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
  71 #ifdef SHARED
  72 static const struct pthread_functions pthread_functions =
  73   {
  74     .ptr_pthread_attr_destroy = __pthread_attr_destroy,
  75 # if SHLIB_COMPAT(libpthread, GLIBC_2_0, GLIBC_2_1)
  76     .ptr___pthread_attr_init_2_0 = __pthread_attr_init_2_0,
  77 # endif
  78     .ptr___pthread_attr_init_2_1 = __pthread_attr_init_2_1,
  79     .ptr_pthread_attr_getdetachstate = __pthread_attr_getdetachstate,
  80     .ptr_pthread_attr_setdetachstate = __pthread_attr_setdetachstate,
  81     .ptr_pthread_attr_getinheritsched = __pthread_attr_getinheritsched,
  82     .ptr_pthread_attr_setinheritsched = __pthread_attr_setinheritsched,
  83     .ptr_pthread_attr_getschedparam = __pthread_attr_getschedparam,
  84     .ptr_pthread_attr_setschedparam = __pthread_attr_setschedparam,
  85     .ptr_pthread_attr_getschedpolicy = __pthread_attr_getschedpolicy,
  86     .ptr_pthread_attr_setschedpolicy = __pthread_attr_setschedpolicy,
  87     .ptr_pthread_attr_getscope = __pthread_attr_getscope,
  88     .ptr_pthread_attr_setscope = __pthread_attr_setscope,
  89     .ptr_pthread_condattr_destroy = __pthread_condattr_destroy,
  90     .ptr_pthread_condattr_init = __pthread_condattr_init,
  91     .ptr___pthread_cond_broadcast = __pthread_cond_broadcast,
  92     .ptr___pthread_cond_destroy = __pthread_cond_destroy,
  93     .ptr___pthread_cond_init = __pthread_cond_init,
  94     .ptr___pthread_cond_signal = __pthread_cond_signal,
  95     .ptr___pthread_cond_wait = __pthread_cond_wait,
  96     .ptr___pthread_cond_timedwait = __pthread_cond_timedwait,
  97 # if SHLIB_COMPAT(libpthread, GLIBC_2_0, GLIBC_2_3_2)
  98     .ptr___pthread_cond_broadcast_2_0 = __pthread_cond_broadcast_2_0,
  99     .ptr___pthread_cond_destroy_2_0 = __pthread_cond_destroy_2_0,
 100     .ptr___pthread_cond_init_2_0 = __pthread_cond_init_2_0,
 101     .ptr___pthread_cond_signal_2_0 = __pthread_cond_signal_2_0,
 102     .ptr___pthread_cond_wait_2_0 = __pthread_cond_wait_2_0,
 103     .ptr___pthread_cond_timedwait_2_0 = __pthread_cond_timedwait_2_0,
 104 # endif
 105     .ptr_pthread_equal = __pthread_equal,
 106     .ptr___pthread_exit = __pthread_exit,
 107     .ptr_pthread_getschedparam = __pthread_getschedparam,
 108     .ptr_pthread_setschedparam = __pthread_setschedparam,
 109     .ptr_pthread_mutex_destroy = INTUSE(__pthread_mutex_destroy),
 110     .ptr_pthread_mutex_init = INTUSE(__pthread_mutex_init),
 111     .ptr_pthread_mutex_lock = INTUSE(__pthread_mutex_lock),
 112     .ptr_pthread_mutex_unlock = INTUSE(__pthread_mutex_unlock),
 113     .ptr_pthread_self = __pthread_self,
 114     .ptr_pthread_setcancelstate = __pthread_setcancelstate,
 115     .ptr_pthread_setcanceltype = __pthread_setcanceltype,
 116     .ptr___pthread_cleanup_upto = __pthread_cleanup_upto,
 117     .ptr___pthread_once = __pthread_once_internal,
 118     .ptr___pthread_rwlock_rdlock = __pthread_rwlock_rdlock_internal,
 119     .ptr___pthread_rwlock_wrlock = __pthread_rwlock_wrlock_internal,
 120     .ptr___pthread_rwlock_unlock = __pthread_rwlock_unlock_internal,
 121     .ptr___pthread_key_create = __pthread_key_create_internal,
 122     .ptr___pthread_getspecific = __pthread_getspecific_internal,
 123     .ptr___pthread_setspecific = __pthread_setspecific_internal,
 124     .ptr__pthread_cleanup_push_defer = __pthread_cleanup_push_defer,
 125     .ptr__pthread_cleanup_pop_restore = __pthread_cleanup_pop_restore,
 126     .ptr_nthreads = &__nptl_nthreads,
 127     .ptr___pthread_unwind = &__pthread_unwind,
 128     .ptr__nptl_deallocate_tsd = __nptl_deallocate_tsd,
 129     .ptr__nptl_setxid = __nptl_setxid,
 130     /* For now only the stack cache needs to be freed.  */
 131     .ptr_freeres = __free_stack_cache
 132   };
 133 # define ptr_pthread_functions &pthread_functions
 134 #else
 135 # define ptr_pthread_functions NULL
 136 #endif
 137
 138
 139 /* For asynchronous cancellation we use a signal.  This is the handler.  */
 140 static void
 141 sigcancel_handler (int sig, siginfo_t *si, void *ctx)
 142 {
 143 #ifdef __ASSUME_CORRECT_SI_PID
 144   /* Determine the process ID.  It might be negative if the thread is
 145      in the middle of a fork() call.  */
 146   pid_t pid = THREAD_GETMEM (THREAD_SELF, pid);
 147   if (__builtin_expect (pid < 0, 0))
 148     pid = -pid;
 149 #endif
 150
 151   /* Safety check.  It would be possible to call this function for
 152      other signals and send a signal from another process.  This is not
 153      correct and might even be a security problem.  Try to catch as
 154      many incorrect invocations as possible.  */
 155   if (sig != SIGCANCEL
 156 #ifdef __ASSUME_CORRECT_SI_PID
 157       /* Kernels before 2.5.75 stored the thread ID and not the process
 158          ID in si_pid so we skip this test.  */
 159       || si->si_pid != pid
 160 #endif
 161       || si->si_code != SI_TKILL)
 162     return;
 163
 164   struct pthread *self = THREAD_SELF;
 165
 166   int oldval = THREAD_GETMEM (self, cancelhandling);
 167   while (1)
 168     {
 169       /* We are canceled now.  When canceled by another thread this flag
 170          is already set but if the signal is directly send (internally or
 171          from another process) is has to be done here.  */
 172       int newval = oldval | CANCELING_BITMASK | CANCELED_BITMASK;
 173
 174       if (oldval == newval || (oldval & EXITING_BITMASK) != 0)
 175         /* Already canceled or exiting.  */
 176         break;
 177
 178       int curval = THREAD_ATOMIC_CMPXCHG_VAL (self, cancelhandling, newval,
 179                                               oldval);
 180       if (curval == oldval)
 181         {
 182           /* Set the return value.  */
 183           THREAD_SETMEM (self, result, PTHREAD_CANCELED);
 184
 185           /* Make sure asynchronous cancellation is still enabled.  */
 186           if ((newval & CANCELTYPE_BITMASK) != 0)
 187             /* Run the registered destructors and terminate the thread.  */
 188             __do_cancel ();
 189
 190           break;
 191         }
 192
 193       oldval = curval;
 194     }
 195 }
 196
 197
 198 struct xid_command *__xidcmd attribute_hidden;
 199
 200 /* For asynchronous cancellation we use a signal.  This is the handler.  */
 201 static void
 202 sighandler_setxid (int sig, siginfo_t *si, void *ctx)
 203 {
 204 #ifdef __ASSUME_CORRECT_SI_PID
 205   /* Determine the process ID.  It might be negative if the thread is
 206      in the middle of a fork() call.  */
 207   pid_t pid = THREAD_GETMEM (THREAD_SELF, pid);
 208   if (__builtin_expect (pid < 0, 0))
 209     pid = -pid;
 210 #endif
 211
 212   /* Safety check.  It would be possible to call this function for
 213      other signals and send a signal from another process.  This is not
 214      correct and might even be a security problem.  Try to catch as
 215      many incorrect invocations as possible.  */
 216   if (sig != SIGSETXID
 217 #ifdef __ASSUME_CORRECT_SI_PID
 218       /* Kernels before 2.5.75 stored the thread ID and not the process
 219          ID in si_pid so we skip this test.  */
 220       || si->si_pid != pid
 221 #endif
 222       || si->si_code != SI_TKILL)
 223     return;
 224
 225   INTERNAL_SYSCALL_DECL (err);
 226   INTERNAL_SYSCALL_NCS (__xidcmd->syscall_no, err, 3, __xidcmd->id[0],
 227                         __xidcmd->id[1], __xidcmd->id[2]);
 228
 229   if (atomic_decrement_val (&__xidcmd->cntr) == 0)
 230     lll_futex_wake (&__xidcmd->cntr, 1, LLL_PRIVATE);
 231
 232   /* Reset the SETXID flag.  */
 233   struct pthread *self = THREAD_SELF;
 234   int flags = THREAD_GETMEM (self, cancelhandling);
 235   THREAD_SETMEM (self, cancelhandling, flags & ~SETXID_BITMASK);
 236
 237   /* And release the futex.  */
 238   self->setxid_futex = 1;
 239   lll_futex_wake (&self->setxid_futex, 1, LLL_PRIVATE);
 240 }
 241
 242
 243 /* When using __thread for this, we do it in libc so as not
 244    to give libpthread its own TLS segment just for this.  */
 245 extern void **__libc_dl_error_tsd (void) __attribute__ ((const));
 246
 247
 248 /* This can be set by the debugger before initialization is complete.  */
 249 static bool __nptl_initial_report_events;
 250
 251 void
 252 __pthread_initialize_minimal_internal (void)
 253 {
 254 #ifndef SHARED
 255   /* Unlike in the dynamically linked case the dynamic linker has not
 256      taken care of initializing the TLS data structures.  */
 257   __libc_setup_tls (TLS_TCB_SIZE, TLS_TCB_ALIGN);
 258
 259   /* We must prevent gcc from being clever and move any of the
 260      following code ahead of the __libc_setup_tls call.  This function
 261      will initialize the thread register which is subsequently
 262      used.  */
 263   __asm __volatile ("");
 264 #endif
 265
 266   /* Minimal initialization of the thread descriptor.  */
 267   struct pthread *pd = THREAD_SELF;
 268   INTERNAL_SYSCALL_DECL (err);
 269   pd->pid = pd->tid = INTERNAL_SYSCALL (set_tid_address, err, 1, &pd->tid);
 270   THREAD_SETMEM (pd, specific[0], &pd->specific_1stblock[0]);
 271   THREAD_SETMEM (pd, user_stack, true);
 272   if (LLL_LOCK_INITIALIZER != 0)
 273     THREAD_SETMEM (pd, lock, LLL_LOCK_INITIALIZER);
 274 #if HP_TIMING_AVAIL
 275   THREAD_SETMEM (pd, cpuclock_offset, GL(dl_cpuclock_offset));
 276 #endif
 277
 278   /* Initialize the robust mutex data.  */
 279 #ifdef __PTHREAD_MUTEX_HAVE_PREV
 280   pd->robust_prev = &pd->robust_head;
 281 #endif
 282   pd->robust_head.list = &pd->robust_head;
 283 #ifdef __NR_set_robust_list
 284   pd->robust_head.futex_offset = (offsetof (pthread_mutex_t, __data.__lock)
 285                                   - offsetof (pthread_mutex_t,
 286                                               __data.__list.__next));
 287   int res = INTERNAL_SYSCALL (set_robust_list, err, 2, &pd->robust_head,
 288                               sizeof (struct robust_list_head));
 289   if (INTERNAL_SYSCALL_ERROR_P (res, err))
 290 #endif
 291     set_robust_list_not_avail ();
 292
 293 #ifndef __ASSUME_PRIVATE_FUTEX
 294   /* Private futexes are always used (at least internally) so that
 295      doing the test once this early is beneficial.  */
 296   {
 297     int word = 0;
 298     word = INTERNAL_SYSCALL (futex, err, 3, &word,
 299                             FUTEX_WAKE | FUTEX_PRIVATE_FLAG, 1);
 300     if (!INTERNAL_SYSCALL_ERROR_P (word, err))
 301       THREAD_SETMEM (pd, header.private_futex, FUTEX_PRIVATE_FLAG);
 302   }
 303 #endif
 304
 305 #ifndef __ASSUME_FUTEX_CLOCK_REALTIME
 306   {
 307     int word = 0;
 308     word = INTERNAL_SYSCALL (futex, err, 6, &word,
 309                              FUTEX_WAIT_BITSET | FUTEX_CLOCK_REALTIME
 310                              | FUTEX_PRIVATE_FLAG, 1, NULL, 0,
 311                              FUTEX_BITSET_MATCH_ANY);
 312     if (!INTERNAL_SYSCALL_ERROR_P (word, err)
 313         || (INTERNAL_SYSCALL_ERRNO (word, err) != ENOSYS
 314             && INTERNAL_SYSCALL_ERRNO (word, err) != EINVAL))
 315       __set_futex_clock_realtime ();
 316   }
 317 #endif
 318
 319   /* Set initial thread's stack block from 0 up to __libc_stack_end.
 320      It will be bigger than it actually is, but for unwind.c/pt-longjmp.c
 321      purposes this is good enough.  */
 322   THREAD_SETMEM (pd, stackblock_size, (size_t) __libc_stack_end);
 323
 324   /* Initialize the list of all running threads with the main thread.  */
 325   INIT_LIST_HEAD (&__stack_user);
 326   list_add (&pd->list, &__stack_user);
 327
 328   /* Before initializing __stack_user, the debugger could not find us and
 329      had to set __nptl_initial_report_events.  Propagate its setting.  */
 330   THREAD_SETMEM (pd, report_events, __nptl_initial_report_events);
 331
 332   /* Install the cancellation signal handler.  If for some reason we
 333      cannot install the handler we do not abort.  Maybe we should, but
 334      it is only asynchronous cancellation which is affected.  */
 335   struct sigaction sa;
 336   sa.sa_sigaction = sigcancel_handler;
 337   sa.sa_flags = SA_SIGINFO;
 338   __sigemptyset (&sa.sa_mask);
 339
 340   (void) __libc_sigaction (SIGCANCEL, &sa, NULL);
 341
 342   /* Install the handle to change the threads' uid/gid.  */
 343   sa.sa_sigaction = sighandler_setxid;
 344   sa.sa_flags = SA_SIGINFO | SA_RESTART;
 345
 346   (void) __libc_sigaction (SIGSETXID, &sa, NULL);
 347
 348   /* The parent process might have left the signals blocked.  Just in
 349      case, unblock it.  We reuse the signal mask in the sigaction
 350      structure.  It is already cleared.  */
 351   __sigaddset (&sa.sa_mask, SIGCANCEL);
 352   __sigaddset (&sa.sa_mask, SIGSETXID);
 353   (void) INTERNAL_SYSCALL (rt_sigprocmask, err, 4, SIG_UNBLOCK, &sa.sa_mask,
 354                            NULL, _NSIG / 8);
 355
 356   /* Get the size of the static and alignment requirements for the TLS
 357      block.  */
 358   size_t static_tls_align;
 359   _dl_get_tls_static_info (&__static_tls_size, &static_tls_align);
 360
 361   /* Make sure the size takes all the alignments into account.  */
 362   if (STACK_ALIGN > static_tls_align)
 363     static_tls_align = STACK_ALIGN;
 364   __static_tls_align_m1 = static_tls_align - 1;
 365
 366   __static_tls_size = roundup (__static_tls_size, static_tls_align);
 367
 368   /* Determine the default allowed stack size.  This is the size used
 369      in case the user does not specify one.  */
 370   struct rlimit limit;
 371   if (getrlimit (RLIMIT_STACK, &limit) != 0
 372       || limit.rlim_cur == RLIM_INFINITY)
 373     /* The system limit is not usable.  Use an architecture-specific
 374        default.  */
 375     limit.rlim_cur = ARCH_STACK_DEFAULT_SIZE;
 376   else if (limit.rlim_cur < PTHREAD_STACK_MIN)
 377     /* The system limit is unusably small.
 378        Use the minimal size acceptable.  */
 379     limit.rlim_cur = PTHREAD_STACK_MIN;
 380
 381   /* Make sure it meets the minimum size that allocate_stack
 382      (allocatestack.c) will demand, which depends on the page size.  */
 383   const uintptr_t pagesz = __sysconf (_SC_PAGESIZE);
 384   const size_t minstack = pagesz + __static_tls_size + MINIMAL_REST_STACK;
 385   if (limit.rlim_cur < minstack)
 386     limit.rlim_cur = minstack;
 387
 388   /* Round the resource limit up to page size.  */
 389   limit.rlim_cur = (limit.rlim_cur + pagesz - 1) & -pagesz;
 390   __default_stacksize = limit.rlim_cur;
 391
 392 #ifdef SHARED
 393   /* Transfer the old value from the dynamic linker's internal location.  */
 394   *__libc_dl_error_tsd () = *(*GL(dl_error_catch_tsd)) ();
 395   GL(dl_error_catch_tsd) = &__libc_dl_error_tsd;
 396
 397   /* Make __rtld_lock_{,un}lock_recursive use pthread_mutex_{,un}lock,
 398      keep the lock count from the ld.so implementation.  */
 399   GL(dl_rtld_lock_recursive) = (void *) INTUSE (__pthread_mutex_lock);
 400   GL(dl_rtld_unlock_recursive) = (void *) INTUSE (__pthread_mutex_unlock);
 401   unsigned int rtld_lock_count = GL(dl_load_lock).mutex.__data.__count;
 402   GL(dl_load_lock).mutex.__data.__count = 0;
 403   while (rtld_lock_count-- > 0)
 404     INTUSE (__pthread_mutex_lock) (&GL(dl_load_lock).mutex);
 405
 406   GL(dl_make_stack_executable_hook) = &__make_stacks_executable;
 407 #endif
 408
 409   GL(dl_init_static_tls) = &__pthread_init_static_tls;
 410
 411   GL(dl_wait_lookup_done) = &__wait_lookup_done;
 412
 413   /* Register the fork generation counter with the libc.  */
 414 #ifndef TLS_MULTIPLE_THREADS_IN_TCB
 415   __libc_multiple_threads_ptr =
 416 #endif
 417     __libc_pthread_init (&__fork_generation, __reclaim_stacks,
 418                          ptr_pthread_functions);
 419
 420   /* Determine whether the machine is SMP or not.  */
 421   __is_smp = is_smp_system ();
 422 }
 423 strong_alias (__pthread_initialize_minimal_internal,
 424               __pthread_initialize_minimal)