nptl/nptl-init.c

   1 /* Copyright (C) 2002-2012 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   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   /* Determine the process ID.  It might be negative if the thread is
 176      in the middle of a fork() call.  */
 177   pid_t pid = THREAD_GETMEM (THREAD_SELF, pid);
 178   if (__builtin_expect (pid < 0, 0))
 179     pid = -pid;
 180
 181   /* Safety check.  It would be possible to call this function for
 182      other signals and send a signal from another process.  This is not
 183      correct and might even be a security problem.  Try to catch as
 184      many incorrect invocations as possible.  */
 185   if (sig != SIGCANCEL
 186       || si->si_pid != pid
 187       || si->si_code != SI_TKILL)
 188     return;
 189
 190   struct pthread *self = THREAD_SELF;
 191
 192   int oldval = THREAD_GETMEM (self, cancelhandling);
 193   while (1)
 194     {
 195       /* We are canceled now.  When canceled by another thread this flag
 196          is already set but if the signal is directly send (internally or
 197          from another process) is has to be done here.  */
 198       int newval = oldval | CANCELING_BITMASK | CANCELED_BITMASK;
 199
 200       if (oldval == newval || (oldval & EXITING_BITMASK) != 0)
 201         /* Already canceled or exiting.  */
 202         break;
 203
 204       int curval = THREAD_ATOMIC_CMPXCHG_VAL (self, cancelhandling, newval,
 205                                               oldval);
 206       if (curval == oldval)
 207         {
 208           /* Set the return value.  */
 209           THREAD_SETMEM (self, result, PTHREAD_CANCELED);
 210
 211           /* Make sure asynchronous cancellation is still enabled.  */
 212           if ((newval & CANCELTYPE_BITMASK) != 0)
 213             /* Run the registered destructors and terminate the thread.  */
 214             __do_cancel ();
 215
 216           break;
 217         }
 218
 219       oldval = curval;
 220     }
 221 }
 222
 223
 224 struct xid_command *__xidcmd attribute_hidden;
 225
 226 /* We use the SIGSETXID signal in the setuid, setgid, etc. implementations to
 227    tell each thread to call the respective setxid syscall on itself.  This is
 228    the handler.  */
 229 static void
 230 sighandler_setxid (int sig, siginfo_t *si, void *ctx)
 231 {
 232   /* Determine the process ID.  It might be negative if the thread is
 233      in the middle of a fork() call.  */
 234   pid_t pid = THREAD_GETMEM (THREAD_SELF, pid);
 235   if (__builtin_expect (pid < 0, 0))
 236     pid = -pid;
 237
 238   /* Safety check.  It would be possible to call this function for
 239      other signals and send a signal from another process.  This is not
 240      correct and might even be a security problem.  Try to catch as
 241      many incorrect invocations as possible.  */
 242   if (sig != SIGSETXID
 243       || si->si_pid != pid
 244       || si->si_code != SI_TKILL)
 245     return;
 246
 247   INTERNAL_SYSCALL_DECL (err);
 248   INTERNAL_SYSCALL_NCS (__xidcmd->syscall_no, err, 3, __xidcmd->id[0],
 249                         __xidcmd->id[1], __xidcmd->id[2]);
 250
 251   /* Reset the SETXID flag.  */
 252   struct pthread *self = THREAD_SELF;
 253   int flags, newval;
 254   do
 255     {
 256       flags = THREAD_GETMEM (self, cancelhandling);
 257       newval = THREAD_ATOMIC_CMPXCHG_VAL (self, cancelhandling,
 258                                           flags & ~SETXID_BITMASK, flags);
 259     }
 260   while (flags != newval);
 261
 262   /* And release the futex.  */
 263   self->setxid_futex = 1;
 264   lll_futex_wake (&self->setxid_futex, 1, LLL_PRIVATE);
 265
 266   if (atomic_decrement_val (&__xidcmd->cntr) == 0)
 267     lll_futex_wake (&__xidcmd->cntr, 1, LLL_PRIVATE);
 268 }
 269
 270
 271 /* When using __thread for this, we do it in libc so as not
 272    to give libpthread its own TLS segment just for this.  */
 273 extern void **__libc_dl_error_tsd (void) __attribute__ ((const));
 274
 275
 276 /* This can be set by the debugger before initialization is complete.  */
 277 static bool __nptl_initial_report_events __attribute_used__;
 278
 279 void
 280 __pthread_initialize_minimal_internal (void)
 281 {
 282 #ifndef SHARED
 283   /* Unlike in the dynamically linked case the dynamic linker has not
 284      taken care of initializing the TLS data structures.  */
 285   __libc_setup_tls (TLS_TCB_SIZE, TLS_TCB_ALIGN);
 286
 287   /* We must prevent gcc from being clever and move any of the
 288      following code ahead of the __libc_setup_tls call.  This function
 289      will initialize the thread register which is subsequently
 290      used.  */
 291   __asm __volatile ("");
 292 #endif
 293
 294   /* Minimal initialization of the thread descriptor.  */
 295   struct pthread *pd = THREAD_SELF;
 296   INTERNAL_SYSCALL_DECL (err);
 297   pd->pid = pd->tid = INTERNAL_SYSCALL (set_tid_address, err, 1, &pd->tid);
 298   THREAD_SETMEM (pd, specific[0], &pd->specific_1stblock[0]);
 299   THREAD_SETMEM (pd, user_stack, true);
 300   if (LLL_LOCK_INITIALIZER != 0)
 301     THREAD_SETMEM (pd, lock, LLL_LOCK_INITIALIZER);
 302 #if HP_TIMING_AVAIL
 303   THREAD_SETMEM (pd, cpuclock_offset, GL(dl_cpuclock_offset));
 304 #endif
 305
 306   /* Initialize the robust mutex data.  */
 307 #ifdef __PTHREAD_MUTEX_HAVE_PREV
 308   pd->robust_prev = &pd->robust_head;
 309 #endif
 310   pd->robust_head.list = &pd->robust_head;
 311 #ifdef __NR_set_robust_list
 312   pd->robust_head.futex_offset = (offsetof (pthread_mutex_t, __data.__lock)
 313                                   - offsetof (pthread_mutex_t,
 314                                               __data.__list.__next));
 315   int res = INTERNAL_SYSCALL (set_robust_list, err, 2, &pd->robust_head,
 316                               sizeof (struct robust_list_head));
 317   if (INTERNAL_SYSCALL_ERROR_P (res, err))
 318 #endif
 319     set_robust_list_not_avail ();
 320
 321 #ifndef __ASSUME_PRIVATE_FUTEX
 322   /* Private futexes are always used (at least internally) so that
 323      doing the test once this early is beneficial.  */
 324   {
 325     int word = 0;
 326     word = INTERNAL_SYSCALL (futex, err, 3, &word,
 327                             FUTEX_WAKE | FUTEX_PRIVATE_FLAG, 1);
 328     if (!INTERNAL_SYSCALL_ERROR_P (word, err))
 329       THREAD_SETMEM (pd, header.private_futex, FUTEX_PRIVATE_FLAG);
 330   }
 331
 332   /* Private futexes have been introduced earlier than the
 333      FUTEX_CLOCK_REALTIME flag.  We don't have to run the test if we
 334      know the former are not supported.  This also means we know the
 335      kernel will return ENOSYS for unknown operations.  */
 336   if (THREAD_GETMEM (pd, header.private_futex) != 0)
 337 #endif
 338 #ifndef __ASSUME_FUTEX_CLOCK_REALTIME
 339     {
 340       int word = 0;
 341       /* NB: the syscall actually takes six parameters.  The last is the
 342          bit mask.  But since we will not actually wait at all the value
 343          is irrelevant.  Given that passing six parameters is difficult
 344          on some architectures we just pass whatever random value the
 345          calling convention calls for to the kernel.  It causes no harm.  */
 346       word = INTERNAL_SYSCALL (futex, err, 5, &word,
 347                                FUTEX_WAIT_BITSET | FUTEX_CLOCK_REALTIME
 348                                | FUTEX_PRIVATE_FLAG, 1, NULL, 0);
 349       assert (INTERNAL_SYSCALL_ERROR_P (word, err));
 350       if (INTERNAL_SYSCALL_ERRNO (word, err) != ENOSYS)
 351         __set_futex_clock_realtime ();
 352     }
 353 #endif
 354
 355   /* Set initial thread's stack block from 0 up to __libc_stack_end.
 356      It will be bigger than it actually is, but for unwind.c/pt-longjmp.c
 357      purposes this is good enough.  */
 358   THREAD_SETMEM (pd, stackblock_size, (size_t) __libc_stack_end);
 359
 360   /* Initialize the list of all running threads with the main thread.  */
 361   INIT_LIST_HEAD (&__stack_user);
 362   list_add (&pd->list, &__stack_user);
 363
 364   /* Before initializing __stack_user, the debugger could not find us and
 365      had to set __nptl_initial_report_events.  Propagate its setting.  */
 366   THREAD_SETMEM (pd, report_events, __nptl_initial_report_events);
 367
 368   /* Install the cancellation signal handler.  If for some reason we
 369      cannot install the handler we do not abort.  Maybe we should, but
 370      it is only asynchronous cancellation which is affected.  */
 371   struct sigaction sa;
 372   sa.sa_sigaction = sigcancel_handler;
 373   sa.sa_flags = SA_SIGINFO;
 374   __sigemptyset (&sa.sa_mask);
 375
 376   (void) __libc_sigaction (SIGCANCEL, &sa, NULL);
 377
 378   /* Install the handle to change the threads' uid/gid.  */
 379   sa.sa_sigaction = sighandler_setxid;
 380   sa.sa_flags = SA_SIGINFO | SA_RESTART;
 381
 382   (void) __libc_sigaction (SIGSETXID, &sa, NULL);
 383
 384   /* The parent process might have left the signals blocked.  Just in
 385      case, unblock it.  We reuse the signal mask in the sigaction
 386      structure.  It is already cleared.  */
 387   __sigaddset (&sa.sa_mask, SIGCANCEL);
 388   __sigaddset (&sa.sa_mask, SIGSETXID);
 389   (void) INTERNAL_SYSCALL (rt_sigprocmask, err, 4, SIG_UNBLOCK, &sa.sa_mask,
 390                            NULL, _NSIG / 8);
 391
 392   /* Get the size of the static and alignment requirements for the TLS
 393      block.  */
 394   size_t static_tls_align;
 395   _dl_get_tls_static_info (&__static_tls_size, &static_tls_align);
 396
 397   /* Make sure the size takes all the alignments into account.  */
 398   if (STACK_ALIGN > static_tls_align)
 399     static_tls_align = STACK_ALIGN;
 400   __static_tls_align_m1 = static_tls_align - 1;
 401
 402   __static_tls_size = roundup (__static_tls_size, static_tls_align);
 403
 404   /* Determine the default allowed stack size.  This is the size used
 405      in case the user does not specify one.  */
 406   struct rlimit limit;
 407   if (getrlimit (RLIMIT_STACK, &limit) != 0
 408       || limit.rlim_cur == RLIM_INFINITY)
 409     /* The system limit is not usable.  Use an architecture-specific
 410        default.  */
 411     limit.rlim_cur = ARCH_STACK_DEFAULT_SIZE;
 412   else if (limit.rlim_cur < PTHREAD_STACK_MIN)
 413     /* The system limit is unusably small.
 414        Use the minimal size acceptable.  */
 415     limit.rlim_cur = PTHREAD_STACK_MIN;
 416
 417   /* Make sure it meets the minimum size that allocate_stack
 418      (allocatestack.c) will demand, which depends on the page size.  */
 419   const uintptr_t pagesz = GLRO(dl_pagesize);
 420   const size_t minstack = pagesz + __static_tls_size + MINIMAL_REST_STACK;
 421   if (limit.rlim_cur < minstack)
 422     limit.rlim_cur = minstack;
 423
 424   /* Round the resource limit up to page size.  */
 425   limit.rlim_cur = (limit.rlim_cur + pagesz - 1) & -pagesz;
 426   __default_stacksize = limit.rlim_cur;
 427
 428 #ifdef SHARED
 429   /* Transfer the old value from the dynamic linker's internal location.  */
 430   *__libc_dl_error_tsd () = *(*GL(dl_error_catch_tsd)) ();
 431   GL(dl_error_catch_tsd) = &__libc_dl_error_tsd;
 432
 433   /* Make __rtld_lock_{,un}lock_recursive use pthread_mutex_{,un}lock,
 434      keep the lock count from the ld.so implementation.  */
 435   GL(dl_rtld_lock_recursive) = (void *) __pthread_mutex_lock;
 436   GL(dl_rtld_unlock_recursive) = (void *) __pthread_mutex_unlock;
 437   unsigned int rtld_lock_count = GL(dl_load_lock).mutex.__data.__count;
 438   GL(dl_load_lock).mutex.__data.__count = 0;
 439   while (rtld_lock_count-- > 0)
 440     __pthread_mutex_lock (&GL(dl_load_lock).mutex);
 441
 442   GL(dl_make_stack_executable_hook) = &__make_stacks_executable;
 443 #endif
 444
 445   GL(dl_init_static_tls) = &__pthread_init_static_tls;
 446
 447   GL(dl_wait_lookup_done) = &__wait_lookup_done;
 448
 449   /* Register the fork generation counter with the libc.  */
 450 #ifndef TLS_MULTIPLE_THREADS_IN_TCB
 451   __libc_multiple_threads_ptr =
 452 #endif
 453     __libc_pthread_init (&__fork_generation, __reclaim_stacks,
 454                          ptr_pthread_functions);
 455
 456   /* Determine whether the machine is SMP or not.  */
 457   __is_smp = is_smp_system ();
 458 }
 459 strong_alias (__pthread_initialize_minimal_internal,
 460               __pthread_initialize_minimal)
 461
 462
 463 size_t
 464 __pthread_get_minstack (const pthread_attr_t *attr)
 465 {
 466   struct pthread_attr *iattr = (struct pthread_attr *) attr;
 467
 468   return (GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN
 469           + iattr->guardsize);
 470 }