sysdeps/generic/dl-tls.c

   1 /* Thread-local storage handling in the ELF dynamic linker.  Generic version.
   2    Copyright (C) 2002 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   4
   5    The GNU C Library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 2.1 of the License, or (at your option) any later version.
   9
  10    The GNU C Library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with the GNU C Library; if not, write to the Free
  17    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  18    02111-1307 USA.  */
  19
  20 #include <assert.h>
  21 #include <signal.h>
  22 #include <stdlib.h>
  23 #include <unistd.h>
  24 #include <sys/param.h>
  25
  26 #include <abort-instr.h>
  27 #include <tls.h>
  28
  29 /* We don't need any of this if TLS is not supported.  */
  30 #ifdef USE_TLS
  31
  32 # include <dl-tls.h>
  33 # include <ldsodefs.h>
  34
  35 /* Value used for dtv entries for which the allocation is delayed.  */
  36 # define TLS_DTV_UNALLOCATED    ((void *) -1l)
  37
  38
  39 /* Out-of-memory handler.  */
  40 # ifdef SHARED
  41 static void
  42 __attribute__ ((__noreturn__))
  43 oom (void)
  44 {
  45   static const char msg[] = "\
  46 cannot allocate memory for thread-local data: ABORT\n";
  47
  48   __libc_write (STDERR_FILENO, msg, sizeof (msg) - 1);
  49
  50   /* Kill ourself.  */
  51   __kill (__getpid (), SIGKILL);
  52
  53   /* Just in case something goes wrong with the kill.  */
  54   while (1)
  55     {
  56 #  ifdef ABORT_INSTRUCTION
  57       ABORT_INSTRUCTION;
  58 #  endif
  59     }
  60 }
  61 # endif
  62
  63
  64
  65 size_t
  66 internal_function
  67 _dl_next_tls_modid (void)
  68 {
  69   size_t result;
  70
  71   if (__builtin_expect (GL(dl_tls_dtv_gaps), false))
  72     {
  73       size_t disp = 0;
  74       struct dtv_slotinfo_list *runp = GL(dl_tls_dtv_slotinfo_list);
  75
  76       /* Note that this branch will never be executed during program
  77          start since there are no gaps at that time.  Therefore it
  78          does not matter that the dl_tls_dtv_slotinfo is not allocated
  79          yet when the function is called for the first times.  */
  80       result = GL(dl_tls_static_nelem) + 1;
  81       /* If the following would not be true we mustn't have assumed
  82          there is a gap.  */
  83       assert (result <= GL(dl_tls_max_dtv_idx));
  84       do
  85         {
  86           while (result - disp < runp->len)
  87             {
  88               if (runp->slotinfo[result - disp].map == NULL)
  89                 break;
  90
  91               ++result;
  92               assert (result <= GL(dl_tls_max_dtv_idx) + 1);
  93             }
  94
  95           if (result - disp < runp->len)
  96             break;
  97
  98           disp += runp->len;
  99         }
 100       while ((runp = runp->next) != NULL);
 101
 102       if (result >= GL(dl_tls_max_dtv_idx))
 103         {
 104           /* The new index must indeed be exactly one higher than the
 105              previous high.  */
 106           assert (result == GL(dl_tls_max_dtv_idx));
 107
 108           /* There is no gap anymore.  */
 109           GL(dl_tls_dtv_gaps) = false;
 110
 111           goto nogaps;
 112         }
 113     }
 114   else
 115     {
 116       /* No gaps, allocate a new entry.  */
 117     nogaps:
 118       result = ++GL(dl_tls_max_dtv_idx);
 119     }
 120
 121   return result;
 122 }
 123
 124
 125 void
 126 internal_function
 127 _dl_determine_tlsoffset (void)
 128 {
 129   struct dtv_slotinfo *slotinfo;
 130   size_t max_align = __alignof__ (void *);
 131   size_t offset;
 132   size_t cnt;
 133
 134   /* The first element of the dtv slot info list is allocated.  */
 135   assert (GL(dl_tls_dtv_slotinfo_list) != NULL);
 136   /* There is at this point only one element in the
 137      dl_tls_dtv_slotinfo_list list.  */
 138   assert (GL(dl_tls_dtv_slotinfo_list)->next == NULL);
 139
 140 # if TLS_TCB_AT_TP
 141   /* We simply start with zero.  */
 142   offset = 0;
 143
 144   slotinfo = GL(dl_tls_dtv_slotinfo_list)->slotinfo;
 145   for (cnt = 1; slotinfo[cnt].map != NULL; ++cnt)
 146     {
 147       assert (cnt < GL(dl_tls_dtv_slotinfo_list)->len);
 148
 149       max_align = MAX (max_align, slotinfo[cnt].map->l_tls_align);
 150
 151       /* Compute the offset of the next TLS block.  */
 152       offset = roundup (offset + slotinfo[cnt].map->l_tls_blocksize,
 153                         slotinfo[cnt].map->l_tls_align);
 154
 155       /* XXX For some architectures we perhaps should store the
 156          negative offset.  */
 157       slotinfo[cnt].map->l_tls_offset = offset;
 158     }
 159
 160   /* The thread descriptor (pointed to by the thread pointer) has its
 161      own alignment requirement.  Adjust the static TLS size
 162      and TLS offsets appropriately.  */
 163   // XXX How to deal with this.  We cannot simply add zero bytes
 164   // XXX after the first (closest to the TCB) TLS block since this
 165   // XXX would invalidate the offsets the linker creates for the LE
 166   // XXX model.
 167
 168   GL(dl_tls_static_size) = offset + TLS_TCB_SIZE;
 169 # elif TLS_DTV_AT_TP
 170   /* The TLS blocks start right after the TCB.  */
 171   offset = TLS_TCB_SIZE;
 172
 173   /* The first block starts right after the TCB.  */
 174   slotinfo = GL(dl_tls_dtv_slotinfo_list)->slotinfo;
 175   if (slotinfo[1].map != NULL)
 176     {
 177       size_t prev_size
 178
 179       offset = roundup (offset, slotinfo[1].map->l_tls_align);
 180       slotinfo[1].map->l_tls_offset = offset;
 181       max_align = slotinfo[1].map->l_tls_align;
 182       prev_size = slotinfo[1].map->l_tls_blocksize;
 183
 184       for (cnt = 2; slotinfo[cnt].map != NULL; ++cnt)
 185         {
 186           assert (cnt < GL(dl_tls_dtv_slotinfo_list)->len);
 187
 188           max_align = MAX (max_align, slotinfo[cnt].map->l_tls_align);
 189
 190           /* Compute the offset of the next TLS block.  */
 191           offset = roundup (offset + prev_size,
 192                             slotinfo[cnt].map->l_tls_align);
 193
 194           /* XXX For some architectures we perhaps should store the
 195              negative offset.  */
 196           slotinfo[cnt].map->l_tls_offset = offset;
 197
 198           prev_size = slotinfo[cnt].map->l_tls_blocksize;
 199         }
 200
 201       offset += prev_size;
 202     }
 203
 204   GL(dl_tls_static_size) = offset;
 205 # else
 206 #  error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined"
 207 # endif
 208
 209   /* The alignment requirement for the static TLS block.  */
 210   GL(dl_tls_static_align) = MAX (TLS_TCB_ALIGN, max_align);
 211 }
 212
 213
 214 void *
 215 internal_function
 216 _dl_allocate_tls (void)
 217 {
 218   void *result;
 219   dtv_t *dtv;
 220   size_t dtv_length;
 221
 222   /* Allocate a correctly aligned chunk of memory.  */
 223   /* XXX For now */
 224   assert (GL(dl_tls_static_align) <= GL(dl_pagesize));
 225 # ifdef MAP_ANON
 226 #  define _dl_zerofd (-1)
 227 # else
 228 #  define _dl_zerofd GL(dl_zerofd)
 229   if ((dl_zerofd) == -1)
 230     GL(dl_zerofd) = _dl_sysdep_open_zero_fill ();
 231 #  define MAP_ANON 0
 232 # endif
 233   result = __mmap (0, GL(dl_tls_static_size), PROT_READ|PROT_WRITE,
 234                    MAP_ANON|MAP_PRIVATE, _dl_zerofd, 0);
 235
 236   /* We allocate a few more elements in the dtv than are needed for the
 237      initial set of modules.  This should avoid in most cases expansions
 238      of the dtv.  */
 239   dtv_length = GL(dl_tls_max_dtv_idx) + DTV_SURPLUS;
 240   dtv = (dtv_t *) malloc ((dtv_length + 2) * sizeof (dtv_t));
 241   if (result != MAP_FAILED && dtv != NULL)
 242     {
 243       struct dtv_slotinfo_list *listp;
 244       bool first_block = true;
 245       size_t total = 0;
 246
 247 # if TLS_TCB_AT_TP
 248       /* The TCB follows the TLS blocks.  */
 249       result = (char *) result + GL(dl_tls_static_size) - TLS_TCB_SIZE;
 250 # endif
 251
 252       /* This is the initial length of the dtv.  */
 253       dtv[0].counter = dtv_length;
 254       /* Fill in the generation number.  */
 255       dtv[1].counter = GL(dl_tls_generation) = 0;
 256       /* Initialize all of the rest of the dtv with zero to indicate
 257          nothing there.  */
 258       memset (dtv + 2, '\0', dtv_length * sizeof (dtv_t));
 259
 260       /* We have to look prepare the dtv for all currently loaded
 261          modules using TLS.  For those which are dynamically loaded we
 262          add the values indicating deferred allocation.  */
 263       listp = GL(dl_tls_dtv_slotinfo_list);
 264       while (1)
 265         {
 266           size_t cnt;
 267
 268           for (cnt = first_block ? 1 : 0; cnt < listp->len; ++cnt)
 269             {
 270               struct link_map *map;
 271               void *dest;
 272
 273               /* Check for the total number of used slots.  */
 274               if (total + cnt >= GL(dl_tls_max_dtv_idx))
 275                 break;
 276
 277               map = listp->slotinfo[cnt].map;
 278               if (map == NULL)
 279                 /* Unused entry.  */
 280                 continue;
 281
 282               if (map->l_type == lt_loaded)
 283                 {
 284                   /* For dynamically loaded modules we simply store
 285                      the value indicating deferred allocation.  */
 286                   dtv[1 + map->l_tls_modid].pointer = TLS_DTV_UNALLOCATED;
 287                   continue;
 288                 }
 289
 290               assert (map->l_tls_modid == cnt);
 291               assert (map->l_tls_blocksize >= map->l_tls_initimage_size);
 292 # if TLS_TCB_AT_TP
 293               assert (map->l_tls_offset >= map->l_tls_blocksize);
 294               dest = (char *) result - map->l_tls_offset;
 295 # elif TLS_DTV_AT_TP
 296               dest = (char *) result + map->l_tls_offset;
 297 # else
 298 #  error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined"
 299 # endif
 300
 301               /* We don't have to clear the BSS part of the TLS block
 302                  since mmap is used to allocate the memory which
 303                  guarantees it is initialized to zero.  */
 304               dtv[1 + cnt].pointer = memcpy (dest, map->l_tls_initimage,
 305                                              map->l_tls_initimage_size);
 306             }
 307
 308           total += cnt;
 309           if (total >= GL(dl_tls_max_dtv_idx))
 310             break;
 311
 312           listp = listp->next;
 313           assert (listp != NULL);
 314         }
 315
 316       /* Add the dtv to the thread data structures.  */
 317       INSTALL_DTV (result, dtv);
 318     }
 319   else if (result != NULL)
 320     {
 321       free (result);
 322       result = NULL;
 323     }
 324
 325   return result;
 326 }
 327 INTDEF(_dl_allocate_tls)
 328
 329
 330 void
 331 internal_function
 332 _dl_deallocate_tls (void *tcb)
 333 {
 334   dtv_t *dtv = GET_DTV (tcb);
 335
 336   /* The array starts with dtv[-1].  */
 337   free (dtv - 1);
 338
 339   munmap (tcb, GL(dl_tls_static_size));
 340 }
 341
 342
 343
 344 # ifdef SHARED
 345 /* The __tls_get_addr function has two basic forms which differ in the
 346    arguments.  The IA-64 form takes two parameters, the module ID and
 347    offset.  The form used, among others, on IA-32 takes a reference to
 348    a special structure which contain the same information.  The second
 349    form seems to be more often used (in the moment) so we default to
 350    it.  Users of the IA-64 form have to provide adequate definitions
 351    of the following macros.  */
 352 #  ifndef GET_ADDR_ARGS
 353 #   define GET_ADDR_ARGS tls_index *ti
 354 #  endif
 355 #  ifndef GET_ADDR_MODULE
 356 #   define GET_ADDR_MODULE ti->ti_module
 357 #  endif
 358 #  ifndef GET_ADDR_OFFSET
 359 #   define GET_ADDR_OFFSET ti->ti_offset
 360 #  endif
 361 /* Systems which do not have tls_index also probably have to define
 362    DONT_USE_TLS_INDEX.  */
 363
 364 #  ifndef __TLS_GET_ADDR
 365 #   define __TLS_GET_ADDR __tls_get_addr
 366 #  endif
 367
 368
 369 /* Return the symbol address given the map of the module it is in and
 370    the symbol record.  This is used in dl-sym.c.  */
 371 void *
 372 internal_function
 373 _dl_tls_symaddr (struct link_map *map, const ElfW(Sym) *ref)
 374 {
 375 #  ifndef DONT_USE_TLS_INDEX
 376   tls_index tmp =
 377     {
 378       .ti_module = map->l_tls_modid,
 379       .ti_offset = ref->st_value
 380     };
 381
 382   return __TLS_GET_ADDR (&tmp);
 383 #  else
 384   return __TLS_GET_ADDR (map->l_tls_modid, ref->st_value);
 385 #  endif
 386 }
 387
 388
 389 static void *
 390 allocate_and_init (struct link_map *map)
 391 {
 392   void *newp;
 393
 394   newp = __libc_memalign (map->l_tls_align, map->l_tls_blocksize);
 395   if (newp == NULL)
 396     oom ();
 397
 398   /* Initialize the memory.  */
 399   memset (__mempcpy (newp, map->l_tls_initimage, map->l_tls_initimage_size),
 400           '\0', map->l_tls_blocksize - map->l_tls_initimage_size);
 401
 402   return newp;
 403 }
 404
 405
 406 /* The generic dynamic and local dynamic model cannot be used in
 407    statically linked applications.  */
 408 void *
 409 __tls_get_addr (GET_ADDR_ARGS)
 410 {
 411   dtv_t *dtv = THREAD_DTV ();
 412   struct link_map *the_map = NULL;
 413   void *p;
 414
 415   if (__builtin_expect (dtv[0].counter != GL(dl_tls_generation), 0))
 416     {
 417       struct dtv_slotinfo_list *listp;
 418       size_t idx;
 419
 420       /* The global dl_tls_dtv_slotinfo array contains for each module
 421          index the generation counter current when the entry was
 422          created.  This array never shrinks so that all module indices
 423          which were valid at some time can be used to access it.
 424          Before the first use of a new module index in this function
 425          the array was extended appropriately.  Access also does not
 426          have to be guarded against modifications of the array.  It is
 427          assumed that pointer-size values can be read atomically even
 428          in SMP environments.  It is possible that other threads at
 429          the same time dynamically load code and therefore add to the
 430          slotinfo list.  This is a problem since we must not pick up
 431          any information about incomplete work.  The solution to this
 432          is to ignore all dtv slots which were created after the one
 433          we are currently interested.  We know that dynamic loading
 434          for this module is completed and this is the last load
 435          operation we know finished.  */
 436       idx = GET_ADDR_MODULE;
 437       listp = GL(dl_tls_dtv_slotinfo_list);
 438       while (idx >= listp->len)
 439         {
 440           idx -= listp->len;
 441           listp = listp->next;
 442         }
 443
 444       if (dtv[0].counter < listp->slotinfo[idx].gen)
 445         {
 446           /* The generation counter for the slot is higher than what
 447              the current dtv implements.  We have to update the whole
 448              dtv but only those entries with a generation counter <=
 449              the one for the entry we need.  */
 450           size_t new_gen = listp->slotinfo[idx].gen;
 451           size_t total = 0;
 452
 453           /* We have to look through the entire dtv slotinfo list.  */
 454           listp =  GL(dl_tls_dtv_slotinfo_list);
 455           do
 456             {
 457               size_t cnt;
 458
 459               for (cnt = total = 0 ? 1 : 0; cnt < listp->len; ++cnt)
 460                 {
 461                   size_t gen = listp->slotinfo[cnt].gen;
 462                   struct link_map *map;
 463                   size_t modid;
 464
 465                   if (gen > new_gen)
 466                     /* This is a slot for a generation younger than
 467                        the one we are handling now.  It might be
 468                        incompletely set up so ignore it.  */
 469                     continue;
 470
 471                   /* If the entry is older than the current dtv layout
 472                      we know we don't have to handle it.  */
 473                   if (gen <= dtv[0].counter)
 474                     continue;
 475
 476                   /* If there is no map this means the entry is empty.  */
 477                   map = listp->slotinfo[cnt].map;
 478                   if (map == NULL)
 479                     {
 480                       /* If this modid was used at some point the memory
 481                          might still be allocated.  */
 482                       if (dtv[total + cnt].pointer != TLS_DTV_UNALLOCATED)
 483                         free (dtv[total + cnt].pointer);
 484
 485                       continue;
 486                     }
 487
 488                   /* Check whether the current dtv array is large enough.  */
 489                   modid = map->l_tls_modid;
 490                   assert (total + cnt == modid);
 491                   if (dtv[-1].counter < modid)
 492                     {
 493                       /* Reallocate the dtv.  */
 494                       dtv_t *newp;
 495                       size_t newsize = GL(dl_tls_max_dtv_idx) + DTV_SURPLUS;
 496                       size_t oldsize = dtv[-1].counter;
 497
 498                       assert (map->l_tls_modid <= newsize);
 499
 500                       newp = (dtv_t *) realloc (&dtv[-1],
 501                                                 (2 + newsize)
 502                                                 * sizeof (dtv_t));
 503                       if (newp == NULL)
 504                         oom ();
 505
 506                       newp[0].counter = newsize;
 507
 508                       /* Clear the newly allocate part.  */
 509                       memset (newp + 2 + oldsize, '\0',
 510                               (newsize - oldsize) * sizeof (dtv_t));
 511
 512                       /* Point dtv to the generation counter.  */
 513                       dtv = &newp[1];
 514
 515                       /* Install this new dtv in the thread data
 516                          structures.  */
 517                       INSTALL_NEW_DTV (dtv);
 518                     }
 519
 520                   /* If there is currently memory allocate for this
 521                      dtv entry free it.  */
 522                   /* XXX Ideally we will at some point create a memory
 523                      pool.  */
 524                   if (dtv[modid].pointer != TLS_DTV_UNALLOCATED)
 525                     /* Note that free is called for NULL is well.  We
 526                        deallocate even if it is this dtv entry we are
 527                        supposed to load.  The reason is that we call
 528                        memalign and not malloc.  */
 529                     free (dtv[modid].pointer);
 530
 531                   /* This module is loaded dynamically- We defer
 532                      memory allocation.  */
 533                   dtv[modid].pointer = TLS_DTV_UNALLOCATED;
 534
 535                   if (modid == GET_ADDR_MODULE)
 536                     the_map = map;
 537                 }
 538
 539               total += listp->len;
 540             }
 541           while ((listp = listp->next) != NULL);
 542
 543           /* This will be the new maximum generation counter.  */
 544           dtv[0].counter = new_gen;
 545         }
 546     }
 547
 548   p = dtv[GET_ADDR_MODULE].pointer;
 549
 550   if (__builtin_expect (p == TLS_DTV_UNALLOCATED, 0))
 551     {
 552       /* The allocation was deferred.  Do it now.  */
 553       if (the_map == NULL)
 554         {
 555           /* Find the link map for this module.  */
 556           size_t idx = GET_ADDR_MODULE;
 557           struct dtv_slotinfo_list *listp = GL(dl_tls_dtv_slotinfo_list);
 558
 559           while (idx >= listp->len)
 560             {
 561               idx -= listp->len;
 562               listp = listp->next;
 563             }
 564
 565           the_map = listp->slotinfo[idx].map;
 566         }
 567
 568       p = dtv[GET_ADDR_MODULE].pointer = allocate_and_init (the_map);
 569     }
 570
 571   return (char *) p + GET_ADDR_OFFSET;
 572 }
 573 # endif
 574
 575 #endif  /* use TLS */