nptl/nptl-init.c

   1 /* Copyright (C) 2002-2015 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 <futex-internal.h>
  38 #include <kernel-features.h>
  39 #include <libc-internal.h>
  40 #include <pthread-pids.h>
  41
  42 #ifndef TLS_MULTIPLE_THREADS_IN_TCB
  43 /* Pointer to the corresponding variable in libc.  */
  44 int *__libc_multiple_threads_ptr attribute_hidden;
  45 #endif
  46
  47 /* Size and alignment of static TLS block.  */
  48 size_t __static_tls_size;
  49 size_t __static_tls_align_m1;
  50
  51 #ifndef __ASSUME_SET_ROBUST_LIST
  52 /* Negative if we do not have the system call and we can use it.  */
  53 int __set_robust_list_avail;
  54 # define set_robust_list_not_avail() \
  55   __set_robust_list_avail = -1
  56 #else
  57 # define set_robust_list_not_avail() do { } while (0)
  58 #endif
  59
  60 #ifndef __ASSUME_FUTEX_CLOCK_REALTIME
  61 /* Nonzero if we do not have FUTEX_CLOCK_REALTIME.  */
  62 int __have_futex_clock_realtime;
  63 # define __set_futex_clock_realtime() \
  64   __have_futex_clock_realtime = 1
  65 #else
  66 #define __set_futex_clock_realtime() do { } while (0)
  67 #endif
  68
  69 /* Version of the library, used in libthread_db to detect mismatches.  */
  70 static const char nptl_version[] __attribute_used__ = VERSION;
  71
  72
  73 #ifndef SHARED
  74 extern void __libc_setup_tls (size_t tcbsize, size_t tcbalign);
  75 #endif
  76
  77 #ifdef SHARED
  78 static
  79 #else
  80 extern
  81 #endif
  82 void __nptl_set_robust (struct pthread *);
  83
  84 #ifdef SHARED
  85 static void nptl_freeres (void);
  86
  87
  88 static const struct pthread_functions pthread_functions =
  89   {
  90     .ptr_pthread_attr_destroy = __pthread_attr_destroy,
  91 # if SHLIB_COMPAT(libpthread, GLIBC_2_0, GLIBC_2_1)
  92     .ptr___pthread_attr_init_2_0 = __pthread_attr_init_2_0,
  93 # endif
  94     .ptr___pthread_attr_init_2_1 = __pthread_attr_init_2_1,
  95     .ptr_pthread_attr_getdetachstate = __pthread_attr_getdetachstate,
  96     .ptr_pthread_attr_setdetachstate = __pthread_attr_setdetachstate,
  97     .ptr_pthread_attr_getinheritsched = __pthread_attr_getinheritsched,
  98     .ptr_pthread_attr_setinheritsched = __pthread_attr_setinheritsched,
  99     .ptr_pthread_attr_getschedparam = __pthread_attr_getschedparam,
 100     .ptr_pthread_attr_setschedparam = __pthread_attr_setschedparam,
 101     .ptr_pthread_attr_getschedpolicy = __pthread_attr_getschedpolicy,
 102     .ptr_pthread_attr_setschedpolicy = __pthread_attr_setschedpolicy,
 103     .ptr_pthread_attr_getscope = __pthread_attr_getscope,
 104     .ptr_pthread_attr_setscope = __pthread_attr_setscope,
 105     .ptr_pthread_condattr_destroy = __pthread_condattr_destroy,
 106     .ptr_pthread_condattr_init = __pthread_condattr_init,
 107     .ptr___pthread_cond_broadcast = __pthread_cond_broadcast,
 108     .ptr___pthread_cond_destroy = __pthread_cond_destroy,
 109     .ptr___pthread_cond_init = __pthread_cond_init,
 110     .ptr___pthread_cond_signal = __pthread_cond_signal,
 111     .ptr___pthread_cond_wait = __pthread_cond_wait,
 112     .ptr___pthread_cond_timedwait = __pthread_cond_timedwait,
 113 # if SHLIB_COMPAT(libpthread, GLIBC_2_0, GLIBC_2_3_2)
 114     .ptr___pthread_cond_broadcast_2_0 = __pthread_cond_broadcast_2_0,
 115     .ptr___pthread_cond_destroy_2_0 = __pthread_cond_destroy_2_0,
 116     .ptr___pthread_cond_init_2_0 = __pthread_cond_init_2_0,
 117     .ptr___pthread_cond_signal_2_0 = __pthread_cond_signal_2_0,
 118     .ptr___pthread_cond_wait_2_0 = __pthread_cond_wait_2_0,
 119     .ptr___pthread_cond_timedwait_2_0 = __pthread_cond_timedwait_2_0,
 120 # endif
 121     .ptr_pthread_equal = __pthread_equal,
 122     .ptr___pthread_exit = __pthread_exit,
 123     .ptr_pthread_getschedparam = __pthread_getschedparam,
 124     .ptr_pthread_setschedparam = __pthread_setschedparam,
 125     .ptr_pthread_mutex_destroy = __pthread_mutex_destroy,
 126     .ptr_pthread_mutex_init = __pthread_mutex_init,
 127     .ptr_pthread_mutex_lock = __pthread_mutex_lock,
 128     .ptr_pthread_mutex_unlock = __pthread_mutex_unlock,
 129     .ptr_pthread_self = __pthread_self,
 130     .ptr___pthread_setcancelstate = __pthread_setcancelstate,
 131     .ptr_pthread_setcanceltype = __pthread_setcanceltype,
 132     .ptr___pthread_cleanup_upto = __pthread_cleanup_upto,
 133     .ptr___pthread_once = __pthread_once,
 134     .ptr___pthread_rwlock_rdlock = __pthread_rwlock_rdlock,
 135     .ptr___pthread_rwlock_wrlock = __pthread_rwlock_wrlock,
 136     .ptr___pthread_rwlock_unlock = __pthread_rwlock_unlock,
 137     .ptr___pthread_key_create = __pthread_key_create,
 138     .ptr___pthread_getspecific = __pthread_getspecific,
 139     .ptr___pthread_setspecific = __pthread_setspecific,
 140     .ptr__pthread_cleanup_push_defer = __pthread_cleanup_push_defer,
 141     .ptr__pthread_cleanup_pop_restore = __pthread_cleanup_pop_restore,
 142     .ptr_nthreads = &__nptl_nthreads,
 143     .ptr___pthread_unwind = &__pthread_unwind,
 144     .ptr__nptl_deallocate_tsd = __nptl_deallocate_tsd,
 145 # ifdef SIGSETXID
 146     .ptr__nptl_setxid = __nptl_setxid,
 147 # endif
 148     /* For now only the stack cache needs to be freed.  */
 149     .ptr_freeres = nptl_freeres,
 150     .ptr_set_robust = __nptl_set_robust
 151   };
 152 # define ptr_pthread_functions &pthread_functions
 153 #else
 154 # define ptr_pthread_functions NULL
 155 #endif
 156
 157
 158 #ifdef SHARED
 159 /* This function is called indirectly from the freeres code in libc.  */
 160 static void
 161 __libc_freeres_fn_section
 162 nptl_freeres (void)
 163 {
 164   __unwind_freeres ();
 165   __free_stacks (0);
 166 }
 167
 168
 169 static
 170 #endif
 171 void
 172 __nptl_set_robust (struct pthread *self)
 173 {
 174 #ifdef __NR_set_robust_list
 175   INTERNAL_SYSCALL_DECL (err);
 176   INTERNAL_SYSCALL (set_robust_list, err, 2, &self->robust_head,
 177                     sizeof (struct robust_list_head));
 178 #endif
 179 }
 180
 181
 182 #ifdef SIGCANCEL
 183 /* For asynchronous cancellation we use a signal.  This is the handler.  */
 184 static void
 185 sigcancel_handler (int sig, siginfo_t *si, void *ctx)
 186 {
 187   /* Determine the process ID.  It might be negative if the thread is
 188      in the middle of a fork() call.  */
 189   pid_t pid = THREAD_GETMEM (THREAD_SELF, pid);
 190   if (__glibc_unlikely (pid < 0))
 191     pid = -pid;
 192
 193   /* Safety check.  It would be possible to call this function for
 194      other signals and send a signal from another process.  This is not
 195      correct and might even be a security problem.  Try to catch as
 196      many incorrect invocations as possible.  */
 197   if (sig != SIGCANCEL
 198       || si->si_pid != pid
 199       || si->si_code != SI_TKILL)
 200     return;
 201
 202   struct pthread *self = THREAD_SELF;
 203
 204   int oldval = THREAD_GETMEM (self, cancelhandling);
 205   while (1)
 206     {
 207       /* We are canceled now.  When canceled by another thread this flag
 208          is already set but if the signal is directly send (internally or
 209          from another process) is has to be done here.  */
 210       int newval = oldval | CANCELING_BITMASK | CANCELED_BITMASK;
 211
 212       if (oldval == newval || (oldval & EXITING_BITMASK) != 0)
 213         /* Already canceled or exiting.  */
 214         break;
 215
 216       int curval = THREAD_ATOMIC_CMPXCHG_VAL (self, cancelhandling, newval,
 217                                               oldval);
 218       if (curval == oldval)
 219         {
 220           /* Set the return value.  */
 221           THREAD_SETMEM (self, result, PTHREAD_CANCELED);
 222
 223           /* Make sure asynchronous cancellation is still enabled.  */
 224           if ((newval & CANCELTYPE_BITMASK) != 0)
 225             /* Run the registered destructors and terminate the thread.  */
 226             __do_cancel ();
 227
 228           break;
 229         }
 230
 231       oldval = curval;
 232     }
 233 }
 234 #endif
 235
 236
 237 #ifdef SIGSETXID
 238 struct xid_command *__xidcmd attribute_hidden;
 239
 240 /* We use the SIGSETXID signal in the setuid, setgid, etc. implementations to
 241    tell each thread to call the respective setxid syscall on itself.  This is
 242    the handler.  */
 243 static void
 244 sighandler_setxid (int sig, siginfo_t *si, void *ctx)
 245 {
 246   /* Determine the process ID.  It might be negative if the thread is
 247      in the middle of a fork() call.  */
 248   pid_t pid = THREAD_GETMEM (THREAD_SELF, pid);
 249   int result;
 250   if (__glibc_unlikely (pid < 0))
 251     pid = -pid;
 252
 253   /* Safety check.  It would be possible to call this function for
 254      other signals and send a signal from another process.  This is not
 255      correct and might even be a security problem.  Try to catch as
 256      many incorrect invocations as possible.  */
 257   if (sig != SIGSETXID
 258       || si->si_pid != pid
 259       || si->si_code != SI_TKILL)
 260     return;
 261
 262   INTERNAL_SYSCALL_DECL (err);
 263   result = INTERNAL_SYSCALL_NCS (__xidcmd->syscall_no, err, 3, __xidcmd->id[0],
 264                                  __xidcmd->id[1], __xidcmd->id[2]);
 265   int error = 0;
 266   if (__glibc_unlikely (INTERNAL_SYSCALL_ERROR_P (result, err)))
 267     error = INTERNAL_SYSCALL_ERRNO (result, err);
 268   __nptl_setxid_error (__xidcmd, error);
 269
 270   /* Reset the SETXID flag.  */
 271   struct pthread *self = THREAD_SELF;
 272   int flags, newval;
 273   do
 274     {
 275       flags = THREAD_GETMEM (self, cancelhandling);
 276       newval = THREAD_ATOMIC_CMPXCHG_VAL (self, cancelhandling,
 277                                           flags & ~SETXID_BITMASK, flags);
 278     }
 279   while (flags != newval);
 280
 281   /* And release the futex.  */
 282   self->setxid_futex = 1;
 283   futex_wake (&self->setxid_futex, 1, FUTEX_PRIVATE);
 284
 285   if (atomic_decrement_val (&__xidcmd->cntr) == 0)
 286     futex_wake ((unsigned int *) &__xidcmd->cntr, 1, FUTEX_PRIVATE);
 287 }
 288 #endif
 289
 290
 291 /* When using __thread for this, we do it in libc so as not
 292    to give libpthread its own TLS segment just for this.  */
 293 extern void **__libc_dl_error_tsd (void) __attribute__ ((const));
 294
 295
 296 /* This can be set by the debugger before initialization is complete.  */
 297 static bool __nptl_initial_report_events __attribute_used__;
 298
 299 void
 300 __pthread_initialize_minimal_internal (void)
 301 {
 302 #ifndef SHARED
 303   /* Unlike in the dynamically linked case the dynamic linker has not
 304      taken care of initializing the TLS data structures.  */
 305   __libc_setup_tls (TLS_TCB_SIZE, TLS_TCB_ALIGN);
 306
 307   /* We must prevent gcc from being clever and move any of the
 308      following code ahead of the __libc_setup_tls call.  This function
 309      will initialize the thread register which is subsequently
 310      used.  */
 311   __asm __volatile ("");
 312 #endif
 313
 314   /* Minimal initialization of the thread descriptor.  */
 315   struct pthread *pd = THREAD_SELF;
 316   __pthread_initialize_pids (pd);
 317   THREAD_SETMEM (pd, specific[0], &pd->specific_1stblock[0]);
 318   THREAD_SETMEM (pd, user_stack, true);
 319   if (LLL_LOCK_INITIALIZER != 0)
 320     THREAD_SETMEM (pd, lock, LLL_LOCK_INITIALIZER);
 321 #if HP_TIMING_AVAIL
 322   THREAD_SETMEM (pd, cpuclock_offset, GL(dl_cpuclock_offset));
 323 #endif
 324
 325   /* Initialize the robust mutex data.  */
 326   {
 327 #ifdef __PTHREAD_MUTEX_HAVE_PREV
 328     pd->robust_prev = &pd->robust_head;
 329 #endif
 330     pd->robust_head.list = &pd->robust_head;
 331 #ifdef __NR_set_robust_list
 332     pd->robust_head.futex_offset = (offsetof (pthread_mutex_t, __data.__lock)
 333                                     - offsetof (pthread_mutex_t,
 334                                                 __data.__list.__next));
 335     INTERNAL_SYSCALL_DECL (err);
 336     int res = INTERNAL_SYSCALL (set_robust_list, err, 2, &pd->robust_head,
 337                                 sizeof (struct robust_list_head));
 338     if (INTERNAL_SYSCALL_ERROR_P (res, err))
 339 #endif
 340       set_robust_list_not_avail ();
 341   }
 342
 343 #ifdef __NR_futex
 344 # ifndef __ASSUME_PRIVATE_FUTEX
 345   /* Private futexes are always used (at least internally) so that
 346      doing the test once this early is beneficial.  */
 347   {
 348     int word = 0;
 349     INTERNAL_SYSCALL_DECL (err);
 350     word = INTERNAL_SYSCALL (futex, err, 3, &word,
 351                             FUTEX_WAKE | FUTEX_PRIVATE_FLAG, 1);
 352     if (!INTERNAL_SYSCALL_ERROR_P (word, err))
 353       THREAD_SETMEM (pd, header.private_futex, FUTEX_PRIVATE_FLAG);
 354   }
 355
 356   /* Private futexes have been introduced earlier than the
 357      FUTEX_CLOCK_REALTIME flag.  We don't have to run the test if we
 358      know the former are not supported.  This also means we know the
 359      kernel will return ENOSYS for unknown operations.  */
 360   if (THREAD_GETMEM (pd, header.private_futex) != 0)
 361 # endif
 362 # ifndef __ASSUME_FUTEX_CLOCK_REALTIME
 363     {
 364       int word = 0;
 365       /* NB: the syscall actually takes six parameters.  The last is the
 366          bit mask.  But since we will not actually wait at all the value
 367          is irrelevant.  Given that passing six parameters is difficult
 368          on some architectures we just pass whatever random value the
 369          calling convention calls for to the kernel.  It causes no harm.  */
 370       INTERNAL_SYSCALL_DECL (err);
 371       word = INTERNAL_SYSCALL (futex, err, 5, &word,
 372                                FUTEX_WAIT_BITSET | FUTEX_CLOCK_REALTIME
 373                                | FUTEX_PRIVATE_FLAG, 1, NULL, 0);
 374       assert (INTERNAL_SYSCALL_ERROR_P (word, err));
 375       if (INTERNAL_SYSCALL_ERRNO (word, err) != ENOSYS)
 376         __set_futex_clock_realtime ();
 377     }
 378 # endif
 379 #endif
 380
 381   /* Set initial thread's stack block from 0 up to __libc_stack_end.
 382      It will be bigger than it actually is, but for unwind.c/pt-longjmp.c
 383      purposes this is good enough.  */
 384   THREAD_SETMEM (pd, stackblock_size, (size_t) __libc_stack_end);
 385
 386   /* Initialize the list of all running threads with the main thread.  */
 387   INIT_LIST_HEAD (&__stack_user);
 388   list_add (&pd->list, &__stack_user);
 389
 390   /* Before initializing __stack_user, the debugger could not find us and
 391      had to set __nptl_initial_report_events.  Propagate its setting.  */
 392   THREAD_SETMEM (pd, report_events, __nptl_initial_report_events);
 393
 394 #if defined SIGCANCEL || defined SIGSETXID
 395   struct sigaction sa;
 396   __sigemptyset (&sa.sa_mask);
 397
 398 # ifdef SIGCANCEL
 399   /* Install the cancellation signal handler.  If for some reason we
 400      cannot install the handler we do not abort.  Maybe we should, but
 401      it is only asynchronous cancellation which is affected.  */
 402   sa.sa_sigaction = sigcancel_handler;
 403   sa.sa_flags = SA_SIGINFO;
 404   (void) __libc_sigaction (SIGCANCEL, &sa, NULL);
 405 # endif
 406
 407 # ifdef SIGSETXID
 408   /* Install the handle to change the threads' uid/gid.  */
 409   sa.sa_sigaction = sighandler_setxid;
 410   sa.sa_flags = SA_SIGINFO | SA_RESTART;
 411   (void) __libc_sigaction (SIGSETXID, &sa, NULL);
 412 # endif
 413
 414   /* The parent process might have left the signals blocked.  Just in
 415      case, unblock it.  We reuse the signal mask in the sigaction
 416      structure.  It is already cleared.  */
 417 # ifdef SIGCANCEL
 418   __sigaddset (&sa.sa_mask, SIGCANCEL);
 419 # endif
 420 # ifdef SIGSETXID
 421   __sigaddset (&sa.sa_mask, SIGSETXID);
 422 # endif
 423   {
 424     INTERNAL_SYSCALL_DECL (err);
 425     (void) INTERNAL_SYSCALL (rt_sigprocmask, err, 4, SIG_UNBLOCK, &sa.sa_mask,
 426                              NULL, _NSIG / 8);
 427   }
 428 #endif
 429
 430   /* Get the size of the static and alignment requirements for the TLS
 431      block.  */
 432   size_t static_tls_align;
 433   _dl_get_tls_static_info (&__static_tls_size, &static_tls_align);
 434
 435   /* Make sure the size takes all the alignments into account.  */
 436   if (STACK_ALIGN > static_tls_align)
 437     static_tls_align = STACK_ALIGN;
 438   __static_tls_align_m1 = static_tls_align - 1;
 439
 440   __static_tls_size = roundup (__static_tls_size, static_tls_align);
 441
 442   /* Determine the default allowed stack size.  This is the size used
 443      in case the user does not specify one.  */
 444   struct rlimit limit;
 445   if (__getrlimit (RLIMIT_STACK, &limit) != 0
 446       || limit.rlim_cur == RLIM_INFINITY)
 447     /* The system limit is not usable.  Use an architecture-specific
 448        default.  */
 449     limit.rlim_cur = ARCH_STACK_DEFAULT_SIZE;
 450   else if (limit.rlim_cur < PTHREAD_STACK_MIN)
 451     /* The system limit is unusably small.
 452        Use the minimal size acceptable.  */
 453     limit.rlim_cur = PTHREAD_STACK_MIN;
 454
 455   /* Make sure it meets the minimum size that allocate_stack
 456      (allocatestack.c) will demand, which depends on the page size.  */
 457   const uintptr_t pagesz = GLRO(dl_pagesize);
 458   const size_t minstack = pagesz + __static_tls_size + MINIMAL_REST_STACK;
 459   if (limit.rlim_cur < minstack)
 460     limit.rlim_cur = minstack;
 461
 462   /* Round the resource limit up to page size.  */
 463   limit.rlim_cur = ALIGN_UP (limit.rlim_cur, pagesz);
 464   lll_lock (__default_pthread_attr_lock, LLL_PRIVATE);
 465   __default_pthread_attr.stacksize = limit.rlim_cur;
 466   __default_pthread_attr.guardsize = GLRO (dl_pagesize);
 467   lll_unlock (__default_pthread_attr_lock, LLL_PRIVATE);
 468
 469 #ifdef SHARED
 470   /* Transfer the old value from the dynamic linker's internal location.  */
 471   *__libc_dl_error_tsd () = *(*GL(dl_error_catch_tsd)) ();
 472   GL(dl_error_catch_tsd) = &__libc_dl_error_tsd;
 473
 474   /* Make __rtld_lock_{,un}lock_recursive use pthread_mutex_{,un}lock,
 475      keep the lock count from the ld.so implementation.  */
 476   GL(dl_rtld_lock_recursive) = (void *) __pthread_mutex_lock;
 477   GL(dl_rtld_unlock_recursive) = (void *) __pthread_mutex_unlock;
 478   unsigned int rtld_lock_count = GL(dl_load_lock).mutex.__data.__count;
 479   GL(dl_load_lock).mutex.__data.__count = 0;
 480   while (rtld_lock_count-- > 0)
 481     __pthread_mutex_lock (&GL(dl_load_lock).mutex);
 482
 483   GL(dl_make_stack_executable_hook) = &__make_stacks_executable;
 484 #endif
 485
 486   GL(dl_init_static_tls) = &__pthread_init_static_tls;
 487
 488   GL(dl_wait_lookup_done) = &__wait_lookup_done;
 489
 490   /* Register the fork generation counter with the libc.  */
 491 #ifndef TLS_MULTIPLE_THREADS_IN_TCB
 492   __libc_multiple_threads_ptr =
 493 #endif
 494     __libc_pthread_init (&__fork_generation, __reclaim_stacks,
 495                          ptr_pthread_functions);
 496
 497   /* Determine whether the machine is SMP or not.  */
 498   __is_smp = is_smp_system ();
 499 }
 500 strong_alias (__pthread_initialize_minimal_internal,
 501               __pthread_initialize_minimal)
 502
 503
 504 size_t
 505 __pthread_get_minstack (const pthread_attr_t *attr)
 506 {
 507   struct pthread_attr *iattr = (struct pthread_attr *) attr;
 508
 509   return (GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN
 510           + iattr->guardsize);
 511 }