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   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 (__glibc_unlikely (pid < 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   int result;
 236   if (__glibc_unlikely (pid < 0))
 237     pid = -pid;
 238
 239   /* Safety check.  It would be possible to call this function for
 240      other signals and send a signal from another process.  This is not
 241      correct and might even be a security problem.  Try to catch as
 242      many incorrect invocations as possible.  */
 243   if (sig != SIGSETXID
 244       || si->si_pid != pid
 245       || si->si_code != SI_TKILL)
 246     return;
 247
 248   INTERNAL_SYSCALL_DECL (err);
 249   result = INTERNAL_SYSCALL_NCS (__xidcmd->syscall_no, err, 3, __xidcmd->id[0],
 250                                  __xidcmd->id[1], __xidcmd->id[2]);
 251   if (__glibc_unlikely (INTERNAL_SYSCALL_ERROR_P (result, err)))
 252     /* Safety check.  This should never happen if the setxid system
 253        calls are only ever called through their glibc wrappers.  */
 254     abort ();
 255
 256   /* Reset the SETXID flag.  */
 257   struct pthread *self = THREAD_SELF;
 258   int flags, newval;
 259   do
 260     {
 261       flags = THREAD_GETMEM (self, cancelhandling);
 262       newval = THREAD_ATOMIC_CMPXCHG_VAL (self, cancelhandling,
 263                                           flags & ~SETXID_BITMASK, flags);
 264     }
 265   while (flags != newval);
 266
 267   /* And release the futex.  */
 268   self->setxid_futex = 1;
 269   lll_futex_wake (&self->setxid_futex, 1, LLL_PRIVATE);
 270
 271   if (atomic_decrement_val (&__xidcmd->cntr) == 0)
 272     lll_futex_wake (&__xidcmd->cntr, 1, LLL_PRIVATE);
 273 }
 274
 275
 276 /* When using __thread for this, we do it in libc so as not
 277    to give libpthread its own TLS segment just for this.  */
 278 extern void **__libc_dl_error_tsd (void) __attribute__ ((const));
 279
 280
 281 /* This can be set by the debugger before initialization is complete.  */
 282 static bool __nptl_initial_report_events __attribute_used__;
 283
 284 void
 285 __pthread_initialize_minimal_internal (void)
 286 {
 287 #ifndef SHARED
 288   /* Unlike in the dynamically linked case the dynamic linker has not
 289      taken care of initializing the TLS data structures.  */
 290   __libc_setup_tls (TLS_TCB_SIZE, TLS_TCB_ALIGN);
 291
 292   /* We must prevent gcc from being clever and move any of the
 293      following code ahead of the __libc_setup_tls call.  This function
 294      will initialize the thread register which is subsequently
 295      used.  */
 296   __asm __volatile ("");
 297 #endif
 298
 299   /* Minimal initialization of the thread descriptor.  */
 300   struct pthread *pd = THREAD_SELF;
 301   INTERNAL_SYSCALL_DECL (err);
 302   pd->pid = pd->tid = INTERNAL_SYSCALL (set_tid_address, err, 1, &pd->tid);
 303   THREAD_SETMEM (pd, specific[0], &pd->specific_1stblock[0]);
 304   THREAD_SETMEM (pd, user_stack, true);
 305   if (LLL_LOCK_INITIALIZER != 0)
 306     THREAD_SETMEM (pd, lock, LLL_LOCK_INITIALIZER);
 307 #if HP_TIMING_AVAIL
 308   THREAD_SETMEM (pd, cpuclock_offset, GL(dl_cpuclock_offset));
 309 #endif
 310
 311   /* Initialize the robust mutex data.  */
 312 #ifdef __PTHREAD_MUTEX_HAVE_PREV
 313   pd->robust_prev = &pd->robust_head;
 314 #endif
 315   pd->robust_head.list = &pd->robust_head;
 316 #ifdef __NR_set_robust_list
 317   pd->robust_head.futex_offset = (offsetof (pthread_mutex_t, __data.__lock)
 318                                   - offsetof (pthread_mutex_t,
 319                                               __data.__list.__next));
 320   int res = INTERNAL_SYSCALL (set_robust_list, err, 2, &pd->robust_head,
 321                               sizeof (struct robust_list_head));
 322   if (INTERNAL_SYSCALL_ERROR_P (res, err))
 323 #endif
 324     set_robust_list_not_avail ();
 325
 326 #ifndef __ASSUME_PRIVATE_FUTEX
 327   /* Private futexes are always used (at least internally) so that
 328      doing the test once this early is beneficial.  */
 329   {
 330     int word = 0;
 331     word = INTERNAL_SYSCALL (futex, err, 3, &word,
 332                             FUTEX_WAKE | FUTEX_PRIVATE_FLAG, 1);
 333     if (!INTERNAL_SYSCALL_ERROR_P (word, err))
 334       THREAD_SETMEM (pd, header.private_futex, FUTEX_PRIVATE_FLAG);
 335   }
 336
 337   /* Private futexes have been introduced earlier than the
 338      FUTEX_CLOCK_REALTIME flag.  We don't have to run the test if we
 339      know the former are not supported.  This also means we know the
 340      kernel will return ENOSYS for unknown operations.  */
 341   if (THREAD_GETMEM (pd, header.private_futex) != 0)
 342 #endif
 343 #ifndef __ASSUME_FUTEX_CLOCK_REALTIME
 344     {
 345       int word = 0;
 346       /* NB: the syscall actually takes six parameters.  The last is the
 347          bit mask.  But since we will not actually wait at all the value
 348          is irrelevant.  Given that passing six parameters is difficult
 349          on some architectures we just pass whatever random value the
 350          calling convention calls for to the kernel.  It causes no harm.  */
 351       word = INTERNAL_SYSCALL (futex, err, 5, &word,
 352                                FUTEX_WAIT_BITSET | FUTEX_CLOCK_REALTIME
 353                                | FUTEX_PRIVATE_FLAG, 1, NULL, 0);
 354       assert (INTERNAL_SYSCALL_ERROR_P (word, err));
 355       if (INTERNAL_SYSCALL_ERRNO (word, err) != ENOSYS)
 356         __set_futex_clock_realtime ();
 357     }
 358 #endif
 359
 360   /* Set initial thread's stack block from 0 up to __libc_stack_end.
 361      It will be bigger than it actually is, but for unwind.c/pt-longjmp.c
 362      purposes this is good enough.  */
 363   THREAD_SETMEM (pd, stackblock_size, (size_t) __libc_stack_end);
 364
 365   /* Initialize the list of all running threads with the main thread.  */
 366   INIT_LIST_HEAD (&__stack_user);
 367   list_add (&pd->list, &__stack_user);
 368
 369   /* Before initializing __stack_user, the debugger could not find us and
 370      had to set __nptl_initial_report_events.  Propagate its setting.  */
 371   THREAD_SETMEM (pd, report_events, __nptl_initial_report_events);
 372
 373   /* Install the cancellation signal handler.  If for some reason we
 374      cannot install the handler we do not abort.  Maybe we should, but
 375      it is only asynchronous cancellation which is affected.  */
 376   struct sigaction sa;
 377   sa.sa_sigaction = sigcancel_handler;
 378   sa.sa_flags = SA_SIGINFO;
 379   __sigemptyset (&sa.sa_mask);
 380
 381   (void) __libc_sigaction (SIGCANCEL, &sa, NULL);
 382
 383   /* Install the handle to change the threads' uid/gid.  */
 384   sa.sa_sigaction = sighandler_setxid;
 385   sa.sa_flags = SA_SIGINFO | SA_RESTART;
 386
 387   (void) __libc_sigaction (SIGSETXID, &sa, NULL);
 388
 389   /* The parent process might have left the signals blocked.  Just in
 390      case, unblock it.  We reuse the signal mask in the sigaction
 391      structure.  It is already cleared.  */
 392   __sigaddset (&sa.sa_mask, SIGCANCEL);
 393   __sigaddset (&sa.sa_mask, SIGSETXID);
 394   (void) INTERNAL_SYSCALL (rt_sigprocmask, err, 4, SIG_UNBLOCK, &sa.sa_mask,
 395                            NULL, _NSIG / 8);
 396
 397   /* Get the size of the static and alignment requirements for the TLS
 398      block.  */
 399   size_t static_tls_align;
 400   _dl_get_tls_static_info (&__static_tls_size, &static_tls_align);
 401
 402   /* Make sure the size takes all the alignments into account.  */
 403   if (STACK_ALIGN > static_tls_align)
 404     static_tls_align = STACK_ALIGN;
 405   __static_tls_align_m1 = static_tls_align - 1;
 406
 407   __static_tls_size = roundup (__static_tls_size, static_tls_align);
 408
 409   /* Determine the default allowed stack size.  This is the size used
 410      in case the user does not specify one.  */
 411   struct rlimit limit;
 412   if (getrlimit (RLIMIT_STACK, &limit) != 0
 413       || limit.rlim_cur == RLIM_INFINITY)
 414     /* The system limit is not usable.  Use an architecture-specific
 415        default.  */
 416     limit.rlim_cur = ARCH_STACK_DEFAULT_SIZE;
 417   else if (limit.rlim_cur < PTHREAD_STACK_MIN)
 418     /* The system limit is unusably small.
 419        Use the minimal size acceptable.  */
 420     limit.rlim_cur = PTHREAD_STACK_MIN;
 421
 422   /* Make sure it meets the minimum size that allocate_stack
 423      (allocatestack.c) will demand, which depends on the page size.  */
 424   const uintptr_t pagesz = GLRO(dl_pagesize);
 425   const size_t minstack = pagesz + __static_tls_size + MINIMAL_REST_STACK;
 426   if (limit.rlim_cur < minstack)
 427     limit.rlim_cur = minstack;
 428
 429   /* Round the resource limit up to page size.  */
 430   limit.rlim_cur = (limit.rlim_cur + pagesz - 1) & -pagesz;
 431   lll_lock (__default_pthread_attr_lock, LLL_PRIVATE);
 432   __default_pthread_attr.stacksize = limit.rlim_cur;
 433   __default_pthread_attr.guardsize = GLRO (dl_pagesize);
 434   lll_unlock (__default_pthread_attr_lock, LLL_PRIVATE);
 435
 436 #ifdef SHARED
 437   /* Transfer the old value from the dynamic linker's internal location.  */
 438   *__libc_dl_error_tsd () = *(*GL(dl_error_catch_tsd)) ();
 439   GL(dl_error_catch_tsd) = &__libc_dl_error_tsd;
 440
 441   /* Make __rtld_lock_{,un}lock_recursive use pthread_mutex_{,un}lock,
 442      keep the lock count from the ld.so implementation.  */
 443   GL(dl_rtld_lock_recursive) = (void *) __pthread_mutex_lock;
 444   GL(dl_rtld_unlock_recursive) = (void *) __pthread_mutex_unlock;
 445   unsigned int rtld_lock_count = GL(dl_load_lock).mutex.__data.__count;
 446   GL(dl_load_lock).mutex.__data.__count = 0;
 447   while (rtld_lock_count-- > 0)
 448     __pthread_mutex_lock (&GL(dl_load_lock).mutex);
 449
 450   GL(dl_make_stack_executable_hook) = &__make_stacks_executable;
 451 #endif
 452
 453   GL(dl_init_static_tls) = &__pthread_init_static_tls;
 454
 455   GL(dl_wait_lookup_done) = &__wait_lookup_done;
 456
 457   /* Register the fork generation counter with the libc.  */
 458 #ifndef TLS_MULTIPLE_THREADS_IN_TCB
 459   __libc_multiple_threads_ptr =
 460 #endif
 461     __libc_pthread_init (&__fork_generation, __reclaim_stacks,
 462                          ptr_pthread_functions);
 463
 464   /* Determine whether the machine is SMP or not.  */
 465   __is_smp = is_smp_system ();
 466 }
 467 strong_alias (__pthread_initialize_minimal_internal,
 468               __pthread_initialize_minimal)
 469
 470
 471 size_t
 472 __pthread_get_minstack (const pthread_attr_t *attr)
 473 {
 474   struct pthread_attr *iattr = (struct pthread_attr *) attr;
 475
 476   return (GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN
 477           + iattr->guardsize);
 478 }