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         proc_clear_tty(current);
 389
 390         /* Block and flush all signals */
 391         sigfillset(&blocked);
 392         sigprocmask(SIG_BLOCK, &blocked, NULL);
 393         flush_signals(current);
 394
 395         /* Become as one with the init task */
 396
 397         exit_fs(current);       /* current->fs->count--; */
 398         fs = init_task.fs;
 399         current->fs = fs;
 400         atomic_inc(&fs->count);
 401         exit_namespace(current);
 402         current->namespace = init_task.namespace;
 403         get_namespace(current->namespace);
 404         exit_files(current);
 405         current->files = init_task.files;
 406         atomic_inc(&current->files->count);
 407
 408         reparent_to_init();
 409 }
 410
 411 EXPORT_SYMBOL(daemonize);
 412
 413 static void close_files(struct files_struct * files)
 414 {
 415         int i, j;
 416         struct fdtable *fdt;
 417
 418         j = 0;
 419
 420         /*
 421          * It is safe to dereference the fd table without RCU or
 422          * ->file_lock because this is the last reference to the
 423          * files structure.
 424          */
 425         fdt = files_fdtable(files);
 426         for (;;) {
 427                 unsigned long set;
 428                 i = j * __NFDBITS;
 429                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 430                         break;
 431                 set = fdt->open_fds->fds_bits[j++];
 432                 while (set) {
 433                         if (set & 1) {
 434                                 struct file * file = xchg(&fdt->fd[i], NULL);
 435                                 if (file)
 436                                         filp_close(file, files);
 437                                 vx_openfd_dec(i);
 438                         }
 439                         i++;
 440                         set >>= 1;
 441                         cond_resched();
 442                 }
 443         }
 444 }
 445
 446 struct files_struct *get_files_struct(struct task_struct *task)
 447 {
 448         struct files_struct *files;
 449
 450         task_lock(task);
 451         files = task->files;
 452         if (files)
 453                 atomic_inc(&files->count);
 454         task_unlock(task);
 455
 456         return files;
 457 }
 458
 459 void fastcall put_files_struct(struct files_struct *files)
 460 {
 461         struct fdtable *fdt;
 462
 463         if (atomic_dec_and_test(&files->count)) {
 464                 close_files(files);
 465                 /*
 466                  * Free the fd and fdset arrays if we expanded them.
 467                  * If the fdtable was embedded, pass files for freeing
 468                  * at the end of the RCU grace period. Otherwise,
 469                  * you can free files immediately.
 470                  */
 471                 fdt = files_fdtable(files);
 472                 if (fdt == &files->fdtab)
 473                         fdt->free_files = files;
 474                 else
 475                         kmem_cache_free(files_cachep, files);
 476                 free_fdtable(fdt);
 477         }
 478 }
 479
 480 EXPORT_SYMBOL(put_files_struct);
 481
 482 static inline void __exit_files(struct task_struct *tsk)
 483 {
 484         struct files_struct * files = tsk->files;
 485
 486         if (files) {
 487                 task_lock(tsk);
 488                 tsk->files = NULL;
 489                 task_unlock(tsk);
 490                 put_files_struct(files);
 491         }
 492 }
 493
 494 void exit_files(struct task_struct *tsk)
 495 {
 496         __exit_files(tsk);
 497 }
 498
 499 static inline void __put_fs_struct(struct fs_struct *fs)
 500 {
 501         /* No need to hold fs->lock if we are killing it */
 502         if (atomic_dec_and_test(&fs->count)) {
 503                 dput(fs->root);
 504                 mntput(fs->rootmnt);
 505                 dput(fs->pwd);
 506                 mntput(fs->pwdmnt);
 507                 if (fs->altroot) {
 508                         dput(fs->altroot);
 509                         mntput(fs->altrootmnt);
 510                 }
 511                 kmem_cache_free(fs_cachep, fs);
 512         }
 513 }
 514
 515 void put_fs_struct(struct fs_struct *fs)
 516 {
 517         __put_fs_struct(fs);
 518 }
 519
 520 static inline void __exit_fs(struct task_struct *tsk)
 521 {
 522         struct fs_struct * fs = tsk->fs;
 523
 524         if (fs) {
 525                 task_lock(tsk);
 526                 tsk->fs = NULL;
 527                 task_unlock(tsk);
 528                 __put_fs_struct(fs);
 529         }
 530 }
 531
 532 void exit_fs(struct task_struct *tsk)
 533 {
 534         __exit_fs(tsk);
 535 }
 536
 537 EXPORT_SYMBOL_GPL(exit_fs);
 538
 539 /*
 540  * Turn us into a lazy TLB process if we
 541  * aren't already..
 542  */
 543 static void exit_mm(struct task_struct * tsk)
 544 {
 545         struct mm_struct *mm = tsk->mm;
 546
 547         mm_release(tsk, mm);
 548         if (!mm)
 549                 return;
 550         /*
 551          * Serialize with any possible pending coredump.
 552          * We must hold mmap_sem around checking core_waiters
 553          * and clearing tsk->mm.  The core-inducing thread
 554          * will increment core_waiters for each thread in the
 555          * group with ->mm != NULL.
 556          */
 557         down_read(&mm->mmap_sem);
 558         if (mm->core_waiters) {
 559                 up_read(&mm->mmap_sem);
 560                 down_write(&mm->mmap_sem);
 561                 if (!--mm->core_waiters)
 562                         complete(mm->core_startup_done);
 563                 up_write(&mm->mmap_sem);
 564
 565                 wait_for_completion(&mm->core_done);
 566                 down_read(&mm->mmap_sem);
 567         }
 568         atomic_inc(&mm->mm_count);
 569         BUG_ON(mm != tsk->active_mm);
 570         /* more a memory barrier than a real lock */
 571         task_lock(tsk);
 572         tsk->mm = NULL;
 573         up_read(&mm->mmap_sem);
 574         enter_lazy_tlb(mm, current);
 575         task_unlock(tsk);
 576         mmput(mm);
 577 }
 578
 579 static inline void
 580 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 581 {
 582         /* check for reaper context */
 583         vxwprintk((p->xid != reaper->xid) && (reaper != child_reaper),
 584                 "rogue reaper: %p[%d,#%u] <> %p[%d,#%u]",
 585                 p, p->pid, p->xid, reaper, reaper->pid, reaper->xid);
 586
 587         /*
 588          * Make sure we're not reparenting to ourselves and that
 589          * the parent is not a zombie.
 590          */
 591         BUG_ON(p == reaper || reaper->exit_state);
 592         p->parent = reaper;
 593 }
 594
 595 static void
 596 reparent_thread(struct task_struct *p, struct task_struct *father)
 597 {
 598         /* We don't want people slaying init.  */
 599         if (p->exit_signal != -1)
 600                 p->exit_signal = SIGCHLD;
 601
 602         if (p->pdeath_signal)
 603                 /* We already hold the tasklist_lock here.  */
 604                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 605
 606         /* Move the child from its dying parent to the new one.  */
 607         list_move_tail(&p->sibling, &p->parent->children);
 608
 609         /* If we'd notified the old parent about this child's death,
 610          * also notify the new parent.
 611          */
 612         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 613             thread_group_empty(p))
 614                 do_notify_parent(p, p->exit_signal);
 615
 616         /*
 617          * process group orphan check
 618          * Case ii: Our child is in a different pgrp
 619          * than we are, and it was the only connection
 620          * outside, so the child pgrp is now orphaned.
 621          */
 622         if ((process_group(p) != process_group(father)) &&
 623             (p->signal->session == father->signal->session)) {
 624                 int pgrp = process_group(p);
 625
 626                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 627                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 628                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 629                 }
 630         }
 631 }
 632
 633 /*
 634  * When we die, we re-parent all our children.
 635  * Try to give them to another thread in our thread
 636  * group, and if no such member exists, give it to
 637  * the global child reaper process (ie "init")
 638  */
 639 static void
 640 forget_original_parent(struct task_struct *father)
 641 {
 642         struct task_struct *p, *reaper = father;
 643         struct list_head *_p, *_n;
 644
 645         do {
 646                 reaper = next_thread(reaper);
 647                 if (reaper == father) {
 648                         reaper = vx_child_reaper(father);
 649                         break;
 650                 }
 651         } while (reaper->exit_state);
 652
 653         list_for_each_safe(_p, _n, &father->children) {
 654                 p = list_entry(_p, struct task_struct, sibling);
 655                 choose_new_parent(p, vx_child_reaper(p));
 656                 reparent_thread(p, father);
 657         }
 658 }
 659
 660 /*
 661  * Send signals to all our closest relatives so that they know
 662  * to properly mourn us..
 663  */
 664 static void exit_notify(struct task_struct *tsk)
 665 {
 666         int state;
 667         struct task_struct *t;
 668         int noreap;
 669         void *cookie;
 670
 671         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 672             && !thread_group_empty(tsk)) {
 673                 /*
 674                  * This occurs when there was a race between our exit
 675                  * syscall and a group signal choosing us as the one to
 676                  * wake up.  It could be that we are the only thread
 677                  * alerted to check for pending signals, but another thread
 678                  * should be woken now to take the signal since we will not.
 679                  * Now we'll wake all the threads in the group just to make
 680                  * sure someone gets all the pending signals.
 681                  */
 682                 read_lock(&tasklist_lock);
 683                 spin_lock_irq(&tsk->sighand->siglock);
 684                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 685                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 686                                 recalc_sigpending_tsk(t);
 687                                 if (signal_pending(t))
 688                                         signal_wake_up(t, 0);
 689                         }
 690                 spin_unlock_irq(&tsk->sighand->siglock);
 691                 read_unlock(&tasklist_lock);
 692         }
 693
 694         write_lock_irq(&tasklist_lock);
 695
 696         /*
 697          * This does two things:
 698          *
 699          * A.  Make init inherit all the child processes
 700          * B.  Check to see if any process groups have become orphaned
 701          *      as a result of our exiting, and if they have any stopped
 702          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 703          */
 704
 705         forget_original_parent(tsk);
 706         BUG_ON(!list_empty(&tsk->children));
 707
 708         /*
 709          * Check to see if any process groups have become orphaned
 710          * as a result of our exiting, and if they have any stopped
 711          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 712          *
 713          * Case i: Our father is in a different pgrp than we are
 714          * and we were the only connection outside, so our pgrp
 715          * is about to become orphaned.
 716          */
 717
 718         t = tsk->parent;
 719
 720         if ((process_group(t) != process_group(tsk)) &&
 721             (t->signal->session == tsk->signal->session) &&
 722             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 723             has_stopped_jobs(process_group(tsk))) {
 724                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 725                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 726         }
 727
 728         /* Let father know we died
 729          *
 730          * Thread signals are configurable, but you aren't going to use
 731          * that to send signals to arbitary processes.
 732          * That stops right now.
 733          *
 734          * If the parent exec id doesn't match the exec id we saved
 735          * when we started then we know the parent has changed security
 736          * domain.
 737          *
 738          * If our self_exec id doesn't match our parent_exec_id then
 739          * we have changed execution domain as these two values started
 740          * the same after a fork.
 741          *
 742          */
 743
 744         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 745             ( tsk->parent_exec_id != t->self_exec_id  ||
 746               tsk->self_exec_id != tsk->parent_exec_id)
 747             && !capable(CAP_KILL))
 748                 tsk->exit_signal = SIGCHLD;
 749
 750         if (!tracehook_notify_death(tsk, &noreap, &cookie)
 751             && tsk->exit_signal != -1 && thread_group_empty(tsk))
 752                 do_notify_parent(tsk, tsk->exit_signal);
 753
 754         state = EXIT_ZOMBIE;
 755         if (tsk->exit_signal == -1 && !noreap)
 756                 state = EXIT_DEAD;
 757         tsk->exit_state = state;
 758
 759         write_unlock_irq(&tasklist_lock);
 760
 761         tracehook_report_death(tsk, state, cookie);
 762
 763         /* If the process is dead, release it - nobody will wait for it */
 764         if (state == EXIT_DEAD)
 765                 release_task(tsk);
 766 }
 767
 768 fastcall NORET_TYPE void do_exit(long code)
 769 {
 770         struct task_struct *tsk = current;
 771         struct taskstats *tidstats;
 772         int group_dead;
 773         unsigned int mycpu;
 774
 775         profile_task_exit(tsk);
 776
 777         WARN_ON(atomic_read(&tsk->fs_excl));
 778
 779         if (unlikely(in_interrupt()))
 780                 panic("Aiee, killing interrupt handler!");
 781         if (unlikely(!tsk->pid))
 782                 panic("Attempted to kill the idle task!");
 783         if (unlikely(tsk == child_reaper))
 784                 panic("Attempted to kill init!");
 785
 786         tracehook_report_exit(&code);
 787
 788         /*
 789          * We're taking recursive faults here in do_exit. Safest is to just
 790          * leave this task alone and wait for reboot.
 791          */
 792         if (unlikely(tsk->flags & PF_EXITING)) {
 793                 printk(KERN_ALERT
 794                         "Fixing recursive fault but reboot is needed!\n");
 795                 if (tsk->io_context)
 796                         exit_io_context();
 797                 set_current_state(TASK_UNINTERRUPTIBLE);
 798                 schedule();
 799         }
 800
 801         tsk->flags |= PF_EXITING;
 802
 803         ptrace_exit(tsk);
 804
 805         if (unlikely(in_atomic()))
 806                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 807                                 current->comm, current->pid,
 808                                 preempt_count());
 809
 810         taskstats_exit_alloc(&tidstats, &mycpu);
 811
 812         acct_update_integrals(tsk);
 813         if (tsk->mm) {
 814                 update_hiwater_rss(tsk->mm);
 815                 update_hiwater_vm(tsk->mm);
 816         }
 817         group_dead = atomic_dec_and_test(&tsk->signal->live);
 818         if (group_dead) {
 819                 hrtimer_cancel(&tsk->signal->real_timer);
 820                 exit_itimers(tsk->signal);
 821         }
 822
 823         if (current->tux_info) {
 824 #ifdef CONFIG_TUX_DEBUG
 825                 printk("Possibly unexpected TUX-thread exit(%ld) at %p?\n",
 826                         code, __builtin_return_address(0));
 827 #endif
 828                 current->tux_exit();
 829         }
 830
 831         acct_collect(code, group_dead);
 832         if (unlikely(tsk->robust_list))
 833                 exit_robust_list(tsk);
 834 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 835         if (unlikely(tsk->compat_robust_list))
 836                 compat_exit_robust_list(tsk);
 837 #endif
 838         if (unlikely(tsk->audit_context))
 839                 audit_free(tsk);
 840         taskstats_exit_send(tsk, tidstats, group_dead, mycpu);
 841         taskstats_exit_free(tidstats);
 842
 843         exit_mm(tsk);
 844
 845         if (group_dead)
 846                 acct_process();
 847         exit_sem(tsk);
 848         __exit_files(tsk);
 849         __exit_fs(tsk);
 850         exit_namespace(tsk);
 851         exit_thread();
 852         cpuset_exit(tsk);
 853         exit_keys(tsk);
 854
 855         if (group_dead && tsk->signal->leader)
 856                 disassociate_ctty(1);
 857
 858         module_put(task_thread_info(tsk)->exec_domain->module);
 859         if (tsk->binfmt)
 860                 module_put(tsk->binfmt->module);
 861
 862         tsk->exit_code = code;
 863         proc_exit_connector(tsk);
 864         /* needs to stay before exit_notify() */
 865         exit_vx_info_early(tsk, code);
 866         exit_notify(tsk);
 867 #ifdef CONFIG_NUMA
 868         mpol_free(tsk->mempolicy);
 869         tsk->mempolicy = NULL;
 870 #endif
 871         /*
 872          * This must happen late, after the PID is not
 873          * hashed anymore:
 874          */
 875         if (unlikely(!list_empty(&tsk->pi_state_list)))
 876                 exit_pi_state_list(tsk);
 877         if (unlikely(current->pi_state_cache))
 878                 kfree(current->pi_state_cache);
 879         /*
 880          * Make sure we are holding no locks:
 881          */
 882         debug_check_no_locks_held(tsk);
 883
 884         if (tsk->io_context)
 885                 exit_io_context();
 886
 887         if (tsk->splice_pipe)
 888                 __free_pipe_info(tsk->splice_pipe);
 889
 890         /* needs to stay after exit_notify() */
 891         exit_vx_info(tsk, code);
 892         exit_nx_info(tsk);
 893
 894         /* PF_DEAD causes final put_task_struct after we schedule. */
 895         preempt_disable();
 896         BUG_ON(tsk->flags & PF_DEAD);
 897         tsk->flags |= PF_DEAD;
 898
 899         schedule();
 900         BUG();
 901         /* Avoid "noreturn function does return".  */
 902         for (;;) ;
 903 }
 904
 905 EXPORT_SYMBOL_GPL(do_exit);
 906
 907 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 908 {
 909         if (comp)
 910                 complete(comp);
 911
 912         do_exit(code);
 913 }
 914
 915 EXPORT_SYMBOL(complete_and_exit);
 916
 917 asmlinkage long sys_exit(int error_code)
 918 {
 919         do_exit((error_code&0xff)<<8);
 920 }
 921
 922 /*
 923  * Take down every thread in the group.  This is called by fatal signals
 924  * as well as by sys_exit_group (below).
 925  */
 926 NORET_TYPE void
 927 do_group_exit(int exit_code)
 928 {
 929         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 930
 931         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 932                 exit_code = current->signal->group_exit_code;
 933         else if (!thread_group_empty(current)) {
 934                 struct signal_struct *const sig = current->signal;
 935                 struct sighand_struct *const sighand = current->sighand;
 936                 spin_lock_irq(&sighand->siglock);
 937                 if (sig->flags & SIGNAL_GROUP_EXIT)
 938                         /* Another thread got here before we took the lock.  */
 939                         exit_code = sig->group_exit_code;
 940                 else {
 941                         sig->group_exit_code = exit_code;
 942                         zap_other_threads(current);
 943                 }
 944                 spin_unlock_irq(&sighand->siglock);
 945         }
 946
 947         do_exit(exit_code);
 948         /* NOTREACHED */
 949 }
 950
 951 /*
 952  * this kills every thread in the thread group. Note that any externally
 953  * wait4()-ing process will get the correct exit code - even if this
 954  * thread is not the thread group leader.
 955  */
 956 asmlinkage void sys_exit_group(int error_code)
 957 {
 958         do_group_exit((error_code & 0xff) << 8);
 959 }
 960
 961 static int eligible_child(pid_t pid, int options, struct task_struct *p)
 962 {
 963         if (pid > 0) {
 964                 if (p->pid != pid)
 965                         return 0;
 966         } else if (!pid) {
 967                 if (process_group(p) != process_group(current))
 968                         return 0;
 969         } else if (pid != -1) {
 970                 if (process_group(p) != -pid)
 971                         return 0;
 972         }
 973
 974         /*
 975          * Do not consider detached threads.
 976          */
 977         if (p->exit_signal == -1)
 978                 return 0;
 979
 980         /* Wait for all children (clone and not) if __WALL is set;
 981          * otherwise, wait for clone children *only* if __WCLONE is
 982          * set; otherwise, wait for non-clone children *only*.  (Note:
 983          * A "clone" child here is one that reports to its parent
 984          * using a signal other than SIGCHLD.) */
 985         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 986             && !(options & __WALL))
 987                 return 0;
 988         /*
 989          * Do not consider thread group leaders that are
 990          * in a non-empty thread group:
 991          */
 992         if (delay_group_leader(p))
 993                 return 2;
 994
 995         if (security_task_wait(p))
 996                 return 0;
 997
 998         return 1;
 999 }
1000
1001 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1002                                int why, int status,
1003                                struct siginfo __user *infop,
1004                                struct rusage __user *rusagep)
1005 {
1006         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1007
1008         put_task_struct(p);
1009         if (!retval)
1010                 retval = put_user(SIGCHLD, &infop->si_signo);
1011         if (!retval)
1012                 retval = put_user(0, &infop->si_errno);
1013         if (!retval)
1014                 retval = put_user((short)why, &infop->si_code);
1015         if (!retval)
1016                 retval = put_user(pid, &infop->si_pid);
1017         if (!retval)
1018                 retval = put_user(uid, &infop->si_uid);
1019         if (!retval)
1020                 retval = put_user(status, &infop->si_status);
1021         if (!retval)
1022                 retval = pid;
1023         return retval;
1024 }
1025
1026 /*
1027  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1028  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1029  * the lock and this task is uninteresting.  If we return nonzero, we have
1030  * released the lock and the system call should return.
1031  */
1032 static int wait_task_zombie(struct task_struct *p, int noreap,
1033                             struct siginfo __user *infop,
1034                             int __user *stat_addr, struct rusage __user *ru)
1035 {
1036         unsigned long state;
1037         int retval;
1038         int status;
1039
1040         if (unlikely(noreap)) {
1041                 pid_t pid = p->pid;
1042                 uid_t uid = p->uid;
1043                 int exit_code = p->exit_code;
1044                 int why, status;
1045
1046                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1047                         return 0;
1048                 if (unlikely(p->exit_signal == -1))
1049                         return 0;
1050                 get_task_struct(p);
1051                 read_unlock(&tasklist_lock);
1052                 if ((exit_code & 0x7f) == 0) {
1053                         why = CLD_EXITED;
1054                         status = exit_code >> 8;
1055                 } else {
1056                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1057                         status = exit_code & 0x7f;
1058                 }
1059                 return wait_noreap_copyout(p, pid, uid, why,
1060                                            status, infop, ru);
1061         }
1062
1063         /*
1064          * Try to move the task's state to DEAD
1065          * only one thread is allowed to do this:
1066          */
1067         state = xchg(&p->exit_state, EXIT_DEAD);
1068         if (state != EXIT_ZOMBIE) {
1069                 BUG_ON(state != EXIT_DEAD);
1070                 return 0;
1071         }
1072         BUG_ON(p->exit_signal == -1);
1073
1074         if (likely(p->signal)) {
1075                 struct signal_struct *psig;
1076                 struct signal_struct *sig;
1077
1078                 /*
1079                  * The resource counters for the group leader are in its
1080                  * own task_struct.  Those for dead threads in the group
1081                  * are in its signal_struct, as are those for the child
1082                  * processes it has previously reaped.  All these
1083                  * accumulate in the parent's signal_struct c* fields.
1084                  *
1085                  * We don't bother to take a lock here to protect these
1086                  * p->signal fields, because they are only touched by
1087                  * __exit_signal, which runs with tasklist_lock
1088                  * write-locked anyway, and so is excluded here.  We do
1089                  * need to protect the access to p->parent->signal fields,
1090                  * as other threads in the parent group can be right
1091                  * here reaping other children at the same time.
1092                  */
1093                 spin_lock_irq(&p->parent->sighand->siglock);
1094                 psig = p->parent->signal;
1095                 sig = p->signal;
1096                 psig->cutime =
1097                         cputime_add(psig->cutime,
1098                         cputime_add(p->utime,
1099                         cputime_add(sig->utime,
1100                                     sig->cutime)));
1101                 psig->cstime =
1102                         cputime_add(psig->cstime,
1103                         cputime_add(p->stime,
1104                         cputime_add(sig->stime,
1105                                     sig->cstime)));
1106                 psig->cmin_flt +=
1107                         p->min_flt + sig->min_flt + sig->cmin_flt;
1108                 psig->cmaj_flt +=
1109                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1110                 psig->cnvcsw +=
1111                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1112                 psig->cnivcsw +=
1113                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1114                 spin_unlock_irq(&p->parent->sighand->siglock);
1115         }
1116
1117         /*
1118          * Now we are sure this task is interesting, and no other
1119          * thread can reap it because we set its state to EXIT_DEAD.
1120          */
1121         read_unlock(&tasklist_lock);
1122
1123         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1124         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1125                 ? p->signal->group_exit_code : p->exit_code;
1126         if (!retval && stat_addr)
1127                 retval = put_user(status, stat_addr);
1128         if (!retval && infop)
1129                 retval = put_user(SIGCHLD, &infop->si_signo);
1130         if (!retval && infop)
1131                 retval = put_user(0, &infop->si_errno);
1132         if (!retval && infop) {
1133                 int why;
1134
1135                 if ((status & 0x7f) == 0) {
1136                         why = CLD_EXITED;
1137                         status >>= 8;
1138                 } else {
1139                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1140                         status &= 0x7f;
1141                 }
1142                 retval = put_user((short)why, &infop->si_code);
1143                 if (!retval)
1144                         retval = put_user(status, &infop->si_status);
1145         }
1146         if (!retval && infop)
1147                 retval = put_user(p->pid, &infop->si_pid);
1148         if (!retval && infop)
1149                 retval = put_user(p->uid, &infop->si_uid);
1150         if (retval) {
1151                 // TODO: is this safe?
1152                 p->exit_state = EXIT_ZOMBIE;
1153                 return retval;
1154         }
1155         retval = p->pid;
1156         release_task(p);
1157
1158         BUG_ON(!retval);
1159         return retval;
1160 }
1161
1162 /*
1163  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1164  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1165  * the lock and this task is uninteresting.  If we return nonzero, we have
1166  * released the lock and the system call should return.
1167  */
1168 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1169                              int noreap, struct siginfo __user *infop,
1170                              int __user *stat_addr, struct rusage __user *ru)
1171 {
1172         int retval, exit_code;
1173
1174         if (!p->exit_code)
1175                 return 0;
1176         if (delayed_group_leader &&
1177             p->signal && p->signal->group_stop_count > 0)
1178                 /*
1179                  * A group stop is in progress and this is the group leader.
1180                  * We won't report until all threads have stopped.
1181                  */
1182                 return 0;
1183
1184         /*
1185          * Now we are pretty sure this task is interesting.
1186          * Make sure it doesn't get reaped out from under us while we
1187          * give up the lock and then examine it below.  We don't want to
1188          * keep holding onto the tasklist_lock while we call getrusage and
1189          * possibly take page faults for user memory.
1190          */
1191         get_task_struct(p);
1192         read_unlock(&tasklist_lock);
1193
1194         if (unlikely(noreap)) {
1195                 pid_t pid = p->pid;
1196                 uid_t uid = p->uid;
1197
1198                 exit_code = p->exit_code;
1199                 if (unlikely(!exit_code) ||
1200                     unlikely(p->state & TASK_TRACED))
1201                         goto bail_ref;
1202                 return wait_noreap_copyout(p, pid, uid, CLD_STOPPED,
1203                                            (exit_code << 8) | 0x7f,
1204                                            infop, ru);
1205         }
1206
1207         write_lock_irq(&tasklist_lock);
1208
1209         /*
1210          * This uses xchg to be atomic with the thread resuming and setting
1211          * it.  It must also be done with the write lock held to prevent a
1212          * race with the EXIT_ZOMBIE case.
1213          */
1214         exit_code = xchg(&p->exit_code, 0);
1215         if (unlikely(p->exit_state)) {
1216                 /*
1217                  * The task resumed and then died.  Let the next iteration
1218                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1219                  * already be zero here if it resumed and did _exit(0).
1220                  * The task itself is dead and won't touch exit_code again;
1221                  * other processors in this function are locked out.
1222                  */
1223                 p->exit_code = exit_code;
1224                 exit_code = 0;
1225         }
1226         if (unlikely(exit_code == 0)) {
1227                 /*
1228                  * Another thread in this function got to it first, or it
1229                  * resumed, or it resumed and then died.
1230                  */
1231                 write_unlock_irq(&tasklist_lock);
1232 bail_ref:
1233                 put_task_struct(p);
1234                 /*
1235                  * We are returning to the wait loop without having successfully
1236                  * removed the process and having released the lock. We cannot
1237                  * continue, since the "p" task pointer is potentially stale.
1238                  *
1239                  * Return -EAGAIN, and do_wait() will restart the loop from the
1240                  * beginning. Do _not_ re-acquire the lock.
1241                  */
1242                 return -EAGAIN;
1243         }
1244
1245         /* move to end of parent's list to avoid starvation */
1246         remove_parent(p);
1247         add_parent(p);
1248
1249         write_unlock_irq(&tasklist_lock);
1250
1251         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1252         if (!retval && stat_addr)
1253                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1254         if (!retval && infop)
1255                 retval = put_user(SIGCHLD, &infop->si_signo);
1256         if (!retval && infop)
1257                 retval = put_user(0, &infop->si_errno);
1258         if (!retval && infop)
1259                 retval = put_user((short)CLD_STOPPED, &infop->si_code);
1260         if (!retval && infop)
1261                 retval = put_user(exit_code, &infop->si_status);
1262         if (!retval && infop)
1263                 retval = put_user(p->pid, &infop->si_pid);
1264         if (!retval && infop)
1265                 retval = put_user(p->uid, &infop->si_uid);
1266         if (!retval)
1267                 retval = p->pid;
1268         put_task_struct(p);
1269
1270         BUG_ON(!retval);
1271         return retval;
1272 }
1273
1274 /*
1275  * Handle do_wait work for one task in a live, non-stopped state.
1276  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1277  * the lock and this task is uninteresting.  If we return nonzero, we have
1278  * released the lock and the system call should return.
1279  */
1280 static int wait_task_continued(struct task_struct *p, int noreap,
1281                                struct siginfo __user *infop,
1282                                int __user *stat_addr, struct rusage __user *ru)
1283 {
1284         int retval;
1285         pid_t pid;
1286         uid_t uid;
1287
1288         if (unlikely(!p->signal))
1289                 return 0;
1290
1291         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1292                 return 0;
1293
1294         spin_lock_irq(&p->sighand->siglock);
1295         /* Re-check with the lock held.  */
1296         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1297                 spin_unlock_irq(&p->sighand->siglock);
1298                 return 0;
1299         }
1300         if (!noreap)
1301                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1302         spin_unlock_irq(&p->sighand->siglock);
1303
1304         pid = p->pid;
1305         uid = p->uid;
1306         get_task_struct(p);
1307         read_unlock(&tasklist_lock);
1308
1309         if (!infop) {
1310                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1311                 put_task_struct(p);
1312                 if (!retval && stat_addr)
1313                         retval = put_user(0xffff, stat_addr);
1314                 if (!retval)
1315                         retval = p->pid;
1316         } else {
1317                 retval = wait_noreap_copyout(p, pid, uid,
1318                                              CLD_CONTINUED, SIGCONT,
1319                                              infop, ru);
1320                 BUG_ON(retval == 0);
1321         }
1322
1323         return retval;
1324 }
1325
1326
1327 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1328                     int __user *stat_addr, struct rusage __user *ru)
1329 {
1330         DECLARE_WAITQUEUE(wait, current);
1331         struct task_struct *tsk;
1332         int flag, retval;
1333
1334         add_wait_queue(&current->signal->wait_chldexit,&wait);
1335 repeat:
1336         /*
1337          * We will set this flag if we see any child that might later
1338          * match our criteria, even if we are not able to reap it yet.
1339          */
1340         flag = 0;
1341         current->state = TASK_INTERRUPTIBLE;
1342         read_lock(&tasklist_lock);
1343         tsk = current;
1344         do {
1345                 struct task_struct *p;
1346                 struct list_head *_p;
1347                 int ret;
1348
1349                 list_for_each(_p,&tsk->children) {
1350                         p = list_entry(_p, struct task_struct, sibling);
1351
1352                         ret = eligible_child(pid, options, p);
1353                         if (!ret)
1354                                 continue;
1355
1356                         switch (p->state) {
1357                         case TASK_TRACED:
1358                                 flag = 1;
1359                                 continue;
1360                         case TASK_STOPPED:
1361                                 /*
1362                                  * It's stopped now, so it might later
1363                                  * continue, exit, or stop again.
1364                                  */
1365                                 flag = 1;
1366                                 if (!(options & WUNTRACED))
1367                                         continue;
1368                                 if (tracehook_inhibit_wait_stopped(p))
1369                                         continue;
1370                                 retval = wait_task_stopped(p, ret == 2,
1371                                                            (options & WNOWAIT),
1372                                                            infop,
1373                                                            stat_addr, ru);
1374                                 if (retval == -EAGAIN)
1375                                         goto repeat;
1376                                 if (retval != 0) /* He released the lock.  */
1377                                         goto end;
1378                                 break;
1379                         default:
1380                         // case EXIT_DEAD:
1381                                 if (p->exit_state == EXIT_DEAD)
1382                                         continue;
1383                         // case EXIT_ZOMBIE:
1384                                 if (p->exit_state == EXIT_ZOMBIE) {
1385                                         /*
1386                                          * Eligible but we cannot release
1387                                          * it yet:
1388                                          */
1389                                         if (ret == 2)
1390                                                 goto check_continued;
1391                                         if (!likely(options & WEXITED))
1392                                                 continue;
1393                                         if (tracehook_inhibit_wait_zombie(p)) {
1394                                                 flag = 1;
1395                                                 continue;
1396                                         }
1397                                         retval = wait_task_zombie(
1398                                                 p, (options & WNOWAIT),
1399                                                 infop, stat_addr, ru);
1400                                         /* He released the lock.  */
1401                                         if (retval != 0)
1402                                                 goto end;
1403                                         break;
1404                                 }
1405 check_continued:
1406                                 /*
1407                                  * It's running now, so it might later
1408                                  * exit, stop, or stop and then continue.
1409                                  */
1410                                 flag = 1;
1411                                 if (!unlikely(options & WCONTINUED))
1412                                         continue;
1413                                 if (tracehook_inhibit_wait_continued(p))
1414                                         continue;
1415                                 retval = wait_task_continued(
1416                                         p, (options & WNOWAIT),
1417                                         infop, stat_addr, ru);
1418                                 if (retval != 0) /* He released the lock.  */
1419                                         goto end;
1420                                 break;
1421                         }
1422                 }
1423
1424                 retval = ptrace_do_wait(tsk, pid, options,
1425                                         infop, stat_addr, ru);
1426                 if (retval != -ECHILD) {
1427                         flag = 1;
1428                         if (retval != 0) /* He released the lock.  */
1429                                 goto end;
1430                 }
1431
1432                 if (options & __WNOTHREAD)
1433                         break;
1434                 tsk = next_thread(tsk);
1435                 BUG_ON(tsk->signal != current->signal);
1436         } while (tsk != current);
1437
1438         read_unlock(&tasklist_lock);
1439         if (flag) {
1440                 retval = 0;
1441                 if (options & WNOHANG)
1442                         goto end;
1443                 retval = -ERESTARTSYS;
1444                 if (signal_pending(current))
1445                         goto end;
1446                 schedule();
1447                 goto repeat;
1448         }
1449         retval = -ECHILD;
1450 end:
1451         current->state = TASK_RUNNING;
1452         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1453         if (infop) {
1454                 if (retval > 0)
1455                         retval = 0;
1456                 else {
1457                         /*
1458                          * For a WNOHANG return, clear out all the fields
1459                          * we would set so the user can easily tell the
1460                          * difference.
1461                          */
1462                         if (!retval)
1463                                 retval = put_user(0, &infop->si_signo);
1464                         if (!retval)
1465                                 retval = put_user(0, &infop->si_errno);
1466                         if (!retval)
1467                                 retval = put_user(0, &infop->si_code);
1468                         if (!retval)
1469                                 retval = put_user(0, &infop->si_pid);
1470                         if (!retval)
1471                                 retval = put_user(0, &infop->si_uid);
1472                         if (!retval)
1473                                 retval = put_user(0, &infop->si_status);
1474                 }
1475         }
1476         return retval;
1477 }
1478
1479 asmlinkage long sys_waitid(int which, pid_t pid,
1480                            struct siginfo __user *infop, int options,
1481                            struct rusage __user *ru)
1482 {
1483         long ret;
1484
1485         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1486                 return -EINVAL;
1487         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1488                 return -EINVAL;
1489
1490         switch (which) {
1491         case P_ALL:
1492                 pid = -1;
1493                 break;
1494         case P_PID:
1495                 if (pid <= 0)
1496                         return -EINVAL;
1497                 break;
1498         case P_PGID:
1499                 if (pid <= 0)
1500                         return -EINVAL;
1501                 pid = -pid;
1502                 break;
1503         default:
1504                 return -EINVAL;
1505         }
1506
1507         ret = do_wait(pid, options, infop, NULL, ru);
1508
1509         /* avoid REGPARM breakage on x86: */
1510         prevent_tail_call(ret);
1511         return ret;
1512 }
1513
1514 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1515                           int options, struct rusage __user *ru)
1516 {
1517         long ret;
1518
1519         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1520                         __WNOTHREAD|__WCLONE|__WALL))
1521                 return -EINVAL;
1522         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1523
1524         /* avoid REGPARM breakage on x86: */
1525         prevent_tail_call(ret);
1526         return ret;
1527 }
1528
1529 #ifdef __ARCH_WANT_SYS_WAITPID
1530
1531 /*
1532  * sys_waitpid() remains for compatibility. waitpid() should be
1533  * implemented by calling sys_wait4() from libc.a.
1534  */
1535 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1536 {
1537         return sys_wait4(pid, stat_addr, options, NULL);
1538 }
1539
1540 #endif