nptl/pthread_create.c

   1 /* Copyright (C) 2002-2012 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, see
  17    <http://www.gnu.org/licenses/>.  */
  18
  19 #include <ctype.h>
  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 #include <stap-probe.h>
  35
  36
  37 /* Local function to start thread and handle cleanup.  */
  38 static int start_thread (void *arg);
  39
  40
  41 /* Nozero if debugging mode is enabled.  */
  42 int __pthread_debug;
  43
  44 /* Globally enabled events.  */
  45 static td_thr_events_t __nptl_threads_events __attribute_used__;
  46
  47 /* Pointer to descriptor with the last event.  */
  48 static struct pthread *__nptl_last_event __attribute_used__;
  49
  50 /* Number of threads running.  */
  51 unsigned int __nptl_nthreads = 1;
  52
  53
  54 /* Code to allocate and deallocate a stack.  */
  55 #include "allocatestack.c"
  56
  57 /* Code to create the thread.  */
  58 #include <createthread.c>
  59
  60
  61 struct pthread *
  62 internal_function
  63 __find_in_stack_list (pd)
  64      struct pthread *pd;
  65 {
  66   list_t *entry;
  67   struct pthread *result = NULL;
  68
  69   lll_lock (stack_cache_lock, LLL_PRIVATE);
  70
  71   list_for_each (entry, &stack_used)
  72     {
  73       struct pthread *curp;
  74
  75       curp = list_entry (entry, struct pthread, list);
  76       if (curp == pd)
  77         {
  78           result = curp;
  79           break;
  80         }
  81     }
  82
  83   if (result == NULL)
  84     list_for_each (entry, &__stack_user)
  85       {
  86         struct pthread *curp;
  87
  88         curp = list_entry (entry, struct pthread, list);
  89         if (curp == pd)
  90           {
  91             result = curp;
  92             break;
  93           }
  94       }
  95
  96   lll_unlock (stack_cache_lock, LLL_PRIVATE);
  97
  98   return result;
  99 }
 100
 101
 102 /* Deallocate POSIX thread-local-storage.  */
 103 void
 104 attribute_hidden
 105 __nptl_deallocate_tsd (void)
 106 {
 107   struct pthread *self = THREAD_SELF;
 108
 109   /* Maybe no data was ever allocated.  This happens often so we have
 110      a flag for this.  */
 111   if (THREAD_GETMEM (self, specific_used))
 112     {
 113       size_t round;
 114       size_t cnt;
 115
 116       round = 0;
 117       do
 118         {
 119           size_t idx;
 120
 121           /* So far no new nonzero data entry.  */
 122           THREAD_SETMEM (self, specific_used, false);
 123
 124           for (cnt = idx = 0; cnt < PTHREAD_KEY_1STLEVEL_SIZE; ++cnt)
 125             {
 126               struct pthread_key_data *level2;
 127
 128               level2 = THREAD_GETMEM_NC (self, specific, cnt);
 129
 130               if (level2 != NULL)
 131                 {
 132                   size_t inner;
 133
 134                   for (inner = 0; inner < PTHREAD_KEY_2NDLEVEL_SIZE;
 135                        ++inner, ++idx)
 136                     {
 137                       void *data = level2[inner].data;
 138
 139                       if (data != NULL)
 140                         {
 141                           /* Always clear the data.  */
 142                           level2[inner].data = NULL;
 143
 144                           /* Make sure the data corresponds to a valid
 145                              key.  This test fails if the key was
 146                              deallocated and also if it was
 147                              re-allocated.  It is the user's
 148                              responsibility to free the memory in this
 149                              case.  */
 150                           if (level2[inner].seq
 151                               == __pthread_keys[idx].seq
 152                               /* It is not necessary to register a destructor
 153                                  function.  */
 154                               && __pthread_keys[idx].destr != NULL)
 155                             /* Call the user-provided destructor.  */
 156                             __pthread_keys[idx].destr (data);
 157                         }
 158                     }
 159                 }
 160               else
 161                 idx += PTHREAD_KEY_1STLEVEL_SIZE;
 162             }
 163
 164           if (THREAD_GETMEM (self, specific_used) == 0)
 165             /* No data has been modified.  */
 166             goto just_free;
 167         }
 168       /* We only repeat the process a fixed number of times.  */
 169       while (__builtin_expect (++round < PTHREAD_DESTRUCTOR_ITERATIONS, 0));
 170
 171       /* Just clear the memory of the first block for reuse.  */
 172       memset (&THREAD_SELF->specific_1stblock, '\0',
 173               sizeof (self->specific_1stblock));
 174
 175     just_free:
 176       /* Free the memory for the other blocks.  */
 177       for (cnt = 1; cnt < PTHREAD_KEY_1STLEVEL_SIZE; ++cnt)
 178         {
 179           struct pthread_key_data *level2;
 180
 181           level2 = THREAD_GETMEM_NC (self, specific, cnt);
 182           if (level2 != NULL)
 183             {
 184               /* The first block is allocated as part of the thread
 185                  descriptor.  */
 186               free (level2);
 187               THREAD_SETMEM_NC (self, specific, cnt, NULL);
 188             }
 189         }
 190
 191       THREAD_SETMEM (self, specific_used, false);
 192     }
 193 }
 194
 195
 196 /* Deallocate a thread's stack after optionally making sure the thread
 197    descriptor is still valid.  */
 198 void
 199 internal_function
 200 __free_tcb (struct pthread *pd)
 201 {
 202   /* The thread is exiting now.  */
 203   if (__builtin_expect (atomic_bit_test_set (&pd->cancelhandling,
 204                                              TERMINATED_BIT) == 0, 1))
 205     {
 206       /* Remove the descriptor from the list.  */
 207       if (DEBUGGING_P && __find_in_stack_list (pd) == NULL)
 208         /* Something is really wrong.  The descriptor for a still
 209            running thread is gone.  */
 210         abort ();
 211
 212       /* Free TPP data.  */
 213       if (__builtin_expect (pd->tpp != NULL, 0))
 214         {
 215           struct priority_protection_data *tpp = pd->tpp;
 216
 217           pd->tpp = NULL;
 218           free (tpp);
 219         }
 220
 221       /* Queue the stack memory block for reuse and exit the process.  The
 222          kernel will signal via writing to the address returned by
 223          QUEUE-STACK when the stack is available.  */
 224       __deallocate_stack (pd);
 225     }
 226 }
 227
 228
 229 static int
 230 start_thread (void *arg)
 231 {
 232   struct pthread *pd = (struct pthread *) arg;
 233
 234 #if HP_TIMING_AVAIL
 235   /* Remember the time when the thread was started.  */
 236   hp_timing_t now;
 237   HP_TIMING_NOW (now);
 238   THREAD_SETMEM (pd, cpuclock_offset, now);
 239 #endif
 240
 241   /* Initialize resolver state pointer.  */
 242   __resp = &pd->res;
 243
 244   /* Initialize pointers to locale data.  */
 245   __ctype_init ();
 246
 247   /* Allow setxid from now onwards.  */
 248   if (__builtin_expect (atomic_exchange_acq (&pd->setxid_futex, 0) == -2, 0))
 249     lll_futex_wake (&pd->setxid_futex, 1, LLL_PRIVATE);
 250
 251 #ifdef __NR_set_robust_list
 252 # ifndef __ASSUME_SET_ROBUST_LIST
 253   if (__set_robust_list_avail >= 0)
 254 # endif
 255     {
 256       INTERNAL_SYSCALL_DECL (err);
 257       /* This call should never fail because the initial call in init.c
 258          succeeded.  */
 259       INTERNAL_SYSCALL (set_robust_list, err, 2, &pd->robust_head,
 260                         sizeof (struct robust_list_head));
 261     }
 262 #endif
 263
 264   /* If the parent was running cancellation handlers while creating
 265      the thread the new thread inherited the signal mask.  Reset the
 266      cancellation signal mask.  */
 267   if (__builtin_expect (pd->parent_cancelhandling & CANCELING_BITMASK, 0))
 268     {
 269       INTERNAL_SYSCALL_DECL (err);
 270       sigset_t mask;
 271       __sigemptyset (&mask);
 272       __sigaddset (&mask, SIGCANCEL);
 273       (void) INTERNAL_SYSCALL (rt_sigprocmask, err, 4, SIG_UNBLOCK, &mask,
 274                                NULL, _NSIG / 8);
 275     }
 276
 277   /* This is where the try/finally block should be created.  For
 278      compilers without that support we do use setjmp.  */
 279   struct pthread_unwind_buf unwind_buf;
 280
 281   /* No previous handlers.  */
 282   unwind_buf.priv.data.prev = NULL;
 283   unwind_buf.priv.data.cleanup = NULL;
 284
 285   int not_first_call;
 286   not_first_call = setjmp ((struct __jmp_buf_tag *) unwind_buf.cancel_jmp_buf);
 287   if (__builtin_expect (! not_first_call, 1))
 288     {
 289       /* Store the new cleanup handler info.  */
 290       THREAD_SETMEM (pd, cleanup_jmp_buf, &unwind_buf);
 291
 292       if (__builtin_expect (pd->stopped_start, 0))
 293         {
 294           int oldtype = CANCEL_ASYNC ();
 295
 296           /* Get the lock the parent locked to force synchronization.  */
 297           lll_lock (pd->lock, LLL_PRIVATE);
 298           /* And give it up right away.  */
 299           lll_unlock (pd->lock, LLL_PRIVATE);
 300
 301           CANCEL_RESET (oldtype);
 302         }
 303
 304       LIBC_PROBE (pthread_start, 3, (pthread_t) pd, pd->start_routine, pd->arg);
 305
 306       /* Run the code the user provided.  */
 307 #ifdef CALL_THREAD_FCT
 308       THREAD_SETMEM (pd, result, CALL_THREAD_FCT (pd));
 309 #else
 310       THREAD_SETMEM (pd, result, pd->start_routine (pd->arg));
 311 #endif
 312     }
 313
 314   /* Run the destructor for the thread-local data.  */
 315   __nptl_deallocate_tsd ();
 316
 317   /* Clean up any state libc stored in thread-local variables.  */
 318   __libc_thread_freeres ();
 319
 320   /* If this is the last thread we terminate the process now.  We
 321      do not notify the debugger, it might just irritate it if there
 322      is no thread left.  */
 323   if (__builtin_expect (atomic_decrement_and_test (&__nptl_nthreads), 0))
 324     /* This was the last thread.  */
 325     exit (0);
 326
 327   /* Report the death of the thread if this is wanted.  */
 328   if (__builtin_expect (pd->report_events, 0))
 329     {
 330       /* See whether TD_DEATH is in any of the mask.  */
 331       const int idx = __td_eventword (TD_DEATH);
 332       const uint32_t mask = __td_eventmask (TD_DEATH);
 333
 334       if ((mask & (__nptl_threads_events.event_bits[idx]
 335                    | pd->eventbuf.eventmask.event_bits[idx])) != 0)
 336         {
 337           /* Yep, we have to signal the death.  Add the descriptor to
 338              the list but only if it is not already on it.  */
 339           if (pd->nextevent == NULL)
 340             {
 341               pd->eventbuf.eventnum = TD_DEATH;
 342               pd->eventbuf.eventdata = pd;
 343
 344               do
 345                 pd->nextevent = __nptl_last_event;
 346               while (atomic_compare_and_exchange_bool_acq (&__nptl_last_event,
 347                                                            pd, pd->nextevent));
 348             }
 349
 350           /* Now call the function to signal the event.  */
 351           __nptl_death_event ();
 352         }
 353     }
 354
 355   /* The thread is exiting now.  Don't set this bit until after we've hit
 356      the event-reporting breakpoint, so that td_thr_get_info on us while at
 357      the breakpoint reports TD_THR_RUN state rather than TD_THR_ZOMBIE.  */
 358   atomic_bit_set (&pd->cancelhandling, EXITING_BIT);
 359
 360 #ifndef __ASSUME_SET_ROBUST_LIST
 361   /* If this thread has any robust mutexes locked, handle them now.  */
 362 # ifdef __PTHREAD_MUTEX_HAVE_PREV
 363   void *robust = pd->robust_head.list;
 364 # else
 365   __pthread_slist_t *robust = pd->robust_list.__next;
 366 # endif
 367   /* We let the kernel do the notification if it is able to do so.
 368      If we have to do it here there for sure are no PI mutexes involved
 369      since the kernel support for them is even more recent.  */
 370   if (__set_robust_list_avail < 0
 371       && __builtin_expect (robust != (void *) &pd->robust_head, 0))
 372     {
 373       do
 374         {
 375           struct __pthread_mutex_s *this = (struct __pthread_mutex_s *)
 376             ((char *) robust - offsetof (struct __pthread_mutex_s,
 377                                          __list.__next));
 378           robust = *((void **) robust);
 379
 380 # ifdef __PTHREAD_MUTEX_HAVE_PREV
 381           this->__list.__prev = NULL;
 382 # endif
 383           this->__list.__next = NULL;
 384
 385           lll_robust_dead (this->__lock, /* XYZ */ LLL_SHARED);
 386         }
 387       while (robust != (void *) &pd->robust_head);
 388     }
 389 #endif
 390
 391   /* Mark the memory of the stack as usable to the kernel.  We free
 392      everything except for the space used for the TCB itself.  */
 393   size_t pagesize_m1 = __getpagesize () - 1;
 394 #ifdef _STACK_GROWS_DOWN
 395   char *sp = CURRENT_STACK_FRAME;
 396   size_t freesize = (sp - (char *) pd->stackblock) & ~pagesize_m1;
 397 #else
 398 # error "to do"
 399 #endif
 400   assert (freesize < pd->stackblock_size);
 401   if (freesize > PTHREAD_STACK_MIN)
 402     madvise (pd->stackblock, freesize - PTHREAD_STACK_MIN, MADV_DONTNEED);
 403
 404   /* If the thread is detached free the TCB.  */
 405   if (IS_DETACHED (pd))
 406     /* Free the TCB.  */
 407     __free_tcb (pd);
 408   else if (__builtin_expect (pd->cancelhandling & SETXID_BITMASK, 0))
 409     {
 410       /* Some other thread might call any of the setXid functions and expect
 411          us to reply.  In this case wait until we did that.  */
 412       do
 413         lll_futex_wait (&pd->setxid_futex, 0, LLL_PRIVATE);
 414       while (pd->cancelhandling & SETXID_BITMASK);
 415
 416       /* Reset the value so that the stack can be reused.  */
 417       pd->setxid_futex = 0;
 418     }
 419
 420   /* We cannot call '_exit' here.  '_exit' will terminate the process.
 421
 422      The 'exit' implementation in the kernel will signal when the
 423      process is really dead since 'clone' got passed the CLONE_CLEARTID
 424      flag.  The 'tid' field in the TCB will be set to zero.
 425
 426      The exit code is zero since in case all threads exit by calling
 427      'pthread_exit' the exit status must be 0 (zero).  */
 428   __exit_thread_inline (0);
 429
 430   /* NOTREACHED */
 431   return 0;
 432 }
 433
 434
 435 /* Default thread attributes for the case when the user does not
 436    provide any.  */
 437 static const struct pthread_attr default_attr =
 438   {
 439     /* Just some value > 0 which gets rounded to the nearest page size.  */
 440     .guardsize = 1,
 441   };
 442
 443
 444 int
 445 __pthread_create_2_1 (newthread, attr, start_routine, arg)
 446      pthread_t *newthread;
 447      const pthread_attr_t *attr;
 448      void *(*start_routine) (void *);
 449      void *arg;
 450 {
 451   STACK_VARIABLES;
 452
 453   const struct pthread_attr *iattr = (struct pthread_attr *) attr;
 454   if (iattr == NULL)
 455     /* Is this the best idea?  On NUMA machines this could mean
 456        accessing far-away memory.  */
 457     iattr = &default_attr;
 458
 459   struct pthread *pd = NULL;
 460   int err = ALLOCATE_STACK (iattr, &pd);
 461   if (__builtin_expect (err != 0, 0))
 462     /* Something went wrong.  Maybe a parameter of the attributes is
 463        invalid or we could not allocate memory.  Note we have to
 464        translate error codes.  */
 465     return err == ENOMEM ? EAGAIN : err;
 466
 467
 468   /* Initialize the TCB.  All initializations with zero should be
 469      performed in 'get_cached_stack'.  This way we avoid doing this if
 470      the stack freshly allocated with 'mmap'.  */
 471
 472 #ifdef TLS_TCB_AT_TP
 473   /* Reference to the TCB itself.  */
 474   pd->header.self = pd;
 475
 476   /* Self-reference for TLS.  */
 477   pd->header.tcb = pd;
 478 #endif
 479
 480   /* Store the address of the start routine and the parameter.  Since
 481      we do not start the function directly the stillborn thread will
 482      get the information from its thread descriptor.  */
 483   pd->start_routine = start_routine;
 484   pd->arg = arg;
 485
 486   /* Copy the thread attribute flags.  */
 487   struct pthread *self = THREAD_SELF;
 488   pd->flags = ((iattr->flags & ~(ATTR_FLAG_SCHED_SET | ATTR_FLAG_POLICY_SET))
 489                | (self->flags & (ATTR_FLAG_SCHED_SET | ATTR_FLAG_POLICY_SET)));
 490
 491   /* Initialize the field for the ID of the thread which is waiting
 492      for us.  This is a self-reference in case the thread is created
 493      detached.  */
 494   pd->joinid = iattr->flags & ATTR_FLAG_DETACHSTATE ? pd : NULL;
 495
 496   /* The debug events are inherited from the parent.  */
 497   pd->eventbuf = self->eventbuf;
 498
 499
 500   /* Copy the parent's scheduling parameters.  The flags will say what
 501      is valid and what is not.  */
 502   pd->schedpolicy = self->schedpolicy;
 503   pd->schedparam = self->schedparam;
 504
 505   /* Copy the stack guard canary.  */
 506 #ifdef THREAD_COPY_STACK_GUARD
 507   THREAD_COPY_STACK_GUARD (pd);
 508 #endif
 509
 510   /* Copy the pointer guard value.  */
 511 #ifdef THREAD_COPY_POINTER_GUARD
 512   THREAD_COPY_POINTER_GUARD (pd);
 513 #endif
 514
 515   /* Determine scheduling parameters for the thread.  */
 516   if (attr != NULL
 517       && __builtin_expect ((iattr->flags & ATTR_FLAG_NOTINHERITSCHED) != 0, 0)
 518       && (iattr->flags & (ATTR_FLAG_SCHED_SET | ATTR_FLAG_POLICY_SET)) != 0)
 519     {
 520       INTERNAL_SYSCALL_DECL (scerr);
 521
 522       /* Use the scheduling parameters the user provided.  */
 523       if (iattr->flags & ATTR_FLAG_POLICY_SET)
 524         pd->schedpolicy = iattr->schedpolicy;
 525       else if ((pd->flags & ATTR_FLAG_POLICY_SET) == 0)
 526         {
 527           pd->schedpolicy = INTERNAL_SYSCALL (sched_getscheduler, scerr, 1, 0);
 528           pd->flags |= ATTR_FLAG_POLICY_SET;
 529         }
 530
 531       if (iattr->flags & ATTR_FLAG_SCHED_SET)
 532         memcpy (&pd->schedparam, &iattr->schedparam,
 533                 sizeof (struct sched_param));
 534       else if ((pd->flags & ATTR_FLAG_SCHED_SET) == 0)
 535         {
 536           INTERNAL_SYSCALL (sched_getparam, scerr, 2, 0, &pd->schedparam);
 537           pd->flags |= ATTR_FLAG_SCHED_SET;
 538         }
 539
 540       /* Check for valid priorities.  */
 541       int minprio = INTERNAL_SYSCALL (sched_get_priority_min, scerr, 1,
 542                                       iattr->schedpolicy);
 543       int maxprio = INTERNAL_SYSCALL (sched_get_priority_max, scerr, 1,
 544                                       iattr->schedpolicy);
 545       if (pd->schedparam.sched_priority < minprio
 546           || pd->schedparam.sched_priority > maxprio)
 547         {
 548           /* Perhaps a thread wants to change the IDs and if waiting
 549              for this stillborn thread.  */
 550           if (__builtin_expect (atomic_exchange_acq (&pd->setxid_futex, 0)
 551                                 == -2, 0))
 552             lll_futex_wake (&pd->setxid_futex, 1, LLL_PRIVATE);
 553
 554           __deallocate_stack (pd);
 555
 556           return EINVAL;
 557         }
 558     }
 559
 560   /* Pass the descriptor to the caller.  */
 561   *newthread = (pthread_t) pd;
 562
 563   LIBC_PROBE (pthread_create, 4, newthread, attr, start_routine, arg);
 564
 565   /* Start the thread.  */
 566   return create_thread (pd, iattr, STACK_VARIABLES_ARGS);
 567 }
 568 versioned_symbol (libpthread, __pthread_create_2_1, pthread_create, GLIBC_2_1);
 569
 570
 571 #if SHLIB_COMPAT(libpthread, GLIBC_2_0, GLIBC_2_1)
 572 int
 573 __pthread_create_2_0 (newthread, attr, start_routine, arg)
 574      pthread_t *newthread;
 575      const pthread_attr_t *attr;
 576      void *(*start_routine) (void *);
 577      void *arg;
 578 {
 579   /* The ATTR attribute is not really of type `pthread_attr_t *'.  It has
 580      the old size and access to the new members might crash the program.
 581      We convert the struct now.  */
 582   struct pthread_attr new_attr;
 583
 584   if (attr != NULL)
 585     {
 586       struct pthread_attr *iattr = (struct pthread_attr *) attr;
 587       size_t ps = __getpagesize ();
 588
 589       /* Copy values from the user-provided attributes.  */
 590       new_attr.schedparam = iattr->schedparam;
 591       new_attr.schedpolicy = iattr->schedpolicy;
 592       new_attr.flags = iattr->flags;
 593
 594       /* Fill in default values for the fields not present in the old
 595          implementation.  */
 596       new_attr.guardsize = ps;
 597       new_attr.stackaddr = NULL;
 598       new_attr.stacksize = 0;
 599       new_attr.cpuset = NULL;
 600
 601       /* We will pass this value on to the real implementation.  */
 602       attr = (pthread_attr_t *) &new_attr;
 603     }
 604
 605   return __pthread_create_2_1 (newthread, attr, start_routine, arg);
 606 }
 607 compat_symbol (libpthread, __pthread_create_2_0, pthread_create,
 608                GLIBC_2_0);
 609 #endif
 610 \f
 611 /* Information for libthread_db.  */
 612
 613 #include "../nptl_db/db_info.c"
 614 \f
 615 /* If pthread_create is present, libgcc_eh.a and libsupc++.a expects some other POSIX thread
 616    functions to be present as well.  */
 617 PTHREAD_STATIC_FN_REQUIRE (pthread_mutex_lock)
 618 PTHREAD_STATIC_FN_REQUIRE (pthread_mutex_trylock)
 619 PTHREAD_STATIC_FN_REQUIRE (pthread_mutex_unlock)
 620
 621 PTHREAD_STATIC_FN_REQUIRE (pthread_once)
 622 PTHREAD_STATIC_FN_REQUIRE (pthread_cancel)
 623
 624 PTHREAD_STATIC_FN_REQUIRE (pthread_key_create)
 625 PTHREAD_STATIC_FN_REQUIRE (pthread_key_delete)
 626 PTHREAD_STATIC_FN_REQUIRE (pthread_setspecific)
 627 PTHREAD_STATIC_FN_REQUIRE (pthread_getspecific)