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 */
 164         if (__pthread_threads_debug && __pthread_sig_debug > 0)
 165           raise(__pthread_sig_debug);
 166         restart(request.req_thread);
 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) {
 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     }
 233   else
 234     {
 235       /* Allocate space for stack and thread descriptor at default address */
 236       new_thread = default_new_thread;
 237       new_thread_bottom = (char *) new_thread - STACK_SIZE;
 238       if (mmap((caddr_t)((char *)(new_thread + 1) - INITIAL_STACK_SIZE),
 239                INITIAL_STACK_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
 240                MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED | MAP_GROWSDOWN,
 241                -1, 0) == MAP_FAILED)
 242         /* Bad luck, this segment is already mapped. */
 243         return -1;
 244       /* We manage to get a stack.  Now see whether we need a guard
 245          and allocate it if necessary.  Notice that the default
 246          attributes (stack_size = STACK_SIZE - pagesize and
 247          guardsize = pagesize) do not need a guard page, since
 248          the RLIMIT_STACK soft limit prevents stacks from
 249          running into one another. */
 250       if (attr == NULL ||
 251           attr->__guardsize == 0 ||
 252           (attr->__guardsize == pagesize &&
 253            attr->__stacksize == STACK_SIZE - pagesize))
 254         {
 255           /* We don't need a guard page. */
 256           guardaddr = NULL;
 257           guardsize = 0;
 258         }
 259       else
 260         {
 261           /* Put a bad page at the bottom of the stack */
 262           stacksize = roundup(attr->__stacksize, pagesize);
 263           if (stacksize >= STACK_SIZE - pagesize)
 264             stacksize = STACK_SIZE - pagesize;
 265           guardaddr = (void *)new_thread - stacksize;
 266           guardsize = attr->__guardsize;
 267           if (mmap ((caddr_t) guardaddr, guardsize, 0, MAP_FIXED, -1, 0)
 268               == MAP_FAILED)
 269             {
 270               /* We don't make this an error.  */
 271               guardaddr = NULL;
 272               guardsize = 0;
 273             }
 274         }
 275     }
 276   *out_new_thread = new_thread;
 277   *out_new_thread_bottom = new_thread_bottom;
 278   *out_guardaddr = guardaddr;
 279   *out_guardsize = guardsize;
 280   return 0;
 281 }
 282
 283 static int pthread_handle_create(pthread_t *thread, const pthread_attr_t *attr,
 284                                  void * (*start_routine)(void *), void *arg,
 285                                  sigset_t * mask, int father_pid)
 286 {
 287   size_t sseg;
 288   int pid;
 289   pthread_descr new_thread;
 290   char * new_thread_bottom;
 291   pthread_t new_thread_id;
 292   char *guardaddr = NULL;
 293   size_t guardsize = 0;
 294   int pagesize = __getpagesize();
 295
 296   /* First check whether we have to change the policy and if yes, whether
 297      we can  do this.  Normally this should be done by examining the
 298      return value of the __sched_setscheduler call in pthread_start_thread
 299      but this is hard to implement.  FIXME  */
 300   if (attr != NULL && attr->__schedpolicy != SCHED_OTHER && geteuid () != 0)
 301     return EPERM;
 302   /* Find a free segment for the thread, and allocate a stack if needed */
 303   for (sseg = 2; ; sseg++)
 304     {
 305       if (sseg >= PTHREAD_THREADS_MAX)
 306         return EAGAIN;
 307       if (__pthread_handles[sseg].h_descr != NULL)
 308         continue;
 309       if (pthread_allocate_stack(attr, thread_segment(sseg), pagesize,
 310                                  &new_thread, &new_thread_bottom,
 311                                  &guardaddr, &guardsize) == 0)
 312         break;
 313     }
 314   /* Allocate new thread identifier */
 315   pthread_threads_counter += PTHREAD_THREADS_MAX;
 316   new_thread_id = sseg + pthread_threads_counter;
 317   /* Initialize the thread descriptor */
 318   new_thread->p_nextwaiting = NULL;
 319   new_thread->p_tid = new_thread_id;
 320   new_thread->p_priority = 0;
 321   new_thread->p_lock = &(__pthread_handles[sseg].h_lock);
 322   new_thread->p_signal = 0;
 323   new_thread->p_signal_jmp = NULL;
 324   new_thread->p_cancel_jmp = NULL;
 325   new_thread->p_terminated = 0;
 326   new_thread->p_detached = attr == NULL ? 0 : attr->__detachstate;
 327   new_thread->p_exited = 0;
 328   new_thread->p_retval = NULL;
 329   new_thread->p_joining = NULL;
 330   new_thread->p_cleanup = NULL;
 331   new_thread->p_cancelstate = PTHREAD_CANCEL_ENABLE;
 332   new_thread->p_canceltype = PTHREAD_CANCEL_DEFERRED;
 333   new_thread->p_canceled = 0;
 334   new_thread->p_errnop = &new_thread->p_errno;
 335   new_thread->p_errno = 0;
 336   new_thread->p_h_errnop = &new_thread->p_h_errno;
 337   new_thread->p_h_errno = 0;
 338   new_thread->p_in_sighandler = NULL;
 339   new_thread->p_sigwaiting = 0;
 340   new_thread->p_guardaddr = guardaddr;
 341   new_thread->p_guardsize = guardsize;
 342   new_thread->p_userstack = attr != NULL && attr->__stackaddr_set;
 343   memset (new_thread->p_specific, '\0',
 344           PTHREAD_KEY_1STLEVEL_SIZE * sizeof (new_thread->p_specific[0]));
 345   new_thread->p_self = new_thread;
 346   new_thread->p_nr = sseg;
 347   /* Initialize the thread handle */
 348   __pthread_init_lock(&__pthread_handles[sseg].h_lock);
 349   __pthread_handles[sseg].h_descr = new_thread;
 350   __pthread_handles[sseg].h_bottom = new_thread_bottom;
 351   /* Determine scheduling parameters for the thread */
 352   new_thread->p_start_args.schedpolicy = -1;
 353   if (attr != NULL) {
 354     switch(attr->__inheritsched) {
 355     case PTHREAD_EXPLICIT_SCHED:
 356       new_thread->p_start_args.schedpolicy = attr->__schedpolicy;
 357       memcpy (&new_thread->p_start_args.schedparam, &attr->__schedparam,
 358               sizeof (struct sched_param));
 359       break;
 360     case PTHREAD_INHERIT_SCHED:
 361       /* schedpolicy doesn't need to be set, only get priority */
 362       __sched_getparam(father_pid, &new_thread->p_start_args.schedparam);
 363       break;
 364     }
 365     new_thread->p_priority =
 366       new_thread->p_start_args.schedparam.sched_priority;
 367   }
 368   /* Finish setting up arguments to pthread_start_thread */
 369   new_thread->p_start_args.start_routine = start_routine;
 370   new_thread->p_start_args.arg = arg;
 371   new_thread->p_start_args.mask = *mask;
 372   /* Raise priority of thread manager if needed */
 373   __pthread_manager_adjust_prio(new_thread->p_priority);
 374   /* Do the cloning */
 375   pid = __clone(pthread_start_thread, (void **) new_thread,
 376                 CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_SIGHAND |
 377 #ifdef CLONE_PTRACE
 378                 CLONE_PTRACE |
 379 #endif
 380                 __pthread_sig_cancel, new_thread);
 381   /* Check if cloning succeeded */
 382   if (pid == -1) {
 383     /* Free the stack if we allocated it */
 384     if (attr == NULL || !attr->__stackaddr_set)
 385       {
 386         munmap((caddr_t)((char *)(new_thread+1) - INITIAL_STACK_SIZE),
 387                INITIAL_STACK_SIZE);
 388         if (new_thread->p_guardsize != 0)
 389           munmap(new_thread->p_guardaddr, new_thread->p_guardsize);
 390       }
 391     __pthread_handles[sseg].h_descr = NULL;
 392     __pthread_handles[sseg].h_bottom = NULL;
 393     __pthread_handles_num--;
 394     return errno;
 395   }
 396   /* Insert new thread in doubly linked list of active threads */
 397   new_thread->p_prevlive = __pthread_main_thread;
 398   new_thread->p_nextlive = __pthread_main_thread->p_nextlive;
 399   __pthread_main_thread->p_nextlive->p_prevlive = new_thread;
 400   __pthread_main_thread->p_nextlive = new_thread;
 401   /* Set pid field of the new thread, in case we get there before the
 402      child starts. */
 403   new_thread->p_pid = pid;
 404   /* We're all set */
 405   *thread = new_thread_id;
 406   return 0;
 407 }
 408
 409
 410 /* Try to free the resources of a thread when requested by pthread_join
 411    or pthread_detach on a terminated thread. */
 412
 413 static void pthread_free(pthread_descr th)
 414 {
 415   pthread_handle handle;
 416   ASSERT(th->p_exited);
 417   /* Make the handle invalid */
 418   handle =  thread_handle(th->p_tid);
 419   __pthread_lock(&handle->h_lock, NULL);
 420   handle->h_descr = NULL;
 421   handle->h_bottom = (char *)(-1L);
 422   __pthread_unlock(&handle->h_lock);
 423 #ifdef FREE_THREAD_SELF
 424   FREE_THREAD_SELF(th, th->p_nr);
 425 #endif
 426   /* One fewer threads in __pthread_handles */
 427   __pthread_handles_num--;
 428   /* If initial thread, nothing to free */
 429   if (th == &__pthread_initial_thread) return;
 430   if (!th->p_userstack)
 431     {
 432       /* Free the stack and thread descriptor area */
 433       if (th->p_guardsize != 0)
 434         munmap(th->p_guardaddr, th->p_guardsize);
 435       munmap((caddr_t) ((char *)(th+1) - STACK_SIZE), STACK_SIZE);
 436     }
 437 }
 438
 439 /* Handle threads that have exited */
 440
 441 static void pthread_exited(pid_t pid)
 442 {
 443   pthread_descr th;
 444   int detached;
 445   /* Find thread with that pid */
 446   for (th = __pthread_main_thread->p_nextlive;
 447        th != __pthread_main_thread;
 448        th = th->p_nextlive) {
 449     if (th->p_pid == pid) {
 450       /* Remove thread from list of active threads */
 451       th->p_nextlive->p_prevlive = th->p_prevlive;
 452       th->p_prevlive->p_nextlive = th->p_nextlive;
 453       /* Mark thread as exited, and if detached, free its resources */
 454       __pthread_lock(th->p_lock, NULL);
 455       th->p_exited = 1;
 456       detached = th->p_detached;
 457       __pthread_unlock(th->p_lock);
 458       if (detached)
 459         pthread_free(th);
 460       break;
 461     }
 462   }
 463   /* If all threads have exited and the main thread is pending on a
 464      pthread_exit, wake up the main thread and terminate ourselves. */
 465   if (main_thread_exiting &&
 466       __pthread_main_thread->p_nextlive == __pthread_main_thread) {
 467     restart(__pthread_main_thread);
 468     _exit(0);
 469   }
 470 }
 471
 472 static void pthread_reap_children(void)
 473 {
 474   pid_t pid;
 475   int status;
 476
 477   while ((pid = __libc_waitpid(-1, &status, WNOHANG | __WCLONE)) > 0) {
 478     pthread_exited(pid);
 479     if (WIFSIGNALED(status)) {
 480       /* If a thread died due to a signal, send the same signal to
 481          all other threads, including the main thread. */
 482       pthread_kill_all_threads(WTERMSIG(status), 1);
 483       _exit(0);
 484     }
 485   }
 486 }
 487
 488 /* Try to free the resources of a thread when requested by pthread_join
 489    or pthread_detach on a terminated thread. */
 490
 491 static void pthread_handle_free(pthread_t th_id)
 492 {
 493   pthread_handle handle = thread_handle(th_id);
 494   pthread_descr th;
 495
 496   __pthread_lock(&handle->h_lock, NULL);
 497   if (invalid_handle(handle, th_id)) {
 498     /* pthread_reap_children has deallocated the thread already,
 499        nothing needs to be done */
 500     __pthread_unlock(&handle->h_lock);
 501     return;
 502   }
 503   th = handle->h_descr;
 504   if (th->p_exited) {
 505     __pthread_unlock(&handle->h_lock);
 506     pthread_free(th);
 507   } else {
 508     /* The Unix process of the thread is still running.
 509        Mark the thread as detached so that the thread manager will
 510        deallocate its resources when the Unix process exits. */
 511     th->p_detached = 1;
 512     __pthread_unlock(&handle->h_lock);
 513   }
 514 }
 515
 516 /* Send a signal to all running threads */
 517
 518 static void pthread_kill_all_threads(int sig, int main_thread_also)
 519 {
 520   pthread_descr th;
 521   for (th = __pthread_main_thread->p_nextlive;
 522        th != __pthread_main_thread;
 523        th = th->p_nextlive) {
 524     kill(th->p_pid, sig);
 525   }
 526   if (main_thread_also) {
 527     kill(__pthread_main_thread->p_pid, sig);
 528   }
 529 }
 530
 531 /* Process-wide exit() */
 532
 533 static void pthread_handle_exit(pthread_descr issuing_thread, int exitcode)
 534 {
 535   pthread_descr th;
 536   __pthread_exit_requested = 1;
 537   __pthread_exit_code = exitcode;
 538   /* Send the CANCEL signal to all running threads, including the main
 539      thread, but excluding the thread from which the exit request originated
 540      (that thread must complete the exit, e.g. calling atexit functions
 541      and flushing stdio buffers). */
 542   for (th = issuing_thread->p_nextlive;
 543        th != issuing_thread;
 544        th = th->p_nextlive) {
 545     kill(th->p_pid, __pthread_sig_cancel);
 546   }
 547   /* Now, wait for all these threads, so that they don't become zombies
 548      and their times are properly added to the thread manager's times. */
 549   for (th = issuing_thread->p_nextlive;
 550        th != issuing_thread;
 551        th = th->p_nextlive) {
 552     waitpid(th->p_pid, NULL, __WCLONE);
 553   }
 554   restart(issuing_thread);
 555   _exit(0);
 556 }
 557
 558 /* Handler for __pthread_sig_cancel in thread manager thread */
 559
 560 void __pthread_manager_sighandler(int sig)
 561 {
 562   terminated_children = 1;
 563 }
 564
 565 /* Adjust priority of thread manager so that it always run at a priority
 566    higher than all threads */
 567
 568 void __pthread_manager_adjust_prio(int thread_prio)
 569 {
 570   struct sched_param param;
 571
 572   if (thread_prio <= __pthread_manager_thread.p_priority) return;
 573   param.sched_priority =
 574     thread_prio < __sched_get_priority_max(SCHED_FIFO)
 575     ? thread_prio + 1 : thread_prio;
 576   __sched_setscheduler(__pthread_manager_thread.p_pid, SCHED_FIFO, &param);
 577   __pthread_manager_thread.p_priority = thread_prio;
 578 }