nptl/nptl-init.c

   1 /* Copyright (C) 2002-2014 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 = __pthread_mutex_destroy,
 119     .ptr_pthread_mutex_init = __pthread_mutex_init,
 120     .ptr_pthread_mutex_lock = __pthread_mutex_lock,
 121     .ptr_pthread_mutex_unlock = __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,
 127     .ptr___pthread_rwlock_rdlock = __pthread_rwlock_rdlock,
 128     .ptr___pthread_rwlock_wrlock = __pthread_rwlock_wrlock,
 129     .ptr___pthread_rwlock_unlock = __pthread_rwlock_unlock,
 130     .ptr___pthread_key_create = __pthread_key_create,
 131     .ptr___pthread_getspecific = __pthread_getspecific,
 132     .ptr___pthread_setspecific = __pthread_setspecific,
 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 #ifdef __NR_set_robust_list
 166   INTERNAL_SYSCALL_DECL (err);
 167   INTERNAL_SYSCALL (set_robust_list, err, 2, &self->robust_head,
 168                     sizeof (struct robust_list_head));
 169 #endif
 170 }
 171
 172
 173 #ifdef SIGCANCEL
 174 /* For asynchronous cancellation we use a signal.  This is the handler.  */
 175 static void
 176 sigcancel_handler (int sig, siginfo_t *si, void *ctx)
 177 {
 178   /* Determine the process ID.  It might be negative if the thread is
 179      in the middle of a fork() call.  */
 180   pid_t pid = THREAD_GETMEM (THREAD_SELF, pid);
 181   if (__glibc_unlikely (pid < 0))
 182     pid = -pid;
 183
 184   /* Safety check.  It would be possible to call this function for
 185      other signals and send a signal from another process.  This is not
 186      correct and might even be a security problem.  Try to catch as
 187      many incorrect invocations as possible.  */
 188   if (sig != SIGCANCEL
 189       || si->si_pid != pid
 190       || si->si_code != SI_TKILL)
 191     return;
 192
 193   struct pthread *self = THREAD_SELF;
 194
 195   int oldval = THREAD_GETMEM (self, cancelhandling);
 196   while (1)
 197     {
 198       /* We are canceled now.  When canceled by another thread this flag
 199          is already set but if the signal is directly send (internally or
 200          from another process) is has to be done here.  */
 201       int newval = oldval | CANCELING_BITMASK | CANCELED_BITMASK;
 202
 203       if (oldval == newval || (oldval & EXITING_BITMASK) != 0)
 204         /* Already canceled or exiting.  */
 205         break;
 206
 207       int curval = THREAD_ATOMIC_CMPXCHG_VAL (self, cancelhandling, newval,
 208                                               oldval);
 209       if (curval == oldval)
 210         {
 211           /* Set the return value.  */
 212           THREAD_SETMEM (self, result, PTHREAD_CANCELED);
 213
 214           /* Make sure asynchronous cancellation is still enabled.  */
 215           if ((newval & CANCELTYPE_BITMASK) != 0)
 216             /* Run the registered destructors and terminate the thread.  */
 217             __do_cancel ();
 218
 219           break;
 220         }
 221
 222       oldval = curval;
 223     }
 224 }
 225 #endif
 226
 227
 228 #ifdef SIGSETXID
 229 struct xid_command *__xidcmd attribute_hidden;
 230
 231 /* We use the SIGSETXID signal in the setuid, setgid, etc. implementations to
 232    tell each thread to call the respective setxid syscall on itself.  This is
 233    the handler.  */
 234 static void
 235 sighandler_setxid (int sig, siginfo_t *si, void *ctx)
 236 {
 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   int result;
 241   if (__glibc_unlikely (pid < 0))
 242     pid = -pid;
 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       || si->si_pid != pid
 250       || si->si_code != SI_TKILL)
 251     return;
 252
 253   INTERNAL_SYSCALL_DECL (err);
 254   result = INTERNAL_SYSCALL_NCS (__xidcmd->syscall_no, err, 3, __xidcmd->id[0],
 255                                  __xidcmd->id[1], __xidcmd->id[2]);
 256   int error = 0;
 257   if (__glibc_unlikely (INTERNAL_SYSCALL_ERROR_P (result, err)))
 258     error = INTERNAL_SYSCALL_ERRNO (result, err);
 259   __nptl_setxid_error (__xidcmd, error);
 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 #endif
 280
 281
 282 /* When using __thread for this, we do it in libc so as not
 283    to give libpthread its own TLS segment just for this.  */
 284 extern void **__libc_dl_error_tsd (void) __attribute__ ((const));
 285
 286
 287 /* This can be set by the debugger before initialization is complete.  */
 288 static bool __nptl_initial_report_events __attribute_used__;
 289
 290 void
 291 __pthread_initialize_minimal_internal (void)
 292 {
 293 #ifndef SHARED
 294   /* Unlike in the dynamically linked case the dynamic linker has not
 295      taken care of initializing the TLS data structures.  */
 296   __libc_setup_tls (TLS_TCB_SIZE, TLS_TCB_ALIGN);
 297
 298   /* We must prevent gcc from being clever and move any of the
 299      following code ahead of the __libc_setup_tls call.  This function
 300      will initialize the thread register which is subsequently
 301      used.  */
 302   __asm __volatile ("");
 303 #endif
 304
 305   /* Minimal initialization of the thread descriptor.  */
 306   struct pthread *pd = THREAD_SELF;
 307 #ifdef __NR_set_tid_address
 308   INTERNAL_SYSCALL_DECL (err);
 309   pd->pid = pd->tid = INTERNAL_SYSCALL (set_tid_address, err, 1, &pd->tid);
 310 #endif
 311   THREAD_SETMEM (pd, specific[0], &pd->specific_1stblock[0]);
 312   THREAD_SETMEM (pd, user_stack, true);
 313   if (LLL_LOCK_INITIALIZER != 0)
 314     THREAD_SETMEM (pd, lock, LLL_LOCK_INITIALIZER);
 315 #if HP_TIMING_AVAIL
 316   THREAD_SETMEM (pd, cpuclock_offset, GL(dl_cpuclock_offset));
 317 #endif
 318
 319   /* Initialize the robust mutex data.  */
 320 #ifdef __PTHREAD_MUTEX_HAVE_PREV
 321   pd->robust_prev = &pd->robust_head;
 322 #endif
 323   pd->robust_head.list = &pd->robust_head;
 324 #ifdef __NR_set_robust_list
 325   pd->robust_head.futex_offset = (offsetof (pthread_mutex_t, __data.__lock)
 326                                   - offsetof (pthread_mutex_t,
 327                                               __data.__list.__next));
 328   int res = INTERNAL_SYSCALL (set_robust_list, err, 2, &pd->robust_head,
 329                               sizeof (struct robust_list_head));
 330   if (INTERNAL_SYSCALL_ERROR_P (res, err))
 331 #endif
 332     set_robust_list_not_avail ();
 333
 334 #ifdef __NR_futex
 335 # ifndef __ASSUME_PRIVATE_FUTEX
 336   /* Private futexes are always used (at least internally) so that
 337      doing the test once this early is beneficial.  */
 338   {
 339     int word = 0;
 340     word = INTERNAL_SYSCALL (futex, err, 3, &word,
 341                             FUTEX_WAKE | FUTEX_PRIVATE_FLAG, 1);
 342     if (!INTERNAL_SYSCALL_ERROR_P (word, err))
 343       THREAD_SETMEM (pd, header.private_futex, FUTEX_PRIVATE_FLAG);
 344   }
 345
 346   /* Private futexes have been introduced earlier than the
 347      FUTEX_CLOCK_REALTIME flag.  We don't have to run the test if we
 348      know the former are not supported.  This also means we know the
 349      kernel will return ENOSYS for unknown operations.  */
 350   if (THREAD_GETMEM (pd, header.private_futex) != 0)
 351 # endif
 352 # ifndef __ASSUME_FUTEX_CLOCK_REALTIME
 353     {
 354       int word = 0;
 355       /* NB: the syscall actually takes six parameters.  The last is the
 356          bit mask.  But since we will not actually wait at all the value
 357          is irrelevant.  Given that passing six parameters is difficult
 358          on some architectures we just pass whatever random value the
 359          calling convention calls for to the kernel.  It causes no harm.  */
 360       word = INTERNAL_SYSCALL (futex, err, 5, &word,
 361                                FUTEX_WAIT_BITSET | FUTEX_CLOCK_REALTIME
 362                                | FUTEX_PRIVATE_FLAG, 1, NULL, 0);
 363       assert (INTERNAL_SYSCALL_ERROR_P (word, err));
 364       if (INTERNAL_SYSCALL_ERRNO (word, err) != ENOSYS)
 365         __set_futex_clock_realtime ();
 366     }
 367 # endif
 368 #endif
 369
 370   /* Set initial thread's stack block from 0 up to __libc_stack_end.
 371      It will be bigger than it actually is, but for unwind.c/pt-longjmp.c
 372      purposes this is good enough.  */
 373   THREAD_SETMEM (pd, stackblock_size, (size_t) __libc_stack_end);
 374
 375   /* Initialize the list of all running threads with the main thread.  */
 376   INIT_LIST_HEAD (&__stack_user);
 377   list_add (&pd->list, &__stack_user);
 378
 379   /* Before initializing __stack_user, the debugger could not find us and
 380      had to set __nptl_initial_report_events.  Propagate its setting.  */
 381   THREAD_SETMEM (pd, report_events, __nptl_initial_report_events);
 382
 383 #if defined SIGCANCEL || defined SIGSETXID
 384   struct sigaction sa;
 385   __sigemptyset (&sa.sa_mask);
 386
 387 # ifdef SIGCANCEL
 388   /* Install the cancellation signal handler.  If for some reason we
 389      cannot install the handler we do not abort.  Maybe we should, but
 390      it is only asynchronous cancellation which is affected.  */
 391   sa.sa_sigaction = sigcancel_handler;
 392   sa.sa_flags = SA_SIGINFO;
 393   (void) __libc_sigaction (SIGCANCEL, &sa, NULL);
 394 # endif
 395
 396 # ifdef SIGSETXID
 397   /* Install the handle to change the threads' uid/gid.  */
 398   sa.sa_sigaction = sighandler_setxid;
 399   sa.sa_flags = SA_SIGINFO | SA_RESTART;
 400   (void) __libc_sigaction (SIGSETXID, &sa, NULL);
 401 # endif
 402
 403   /* The parent process might have left the signals blocked.  Just in
 404      case, unblock it.  We reuse the signal mask in the sigaction
 405      structure.  It is already cleared.  */
 406 # ifdef SIGCANCEL
 407   __sigaddset (&sa.sa_mask, SIGCANCEL);
 408 # endif
 409 # ifdef SIGSETXID
 410   __sigaddset (&sa.sa_mask, SIGSETXID);
 411 # endif
 412   (void) INTERNAL_SYSCALL (rt_sigprocmask, err, 4, SIG_UNBLOCK, &sa.sa_mask,
 413                            NULL, _NSIG / 8);
 414 #endif
 415
 416   /* Get the size of the static and alignment requirements for the TLS
 417      block.  */
 418   size_t static_tls_align;
 419   _dl_get_tls_static_info (&__static_tls_size, &static_tls_align);
 420
 421   /* Make sure the size takes all the alignments into account.  */
 422   if (STACK_ALIGN > static_tls_align)
 423     static_tls_align = STACK_ALIGN;
 424   __static_tls_align_m1 = static_tls_align - 1;
 425
 426   __static_tls_size = roundup (__static_tls_size, static_tls_align);
 427
 428   /* Determine the default allowed stack size.  This is the size used
 429      in case the user does not specify one.  */
 430   struct rlimit limit;
 431   if (getrlimit (RLIMIT_STACK, &limit) != 0
 432       || limit.rlim_cur == RLIM_INFINITY)
 433     /* The system limit is not usable.  Use an architecture-specific
 434        default.  */
 435     limit.rlim_cur = ARCH_STACK_DEFAULT_SIZE;
 436   else if (limit.rlim_cur < PTHREAD_STACK_MIN)
 437     /* The system limit is unusably small.
 438        Use the minimal size acceptable.  */
 439     limit.rlim_cur = PTHREAD_STACK_MIN;
 440
 441   /* Make sure it meets the minimum size that allocate_stack
 442      (allocatestack.c) will demand, which depends on the page size.  */
 443   const uintptr_t pagesz = GLRO(dl_pagesize);
 444   const size_t minstack = pagesz + __static_tls_size + MINIMAL_REST_STACK;
 445   if (limit.rlim_cur < minstack)
 446     limit.rlim_cur = minstack;
 447
 448   /* Round the resource limit up to page size.  */
 449   limit.rlim_cur = (limit.rlim_cur + pagesz - 1) & -pagesz;
 450   lll_lock (__default_pthread_attr_lock, LLL_PRIVATE);
 451   __default_pthread_attr.stacksize = limit.rlim_cur;
 452   __default_pthread_attr.guardsize = GLRO (dl_pagesize);
 453   lll_unlock (__default_pthread_attr_lock, LLL_PRIVATE);
 454
 455 #ifdef SHARED
 456   /* Transfer the old value from the dynamic linker's internal location.  */
 457   *__libc_dl_error_tsd () = *(*GL(dl_error_catch_tsd)) ();
 458   GL(dl_error_catch_tsd) = &__libc_dl_error_tsd;
 459
 460   /* Make __rtld_lock_{,un}lock_recursive use pthread_mutex_{,un}lock,
 461      keep the lock count from the ld.so implementation.  */
 462   GL(dl_rtld_lock_recursive) = (void *) __pthread_mutex_lock;
 463   GL(dl_rtld_unlock_recursive) = (void *) __pthread_mutex_unlock;
 464   unsigned int rtld_lock_count = GL(dl_load_lock).mutex.__data.__count;
 465   GL(dl_load_lock).mutex.__data.__count = 0;
 466   while (rtld_lock_count-- > 0)
 467     __pthread_mutex_lock (&GL(dl_load_lock).mutex);
 468
 469   GL(dl_make_stack_executable_hook) = &__make_stacks_executable;
 470 #endif
 471
 472   GL(dl_init_static_tls) = &__pthread_init_static_tls;
 473
 474   GL(dl_wait_lookup_done) = &__wait_lookup_done;
 475
 476   /* Register the fork generation counter with the libc.  */
 477 #ifndef TLS_MULTIPLE_THREADS_IN_TCB
 478   __libc_multiple_threads_ptr =
 479 #endif
 480     __libc_pthread_init (&__fork_generation, __reclaim_stacks,
 481                          ptr_pthread_functions);
 482
 483   /* Determine whether the machine is SMP or not.  */
 484   __is_smp = is_smp_system ();
 485 }
 486 strong_alias (__pthread_initialize_minimal_internal,
 487               __pthread_initialize_minimal)
 488
 489
 490 size_t
 491 __pthread_get_minstack (const pthread_attr_t *attr)
 492 {
 493   struct pthread_attr *iattr = (struct pthread_attr *) attr;
 494
 495   return (GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN
 496           + iattr->guardsize);
 497 }