linuxthreads/manager.c

   1 /* Linuxthreads - a simple clone()-based implementation of Posix        */
   2 /* threads for Linux.                                                   */
   3 /* Copyright (C) 1996 Xavier Leroy (Xavier.Leroy@inria.fr)              */
   4 /*                                                                      */
   5 /* This program is free software; you can redistribute it and/or        */
   6 /* modify it under the terms of the GNU Library General Public License  */
   7 /* as published by the Free Software Foundation; either version 2       */
   8 /* of the License, or (at your option) any later version.               */
   9 /*                                                                      */
  10 /* This program 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        */
  13 /* GNU Library General Public License for more details.                 */
  14
  15 /* The "thread manager" thread: manages creation and termination of threads */
  16
  17 #include <errno.h>
  18 #include <sched.h>
  19 #include <stddef.h>
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <sys/poll.h>           /* for poll */
  25 #include <sys/mman.h>           /* for mmap */
  26 #include <sys/param.h>
  27 #include <sys/time.h>
  28 #include <sys/wait.h>           /* for waitpid macros */
  29 #include <linux/tasks.h>
  30
  31 #include "pthread.h"
  32 #include "internals.h"
  33 #include "spinlock.h"
  34 #include "restart.h"
  35 #include "semaphore.h"
  36
  37 /* Array of active threads. Entry 0 is reserved for the initial thread. */
  38 struct pthread_handle_struct __pthread_handles[PTHREAD_THREADS_MAX] =
  39 { { LOCK_INITIALIZER, &__pthread_initial_thread, 0},
  40   { LOCK_INITIALIZER, &__pthread_manager_thread, 0}, /* All NULLs */ };
  41
  42 /* Indicate whether at least one thread has a user-defined stack (if 1),
  43    or if all threads have stacks supplied by LinuxThreads (if 0). */
  44 int __pthread_nonstandard_stacks = 0;
  45
  46 /* Number of active entries in __pthread_handles (used by gdb) */
  47 volatile int __pthread_handles_num = 2;
  48
  49 /* Whether to use debugger additional actions for thread creation
  50    (set to 1 by gdb) */
  51 volatile int __pthread_threads_debug = 0;
  52
  53 /* Mapping from stack segment to thread descriptor. */
  54 /* Stack segment numbers are also indices into the __pthread_handles array. */
  55 /* Stack segment number 0 is reserved for the initial thread. */
  56
  57 static inline pthread_descr thread_segment(int seg)
  58 {
  59   return (pthread_descr)(THREAD_STACK_START_ADDRESS - (seg - 1) * STACK_SIZE)
  60          - 1;
  61 }
  62
  63 /* Flag set in signal handler to record child termination */
  64
  65 static volatile int terminated_children = 0;
  66
  67 /* Flag set when the initial thread is blocked on pthread_exit waiting
  68    for all other threads to terminate */
  69
  70 static int main_thread_exiting = 0;
  71
  72 /* Counter used to generate unique thread identifier.
  73    Thread identifier is pthread_threads_counter + segment. */
  74
  75 static pthread_t pthread_threads_counter = 0;
  76
  77 /* Forward declarations */
  78
  79 static int pthread_handle_create(pthread_t *thread, const pthread_attr_t *attr,
  80                                  void * (*start_routine)(void *), void *arg,
  81                                  sigset_t *mask, int father_pid);
  82 static void pthread_handle_free(pthread_t th_id);
  83 static void pthread_handle_exit(pthread_descr issuing_thread, int exitcode);
  84 static void pthread_reap_children(void);
  85 static void pthread_kill_all_threads(int sig, int main_thread_also);
  86
  87 /* The server thread managing requests for thread creation and termination */
  88
  89 int __pthread_manager(void *arg)
  90 {
  91   int reqfd = (int)arg;
  92   struct pollfd ufd;
  93   sigset_t mask;
  94   int n;
  95   struct pthread_request request;
  96
  97   /* If we have special thread_self processing, initialize it.  */
  98 #ifdef INIT_THREAD_SELF
  99   INIT_THREAD_SELF(&__pthread_manager_thread, 1);
 100 #endif
 101   /* Set the error variable.  */
 102   __pthread_manager_thread.p_errnop = &__pthread_manager_thread.p_errno;
 103   __pthread_manager_thread.p_h_errnop = &__pthread_manager_thread.p_h_errno;
 104   /* Block all signals except __pthread_sig_restart, __pthread_sig_cancel
 105      and SIGTRAP */
 106   sigfillset(&mask);
 107   sigdelset(&mask, __pthread_sig_restart);
 108   sigdelset(&mask, __pthread_sig_cancel); /* for debugging new threads */
 109   sigdelset(&mask, SIGTRAP);            /* for debugging purposes */
 110   sigprocmask(SIG_SETMASK, &mask, NULL);
 111   /* Raise our priority to match that of main thread */
 112   __pthread_manager_adjust_prio(__pthread_main_thread->p_priority);
 113   /* Synchronize debugging of the thread manager */
 114   n = __libc_read(reqfd, (char *)&request, sizeof(request));
 115   ASSERT(n == sizeof(request) && request.req_kind == REQ_DEBUG);
 116   ufd.fd = reqfd;
 117   ufd.events = POLLIN;
 118   /* Enter server loop */
 119   while(1) {
 120     n = __poll(&ufd, 1, 2000);
 121
 122     /* Check for termination of the main thread */
 123     if (getppid() == 1) {
 124       pthread_kill_all_threads(SIGKILL, 0);
 125       _exit(0);
 126     }
 127     /* Check for dead children */
 128     if (terminated_children) {
 129       terminated_children = 0;
 130       pthread_reap_children();
 131     }
 132     /* Read and execute request */
 133     if (n == 1 && (ufd.revents & POLLIN)) {
 134       n = __libc_read(reqfd, (char *)&request, sizeof(request));
 135       ASSERT(n == sizeof(request));
 136       switch(request.req_kind) {
 137       case REQ_CREATE:
 138         request.req_thread->p_retcode =
 139           pthread_handle_create((pthread_t *) &request.req_thread->p_retval,
 140                                 request.req_args.create.attr,
 141                                 request.req_args.create.fn,
 142                                 request.req_args.create.arg,
 143                                 &request.req_args.create.mask,
 144                                 request.req_thread->p_pid);
 145         restart(request.req_thread);
 146         break;
 147       case REQ_FREE:
 148         pthread_handle_free(request.req_args.free.thread_id);
 149         break;
 150       case REQ_PROCESS_EXIT:
 151         pthread_handle_exit(request.req_thread,
 152                             request.req_args.exit.code);
 153         break;
 154       case REQ_MAIN_THREAD_EXIT:
 155         main_thread_exiting = 1;
 156         if (__pthread_main_thread->p_nextlive == __pthread_main_thread) {
 157           restart(__pthread_main_thread);
 158           return 0;
 159         }
 160         break;
 161       case REQ_POST:
 162         sem_post(request.req_args.post);
 163         break;
 164       case REQ_DEBUG:
 165         /* Make gdb aware of new thread */
 166         if (__pthread_threads_debug) raise(__pthread_sig_cancel);
 167         restart(request.req_thread);
 168         break;
 169       }
 170     }
 171   }
 172 }
 173
 174 /* Process creation */
 175
 176 static int pthread_start_thread(void *arg)
 177 {
 178   pthread_descr self = (pthread_descr) arg;
 179   struct pthread_request request;
 180   void * outcome;
 181   /* Initialize special thread_self processing, if any.  */
 182 #ifdef INIT_THREAD_SELF
 183   INIT_THREAD_SELF(self, self->p_nr);
 184 #endif
 185   /* Make sure our pid field is initialized, just in case we get there
 186      before our father has initialized it. */
 187   THREAD_SETMEM(self, p_pid, __getpid());
 188   /* Initial signal mask is that of the creating thread. (Otherwise,
 189      we'd just inherit the mask of the thread manager.) */
 190   sigprocmask(SIG_SETMASK, &self->p_start_args.mask, NULL);
 191   /* Set the scheduling policy and priority for the new thread, if needed */
 192   if (THREAD_GETMEM(self, p_start_args.schedpolicy) >= 0)
 193     __sched_setscheduler(THREAD_GETMEM(self, p_pid),
 194                          THREAD_GETMEM(self, p_start_args.schedpolicy),
 195                          &self->p_start_args.schedparam);
 196   /* Make gdb aware of new thread */
 197   if (__pthread_threads_debug) {
 198     request.req_thread = self;
 199     request.req_kind = REQ_DEBUG;
 200     __libc_write(__pthread_manager_request,
 201                  (char *) &request, sizeof(request));
 202     suspend(self);
 203   }
 204   /* Run the thread code */
 205   outcome = self->p_start_args.start_routine(THREAD_GETMEM(self,
 206                                                            p_start_args.arg));
 207   /* Exit with the given return value */
 208   pthread_exit(outcome);
 209   return 0;
 210 }
 211
 212 static int pthread_allocate_stack(const pthread_attr_t *attr,
 213                                   pthread_descr default_new_thread,
 214                                   int pagesize,
 215                                   pthread_descr * out_new_thread,
 216                                   char ** out_new_thread_bottom,
 217                                   char ** out_guardaddr,
 218                                   size_t * out_guardsize)
 219 {
 220   pthread_descr new_thread;
 221   char * new_thread_bottom;
 222   char * guardaddr;
 223   size_t stacksize, guardsize;
 224
 225   if (attr != NULL && attr->__stackaddr_set)
 226     {
 227       /* The user provided a stack. */
 228       new_thread =
 229         (pthread_descr) ((long)(attr->__stackaddr) & -sizeof(void *)) - 1;
 230       new_thread_bottom = (char *) attr->__stackaddr - attr->__stacksize;
 231       guardaddr = NULL;
 232       guardsize = 0;
 233     }
 234   else
 235     {
 236       /* Allocate space for stack and thread descriptor at default address */
 237       new_thread = default_new_thread;
 238       new_thread_bottom = (char *) new_thread - STACK_SIZE;
 239       if (mmap((caddr_t)((char *)(new_thread + 1) - INITIAL_STACK_SIZE),
 240                INITIAL_STACK_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
 241                MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED | MAP_GROWSDOWN,
 242                -1, 0) == MAP_FAILED)
 243         /* Bad luck, this segment is already mapped. */
 244         return -1;
 245       /* We manage to get a stack.  Now see whether we need a guard
 246          and allocate it if necessary.  Notice that the default
 247          attributes (stack_size = STACK_SIZE - pagesize and
 248          guardsize = pagesize) do not need a guard page, since
 249          the RLIMIT_STACK soft limit prevents stacks from
 250          running into one another. */
 251       if (attr == NULL ||
 252           attr->__guardsize == 0 ||
 253           (attr->__guardsize == pagesize &&
 254            attr->__stacksize == STACK_SIZE - pagesize))
 255         {
 256           /* We don't need a guard page. */
 257           guardaddr = NULL;
 258           guardsize = 0;
 259         }
 260       else
 261         {
 262           /* Put a bad page at the bottom of the stack */
 263           stacksize = roundup(attr->__stacksize, pagesize);
 264           if (stacksize >= STACK_SIZE - pagesize)
 265             stacksize = STACK_SIZE - pagesize;
 266           guardaddr = (void *)new_thread - stacksize;
 267           guardsize = attr->__guardsize;
 268           if (mmap ((caddr_t) guardaddr, guardsize, 0, MAP_FIXED, -1, 0)
 269               == MAP_FAILED)
 270             {
 271               /* We don't make this an error.  */
 272               guardaddr = NULL;
 273               guardsize = 0;
 274             }
 275         }
 276     }
 277   *out_new_thread = new_thread;
 278   *out_new_thread_bottom = new_thread_bottom;
 279   *out_guardaddr = guardaddr;
 280   *out_guardsize = guardsize;
 281   return 0;
 282 }
 283
 284 static int pthread_handle_create(pthread_t *thread, const pthread_attr_t *attr,
 285                                  void * (*start_routine)(void *), void *arg,
 286                                  sigset_t * mask, int father_pid)
 287 {
 288   size_t sseg;
 289   int pid;
 290   pthread_descr new_thread;
 291   char * new_thread_bottom;
 292   pthread_t new_thread_id;
 293   char *guardaddr = NULL;
 294   size_t guardsize = 0;
 295   int pagesize = __getpagesize();
 296
 297   /* First check whether we have to change the policy and if yes, whether
 298      we can  do this.  Normally this should be done by examining the
 299      return value of the __sched_setscheduler call in pthread_start_thread
 300      but this is hard to implement.  FIXME  */
 301   if (attr != NULL && attr->__schedpolicy != SCHED_OTHER && geteuid () != 0)
 302     return EPERM;
 303   /* Find a free segment for the thread, and allocate a stack if needed */
 304   for (sseg = 2; ; sseg++)
 305     {
 306       if (sseg >= PTHREAD_THREADS_MAX)
 307         return EAGAIN;
 308       if (__pthread_handles[sseg].h_descr != NULL)
 309         continue;
 310       if (pthread_allocate_stack(attr, thread_segment(sseg), pagesize,
 311                                  &new_thread, &new_thread_bottom,
 312                                  &guardaddr, &guardsize) == 0)
 313         break;
 314     }
 315   /* Allocate new thread identifier */
 316   pthread_threads_counter += PTHREAD_THREADS_MAX;
 317   new_thread_id = sseg + pthread_threads_counter;
 318   /* Initialize the thread descriptor */
 319   new_thread->p_nextwaiting = NULL;
 320   new_thread->p_tid = new_thread_id;
 321   new_thread->p_priority = 0;
 322   new_thread->p_lock = &(__pthread_handles[sseg].h_lock);
 323   new_thread->p_signal = 0;
 324   new_thread->p_signal_jmp = NULL;
 325   new_thread->p_cancel_jmp = NULL;
 326   new_thread->p_terminated = 0;
 327   new_thread->p_detached = attr == NULL ? 0 : attr->__detachstate;
 328   new_thread->p_exited = 0;
 329   new_thread->p_retval = NULL;
 330   new_thread->p_joining = NULL;
 331   new_thread->p_cleanup = NULL;
 332   new_thread->p_cancelstate = PTHREAD_CANCEL_ENABLE;
 333   new_thread->p_canceltype = PTHREAD_CANCEL_DEFERRED;
 334   new_thread->p_canceled = 0;
 335   new_thread->p_errnop = &new_thread->p_errno;
 336   new_thread->p_errno = 0;
 337   new_thread->p_h_errnop = &new_thread->p_h_errno;
 338   new_thread->p_h_errno = 0;
 339   new_thread->p_in_sighandler = NULL;
 340   new_thread->p_sigwaiting = 0;
 341   new_thread->p_guardaddr = guardaddr;
 342   new_thread->p_guardsize = guardsize;
 343   new_thread->p_userstack = attr != NULL && attr->__stackaddr_set;
 344   memset (new_thread->p_specific, '\0',
 345           PTHREAD_KEY_1STLEVEL_SIZE * sizeof (new_thread->p_specific[0]));
 346   new_thread->p_self = new_thread;
 347   new_thread->p_nr = sseg;
 348   /* Initialize the thread handle */
 349   __pthread_init_lock(&__pthread_handles[sseg].h_lock);
 350   __pthread_handles[sseg].h_descr = new_thread;
 351   __pthread_handles[sseg].h_bottom = new_thread_bottom;
 352   /* Determine scheduling parameters for the thread */
 353   new_thread->p_start_args.schedpolicy = -1;
 354   if (attr != NULL) {
 355     switch(attr->__inheritsched) {
 356     case PTHREAD_EXPLICIT_SCHED:
 357       new_thread->p_start_args.schedpolicy = attr->__schedpolicy;
 358       memcpy (&new_thread->p_start_args.schedparam, &attr->__schedparam,
 359               sizeof (struct sched_param));
 360       break;
 361     case PTHREAD_INHERIT_SCHED:
 362       /* schedpolicy doesn't need to be set, only get priority */
 363       __sched_getparam(father_pid, &new_thread->p_start_args.schedparam);
 364       break;
 365     }
 366     new_thread->p_priority =
 367       new_thread->p_start_args.schedparam.sched_priority;
 368   }
 369   /* Finish setting up arguments to pthread_start_thread */
 370   new_thread->p_start_args.start_routine = start_routine;
 371   new_thread->p_start_args.arg = arg;
 372   new_thread->p_start_args.mask = *mask;
 373   /* Raise priority of thread manager if needed */
 374   __pthread_manager_adjust_prio(new_thread->p_priority);
 375   /* Do the cloning */
 376   pid = __clone(pthread_start_thread, (void **) new_thread,
 377                 CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_SIGHAND |
 378 #ifdef CLONE_PTRACE
 379                 CLONE_PTRACE |
 380 #endif
 381                 __pthread_sig_restart, new_thread);
 382   /* Check if cloning succeeded */
 383   if (pid == -1) {
 384     /* Free the stack if we allocated it */
 385     if (attr == NULL || !attr->__stackaddr_set)
 386       {
 387         munmap((caddr_t)((char *)(new_thread+1) - INITIAL_STACK_SIZE),
 388                INITIAL_STACK_SIZE);
 389         if (new_thread->p_guardsize != 0)
 390           munmap(new_thread->p_guardaddr, new_thread->p_guardsize);
 391       }
 392     __pthread_handles[sseg].h_descr = NULL;
 393     __pthread_handles[sseg].h_bottom = NULL;
 394     __pthread_handles_num--;
 395     return errno;
 396   }
 397   /* Insert new thread in doubly linked list of active threads */
 398   new_thread->p_prevlive = __pthread_main_thread;
 399   new_thread->p_nextlive = __pthread_main_thread->p_nextlive;
 400   __pthread_main_thread->p_nextlive->p_prevlive = new_thread;
 401   __pthread_main_thread->p_nextlive = new_thread;
 402   /* Set pid field of the new thread, in case we get there before the
 403      child starts. */
 404   new_thread->p_pid = pid;
 405   /* We're all set */
 406   *thread = new_thread_id;
 407   return 0;
 408 }
 409
 410
 411 /* Try to free the resources of a thread when requested by pthread_join
 412    or pthread_detach on a terminated thread. */
 413
 414 static void pthread_free(pthread_descr th)
 415 {
 416   pthread_handle handle;
 417   ASSERT(th->p_exited);
 418   /* Make the handle invalid */
 419   handle =  thread_handle(th->p_tid);
 420   __pthread_lock(&handle->h_lock, NULL);
 421   handle->h_descr = NULL;
 422   handle->h_bottom = (char *)(-1L);
 423   __pthread_unlock(&handle->h_lock);
 424 #ifdef FREE_THREAD_SELF
 425   FREE_THREAD_SELF(th, th->p_nr);
 426 #endif
 427   /* One fewer threads in __pthread_handles */
 428   __pthread_handles_num--;
 429   /* If initial thread, nothing to free */
 430   if (th == &__pthread_initial_thread) return;
 431   if (!th->p_userstack)
 432     {
 433       /* Free the stack and thread descriptor area */
 434       if (th->p_guardsize != 0)
 435         munmap(th->p_guardaddr, th->p_guardsize);
 436       munmap((caddr_t) ((char *)(th+1) - STACK_SIZE), STACK_SIZE);
 437     }
 438 }
 439
 440 /* Handle threads that have exited */
 441
 442 static void pthread_exited(pid_t pid)
 443 {
 444   pthread_descr th;
 445   int detached;
 446   /* Find thread with that pid */
 447   for (th = __pthread_main_thread->p_nextlive;
 448        th != __pthread_main_thread;
 449        th = th->p_nextlive) {
 450     if (th->p_pid == pid) {
 451       /* Remove thread from list of active threads */
 452       th->p_nextlive->p_prevlive = th->p_prevlive;
 453       th->p_prevlive->p_nextlive = th->p_nextlive;
 454       /* Mark thread as exited, and if detached, free its resources */
 455       __pthread_lock(th->p_lock, NULL);
 456       th->p_exited = 1;
 457       detached = th->p_detached;
 458       __pthread_unlock(th->p_lock);
 459       if (detached)
 460         pthread_free(th);
 461       break;
 462     }
 463   }
 464   /* If all threads have exited and the main thread is pending on a
 465      pthread_exit, wake up the main thread and terminate ourselves. */
 466   if (main_thread_exiting &&
 467       __pthread_main_thread->p_nextlive == __pthread_main_thread) {
 468     restart(__pthread_main_thread);
 469     _exit(0);
 470   }
 471 }
 472
 473 static void pthread_reap_children(void)
 474 {
 475   pid_t pid;
 476   int status;
 477
 478   while ((pid = __libc_waitpid(-1, &status, WNOHANG | __WCLONE)) > 0) {
 479     pthread_exited(pid);
 480     if (WIFSIGNALED(status)) {
 481       /* If a thread died due to a signal, send the same signal to
 482          all other threads, including the main thread. */
 483       pthread_kill_all_threads(WTERMSIG(status), 1);
 484       _exit(0);
 485     }
 486   }
 487 }
 488
 489 /* Try to free the resources of a thread when requested by pthread_join
 490    or pthread_detach on a terminated thread. */
 491
 492 static void pthread_handle_free(pthread_t th_id)
 493 {
 494   pthread_handle handle = thread_handle(th_id);
 495   pthread_descr th;
 496
 497   __pthread_lock(&handle->h_lock, NULL);
 498   if (invalid_handle(handle, th_id)) {
 499     /* pthread_reap_children has deallocated the thread already,
 500        nothing needs to be done */
 501     __pthread_unlock(&handle->h_lock);
 502     return;
 503   }
 504   th = handle->h_descr;
 505   if (th->p_exited) {
 506     __pthread_unlock(&handle->h_lock);
 507     pthread_free(th);
 508   } else {
 509     /* The Unix process of the thread is still running.
 510        Mark the thread as detached so that the thread manager will
 511        deallocate its resources when the Unix process exits. */
 512     th->p_detached = 1;
 513     __pthread_unlock(&handle->h_lock);
 514   }
 515 }
 516
 517 /* Send a signal to all running threads */
 518
 519 static void pthread_kill_all_threads(int sig, int main_thread_also)
 520 {
 521   pthread_descr th;
 522   for (th = __pthread_main_thread->p_nextlive;
 523        th != __pthread_main_thread;
 524        th = th->p_nextlive) {
 525     kill(th->p_pid, sig);
 526   }
 527   if (main_thread_also) {
 528     kill(__pthread_main_thread->p_pid, sig);
 529   }
 530 }
 531
 532 /* Process-wide exit() */
 533
 534 static void pthread_handle_exit(pthread_descr issuing_thread, int exitcode)
 535 {
 536   pthread_descr th;
 537   __pthread_exit_requested = 1;
 538   __pthread_exit_code = exitcode;
 539   /* Send the CANCEL signal to all running threads, including the main
 540      thread, but excluding the thread from which the exit request originated
 541      (that thread must complete the exit, e.g. calling atexit functions
 542      and flushing stdio buffers). */
 543   for (th = issuing_thread->p_nextlive;
 544        th != issuing_thread;
 545        th = th->p_nextlive) {
 546     kill(th->p_pid, __pthread_sig_cancel);
 547   }
 548   /* Now, wait for all these threads, so that they don't become zombies
 549      and their times are properly added to the thread manager's times. */
 550   for (th = issuing_thread->p_nextlive;
 551        th != issuing_thread;
 552        th = th->p_nextlive) {
 553     waitpid(th->p_pid, NULL, __WCLONE);
 554   }
 555   restart(issuing_thread);
 556   _exit(0);
 557 }
 558
 559 /* Handler for __pthread_sig_restart in thread manager thread */
 560
 561 void __pthread_manager_sighandler(int sig)
 562 {
 563   terminated_children = 1;
 564 }
 565
 566 /* Adjust priority of thread manager so that it always run at a priority
 567    higher than all threads */
 568
 569 void __pthread_manager_adjust_prio(int thread_prio)
 570 {
 571   struct sched_param param;
 572
 573   if (thread_prio <= __pthread_manager_thread.p_priority) return;
 574   param.sched_priority =
 575     thread_prio < __sched_get_priority_max(SCHED_FIFO)
 576     ? thread_prio + 1 : thread_prio;
 577   __sched_setscheduler(__pthread_manager_thread.p_pid, SCHED_FIFO, &param);
 578   __pthread_manager_thread.p_priority = thread_prio;
 579 }