nptl/pthread_create.c

   1 /* Copyright (C) 2002-2007,2008,2009,2010 Free Software Foundation, Inc.
   2    This file is part of the GNU C Library.
   3    Contributed by Ulrich Drepper <drepper@redhat.com>, 2002.
   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 <errno.h>
  21 #include <stdbool.h>
  22 #include <stdlib.h>
  23 #include <string.h>
  24 #include "pthreadP.h"
  25 #include <hp-timing.h>
  26 #include <ldsodefs.h>
  27 #include <atomic.h>
  28 #include <libc-internal.h>
  29 #include <resolv.h>
  30 #include <kernel-features.h>
  31
  32 #include <shlib-compat.h>
  33
  34
  35 /* Local function to start thread and handle cleanup.  */
  36 static int start_thread (void *arg);
  37
  38
  39 /* Nozero if debugging mode is enabled.  */
  40 int __pthread_debug;
  41
  42 /* Globally enabled events.  */
  43 static td_thr_events_t __nptl_threads_events __attribute_used__;
  44
  45 /* Pointer to descriptor with the last event.  */
  46 static struct pthread *__nptl_last_event __attribute_used__;
  47
  48 /* Number of threads running.  */
  49 unsigned int __nptl_nthreads = 1;
  50
  51
  52 /* Code to allocate and deallocate a stack.  */
  53 #include "allocatestack.c"
  54
  55 /* Code to create the thread.  */
  56 #include <createthread.c>
  57
  58
  59 struct pthread *
  60 internal_function
  61 __find_in_stack_list (pd)
  62      struct pthread *pd;
  63 {
  64   list_t *entry;
  65   struct pthread *result = NULL;
  66
  67   lll_lock (stack_cache_lock, LLL_PRIVATE);
  68
  69   list_for_each (entry, &stack_used)
  70     {
  71       struct pthread *curp;
  72
  73       curp = list_entry (entry, struct pthread, list);
  74       if (curp == pd)
  75         {
  76           result = curp;
  77           break;
  78         }
  79     }
  80
  81   if (result == NULL)
  82     list_for_each (entry, &__stack_user)
  83       {
  84         struct pthread *curp;
  85
  86         curp = list_entry (entry, struct pthread, list);
  87         if (curp == pd)
  88           {
  89             result = curp;
  90             break;
  91           }
  92       }
  93
  94   lll_unlock (stack_cache_lock, LLL_PRIVATE);
  95
  96   return result;
  97 }
  98
  99
 100 /* Deallocate POSIX thread-local-storage.  */
 101 void
 102 attribute_hidden
 103 __nptl_deallocate_tsd (void)
 104 {
 105   struct pthread *self = THREAD_SELF;
 106
 107   /* Maybe no data was ever allocated.  This happens often so we have
 108      a flag for this.  */
 109   if (THREAD_GETMEM (self, specific_used))
 110     {
 111       size_t round;
 112       size_t cnt;
 113
 114       round = 0;
 115       do
 116         {
 117           size_t idx;
 118
 119           /* So far no new nonzero data entry.  */
 120           THREAD_SETMEM (self, specific_used, false);
 121
 122           for (cnt = idx = 0; cnt < PTHREAD_KEY_1STLEVEL_SIZE; ++cnt)
 123             {
 124               struct pthread_key_data *level2;
 125
 126               level2 = THREAD_GETMEM_NC (self, specific, cnt);
 127
 128               if (level2 != NULL)
 129                 {
 130                   size_t inner;
 131
 132                   for (inner = 0; inner < PTHREAD_KEY_2NDLEVEL_SIZE;
 133                        ++inner, ++idx)
 134                     {
 135                       void *data = level2[inner].data;
 136
 137                       if (data != NULL)
 138                         {
 139                           /* Always clear the data.  */
 140                           level2[inner].data = NULL;
 141
 142                           /* Make sure the data corresponds to a valid
 143                              key.  This test fails if the key was
 144                              deallocated and also if it was
 145                              re-allocated.  It is the user's
 146                              responsibility to free the memory in this
 147                              case.  */
 148                           if (level2[inner].seq
 149                               == __pthread_keys[idx].seq
 150                               /* It is not necessary to register a destructor
 151                                  function.  */
 152                               && __pthread_keys[idx].destr != NULL)
 153                             /* Call the user-provided destructor.  */
 154                             __pthread_keys[idx].destr (data);
 155                         }
 156                     }
 157                 }
 158               else
 159                 idx += PTHREAD_KEY_1STLEVEL_SIZE;
 160             }
 161
 162           if (THREAD_GETMEM (self, specific_used) == 0)
 163             /* No data has been modified.  */
 164             goto just_free;
 165         }
 166       /* We only repeat the process a fixed number of times.  */
 167       while (__builtin_expect (++round < PTHREAD_DESTRUCTOR_ITERATIONS, 0));
 168
 169       /* Just clear the memory of the first block for reuse.  */
 170       memset (&THREAD_SELF->specific_1stblock, '\0',
 171               sizeof (self->specific_1stblock));
 172
 173     just_free:
 174       /* Free the memory for the other blocks.  */
 175       for (cnt = 1; cnt < PTHREAD_KEY_1STLEVEL_SIZE; ++cnt)
 176         {
 177           struct pthread_key_data *level2;
 178
 179           level2 = THREAD_GETMEM_NC (self, specific, cnt);
 180           if (level2 != NULL)
 181             {
 182               /* The first block is allocated as part of the thread
 183                  descriptor.  */
 184               free (level2);
 185               THREAD_SETMEM_NC (self, specific, cnt, NULL);
 186             }
 187         }
 188
 189       THREAD_SETMEM (self, specific_used, false);
 190     }
 191 }
 192
 193
 194 /* Deallocate a thread's stack after optionally making sure the thread
 195    descriptor is still valid.  */
 196 void
 197 internal_function
 198 __free_tcb (struct pthread *pd)
 199 {
 200   /* The thread is exiting now.  */
 201   if (__builtin_expect (atomic_bit_test_set (&pd->cancelhandling,
 202                                              TERMINATED_BIT) == 0, 1))
 203     {
 204       /* Remove the descriptor from the list.  */
 205       if (DEBUGGING_P && __find_in_stack_list (pd) == NULL)
 206         /* Something is really wrong.  The descriptor for a still
 207            running thread is gone.  */
 208         abort ();
 209
 210       /* Free TPP data.  */
 211       if (__builtin_expect (pd->tpp != NULL, 0))
 212         {
 213           struct priority_protection_data *tpp = pd->tpp;
 214
 215           pd->tpp = NULL;
 216           free (tpp);
 217         }
 218
 219       /* Queue the stack memory block for reuse and exit the process.  The
 220          kernel will signal via writing to the address returned by
 221          QUEUE-STACK when the stack is available.  */
 222       __deallocate_stack (pd);
 223     }
 224 }
 225
 226
 227 static int
 228 start_thread (void *arg)
 229 {
 230   struct pthread *pd = (struct pthread *) arg;
 231
 232 #if HP_TIMING_AVAIL
 233   /* Remember the time when the thread was started.  */
 234   hp_timing_t now;
 235   HP_TIMING_NOW (now);
 236   THREAD_SETMEM (pd, cpuclock_offset, now);
 237 #endif
 238
 239   /* Initialize resolver state pointer.  */
 240   __resp = &pd->res;
 241
 242   /* Allow setxid from now onwards.  */
 243   if (__builtin_expect (atomic_exchange_acq (&pd->setxid_futex, 0) == -2, 0))
 244     lll_futex_wake (&pd->setxid_futex, 1, LLL_PRIVATE);
 245
 246 #ifdef __NR_set_robust_list
 247 # ifndef __ASSUME_SET_ROBUST_LIST
 248   if (__set_robust_list_avail >= 0)
 249 # endif
 250     {
 251       INTERNAL_SYSCALL_DECL (err);
 252       /* This call should never fail because the initial call in init.c
 253          succeeded.  */
 254       INTERNAL_SYSCALL (set_robust_list, err, 2, &pd->robust_head,
 255                         sizeof (struct robust_list_head));
 256     }
 257 #endif
 258
 259   /* If the parent was running cancellation handlers while creating
 260      the thread the new thread inherited the signal mask.  Reset the
 261      cancellation signal mask.  */
 262   if (__builtin_expect (pd->parent_cancelhandling & CANCELING_BITMASK, 0))
 263     {
 264       INTERNAL_SYSCALL_DECL (err);
 265       sigset_t mask;
 266       __sigemptyset (&mask);
 267       __sigaddset (&mask, SIGCANCEL);
 268       (void) INTERNAL_SYSCALL (rt_sigprocmask, err, 4, SIG_UNBLOCK, &mask,
 269                                NULL, _NSIG / 8);
 270     }
 271
 272   /* This is where the try/finally block should be created.  For
 273      compilers without that support we do use setjmp.  */
 274   struct pthread_unwind_buf unwind_buf;
 275
 276   /* No previous handlers.  */
 277   unwind_buf.priv.data.prev = NULL;
 278   unwind_buf.priv.data.cleanup = NULL;
 279
 280   int not_first_call;
 281   not_first_call = setjmp ((struct __jmp_buf_tag *) unwind_buf.cancel_jmp_buf);
 282   if (__builtin_expect (! not_first_call, 1))
 283     {
 284       /* Store the new cleanup handler info.  */
 285       THREAD_SETMEM (pd, cleanup_jmp_buf, &unwind_buf);
 286
 287       if (__builtin_expect (pd->stopped_start, 0))
 288         {
 289           int oldtype = CANCEL_ASYNC ();
 290
 291           /* Get the lock the parent locked to force synchronization.  */
 292           lll_lock (pd->lock, LLL_PRIVATE);
 293           /* And give it up right away.  */
 294           lll_unlock (pd->lock, LLL_PRIVATE);
 295
 296           CANCEL_RESET (oldtype);
 297         }
 298
 299       /* Run the code the user provided.  */
 300 #ifdef CALL_THREAD_FCT
 301       THREAD_SETMEM (pd, result, CALL_THREAD_FCT (pd));
 302 #else
 303       THREAD_SETMEM (pd, result, pd->start_routine (pd->arg));
 304 #endif
 305     }
 306
 307   /* Run the destructor for the thread-local data.  */
 308   __nptl_deallocate_tsd ();
 309
 310   /* Clean up any state libc stored in thread-local variables.  */
 311   __libc_thread_freeres ();
 312
 313   /* If this is the last thread we terminate the process now.  We
 314      do not notify the debugger, it might just irritate it if there
 315      is no thread left.  */
 316   if (__builtin_expect (atomic_decrement_and_test (&__nptl_nthreads), 0))
 317     /* This was the last thread.  */
 318     exit (0);
 319
 320   /* Report the death of the thread if this is wanted.  */
 321   if (__builtin_expect (pd->report_events, 0))
 322     {
 323       /* See whether TD_DEATH is in any of the mask.  */
 324       const int idx = __td_eventword (TD_DEATH);
 325       const uint32_t mask = __td_eventmask (TD_DEATH);
 326
 327       if ((mask & (__nptl_threads_events.event_bits[idx]
 328                    | pd->eventbuf.eventmask.event_bits[idx])) != 0)
 329         {
 330           /* Yep, we have to signal the death.  Add the descriptor to
 331              the list but only if it is not already on it.  */
 332           if (pd->nextevent == NULL)
 333             {
 334               pd->eventbuf.eventnum = TD_DEATH;
 335               pd->eventbuf.eventdata = pd;
 336
 337               do
 338                 pd->nextevent = __nptl_last_event;
 339               while (atomic_compare_and_exchange_bool_acq (&__nptl_last_event,
 340                                                            pd, pd->nextevent));
 341             }
 342
 343           /* Now call the function to signal the event.  */
 344           __nptl_death_event ();
 345         }
 346     }
 347
 348   /* The thread is exiting now.  Don't set this bit until after we've hit
 349      the event-reporting breakpoint, so that td_thr_get_info on us while at
 350      the breakpoint reports TD_THR_RUN state rather than TD_THR_ZOMBIE.  */
 351   atomic_bit_set (&pd->cancelhandling, EXITING_BIT);
 352
 353 #ifndef __ASSUME_SET_ROBUST_LIST
 354   /* If this thread has any robust mutexes locked, handle them now.  */
 355 # if __WORDSIZE == 64
 356   void *robust = pd->robust_head.list;
 357 # else
 358   __pthread_slist_t *robust = pd->robust_list.__next;
 359 # endif
 360   /* We let the kernel do the notification if it is able to do so.
 361      If we have to do it here there for sure are no PI mutexes involved
 362      since the kernel support for them is even more recent.  */
 363   if (__set_robust_list_avail < 0
 364       && __builtin_expect (robust != (void *) &pd->robust_head, 0))
 365     {
 366       do
 367         {
 368           struct __pthread_mutex_s *this = (struct __pthread_mutex_s *)
 369             ((char *) robust - offsetof (struct __pthread_mutex_s,
 370                                          __list.__next));
 371           robust = *((void **) robust);
 372
 373 # ifdef __PTHREAD_MUTEX_HAVE_PREV
 374           this->__list.__prev = NULL;
 375 # endif
 376           this->__list.__next = NULL;
 377
 378           lll_robust_dead (this->__lock, /* XYZ */ LLL_SHARED);
 379         }
 380       while (robust != (void *) &pd->robust_head);
 381     }
 382 #endif
 383
 384   /* Mark the memory of the stack as usable to the kernel.  We free
 385      everything except for the space used for the TCB itself.  */
 386   size_t pagesize_m1 = __getpagesize () - 1;
 387 #ifdef _STACK_GROWS_DOWN
 388   char *sp = CURRENT_STACK_FRAME;
 389   size_t freesize = (sp - (char *) pd->stackblock) & ~pagesize_m1;
 390 #else
 391 # error "to do"
 392 #endif
 393   assert (freesize < pd->stackblock_size);
 394   if (freesize > PTHREAD_STACK_MIN)
 395     madvise (pd->stackblock, freesize - PTHREAD_STACK_MIN, MADV_DONTNEED);
 396
 397   /* If the thread is detached free the TCB.  */
 398   if (IS_DETACHED (pd))
 399     /* Free the TCB.  */
 400     __free_tcb (pd);
 401   else if (__builtin_expect (pd->cancelhandling & SETXID_BITMASK, 0))
 402     {
 403       /* Some other thread might call any of the setXid functions and expect
 404          us to reply.  In this case wait until we did that.  */
 405       do
 406         lll_futex_wait (&pd->setxid_futex, 0, LLL_PRIVATE);
 407       while (pd->cancelhandling & SETXID_BITMASK);
 408
 409       /* Reset the value so that the stack can be reused.  */
 410       pd->setxid_futex = 0;
 411     }
 412
 413   /* We cannot call '_exit' here.  '_exit' will terminate the process.
 414
 415      The 'exit' implementation in the kernel will signal when the
 416      process is really dead since 'clone' got passed the CLONE_CLEARTID
 417      flag.  The 'tid' field in the TCB will be set to zero.
 418
 419      The exit code is zero since in case all threads exit by calling
 420      'pthread_exit' the exit status must be 0 (zero).  */
 421   __exit_thread_inline (0);
 422
 423   /* NOTREACHED */
 424   return 0;
 425 }
 426
 427
 428 /* Default thread attributes for the case when the user does not
 429    provide any.  */
 430 static const struct pthread_attr default_attr =
 431   {
 432     /* Just some value > 0 which gets rounded to the nearest page size.  */
 433     .guardsize = 1,
 434   };
 435
 436
 437 int
 438 __pthread_create_2_1 (newthread, attr, start_routine, arg)
 439      pthread_t *newthread;
 440      const pthread_attr_t *attr;
 441      void *(*start_routine) (void *);
 442      void *arg;
 443 {
 444   STACK_VARIABLES;
 445
 446   const struct pthread_attr *iattr = (struct pthread_attr *) attr;
 447   if (iattr == NULL)
 448     /* Is this the best idea?  On NUMA machines this could mean
 449        accessing far-away memory.  */
 450     iattr = &default_attr;
 451
 452   struct pthread *pd = NULL;
 453   int err = ALLOCATE_STACK (iattr, &pd);
 454   if (__builtin_expect (err != 0, 0))
 455     /* Something went wrong.  Maybe a parameter of the attributes is
 456        invalid or we could not allocate memory.  */
 457     return err;
 458
 459
 460   /* Initialize the TCB.  All initializations with zero should be
 461      performed in 'get_cached_stack'.  This way we avoid doing this if
 462      the stack freshly allocated with 'mmap'.  */
 463
 464 #ifdef TLS_TCB_AT_TP
 465   /* Reference to the TCB itself.  */
 466   pd->header.self = pd;
 467
 468   /* Self-reference for TLS.  */
 469   pd->header.tcb = pd;
 470 #endif
 471
 472   /* Store the address of the start routine and the parameter.  Since
 473      we do not start the function directly the stillborn thread will
 474      get the information from its thread descriptor.  */
 475   pd->start_routine = start_routine;
 476   pd->arg = arg;
 477
 478   /* Copy the thread attribute flags.  */
 479   struct pthread *self = THREAD_SELF;
 480   pd->flags = ((iattr->flags & ~(ATTR_FLAG_SCHED_SET | ATTR_FLAG_POLICY_SET))
 481                | (self->flags & (ATTR_FLAG_SCHED_SET | ATTR_FLAG_POLICY_SET)));
 482
 483   /* Initialize the field for the ID of the thread which is waiting
 484      for us.  This is a self-reference in case the thread is created
 485      detached.  */
 486   pd->joinid = iattr->flags & ATTR_FLAG_DETACHSTATE ? pd : NULL;
 487
 488   /* The debug events are inherited from the parent.  */
 489   pd->eventbuf = self->eventbuf;
 490
 491
 492   /* Copy the parent's scheduling parameters.  The flags will say what
 493      is valid and what is not.  */
 494   pd->schedpolicy = self->schedpolicy;
 495   pd->schedparam = self->schedparam;
 496
 497   /* Copy the stack guard canary.  */
 498 #ifdef THREAD_COPY_STACK_GUARD
 499   THREAD_COPY_STACK_GUARD (pd);
 500 #endif
 501
 502   /* Copy the pointer guard value.  */
 503 #ifdef THREAD_COPY_POINTER_GUARD
 504   THREAD_COPY_POINTER_GUARD (pd);
 505 #endif
 506
 507   /* Determine scheduling parameters for the thread.  */
 508   if (attr != NULL
 509       && __builtin_expect ((iattr->flags & ATTR_FLAG_NOTINHERITSCHED) != 0, 0)
 510       && (iattr->flags & (ATTR_FLAG_SCHED_SET | ATTR_FLAG_POLICY_SET)) != 0)
 511     {
 512       INTERNAL_SYSCALL_DECL (scerr);
 513
 514       /* Use the scheduling parameters the user provided.  */
 515       if (iattr->flags & ATTR_FLAG_POLICY_SET)
 516         pd->schedpolicy = iattr->schedpolicy;
 517       else if ((pd->flags & ATTR_FLAG_POLICY_SET) == 0)
 518         {
 519           pd->schedpolicy = INTERNAL_SYSCALL (sched_getscheduler, scerr, 1, 0);
 520           pd->flags |= ATTR_FLAG_POLICY_SET;
 521         }
 522
 523       if (iattr->flags & ATTR_FLAG_SCHED_SET)
 524         memcpy (&pd->schedparam, &iattr->schedparam,
 525                 sizeof (struct sched_param));
 526       else if ((pd->flags & ATTR_FLAG_SCHED_SET) == 0)
 527         {
 528           INTERNAL_SYSCALL (sched_getparam, scerr, 2, 0, &pd->schedparam);
 529           pd->flags |= ATTR_FLAG_SCHED_SET;
 530         }
 531
 532       /* Check for valid priorities.  */
 533       int minprio = INTERNAL_SYSCALL (sched_get_priority_min, scerr, 1,
 534                                       iattr->schedpolicy);
 535       int maxprio = INTERNAL_SYSCALL (sched_get_priority_max, scerr, 1,
 536                                       iattr->schedpolicy);
 537       if (pd->schedparam.sched_priority < minprio
 538           || pd->schedparam.sched_priority > maxprio)
 539         {
 540           /* Perhaps a thread wants to change the IDs and if waiting
 541              for this stillborn thread.  */
 542           if (__builtin_expect (atomic_exchange_acq (&pd->setxid_futex, 0)
 543                                 == -2, 0))
 544             lll_futex_wake (&pd->setxid_futex, 1, LLL_PRIVATE);
 545
 546           __deallocate_stack (pd);
 547
 548           return EINVAL;
 549         }
 550     }
 551
 552   /* Pass the descriptor to the caller.  */
 553   *newthread = (pthread_t) pd;
 554
 555   /* Start the thread.  */
 556   return create_thread (pd, iattr, STACK_VARIABLES_ARGS);
 557 }
 558 versioned_symbol (libpthread, __pthread_create_2_1, pthread_create, GLIBC_2_1);
 559
 560
 561 #if SHLIB_COMPAT(libpthread, GLIBC_2_0, GLIBC_2_1)
 562 int
 563 __pthread_create_2_0 (newthread, attr, start_routine, arg)
 564      pthread_t *newthread;
 565      const pthread_attr_t *attr;
 566      void *(*start_routine) (void *);
 567      void *arg;
 568 {
 569   /* The ATTR attribute is not really of type `pthread_attr_t *'.  It has
 570      the old size and access to the new members might crash the program.
 571      We convert the struct now.  */
 572   struct pthread_attr new_attr;
 573
 574   if (attr != NULL)
 575     {
 576       struct pthread_attr *iattr = (struct pthread_attr *) attr;
 577       size_t ps = __getpagesize ();
 578
 579       /* Copy values from the user-provided attributes.  */
 580       new_attr.schedparam = iattr->schedparam;
 581       new_attr.schedpolicy = iattr->schedpolicy;
 582       new_attr.flags = iattr->flags;
 583
 584       /* Fill in default values for the fields not present in the old
 585          implementation.  */
 586       new_attr.guardsize = ps;
 587       new_attr.stackaddr = NULL;
 588       new_attr.stacksize = 0;
 589       new_attr.cpuset = NULL;
 590
 591       /* We will pass this value on to the real implementation.  */
 592       attr = (pthread_attr_t *) &new_attr;
 593     }
 594
 595   return __pthread_create_2_1 (newthread, attr, start_routine, arg);
 596 }
 597 compat_symbol (libpthread, __pthread_create_2_0, pthread_create,
 598                GLIBC_2_0);
 599 #endif
 600 \f
 601 /* Information for libthread_db.  */
 602
 603 #include "../nptl_db/db_info.c"
 604 \f
 605 /* If pthread_create is present, libgcc_eh.a and libsupc++.a expects some other POSIX thread
 606    functions to be present as well.  */
 607 PTHREAD_STATIC_FN_REQUIRE (pthread_mutex_lock)
 608 PTHREAD_STATIC_FN_REQUIRE (pthread_mutex_trylock)
 609 PTHREAD_STATIC_FN_REQUIRE (pthread_mutex_unlock)
 610
 611 PTHREAD_STATIC_FN_REQUIRE (pthread_once)
 612 PTHREAD_STATIC_FN_REQUIRE (pthread_cancel)
 613
 614 PTHREAD_STATIC_FN_REQUIRE (pthread_key_create)
 615 PTHREAD_STATIC_FN_REQUIRE (pthread_key_delete)
 616 PTHREAD_STATIC_FN_REQUIRE (pthread_setspecific)
 617 PTHREAD_STATIC_FN_REQUIRE (pthread_getspecific)