nptl/nptl-init.c

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