - Input section associated with the symbol.
3200  *      fmt - NULL, or format string to use.
3201  *
3202  * exit:
3203  *      Sets isp->is_sym_name to the allocated string. Returns the
3204  *      string pointer, or NULL on allocation failure.
3205  */
3206 const char *
3207 ld_stt_section_sym_name(Is_desc *isp)
3208 {
3209         const char      *fmt;
3210         char            *str;
3211         size_t          len;
3212 
3213         if ((isp == NULL) || (isp->is_name == NULL))
3214                 return (NULL);
3215 
3216         if (isp->is_sym_name == NULL) {
3217                 fmt = (isp->is_flags & FLG_IS_GNSTRMRG) ?
3218                     MSG_INTL(MSG_STR_SECTION_MSTR) : MSG_INTL(MSG_STR_SECTION);
3219 
3220                 len = strlen(fmt) + strlen(isp->is_name) + 1;
3221 
3222                 if ((str = libld_malloc(len)) == NULL)
3223                         return (NULL);
3224                 (void) snprintf(str, len, fmt, isp->is_name);
3225                 isp->is_sym_name = str;
3226         }
3227 
3228         return (isp->is_sym_name);
3229 }