kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/mm.h>
   8 #include <linux/slab.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/smp_lock.h>
  11 #include <linux/module.h>
  12 #include <linux/capability.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/tracehook.h>
  25 #include <linux/profile.h>
  26 #include <linux/mount.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/mempolicy.h>
  29 #include <linux/taskstats_kern.h>
  30 #include <linux/delayacct.h>
  31 #include <linux/cpuset.h>
  32 #include <linux/syscalls.h>
  33 #include <linux/signal.h>
  34 #include <linux/posix-timers.h>
  35 #include <linux/cn_proc.h>
  36 #include <linux/mutex.h>
  37 #include <linux/futex.h>
  38 #include <linux/compat.h>
  39 #include <linux/pipe_fs_i.h>
  40 #include <linux/audit.h> /* for audit_free() */
  41 #include <linux/resource.h>
  42 #include <linux/vs_limit.h>
  43 #include <linux/vs_context.h>
  44 #include <linux/vs_network.h>
  45
  46 #include <asm/uaccess.h>
  47 #include <asm/unistd.h>
  48 #include <asm/pgtable.h>
  49 #include <asm/mmu_context.h>
  50
  51 extern void sem_exit (void);
  52 extern struct task_struct *child_reaper;
  53
  54 static void exit_mm(struct task_struct * tsk);
  55
  56 static void __unhash_process(struct task_struct *p)
  57 {
  58         nr_threads--;
  59         detach_pid(p, PIDTYPE_PID);
  60         if (thread_group_leader(p)) {
  61                 detach_pid(p, PIDTYPE_PGID);
  62                 detach_pid(p, PIDTYPE_SID);
  63
  64                 list_del_rcu(&p->tasks);
  65                 __get_cpu_var(process_counts)--;
  66         }
  67         list_del_rcu(&p->thread_group);
  68         remove_parent(p);
  69 }
  70
  71 /*
  72  * This function expects the tasklist_lock write-locked.
  73  */
  74 static void __exit_signal(struct task_struct *tsk)
  75 {
  76         struct signal_struct *sig = tsk->signal;
  77         struct sighand_struct *sighand;
  78
  79         BUG_ON(!sig);
  80         BUG_ON(!atomic_read(&sig->count));
  81
  82         rcu_read_lock();
  83         sighand = rcu_dereference(tsk->sighand);
  84         spin_lock(&sighand->siglock);
  85
  86         posix_cpu_timers_exit(tsk);
  87         if (atomic_dec_and_test(&sig->count))
  88                 posix_cpu_timers_exit_group(tsk);
  89         else {
  90                 /*
  91                  * If there is any task waiting for the group exit
  92                  * then notify it:
  93                  */
  94                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
  95                         wake_up_process(sig->group_exit_task);
  96                         sig->group_exit_task = NULL;
  97                 }
  98                 if (tsk == sig->curr_target)
  99                         sig->curr_target = next_thread(tsk);
 100                 /*
 101                  * Accumulate here the counters for all threads but the
 102                  * group leader as they die, so they can be added into
 103                  * the process-wide totals when those are taken.
 104                  * The group leader stays around as a zombie as long
 105                  * as there are other threads.  When it gets reaped,
 106                  * the exit.c code will add its counts into these totals.
 107                  * We won't ever get here for the group leader, since it
 108                  * will have been the last reference on the signal_struct.
 109                  */
 110                 sig->utime = cputime_add(sig->utime, tsk->utime);
 111                 sig->stime = cputime_add(sig->stime, tsk->stime);
 112                 sig->min_flt += tsk->min_flt;
 113                 sig->maj_flt += tsk->maj_flt;
 114                 sig->nvcsw += tsk->nvcsw;
 115                 sig->nivcsw += tsk->nivcsw;
 116                 sig->sched_time += tsk->sched_time;
 117                 sig = NULL; /* Marker for below. */
 118         }
 119
 120         __unhash_process(tsk);
 121
 122         tsk->signal = NULL;
 123         tsk->sighand = NULL;
 124         spin_unlock(&sighand->siglock);
 125         rcu_read_unlock();
 126
 127         __cleanup_sighand(sighand);
 128         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 129         flush_sigqueue(&tsk->pending);
 130         if (sig) {
 131                 flush_sigqueue(&sig->shared_pending);
 132                 __cleanup_signal(sig);
 133         }
 134 }
 135
 136 static void delayed_put_task_struct(struct rcu_head *rhp)
 137 {
 138         put_task_struct(container_of(rhp, struct task_struct, rcu));
 139 }
 140
 141 void release_task(struct task_struct * p)
 142 {
 143         struct task_struct *leader;
 144         int zap_leader;
 145 repeat:
 146         tracehook_release_task(p);
 147         atomic_dec(&p->user->processes);
 148         write_lock_irq(&tasklist_lock);
 149         BUG_ON(tracehook_check_released(p));
 150         __exit_signal(p);
 151
 152         /*
 153          * If we are the last non-leader member of the thread
 154          * group, and the leader is zombie, then notify the
 155          * group leader's parent process. (if it wants notification.)
 156          */
 157         zap_leader = 0;
 158         leader = p->group_leader;
 159         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 160                 BUG_ON(leader->exit_signal == -1);
 161                 do_notify_parent(leader, leader->exit_signal);
 162                 /*
 163                  * If we were the last child thread and the leader has
 164                  * exited already, and the leader's parent ignores SIGCHLD,
 165                  * then we are the one who should release the leader.
 166                  *
 167                  * do_notify_parent() will have marked it self-reaping in
 168                  * that case.
 169                  */
 170                 zap_leader = (leader->exit_signal == -1);
 171         }
 172
 173         sched_exit(p);
 174         write_unlock_irq(&tasklist_lock);
 175         proc_flush_task(p);
 176         release_thread(p);
 177         call_rcu(&p->rcu, delayed_put_task_struct);
 178
 179         p = leader;
 180         if (unlikely(zap_leader))
 181                 goto repeat;
 182 }
 183
 184 /*
 185  * This checks not only the pgrp, but falls back on the pid if no
 186  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 187  * without this...
 188  */
 189 int session_of_pgrp(int pgrp)
 190 {
 191         struct task_struct *p;
 192         int sid = -1;
 193
 194         read_lock(&tasklist_lock);
 195         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 196                 if (p->signal->session > 0) {
 197                         sid = p->signal->session;
 198                         goto out;
 199                 }
 200         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 201         p = find_task_by_pid(pgrp);
 202         if (p)
 203                 sid = p->signal->session;
 204 out:
 205         read_unlock(&tasklist_lock);
 206
 207         return sid;
 208 }
 209
 210 /*
 211  * Determine if a process group is "orphaned", according to the POSIX
 212  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 213  * by terminal-generated stop signals.  Newly orphaned process groups are
 214  * to receive a SIGHUP and a SIGCONT.
 215  *
 216  * "I ask you, have you ever known what it is to be an orphan?"
 217  */
 218 static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
 219 {
 220         struct task_struct *p;
 221         int ret = 1;
 222
 223         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 224                 if (p == ignored_task
 225                                 || p->exit_state
 226                                 || p->parent->pid == 1)
 227                         continue;
 228                 if (process_group(p->parent) != pgrp
 229                             && p->parent->signal->session == p->signal->session) {
 230                         ret = 0;
 231                         break;
 232                 }
 233         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 234         return ret;     /* (sighing) "Often!" */
 235 }
 236
 237 int is_orphaned_pgrp(int pgrp)
 238 {
 239         int retval;
 240
 241         read_lock(&tasklist_lock);
 242         retval = will_become_orphaned_pgrp(pgrp, NULL);
 243         read_unlock(&tasklist_lock);
 244
 245         return retval;
 246 }
 247
 248 static int has_stopped_jobs(int pgrp)
 249 {
 250         int retval = 0;
 251         struct task_struct *p;
 252
 253         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 254                 if (p->state != TASK_STOPPED)
 255                         continue;
 256
 257                 retval = 1;
 258                 break;
 259         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 260         return retval;
 261 }
 262
 263 /**
 264  * reparent_to_init - Reparent the calling kernel thread to the init task.
 265  *
 266  * If a kernel thread is launched as a result of a system call, or if
 267  * it ever exits, it should generally reparent itself to init so that
 268  * it is correctly cleaned up on exit.
 269  *
 270  * The various task state such as scheduling policy and priority may have
 271  * been inherited from a user process, so we reset them to sane values here.
 272  *
 273  * NOTE that reparent_to_init() gives the caller full capabilities.
 274  */
 275 static void reparent_to_init(void)
 276 {
 277         write_lock_irq(&tasklist_lock);
 278
 279         /* Reparent to init */
 280         remove_parent(current);
 281         current->parent = child_reaper;
 282         add_parent(current);
 283
 284         /* Set the exit signal to SIGCHLD so we signal init on exit */
 285         current->exit_signal = SIGCHLD;
 286
 287         if ((current->policy == SCHED_NORMAL ||
 288                         current->policy == SCHED_BATCH)
 289                                 && (task_nice(current) < 0))
 290                 set_user_nice(current, 0);
 291         /* cpus_allowed? */
 292         /* rt_priority? */
 293         /* signals? */
 294         security_task_reparent_to_init(current);
 295         memcpy(current->signal->rlim, init_task.signal->rlim,
 296                sizeof(current->signal->rlim));
 297         atomic_inc(&(INIT_USER->__count));
 298         write_unlock_irq(&tasklist_lock);
 299         switch_uid(INIT_USER);
 300 }
 301
 302 void __set_special_pids(pid_t session, pid_t pgrp)
 303 {
 304         struct task_struct *curr = current->group_leader;
 305
 306         if (curr->signal->session != session) {
 307                 detach_pid(curr, PIDTYPE_SID);
 308                 curr->signal->session = session;
 309                 attach_pid(curr, PIDTYPE_SID, session);
 310         }
 311         if (process_group(curr) != pgrp) {
 312                 detach_pid(curr, PIDTYPE_PGID);
 313                 curr->signal->pgrp = pgrp;
 314                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 315         }
 316 }
 317
 318 void set_special_pids(pid_t session, pid_t pgrp)
 319 {
 320         write_lock_irq(&tasklist_lock);
 321         __set_special_pids(session, pgrp);
 322         write_unlock_irq(&tasklist_lock);
 323 }
 324
 325 /*
 326  * Let kernel threads use this to say that they
 327  * allow a certain signal (since daemonize() will
 328  * have disabled all of them by default).
 329  */
 330 int allow_signal(int sig)
 331 {
 332         if (!valid_signal(sig) || sig < 1)
 333                 return -EINVAL;
 334
 335         spin_lock_irq(&current->sighand->siglock);
 336         sigdelset(&current->blocked, sig);
 337         if (!current->mm) {
 338                 /* Kernel threads handle their own signals.
 339                    Let the signal code know it'll be handled, so
 340                    that they don't get converted to SIGKILL or
 341                    just silently dropped */
 342                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 343         }
 344         recalc_sigpending();
 345         spin_unlock_irq(&current->sighand->siglock);
 346         return 0;
 347 }
 348
 349 EXPORT_SYMBOL(allow_signal);
 350
 351 int disallow_signal(int sig)
 352 {
 353         if (!valid_signal(sig) || sig < 1)
 354                 return -EINVAL;
 355
 356         spin_lock_irq(&current->sighand->siglock);
 357         sigaddset(&current->blocked, sig);
 358         recalc_sigpending();
 359         spin_unlock_irq(&current->sighand->siglock);
 360         return 0;
 361 }
 362
 363 EXPORT_SYMBOL(disallow_signal);
 364
 365 /*
 366  *      Put all the gunge required to become a kernel thread without
 367  *      attached user resources in one place where it belongs.
 368  */
 369
 370 void daemonize(const char *name, ...)
 371 {
 372         va_list args;
 373         struct fs_struct *fs;
 374         sigset_t blocked;
 375
 376         va_start(args, name);
 377         vsnprintf(current->comm, sizeof(current->comm), name, args);
 378         va_end(args);
 379
 380         /*
 381          * If we were started as result of loading a module, close all of the
 382          * user space pages.  We don't need them, and if we didn't close them
 383          * they would be locked into memory.
 384          */
 385         exit_mm(current);
 386
 387         set_special_pids(1, 1);
 388         mutex_lock(&tty_mutex);
 389         current->signal->tty = NULL;
 390         mutex_unlock(&tty_mutex);
 391
 392         /* Block and flush all signals */
 393         sigfillset(&blocked);
 394         sigprocmask(SIG_BLOCK, &blocked, NULL);
 395         flush_signals(current);
 396
 397         /* Become as one with the init task */
 398
 399         exit_fs(current);       /* current->fs->count--; */
 400         fs = init_task.fs;
 401         current->fs = fs;
 402         atomic_inc(&fs->count);
 403         exit_namespace(current);
 404         current->namespace = init_task.namespace;
 405         get_namespace(current->namespace);
 406         exit_files(current);
 407         current->files = init_task.files;
 408         atomic_inc(&current->files->count);
 409
 410         reparent_to_init();
 411 }
 412
 413 EXPORT_SYMBOL(daemonize);
 414
 415 static void close_files(struct files_struct * files)
 416 {
 417         int i, j;
 418         struct fdtable *fdt;
 419
 420         j = 0;
 421
 422         /*
 423          * It is safe to dereference the fd table without RCU or
 424          * ->file_lock because this is the last reference to the
 425          * files structure.
 426          */
 427         fdt = files_fdtable(files);
 428         for (;;) {
 429                 unsigned long set;
 430                 i = j * __NFDBITS;
 431                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 432                         break;
 433                 set = fdt->open_fds->fds_bits[j++];
 434                 while (set) {
 435                         if (set & 1) {
 436                                 struct file * file = xchg(&fdt->fd[i], NULL);
 437                                 if (file)
 438                                         filp_close(file, files);
 439                                 vx_openfd_dec(i);
 440                         }
 441                         i++;
 442                         set >>= 1;
 443                         cond_resched();
 444                 }
 445         }
 446 }
 447
 448 struct files_struct *get_files_struct(struct task_struct *task)
 449 {
 450         struct files_struct *files;
 451
 452         task_lock(task);
 453         files = task->files;
 454         if (files)
 455                 atomic_inc(&files->count);
 456         task_unlock(task);
 457
 458         return files;
 459 }
 460
 461 void fastcall put_files_struct(struct files_struct *files)
 462 {
 463         struct fdtable *fdt;
 464
 465         if (atomic_dec_and_test(&files->count)) {
 466                 close_files(files);
 467                 /*
 468                  * Free the fd and fdset arrays if we expanded them.
 469                  * If the fdtable was embedded, pass files for freeing
 470                  * at the end of the RCU grace period. Otherwise,
 471                  * you can free files immediately.
 472                  */
 473                 fdt = files_fdtable(files);
 474                 if (fdt == &files->fdtab)
 475                         fdt->free_files = files;
 476                 else
 477                         kmem_cache_free(files_cachep, files);
 478                 free_fdtable(fdt);
 479         }
 480 }
 481
 482 EXPORT_SYMBOL(put_files_struct);
 483
 484 static inline void __exit_files(struct task_struct *tsk)
 485 {
 486         struct files_struct * files = tsk->files;
 487
 488         if (files) {
 489                 task_lock(tsk);
 490                 tsk->files = NULL;
 491                 task_unlock(tsk);
 492                 put_files_struct(files);
 493         }
 494 }
 495
 496 void exit_files(struct task_struct *tsk)
 497 {
 498         __exit_files(tsk);
 499 }
 500
 501 static inline void __put_fs_struct(struct fs_struct *fs)
 502 {
 503         /* No need to hold fs->lock if we are killing it */
 504         if (atomic_dec_and_test(&fs->count)) {
 505                 dput(fs->root);
 506                 mntput(fs->rootmnt);
 507                 dput(fs->pwd);
 508                 mntput(fs->pwdmnt);
 509                 if (fs->altroot) {
 510                         dput(fs->altroot);
 511                         mntput(fs->altrootmnt);
 512                 }
 513                 kmem_cache_free(fs_cachep, fs);
 514         }
 515 }
 516
 517 void put_fs_struct(struct fs_struct *fs)
 518 {
 519         __put_fs_struct(fs);
 520 }
 521
 522 static inline void __exit_fs(struct task_struct *tsk)
 523 {
 524         struct fs_struct * fs = tsk->fs;
 525
 526         if (fs) {
 527                 task_lock(tsk);
 528                 tsk->fs = NULL;
 529                 task_unlock(tsk);
 530                 __put_fs_struct(fs);
 531         }
 532 }
 533
 534 void exit_fs(struct task_struct *tsk)
 535 {
 536         __exit_fs(tsk);
 537 }
 538
 539 EXPORT_SYMBOL_GPL(exit_fs);
 540
 541 /*
 542  * Turn us into a lazy TLB process if we
 543  * aren't already..
 544  */
 545 static void exit_mm(struct task_struct * tsk)
 546 {
 547         struct mm_struct *mm = tsk->mm;
 548
 549         mm_release(tsk, mm);
 550         if (!mm)
 551                 return;
 552         /*
 553          * Serialize with any possible pending coredump.
 554          * We must hold mmap_sem around checking core_waiters
 555          * and clearing tsk->mm.  The core-inducing thread
 556          * will increment core_waiters for each thread in the
 557          * group with ->mm != NULL.
 558          */
 559         down_read(&mm->mmap_sem);
 560         if (mm->core_waiters) {
 561                 up_read(&mm->mmap_sem);
 562                 down_write(&mm->mmap_sem);
 563                 if (!--mm->core_waiters)
 564                         complete(mm->core_startup_done);
 565                 up_write(&mm->mmap_sem);
 566
 567                 wait_for_completion(&mm->core_done);
 568                 down_read(&mm->mmap_sem);
 569         }
 570         atomic_inc(&mm->mm_count);
 571         BUG_ON(mm != tsk->active_mm);
 572         /* more a memory barrier than a real lock */
 573         task_lock(tsk);
 574         tsk->mm = NULL;
 575         up_read(&mm->mmap_sem);
 576         enter_lazy_tlb(mm, current);
 577         task_unlock(tsk);
 578         mmput(mm);
 579 }
 580
 581 static inline void
 582 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 583 {
 584         /* check for reaper context */
 585         vxwprintk((p->xid != reaper->xid) && (reaper != child_reaper),
 586                 "rogue reaper: %p[%d,#%u] <> %p[%d,#%u]",
 587                 p, p->pid, p->xid, reaper, reaper->pid, reaper->xid);
 588
 589         /*
 590          * Make sure we're not reparenting to ourselves and that
 591          * the parent is not a zombie.
 592          */
 593         BUG_ON(p == reaper || reaper->exit_state);
 594         p->parent = reaper;
 595 }
 596
 597 static void
 598 reparent_thread(struct task_struct *p, struct task_struct *father)
 599 {
 600         /* We don't want people slaying init.  */
 601         if (p->exit_signal != -1)
 602                 p->exit_signal = SIGCHLD;
 603
 604         if (p->pdeath_signal)
 605                 /* We already hold the tasklist_lock here.  */
 606                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 607
 608         /* Move the child from its dying parent to the new one.  */
 609         list_move_tail(&p->sibling, &p->parent->children);
 610
 611         /* If we'd notified the old parent about this child's death,
 612          * also notify the new parent.
 613          */
 614         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 615             thread_group_empty(p))
 616                 do_notify_parent(p, p->exit_signal);
 617
 618         /*
 619          * process group orphan check
 620          * Case ii: Our child is in a different pgrp
 621          * than we are, and it was the only connection
 622          * outside, so the child pgrp is now orphaned.
 623          */
 624         if ((process_group(p) != process_group(father)) &&
 625             (p->signal->session == father->signal->session)) {
 626                 int pgrp = process_group(p);
 627
 628                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 629                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 630                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 631                 }
 632         }
 633 }
 634
 635 /*
 636  * When we die, we re-parent all our children.
 637  * Try to give them to another thread in our thread
 638  * group, and if no such member exists, give it to
 639  * the global child reaper process (ie "init")
 640  */
 641 static void
 642 forget_original_parent(struct task_struct *father)
 643 {
 644         struct task_struct *p, *reaper = father;
 645         struct list_head *_p, *_n;
 646
 647         do {
 648                 reaper = next_thread(reaper);
 649                 if (reaper == father) {
 650                         reaper = vx_child_reaper(father);
 651                         break;
 652                 }
 653         } while (reaper->exit_state);
 654
 655         list_for_each_safe(_p, _n, &father->children) {
 656                 p = list_entry(_p, struct task_struct, sibling);
 657                 choose_new_parent(p, vx_child_reaper(p));
 658                 reparent_thread(p, father);
 659         }
 660 }
 661
 662 /*
 663  * Send signals to all our closest relatives so that they know
 664  * to properly mourn us..
 665  */
 666 static void exit_notify(struct task_struct *tsk)
 667 {
 668         int state;
 669         struct task_struct *t;
 670         int noreap;
 671         void *cookie;
 672
 673         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 674             && !thread_group_empty(tsk)) {
 675                 /*
 676                  * This occurs when there was a race between our exit
 677                  * syscall and a group signal choosing us as the one to
 678                  * wake up.  It could be that we are the only thread
 679                  * alerted to check for pending signals, but another thread
 680                  * should be woken now to take the signal since we will not.
 681                  * Now we'll wake all the threads in the group just to make
 682                  * sure someone gets all the pending signals.
 683                  */
 684                 read_lock(&tasklist_lock);
 685                 spin_lock_irq(&tsk->sighand->siglock);
 686                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 687                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 688                                 recalc_sigpending_tsk(t);
 689                                 if (signal_pending(t))
 690                                         signal_wake_up(t, 0);
 691                         }
 692                 spin_unlock_irq(&tsk->sighand->siglock);
 693                 read_unlock(&tasklist_lock);
 694         }
 695
 696         write_lock_irq(&tasklist_lock);
 697
 698         /*
 699          * This does two things:
 700          *
 701          * A.  Make init inherit all the child processes
 702          * B.  Check to see if any process groups have become orphaned
 703          *      as a result of our exiting, and if they have any stopped
 704          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 705          */
 706
 707         forget_original_parent(tsk);
 708         BUG_ON(!list_empty(&tsk->children));
 709
 710         /*
 711          * Check to see if any process groups have become orphaned
 712          * as a result of our exiting, and if they have any stopped
 713          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 714          *
 715          * Case i: Our father is in a different pgrp than we are
 716          * and we were the only connection outside, so our pgrp
 717          * is about to become orphaned.
 718          */
 719
 720         t = tsk->parent;
 721
 722         if ((process_group(t) != process_group(tsk)) &&
 723             (t->signal->session == tsk->signal->session) &&
 724             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 725             has_stopped_jobs(process_group(tsk))) {
 726                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 727                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 728         }
 729
 730         /* Let father know we died
 731          *
 732          * Thread signals are configurable, but you aren't going to use
 733          * that to send signals to arbitary processes.
 734          * That stops right now.
 735          *
 736          * If the parent exec id doesn't match the exec id we saved
 737          * when we started then we know the parent has changed security
 738          * domain.
 739          *
 740          * If our self_exec id doesn't match our parent_exec_id then
 741          * we have changed execution domain as these two values started
 742          * the same after a fork.
 743          *
 744          */
 745
 746         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 747             ( tsk->parent_exec_id != t->self_exec_id  ||
 748               tsk->self_exec_id != tsk->parent_exec_id)
 749             && !capable(CAP_KILL))
 750                 tsk->exit_signal = SIGCHLD;
 751
 752         if (!tracehook_notify_death(tsk, &noreap, &cookie)
 753             && tsk->exit_signal != -1 && thread_group_empty(tsk))
 754                 do_notify_parent(tsk, tsk->exit_signal);
 755
 756         state = EXIT_ZOMBIE;
 757         if (tsk->exit_signal == -1 && !noreap)
 758                 state = EXIT_DEAD;
 759         tsk->exit_state = state;
 760
 761         write_unlock_irq(&tasklist_lock);
 762
 763         tracehook_report_death(tsk, state, cookie);
 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
 770 fastcall NORET_TYPE void do_exit(long code)
 771 {
 772         struct task_struct *tsk = current;
 773         struct taskstats *tidstats;
 774         int group_dead;
 775         unsigned int mycpu;
 776
 777         profile_task_exit(tsk);
 778
 779         WARN_ON(atomic_read(&tsk->fs_excl));
 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 == child_reaper))
 786                 panic("Attempted to kill init!");
 787
 788         tracehook_report_exit(&code);
 789
 790         /*
 791          * We're taking recursive faults here in do_exit. Safest is to just
 792          * leave this task alone and wait for reboot.
 793          */
 794         if (unlikely(tsk->flags & PF_EXITING)) {
 795                 printk(KERN_ALERT
 796                         "Fixing recursive fault but reboot is needed!\n");
 797                 if (tsk->io_context)
 798                         exit_io_context();
 799                 set_current_state(TASK_UNINTERRUPTIBLE);
 800                 schedule();
 801         }
 802
 803         tsk->flags |= PF_EXITING;
 804
 805         ptrace_exit(tsk);
 806
 807         if (unlikely(in_atomic()))
 808                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 809                                 current->comm, current->pid,
 810                                 preempt_count());
 811
 812         taskstats_exit_alloc(&tidstats, &mycpu);
 813
 814         acct_update_integrals(tsk);
 815         if (tsk->mm) {
 816                 update_hiwater_rss(tsk->mm);
 817                 update_hiwater_vm(tsk->mm);
 818         }
 819         group_dead = atomic_dec_and_test(&tsk->signal->live);
 820         if (group_dead) {
 821                 hrtimer_cancel(&tsk->signal->real_timer);
 822                 exit_itimers(tsk->signal);
 823         }
 824
 825         if (current->tux_info) {
 826 #ifdef CONFIG_TUX_DEBUG
 827                 printk("Possibly unexpected TUX-thread exit(%ld) at %p?\n",
 828                         code, __builtin_return_address(0));
 829 #endif
 830                 current->tux_exit();
 831         }
 832
 833         acct_collect(code, group_dead);
 834         if (unlikely(tsk->robust_list))
 835                 exit_robust_list(tsk);
 836 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 837         if (unlikely(tsk->compat_robust_list))
 838                 compat_exit_robust_list(tsk);
 839 #endif
 840         if (unlikely(tsk->audit_context))
 841                 audit_free(tsk);
 842         taskstats_exit_send(tsk, tidstats, group_dead, mycpu);
 843         taskstats_exit_free(tidstats);
 844
 845         exit_mm(tsk);
 846
 847         if (group_dead)
 848                 acct_process();
 849         exit_sem(tsk);
 850         __exit_files(tsk);
 851         __exit_fs(tsk);
 852         exit_namespace(tsk);
 853         exit_thread();
 854         cpuset_exit(tsk);
 855         exit_keys(tsk);
 856
 857         if (group_dead && tsk->signal->leader)
 858                 disassociate_ctty(1);
 859
 860         module_put(task_thread_info(tsk)->exec_domain->module);
 861         if (tsk->binfmt)
 862                 module_put(tsk->binfmt->module);
 863
 864         tsk->exit_code = code;
 865         proc_exit_connector(tsk);
 866         /* needs to stay before exit_notify() */
 867         exit_vx_info_early(tsk, code);
 868         exit_notify(tsk);
 869 #ifdef CONFIG_NUMA
 870         mpol_free(tsk->mempolicy);
 871         tsk->mempolicy = NULL;
 872 #endif
 873         /*
 874          * This must happen late, after the PID is not
 875          * hashed anymore:
 876          */
 877         if (unlikely(!list_empty(&tsk->pi_state_list)))
 878                 exit_pi_state_list(tsk);
 879         if (unlikely(current->pi_state_cache))
 880                 kfree(current->pi_state_cache);
 881         /*
 882          * Make sure we are holding no locks:
 883          */
 884         debug_check_no_locks_held(tsk);
 885
 886         if (tsk->io_context)
 887                 exit_io_context();
 888
 889         if (tsk->splice_pipe)
 890                 __free_pipe_info(tsk->splice_pipe);
 891
 892         /* needs to stay after exit_notify() */
 893         exit_vx_info(tsk, code);
 894         exit_nx_info(tsk);
 895
 896         /* PF_DEAD causes final put_task_struct after we schedule. */
 897         preempt_disable();
 898         BUG_ON(tsk->flags & PF_DEAD);
 899         tsk->flags |= PF_DEAD;
 900
 901         schedule();
 902         BUG();
 903         /* Avoid "noreturn function does return".  */
 904         for (;;) ;
 905 }
 906
 907 EXPORT_SYMBOL_GPL(do_exit);
 908
 909 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 910 {
 911         if (comp)
 912                 complete(comp);
 913
 914         do_exit(code);
 915 }
 916
 917 EXPORT_SYMBOL(complete_and_exit);
 918
 919 asmlinkage long sys_exit(int error_code)
 920 {
 921         do_exit((error_code&0xff)<<8);
 922 }
 923
 924 /*
 925  * Take down every thread in the group.  This is called by fatal signals
 926  * as well as by sys_exit_group (below).
 927  */
 928 NORET_TYPE void
 929 do_group_exit(int exit_code)
 930 {
 931         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 932
 933         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 934                 exit_code = current->signal->group_exit_code;
 935         else if (!thread_group_empty(current)) {
 936                 struct signal_struct *const sig = current->signal;
 937                 struct sighand_struct *const sighand = current->sighand;
 938                 spin_lock_irq(&sighand->siglock);
 939                 if (sig->flags & SIGNAL_GROUP_EXIT)
 940                         /* Another thread got here before we took the lock.  */
 941                         exit_code = sig->group_exit_code;
 942                 else {
 943                         sig->group_exit_code = exit_code;
 944                         zap_other_threads(current);
 945                 }
 946                 spin_unlock_irq(&sighand->siglock);
 947         }
 948
 949         do_exit(exit_code);
 950         /* NOTREACHED */
 951 }
 952
 953 /*
 954  * this kills every thread in the thread group. Note that any externally
 955  * wait4()-ing process will get the correct exit code - even if this
 956  * thread is not the thread group leader.
 957  */
 958 asmlinkage void sys_exit_group(int error_code)
 959 {
 960         do_group_exit((error_code & 0xff) << 8);
 961 }
 962
 963 static int eligible_child(pid_t pid, int options, struct task_struct *p)
 964 {
 965         if (pid > 0) {
 966                 if (p->pid != pid)
 967                         return 0;
 968         } else if (!pid) {
 969                 if (process_group(p) != process_group(current))
 970                         return 0;
 971         } else if (pid != -1) {
 972                 if (process_group(p) != -pid)
 973                         return 0;
 974         }
 975
 976         /*
 977          * Do not consider detached threads.
 978          */
 979         if (p->exit_signal == -1)
 980                 return 0;
 981
 982         /* Wait for all children (clone and not) if __WALL is set;
 983          * otherwise, wait for clone children *only* if __WCLONE is
 984          * set; otherwise, wait for non-clone children *only*.  (Note:
 985          * A "clone" child here is one that reports to its parent
 986          * using a signal other than SIGCHLD.) */
 987         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 988             && !(options & __WALL))
 989                 return 0;
 990         /*
 991          * Do not consider thread group leaders that are
 992          * in a non-empty thread group:
 993          */
 994         if (delay_group_leader(p))
 995                 return 2;
 996
 997         if (security_task_wait(p))
 998                 return 0;
 999
1000         return 1;
1001 }
1002
1003 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1004                                int why, int status,
1005                                struct siginfo __user *infop,
1006                                struct rusage __user *rusagep)
1007 {
1008         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1009
1010         put_task_struct(p);
1011         if (!retval)
1012                 retval = put_user(SIGCHLD, &infop->si_signo);
1013         if (!retval)
1014                 retval = put_user(0, &infop->si_errno);
1015         if (!retval)
1016                 retval = put_user((short)why, &infop->si_code);
1017         if (!retval)
1018                 retval = put_user(pid, &infop->si_pid);
1019         if (!retval)
1020                 retval = put_user(uid, &infop->si_uid);
1021         if (!retval)
1022                 retval = put_user(status, &infop->si_status);
1023         if (!retval)
1024                 retval = pid;
1025         return retval;
1026 }
1027
1028 /*
1029  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1030  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1031  * the lock and this task is uninteresting.  If we return nonzero, we have
1032  * released the lock and the system call should return.
1033  */
1034 static int wait_task_zombie(struct task_struct *p, int noreap,
1035                             struct siginfo __user *infop,
1036                             int __user *stat_addr, struct rusage __user *ru)
1037 {
1038         unsigned long state;
1039         int retval;
1040         int status;
1041
1042         if (unlikely(noreap)) {
1043                 pid_t pid = p->pid;
1044                 uid_t uid = p->uid;
1045                 int exit_code = p->exit_code;
1046                 int why, status;
1047
1048                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1049                         return 0;
1050                 if (unlikely(p->exit_signal == -1))
1051                         return 0;
1052                 get_task_struct(p);
1053                 read_unlock(&tasklist_lock);
1054                 if ((exit_code & 0x7f) == 0) {
1055                         why = CLD_EXITED;
1056                         status = exit_code >> 8;
1057                 } else {
1058                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1059                         status = exit_code & 0x7f;
1060                 }
1061                 return wait_noreap_copyout(p, pid, uid, why,
1062                                            status, infop, ru);
1063         }
1064
1065         /*
1066          * Try to move the task's state to DEAD
1067          * only one thread is allowed to do this:
1068          */
1069         state = xchg(&p->exit_state, EXIT_DEAD);
1070         if (state != EXIT_ZOMBIE) {
1071                 BUG_ON(state != EXIT_DEAD);
1072                 return 0;
1073         }
1074         BUG_ON(p->exit_signal == -1);
1075
1076         if (likely(p->signal)) {
1077                 struct signal_struct *psig;
1078                 struct signal_struct *sig;
1079
1080                 /*
1081                  * The resource counters for the group leader are in its
1082                  * own task_struct.  Those for dead threads in the group
1083                  * are in its signal_struct, as are those for the child
1084                  * processes it has previously reaped.  All these
1085                  * accumulate in the parent's signal_struct c* fields.
1086                  *
1087                  * We don't bother to take a lock here to protect these
1088                  * p->signal fields, because they are only touched by
1089                  * __exit_signal, which runs with tasklist_lock
1090                  * write-locked anyway, and so is excluded here.  We do
1091                  * need to protect the access to p->parent->signal fields,
1092                  * as other threads in the parent group can be right
1093                  * here reaping other children at the same time.
1094                  */
1095                 spin_lock_irq(&p->parent->sighand->siglock);
1096                 psig = p->parent->signal;
1097                 sig = p->signal;
1098                 psig->cutime =
1099                         cputime_add(psig->cutime,
1100                         cputime_add(p->utime,
1101                         cputime_add(sig->utime,
1102                                     sig->cutime)));
1103                 psig->cstime =
1104                         cputime_add(psig->cstime,
1105                         cputime_add(p->stime,
1106                         cputime_add(sig->stime,
1107                                     sig->cstime)));
1108                 psig->cmin_flt +=
1109                         p->min_flt + sig->min_flt + sig->cmin_flt;
1110                 psig->cmaj_flt +=
1111                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1112                 psig->cnvcsw +=
1113                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1114                 psig->cnivcsw +=
1115                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1116                 spin_unlock_irq(&p->parent->sighand->siglock);
1117         }
1118
1119         /*
1120          * Now we are sure this task is interesting, and no other
1121          * thread can reap it because we set its state to EXIT_DEAD.
1122          */
1123         read_unlock(&tasklist_lock);
1124
1125         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1126         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1127                 ? p->signal->group_exit_code : p->exit_code;
1128         if (!retval && stat_addr)
1129                 retval = put_user(status, stat_addr);
1130         if (!retval && infop)
1131                 retval = put_user(SIGCHLD, &infop->si_signo);
1132         if (!retval && infop)
1133                 retval = put_user(0, &infop->si_errno);
1134         if (!retval && infop) {
1135                 int why;
1136
1137                 if ((status & 0x7f) == 0) {
1138                         why = CLD_EXITED;
1139                         status >>= 8;
1140                 } else {
1141                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1142                         status &= 0x7f;
1143                 }
1144                 retval = put_user((short)why, &infop->si_code);
1145                 if (!retval)
1146                         retval = put_user(status, &infop->si_status);
1147         }
1148         if (!retval && infop)
1149                 retval = put_user(p->pid, &infop->si_pid);
1150         if (!retval && infop)
1151                 retval = put_user(p->uid, &infop->si_uid);
1152         if (retval) {
1153                 // TODO: is this safe?
1154                 p->exit_state = EXIT_ZOMBIE;
1155                 return retval;
1156         }
1157         retval = p->pid;
1158         release_task(p);
1159
1160         BUG_ON(!retval);
1161         return retval;
1162 }
1163
1164 /*
1165  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1166  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1167  * the lock and this task is uninteresting.  If we return nonzero, we have
1168  * released the lock and the system call should return.
1169  */
1170 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1171                              int noreap, struct siginfo __user *infop,
1172                              int __user *stat_addr, struct rusage __user *ru)
1173 {
1174         int retval, exit_code;
1175
1176         if (!p->exit_code)
1177                 return 0;
1178         if (delayed_group_leader &&
1179             p->signal && p->signal->group_stop_count > 0)
1180                 /*
1181                  * A group stop is in progress and this is the group leader.
1182                  * We won't report until all threads have stopped.
1183                  */
1184                 return 0;
1185
1186         /*
1187          * Now we are pretty sure this task is interesting.
1188          * Make sure it doesn't get reaped out from under us while we
1189          * give up the lock and then examine it below.  We don't want to
1190          * keep holding onto the tasklist_lock while we call getrusage and
1191          * possibly take page faults for user memory.
1192          */
1193         get_task_struct(p);
1194         read_unlock(&tasklist_lock);
1195
1196         if (unlikely(noreap)) {
1197                 pid_t pid = p->pid;
1198                 uid_t uid = p->uid;
1199
1200                 exit_code = p->exit_code;
1201                 if (unlikely(!exit_code) ||
1202                     unlikely(p->state & TASK_TRACED))
1203                         goto bail_ref;
1204                 return wait_noreap_copyout(p, pid, uid, CLD_STOPPED,
1205                                            (exit_code << 8) | 0x7f,
1206                                            infop, ru);
1207         }
1208
1209         write_lock_irq(&tasklist_lock);
1210
1211         /*
1212          * This uses xchg to be atomic with the thread resuming and setting
1213          * it.  It must also be done with the write lock held to prevent a
1214          * race with the EXIT_ZOMBIE case.
1215          */
1216         exit_code = xchg(&p->exit_code, 0);
1217         if (unlikely(p->exit_state)) {
1218                 /*
1219                  * The task resumed and then died.  Let the next iteration
1220                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1221                  * already be zero here if it resumed and did _exit(0).
1222                  * The task itself is dead and won't touch exit_code again;
1223                  * other processors in this function are locked out.
1224                  */
1225                 p->exit_code = exit_code;
1226                 exit_code = 0;
1227         }
1228         if (unlikely(exit_code == 0)) {
1229                 /*
1230                  * Another thread in this function got to it first, or it
1231                  * resumed, or it resumed and then died.
1232                  */
1233                 write_unlock_irq(&tasklist_lock);
1234 bail_ref:
1235                 put_task_struct(p);
1236                 /*
1237                  * We are returning to the wait loop without having successfully
1238                  * removed the process and having released the lock. We cannot
1239                  * continue, since the "p" task pointer is potentially stale.
1240                  *
1241                  * Return -EAGAIN, and do_wait() will restart the loop from the
1242                  * beginning. Do _not_ re-acquire the lock.
1243                  */
1244                 return -EAGAIN;
1245         }
1246
1247         /* move to end of parent's list to avoid starvation */
1248         remove_parent(p);
1249         add_parent(p);
1250
1251         write_unlock_irq(&tasklist_lock);
1252
1253         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1254         if (!retval && stat_addr)
1255                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1256         if (!retval && infop)
1257                 retval = put_user(SIGCHLD, &infop->si_signo);
1258         if (!retval && infop)
1259                 retval = put_user(0, &infop->si_errno);
1260         if (!retval && infop)
1261                 retval = put_user((short)CLD_STOPPED, &infop->si_code);
1262         if (!retval && infop)
1263                 retval = put_user(exit_code, &infop->si_status);
1264         if (!retval && infop)
1265                 retval = put_user(p->pid, &infop->si_pid);
1266         if (!retval && infop)
1267                 retval = put_user(p->uid, &infop->si_uid);
1268         if (!retval)
1269                 retval = p->pid;
1270         put_task_struct(p);
1271
1272         BUG_ON(!retval);
1273         return retval;
1274 }
1275
1276 /*
1277  * Handle do_wait work for one task in a live, non-stopped state.
1278  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1279  * the lock and this task is uninteresting.  If we return nonzero, we have
1280  * released the lock and the system call should return.
1281  */
1282 static int wait_task_continued(struct task_struct *p, int noreap,
1283                                struct siginfo __user *infop,
1284                                int __user *stat_addr, struct rusage __user *ru)
1285 {
1286         int retval;
1287         pid_t pid;
1288         uid_t uid;
1289
1290         if (unlikely(!p->signal))
1291                 return 0;
1292
1293         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1294                 return 0;
1295
1296         spin_lock_irq(&p->sighand->siglock);
1297         /* Re-check with the lock held.  */
1298         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1299                 spin_unlock_irq(&p->sighand->siglock);
1300                 return 0;
1301         }
1302         if (!noreap)
1303                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1304         spin_unlock_irq(&p->sighand->siglock);
1305
1306         pid = p->pid;
1307         uid = p->uid;
1308         get_task_struct(p);
1309         read_unlock(&tasklist_lock);
1310
1311         if (!infop) {
1312                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1313                 put_task_struct(p);
1314                 if (!retval && stat_addr)
1315                         retval = put_user(0xffff, stat_addr);
1316                 if (!retval)
1317                         retval = p->pid;
1318         } else {
1319                 retval = wait_noreap_copyout(p, pid, uid,
1320                                              CLD_CONTINUED, SIGCONT,
1321                                              infop, ru);
1322                 BUG_ON(retval == 0);
1323         }
1324
1325         return retval;
1326 }
1327
1328
1329 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1330                     int __user *stat_addr, struct rusage __user *ru)
1331 {
1332         DECLARE_WAITQUEUE(wait, current);
1333         struct task_struct *tsk;
1334         int flag, retval;
1335
1336         add_wait_queue(&current->signal->wait_chldexit,&wait);
1337 repeat:
1338         /*
1339          * We will set this flag if we see any child that might later
1340          * match our criteria, even if we are not able to reap it yet.
1341          */
1342         flag = 0;
1343         current->state = TASK_INTERRUPTIBLE;
1344         read_lock(&tasklist_lock);
1345         tsk = current;
1346         do {
1347                 struct task_struct *p;
1348                 struct list_head *_p;
1349                 int ret;
1350
1351                 list_for_each(_p,&tsk->children) {
1352                         p = list_entry(_p, struct task_struct, sibling);
1353
1354                         ret = eligible_child(pid, options, p);
1355                         if (!ret)
1356                                 continue;
1357
1358                         switch (p->state) {
1359                         case TASK_TRACED:
1360                                 flag = 1;
1361                                 continue;
1362                         case TASK_STOPPED:
1363                                 /*
1364                                  * It's stopped now, so it might later
1365                                  * continue, exit, or stop again.
1366                                  */
1367                                 flag = 1;
1368                                 if (!(options & WUNTRACED))
1369                                         continue;
1370                                 if (tracehook_inhibit_wait_stopped(p))
1371                                         continue;
1372                                 retval = wait_task_stopped(p, ret == 2,
1373                                                            (options & WNOWAIT),
1374                                                            infop,
1375                                                            stat_addr, ru);
1376                                 if (retval == -EAGAIN)
1377                                         goto repeat;
1378                                 if (retval != 0) /* He released the lock.  */
1379                                         goto end;
1380                                 break;
1381                         default:
1382                         // case EXIT_DEAD:
1383                                 if (p->exit_state == EXIT_DEAD)
1384                                         continue;
1385                         // case EXIT_ZOMBIE:
1386                                 if (p->exit_state == EXIT_ZOMBIE) {
1387                                         /*
1388                                          * Eligible but we cannot release
1389                                          * it yet:
1390                                          */
1391                                         if (ret == 2)
1392                                                 goto check_continued;
1393                                         if (!likely(options & WEXITED))
1394                                                 continue;
1395                                         if (tracehook_inhibit_wait_zombie(p)) {
1396                                                 flag = 1;
1397                                                 continue;
1398                                         }
1399                                         retval = wait_task_zombie(
1400                                                 p, (options & WNOWAIT),
1401                                                 infop, stat_addr, ru);
1402                                         /* He released the lock.  */
1403                                         if (retval != 0)
1404                                                 goto end;
1405                                         break;
1406                                 }
1407 check_continued:
1408                                 /*
1409                                  * It's running now, so it might later
1410                                  * exit, stop, or stop and then continue.
1411                                  */
1412                                 flag = 1;
1413                                 if (!unlikely(options & WCONTINUED))
1414                                         continue;
1415                                 if (tracehook_inhibit_wait_continued(p))
1416                                         continue;
1417                                 retval = wait_task_continued(
1418                                         p, (options & WNOWAIT),
1419                                         infop, stat_addr, ru);
1420                                 if (retval != 0) /* He released the lock.  */
1421                                         goto end;
1422                                 break;
1423                         }
1424                 }
1425
1426                 retval = ptrace_do_wait(tsk, pid, options,
1427                                         infop, stat_addr, ru);
1428                 if (retval != -ECHILD) {
1429                         flag = 1;
1430                         if (retval != 0) /* He released the lock.  */
1431                                 goto end;
1432                 }
1433
1434                 if (options & __WNOTHREAD)
1435                         break;
1436                 tsk = next_thread(tsk);
1437                 BUG_ON(tsk->signal != current->signal);
1438         } while (tsk != current);
1439
1440         read_unlock(&tasklist_lock);
1441         if (flag) {
1442                 retval = 0;
1443                 if (options & WNOHANG)
1444                         goto end;
1445                 retval = -ERESTARTSYS;
1446                 if (signal_pending(current))
1447                         goto end;
1448                 schedule();
1449                 goto repeat;
1450         }
1451         retval = -ECHILD;
1452 end:
1453         current->state = TASK_RUNNING;
1454         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1455         if (infop) {
1456                 if (retval > 0)
1457                         retval = 0;
1458                 else {
1459                         /*
1460                          * For a WNOHANG return, clear out all the fields
1461                          * we would set so the user can easily tell the
1462                          * difference.
1463                          */
1464                         if (!retval)
1465                                 retval = put_user(0, &infop->si_signo);
1466                         if (!retval)
1467                                 retval = put_user(0, &infop->si_errno);
1468                         if (!retval)
1469                                 retval = put_user(0, &infop->si_code);
1470                         if (!retval)
1471                                 retval = put_user(0, &infop->si_pid);
1472                         if (!retval)
1473                                 retval = put_user(0, &infop->si_uid);
1474                         if (!retval)
1475                                 retval = put_user(0, &infop->si_status);
1476                 }
1477         }
1478         return retval;
1479 }
1480
1481 asmlinkage long sys_waitid(int which, pid_t pid,
1482                            struct siginfo __user *infop, int options,
1483                            struct rusage __user *ru)
1484 {
1485         long ret;
1486
1487         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1488                 return -EINVAL;
1489         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1490                 return -EINVAL;
1491
1492         switch (which) {
1493         case P_ALL:
1494                 pid = -1;
1495                 break;
1496         case P_PID:
1497                 if (pid <= 0)
1498                         return -EINVAL;
1499                 break;
1500         case P_PGID:
1501                 if (pid <= 0)
1502                         return -EINVAL;
1503                 pid = -pid;
1504                 break;
1505         default:
1506                 return -EINVAL;
1507         }
1508
1509         ret = do_wait(pid, options, infop, NULL, ru);
1510
1511         /* avoid REGPARM breakage on x86: */
1512         prevent_tail_call(ret);
1513         return ret;
1514 }
1515
1516 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1517                           int options, struct rusage __user *ru)
1518 {
1519         long ret;
1520
1521         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1522                         __WNOTHREAD|__WCLONE|__WALL))
1523                 return -EINVAL;
1524         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1525
1526         /* avoid REGPARM breakage on x86: */
1527         prevent_tail_call(ret);
1528         return ret;
1529 }
1530
1531 #ifdef __ARCH_WANT_SYS_WAITPID
1532
1533 /*
1534  * sys_waitpid() remains for compatibility. waitpid() should be
1535  * implemented by calling sys_wait4() from libc.a.
1536  */
1537 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1538 {
1539         return sys_wait4(pid, stat_addr, options, NULL);
1540 }
1541
1542 #endif