nptl/nptl-init.c

   1 /* Copyright (C) 2002-2007, 2008, 2009, 2011 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 <assert.h>
  20 #include <errno.h>
  21 #include <limits.h>
  22 #include <signal.h>
  23 #include <stdlib.h>
  24 #include <unistd.h>
  25 #include <sys/param.h>
  26 #include <sys/resource.h>
  27 #include <pthreadP.h>
  28 #include <atomic.h>
  29 #include <ldsodefs.h>
  30 #include <tls.h>
  31 #include <list.h>
  32 #include <fork.h>
  33 #include <version.h>
  34 #include <shlib-compat.h>
  35 #include <smp.h>
  36 #include <lowlevellock.h>
  37 #include <kernel-features.h>
  38
  39
  40 /* Size and alignment of static TLS block.  */
  41 size_t __static_tls_size;
  42 size_t __static_tls_align_m1;
  43
  44 #ifndef __ASSUME_SET_ROBUST_LIST
  45 /* Negative if we do not have the system call and we can use it.  */
  46 int __set_robust_list_avail;
  47 # define set_robust_list_not_avail() \
  48   __set_robust_list_avail = -1
  49 #else
  50 # define set_robust_list_not_avail() do { } while (0)
  51 #endif
  52
  53 #ifndef __ASSUME_FUTEX_CLOCK_REALTIME
  54 /* Nonzero if we do not have FUTEX_CLOCK_REALTIME.  */
  55 int __have_futex_clock_realtime;
  56 # define __set_futex_clock_realtime() \
  57   __have_futex_clock_realtime = 1
  58 #else
  59 #define __set_futex_clock_realtime() do { } while (0)
  60 #endif
  61
  62 /* Version of the library, used in libthread_db to detect mismatches.  */
  63 static const char nptl_version[] __attribute_used__ = VERSION;
  64
  65
  66 #ifndef SHARED
  67 extern void __libc_setup_tls (size_t tcbsize, size_t tcbalign);
  68 #endif
  69
  70 #ifdef SHARED
  71 static
  72 #else
  73 extern
  74 #endif
  75 void __nptl_set_robust (struct pthread *);
  76
  77 #ifdef SHARED
  78 static void nptl_freeres (void);
  79
  80
  81 static const struct pthread_functions pthread_functions =
  82   {
  83     .ptr_pthread_attr_destroy = __pthread_attr_destroy,
  84 # if SHLIB_COMPAT(libpthread, GLIBC_2_0, GLIBC_2_1)
  85     .ptr___pthread_attr_init_2_0 = __pthread_attr_init_2_0,
  86 # endif
  87     .ptr___pthread_attr_init_2_1 = __pthread_attr_init_2_1,
  88     .ptr_pthread_attr_getdetachstate = __pthread_attr_getdetachstate,
  89     .ptr_pthread_attr_setdetachstate = __pthread_attr_setdetachstate,
  90     .ptr_pthread_attr_getinheritsched = __pthread_attr_getinheritsched,
  91     .ptr_pthread_attr_setinheritsched = __pthread_attr_setinheritsched,
  92     .ptr_pthread_attr_getschedparam = __pthread_attr_getschedparam,
  93     .ptr_pthread_attr_setschedparam = __pthread_attr_setschedparam,
  94     .ptr_pthread_attr_getschedpolicy = __pthread_attr_getschedpolicy,
  95     .ptr_pthread_attr_setschedpolicy = __pthread_attr_setschedpolicy,
  96     .ptr_pthread_attr_getscope = __pthread_attr_getscope,
  97     .ptr_pthread_attr_setscope = __pthread_attr_setscope,
  98     .ptr_pthread_condattr_destroy = __pthread_condattr_destroy,
  99     .ptr_pthread_condattr_init = __pthread_condattr_init,
 100     .ptr___pthread_cond_broadcast = __pthread_cond_broadcast,
 101     .ptr___pthread_cond_destroy = __pthread_cond_destroy,
 102     .ptr___pthread_cond_init = __pthread_cond_init,
 103     .ptr___pthread_cond_signal = __pthread_cond_signal,
 104     .ptr___pthread_cond_wait = __pthread_cond_wait,
 105     .ptr___pthread_cond_timedwait = __pthread_cond_timedwait,
 106 # if SHLIB_COMPAT(libpthread, GLIBC_2_0, GLIBC_2_3_2)
 107     .ptr___pthread_cond_broadcast_2_0 = __pthread_cond_broadcast_2_0,
 108     .ptr___pthread_cond_destroy_2_0 = __pthread_cond_destroy_2_0,
 109     .ptr___pthread_cond_init_2_0 = __pthread_cond_init_2_0,
 110     .ptr___pthread_cond_signal_2_0 = __pthread_cond_signal_2_0,
 111     .ptr___pthread_cond_wait_2_0 = __pthread_cond_wait_2_0,
 112     .ptr___pthread_cond_timedwait_2_0 = __pthread_cond_timedwait_2_0,
 113 # endif
 114     .ptr_pthread_equal = __pthread_equal,
 115     .ptr___pthread_exit = __pthread_exit,
 116     .ptr_pthread_getschedparam = __pthread_getschedparam,
 117     .ptr_pthread_setschedparam = __pthread_setschedparam,
 118     .ptr_pthread_mutex_destroy = INTUSE(__pthread_mutex_destroy),
 119     .ptr_pthread_mutex_init = INTUSE(__pthread_mutex_init),
 120     .ptr_pthread_mutex_lock = INTUSE(__pthread_mutex_lock),
 121     .ptr_pthread_mutex_unlock = INTUSE(__pthread_mutex_unlock),
 122     .ptr_pthread_self = __pthread_self,
 123     .ptr_pthread_setcancelstate = __pthread_setcancelstate,
 124     .ptr_pthread_setcanceltype = __pthread_setcanceltype,
 125     .ptr___pthread_cleanup_upto = __pthread_cleanup_upto,
 126     .ptr___pthread_once = __pthread_once_internal,
 127     .ptr___pthread_rwlock_rdlock = __pthread_rwlock_rdlock_internal,
 128     .ptr___pthread_rwlock_wrlock = __pthread_rwlock_wrlock_internal,
 129     .ptr___pthread_rwlock_unlock = __pthread_rwlock_unlock_internal,
 130     .ptr___pthread_key_create = __pthread_key_create_internal,
 131     .ptr___pthread_getspecific = __pthread_getspecific_internal,
 132     .ptr___pthread_setspecific = __pthread_setspecific_internal,
 133     .ptr__pthread_cleanup_push_defer = __pthread_cleanup_push_defer,
 134     .ptr__pthread_cleanup_pop_restore = __pthread_cleanup_pop_restore,
 135     .ptr_nthreads = &__nptl_nthreads,
 136     .ptr___pthread_unwind = &__pthread_unwind,
 137     .ptr__nptl_deallocate_tsd = __nptl_deallocate_tsd,
 138     .ptr__nptl_setxid = __nptl_setxid,
 139     /* For now only the stack cache needs to be freed.  */
 140     .ptr_freeres = nptl_freeres,
 141     .ptr_set_robust = __nptl_set_robust
 142   };
 143 # define ptr_pthread_functions &pthread_functions
 144 #else
 145 # define ptr_pthread_functions NULL
 146 #endif
 147
 148
 149 #ifdef SHARED
 150 /* This function is called indirectly from the freeres code in libc.  */
 151 static void
 152 __libc_freeres_fn_section
 153 nptl_freeres (void)
 154 {
 155   __unwind_freeres ();
 156   __free_stacks (0);
 157 }
 158
 159
 160 static
 161 #endif
 162 void
 163 __nptl_set_robust (struct pthread *self)
 164 {
 165   INTERNAL_SYSCALL_DECL (err);
 166   INTERNAL_SYSCALL (set_robust_list, err, 2, &self->robust_head,
 167                     sizeof (struct robust_list_head));
 168 }
 169
 170
 171 /* For asynchronous cancellation we use a signal.  This is the handler.  */
 172 static void
 173 sigcancel_handler (int sig, siginfo_t *si, void *ctx)
 174 {
 175 #ifdef __ASSUME_CORRECT_SI_PID
 176   /* Determine the process ID.  It might be negative if the thread is
 177      in the middle of a fork() call.  */
 178   pid_t pid = THREAD_GETMEM (THREAD_SELF, pid);
 179   if (__builtin_expect (pid < 0, 0))
 180     pid = -pid;
 181 #endif
 182
 183   /* Safety check.  It would be possible to call this function for
 184      other signals and send a signal from another process.  This is not
 185      correct and might even be a security problem.  Try to catch as
 186      many incorrect invocations as possible.  */
 187   if (sig != SIGCANCEL
 188 #ifdef __ASSUME_CORRECT_SI_PID
 189       /* Kernels before 2.5.75 stored the thread ID and not the process
 190          ID in si_pid so we skip this test.  */
 191       || si->si_pid != pid
 192 #endif
 193       || si->si_code != SI_TKILL)
 194     return;
 195
 196   struct pthread *self = THREAD_SELF;
 197
 198   int oldval = THREAD_GETMEM (self, cancelhandling);
 199   while (1)
 200     {
 201       /* We are canceled now.  When canceled by another thread this flag
 202          is already set but if the signal is directly send (internally or
 203          from another process) is has to be done here.  */
 204       int newval = oldval | CANCELING_BITMASK | CANCELED_BITMASK;
 205
 206       if (oldval == newval || (oldval & EXITING_BITMASK) != 0)
 207         /* Already canceled or exiting.  */
 208         break;
 209
 210       int curval = THREAD_ATOMIC_CMPXCHG_VAL (self, cancelhandling, newval,
 211                                               oldval);
 212       if (curval == oldval)
 213         {
 214           /* Set the return value.  */
 215           THREAD_SETMEM (self, result, PTHREAD_CANCELED);
 216
 217           /* Make sure asynchronous cancellation is still enabled.  */
 218           if ((newval & CANCELTYPE_BITMASK) != 0)
 219             /* Run the registered destructors and terminate the thread.  */
 220             __do_cancel ();
 221
 222           break;
 223         }
 224
 225       oldval = curval;
 226     }
 227 }
 228
 229
 230 struct xid_command *__xidcmd attribute_hidden;
 231
 232 /* For asynchronous cancellation we use a signal.  This is the handler.  */
 233 static void
 234 sighandler_setxid (int sig, siginfo_t *si, void *ctx)
 235 {
 236 #ifdef __ASSUME_CORRECT_SI_PID
 237   /* Determine the process ID.  It might be negative if the thread is
 238      in the middle of a fork() call.  */
 239   pid_t pid = THREAD_GETMEM (THREAD_SELF, pid);
 240   if (__builtin_expect (pid < 0, 0))
 241     pid = -pid;
 242 #endif
 243
 244   /* Safety check.  It would be possible to call this function for
 245      other signals and send a signal from another process.  This is not
 246      correct and might even be a security problem.  Try to catch as
 247      many incorrect invocations as possible.  */
 248   if (sig != SIGSETXID
 249 #ifdef __ASSUME_CORRECT_SI_PID
 250       /* Kernels before 2.5.75 stored the thread ID and not the process
 251          ID in si_pid so we skip this test.  */
 252       || si->si_pid != pid
 253 #endif
 254       || si->si_code != SI_TKILL)
 255     return;
 256
 257   INTERNAL_SYSCALL_DECL (err);
 258   INTERNAL_SYSCALL_NCS (__xidcmd->syscall_no, err, 3, __xidcmd->id[0],
 259                         __xidcmd->id[1], __xidcmd->id[2]);
 260
 261   /* Reset the SETXID flag.  */
 262   struct pthread *self = THREAD_SELF;
 263   int flags, newval;
 264   do
 265     {
 266       flags = THREAD_GETMEM (self, cancelhandling);
 267       newval = THREAD_ATOMIC_CMPXCHG_VAL (self, cancelhandling,
 268                                           flags & ~SETXID_BITMASK, flags);
 269     }
 270   while (flags != newval);
 271
 272   /* And release the futex.  */
 273   self->setxid_futex = 1;
 274   lll_futex_wake (&self->setxid_futex, 1, LLL_PRIVATE);
 275
 276   if (atomic_decrement_val (&__xidcmd->cntr) == 0)
 277     lll_futex_wake (&__xidcmd->cntr, 1, LLL_PRIVATE);
 278 }
 279
 280
 281 /* When using __thread for this, we do it in libc so as not
 282    to give libpthread its own TLS segment just for this.  */
 283 extern void **__libc_dl_error_tsd (void) __attribute__ ((const));
 284
 285
 286 /* This can be set by the debugger before initialization is complete.  */
 287 static bool __nptl_initial_report_events __attribute_used__;
 288
 289 void
 290 __pthread_initialize_minimal_internal (void)
 291 {
 292 #ifndef SHARED
 293   /* Unlike in the dynamically linked case the dynamic linker has not
 294      taken care of initializing the TLS data structures.  */
 295   __libc_setup_tls (TLS_TCB_SIZE, TLS_TCB_ALIGN);
 296
 297   /* We must prevent gcc from being clever and move any of the
 298      following code ahead of the __libc_setup_tls call.  This function
 299      will initialize the thread register which is subsequently
 300      used.  */
 301   __asm __volatile ("");
 302 #endif
 303
 304   /* Minimal initialization of the thread descriptor.  */
 305   struct pthread *pd = THREAD_SELF;
 306   INTERNAL_SYSCALL_DECL (err);
 307   pd->pid = pd->tid = INTERNAL_SYSCALL (set_tid_address, err, 1, &pd->tid);
 308   THREAD_SETMEM (pd, specific[0], &pd->specific_1stblock[0]);
 309   THREAD_SETMEM (pd, user_stack, true);
 310   if (LLL_LOCK_INITIALIZER != 0)
 311     THREAD_SETMEM (pd, lock, LLL_LOCK_INITIALIZER);
 312 #if HP_TIMING_AVAIL
 313   THREAD_SETMEM (pd, cpuclock_offset, GL(dl_cpuclock_offset));
 314 #endif
 315
 316   /* Initialize the robust mutex data.  */
 317 #ifdef __PTHREAD_MUTEX_HAVE_PREV
 318   pd->robust_prev = &pd->robust_head;
 319 #endif
 320   pd->robust_head.list = &pd->robust_head;
 321 #ifdef __NR_set_robust_list
 322   pd->robust_head.futex_offset = (offsetof (pthread_mutex_t, __data.__lock)
 323                                   - offsetof (pthread_mutex_t,
 324                                               __data.__list.__next));
 325   int res = INTERNAL_SYSCALL (set_robust_list, err, 2, &pd->robust_head,
 326                               sizeof (struct robust_list_head));
 327   if (INTERNAL_SYSCALL_ERROR_P (res, err))
 328 #endif
 329     set_robust_list_not_avail ();
 330
 331 #ifndef __ASSUME_PRIVATE_FUTEX
 332   /* Private futexes are always used (at least internally) so that
 333      doing the test once this early is beneficial.  */
 334   {
 335     int word = 0;
 336     word = INTERNAL_SYSCALL (futex, err, 3, &word,
 337                             FUTEX_WAKE | FUTEX_PRIVATE_FLAG, 1);
 338     if (!INTERNAL_SYSCALL_ERROR_P (word, err))
 339       THREAD_SETMEM (pd, header.private_futex, FUTEX_PRIVATE_FLAG);
 340   }
 341
 342   /* Private futexes have been introduced earlier than the
 343      FUTEX_CLOCK_REALTIME flag.  We don't have to run the test if we
 344      know the former are not supported.  This also means we know the
 345      kernel will return ENOSYS for unknown operations.  */
 346   if (THREAD_GETMEM (pd, header.private_futex) != 0)
 347 #endif
 348 #ifndef __ASSUME_FUTEX_CLOCK_REALTIME
 349     {
 350       int word = 0;
 351       /* NB: the syscall actually takes six parameters.  The last is the
 352          bit mask.  But since we will not actually wait at all the value
 353          is irrelevant.  Given that passing six parameters is difficult
 354          on some architectures we just pass whatever random value the
 355          calling convention calls for to the kernel.  It causes no harm.  */
 356       word = INTERNAL_SYSCALL (futex, err, 5, &word,
 357                                FUTEX_WAIT_BITSET | FUTEX_CLOCK_REALTIME
 358                                | FUTEX_PRIVATE_FLAG, 1, NULL, 0);
 359       assert (INTERNAL_SYSCALL_ERROR_P (word, err));
 360       if (INTERNAL_SYSCALL_ERRNO (word, err) != ENOSYS)
 361         __set_futex_clock_realtime ();
 362     }
 363 #endif
 364
 365   /* Set initial thread's stack block from 0 up to __libc_stack_end.
 366      It will be bigger than it actually is, but for unwind.c/pt-longjmp.c
 367      purposes this is good enough.  */
 368   THREAD_SETMEM (pd, stackblock_size, (size_t) __libc_stack_end);
 369
 370   /* Initialize the list of all running threads with the main thread.  */
 371   INIT_LIST_HEAD (&__stack_user);
 372   list_add (&pd->list, &__stack_user);
 373
 374   /* Before initializing __stack_user, the debugger could not find us and
 375      had to set __nptl_initial_report_events.  Propagate its setting.  */
 376   THREAD_SETMEM (pd, report_events, __nptl_initial_report_events);
 377
 378   /* Install the cancellation signal handler.  If for some reason we
 379      cannot install the handler we do not abort.  Maybe we should, but
 380      it is only asynchronous cancellation which is affected.  */
 381   struct sigaction sa;
 382   sa.sa_sigaction = sigcancel_handler;
 383   sa.sa_flags = SA_SIGINFO;
 384   __sigemptyset (&sa.sa_mask);
 385
 386   (void) __libc_sigaction (SIGCANCEL, &sa, NULL);
 387
 388   /* Install the handle to change the threads' uid/gid.  */
 389   sa.sa_sigaction = sighandler_setxid;
 390   sa.sa_flags = SA_SIGINFO | SA_RESTART;
 391
 392   (void) __libc_sigaction (SIGSETXID, &sa, NULL);
 393
 394   /* The parent process might have left the signals blocked.  Just in
 395      case, unblock it.  We reuse the signal mask in the sigaction
 396      structure.  It is already cleared.  */
 397   __sigaddset (&sa.sa_mask, SIGCANCEL);
 398   __sigaddset (&sa.sa_mask, SIGSETXID);
 399   (void) INTERNAL_SYSCALL (rt_sigprocmask, err, 4, SIG_UNBLOCK, &sa.sa_mask,
 400                            NULL, _NSIG / 8);
 401
 402   /* Get the size of the static and alignment requirements for the TLS
 403      block.  */
 404   size_t static_tls_align;
 405   _dl_get_tls_static_info (&__static_tls_size, &static_tls_align);
 406
 407   /* Make sure the size takes all the alignments into account.  */
 408   if (STACK_ALIGN > static_tls_align)
 409     static_tls_align = STACK_ALIGN;
 410   __static_tls_align_m1 = static_tls_align - 1;
 411
 412   __static_tls_size = roundup (__static_tls_size, static_tls_align);
 413
 414   /* Determine the default allowed stack size.  This is the size used
 415      in case the user does not specify one.  */
 416   struct rlimit limit;
 417   if (getrlimit (RLIMIT_STACK, &limit) != 0
 418       || limit.rlim_cur == RLIM_INFINITY)
 419     /* The system limit is not usable.  Use an architecture-specific
 420        default.  */
 421     limit.rlim_cur = ARCH_STACK_DEFAULT_SIZE;
 422   else if (limit.rlim_cur < PTHREAD_STACK_MIN)
 423     /* The system limit is unusably small.
 424        Use the minimal size acceptable.  */
 425     limit.rlim_cur = PTHREAD_STACK_MIN;
 426
 427   /* Make sure it meets the minimum size that allocate_stack
 428      (allocatestack.c) will demand, which depends on the page size.  */
 429   const uintptr_t pagesz = GLRO(dl_pagesize);
 430   const size_t minstack = pagesz + __static_tls_size + MINIMAL_REST_STACK;
 431   if (limit.rlim_cur < minstack)
 432     limit.rlim_cur = minstack;
 433
 434   /* Round the resource limit up to page size.  */
 435   limit.rlim_cur = (limit.rlim_cur + pagesz - 1) & -pagesz;
 436   __default_stacksize = limit.rlim_cur;
 437
 438 #ifdef SHARED
 439   /* Transfer the old value from the dynamic linker's internal location.  */
 440   *__libc_dl_error_tsd () = *(*GL(dl_error_catch_tsd)) ();
 441   GL(dl_error_catch_tsd) = &__libc_dl_error_tsd;
 442
 443   /* Make __rtld_lock_{,un}lock_recursive use pthread_mutex_{,un}lock,
 444      keep the lock count from the ld.so implementation.  */
 445   GL(dl_rtld_lock_recursive) = (void *) INTUSE (__pthread_mutex_lock);
 446   GL(dl_rtld_unlock_recursive) = (void *) INTUSE (__pthread_mutex_unlock);
 447   unsigned int rtld_lock_count = GL(dl_load_lock).mutex.__data.__count;
 448   GL(dl_load_lock).mutex.__data.__count = 0;
 449   while (rtld_lock_count-- > 0)
 450     INTUSE (__pthread_mutex_lock) (&GL(dl_load_lock).mutex);
 451
 452   GL(dl_make_stack_executable_hook) = &__make_stacks_executable;
 453 #endif
 454
 455   GL(dl_init_static_tls) = &__pthread_init_static_tls;
 456
 457   GL(dl_wait_lookup_done) = &__wait_lookup_done;
 458
 459   /* Register the fork generation counter with the libc.  */
 460 #ifndef TLS_MULTIPLE_THREADS_IN_TCB
 461   __libc_multiple_threads_ptr =
 462 #endif
 463     __libc_pthread_init (&__fork_generation, __reclaim_stacks,
 464                          ptr_pthread_functions);
 465
 466   /* Determine whether the machine is SMP or not.  */
 467   __is_smp = is_smp_system ();
 468 }
 469 strong_alias (__pthread_initialize_minimal_internal,
 470               __pthread_initialize_minimal)
 471
 472
 473 size_t
 474 __pthread_get_minstack (const pthread_attr_t *attr)
 475 {
 476   struct pthread_attr *iattr = (struct pthread_attr *) attr;
 477
 478   return (GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN
 479           + iattr->guardsize);
 480 }