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