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/ckrm_events.h>
  29 #include <linux/syscalls.h>
  30 #include <linux/ckrm_mem_inline.h>
  31
  32 #include <asm/uaccess.h>
  33 #include <asm/unistd.h>
  34 #include <asm/pgtable.h>
  35 #include <asm/mmu_context.h>
  36
  37 extern void sem_exit (void);
  38 extern struct task_struct *child_reaper;
  39
  40 int getrusage(struct task_struct *, int, struct rusage __user *);
  41
  42 static void __unhash_process(struct task_struct *p)
  43 {
  44         nr_threads--;
  45         detach_pid(p, PIDTYPE_PID);
  46         detach_pid(p, PIDTYPE_TGID);
  47         if (thread_group_leader(p)) {
  48                 detach_pid(p, PIDTYPE_PGID);
  49                 detach_pid(p, PIDTYPE_SID);
  50                 if (p->pid)
  51                         __get_cpu_var(process_counts)--;
  52         }
  53
  54         REMOVE_LINKS(p);
  55 }
  56
  57 void release_task(struct task_struct * p)
  58 {
  59         int zap_leader;
  60         task_t *leader;
  61         struct dentry *proc_dentry;
  62
  63 repeat:
  64         atomic_dec(&p->user->processes);
  65         spin_lock(&p->proc_lock);
  66         proc_dentry = proc_pid_unhash(p);
  67         write_lock_irq(&tasklist_lock);
  68         if (unlikely(p->ptrace))
  69                 __ptrace_unlink(p);
  70         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  71         __exit_signal(p);
  72         __exit_sighand(p);
  73         __unhash_process(p);
  74
  75         /*
  76          * If we are the last non-leader member of the thread
  77          * group, and the leader is zombie, then notify the
  78          * group leader's parent process. (if it wants notification.)
  79          */
  80         zap_leader = 0;
  81         leader = p->group_leader;
  82         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
  83                 BUG_ON(leader->exit_signal == -1);
  84                 do_notify_parent(leader, leader->exit_signal);
  85                 /*
  86                  * If we were the last child thread and the leader has
  87                  * exited already, and the leader's parent ignores SIGCHLD,
  88                  * then we are the one who should release the leader.
  89                  *
  90                  * do_notify_parent() will have marked it self-reaping in
  91                  * that case.
  92                  */
  93                 zap_leader = (leader->exit_signal == -1);
  94         }
  95
  96         sched_exit(p);
  97         write_unlock_irq(&tasklist_lock);
  98         spin_unlock(&p->proc_lock);
  99         proc_pid_flush(proc_dentry);
 100         release_thread(p);
 101         put_task_struct(p);
 102
 103         p = leader;
 104         if (unlikely(zap_leader))
 105                 goto repeat;
 106 }
 107
 108 /* we are using it only for SMP init */
 109
 110 void unhash_process(struct task_struct *p)
 111 {
 112         struct dentry *proc_dentry;
 113
 114         spin_lock(&p->proc_lock);
 115         proc_dentry = proc_pid_unhash(p);
 116         write_lock_irq(&tasklist_lock);
 117         __unhash_process(p);
 118         write_unlock_irq(&tasklist_lock);
 119         spin_unlock(&p->proc_lock);
 120         proc_pid_flush(proc_dentry);
 121 }
 122
 123 /*
 124  * This checks not only the pgrp, but falls back on the pid if no
 125  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 126  * without this...
 127  */
 128 int session_of_pgrp(int pgrp)
 129 {
 130         struct task_struct *p;
 131         int sid = -1;
 132
 133         read_lock(&tasklist_lock);
 134         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 135                 if (p->signal->session > 0) {
 136                         sid = p->signal->session;
 137                         goto out;
 138                 }
 139         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 140         p = find_task_by_pid(pgrp);
 141         if (p)
 142                 sid = p->signal->session;
 143 out:
 144         read_unlock(&tasklist_lock);
 145
 146         return sid;
 147 }
 148
 149 /*
 150  * Determine if a process group is "orphaned", according to the POSIX
 151  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 152  * by terminal-generated stop signals.  Newly orphaned process groups are
 153  * to receive a SIGHUP and a SIGCONT.
 154  *
 155  * "I ask you, have you ever known what it is to be an orphan?"
 156  */
 157 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 158 {
 159         struct task_struct *p;
 160         int ret = 1;
 161
 162         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 163                 if (p == ignored_task
 164                                 || p->exit_state >= EXIT_ZOMBIE
 165                                 || p->real_parent->pid == 1)
 166                         continue;
 167                 if (process_group(p->real_parent) != pgrp
 168                             && p->real_parent->signal->session == p->signal->session) {
 169                         ret = 0;
 170                         break;
 171                 }
 172         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 173         return ret;     /* (sighing) "Often!" */
 174 }
 175
 176 int is_orphaned_pgrp(int pgrp)
 177 {
 178         int retval;
 179
 180         read_lock(&tasklist_lock);
 181         retval = will_become_orphaned_pgrp(pgrp, NULL);
 182         read_unlock(&tasklist_lock);
 183
 184         return retval;
 185 }
 186
 187 static inline int has_stopped_jobs(int pgrp)
 188 {
 189         int retval = 0;
 190         struct task_struct *p;
 191
 192         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 193                 if (p->state != TASK_STOPPED)
 194                         continue;
 195
 196                 /* If p is stopped by a debugger on a signal that won't
 197                    stop it, then don't count p as stopped.  This isn't
 198                    perfect but it's a good approximation.  */
 199                 if (unlikely (p->ptrace)
 200                     && p->exit_code != SIGSTOP
 201                     && p->exit_code != SIGTSTP
 202                     && p->exit_code != SIGTTOU
 203                     && p->exit_code != SIGTTIN)
 204                         continue;
 205
 206                 retval = 1;
 207                 break;
 208         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 209         return retval;
 210 }
 211
 212 /**
 213  * reparent_to_init() - Reparent the calling kernel thread to the init task.
 214  *
 215  * If a kernel thread is launched as a result of a system call, or if
 216  * it ever exits, it should generally reparent itself to init so that
 217  * it is correctly cleaned up on exit.
 218  *
 219  * The various task state such as scheduling policy and priority may have
 220  * been inherited from a user process, so we reset them to sane values here.
 221  *
 222  * NOTE that reparent_to_init() gives the caller full capabilities.
 223  */
 224 void reparent_to_init(void)
 225 {
 226         write_lock_irq(&tasklist_lock);
 227
 228         ptrace_unlink(current);
 229         /* Reparent to init */
 230         REMOVE_LINKS(current);
 231         current->parent = child_reaper;
 232         current->real_parent = child_reaper;
 233         SET_LINKS(current);
 234
 235         /* Set the exit signal to SIGCHLD so we signal init on exit */
 236         current->exit_signal = SIGCHLD;
 237
 238         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
 239                 set_user_nice(current, 0);
 240         /* cpus_allowed? */
 241         /* rt_priority? */
 242         /* signals? */
 243         security_task_reparent_to_init(current);
 244         memcpy(current->signal->rlim, init_task.signal->rlim,
 245                sizeof(current->signal->rlim));
 246         atomic_inc(&(INIT_USER->__count));
 247         write_unlock_irq(&tasklist_lock);
 248         switch_uid(INIT_USER);
 249 }
 250
 251 void __set_special_pids(pid_t session, pid_t pgrp)
 252 {
 253         struct task_struct *curr = current;
 254
 255         if (curr->signal->session != session) {
 256                 detach_pid(curr, PIDTYPE_SID);
 257                 curr->signal->session = session;
 258                 attach_pid(curr, PIDTYPE_SID, session);
 259         }
 260         if (process_group(curr) != pgrp) {
 261                 detach_pid(curr, PIDTYPE_PGID);
 262                 curr->signal->pgrp = pgrp;
 263                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 264         }
 265 }
 266
 267 void set_special_pids(pid_t session, pid_t pgrp)
 268 {
 269         write_lock_irq(&tasklist_lock);
 270         __set_special_pids(session, pgrp);
 271         write_unlock_irq(&tasklist_lock);
 272 }
 273
 274 /*
 275  * Let kernel threads use this to say that they
 276  * allow a certain signal (since daemonize() will
 277  * have disabled all of them by default).
 278  */
 279 int allow_signal(int sig)
 280 {
 281         if (sig < 1 || sig > _NSIG)
 282                 return -EINVAL;
 283
 284         spin_lock_irq(&current->sighand->siglock);
 285         sigdelset(&current->blocked, sig);
 286         if (!current->mm) {
 287                 /* Kernel threads handle their own signals.
 288                    Let the signal code know it'll be handled, so
 289                    that they don't get converted to SIGKILL or
 290                    just silently dropped */
 291                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 292         }
 293         recalc_sigpending();
 294         spin_unlock_irq(&current->sighand->siglock);
 295         return 0;
 296 }
 297
 298 EXPORT_SYMBOL(allow_signal);
 299
 300 int disallow_signal(int sig)
 301 {
 302         if (sig < 1 || sig > _NSIG)
 303                 return -EINVAL;
 304
 305         spin_lock_irq(&current->sighand->siglock);
 306         sigaddset(&current->blocked, sig);
 307         recalc_sigpending();
 308         spin_unlock_irq(&current->sighand->siglock);
 309         return 0;
 310 }
 311
 312 EXPORT_SYMBOL(disallow_signal);
 313
 314 /*
 315  *      Put all the gunge required to become a kernel thread without
 316  *      attached user resources in one place where it belongs.
 317  */
 318
 319 void daemonize(const char *name, ...)
 320 {
 321         va_list args;
 322         struct fs_struct *fs;
 323         sigset_t blocked;
 324
 325         va_start(args, name);
 326         vsnprintf(current->comm, sizeof(current->comm), name, args);
 327         va_end(args);
 328
 329         /*
 330          * If we were started as result of loading a module, close all of the
 331          * user space pages.  We don't need them, and if we didn't close them
 332          * they would be locked into memory.
 333          */
 334         exit_mm(current);
 335
 336         set_special_pids(1, 1);
 337         current->signal->tty = NULL;
 338
 339         /* Block and flush all signals */
 340         sigfillset(&blocked);
 341         sigprocmask(SIG_BLOCK, &blocked, NULL);
 342         flush_signals(current);
 343
 344         /* Become as one with the init task */
 345
 346         exit_fs(current);       /* current->fs->count--; */
 347         fs = init_task.fs;
 348         current->fs = fs;
 349         atomic_inc(&fs->count);
 350         exit_files(current);
 351         current->files = init_task.files;
 352         atomic_inc(&current->files->count);
 353
 354         reparent_to_init();
 355 }
 356
 357 EXPORT_SYMBOL(daemonize);
 358
 359 static inline void close_files(struct files_struct * files)
 360 {
 361         int i, j;
 362
 363         j = 0;
 364         for (;;) {
 365                 unsigned long set;
 366                 i = j * __NFDBITS;
 367                 if (i >= files->max_fdset || i >= files->max_fds)
 368                         break;
 369                 set = files->open_fds->fds_bits[j++];
 370                 while (set) {
 371                         if (set & 1) {
 372                                 struct file * file = xchg(&files->fd[i], NULL);
 373                                 if (file)
 374                                         filp_close(file, files);
 375                         }
 376                         i++;
 377                         set >>= 1;
 378                 }
 379         }
 380 }
 381
 382 struct files_struct *get_files_struct(struct task_struct *task)
 383 {
 384         struct files_struct *files;
 385
 386         task_lock(task);
 387         files = task->files;
 388         if (files)
 389                 atomic_inc(&files->count);
 390         task_unlock(task);
 391
 392         return files;
 393 }
 394
 395 void fastcall put_files_struct(struct files_struct *files)
 396 {
 397         if (atomic_dec_and_test(&files->count)) {
 398                 close_files(files);
 399                 /*
 400                  * Free the fd and fdset arrays if we expanded them.
 401                  */
 402                 if (files->fd != &files->fd_array[0])
 403                         free_fd_array(files->fd, files->max_fds);
 404                 if (files->max_fdset > __FD_SETSIZE) {
 405                         free_fdset(files->open_fds, files->max_fdset);
 406                         free_fdset(files->close_on_exec, files->max_fdset);
 407                 }
 408                 kmem_cache_free(files_cachep, files);
 409         }
 410 }
 411
 412 EXPORT_SYMBOL(put_files_struct);
 413
 414 static inline void __exit_files(struct task_struct *tsk)
 415 {
 416         struct files_struct * files = tsk->files;
 417
 418         if (files) {
 419                 task_lock(tsk);
 420                 tsk->files = NULL;
 421                 task_unlock(tsk);
 422                 put_files_struct(files);
 423         }
 424 }
 425
 426 void exit_files(struct task_struct *tsk)
 427 {
 428         __exit_files(tsk);
 429 }
 430
 431 static inline void __put_fs_struct(struct fs_struct *fs)
 432 {
 433         /* No need to hold fs->lock if we are killing it */
 434         if (atomic_dec_and_test(&fs->count)) {
 435                 dput(fs->root);
 436                 mntput(fs->rootmnt);
 437                 dput(fs->pwd);
 438                 mntput(fs->pwdmnt);
 439                 if (fs->altroot) {
 440                         dput(fs->altroot);
 441                         mntput(fs->altrootmnt);
 442                 }
 443                 kmem_cache_free(fs_cachep, fs);
 444         }
 445 }
 446
 447 void put_fs_struct(struct fs_struct *fs)
 448 {
 449         __put_fs_struct(fs);
 450 }
 451
 452 static inline void __exit_fs(struct task_struct *tsk)
 453 {
 454         struct fs_struct * fs = tsk->fs;
 455
 456         if (fs) {
 457                 task_lock(tsk);
 458                 tsk->fs = NULL;
 459                 task_unlock(tsk);
 460                 __put_fs_struct(fs);
 461         }
 462 }
 463
 464 void exit_fs(struct task_struct *tsk)
 465 {
 466         __exit_fs(tsk);
 467 }
 468
 469 EXPORT_SYMBOL_GPL(exit_fs);
 470
 471 /*
 472  * Turn us into a lazy TLB process if we
 473  * aren't already..
 474  */
 475 static inline void __exit_mm(struct task_struct * tsk)
 476 {
 477         struct mm_struct *mm = tsk->mm;
 478
 479         mm_release(tsk, mm);
 480         if (!mm)
 481                 return;
 482         /*
 483          * Serialize with any possible pending coredump.
 484          * We must hold mmap_sem around checking core_waiters
 485          * and clearing tsk->mm.  The core-inducing thread
 486          * will increment core_waiters for each thread in the
 487          * group with ->mm != NULL.
 488          */
 489         down_read(&mm->mmap_sem);
 490         if (mm->core_waiters) {
 491                 up_read(&mm->mmap_sem);
 492                 down_write(&mm->mmap_sem);
 493                 if (!--mm->core_waiters)
 494                         complete(mm->core_startup_done);
 495                 up_write(&mm->mmap_sem);
 496
 497                 wait_for_completion(&mm->core_done);
 498                 down_read(&mm->mmap_sem);
 499         }
 500         atomic_inc(&mm->mm_count);
 501         if (mm != tsk->active_mm) BUG();
 502         /* more a memory barrier than a real lock */
 503         task_lock(tsk);
 504         tsk->mm = NULL;
 505         up_read(&mm->mmap_sem);
 506         ckrm_task_mm_clear(tsk, mm);
 507         enter_lazy_tlb(mm, current);
 508         task_unlock(tsk);
 509         mmput(mm);
 510 }
 511
 512 void exit_mm(struct task_struct *tsk)
 513 {
 514         __exit_mm(tsk);
 515 }
 516
 517 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 518 {
 519         /*
 520          * Make sure we're not reparenting to ourselves and that
 521          * the parent is not a zombie.
 522          */
 523         BUG_ON(p == reaper || reaper->state >= EXIT_ZOMBIE || reaper->exit_state >= EXIT_ZOMBIE);
 524         p->real_parent = reaper;
 525         if (p->parent == p->real_parent)
 526                 BUG();
 527 }
 528
 529 static inline void reparent_thread(task_t *p, task_t *father, int traced)
 530 {
 531         /* We don't want people slaying init.  */
 532         if (p->exit_signal != -1)
 533                 p->exit_signal = SIGCHLD;
 534
 535         if (p->pdeath_signal)
 536                 /* We already hold the tasklist_lock here.  */
 537                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
 538
 539         /* Move the child from its dying parent to the new one.  */
 540         if (unlikely(traced)) {
 541                 /* Preserve ptrace links if someone else is tracing this child.  */
 542                 list_del_init(&p->ptrace_list);
 543                 if (p->parent != p->real_parent)
 544                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 545         } else {
 546                 /* If this child is being traced, then we're the one tracing it
 547                  * anyway, so let go of it.
 548                  */
 549                 p->ptrace = 0;
 550                 list_del_init(&p->sibling);
 551                 p->parent = p->real_parent;
 552                 list_add_tail(&p->sibling, &p->parent->children);
 553
 554                 /* If we'd notified the old parent about this child's death,
 555                  * also notify the new parent.
 556                  */
 557                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 558                     thread_group_empty(p))
 559                         do_notify_parent(p, p->exit_signal);
 560                 else if (p->state == TASK_TRACED) {
 561                         /*
 562                          * If it was at a trace stop, turn it into
 563                          * a normal stop since it's no longer being
 564                          * traced.
 565                          */
 566                         p->state = TASK_STOPPED;
 567                 }
 568         }
 569
 570         /*
 571          * process group orphan check
 572          * Case ii: Our child is in a different pgrp
 573          * than we are, and it was the only connection
 574          * outside, so the child pgrp is now orphaned.
 575          */
 576         if ((process_group(p) != process_group(father)) &&
 577             (p->signal->session == father->signal->session)) {
 578                 int pgrp = process_group(p);
 579
 580                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 581                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
 582                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
 583                 }
 584         }
 585 }
 586
 587 /*
 588  * When we die, we re-parent all our children.
 589  * Try to give them to another thread in our thread
 590  * group, and if no such member exists, give it to
 591  * the global child reaper process (ie "init")
 592  */
 593 static inline void forget_original_parent(struct task_struct * father,
 594                                           struct list_head *to_release)
 595 {
 596         struct task_struct *p, *reaper = father;
 597         struct list_head *_p, *_n;
 598
 599         do {
 600                 reaper = next_thread(reaper);
 601                 if (reaper == father) {
 602                         reaper = child_reaper;
 603                         break;
 604                 }
 605         } while (reaper->exit_state >= EXIT_ZOMBIE);
 606
 607         /*
 608          * There are only two places where our children can be:
 609          *
 610          * - in our child list
 611          * - in our ptraced child list
 612          *
 613          * Search them and reparent children.
 614          */
 615         list_for_each_safe(_p, _n, &father->children) {
 616                 int ptrace;
 617                 p = list_entry(_p,struct task_struct,sibling);
 618
 619                 ptrace = p->ptrace;
 620
 621                 /* if father isn't the real parent, then ptrace must be enabled */
 622                 BUG_ON(father != p->real_parent && !ptrace);
 623
 624                 if (father == p->real_parent) {
 625                         /* reparent with a reaper, real father it's us */
 626                         choose_new_parent(p, reaper, child_reaper);
 627                         reparent_thread(p, father, 0);
 628                 } else {
 629                         /* reparent ptraced task to its real parent */
 630                         __ptrace_unlink (p);
 631                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 632                             thread_group_empty(p))
 633                                 do_notify_parent(p, p->exit_signal);
 634                 }
 635
 636                 /*
 637                  * if the ptraced child is a zombie with exit_signal == -1
 638                  * we must collect it before we exit, or it will remain
 639                  * zombie forever since we prevented it from self-reap itself
 640                  * while it was being traced by us, to be able to see it in wait4.
 641                  */
 642                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 643                         list_add(&p->ptrace_list, to_release);
 644         }
 645         list_for_each_safe(_p, _n, &father->ptrace_children) {
 646                 p = list_entry(_p,struct task_struct,ptrace_list);
 647                 choose_new_parent(p, reaper, child_reaper);
 648                 reparent_thread(p, father, 1);
 649         }
 650 }
 651
 652 /*
 653  * Send signals to all our closest relatives so that they know
 654  * to properly mourn us..
 655  */
 656 static void exit_notify(struct task_struct *tsk)
 657 {
 658         int state;
 659         struct task_struct *t;
 660         struct list_head ptrace_dead, *_p, *_n;
 661
 662         ckrm_cb_exit(tsk);
 663
 664         if (signal_pending(tsk) && !tsk->signal->group_exit
 665             && !thread_group_empty(tsk)) {
 666                 /*
 667                  * This occurs when there was a race between our exit
 668                  * syscall and a group signal choosing us as the one to
 669                  * wake up.  It could be that we are the only thread
 670                  * alerted to check for pending signals, but another thread
 671                  * should be woken now to take the signal since we will not.
 672                  * Now we'll wake all the threads in the group just to make
 673                  * sure someone gets all the pending signals.
 674                  */
 675                 read_lock(&tasklist_lock);
 676                 spin_lock_irq(&tsk->sighand->siglock);
 677                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 678                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 679                                 recalc_sigpending_tsk(t);
 680                                 if (signal_pending(t))
 681                                         signal_wake_up(t, 0);
 682                         }
 683                 spin_unlock_irq(&tsk->sighand->siglock);
 684                 read_unlock(&tasklist_lock);
 685         }
 686
 687         write_lock_irq(&tasklist_lock);
 688
 689         /*
 690          * This does two things:
 691          *
 692          * A.  Make init inherit all the child processes
 693          * B.  Check to see if any process groups have become orphaned
 694          *      as a result of our exiting, and if they have any stopped
 695          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 696          */
 697
 698         INIT_LIST_HEAD(&ptrace_dead);
 699         forget_original_parent(tsk, &ptrace_dead);
 700         BUG_ON(!list_empty(&tsk->children));
 701         BUG_ON(!list_empty(&tsk->ptrace_children));
 702
 703         /*
 704          * Check to see if any process groups have become orphaned
 705          * as a result of our exiting, and if they have any stopped
 706          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 707          *
 708          * Case i: Our father is in a different pgrp than we are
 709          * and we were the only connection outside, so our pgrp
 710          * is about to become orphaned.
 711          */
 712
 713         t = tsk->real_parent;
 714
 715         if ((process_group(t) != process_group(tsk)) &&
 716             (t->signal->session == tsk->signal->session) &&
 717             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 718             has_stopped_jobs(process_group(tsk))) {
 719                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
 720                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
 721         }
 722
 723         /* Let father know we died
 724          *
 725          * Thread signals are configurable, but you aren't going to use
 726          * that to send signals to arbitary processes.
 727          * That stops right now.
 728          *
 729          * If the parent exec id doesn't match the exec id we saved
 730          * when we started then we know the parent has changed security
 731          * domain.
 732          *
 733          * If our self_exec id doesn't match our parent_exec_id then
 734          * we have changed execution domain as these two values started
 735          * the same after a fork.
 736          *
 737          */
 738
 739         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 740             ( tsk->parent_exec_id != t->self_exec_id  ||
 741               tsk->self_exec_id != tsk->parent_exec_id)
 742             && !capable(CAP_KILL))
 743                 tsk->exit_signal = SIGCHLD;
 744
 745
 746         /* If something other than our normal parent is ptracing us, then
 747          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 748          * only has special meaning to our real parent.
 749          */
 750         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 751                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 752                 do_notify_parent(tsk, signal);
 753         } else if (tsk->ptrace) {
 754                 do_notify_parent(tsk, SIGCHLD);
 755         }
 756
 757         state = EXIT_ZOMBIE;
 758         if (tsk->exit_signal == -1 && tsk->ptrace == 0)
 759                 state = EXIT_DEAD;
 760         tsk->exit_state = state;
 761
 762         /*
 763          * Clear these here so that update_process_times() won't try to deliver
 764          * itimer, profile or rlimit signals to this task while it is in late exit.
 765          */
 766         tsk->it_virt_value = 0;
 767         tsk->it_prof_value = 0;
 768
 769         write_unlock_irq(&tasklist_lock);
 770
 771         list_for_each_safe(_p, _n, &ptrace_dead) {
 772                 list_del_init(_p);
 773                 t = list_entry(_p,struct task_struct,ptrace_list);
 774                 release_task(t);
 775         }
 776
 777         /* If the process is dead, release it - nobody will wait for it */
 778         if (state == EXIT_DEAD)
 779                 release_task(tsk);
 780
 781         /* PF_DEAD causes final put_task_struct after we schedule. */
 782         preempt_disable();
 783         tsk->flags |= PF_DEAD;
 784 }
 785
 786 fastcall NORET_TYPE void do_exit(long code)
 787 {
 788         struct task_struct *tsk = current;
 789         int group_dead;
 790
 791         profile_task_exit(tsk);
 792
 793         if (unlikely(in_interrupt()))
 794                 panic("Aiee, killing interrupt handler!");
 795         if (unlikely(!tsk->pid))
 796                 panic("Attempted to kill the idle task!");
 797         if (unlikely(tsk->pid == 1))
 798                 panic("Attempted to kill init!");
 799         if (tsk->io_context)
 800                 exit_io_context();
 801         tsk->flags |= PF_EXITING;
 802         del_timer_sync(&tsk->real_timer);
 803
 804         if (unlikely(in_atomic()))
 805                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 806                                 current->comm, current->pid,
 807                                 preempt_count());
 808
 809         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 810                 current->ptrace_message = code;
 811                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 812         }
 813
 814         group_dead = atomic_dec_and_test(&tsk->signal->live);
 815         if (group_dead)
 816                 acct_process(code);
 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         exit_keys(tsk);
 825
 826         if (group_dead && tsk->signal->leader)
 827                 disassociate_ctty(1);
 828
 829         module_put(tsk->thread_info->exec_domain->module);
 830         if (tsk->binfmt)
 831                 module_put(tsk->binfmt->module);
 832
 833         tsk->exit_code = code;
 834         exit_notify(tsk);
 835 #ifdef CONFIG_NUMA
 836         mpol_free(tsk->mempolicy);
 837         tsk->mempolicy = NULL;
 838 #endif
 839
 840         BUG_ON(!(current->flags & PF_DEAD));
 841         schedule();
 842         BUG();
 843         /* Avoid "noreturn function does return".  */
 844         for (;;) ;
 845 }
 846
 847 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 848 {
 849         if (comp)
 850                 complete(comp);
 851
 852         do_exit(code);
 853 }
 854
 855 EXPORT_SYMBOL(complete_and_exit);
 856
 857 asmlinkage long sys_exit(int error_code)
 858 {
 859         do_exit((error_code&0xff)<<8);
 860 }
 861
 862 task_t fastcall *next_thread(const task_t *p)
 863 {
 864 #ifdef CONFIG_SMP
 865         if (!p->sighand)
 866                 BUG();
 867         if (!spin_is_locked(&p->sighand->siglock) &&
 868                                 !rwlock_is_locked(&tasklist_lock))
 869                 BUG();
 870 #endif
 871         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 872 }
 873
 874 EXPORT_SYMBOL(next_thread);
 875
 876 /*
 877  * Take down every thread in the group.  This is called by fatal signals
 878  * as well as by sys_exit_group (below).
 879  */
 880 NORET_TYPE void
 881 do_group_exit(int exit_code)
 882 {
 883         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 884
 885         if (current->signal->group_exit)
 886                 exit_code = current->signal->group_exit_code;
 887         else if (!thread_group_empty(current)) {
 888                 struct signal_struct *const sig = current->signal;
 889                 struct sighand_struct *const sighand = current->sighand;
 890                 read_lock(&tasklist_lock);
 891                 spin_lock_irq(&sighand->siglock);
 892                 if (sig->group_exit)
 893                         /* Another thread got here before we took the lock.  */
 894                         exit_code = sig->group_exit_code;
 895                 else {
 896                         sig->group_exit = 1;
 897                         sig->group_exit_code = exit_code;
 898                         zap_other_threads(current);
 899                 }
 900                 spin_unlock_irq(&sighand->siglock);
 901                 read_unlock(&tasklist_lock);
 902         }
 903
 904         do_exit(exit_code);
 905         /* NOTREACHED */
 906 }
 907
 908 /*
 909  * this kills every thread in the thread group. Note that any externally
 910  * wait4()-ing process will get the correct exit code - even if this
 911  * thread is not the thread group leader.
 912  */
 913 asmlinkage void sys_exit_group(int error_code)
 914 {
 915         do_group_exit((error_code & 0xff) << 8);
 916 }
 917
 918 static int eligible_child(pid_t pid, int options, task_t *p)
 919 {
 920         if (pid > 0) {
 921                 if (p->pid != pid)
 922                         return 0;
 923         } else if (!pid) {
 924                 if (process_group(p) != process_group(current))
 925                         return 0;
 926         } else if (pid != -1) {
 927                 if (process_group(p) != -pid)
 928                         return 0;
 929         }
 930
 931         /*
 932          * Do not consider detached threads that are
 933          * not ptraced:
 934          */
 935         if (p->exit_signal == -1 && !p->ptrace)
 936                 return 0;
 937
 938         /* Wait for all children (clone and not) if __WALL is set;
 939          * otherwise, wait for clone children *only* if __WCLONE is
 940          * set; otherwise, wait for non-clone children *only*.  (Note:
 941          * A "clone" child here is one that reports to its parent
 942          * using a signal other than SIGCHLD.) */
 943         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 944             && !(options & __WALL))
 945                 return 0;
 946         /*
 947          * Do not consider thread group leaders that are
 948          * in a non-empty thread group:
 949          */
 950         if (current->tgid != p->tgid && delay_group_leader(p))
 951                 return 2;
 952
 953         if (security_task_wait(p))
 954                 return 0;
 955
 956         return 1;
 957 }
 958
 959 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 960                                int why, int status,
 961                                struct siginfo __user *infop,
 962                                struct rusage __user *rusagep)
 963 {
 964         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
 965         put_task_struct(p);
 966         if (!retval)
 967                 retval = put_user(SIGCHLD, &infop->si_signo);
 968         if (!retval)
 969                 retval = put_user(0, &infop->si_errno);
 970         if (!retval)
 971                 retval = put_user((short)why, &infop->si_code);
 972         if (!retval)
 973                 retval = put_user(pid, &infop->si_pid);
 974         if (!retval)
 975                 retval = put_user(uid, &infop->si_uid);
 976         if (!retval)
 977                 retval = put_user(status, &infop->si_status);
 978         if (!retval)
 979                 retval = pid;
 980         return retval;
 981 }
 982
 983 /*
 984  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
 985  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
 986  * the lock and this task is uninteresting.  If we return nonzero, we have
 987  * released the lock and the system call should return.
 988  */
 989 static int wait_task_zombie(task_t *p, int noreap,
 990                             struct siginfo __user *infop,
 991                             int __user *stat_addr, struct rusage __user *ru)
 992 {
 993         unsigned long state;
 994         int retval;
 995         int status;
 996
 997         if (unlikely(noreap)) {
 998                 pid_t pid = p->pid;
 999                 uid_t uid = p->uid;
1000                 int exit_code = p->exit_code;
1001                 int why, status;
1002
1003                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1004                         return 0;
1005                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1006                         return 0;
1007                 get_task_struct(p);
1008                 read_unlock(&tasklist_lock);
1009                 if ((exit_code & 0x7f) == 0) {
1010                         why = CLD_EXITED;
1011                         status = exit_code >> 8;
1012                 } else {
1013                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1014                         status = exit_code & 0x7f;
1015                 }
1016                 return wait_noreap_copyout(p, pid, uid, why,
1017                                            status, infop, ru);
1018         }
1019
1020         /*
1021          * Try to move the task's state to DEAD
1022          * only one thread is allowed to do this:
1023          */
1024         state = xchg(&p->exit_state, EXIT_DEAD);
1025         if (state != EXIT_ZOMBIE) {
1026                 BUG_ON(state != EXIT_DEAD);
1027                 return 0;
1028         }
1029         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1030                 /*
1031                  * This can only happen in a race with a ptraced thread
1032                  * dying on another processor.
1033                  */
1034                 return 0;
1035         }
1036
1037         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1038                 /*
1039                  * The resource counters for the group leader are in its
1040                  * own task_struct.  Those for dead threads in the group
1041                  * are in its signal_struct, as are those for the child
1042                  * processes it has previously reaped.  All these
1043                  * accumulate in the parent's signal_struct c* fields.
1044                  *
1045                  * We don't bother to take a lock here to protect these
1046                  * p->signal fields, because they are only touched by
1047                  * __exit_signal, which runs with tasklist_lock
1048                  * write-locked anyway, and so is excluded here.  We do
1049                  * need to protect the access to p->parent->signal fields,
1050                  * as other threads in the parent group can be right
1051                  * here reaping other children at the same time.
1052                  */
1053                 spin_lock_irq(&p->parent->sighand->siglock);
1054                 p->parent->signal->cutime +=
1055                         p->utime + p->signal->utime + p->signal->cutime;
1056                 p->parent->signal->cstime +=
1057                         p->stime + p->signal->stime + 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->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 >= EXIT_ZOMBIE)) {
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->stop_state >= 0)
1267                 return 0;
1268
1269         spin_lock_irq(&p->sighand->siglock);
1270         if (p->signal->stop_state >= 0) { /* Re-check with the lock held.  */
1271                 spin_unlock_irq(&p->sighand->siglock);
1272                 return 0;
1273         }
1274         if (!noreap)
1275                 p->signal->stop_state = 0;
1276         spin_unlock_irq(&p->sighand->siglock);
1277
1278         pid = p->pid;
1279         uid = p->uid;
1280         get_task_struct(p);
1281         read_unlock(&tasklist_lock);
1282
1283         if (!infop) {
1284                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1285                 put_task_struct(p);
1286                 if (!retval && stat_addr)
1287                         retval = put_user(0xffff, stat_addr);
1288                 if (!retval)
1289                         retval = p->pid;
1290         } else {
1291                 retval = wait_noreap_copyout(p, pid, uid,
1292                                              CLD_CONTINUED, SIGCONT,
1293                                              infop, ru);
1294                 BUG_ON(retval == 0);
1295         }
1296
1297         return retval;
1298 }
1299
1300
1301 static inline int my_ptrace_child(struct task_struct *p)
1302 {
1303         if (!(p->ptrace & PT_PTRACED))
1304                 return 0;
1305         if (!(p->ptrace & PT_ATTACHED))
1306                 return 1;
1307         /*
1308          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1309          * we are the attacher.  If we are the real parent, this is a race
1310          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1311          * which we have to switch the parent links, but has already set
1312          * the flags in p->ptrace.
1313          */
1314         return (p->parent != p->real_parent);
1315 }
1316
1317 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1318                     int __user *stat_addr, struct rusage __user *ru)
1319 {
1320         DECLARE_WAITQUEUE(wait, current);
1321         struct task_struct *tsk;
1322         int flag, retval;
1323
1324         add_wait_queue(&current->wait_chldexit,&wait);
1325 repeat:
1326         /*
1327          * We will set this flag if we see any child that might later
1328          * match our criteria, even if we are not able to reap it yet.
1329          */
1330         flag = 0;
1331         current->state = TASK_INTERRUPTIBLE;
1332         read_lock(&tasklist_lock);
1333         tsk = current;
1334         do {
1335                 struct task_struct *p;
1336                 struct list_head *_p;
1337                 int ret;
1338
1339                 list_for_each(_p,&tsk->children) {
1340                         p = list_entry(_p,struct task_struct,sibling);
1341
1342                         ret = eligible_child(pid, options, p);
1343                         if (!ret)
1344                                 continue;
1345
1346                         switch (p->state) {
1347                         case TASK_TRACED:
1348                                 if (!my_ptrace_child(p))
1349                                         continue;
1350                                 /*FALLTHROUGH*/
1351                         case TASK_STOPPED:
1352                                 /*
1353                                  * It's stopped now, so it might later
1354                                  * continue, exit, or stop again.
1355                                  */
1356                                 flag = 1;
1357                                 if (!(options & WUNTRACED) &&
1358                                     !my_ptrace_child(p))
1359                                         continue;
1360                                 retval = wait_task_stopped(p, ret == 2,
1361                                                            (options & WNOWAIT),
1362                                                            infop,
1363                                                            stat_addr, ru);
1364                                 if (retval == -EAGAIN)
1365                                         goto repeat;
1366                                 if (retval != 0) /* He released the lock.  */
1367                                         goto end;
1368                                 break;
1369                         default:
1370                         // case EXIT_DEAD:
1371                                 if (p->exit_state == EXIT_DEAD)
1372                                         continue;
1373                         // case EXIT_ZOMBIE:
1374                                 if (p->exit_state == EXIT_ZOMBIE) {
1375                                         /*
1376                                          * Eligible but we cannot release
1377                                          * it yet:
1378                                          */
1379                                         if (ret == 2)
1380                                                 goto check_continued;
1381                                         if (!likely(options & WEXITED))
1382                                                 continue;
1383                                         retval = wait_task_zombie(
1384                                                 p, (options & WNOWAIT),
1385                                                 infop, stat_addr, ru);
1386                                         /* He released the lock.  */
1387                                         if (retval != 0)
1388                                                 goto end;
1389                                         break;
1390                                 }
1391 check_continued:
1392                                 /*
1393                                  * It's running now, so it might later
1394                                  * exit, stop, or stop and then continue.
1395                                  */
1396                                 flag = 1;
1397                                 if (!unlikely(options & WCONTINUED))
1398                                         continue;
1399                                 retval = wait_task_continued(
1400                                         p, (options & WNOWAIT),
1401                                         infop, stat_addr, ru);
1402                                 if (retval != 0) /* He released the lock.  */
1403                                         goto end;
1404                                 break;
1405                         }
1406                 }
1407                 if (!flag) {
1408                         list_for_each(_p, &tsk->ptrace_children) {
1409                                 p = list_entry(_p, struct task_struct,
1410                                                 ptrace_list);
1411                                 if (!eligible_child(pid, options, p))
1412                                         continue;
1413                                 flag = 1;
1414                                 break;
1415                         }
1416                 }
1417                 if (options & __WNOTHREAD)
1418                         break;
1419                 tsk = next_thread(tsk);
1420                 if (tsk->signal != current->signal)
1421                         BUG();
1422         } while (tsk != current);
1423
1424         read_unlock(&tasklist_lock);
1425         if (flag) {
1426                 retval = 0;
1427                 if (options & WNOHANG)
1428                         goto end;
1429                 retval = -ERESTARTSYS;
1430                 if (signal_pending(current))
1431                         goto end;
1432                 schedule();
1433                 goto repeat;
1434         }
1435         retval = -ECHILD;
1436 end:
1437         current->state = TASK_RUNNING;
1438         remove_wait_queue(&current->wait_chldexit,&wait);
1439         if (infop) {
1440                 if (retval > 0)
1441                 retval = 0;
1442                 else {
1443                         /*
1444                          * For a WNOHANG return, clear out all the fields
1445                          * we would set so the user can easily tell the
1446                          * difference.
1447                          */
1448                         if (!retval)
1449                                 retval = put_user(0, &infop->si_signo);
1450                         if (!retval)
1451                                 retval = put_user(0, &infop->si_errno);
1452                         if (!retval)
1453                                 retval = put_user(0, &infop->si_code);
1454                         if (!retval)
1455                                 retval = put_user(0, &infop->si_pid);
1456                         if (!retval)
1457                                 retval = put_user(0, &infop->si_uid);
1458                         if (!retval)
1459                                 retval = put_user(0, &infop->si_status);
1460                 }
1461         }
1462         return retval;
1463 }
1464
1465 asmlinkage long sys_waitid(int which, pid_t pid,
1466                            struct siginfo __user *infop, int options,
1467                            struct rusage __user *ru)
1468 {
1469         long ret;
1470
1471         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1472                 return -EINVAL;
1473         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1474                 return -EINVAL;
1475
1476         switch (which) {
1477         case P_ALL:
1478                 pid = -1;
1479                 break;
1480         case P_PID:
1481                 if (pid <= 0)
1482                         return -EINVAL;
1483                 break;
1484         case P_PGID:
1485                 if (pid <= 0)
1486                         return -EINVAL;
1487                 pid = -pid;
1488                 break;
1489         default:
1490                 return -EINVAL;
1491         }
1492
1493         ret = do_wait(pid, options, infop, NULL, ru);
1494
1495         /* avoid REGPARM breakage on x86: */
1496         prevent_tail_call(ret);
1497         return ret;
1498 }
1499
1500 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1501                           int options, struct rusage __user *ru)
1502 {
1503         long ret;
1504
1505         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1506                         __WNOTHREAD|__WCLONE|__WALL))
1507                 return -EINVAL;
1508         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1509
1510         /* avoid REGPARM breakage on x86: */
1511         prevent_tail_call(ret);
1512         return ret;
1513 }
1514
1515 #ifdef __ARCH_WANT_SYS_WAITPID
1516
1517 /*
1518  * sys_waitpid() remains for compatibility. waitpid() should be
1519  * implemented by calling sys_wait4() from libc.a.
1520  */
1521 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1522 {
1523         return sys_wait4(pid, stat_addr, options, NULL);
1524 }
1525
1526 #endif