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/key.h>
  18 #include <linux/security.h>
  19 #include <linux/cpu.h>
  20 #include <linux/acct.h>
  21 #include <linux/file.h>
  22 #include <linux/binfmts.h>
  23 #include <linux/ptrace.h>
  24 #include <linux/profile.h>
  25 #include <linux/mount.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/mempolicy.h>
  28 #include <linux/cpuset.h>
  29 #include <linux/syscalls.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/unistd.h>
  33 #include <asm/pgtable.h>
  34 #include <asm/mmu_context.h>
  35
  36 extern void sem_exit (void);
  37 extern struct task_struct *child_reaper;
  38
  39 int getrusage(struct task_struct *, int, struct rusage __user *);
  40
  41 static void __unhash_process(struct task_struct *p)
  42 {
  43         nr_threads--;
  44         detach_pid(p, PIDTYPE_PID);
  45         detach_pid(p, PIDTYPE_TGID);
  46         if (thread_group_leader(p)) {
  47                 detach_pid(p, PIDTYPE_PGID);
  48                 detach_pid(p, PIDTYPE_SID);
  49                 if (p->pid)
  50                         __get_cpu_var(process_counts)--;
  51         }
  52
  53         REMOVE_LINKS(p);
  54 }
  55
  56 void release_task(struct task_struct * p)
  57 {
  58         int zap_leader;
  59         task_t *leader;
  60         struct dentry *proc_dentry;
  61
  62 repeat:
  63         atomic_dec(&p->user->processes);
  64         spin_lock(&p->proc_lock);
  65         proc_dentry = proc_pid_unhash(p);
  66         write_lock_irq(&tasklist_lock);
  67         if (unlikely(p->ptrace))
  68                 __ptrace_unlink(p);
  69         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  70         __exit_signal(p);
  71         __exit_sighand(p);
  72         __unhash_process(p);
  73
  74         /*
  75          * If we are the last non-leader member of the thread
  76          * group, and the leader is zombie, then notify the
  77          * group leader's parent process. (if it wants notification.)
  78          */
  79         zap_leader = 0;
  80         leader = p->group_leader;
  81         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
  82                 BUG_ON(leader->exit_signal == -1);
  83                 do_notify_parent(leader, leader->exit_signal);
  84                 /*
  85                  * If we were the last child thread and the leader has
  86                  * exited already, and the leader's parent ignores SIGCHLD,
  87                  * then we are the one who should release the leader.
  88                  *
  89                  * do_notify_parent() will have marked it self-reaping in
  90                  * that case.
  91                  */
  92                 zap_leader = (leader->exit_signal == -1);
  93         }
  94
  95         sched_exit(p);
  96         write_unlock_irq(&tasklist_lock);
  97         spin_unlock(&p->proc_lock);
  98         proc_pid_flush(proc_dentry);
  99         release_thread(p);
 100         put_task_struct(p);
 101
 102         p = leader;
 103         if (unlikely(zap_leader))
 104                 goto repeat;
 105 }
 106
 107 /* we are using it only for SMP init */
 108
 109 void unhash_process(struct task_struct *p)
 110 {
 111         struct dentry *proc_dentry;
 112
 113         spin_lock(&p->proc_lock);
 114         proc_dentry = proc_pid_unhash(p);
 115         write_lock_irq(&tasklist_lock);
 116         __unhash_process(p);
 117         write_unlock_irq(&tasklist_lock);
 118         spin_unlock(&p->proc_lock);
 119         proc_pid_flush(proc_dentry);
 120 }
 121
 122 /*
 123  * This checks not only the pgrp, but falls back on the pid if no
 124  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 125  * without this...
 126  */
 127 int session_of_pgrp(int pgrp)
 128 {
 129         struct task_struct *p;
 130         int sid = -1;
 131
 132         read_lock(&tasklist_lock);
 133         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 134                 if (p->signal->session > 0) {
 135                         sid = p->signal->session;
 136                         goto out;
 137                 }
 138         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 139         p = find_task_by_pid(pgrp);
 140         if (p)
 141                 sid = p->signal->session;
 142 out:
 143         read_unlock(&tasklist_lock);
 144
 145         return sid;
 146 }
 147
 148 /*
 149  * Determine if a process group is "orphaned", according to the POSIX
 150  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 151  * by terminal-generated stop signals.  Newly orphaned process groups are
 152  * to receive a SIGHUP and a SIGCONT.
 153  *
 154  * "I ask you, have you ever known what it is to be an orphan?"
 155  */
 156 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 157 {
 158         struct task_struct *p;
 159         int ret = 1;
 160
 161         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 162                 if (p == ignored_task
 163                                 || p->exit_state
 164                                 || p->real_parent->pid == 1)
 165                         continue;
 166                 if (process_group(p->real_parent) != pgrp
 167                             && p->real_parent->signal->session == p->signal->session) {
 168                         ret = 0;
 169                         break;
 170                 }
 171         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 172         return ret;     /* (sighing) "Often!" */
 173 }
 174
 175 int is_orphaned_pgrp(int pgrp)
 176 {
 177         int retval;
 178
 179         read_lock(&tasklist_lock);
 180         retval = will_become_orphaned_pgrp(pgrp, NULL);
 181         read_unlock(&tasklist_lock);
 182
 183         return retval;
 184 }
 185
 186 static inline int has_stopped_jobs(int pgrp)
 187 {
 188         int retval = 0;
 189         struct task_struct *p;
 190
 191         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 192                 if (p->state != TASK_STOPPED)
 193                         continue;
 194
 195                 /* If p is stopped by a debugger on a signal that won't
 196                    stop it, then don't count p as stopped.  This isn't
 197                    perfect but it's a good approximation.  */
 198                 if (unlikely (p->ptrace)
 199                     && p->exit_code != SIGSTOP
 200                     && p->exit_code != SIGTSTP
 201                     && p->exit_code != SIGTTOU
 202                     && p->exit_code != SIGTTIN)
 203                         continue;
 204
 205                 retval = 1;
 206                 break;
 207         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 208         return retval;
 209 }
 210
 211 /**
 212  * reparent_to_init() - Reparent the calling kernel thread to the init task.
 213  *
 214  * If a kernel thread is launched as a result of a system call, or if
 215  * it ever exits, it should generally reparent itself to init so that
 216  * it is correctly cleaned up on exit.
 217  *
 218  * The various task state such as scheduling policy and priority may have
 219  * been inherited from a user process, so we reset them to sane values here.
 220  *
 221  * NOTE that reparent_to_init() gives the caller full capabilities.
 222  */
 223 void reparent_to_init(void)
 224 {
 225         write_lock_irq(&tasklist_lock);
 226
 227         ptrace_unlink(current);
 228         /* Reparent to init */
 229         REMOVE_LINKS(current);
 230         current->parent = child_reaper;
 231         current->real_parent = child_reaper;
 232         SET_LINKS(current);
 233
 234         /* Set the exit signal to SIGCHLD so we signal init on exit */
 235         current->exit_signal = SIGCHLD;
 236
 237         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
 238                 set_user_nice(current, 0);
 239         /* cpus_allowed? */
 240         /* rt_priority? */
 241         /* signals? */
 242         security_task_reparent_to_init(current);
 243         memcpy(current->signal->rlim, init_task.signal->rlim,
 244                sizeof(current->signal->rlim));
 245         atomic_inc(&(INIT_USER->__count));
 246         write_unlock_irq(&tasklist_lock);
 247         switch_uid(INIT_USER);
 248 }
 249
 250 void __set_special_pids(pid_t session, pid_t pgrp)
 251 {
 252         struct task_struct *curr = current;
 253
 254         if (curr->signal->session != session) {
 255                 detach_pid(curr, PIDTYPE_SID);
 256                 curr->signal->session = session;
 257                 attach_pid(curr, PIDTYPE_SID, session);
 258         }
 259         if (process_group(curr) != pgrp) {
 260                 detach_pid(curr, PIDTYPE_PGID);
 261                 curr->signal->pgrp = pgrp;
 262                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 263         }
 264 }
 265
 266 void set_special_pids(pid_t session, pid_t pgrp)
 267 {
 268         write_lock_irq(&tasklist_lock);
 269         __set_special_pids(session, pgrp);
 270         write_unlock_irq(&tasklist_lock);
 271 }
 272
 273 /*
 274  * Let kernel threads use this to say that they
 275  * allow a certain signal (since daemonize() will
 276  * have disabled all of them by default).
 277  */
 278 int allow_signal(int sig)
 279 {
 280         if (sig < 1 || sig > _NSIG)
 281                 return -EINVAL;
 282
 283         spin_lock_irq(&current->sighand->siglock);
 284         sigdelset(&current->blocked, sig);
 285         if (!current->mm) {
 286                 /* Kernel threads handle their own signals.
 287                    Let the signal code know it'll be handled, so
 288                    that they don't get converted to SIGKILL or
 289                    just silently dropped */
 290                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 291         }
 292         recalc_sigpending();
 293         spin_unlock_irq(&current->sighand->siglock);
 294         return 0;
 295 }
 296
 297 EXPORT_SYMBOL(allow_signal);
 298
 299 int disallow_signal(int sig)
 300 {
 301         if (sig < 1 || sig > _NSIG)
 302                 return -EINVAL;
 303
 304         spin_lock_irq(&current->sighand->siglock);
 305         sigaddset(&current->blocked, sig);
 306         recalc_sigpending();
 307         spin_unlock_irq(&current->sighand->siglock);
 308         return 0;
 309 }
 310
 311 EXPORT_SYMBOL(disallow_signal);
 312
 313 /*
 314  *      Put all the gunge required to become a kernel thread without
 315  *      attached user resources in one place where it belongs.
 316  */
 317
 318 void daemonize(const char *name, ...)
 319 {
 320         va_list args;
 321         struct fs_struct *fs;
 322         sigset_t blocked;
 323
 324         va_start(args, name);
 325         vsnprintf(current->comm, sizeof(current->comm), name, args);
 326         va_end(args);
 327
 328         /*
 329          * If we were started as result of loading a module, close all of the
 330          * user space pages.  We don't need them, and if we didn't close them
 331          * they would be locked into memory.
 332          */
 333         exit_mm(current);
 334
 335         set_special_pids(1, 1);
 336         down(&tty_sem);
 337         current->signal->tty = NULL;
 338         up(&tty_sem);
 339
 340         /* Block and flush all signals */
 341         sigfillset(&blocked);
 342         sigprocmask(SIG_BLOCK, &blocked, NULL);
 343         flush_signals(current);
 344
 345         /* Become as one with the init task */
 346
 347         exit_fs(current);       /* current->fs->count--; */
 348         fs = init_task.fs;
 349         current->fs = fs;
 350         atomic_inc(&fs->count);
 351         exit_files(current);
 352         current->files = init_task.files;
 353         atomic_inc(&current->files->count);
 354
 355         reparent_to_init();
 356 }
 357
 358 EXPORT_SYMBOL(daemonize);
 359
 360 static inline void close_files(struct files_struct * files)
 361 {
 362         int i, j;
 363
 364         j = 0;
 365         for (;;) {
 366                 unsigned long set;
 367                 i = j * __NFDBITS;
 368                 if (i >= files->max_fdset || i >= files->max_fds)
 369                         break;
 370                 set = files->open_fds->fds_bits[j++];
 371                 while (set) {
 372                         if (set & 1) {
 373                                 struct file * file = xchg(&files->fd[i], NULL);
 374                                 if (file)
 375                                         filp_close(file, files);
 376                         }
 377                         i++;
 378                         set >>= 1;
 379                 }
 380         }
 381 }
 382
 383 struct files_struct *get_files_struct(struct task_struct *task)
 384 {
 385         struct files_struct *files;
 386
 387         task_lock(task);
 388         files = task->files;
 389         if (files)
 390                 atomic_inc(&files->count);
 391         task_unlock(task);
 392
 393         return files;
 394 }
 395
 396 void fastcall put_files_struct(struct files_struct *files)
 397 {
 398         if (atomic_dec_and_test(&files->count)) {
 399                 close_files(files);
 400                 /*
 401                  * Free the fd and fdset arrays if we expanded them.
 402                  */
 403                 if (files->fd != &files->fd_array[0])
 404                         free_fd_array(files->fd, files->max_fds);
 405                 if (files->max_fdset > __FD_SETSIZE) {
 406                         free_fdset(files->open_fds, files->max_fdset);
 407                         free_fdset(files->close_on_exec, files->max_fdset);
 408                 }
 409                 kmem_cache_free(files_cachep, files);
 410         }
 411 }
 412
 413 EXPORT_SYMBOL(put_files_struct);
 414
 415 static inline void __exit_files(struct task_struct *tsk)
 416 {
 417         struct files_struct * files = tsk->files;
 418
 419         if (files) {
 420                 task_lock(tsk);
 421                 tsk->files = NULL;
 422                 task_unlock(tsk);
 423                 put_files_struct(files);
 424         }
 425 }
 426
 427 void exit_files(struct task_struct *tsk)
 428 {
 429         __exit_files(tsk);
 430 }
 431
 432 static inline void __put_fs_struct(struct fs_struct *fs)
 433 {
 434         /* No need to hold fs->lock if we are killing it */
 435         if (atomic_dec_and_test(&fs->count)) {
 436                 dput(fs->root);
 437                 mntput(fs->rootmnt);
 438                 dput(fs->pwd);
 439                 mntput(fs->pwdmnt);
 440                 if (fs->altroot) {
 441                         dput(fs->altroot);
 442                         mntput(fs->altrootmnt);
 443                 }
 444                 kmem_cache_free(fs_cachep, fs);
 445         }
 446 }
 447
 448 void put_fs_struct(struct fs_struct *fs)
 449 {
 450         __put_fs_struct(fs);
 451 }
 452
 453 static inline void __exit_fs(struct task_struct *tsk)
 454 {
 455         struct fs_struct * fs = tsk->fs;
 456
 457         if (fs) {
 458                 task_lock(tsk);
 459                 tsk->fs = NULL;
 460                 task_unlock(tsk);
 461                 __put_fs_struct(fs);
 462         }
 463 }
 464
 465 void exit_fs(struct task_struct *tsk)
 466 {
 467         __exit_fs(tsk);
 468 }
 469
 470 EXPORT_SYMBOL_GPL(exit_fs);
 471
 472 /*
 473  * Turn us into a lazy TLB process if we
 474  * aren't already..
 475  */
 476 void exit_mm(struct task_struct * tsk)
 477 {
 478         struct mm_struct *mm = tsk->mm;
 479
 480         mm_release(tsk, mm);
 481         if (!mm)
 482                 return;
 483         /*
 484          * Serialize with any possible pending coredump.
 485          * We must hold mmap_sem around checking core_waiters
 486          * and clearing tsk->mm.  The core-inducing thread
 487          * will increment core_waiters for each thread in the
 488          * group with ->mm != NULL.
 489          */
 490         down_read(&mm->mmap_sem);
 491         if (mm->core_waiters) {
 492                 up_read(&mm->mmap_sem);
 493                 down_write(&mm->mmap_sem);
 494                 if (!--mm->core_waiters)
 495                         complete(mm->core_startup_done);
 496                 up_write(&mm->mmap_sem);
 497
 498                 wait_for_completion(&mm->core_done);
 499                 down_read(&mm->mmap_sem);
 500         }
 501         atomic_inc(&mm->mm_count);
 502         if (mm != tsk->active_mm) BUG();
 503         /* more a memory barrier than a real lock */
 504         task_lock(tsk);
 505         tsk->mm = NULL;
 506         up_read(&mm->mmap_sem);
 507         enter_lazy_tlb(mm, current);
 508         task_unlock(tsk);
 509         mmput(mm);
 510 }
 511
 512 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 513 {
 514         /*
 515          * Make sure we're not reparenting to ourselves and that
 516          * the parent is not a zombie.
 517          */
 518         BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
 519         p->real_parent = reaper;
 520         if (p->parent == p->real_parent)
 521                 BUG();
 522 }
 523
 524 static inline void reparent_thread(task_t *p, task_t *father, int traced)
 525 {
 526         /* We don't want people slaying init.  */
 527         if (p->exit_signal != -1)
 528                 p->exit_signal = SIGCHLD;
 529
 530         if (p->pdeath_signal)
 531                 /* We already hold the tasklist_lock here.  */
 532                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
 533
 534         /* Move the child from its dying parent to the new one.  */
 535         if (unlikely(traced)) {
 536                 /* Preserve ptrace links if someone else is tracing this child.  */
 537                 list_del_init(&p->ptrace_list);
 538                 if (p->parent != p->real_parent)
 539                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 540         } else {
 541                 /* If this child is being traced, then we're the one tracing it
 542                  * anyway, so let go of it.
 543                  */
 544                 p->ptrace = 0;
 545                 list_del_init(&p->sibling);
 546                 p->parent = p->real_parent;
 547                 list_add_tail(&p->sibling, &p->parent->children);
 548
 549                 /* If we'd notified the old parent about this child's death,
 550                  * also notify the new parent.
 551                  */
 552                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 553                     thread_group_empty(p))
 554                         do_notify_parent(p, p->exit_signal);
 555                 else if (p->state == TASK_TRACED) {
 556                         /*
 557                          * If it was at a trace stop, turn it into
 558                          * a normal stop since it's no longer being
 559                          * traced.
 560                          */
 561                         ptrace_untrace(p);
 562                 }
 563         }
 564
 565         /*
 566          * process group orphan check
 567          * Case ii: Our child is in a different pgrp
 568          * than we are, and it was the only connection
 569          * outside, so the child pgrp is now orphaned.
 570          */
 571         if ((process_group(p) != process_group(father)) &&
 572             (p->signal->session == father->signal->session)) {
 573                 int pgrp = process_group(p);
 574
 575                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 576                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
 577                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
 578                 }
 579         }
 580 }
 581
 582 /*
 583  * When we die, we re-parent all our children.
 584  * Try to give them to another thread in our thread
 585  * group, and if no such member exists, give it to
 586  * the global child reaper process (ie "init")
 587  */
 588 static inline void forget_original_parent(struct task_struct * father,
 589                                           struct list_head *to_release)
 590 {
 591         struct task_struct *p, *reaper = father;
 592         struct list_head *_p, *_n;
 593
 594         do {
 595                 reaper = next_thread(reaper);
 596                 if (reaper == father) {
 597                         reaper = child_reaper;
 598                         break;
 599                 }
 600         } while (reaper->exit_state);
 601
 602         /*
 603          * There are only two places where our children can be:
 604          *
 605          * - in our child list
 606          * - in our ptraced child list
 607          *
 608          * Search them and reparent children.
 609          */
 610         list_for_each_safe(_p, _n, &father->children) {
 611                 int ptrace;
 612                 p = list_entry(_p,struct task_struct,sibling);
 613
 614                 ptrace = p->ptrace;
 615
 616                 /* if father isn't the real parent, then ptrace must be enabled */
 617                 BUG_ON(father != p->real_parent && !ptrace);
 618
 619                 if (father == p->real_parent) {
 620                         /* reparent with a reaper, real father it's us */
 621                         choose_new_parent(p, reaper, child_reaper);
 622                         reparent_thread(p, father, 0);
 623                 } else {
 624                         /* reparent ptraced task to its real parent */
 625                         __ptrace_unlink (p);
 626                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 627                             thread_group_empty(p))
 628                                 do_notify_parent(p, p->exit_signal);
 629                 }
 630
 631                 /*
 632                  * if the ptraced child is a zombie with exit_signal == -1
 633                  * we must collect it before we exit, or it will remain
 634                  * zombie forever since we prevented it from self-reap itself
 635                  * while it was being traced by us, to be able to see it in wait4.
 636                  */
 637                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 638                         list_add(&p->ptrace_list, to_release);
 639         }
 640         list_for_each_safe(_p, _n, &father->ptrace_children) {
 641                 p = list_entry(_p,struct task_struct,ptrace_list);
 642                 choose_new_parent(p, reaper, child_reaper);
 643                 reparent_thread(p, father, 1);
 644         }
 645 }
 646
 647 /*
 648  * Send signals to all our closest relatives so that they know
 649  * to properly mourn us..
 650  */
 651 static void exit_notify(struct task_struct *tsk)
 652 {
 653         int state;
 654         struct task_struct *t;
 655         struct list_head ptrace_dead, *_p, *_n;
 656
 657         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 658             && !thread_group_empty(tsk)) {
 659                 /*
 660                  * This occurs when there was a race between our exit
 661                  * syscall and a group signal choosing us as the one to
 662                  * wake up.  It could be that we are the only thread
 663                  * alerted to check for pending signals, but another thread
 664                  * should be woken now to take the signal since we will not.
 665                  * Now we'll wake all the threads in the group just to make
 666                  * sure someone gets all the pending signals.
 667                  */
 668                 read_lock(&tasklist_lock);
 669                 spin_lock_irq(&tsk->sighand->siglock);
 670                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 671                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 672                                 recalc_sigpending_tsk(t);
 673                                 if (signal_pending(t))
 674                                         signal_wake_up(t, 0);
 675                         }
 676                 spin_unlock_irq(&tsk->sighand->siglock);
 677                 read_unlock(&tasklist_lock);
 678         }
 679
 680         write_lock_irq(&tasklist_lock);
 681
 682         /*
 683          * This does two things:
 684          *
 685          * A.  Make init inherit all the child processes
 686          * B.  Check to see if any process groups have become orphaned
 687          *      as a result of our exiting, and if they have any stopped
 688          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 689          */
 690
 691         INIT_LIST_HEAD(&ptrace_dead);
 692         forget_original_parent(tsk, &ptrace_dead);
 693         BUG_ON(!list_empty(&tsk->children));
 694         BUG_ON(!list_empty(&tsk->ptrace_children));
 695
 696         /*
 697          * Check to see if any process groups have become orphaned
 698          * as a result of our exiting, and if they have any stopped
 699          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 700          *
 701          * Case i: Our father is in a different pgrp than we are
 702          * and we were the only connection outside, so our pgrp
 703          * is about to become orphaned.
 704          */
 705
 706         t = tsk->real_parent;
 707
 708         if ((process_group(t) != process_group(tsk)) &&
 709             (t->signal->session == tsk->signal->session) &&
 710             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 711             has_stopped_jobs(process_group(tsk))) {
 712                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
 713                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
 714         }
 715
 716         /* Let father know we died
 717          *
 718          * Thread signals are configurable, but you aren't going to use
 719          * that to send signals to arbitary processes.
 720          * That stops right now.
 721          *
 722          * If the parent exec id doesn't match the exec id we saved
 723          * when we started then we know the parent has changed security
 724          * domain.
 725          *
 726          * If our self_exec id doesn't match our parent_exec_id then
 727          * we have changed execution domain as these two values started
 728          * the same after a fork.
 729          *
 730          */
 731
 732         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 733             ( tsk->parent_exec_id != t->self_exec_id  ||
 734               tsk->self_exec_id != tsk->parent_exec_id)
 735             && !capable(CAP_KILL))
 736                 tsk->exit_signal = SIGCHLD;
 737
 738
 739         /* If something other than our normal parent is ptracing us, then
 740          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 741          * only has special meaning to our real parent.
 742          */
 743         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 744                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 745                 do_notify_parent(tsk, signal);
 746         } else if (tsk->ptrace) {
 747                 do_notify_parent(tsk, SIGCHLD);
 748         }
 749
 750         state = EXIT_ZOMBIE;
 751         if (tsk->exit_signal == -1 &&
 752             (likely(tsk->ptrace == 0) ||
 753              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 754                 state = EXIT_DEAD;
 755         tsk->exit_state = state;
 756
 757         write_unlock_irq(&tasklist_lock);
 758
 759         list_for_each_safe(_p, _n, &ptrace_dead) {
 760                 list_del_init(_p);
 761                 t = list_entry(_p,struct task_struct,ptrace_list);
 762                 release_task(t);
 763         }
 764
 765         /* If the process is dead, release it - nobody will wait for it */
 766         if (state == EXIT_DEAD)
 767                 release_task(tsk);
 768
 769         /* PF_DEAD causes final put_task_struct after we schedule. */
 770         preempt_disable();
 771         tsk->flags |= PF_DEAD;
 772 }
 773
 774 fastcall NORET_TYPE void do_exit(long code)
 775 {
 776         struct task_struct *tsk = current;
 777         int group_dead;
 778
 779         profile_task_exit(tsk);
 780
 781         if (unlikely(in_interrupt()))
 782                 panic("Aiee, killing interrupt handler!");
 783         if (unlikely(!tsk->pid))
 784                 panic("Attempted to kill the idle task!");
 785         if (unlikely(tsk->pid == 1))
 786                 panic("Attempted to kill init!");
 787         if (tsk->io_context)
 788                 exit_io_context();
 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         tsk->flags |= PF_EXITING;
 796
 797         /*
 798          * Make sure we don't try to process any timer firings
 799          * while we are already exiting.
 800          */
 801         tsk->it_virt_expires = cputime_zero;
 802         tsk->it_prof_expires = cputime_zero;
 803         tsk->it_sched_expires = 0;
 804
 805         if (unlikely(in_atomic()))
 806                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 807                                 current->comm, current->pid,
 808                                 preempt_count());
 809
 810         acct_update_integrals(tsk);
 811         update_mem_hiwater(tsk);
 812         group_dead = atomic_dec_and_test(&tsk->signal->live);
 813         if (group_dead) {
 814                 del_timer_sync(&tsk->signal->real_timer);
 815                 acct_process(code);
 816         }
 817         exit_mm(tsk);
 818
 819         exit_sem(tsk);
 820         __exit_files(tsk);
 821         __exit_fs(tsk);
 822         exit_namespace(tsk);
 823         exit_thread();
 824         cpuset_exit(tsk);
 825         exit_keys(tsk);
 826
 827         if (group_dead && tsk->signal->leader)
 828                 disassociate_ctty(1);
 829
 830         module_put(tsk->thread_info->exec_domain->module);
 831         if (tsk->binfmt)
 832                 module_put(tsk->binfmt->module);
 833
 834         tsk->exit_code = code;
 835         exit_notify(tsk);
 836 #ifdef CONFIG_NUMA
 837         mpol_free(tsk->mempolicy);
 838         tsk->mempolicy = NULL;
 839 #endif
 840
 841         BUG_ON(!(current->flags & PF_DEAD));
 842         schedule();
 843         BUG();
 844         /* Avoid "noreturn function does return".  */
 845         for (;;) ;
 846 }
 847
 848 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 849 {
 850         if (comp)
 851                 complete(comp);
 852
 853         do_exit(code);
 854 }
 855
 856 EXPORT_SYMBOL(complete_and_exit);
 857
 858 asmlinkage long sys_exit(int error_code)
 859 {
 860         do_exit((error_code&0xff)<<8);
 861 }
 862
 863 task_t fastcall *next_thread(const task_t *p)
 864 {
 865         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 866 }
 867
 868 EXPORT_SYMBOL(next_thread);
 869
 870 /*
 871  * Take down every thread in the group.  This is called by fatal signals
 872  * as well as by sys_exit_group (below).
 873  */
 874 NORET_TYPE void
 875 do_group_exit(int exit_code)
 876 {
 877         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 878
 879         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 880                 exit_code = current->signal->group_exit_code;
 881         else if (!thread_group_empty(current)) {
 882                 struct signal_struct *const sig = current->signal;
 883                 struct sighand_struct *const sighand = current->sighand;
 884                 read_lock(&tasklist_lock);
 885                 spin_lock_irq(&sighand->siglock);
 886                 if (sig->flags & SIGNAL_GROUP_EXIT)
 887                         /* Another thread got here before we took the lock.  */
 888                         exit_code = sig->group_exit_code;
 889                 else {
 890                         sig->flags = SIGNAL_GROUP_EXIT;
 891                         sig->group_exit_code = exit_code;
 892                         zap_other_threads(current);
 893                 }
 894                 spin_unlock_irq(&sighand->siglock);
 895                 read_unlock(&tasklist_lock);
 896         }
 897
 898         do_exit(exit_code);
 899         /* NOTREACHED */
 900 }
 901
 902 /*
 903  * this kills every thread in the thread group. Note that any externally
 904  * wait4()-ing process will get the correct exit code - even if this
 905  * thread is not the thread group leader.
 906  */
 907 asmlinkage void sys_exit_group(int error_code)
 908 {
 909         do_group_exit((error_code & 0xff) << 8);
 910 }
 911
 912 static int eligible_child(pid_t pid, int options, task_t *p)
 913 {
 914         if (pid > 0) {
 915                 if (p->pid != pid)
 916                         return 0;
 917         } else if (!pid) {
 918                 if (process_group(p) != process_group(current))
 919                         return 0;
 920         } else if (pid != -1) {
 921                 if (process_group(p) != -pid)
 922                         return 0;
 923         }
 924
 925         /*
 926          * Do not consider detached threads that are
 927          * not ptraced:
 928          */
 929         if (p->exit_signal == -1 && !p->ptrace)
 930                 return 0;
 931
 932         /* Wait for all children (clone and not) if __WALL is set;
 933          * otherwise, wait for clone children *only* if __WCLONE is
 934          * set; otherwise, wait for non-clone children *only*.  (Note:
 935          * A "clone" child here is one that reports to its parent
 936          * using a signal other than SIGCHLD.) */
 937         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 938             && !(options & __WALL))
 939                 return 0;
 940         /*
 941          * Do not consider thread group leaders that are
 942          * in a non-empty thread group:
 943          */
 944         if (current->tgid != p->tgid && delay_group_leader(p))
 945                 return 2;
 946
 947         if (security_task_wait(p))
 948                 return 0;
 949
 950         return 1;
 951 }
 952
 953 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 954                                int why, int status,
 955                                struct siginfo __user *infop,
 956                                struct rusage __user *rusagep)
 957 {
 958         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
 959         put_task_struct(p);
 960         if (!retval)
 961                 retval = put_user(SIGCHLD, &infop->si_signo);
 962         if (!retval)
 963                 retval = put_user(0, &infop->si_errno);
 964         if (!retval)
 965                 retval = put_user((short)why, &infop->si_code);
 966         if (!retval)
 967                 retval = put_user(pid, &infop->si_pid);
 968         if (!retval)
 969                 retval = put_user(uid, &infop->si_uid);
 970         if (!retval)
 971                 retval = put_user(status, &infop->si_status);
 972         if (!retval)
 973                 retval = pid;
 974         return retval;
 975 }
 976
 977 /*
 978  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
 979  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
 980  * the lock and this task is uninteresting.  If we return nonzero, we have
 981  * released the lock and the system call should return.
 982  */
 983 static int wait_task_zombie(task_t *p, int noreap,
 984                             struct siginfo __user *infop,
 985                             int __user *stat_addr, struct rusage __user *ru)
 986 {
 987         unsigned long state;
 988         int retval;
 989         int status;
 990
 991         if (unlikely(noreap)) {
 992                 pid_t pid = p->pid;
 993                 uid_t uid = p->uid;
 994                 int exit_code = p->exit_code;
 995                 int why, status;
 996
 997                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
 998                         return 0;
 999                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1000                         return 0;
1001                 get_task_struct(p);
1002                 read_unlock(&tasklist_lock);
1003                 if ((exit_code & 0x7f) == 0) {
1004                         why = CLD_EXITED;
1005                         status = exit_code >> 8;
1006                 } else {
1007                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1008                         status = exit_code & 0x7f;
1009                 }
1010                 return wait_noreap_copyout(p, pid, uid, why,
1011                                            status, infop, ru);
1012         }
1013
1014         /*
1015          * Try to move the task's state to DEAD
1016          * only one thread is allowed to do this:
1017          */
1018         state = xchg(&p->exit_state, EXIT_DEAD);
1019         if (state != EXIT_ZOMBIE) {
1020                 BUG_ON(state != EXIT_DEAD);
1021                 return 0;
1022         }
1023         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1024                 /*
1025                  * This can only happen in a race with a ptraced thread
1026                  * dying on another processor.
1027                  */
1028                 return 0;
1029         }
1030
1031         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1032                 /*
1033                  * The resource counters for the group leader are in its
1034                  * own task_struct.  Those for dead threads in the group
1035                  * are in its signal_struct, as are those for the child
1036                  * processes it has previously reaped.  All these
1037                  * accumulate in the parent's signal_struct c* fields.
1038                  *
1039                  * We don't bother to take a lock here to protect these
1040                  * p->signal fields, because they are only touched by
1041                  * __exit_signal, which runs with tasklist_lock
1042                  * write-locked anyway, and so is excluded here.  We do
1043                  * need to protect the access to p->parent->signal fields,
1044                  * as other threads in the parent group can be right
1045                  * here reaping other children at the same time.
1046                  */
1047                 spin_lock_irq(&p->parent->sighand->siglock);
1048                 p->parent->signal->cutime =
1049                         cputime_add(p->parent->signal->cutime,
1050                         cputime_add(p->utime,
1051                         cputime_add(p->signal->utime,
1052                                     p->signal->cutime)));
1053                 p->parent->signal->cstime =
1054                         cputime_add(p->parent->signal->cstime,
1055                         cputime_add(p->stime,
1056                         cputime_add(p->signal->stime,
1057                                     p->signal->cstime)));
1058                 p->parent->signal->cmin_flt +=
1059                         p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
1060                 p->parent->signal->cmaj_flt +=
1061                         p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
1062                 p->parent->signal->cnvcsw +=
1063                         p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
1064                 p->parent->signal->cnivcsw +=
1065                         p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
1066                 spin_unlock_irq(&p->parent->sighand->siglock);
1067         }
1068
1069         /*
1070          * Now we are sure this task is interesting, and no other
1071          * thread can reap it because we set its state to EXIT_DEAD.
1072          */
1073         read_unlock(&tasklist_lock);
1074
1075         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1076         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1077                 ? p->signal->group_exit_code : p->exit_code;
1078         if (!retval && stat_addr)
1079                 retval = put_user(status, stat_addr);
1080         if (!retval && infop)
1081                 retval = put_user(SIGCHLD, &infop->si_signo);
1082         if (!retval && infop)
1083                 retval = put_user(0, &infop->si_errno);
1084         if (!retval && infop) {
1085                 int why;
1086
1087                 if ((status & 0x7f) == 0) {
1088                         why = CLD_EXITED;
1089                         status >>= 8;
1090                 } else {
1091                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1092                         status &= 0x7f;
1093                 }
1094                 retval = put_user((short)why, &infop->si_code);
1095                 if (!retval)
1096                         retval = put_user(status, &infop->si_status);
1097         }
1098         if (!retval && infop)
1099                 retval = put_user(p->pid, &infop->si_pid);
1100         if (!retval && infop)
1101                 retval = put_user(p->uid, &infop->si_uid);
1102         if (retval) {
1103                 // TODO: is this safe?
1104                 p->exit_state = EXIT_ZOMBIE;
1105                 return retval;
1106         }
1107         retval = p->pid;
1108         if (p->real_parent != p->parent) {
1109                 write_lock_irq(&tasklist_lock);
1110                 /* Double-check with lock held.  */
1111                 if (p->real_parent != p->parent) {
1112                         __ptrace_unlink(p);
1113                         // TODO: is this safe?
1114                         p->exit_state = EXIT_ZOMBIE;
1115                         /*
1116                          * If this is not a detached task, notify the parent.
1117                          * If it's still not detached after that, don't release
1118                          * it now.
1119                          */
1120                         if (p->exit_signal != -1) {
1121                                 do_notify_parent(p, p->exit_signal);
1122                                 if (p->exit_signal != -1)
1123                                         p = NULL;
1124                         }
1125                 }
1126                 write_unlock_irq(&tasklist_lock);
1127         }
1128         if (p != NULL)
1129                 release_task(p);
1130         BUG_ON(!retval);
1131         return retval;
1132 }
1133
1134 /*
1135  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1136  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1137  * the lock and this task is uninteresting.  If we return nonzero, we have
1138  * released the lock and the system call should return.
1139  */
1140 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1141                              struct siginfo __user *infop,
1142                              int __user *stat_addr, struct rusage __user *ru)
1143 {
1144         int retval, exit_code;
1145
1146         if (!p->exit_code)
1147                 return 0;
1148         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1149             p->signal && p->signal->group_stop_count > 0)
1150                 /*
1151                  * A group stop is in progress and this is the group leader.
1152                  * We won't report until all threads have stopped.
1153                  */
1154                 return 0;
1155
1156         /*
1157          * Now we are pretty sure this task is interesting.
1158          * Make sure it doesn't get reaped out from under us while we
1159          * give up the lock and then examine it below.  We don't want to
1160          * keep holding onto the tasklist_lock while we call getrusage and
1161          * possibly take page faults for user memory.
1162          */
1163         get_task_struct(p);
1164         read_unlock(&tasklist_lock);
1165
1166         if (unlikely(noreap)) {
1167                 pid_t pid = p->pid;
1168                 uid_t uid = p->uid;
1169                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1170
1171                 exit_code = p->exit_code;
1172                 if (unlikely(!exit_code) ||
1173                     unlikely(p->state > TASK_STOPPED))
1174                         goto bail_ref;
1175                 return wait_noreap_copyout(p, pid, uid,
1176                                            why, (exit_code << 8) | 0x7f,
1177                                            infop, ru);
1178         }
1179
1180         write_lock_irq(&tasklist_lock);
1181
1182         /*
1183          * This uses xchg to be atomic with the thread resuming and setting
1184          * it.  It must also be done with the write lock held to prevent a
1185          * race with the EXIT_ZOMBIE case.
1186          */
1187         exit_code = xchg(&p->exit_code, 0);
1188         if (unlikely(p->exit_state)) {
1189                 /*
1190                  * The task resumed and then died.  Let the next iteration
1191                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1192                  * already be zero here if it resumed and did _exit(0).
1193                  * The task itself is dead and won't touch exit_code again;
1194                  * other processors in this function are locked out.
1195                  */
1196                 p->exit_code = exit_code;
1197                 exit_code = 0;
1198         }
1199         if (unlikely(exit_code == 0)) {
1200                 /*
1201                  * Another thread in this function got to it first, or it
1202                  * resumed, or it resumed and then died.
1203                  */
1204                 write_unlock_irq(&tasklist_lock);
1205 bail_ref:
1206                 put_task_struct(p);
1207                 /*
1208                  * We are returning to the wait loop without having successfully
1209                  * removed the process and having released the lock. We cannot
1210                  * continue, since the "p" task pointer is potentially stale.
1211                  *
1212                  * Return -EAGAIN, and do_wait() will restart the loop from the
1213                  * beginning. Do _not_ re-acquire the lock.
1214                  */
1215                 return -EAGAIN;
1216         }
1217
1218         /* move to end of parent's list to avoid starvation */
1219         remove_parent(p);
1220         add_parent(p, p->parent);
1221
1222         write_unlock_irq(&tasklist_lock);
1223
1224         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1225         if (!retval && stat_addr)
1226                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1227         if (!retval && infop)
1228                 retval = put_user(SIGCHLD, &infop->si_signo);
1229         if (!retval && infop)
1230                 retval = put_user(0, &infop->si_errno);
1231         if (!retval && infop)
1232                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1233                                           ? CLD_TRAPPED : CLD_STOPPED),
1234                                   &infop->si_code);
1235         if (!retval && infop)
1236                 retval = put_user(exit_code, &infop->si_status);
1237         if (!retval && infop)
1238                 retval = put_user(p->pid, &infop->si_pid);
1239         if (!retval && infop)
1240                 retval = put_user(p->uid, &infop->si_uid);
1241         if (!retval)
1242                 retval = p->pid;
1243         put_task_struct(p);
1244
1245         BUG_ON(!retval);
1246         return retval;
1247 }
1248
1249 /*
1250  * Handle do_wait work for one task in a live, non-stopped state.
1251  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1252  * the lock and this task is uninteresting.  If we return nonzero, we have
1253  * released the lock and the system call should return.
1254  */
1255 static int wait_task_continued(task_t *p, int noreap,
1256                                struct siginfo __user *infop,
1257                                int __user *stat_addr, struct rusage __user *ru)
1258 {
1259         int retval;
1260         pid_t pid;
1261         uid_t uid;
1262
1263         if (unlikely(!p->signal))
1264                 return 0;
1265
1266         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1267                 return 0;
1268
1269         spin_lock_irq(&p->sighand->siglock);
1270         /* Re-check with the lock held.  */
1271         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1272                 spin_unlock_irq(&p->sighand->siglock);
1273                 return 0;
1274         }
1275         if (!noreap)
1276                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1277         spin_unlock_irq(&p->sighand->siglock);
1278
1279         pid = p->pid;
1280         uid = p->uid;
1281         get_task_struct(p);
1282         read_unlock(&tasklist_lock);
1283
1284         if (!infop) {
1285                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1286                 put_task_struct(p);
1287                 if (!retval && stat_addr)
1288                         retval = put_user(0xffff, stat_addr);
1289                 if (!retval)
1290                         retval = p->pid;
1291         } else {
1292                 retval = wait_noreap_copyout(p, pid, uid,
1293                                              CLD_CONTINUED, SIGCONT,
1294                                              infop, ru);
1295                 BUG_ON(retval == 0);
1296         }
1297
1298         return retval;
1299 }
1300
1301
1302 static inline int my_ptrace_child(struct task_struct *p)
1303 {
1304         if (!(p->ptrace & PT_PTRACED))
1305                 return 0;
1306         if (!(p->ptrace & PT_ATTACHED))
1307                 return 1;
1308         /*
1309          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1310          * we are the attacher.  If we are the real parent, this is a race
1311          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1312          * which we have to switch the parent links, but has already set
1313          * the flags in p->ptrace.
1314          */
1315         return (p->parent != p->real_parent);
1316 }
1317
1318 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1319                     int __user *stat_addr, struct rusage __user *ru)
1320 {
1321         DECLARE_WAITQUEUE(wait, current);
1322         struct task_struct *tsk;
1323         int flag, retval;
1324
1325         add_wait_queue(&current->signal->wait_chldexit,&wait);
1326 repeat:
1327         /*
1328          * We will set this flag if we see any child that might later
1329          * match our criteria, even if we are not able to reap it yet.
1330          */
1331         flag = 0;
1332         current->state = TASK_INTERRUPTIBLE;
1333         read_lock(&tasklist_lock);
1334         tsk = current;
1335         do {
1336                 struct task_struct *p;
1337                 struct list_head *_p;
1338                 int ret;
1339
1340                 list_for_each(_p,&tsk->children) {
1341                         p = list_entry(_p,struct task_struct,sibling);
1342
1343                         ret = eligible_child(pid, options, p);
1344                         if (!ret)
1345                                 continue;
1346
1347                         switch (p->state) {
1348                         case TASK_TRACED:
1349                                 if (!my_ptrace_child(p))
1350                                         continue;
1351                                 /*FALLTHROUGH*/
1352                         case TASK_STOPPED:
1353                                 /*
1354                                  * It's stopped now, so it might later
1355                                  * continue, exit, or stop again.
1356                                  */
1357                                 flag = 1;
1358                                 if (!(options & WUNTRACED) &&
1359                                     !my_ptrace_child(p))
1360                                         continue;
1361                                 retval = wait_task_stopped(p, ret == 2,
1362                                                            (options & WNOWAIT),
1363                                                            infop,
1364                                                            stat_addr, ru);
1365                                 if (retval == -EAGAIN)
1366                                         goto repeat;
1367                                 if (retval != 0) /* He released the lock.  */
1368                                         goto end;
1369                                 break;
1370                         default:
1371                         // case EXIT_DEAD:
1372                                 if (p->exit_state == EXIT_DEAD)
1373                                         continue;
1374                         // case EXIT_ZOMBIE:
1375                                 if (p->exit_state == EXIT_ZOMBIE) {
1376                                         /*
1377                                          * Eligible but we cannot release
1378                                          * it yet:
1379                                          */
1380                                         if (ret == 2)
1381                                                 goto check_continued;
1382                                         if (!likely(options & WEXITED))
1383                                                 continue;
1384                                         retval = wait_task_zombie(
1385                                                 p, (options & WNOWAIT),
1386                                                 infop, stat_addr, ru);
1387                                         /* He released the lock.  */
1388                                         if (retval != 0)
1389                                                 goto end;
1390                                         break;
1391                                 }
1392 check_continued:
1393                                 /*
1394                                  * It's running now, so it might later
1395                                  * exit, stop, or stop and then continue.
1396                                  */
1397                                 flag = 1;
1398                                 if (!unlikely(options & WCONTINUED))
1399                                         continue;
1400                                 retval = wait_task_continued(
1401                                         p, (options & WNOWAIT),
1402                                         infop, stat_addr, ru);
1403                                 if (retval != 0) /* He released the lock.  */
1404                                         goto end;
1405                                 break;
1406                         }
1407                 }
1408                 if (!flag) {
1409                         list_for_each(_p, &tsk->ptrace_children) {
1410                                 p = list_entry(_p, struct task_struct,
1411                                                 ptrace_list);
1412                                 if (!eligible_child(pid, options, p))
1413                                         continue;
1414                                 flag = 1;
1415                                 break;
1416                         }
1417                 }
1418                 if (options & __WNOTHREAD)
1419                         break;
1420                 tsk = next_thread(tsk);
1421                 if (tsk->signal != current->signal)
1422                         BUG();
1423         } while (tsk != current);
1424
1425         read_unlock(&tasklist_lock);
1426         if (flag) {
1427                 retval = 0;
1428                 if (options & WNOHANG)
1429                         goto end;
1430                 retval = -ERESTARTSYS;
1431                 if (signal_pending(current))
1432                         goto end;
1433                 schedule();
1434                 goto repeat;
1435         }
1436         retval = -ECHILD;
1437 end:
1438         current->state = TASK_RUNNING;
1439         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1440         if (infop) {
1441                 if (retval > 0)
1442                 retval = 0;
1443                 else {
1444                         /*
1445                          * For a WNOHANG return, clear out all the fields
1446                          * we would set so the user can easily tell the
1447                          * difference.
1448                          */
1449                         if (!retval)
1450                                 retval = put_user(0, &infop->si_signo);
1451                         if (!retval)
1452                                 retval = put_user(0, &infop->si_errno);
1453                         if (!retval)
1454                                 retval = put_user(0, &infop->si_code);
1455                         if (!retval)
1456                                 retval = put_user(0, &infop->si_pid);
1457                         if (!retval)
1458                                 retval = put_user(0, &infop->si_uid);
1459                         if (!retval)
1460                                 retval = put_user(0, &infop->si_status);
1461                 }
1462         }
1463         return retval;
1464 }
1465
1466 asmlinkage long sys_waitid(int which, pid_t pid,
1467                            struct siginfo __user *infop, int options,
1468                            struct rusage __user *ru)
1469 {
1470         long ret;
1471
1472         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1473                 return -EINVAL;
1474         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1475                 return -EINVAL;
1476
1477         switch (which) {
1478         case P_ALL:
1479                 pid = -1;
1480                 break;
1481         case P_PID:
1482                 if (pid <= 0)
1483                         return -EINVAL;
1484                 break;
1485         case P_PGID:
1486                 if (pid <= 0)
1487                         return -EINVAL;
1488                 pid = -pid;
1489                 break;
1490         default:
1491                 return -EINVAL;
1492         }
1493
1494         ret = do_wait(pid, options, infop, NULL, ru);
1495
1496         /* avoid REGPARM breakage on x86: */
1497         prevent_tail_call(ret);
1498         return ret;
1499 }
1500
1501 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1502                           int options, struct rusage __user *ru)
1503 {
1504         long ret;
1505
1506         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1507                         __WNOTHREAD|__WCLONE|__WALL))
1508                 return -EINVAL;
1509         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1510
1511         /* avoid REGPARM breakage on x86: */
1512         prevent_tail_call(ret);
1513         return ret;
1514 }
1515
1516 #ifdef __ARCH_WANT_SYS_WAITPID
1517
1518 /*
1519  * sys_waitpid() remains for compatibility. waitpid() should be
1520  * implemented by calling sys_wait4() from libc.a.
1521  */
1522 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1523 {
1524         return sys_wait4(pid, stat_addr, options, NULL);
1525 }
1526
1527 #endif