kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/config.h>
   8 #include <linux/mm.h>
   9 #include <linux/slab.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/module.h>
  13 #include <linux/completion.h>
  14 #include <linux/personality.h>
  15 #include <linux/tty.h>
  16 #include <linux/namespace.h>
  17 #include <linux/security.h>
  18 #include <linux/acct.h>
  19 #include <linux/file.h>
  20 #include <linux/binfmts.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/profile.h>
  23 #include <linux/mount.h>
  24 #include <linux/proc_fs.h>
  25 #include <linux/mempolicy.h>
  26
  27 #include <asm/uaccess.h>
  28 #include <asm/unistd.h>
  29 #include <asm/pgtable.h>
  30 #include <asm/mmu_context.h>
  31
  32 extern void sem_exit (void);
  33 extern struct task_struct *child_reaper;
  34
  35 int getrusage(struct task_struct *, int, struct rusage __user *);
  36
  37 static void __unhash_process(struct task_struct *p)
  38 {
  39         nr_threads--;
  40         /* tasklist_lock is held, is this sufficient? */
  41         if (p->vx_info) {
  42                 atomic_dec(&p->vx_info->cacct.nr_threads);
  43                 atomic_dec(&p->vx_info->limit.res[RLIMIT_NPROC]);
  44         }
  45         detach_pid(p, PIDTYPE_PID);
  46         detach_pid(p, PIDTYPE_TGID);
  47         if (thread_group_leader(p)) {
  48                 detach_pid(p, PIDTYPE_PGID);
  49                 detach_pid(p, PIDTYPE_SID);
  50                 if (p->pid)
  51                         __get_cpu_var(process_counts)--;
  52         }
  53
  54         REMOVE_LINKS(p);
  55 }
  56
  57 void release_task(struct task_struct * p)
  58 {
  59         int zap_leader;
  60         task_t *leader;
  61         struct dentry *proc_dentry;
  62
  63 repeat:
  64         BUG_ON(p->state < TASK_ZOMBIE);
  65
  66         atomic_dec(&p->user->processes);
  67         spin_lock(&p->proc_lock);
  68         proc_dentry = proc_pid_unhash(p);
  69         write_lock_irq(&tasklist_lock);
  70         if (unlikely(p->ptrace))
  71                 __ptrace_unlink(p);
  72         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  73         __exit_signal(p);
  74         __exit_sighand(p);
  75         __unhash_process(p);
  76
  77         /*
  78          * If we are the last non-leader member of the thread
  79          * group, and the leader is zombie, then notify the
  80          * group leader's parent process. (if it wants notification.)
  81          */
  82         zap_leader = 0;
  83         leader = p->group_leader;
  84         if (leader != p && thread_group_empty(leader) && leader->state == TASK_ZOMBIE) {
  85                 BUG_ON(leader->exit_signal == -1);
  86                 do_notify_parent(leader, leader->exit_signal);
  87                 /*
  88                  * If we were the last child thread and the leader has
  89                  * exited already, and the leader's parent ignores SIGCHLD,
  90                  * then we are the one who should release the leader.
  91                  *
  92                  * do_notify_parent() will have marked it self-reaping in
  93                  * that case.
  94                  */
  95                 zap_leader = (leader->exit_signal == -1);
  96         }
  97
  98         p->parent->cutime += p->utime + p->cutime;
  99         p->parent->cstime += p->stime + p->cstime;
 100         p->parent->cmin_flt += p->min_flt + p->cmin_flt;
 101         p->parent->cmaj_flt += p->maj_flt + p->cmaj_flt;
 102         p->parent->cnvcsw += p->nvcsw + p->cnvcsw;
 103         p->parent->cnivcsw += p->nivcsw + p->cnivcsw;
 104         sched_exit(p);
 105         write_unlock_irq(&tasklist_lock);
 106         spin_unlock(&p->proc_lock);
 107         proc_pid_flush(proc_dentry);
 108         release_thread(p);
 109         put_task_struct(p);
 110
 111         p = leader;
 112         if (unlikely(zap_leader))
 113                 goto repeat;
 114 }
 115
 116 /* we are using it only for SMP init */
 117
 118 void unhash_process(struct task_struct *p)
 119 {
 120         struct dentry *proc_dentry;
 121
 122         spin_lock(&p->proc_lock);
 123         proc_dentry = proc_pid_unhash(p);
 124         write_lock_irq(&tasklist_lock);
 125         __unhash_process(p);
 126         write_unlock_irq(&tasklist_lock);
 127         spin_unlock(&p->proc_lock);
 128         proc_pid_flush(proc_dentry);
 129 }
 130
 131 /*
 132  * This checks not only the pgrp, but falls back on the pid if no
 133  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 134  * without this...
 135  */
 136 int session_of_pgrp(int pgrp)
 137 {
 138         struct task_struct *p;
 139         struct list_head *l;
 140         struct pid *pid;
 141         int sid = -1;
 142
 143         read_lock(&tasklist_lock);
 144         for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid)
 145                 if (p->signal->session > 0) {
 146                         sid = p->signal->session;
 147                         goto out;
 148                 }
 149         p = find_task_by_pid(pgrp);
 150         if (p)
 151                 sid = p->signal->session;
 152 out:
 153         read_unlock(&tasklist_lock);
 154
 155         return sid;
 156 }
 157
 158 /*
 159  * Determine if a process group is "orphaned", according to the POSIX
 160  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 161  * by terminal-generated stop signals.  Newly orphaned process groups are
 162  * to receive a SIGHUP and a SIGCONT.
 163  *
 164  * "I ask you, have you ever known what it is to be an orphan?"
 165  */
 166 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 167 {
 168         struct task_struct *p;
 169         struct list_head *l;
 170         struct pid *pid;
 171         int ret = 1;
 172
 173         for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
 174                 if (p == ignored_task
 175                                 || p->state >= TASK_ZOMBIE
 176                                 || p->real_parent->pid == 1)
 177                         continue;
 178                 if (process_group(p->real_parent) != pgrp
 179                             && p->real_parent->signal->session == p->signal->session) {
 180                         ret = 0;
 181                         break;
 182                 }
 183         }
 184         return ret;     /* (sighing) "Often!" */
 185 }
 186
 187 int is_orphaned_pgrp(int pgrp)
 188 {
 189         int retval;
 190
 191         read_lock(&tasklist_lock);
 192         retval = will_become_orphaned_pgrp(pgrp, NULL);
 193         read_unlock(&tasklist_lock);
 194
 195         return retval;
 196 }
 197
 198 static inline int has_stopped_jobs(int pgrp)
 199 {
 200         int retval = 0;
 201         struct task_struct *p;
 202         struct list_head *l;
 203         struct pid *pid;
 204
 205         for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
 206                 if (p->state != TASK_STOPPED)
 207                         continue;
 208
 209                 /* If p is stopped by a debugger on a signal that won't
 210                    stop it, then don't count p as stopped.  This isn't
 211                    perfect but it's a good approximation.  */
 212                 if (unlikely (p->ptrace)
 213                     && p->exit_code != SIGSTOP
 214                     && p->exit_code != SIGTSTP
 215                     && p->exit_code != SIGTTOU
 216                     && p->exit_code != SIGTTIN)
 217                         continue;
 218
 219                 retval = 1;
 220                 break;
 221         }
 222         return retval;
 223 }
 224
 225 /**
 226  * reparent_to_init() - Reparent the calling kernel thread to the init task.
 227  *
 228  * If a kernel thread is launched as a result of a system call, or if
 229  * it ever exits, it should generally reparent itself to init so that
 230  * it is correctly cleaned up on exit.
 231  *
 232  * The various task state such as scheduling policy and priority may have
 233  * been inherited from a user process, so we reset them to sane values here.
 234  *
 235  * NOTE that reparent_to_init() gives the caller full capabilities.
 236  */
 237 void reparent_to_init(void)
 238 {
 239         write_lock_irq(&tasklist_lock);
 240
 241         ptrace_unlink(current);
 242         /* Reparent to init */
 243         REMOVE_LINKS(current);
 244         /* FIXME handle vchild_reaper/initpid */
 245         current->parent = child_reaper;
 246         current->real_parent = child_reaper;
 247         SET_LINKS(current);
 248
 249         /* Set the exit signal to SIGCHLD so we signal init on exit */
 250         current->exit_signal = SIGCHLD;
 251
 252         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
 253                 set_user_nice(current, 0);
 254         /* cpus_allowed? */
 255         /* rt_priority? */
 256         /* signals? */
 257         security_task_reparent_to_init(current);
 258         memcpy(current->rlim, init_task.rlim, sizeof(*(current->rlim)));
 259         atomic_inc(&(INIT_USER->__count));
 260         switch_uid(INIT_USER);
 261
 262         write_unlock_irq(&tasklist_lock);
 263 }
 264
 265 void __set_special_pids(pid_t session, pid_t pgrp)
 266 {
 267         struct task_struct *curr = current;
 268
 269         if (curr->signal->session != session) {
 270                 detach_pid(curr, PIDTYPE_SID);
 271                 curr->signal->session = session;
 272                 attach_pid(curr, PIDTYPE_SID, session);
 273         }
 274         if (process_group(curr) != pgrp) {
 275                 detach_pid(curr, PIDTYPE_PGID);
 276                 curr->signal->pgrp = pgrp;
 277                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 278         }
 279 }
 280
 281 void set_special_pids(pid_t session, pid_t pgrp)
 282 {
 283         write_lock_irq(&tasklist_lock);
 284         __set_special_pids(session, pgrp);
 285         write_unlock_irq(&tasklist_lock);
 286 }
 287
 288 /*
 289  * Let kernel threads use this to say that they
 290  * allow a certain signal (since daemonize() will
 291  * have disabled all of them by default).
 292  */
 293 int allow_signal(int sig)
 294 {
 295         if (sig < 1 || sig > _NSIG)
 296                 return -EINVAL;
 297
 298         spin_lock_irq(&current->sighand->siglock);
 299         sigdelset(&current->blocked, sig);
 300         if (!current->mm) {
 301                 /* Kernel threads handle their own signals.
 302                    Let the signal code know it'll be handled, so
 303                    that they don't get converted to SIGKILL or
 304                    just silently dropped */
 305                 current->sighand->action[(sig)-1].sa.sa_handler = (void *)2;
 306         }
 307         recalc_sigpending();
 308         spin_unlock_irq(&current->sighand->siglock);
 309         return 0;
 310 }
 311
 312 EXPORT_SYMBOL(allow_signal);
 313
 314 int disallow_signal(int sig)
 315 {
 316         if (sig < 1 || sig > _NSIG)
 317                 return -EINVAL;
 318
 319         spin_lock_irq(&current->sighand->siglock);
 320         sigaddset(&current->blocked, sig);
 321         recalc_sigpending();
 322         spin_unlock_irq(&current->sighand->siglock);
 323         return 0;
 324 }
 325
 326 EXPORT_SYMBOL(disallow_signal);
 327
 328 /*
 329  *      Put all the gunge required to become a kernel thread without
 330  *      attached user resources in one place where it belongs.
 331  */
 332
 333 void daemonize(const char *name, ...)
 334 {
 335         va_list args;
 336         struct fs_struct *fs;
 337         sigset_t blocked;
 338
 339         va_start(args, name);
 340         vsnprintf(current->comm, sizeof(current->comm), name, args);
 341         va_end(args);
 342
 343         /*
 344          * If we were started as result of loading a module, close all of the
 345          * user space pages.  We don't need them, and if we didn't close them
 346          * they would be locked into memory.
 347          */
 348         exit_mm(current);
 349
 350         set_special_pids(1, 1);
 351         current->signal->tty = NULL;
 352
 353         /* Block and flush all signals */
 354         sigfillset(&blocked);
 355         sigprocmask(SIG_BLOCK, &blocked, NULL);
 356         flush_signals(current);
 357
 358         /* Become as one with the init task */
 359
 360         exit_fs(current);       /* current->fs->count--; */
 361         fs = init_task.fs;
 362         current->fs = fs;
 363         atomic_inc(&fs->count);
 364         exit_files(current);
 365         current->files = init_task.files;
 366         atomic_inc(&current->files->count);
 367
 368         reparent_to_init();
 369 }
 370
 371 EXPORT_SYMBOL(daemonize);
 372
 373 static inline void close_files(struct files_struct * files)
 374 {
 375         int i, j;
 376
 377         j = 0;
 378         for (;;) {
 379                 unsigned long set;
 380                 i = j * __NFDBITS;
 381                 if (i >= files->max_fdset || i >= files->max_fds)
 382                         break;
 383                 set = files->open_fds->fds_bits[j++];
 384                 while (set) {
 385                         if (set & 1) {
 386                                 struct file * file = xchg(&files->fd[i], NULL);
 387                                 if (file)
 388                                         filp_close(file, files);
 389                                 vx_openfd_dec(fd);
 390                         }
 391                         i++;
 392                         set >>= 1;
 393                 }
 394         }
 395 }
 396
 397 struct files_struct *get_files_struct(struct task_struct *task)
 398 {
 399         struct files_struct *files;
 400
 401         task_lock(task);
 402         files = task->files;
 403         if (files)
 404                 atomic_inc(&files->count);
 405         task_unlock(task);
 406
 407         return files;
 408 }
 409
 410 void fastcall put_files_struct(struct files_struct *files)
 411 {
 412         if (atomic_dec_and_test(&files->count)) {
 413                 close_files(files);
 414                 /*
 415                  * Free the fd and fdset arrays if we expanded them.
 416                  */
 417                 if (files->fd != &files->fd_array[0])
 418                         free_fd_array(files->fd, files->max_fds);
 419                 if (files->max_fdset > __FD_SETSIZE) {
 420                         free_fdset(files->open_fds, files->max_fdset);
 421                         free_fdset(files->close_on_exec, files->max_fdset);
 422                 }
 423                 kmem_cache_free(files_cachep, files);
 424         }
 425 }
 426
 427 EXPORT_SYMBOL(put_files_struct);
 428
 429 static inline void __exit_files(struct task_struct *tsk)
 430 {
 431         struct files_struct * files = tsk->files;
 432
 433         if (files) {
 434                 task_lock(tsk);
 435                 tsk->files = NULL;
 436                 task_unlock(tsk);
 437                 put_files_struct(files);
 438         }
 439 }
 440
 441 void exit_files(struct task_struct *tsk)
 442 {
 443         __exit_files(tsk);
 444 }
 445
 446 static inline void __put_fs_struct(struct fs_struct *fs)
 447 {
 448         /* No need to hold fs->lock if we are killing it */
 449         if (atomic_dec_and_test(&fs->count)) {
 450                 dput(fs->root);
 451                 mntput(fs->rootmnt);
 452                 dput(fs->pwd);
 453                 mntput(fs->pwdmnt);
 454                 if (fs->altroot) {
 455                         dput(fs->altroot);
 456                         mntput(fs->altrootmnt);
 457                 }
 458                 kmem_cache_free(fs_cachep, fs);
 459         }
 460 }
 461
 462 void put_fs_struct(struct fs_struct *fs)
 463 {
 464         __put_fs_struct(fs);
 465 }
 466
 467 static inline void __exit_fs(struct task_struct *tsk)
 468 {
 469         struct fs_struct * fs = tsk->fs;
 470
 471         if (fs) {
 472                 task_lock(tsk);
 473                 tsk->fs = NULL;
 474                 task_unlock(tsk);
 475                 __put_fs_struct(fs);
 476         }
 477 }
 478
 479 void exit_fs(struct task_struct *tsk)
 480 {
 481         __exit_fs(tsk);
 482 }
 483
 484 EXPORT_SYMBOL_GPL(exit_fs);
 485
 486 /*
 487  * Turn us into a lazy TLB process if we
 488  * aren't already..
 489  */
 490 static inline void __exit_mm(struct task_struct * tsk)
 491 {
 492         struct mm_struct *mm = tsk->mm;
 493
 494         mm_release(tsk, mm);
 495         if (!mm)
 496                 return;
 497         /*
 498          * Serialize with any possible pending coredump.
 499          * We must hold mmap_sem around checking core_waiters
 500          * and clearing tsk->mm.  The core-inducing thread
 501          * will increment core_waiters for each thread in the
 502          * group with ->mm != NULL.
 503          */
 504         down_read(&mm->mmap_sem);
 505         if (mm->core_waiters) {
 506                 up_read(&mm->mmap_sem);
 507                 down_write(&mm->mmap_sem);
 508                 if (!--mm->core_waiters)
 509                         complete(mm->core_startup_done);
 510                 up_write(&mm->mmap_sem);
 511
 512                 wait_for_completion(&mm->core_done);
 513                 down_read(&mm->mmap_sem);
 514         }
 515         atomic_inc(&mm->mm_count);
 516         if (mm != tsk->active_mm) BUG();
 517         /* more a memory barrier than a real lock */
 518         task_lock(tsk);
 519         tsk->mm = NULL;
 520         up_read(&mm->mmap_sem);
 521         enter_lazy_tlb(mm, current);
 522         task_unlock(tsk);
 523         mmput(mm);
 524 }
 525
 526 void exit_mm(struct task_struct *tsk)
 527 {
 528         __exit_mm(tsk);
 529 }
 530
 531 EXPORT_SYMBOL(exit_mm);
 532
 533 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 534 {
 535         /*
 536          * Make sure we're not reparenting to ourselves and that
 537          * the parent is not a zombie.
 538          */
 539         if (p == reaper || reaper->state >= TASK_ZOMBIE)
 540                 p->real_parent = child_reaper;
 541         else
 542                 p->real_parent = reaper;
 543         if (p->parent == p->real_parent)
 544                 BUG();
 545 }
 546
 547 static inline void reparent_thread(task_t *p, task_t *father, int traced)
 548 {
 549         /* We don't want people slaying init.  */
 550         if (p->exit_signal != -1)
 551                 p->exit_signal = SIGCHLD;
 552         p->self_exec_id++;
 553
 554         if (p->pdeath_signal)
 555                 /* We already hold the tasklist_lock here.  */
 556                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
 557
 558         /* Move the child from its dying parent to the new one.  */
 559         if (unlikely(traced)) {
 560                 /* Preserve ptrace links if someone else is tracing this child.  */
 561                 list_del_init(&p->ptrace_list);
 562                 if (p->parent != p->real_parent)
 563                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 564         } else {
 565                 /* If this child is being traced, then we're the one tracing it
 566                  * anyway, so let go of it.
 567                  */
 568                 p->ptrace = 0;
 569                 list_del_init(&p->sibling);
 570                 p->parent = p->real_parent;
 571                 list_add_tail(&p->sibling, &p->parent->children);
 572
 573                 /* If we'd notified the old parent about this child's death,
 574                  * also notify the new parent.
 575                  */
 576                 if (p->state == TASK_ZOMBIE && p->exit_signal != -1 &&
 577                     thread_group_empty(p))
 578                         do_notify_parent(p, p->exit_signal);
 579         }
 580
 581         /*
 582          * process group orphan check
 583          * Case ii: Our child is in a different pgrp
 584          * than we are, and it was the only connection
 585          * outside, so the child pgrp is now orphaned.
 586          */
 587         if ((process_group(p) != process_group(father)) &&
 588             (p->signal->session == father->signal->session)) {
 589                 int pgrp = process_group(p);
 590
 591                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 592                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
 593                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
 594                 }
 595         }
 596 }
 597
 598 /*
 599  * When we die, we re-parent all our children.
 600  * Try to give them to another thread in our thread
 601  * group, and if no such member exists, give it to
 602  * the global child reaper process (ie "init")
 603  */
 604 static inline void forget_original_parent(struct task_struct * father)
 605 {
 606         struct task_struct *p, *reaper = father;
 607         struct list_head *_p, *_n;
 608
 609         /* FIXME handle vchild_reaper/initpid */
 610         reaper = father->group_leader;
 611         if (reaper == father)
 612                 reaper = child_reaper;
 613
 614         /*
 615          * There are only two places where our children can be:
 616          *
 617          * - in our child list
 618          * - in our ptraced child list
 619          *
 620          * Search them and reparent children.
 621          */
 622         list_for_each_safe(_p, _n, &father->children) {
 623                 p = list_entry(_p,struct task_struct,sibling);
 624                 if (father == p->real_parent) {
 625                         choose_new_parent(p, reaper, child_reaper);
 626                         reparent_thread(p, father, 0);
 627                 } else {
 628                         ptrace_unlink (p);
 629                         if (p->state == TASK_ZOMBIE && p->exit_signal != -1 &&
 630                             thread_group_empty(p))
 631                                 do_notify_parent(p, p->exit_signal);
 632                 }
 633         }
 634         list_for_each_safe(_p, _n, &father->ptrace_children) {
 635                 p = list_entry(_p,struct task_struct,ptrace_list);
 636                 choose_new_parent(p, reaper, child_reaper);
 637                 reparent_thread(p, father, 1);
 638         }
 639 }
 640
 641 /*
 642  * Send signals to all our closest relatives so that they know
 643  * to properly mourn us..
 644  */
 645 static void exit_notify(struct task_struct *tsk)
 646 {
 647         int state;
 648         struct task_struct *t;
 649
 650         if (signal_pending(tsk) && !tsk->signal->group_exit
 651             && !thread_group_empty(tsk)) {
 652                 /*
 653                  * This occurs when there was a race between our exit
 654                  * syscall and a group signal choosing us as the one to
 655                  * wake up.  It could be that we are the only thread
 656                  * alerted to check for pending signals, but another thread
 657                  * should be woken now to take the signal since we will not.
 658                  * Now we'll wake all the threads in the group just to make
 659                  * sure someone gets all the pending signals.
 660                  */
 661                 read_lock(&tasklist_lock);
 662                 spin_lock_irq(&tsk->sighand->siglock);
 663                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 664                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 665                                 recalc_sigpending_tsk(t);
 666                                 if (signal_pending(t))
 667                                         signal_wake_up(t, 0);
 668                         }
 669                 spin_unlock_irq(&tsk->sighand->siglock);
 670                 read_unlock(&tasklist_lock);
 671         }
 672
 673         write_lock_irq(&tasklist_lock);
 674
 675         /*
 676          * This does two things:
 677          *
 678          * A.  Make init inherit all the child processes
 679          * B.  Check to see if any process groups have become orphaned
 680          *      as a result of our exiting, and if they have any stopped
 681          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 682          */
 683
 684         forget_original_parent(tsk);
 685         BUG_ON(!list_empty(&tsk->children));
 686
 687         /*
 688          * Check to see if any process groups have become orphaned
 689          * as a result of our exiting, and if they have any stopped
 690          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 691          *
 692          * Case i: Our father is in a different pgrp than we are
 693          * and we were the only connection outside, so our pgrp
 694          * is about to become orphaned.
 695          */
 696
 697         t = tsk->real_parent;
 698
 699         if ((process_group(t) != process_group(tsk)) &&
 700             (t->signal->session == tsk->signal->session) &&
 701             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 702             has_stopped_jobs(process_group(tsk))) {
 703                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
 704                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
 705         }
 706
 707         /* Let father know we died
 708          *
 709          * Thread signals are configurable, but you aren't going to use
 710          * that to send signals to arbitary processes.
 711          * That stops right now.
 712          *
 713          * If the parent exec id doesn't match the exec id we saved
 714          * when we started then we know the parent has changed security
 715          * domain.
 716          *
 717          * If our self_exec id doesn't match our parent_exec_id then
 718          * we have changed execution domain as these two values started
 719          * the same after a fork.
 720          *
 721          */
 722
 723         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 724             ( tsk->parent_exec_id != t->self_exec_id  ||
 725               tsk->self_exec_id != tsk->parent_exec_id)
 726             && !capable(CAP_KILL))
 727                 tsk->exit_signal = SIGCHLD;
 728
 729
 730         /* If something other than our normal parent is ptracing us, then
 731          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 732          * only has special meaning to our real parent.
 733          */
 734         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 735                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 736                 do_notify_parent(tsk, signal);
 737         } else if (tsk->ptrace) {
 738                 do_notify_parent(tsk, SIGCHLD);
 739         }
 740
 741         state = TASK_ZOMBIE;
 742         if (tsk->exit_signal == -1 && tsk->ptrace == 0)
 743                 state = TASK_DEAD;
 744         tsk->state = state;
 745         tsk->flags |= PF_DEAD;
 746
 747         /*
 748          * In the preemption case it must be impossible for the task
 749          * to get runnable again, so use "_raw_" unlock to keep
 750          * preempt_count elevated until we schedule().
 751          *
 752          * To avoid deadlock on SMP, interrupts must be unmasked.  If we
 753          * don't, subsequently called functions (e.g, wait_task_inactive()
 754          * via release_task()) will spin, with interrupt flags
 755          * unwittingly blocked, until the other task sleeps.  That task
 756          * may itself be waiting for smp_call_function() to answer and
 757          * complete, and with interrupts blocked that will never happen.
 758          */
 759         _raw_write_unlock(&tasklist_lock);
 760         local_irq_enable();
 761
 762         /* If the process is dead, release it - nobody will wait for it */
 763         if (state == TASK_DEAD)
 764                 release_task(tsk);
 765
 766 }
 767
 768 asmlinkage NORET_TYPE void do_exit(long code)
 769 {
 770         struct task_struct *tsk = current;
 771
 772         if (unlikely(in_interrupt()))
 773                 panic("Aiee, killing interrupt handler!");
 774         if (unlikely(!tsk->pid))
 775                 panic("Attempted to kill the idle task!");
 776         if (unlikely(tsk->pid == 1))
 777                 panic("Attempted to kill init!");
 778         if (tsk->io_context)
 779                 exit_io_context();
 780         tsk->flags |= PF_EXITING;
 781         del_timer_sync(&tsk->real_timer);
 782
 783         if (unlikely(in_atomic()))
 784                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 785                                 current->comm, current->pid,
 786                                 preempt_count());
 787
 788         profile_exit_task(tsk);
 789
 790         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 791                 current->ptrace_message = code;
 792                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 793         }
 794
 795         acct_process(code);
 796         if (current->tux_info) {
 797 #ifdef CONFIG_TUX_DEBUG
 798                 printk("Possibly unexpected TUX-thread exit(%ld) at %p?\n",
 799                         code, __builtin_return_address(0));
 800 #endif
 801                 current->tux_exit();
 802         }
 803         __exit_mm(tsk);
 804
 805         exit_sem(tsk);
 806         __exit_files(tsk);
 807         __exit_fs(tsk);
 808         exit_namespace(tsk);
 809         exit_thread();
 810 #ifdef CONFIG_NUMA
 811         mpol_free(tsk->mempolicy);
 812 #endif
 813
 814         if (tsk->signal->leader)
 815                 disassociate_ctty(1);
 816
 817         module_put(tsk->thread_info->exec_domain->module);
 818         if (tsk->binfmt)
 819                 module_put(tsk->binfmt->module);
 820
 821         tsk->exit_code = code;
 822         exit_notify(tsk);
 823         schedule();
 824         BUG();
 825         /* Avoid "noreturn function does return".  */
 826         for (;;) ;
 827 }
 828
 829 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 830 {
 831         if (comp)
 832                 complete(comp);
 833
 834         do_exit(code);
 835 }
 836
 837 EXPORT_SYMBOL(complete_and_exit);
 838
 839 asmlinkage long sys_exit(int error_code)
 840 {
 841         do_exit((error_code&0xff)<<8);
 842 }
 843
 844 task_t fastcall *next_thread(task_t *p)
 845 {
 846         struct pid_link *link = p->pids + PIDTYPE_TGID;
 847         struct list_head *tmp, *head = &link->pidptr->task_list;
 848
 849 #ifdef CONFIG_SMP
 850         if (!p->sighand)
 851                 BUG();
 852         if (!spin_is_locked(&p->sighand->siglock) &&
 853                                 !rwlock_is_locked(&tasklist_lock))
 854                 BUG();
 855 #endif
 856         tmp = link->pid_chain.next;
 857         if (tmp == head)
 858                 tmp = head->next;
 859
 860         return pid_task(tmp, PIDTYPE_TGID);
 861 }
 862
 863 EXPORT_SYMBOL(next_thread);
 864
 865 /*
 866  * Take down every thread in the group.  This is called by fatal signals
 867  * as well as by sys_exit_group (below).
 868  */
 869 NORET_TYPE void
 870 do_group_exit(int exit_code)
 871 {
 872         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 873
 874         if (current->signal->group_exit)
 875                 exit_code = current->signal->group_exit_code;
 876         else if (!thread_group_empty(current)) {
 877                 struct signal_struct *const sig = current->signal;
 878                 struct sighand_struct *const sighand = current->sighand;
 879                 read_lock(&tasklist_lock);
 880                 spin_lock_irq(&sighand->siglock);
 881                 if (sig->group_exit)
 882                         /* Another thread got here before we took the lock.  */
 883                         exit_code = sig->group_exit_code;
 884                 else {
 885                         sig->group_exit = 1;
 886                         sig->group_exit_code = exit_code;
 887                         zap_other_threads(current);
 888                 }
 889                 spin_unlock_irq(&sighand->siglock);
 890                 read_unlock(&tasklist_lock);
 891         }
 892
 893         do_exit(exit_code);
 894         /* NOTREACHED */
 895 }
 896
 897 /*
 898  * this kills every thread in the thread group. Note that any externally
 899  * wait4()-ing process will get the correct exit code - even if this
 900  * thread is not the thread group leader.
 901  */
 902 asmlinkage void sys_exit_group(int error_code)
 903 {
 904         do_group_exit((error_code & 0xff) << 8);
 905 }
 906
 907 static int eligible_child(pid_t pid, int options, task_t *p)
 908 {
 909         if (pid > 0) {
 910                 if (p->pid != pid)
 911                         return 0;
 912         } else if (!pid) {
 913                 if (process_group(p) != process_group(current))
 914                         return 0;
 915         } else if (pid != -1) {
 916                 if (process_group(p) != -pid)
 917                         return 0;
 918         }
 919
 920         /*
 921          * Do not consider detached threads that are
 922          * not ptraced:
 923          */
 924         if (p->exit_signal == -1 && !p->ptrace)
 925                 return 0;
 926
 927         /* Wait for all children (clone and not) if __WALL is set;
 928          * otherwise, wait for clone children *only* if __WCLONE is
 929          * set; otherwise, wait for non-clone children *only*.  (Note:
 930          * A "clone" child here is one that reports to its parent
 931          * using a signal other than SIGCHLD.) */
 932         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 933             && !(options & __WALL))
 934                 return 0;
 935         /*
 936          * Do not consider thread group leaders that are
 937          * in a non-empty thread group:
 938          */
 939         if (current->tgid != p->tgid && delay_group_leader(p))
 940                 return 2;
 941
 942         if (security_task_wait(p))
 943                 return 0;
 944
 945         return 1;
 946 }
 947
 948 /*
 949  * Handle sys_wait4 work for one task in state TASK_ZOMBIE.  We hold
 950  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
 951  * the lock and this task is uninteresting.  If we return nonzero, we have
 952  * released the lock and the system call should return.
 953  */
 954 static int wait_task_zombie(task_t *p, unsigned int __user *stat_addr, struct rusage __user *ru)
 955 {
 956         unsigned long state;
 957         int retval;
 958
 959         /*
 960          * Try to move the task's state to DEAD
 961          * only one thread is allowed to do this:
 962          */
 963         state = xchg(&p->state, TASK_DEAD);
 964         if (state != TASK_ZOMBIE) {
 965                 BUG_ON(state != TASK_DEAD);
 966                 return 0;
 967         }
 968         if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
 969                 /*
 970                  * This can only happen in a race with a ptraced thread
 971                  * dying on another processor.
 972                  */
 973                 return 0;
 974
 975         /*
 976          * Now we are sure this task is interesting, and no other
 977          * thread can reap it because we set its state to TASK_DEAD.
 978          */
 979         read_unlock(&tasklist_lock);
 980
 981         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
 982         if (!retval && stat_addr) {
 983                 if (p->signal->group_exit)
 984                         retval = put_user(p->signal->group_exit_code, stat_addr);
 985                 else
 986                         retval = put_user(p->exit_code, stat_addr);
 987         }
 988         if (retval) {
 989                 p->state = TASK_ZOMBIE;
 990                 return retval;
 991         }
 992         retval = p->pid;
 993         if (p->real_parent != p->parent) {
 994                 write_lock_irq(&tasklist_lock);
 995                 /* Double-check with lock held.  */
 996                 if (p->real_parent != p->parent) {
 997                         __ptrace_unlink(p);
 998                         p->state = TASK_ZOMBIE;
 999                         /* If this is a detached thread, this is where it goes away.  */
1000                         if (p->exit_signal == -1) {
1001                                 /* release_task takes the lock itself.  */
1002                                 write_unlock_irq(&tasklist_lock);
1003                                 release_task (p);
1004                         }
1005                         else {
1006                                 do_notify_parent(p, p->exit_signal);
1007                                 write_unlock_irq(&tasklist_lock);
1008                         }
1009                         p = NULL;
1010                 }
1011                 else
1012                         write_unlock_irq(&tasklist_lock);
1013         }
1014         if (p != NULL)
1015                 release_task(p);
1016         BUG_ON(!retval);
1017         return retval;
1018 }
1019
1020 /*
1021  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1022  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1023  * the lock and this task is uninteresting.  If we return nonzero, we have
1024  * released the lock and the system call should return.
1025  */
1026 static int wait_task_stopped(task_t *p, int delayed_group_leader,
1027                              unsigned int __user *stat_addr,
1028                              struct rusage __user *ru)
1029 {
1030         int retval, exit_code;
1031
1032         if (!p->exit_code)
1033                 return 0;
1034         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1035             p->signal && p->signal->group_stop_count > 0)
1036                 /*
1037                  * A group stop is in progress and this is the group leader.
1038                  * We won't report until all threads have stopped.
1039                  */
1040                 return 0;
1041
1042         /*
1043          * Now we are pretty sure this task is interesting.
1044          * Make sure it doesn't get reaped out from under us while we
1045          * give up the lock and then examine it below.  We don't want to
1046          * keep holding onto the tasklist_lock while we call getrusage and
1047          * possibly take page faults for user memory.
1048          */
1049         get_task_struct(p);
1050         read_unlock(&tasklist_lock);
1051         write_lock_irq(&tasklist_lock);
1052
1053         /*
1054          * This uses xchg to be atomic with the thread resuming and setting
1055          * it.  It must also be done with the write lock held to prevent a
1056          * race with the TASK_ZOMBIE case.
1057          */
1058         exit_code = xchg(&p->exit_code, 0);
1059         if (unlikely(p->state > TASK_STOPPED)) {
1060                 /*
1061                  * The task resumed and then died.  Let the next iteration
1062                  * catch it in TASK_ZOMBIE.  Note that exit_code might
1063                  * already be zero here if it resumed and did _exit(0).
1064                  * The task itself is dead and won't touch exit_code again;
1065                  * other processors in this function are locked out.
1066                  */
1067                 p->exit_code = exit_code;
1068                 exit_code = 0;
1069         }
1070         if (unlikely(exit_code == 0)) {
1071                 /*
1072                  * Another thread in this function got to it first, or it
1073                  * resumed, or it resumed and then died.
1074                  */
1075                 write_unlock_irq(&tasklist_lock);
1076                 put_task_struct(p);
1077                 read_lock(&tasklist_lock);
1078                 return 0;
1079         }
1080
1081         /* move to end of parent's list to avoid starvation */
1082         remove_parent(p);
1083         add_parent(p, p->parent);
1084
1085         write_unlock_irq(&tasklist_lock);
1086
1087         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1088         if (!retval && stat_addr)
1089                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1090         if (!retval)
1091                 retval = p->pid;
1092         put_task_struct(p);
1093
1094         BUG_ON(!retval);
1095         return retval;
1096 }
1097
1098 asmlinkage long sys_wait4(pid_t pid,unsigned int __user *stat_addr, int options, struct rusage __user *ru)
1099 {
1100         DECLARE_WAITQUEUE(wait, current);
1101         struct task_struct *tsk;
1102         int flag, retval;
1103
1104         if (options & ~(WNOHANG|WUNTRACED|__WNOTHREAD|__WCLONE|__WALL))
1105                 return -EINVAL;
1106
1107         add_wait_queue(&current->wait_chldexit,&wait);
1108 repeat:
1109         flag = 0;
1110         current->state = TASK_INTERRUPTIBLE;
1111         read_lock(&tasklist_lock);
1112         tsk = current;
1113         do {
1114                 struct task_struct *p;
1115                 struct list_head *_p;
1116                 int ret;
1117
1118                 list_for_each(_p,&tsk->children) {
1119                         p = list_entry(_p,struct task_struct,sibling);
1120
1121                         ret = eligible_child(pid, options, p);
1122                         if (!ret)
1123                                 continue;
1124                         flag = 1;
1125
1126                         switch (p->state) {
1127                         case TASK_STOPPED:
1128                                 if (!(options & WUNTRACED) &&
1129                                     !(p->ptrace & PT_PTRACED))
1130                                         continue;
1131                                 retval = wait_task_stopped(p, ret == 2,
1132                                                            stat_addr, ru);
1133                                 if (retval != 0) /* He released the lock.  */
1134                                         goto end_wait4;
1135                                 break;
1136                         case TASK_ZOMBIE:
1137                                 /*
1138                                  * Eligible but we cannot release it yet:
1139                                  */
1140                                 if (ret == 2)
1141                                         continue;
1142                                 retval = wait_task_zombie(p, stat_addr, ru);
1143                                 if (retval != 0) /* He released the lock.  */
1144                                         goto end_wait4;
1145                                 break;
1146                         }
1147                 }
1148                 if (!flag) {
1149                         list_for_each (_p,&tsk->ptrace_children) {
1150                                 p = list_entry(_p,struct task_struct,ptrace_list);
1151                                 if (!eligible_child(pid, options, p))
1152                                         continue;
1153                                 flag = 1;
1154                                 break;
1155                         }
1156                 }
1157                 if (options & __WNOTHREAD)
1158                         break;
1159                 tsk = next_thread(tsk);
1160                 if (tsk->signal != current->signal)
1161                         BUG();
1162         } while (tsk != current);
1163         read_unlock(&tasklist_lock);
1164         if (flag) {
1165                 retval = 0;
1166                 if (options & WNOHANG)
1167                         goto end_wait4;
1168                 retval = -ERESTARTSYS;
1169                 if (signal_pending(current))
1170                         goto end_wait4;
1171                 schedule();
1172                 goto repeat;
1173         }
1174         retval = -ECHILD;
1175 end_wait4:
1176         current->state = TASK_RUNNING;
1177         remove_wait_queue(&current->wait_chldexit,&wait);
1178         return retval;
1179 }
1180
1181 #ifdef __ARCH_WANT_SYS_WAITPID
1182
1183 /*
1184  * sys_waitpid() remains for compatibility. waitpid() should be
1185  * implemented by calling sys_wait4() from libc.a.
1186  */
1187 asmlinkage long sys_waitpid(pid_t pid, unsigned __user *stat_addr, int options)
1188 {
1189         return sys_wait4(pid, stat_addr, options, NULL);
1190 }
1191
1192 #endif