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_cancel and SIGTRAP */
 105   sigfillset(&mask);
 106   sigdelset(&mask, __pthread_sig_cancel); /* for thread termination */
 107   sigdelset(&mask, SIGTRAP);            /* for debugging purposes */
 108   sigprocmask(SIG_SETMASK, &mask, NULL);
 109   /* Raise our priority to match that of main thread */
 110   __pthread_manager_adjust_prio(__pthread_main_thread->p_priority);
 111   /* Synchronize debugging of the thread manager */
 112   n = __libc_read(reqfd, (char *)&request, sizeof(request));
 113   ASSERT(n == sizeof(request) && request.req_kind == REQ_DEBUG);
 114   ufd.fd = reqfd;
 115   ufd.events = POLLIN;
 116   /* Enter server loop */
 117   while(1) {
 118     n = __poll(&ufd, 1, 2000);
 119
 120     /* Check for termination of the main thread */
 121     if (getppid() == 1) {
 122       pthread_kill_all_threads(SIGKILL, 0);
 123       _exit(0);
 124     }
 125     /* Check for dead children */
 126     if (terminated_children) {
 127       terminated_children = 0;
 128       pthread_reap_children();
 129     }
 130     /* Read and execute request */
 131     if (n == 1 && (ufd.revents & POLLIN)) {
 132       n = __libc_read(reqfd, (char *)&request, sizeof(request));
 133       ASSERT(n == sizeof(request));
 134       switch(request.req_kind) {
 135       case REQ_CREATE:
 136         request.req_thread->p_retcode =
 137           pthread_handle_create((pthread_t *) &request.req_thread->p_retval,
 138                                 request.req_args.create.attr,
 139                                 request.req_args.create.fn,
 140                                 request.req_args.create.arg,
 141                                 &request.req_args.create.mask,
 142                                 request.req_thread->p_pid);
 143         restart(request.req_thread);
 144         break;
 145       case REQ_FREE:
 146         pthread_handle_free(request.req_args.free.thread_id);
 147         break;
 148       case REQ_PROCESS_EXIT:
 149         pthread_handle_exit(request.req_thread,
 150                             request.req_args.exit.code);
 151         break;
 152       case REQ_MAIN_THREAD_EXIT:
 153         main_thread_exiting = 1;
 154         if (__pthread_main_thread->p_nextlive == __pthread_main_thread) {
 155           restart(__pthread_main_thread);
 156           return 0;
 157         }
 158         break;
 159       case REQ_POST:
 160         sem_post(request.req_args.post);
 161         break;
 162       case REQ_DEBUG:
 163         /* Make gdb aware of new thread and gdb will restart the
 164            new thread when it is ready to handle the new thread. */
 165         if (__pthread_threads_debug && __pthread_sig_debug > 0)
 166           raise(__pthread_sig_debug);
 167         break;
 168       }
 169     }
 170   }
 171 }
 172
 173 /* Process creation */
 174
 175 static int pthread_start_thread(void *arg)
 176 {
 177   pthread_descr self = (pthread_descr) arg;
 178   struct pthread_request request;
 179   void * outcome;
 180   /* Initialize special thread_self processing, if any.  */
 181 #ifdef INIT_THREAD_SELF
 182   INIT_THREAD_SELF(self, self->p_nr);
 183 #endif
 184   /* Make sure our pid field is initialized, just in case we get there
 185      before our father has initialized it. */
 186   THREAD_SETMEM(self, p_pid, __getpid());
 187   /* Initial signal mask is that of the creating thread. (Otherwise,
 188      we'd just inherit the mask of the thread manager.) */
 189   sigprocmask(SIG_SETMASK, &self->p_start_args.mask, NULL);
 190   /* Set the scheduling policy and priority for the new thread, if needed */
 191   if (THREAD_GETMEM(self, p_start_args.schedpolicy) >= 0)
 192     __sched_setscheduler(THREAD_GETMEM(self, p_pid),
 193                          THREAD_GETMEM(self, p_start_args.schedpolicy),
 194                          &self->p_start_args.schedparam);
 195   /* Make gdb aware of new thread */
 196   if (__pthread_threads_debug && __pthread_sig_debug > 0) {
 197     request.req_thread = self;
 198     request.req_kind = REQ_DEBUG;
 199     __libc_write(__pthread_manager_request,
 200                  (char *) &request, sizeof(request));
 201     suspend(self);
 202   }
 203   /* Run the thread code */
 204   outcome = self->p_start_args.start_routine(THREAD_GETMEM(self,
 205                                                            p_start_args.arg));
 206   /* Exit with the given return value */
 207   pthread_exit(outcome);
 208   return 0;
 209 }
 210
 211 static int pthread_allocate_stack(const pthread_attr_t *attr,
 212                                   pthread_descr default_new_thread,
 213                                   int pagesize,
 214                                   pthread_descr * out_new_thread,
 215                                   char ** out_new_thread_bottom,
 216                                   char ** out_guardaddr,
 217                                   size_t * out_guardsize)
 218 {
 219   pthread_descr new_thread;
 220   char * new_thread_bottom;
 221   char * guardaddr;
 222   size_t stacksize, guardsize;
 223
 224   if (attr != NULL && attr->__stackaddr_set)
 225     {
 226       /* The user provided a stack. */
 227       new_thread =
 228         (pthread_descr) ((long)(attr->__stackaddr) & -sizeof(void *)) - 1;
 229       new_thread_bottom = (char *) attr->__stackaddr - attr->__stacksize;
 230       guardaddr = NULL;
 231       guardsize = 0;
 232       __pthread_nonstandard_stacks = 1;
 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   /* Clear the thread data structure.  */
 278   memset (new_thread, '\0', sizeof (*new_thread));
 279   *out_new_thread = new_thread;
 280   *out_new_thread_bottom = new_thread_bottom;
 281   *out_guardaddr = guardaddr;
 282   *out_guardsize = guardsize;
 283   return 0;
 284 }
 285
 286 static int pthread_handle_create(pthread_t *thread, const pthread_attr_t *attr,
 287                                  void * (*start_routine)(void *), void *arg,
 288                                  sigset_t * mask, int father_pid)
 289 {
 290   size_t sseg;
 291   int pid;
 292   pthread_descr new_thread;
 293   char * new_thread_bottom;
 294   pthread_t new_thread_id;
 295   char *guardaddr = NULL;
 296   size_t guardsize = 0;
 297   int pagesize = __getpagesize();
 298
 299   /* First check whether we have to change the policy and if yes, whether
 300      we can  do this.  Normally this should be done by examining the
 301      return value of the __sched_setscheduler call in pthread_start_thread
 302      but this is hard to implement.  FIXME  */
 303   if (attr != NULL && attr->__schedpolicy != SCHED_OTHER && geteuid () != 0)
 304     return EPERM;
 305   /* Find a free segment for the thread, and allocate a stack if needed */
 306   for (sseg = 2; ; sseg++)
 307     {
 308       if (sseg >= PTHREAD_THREADS_MAX)
 309         return EAGAIN;
 310       if (__pthread_handles[sseg].h_descr != NULL)
 311         continue;
 312       if (pthread_allocate_stack(attr, thread_segment(sseg), pagesize,
 313                                  &new_thread, &new_thread_bottom,
 314                                  &guardaddr, &guardsize) == 0)
 315         break;
 316     }
 317   __pthread_handles_num++;
 318   /* Allocate new thread identifier */
 319   pthread_threads_counter += PTHREAD_THREADS_MAX;
 320   new_thread_id = sseg + pthread_threads_counter;
 321   /* Initialize the thread descriptor.  Elements which have to be
 322      initialized to zero already have this value.  */
 323   new_thread->p_tid = new_thread_id;
 324   new_thread->p_lock = &(__pthread_handles[sseg].h_lock);
 325   new_thread->p_cancelstate = PTHREAD_CANCEL_ENABLE;
 326   new_thread->p_canceltype = PTHREAD_CANCEL_DEFERRED;
 327   new_thread->p_errnop = &new_thread->p_errno;
 328   new_thread->p_h_errnop = &new_thread->p_h_errno;
 329   new_thread->p_guardaddr = guardaddr;
 330   new_thread->p_guardsize = guardsize;
 331   new_thread->p_self = new_thread;
 332   new_thread->p_nr = sseg;
 333   /* Initialize the thread handle */
 334   __pthread_init_lock(&__pthread_handles[sseg].h_lock);
 335   __pthread_handles[sseg].h_descr = new_thread;
 336   __pthread_handles[sseg].h_bottom = new_thread_bottom;
 337   /* Determine scheduling parameters for the thread */
 338   new_thread->p_start_args.schedpolicy = -1;
 339   if (attr != NULL) {
 340     new_thread->p_detached = attr->__detachstate;
 341     new_thread->p_userstack = attr->__stackaddr_set;
 342
 343     switch(attr->__inheritsched) {
 344     case PTHREAD_EXPLICIT_SCHED:
 345       new_thread->p_start_args.schedpolicy = attr->__schedpolicy;
 346       memcpy (&new_thread->p_start_args.schedparam, &attr->__schedparam,
 347               sizeof (struct sched_param));
 348       break;
 349     case PTHREAD_INHERIT_SCHED:
 350       /* schedpolicy doesn't need to be set, only get priority */
 351       __sched_getparam(father_pid, &new_thread->p_start_args.schedparam);
 352       break;
 353     }
 354     new_thread->p_priority =
 355       new_thread->p_start_args.schedparam.sched_priority;
 356   }
 357   /* Finish setting up arguments to pthread_start_thread */
 358   new_thread->p_start_args.start_routine = start_routine;
 359   new_thread->p_start_args.arg = arg;
 360   new_thread->p_start_args.mask = *mask;
 361   /* Raise priority of thread manager if needed */
 362   __pthread_manager_adjust_prio(new_thread->p_priority);
 363   /* Do the cloning */
 364   pid = __clone(pthread_start_thread, (void **) new_thread,
 365                 CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_SIGHAND |
 366                 __pthread_sig_cancel, new_thread);
 367   /* Check if cloning succeeded */
 368   if (pid == -1) {
 369     /* Free the stack if we allocated it */
 370     if (attr == NULL || !attr->__stackaddr_set)
 371       {
 372         munmap((caddr_t)((char *)(new_thread+1) - INITIAL_STACK_SIZE),
 373                INITIAL_STACK_SIZE);
 374         if (new_thread->p_guardsize != 0)
 375           munmap(new_thread->p_guardaddr, new_thread->p_guardsize);
 376       }
 377     __pthread_handles[sseg].h_descr = NULL;
 378     __pthread_handles[sseg].h_bottom = NULL;
 379     __pthread_handles_num--;
 380     return errno;
 381   }
 382   /* Insert new thread in doubly linked list of active threads */
 383   new_thread->p_prevlive = __pthread_main_thread;
 384   new_thread->p_nextlive = __pthread_main_thread->p_nextlive;
 385   __pthread_main_thread->p_nextlive->p_prevlive = new_thread;
 386   __pthread_main_thread->p_nextlive = new_thread;
 387   /* Set pid field of the new thread, in case we get there before the
 388      child starts. */
 389   new_thread->p_pid = pid;
 390   /* We're all set */
 391   *thread = new_thread_id;
 392   return 0;
 393 }
 394
 395
 396 /* Try to free the resources of a thread when requested by pthread_join
 397    or pthread_detach on a terminated thread. */
 398
 399 static void pthread_free(pthread_descr th)
 400 {
 401   pthread_handle handle;
 402   ASSERT(th->p_exited);
 403   /* Make the handle invalid */
 404   handle =  thread_handle(th->p_tid);
 405   __pthread_lock(&handle->h_lock, NULL);
 406   handle->h_descr = NULL;
 407   handle->h_bottom = (char *)(-1L);
 408   __pthread_unlock(&handle->h_lock);
 409 #ifdef FREE_THREAD_SELF
 410   FREE_THREAD_SELF(th, th->p_nr);
 411 #endif
 412   /* One fewer threads in __pthread_handles */
 413   __pthread_handles_num--;
 414   /* If initial thread, nothing to free */
 415   if (th == &__pthread_initial_thread) return;
 416   if (!th->p_userstack)
 417     {
 418       /* Free the stack and thread descriptor area */
 419       if (th->p_guardsize != 0)
 420         munmap(th->p_guardaddr, th->p_guardsize);
 421       munmap((caddr_t) ((char *)(th+1) - STACK_SIZE), STACK_SIZE);
 422     }
 423 }
 424
 425 /* Handle threads that have exited */
 426
 427 static void pthread_exited(pid_t pid)
 428 {
 429   pthread_descr th;
 430   int detached;
 431   /* Find thread with that pid */
 432   for (th = __pthread_main_thread->p_nextlive;
 433        th != __pthread_main_thread;
 434        th = th->p_nextlive) {
 435     if (th->p_pid == pid) {
 436       /* Remove thread from list of active threads */
 437       th->p_nextlive->p_prevlive = th->p_prevlive;
 438       th->p_prevlive->p_nextlive = th->p_nextlive;
 439       /* Mark thread as exited, and if detached, free its resources */
 440       __pthread_lock(th->p_lock, NULL);
 441       th->p_exited = 1;
 442       detached = th->p_detached;
 443       __pthread_unlock(th->p_lock);
 444       if (detached)
 445         pthread_free(th);
 446       break;
 447     }
 448   }
 449   /* If all threads have exited and the main thread is pending on a
 450      pthread_exit, wake up the main thread and terminate ourselves. */
 451   if (main_thread_exiting &&
 452       __pthread_main_thread->p_nextlive == __pthread_main_thread) {
 453     restart(__pthread_main_thread);
 454     _exit(0);
 455   }
 456 }
 457
 458 static void pthread_reap_children(void)
 459 {
 460   pid_t pid;
 461   int status;
 462
 463   while ((pid = __libc_waitpid(-1, &status, WNOHANG | __WCLONE)) > 0) {
 464     pthread_exited(pid);
 465     if (WIFSIGNALED(status)) {
 466       /* If a thread died due to a signal, send the same signal to
 467          all other threads, including the main thread. */
 468       pthread_kill_all_threads(WTERMSIG(status), 1);
 469       _exit(0);
 470     }
 471   }
 472 }
 473
 474 /* Try to free the resources of a thread when requested by pthread_join
 475    or pthread_detach on a terminated thread. */
 476
 477 static void pthread_handle_free(pthread_t th_id)
 478 {
 479   pthread_handle handle = thread_handle(th_id);
 480   pthread_descr th;
 481
 482   __pthread_lock(&handle->h_lock, NULL);
 483   if (invalid_handle(handle, th_id)) {
 484     /* pthread_reap_children has deallocated the thread already,
 485        nothing needs to be done */
 486     __pthread_unlock(&handle->h_lock);
 487     return;
 488   }
 489   th = handle->h_descr;
 490   if (th->p_exited) {
 491     __pthread_unlock(&handle->h_lock);
 492     pthread_free(th);
 493   } else {
 494     /* The Unix process of the thread is still running.
 495        Mark the thread as detached so that the thread manager will
 496        deallocate its resources when the Unix process exits. */
 497     th->p_detached = 1;
 498     __pthread_unlock(&handle->h_lock);
 499   }
 500 }
 501
 502 /* Send a signal to all running threads */
 503
 504 static void pthread_kill_all_threads(int sig, int main_thread_also)
 505 {
 506   pthread_descr th;
 507   for (th = __pthread_main_thread->p_nextlive;
 508        th != __pthread_main_thread;
 509        th = th->p_nextlive) {
 510     kill(th->p_pid, sig);
 511   }
 512   if (main_thread_also) {
 513     kill(__pthread_main_thread->p_pid, sig);
 514   }
 515 }
 516
 517 /* Process-wide exit() */
 518
 519 static void pthread_handle_exit(pthread_descr issuing_thread, int exitcode)
 520 {
 521   pthread_descr th;
 522   __pthread_exit_requested = 1;
 523   __pthread_exit_code = exitcode;
 524   /* Send the CANCEL signal to all running threads, including the main
 525      thread, but excluding the thread from which the exit request originated
 526      (that thread must complete the exit, e.g. calling atexit functions
 527      and flushing stdio buffers). */
 528   for (th = issuing_thread->p_nextlive;
 529        th != issuing_thread;
 530        th = th->p_nextlive) {
 531     kill(th->p_pid, __pthread_sig_cancel);
 532   }
 533   /* Now, wait for all these threads, so that they don't become zombies
 534      and their times are properly added to the thread manager's times. */
 535   for (th = issuing_thread->p_nextlive;
 536        th != issuing_thread;
 537        th = th->p_nextlive) {
 538     waitpid(th->p_pid, NULL, __WCLONE);
 539   }
 540   restart(issuing_thread);
 541   _exit(0);
 542 }
 543
 544 /* Handler for __pthread_sig_cancel in thread manager thread */
 545
 546 void __pthread_manager_sighandler(int sig)
 547 {
 548   terminated_children = 1;
 549 }
 550
 551 /* Adjust priority of thread manager so that it always run at a priority
 552    higher than all threads */
 553
 554 void __pthread_manager_adjust_prio(int thread_prio)
 555 {
 556   struct sched_param param;
 557
 558   if (thread_prio <= __pthread_manager_thread.p_priority) return;
 559   param.sched_priority =
 560     thread_prio < __sched_get_priority_max(SCHED_FIFO)
 561     ? thread_prio + 1 : thread_prio;
 562   __sched_setscheduler(__pthread_manager_thread.p_pid, SCHED_FIFO, &param);
 563   __pthread_manager_thread.p_priority = thread_prio;
 564 }