kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/config.h>
   8 #include <linux/mm.h>
   9 #include <linux/slab.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/module.h>
  13 #include <linux/completion.h>
  14 #include <linux/personality.h>
  15 #include <linux/tty.h>
  16 #include <linux/namespace.h>
  17 #include <linux/key.h>
  18 #include <linux/security.h>
  19 #include <linux/cpu.h>
  20 #include <linux/acct.h>
  21 #include <linux/file.h>
  22 #include <linux/binfmts.h>
  23 #include <linux/ptrace.h>
  24 #include <linux/profile.h>
  25 #include <linux/mount.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/mempolicy.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/vs_limit.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/unistd.h>
  33 #include <asm/pgtable.h>
  34 #include <asm/mmu_context.h>
  35
  36 extern void sem_exit (void);
  37 extern struct task_struct *child_reaper;
  38
  39 int getrusage(struct task_struct *, int, struct rusage __user *);
  40
  41 static void __unhash_process(struct task_struct *p)
  42 {
  43         nr_threads--;
  44         /* tasklist_lock is held, is this sufficient? */
  45         if (p->vx_info) {
  46                 atomic_dec(&p->vx_info->cvirt.nr_threads);
  47                 vx_nproc_dec(p);
  48         }
  49         detach_pid(p, PIDTYPE_PID);
  50         detach_pid(p, PIDTYPE_TGID);
  51         if (thread_group_leader(p)) {
  52                 detach_pid(p, PIDTYPE_PGID);
  53                 detach_pid(p, PIDTYPE_SID);
  54                 if (p->pid)
  55                         __get_cpu_var(process_counts)--;
  56         }
  57
  58         REMOVE_LINKS(p);
  59 }
  60
  61 void release_task(struct task_struct * p)
  62 {
  63         int zap_leader;
  64         task_t *leader;
  65         struct dentry *proc_dentry;
  66
  67 repeat:
  68         atomic_dec(&p->user->processes);
  69         spin_lock(&p->proc_lock);
  70         proc_dentry = proc_pid_unhash(p);
  71         write_lock_irq(&tasklist_lock);
  72         if (unlikely(p->ptrace))
  73                 __ptrace_unlink(p);
  74         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  75         __exit_signal(p);
  76         __exit_sighand(p);
  77         __unhash_process(p);
  78
  79         /*
  80          * If we are the last non-leader member of the thread
  81          * group, and the leader is zombie, then notify the
  82          * group leader's parent process. (if it wants notification.)
  83          */
  84         zap_leader = 0;
  85         leader = p->group_leader;
  86         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
  87                 BUG_ON(leader->exit_signal == -1);
  88                 do_notify_parent(leader, leader->exit_signal);
  89                 /*
  90                  * If we were the last child thread and the leader has
  91                  * exited already, and the leader's parent ignores SIGCHLD,
  92                  * then we are the one who should release the leader.
  93                  *
  94                  * do_notify_parent() will have marked it self-reaping in
  95                  * that case.
  96                  */
  97                 zap_leader = (leader->exit_signal == -1);
  98         }
  99
 100         sched_exit(p);
 101         write_unlock_irq(&tasklist_lock);
 102         spin_unlock(&p->proc_lock);
 103         proc_pid_flush(proc_dentry);
 104         release_thread(p);
 105         if (p->vx_info)
 106                 release_vx_info(p->vx_info, p);
 107         put_task_struct(p);
 108
 109         p = leader;
 110         if (unlikely(zap_leader))
 111                 goto repeat;
 112 }
 113
 114 /* we are using it only for SMP init */
 115
 116 void unhash_process(struct task_struct *p)
 117 {
 118         struct dentry *proc_dentry;
 119
 120         spin_lock(&p->proc_lock);
 121         proc_dentry = proc_pid_unhash(p);
 122         write_lock_irq(&tasklist_lock);
 123         __unhash_process(p);
 124         write_unlock_irq(&tasklist_lock);
 125         spin_unlock(&p->proc_lock);
 126         proc_pid_flush(proc_dentry);
 127 }
 128
 129 /*
 130  * This checks not only the pgrp, but falls back on the pid if no
 131  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 132  * without this...
 133  */
 134 int session_of_pgrp(int pgrp)
 135 {
 136         struct task_struct *p;
 137         int sid = -1;
 138
 139         read_lock(&tasklist_lock);
 140         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 141                 if (p->signal->session > 0) {
 142                         sid = p->signal->session;
 143                         goto out;
 144                 }
 145         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 146         p = find_task_by_pid(pgrp);
 147         if (p)
 148                 sid = p->signal->session;
 149 out:
 150         read_unlock(&tasklist_lock);
 151
 152         return sid;
 153 }
 154
 155 /*
 156  * Determine if a process group is "orphaned", according to the POSIX
 157  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 158  * by terminal-generated stop signals.  Newly orphaned process groups are
 159  * to receive a SIGHUP and a SIGCONT.
 160  *
 161  * "I ask you, have you ever known what it is to be an orphan?"
 162  */
 163 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 164 {
 165         struct task_struct *p;
 166         int ret = 1;
 167
 168         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 169                 if (p == ignored_task
 170                                 || p->exit_state
 171                                 || p->real_parent->pid == 1)
 172                         continue;
 173                 if (process_group(p->real_parent) != pgrp
 174                             && p->real_parent->signal->session == p->signal->session) {
 175                         ret = 0;
 176                         break;
 177                 }
 178         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 179         return ret;     /* (sighing) "Often!" */
 180 }
 181
 182 int is_orphaned_pgrp(int pgrp)
 183 {
 184         int retval;
 185
 186         read_lock(&tasklist_lock);
 187         retval = will_become_orphaned_pgrp(pgrp, NULL);
 188         read_unlock(&tasklist_lock);
 189
 190         return retval;
 191 }
 192
 193 static inline int has_stopped_jobs(int pgrp)
 194 {
 195         int retval = 0;
 196         struct task_struct *p;
 197
 198         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 199                 if (p->state != TASK_STOPPED)
 200                         continue;
 201
 202                 /* If p is stopped by a debugger on a signal that won't
 203                    stop it, then don't count p as stopped.  This isn't
 204                    perfect but it's a good approximation.  */
 205                 if (unlikely (p->ptrace)
 206                     && p->exit_code != SIGSTOP
 207                     && p->exit_code != SIGTSTP
 208                     && p->exit_code != SIGTTOU
 209                     && p->exit_code != SIGTTIN)
 210                         continue;
 211
 212                 retval = 1;
 213                 break;
 214         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 215         return retval;
 216 }
 217
 218 /**
 219  * reparent_to_init() - Reparent the calling kernel thread to the init task.
 220  *
 221  * If a kernel thread is launched as a result of a system call, or if
 222  * it ever exits, it should generally reparent itself to init so that
 223  * it is correctly cleaned up on exit.
 224  *
 225  * The various task state such as scheduling policy and priority may have
 226  * been inherited from a user process, so we reset them to sane values here.
 227  *
 228  * NOTE that reparent_to_init() gives the caller full capabilities.
 229  */
 230 void reparent_to_init(void)
 231 {
 232         write_lock_irq(&tasklist_lock);
 233
 234         ptrace_unlink(current);
 235         /* Reparent to init */
 236         REMOVE_LINKS(current);
 237         /* FIXME handle vchild_reaper/initpid */
 238         current->parent = child_reaper;
 239         current->real_parent = child_reaper;
 240         SET_LINKS(current);
 241
 242         /* Set the exit signal to SIGCHLD so we signal init on exit */
 243         current->exit_signal = SIGCHLD;
 244
 245         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
 246                 set_user_nice(current, 0);
 247         /* cpus_allowed? */
 248         /* rt_priority? */
 249         /* signals? */
 250         security_task_reparent_to_init(current);
 251         memcpy(current->signal->rlim, init_task.signal->rlim,
 252                sizeof(current->signal->rlim));
 253         atomic_inc(&(INIT_USER->__count));
 254         write_unlock_irq(&tasklist_lock);
 255         switch_uid(INIT_USER);
 256 }
 257
 258 void __set_special_pids(pid_t session, pid_t pgrp)
 259 {
 260         struct task_struct *curr = current;
 261
 262         if (curr->signal->session != session) {
 263                 detach_pid(curr, PIDTYPE_SID);
 264                 curr->signal->session = session;
 265                 attach_pid(curr, PIDTYPE_SID, session);
 266         }
 267         if (process_group(curr) != pgrp) {
 268                 detach_pid(curr, PIDTYPE_PGID);
 269                 curr->signal->pgrp = pgrp;
 270                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 271         }
 272 }
 273
 274 void set_special_pids(pid_t session, pid_t pgrp)
 275 {
 276         write_lock_irq(&tasklist_lock);
 277         __set_special_pids(session, pgrp);
 278         write_unlock_irq(&tasklist_lock);
 279 }
 280
 281 /*
 282  * Let kernel threads use this to say that they
 283  * allow a certain signal (since daemonize() will
 284  * have disabled all of them by default).
 285  */
 286 int allow_signal(int sig)
 287 {
 288         if (sig < 1 || sig > _NSIG)
 289                 return -EINVAL;
 290
 291         spin_lock_irq(&current->sighand->siglock);
 292         sigdelset(&current->blocked, sig);
 293         if (!current->mm) {
 294                 /* Kernel threads handle their own signals.
 295                    Let the signal code know it'll be handled, so
 296                    that they don't get converted to SIGKILL or
 297                    just silently dropped */
 298                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 299         }
 300         recalc_sigpending();
 301         spin_unlock_irq(&current->sighand->siglock);
 302         return 0;
 303 }
 304
 305 EXPORT_SYMBOL(allow_signal);
 306
 307 int disallow_signal(int sig)
 308 {
 309         if (sig < 1 || sig > _NSIG)
 310                 return -EINVAL;
 311
 312         spin_lock_irq(&current->sighand->siglock);
 313         sigaddset(&current->blocked, sig);
 314         recalc_sigpending();
 315         spin_unlock_irq(&current->sighand->siglock);
 316         return 0;
 317 }
 318
 319 EXPORT_SYMBOL(disallow_signal);
 320
 321 /*
 322  *      Put all the gunge required to become a kernel thread without
 323  *      attached user resources in one place where it belongs.
 324  */
 325
 326 void daemonize(const char *name, ...)
 327 {
 328         va_list args;
 329         struct fs_struct *fs;
 330         sigset_t blocked;
 331
 332         va_start(args, name);
 333         vsnprintf(current->comm, sizeof(current->comm), name, args);
 334         va_end(args);
 335
 336         /*
 337          * If we were started as result of loading a module, close all of the
 338          * user space pages.  We don't need them, and if we didn't close them
 339          * they would be locked into memory.
 340          */
 341         exit_mm(current);
 342
 343         set_special_pids(1, 1);
 344         down(&tty_sem);
 345         current->signal->tty = NULL;
 346         up(&tty_sem);
 347
 348         /* Block and flush all signals */
 349         sigfillset(&blocked);
 350         sigprocmask(SIG_BLOCK, &blocked, NULL);
 351         flush_signals(current);
 352
 353         /* Become as one with the init task */
 354
 355         exit_fs(current);       /* current->fs->count--; */
 356         fs = init_task.fs;
 357         current->fs = fs;
 358         atomic_inc(&fs->count);
 359         exit_files(current);
 360         current->files = init_task.files;
 361         atomic_inc(&current->files->count);
 362
 363         reparent_to_init();
 364 }
 365
 366 EXPORT_SYMBOL(daemonize);
 367
 368 static inline void close_files(struct files_struct * files)
 369 {
 370         int i, j;
 371
 372         j = 0;
 373         for (;;) {
 374                 unsigned long set;
 375                 i = j * __NFDBITS;
 376                 if (i >= files->max_fdset || i >= files->max_fds)
 377                         break;
 378                 set = files->open_fds->fds_bits[j++];
 379                 while (set) {
 380                         if (set & 1) {
 381                                 struct file * file = xchg(&files->fd[i], NULL);
 382                                 if (file)
 383                                         filp_close(file, files);
 384                                 vx_openfd_dec(i);
 385                         }
 386                         i++;
 387                         set >>= 1;
 388                 }
 389         }
 390 }
 391
 392 struct files_struct *get_files_struct(struct task_struct *task)
 393 {
 394         struct files_struct *files;
 395
 396         task_lock(task);
 397         files = task->files;
 398         if (files)
 399                 atomic_inc(&files->count);
 400         task_unlock(task);
 401
 402         return files;
 403 }
 404
 405 void fastcall put_files_struct(struct files_struct *files)
 406 {
 407         if (atomic_dec_and_test(&files->count)) {
 408                 close_files(files);
 409                 /*
 410                  * Free the fd and fdset arrays if we expanded them.
 411                  */
 412                 if (files->fd != &files->fd_array[0])
 413                         free_fd_array(files->fd, files->max_fds);
 414                 if (files->max_fdset > __FD_SETSIZE) {
 415                         free_fdset(files->open_fds, files->max_fdset);
 416                         free_fdset(files->close_on_exec, files->max_fdset);
 417                 }
 418                 kmem_cache_free(files_cachep, files);
 419         }
 420 }
 421
 422 EXPORT_SYMBOL(put_files_struct);
 423
 424 static inline void __exit_files(struct task_struct *tsk)
 425 {
 426         struct files_struct * files = tsk->files;
 427
 428         if (files) {
 429                 task_lock(tsk);
 430                 tsk->files = NULL;
 431                 task_unlock(tsk);
 432                 put_files_struct(files);
 433         }
 434 }
 435
 436 void exit_files(struct task_struct *tsk)
 437 {
 438         __exit_files(tsk);
 439 }
 440
 441 static inline void __put_fs_struct(struct fs_struct *fs)
 442 {
 443         /* No need to hold fs->lock if we are killing it */
 444         if (atomic_dec_and_test(&fs->count)) {
 445                 dput(fs->root);
 446                 mntput(fs->rootmnt);
 447                 dput(fs->pwd);
 448                 mntput(fs->pwdmnt);
 449                 if (fs->altroot) {
 450                         dput(fs->altroot);
 451                         mntput(fs->altrootmnt);
 452                 }
 453                 kmem_cache_free(fs_cachep, fs);
 454         }
 455 }
 456
 457 void put_fs_struct(struct fs_struct *fs)
 458 {
 459         __put_fs_struct(fs);
 460 }
 461
 462 static inline void __exit_fs(struct task_struct *tsk)
 463 {
 464         struct fs_struct * fs = tsk->fs;
 465
 466         if (fs) {
 467                 task_lock(tsk);
 468                 tsk->fs = NULL;
 469                 task_unlock(tsk);
 470                 __put_fs_struct(fs);
 471         }
 472 }
 473
 474 void exit_fs(struct task_struct *tsk)
 475 {
 476         __exit_fs(tsk);
 477 }
 478
 479 EXPORT_SYMBOL_GPL(exit_fs);
 480
 481 /*
 482  * Turn us into a lazy TLB process if we
 483  * aren't already..
 484  */
 485 void exit_mm(struct task_struct * tsk)
 486 {
 487         struct mm_struct *mm = tsk->mm;
 488
 489         mm_release(tsk, mm);
 490         if (!mm)
 491                 return;
 492         /*
 493          * Serialize with any possible pending coredump.
 494          * We must hold mmap_sem around checking core_waiters
 495          * and clearing tsk->mm.  The core-inducing thread
 496          * will increment core_waiters for each thread in the
 497          * group with ->mm != NULL.
 498          */
 499         down_read(&mm->mmap_sem);
 500         if (mm->core_waiters) {
 501                 up_read(&mm->mmap_sem);
 502                 down_write(&mm->mmap_sem);
 503                 if (!--mm->core_waiters)
 504                         complete(mm->core_startup_done);
 505                 up_write(&mm->mmap_sem);
 506
 507                 wait_for_completion(&mm->core_done);
 508                 down_read(&mm->mmap_sem);
 509         }
 510         atomic_inc(&mm->mm_count);
 511         if (mm != tsk->active_mm) BUG();
 512         /* more a memory barrier than a real lock */
 513         task_lock(tsk);
 514         tsk->mm = NULL;
 515         up_read(&mm->mmap_sem);
 516         enter_lazy_tlb(mm, current);
 517         task_unlock(tsk);
 518         mmput(mm);
 519 }
 520
 521 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 522 {
 523         /*
 524          * Make sure we're not reparenting to ourselves and that
 525          * the parent is not a zombie.
 526          */
 527         BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
 528         p->real_parent = reaper;
 529         if (p->parent == p->real_parent)
 530                 BUG();
 531 }
 532
 533 static inline void reparent_thread(task_t *p, task_t *father, int traced)
 534 {
 535         /* We don't want people slaying init.  */
 536         if (p->exit_signal != -1)
 537                 p->exit_signal = SIGCHLD;
 538
 539         if (p->pdeath_signal)
 540                 /* We already hold the tasklist_lock here.  */
 541                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
 542
 543         /* Move the child from its dying parent to the new one.  */
 544         if (unlikely(traced)) {
 545                 /* Preserve ptrace links if someone else is tracing this child.  */
 546                 list_del_init(&p->ptrace_list);
 547                 if (p->parent != p->real_parent)
 548                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 549         } else {
 550                 /* If this child is being traced, then we're the one tracing it
 551                  * anyway, so let go of it.
 552                  */
 553                 p->ptrace = 0;
 554                 list_del_init(&p->sibling);
 555                 p->parent = p->real_parent;
 556                 list_add_tail(&p->sibling, &p->parent->children);
 557
 558                 /* If we'd notified the old parent about this child's death,
 559                  * also notify the new parent.
 560                  */
 561                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 562                     thread_group_empty(p))
 563                         do_notify_parent(p, p->exit_signal);
 564                 else if (p->state == TASK_TRACED) {
 565                         /*
 566                          * If it was at a trace stop, turn it into
 567                          * a normal stop since it's no longer being
 568                          * traced.
 569                          */
 570                         ptrace_untrace(p);
 571                 }
 572         }
 573
 574         /*
 575          * process group orphan check
 576          * Case ii: Our child is in a different pgrp
 577          * than we are, and it was the only connection
 578          * outside, so the child pgrp is now orphaned.
 579          */
 580         if ((process_group(p) != process_group(father)) &&
 581             (p->signal->session == father->signal->session)) {
 582                 int pgrp = process_group(p);
 583
 584                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 585                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
 586                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
 587                 }
 588         }
 589 }
 590
 591 /*
 592  * When we die, we re-parent all our children.
 593  * Try to give them to another thread in our thread
 594  * group, and if no such member exists, give it to
 595  * the global child reaper process (ie "init")
 596  */
 597 static inline void forget_original_parent(struct task_struct * father,
 598                                           struct list_head *to_release)
 599 {
 600         struct task_struct *p, *reaper = father;
 601         struct list_head *_p, *_n;
 602
 603         /* FIXME handle vchild_reaper/initpid */
 604         do {
 605                 reaper = next_thread(reaper);
 606                 if (reaper == father) {
 607                         reaper = child_reaper;
 608                         break;
 609                 }
 610         } while (reaper->exit_state);
 611
 612         /*
 613          * There are only two places where our children can be:
 614          *
 615          * - in our child list
 616          * - in our ptraced child list
 617          *
 618          * Search them and reparent children.
 619          */
 620         list_for_each_safe(_p, _n, &father->children) {
 621                 int ptrace;
 622                 p = list_entry(_p,struct task_struct,sibling);
 623
 624                 ptrace = p->ptrace;
 625
 626                 /* if father isn't the real parent, then ptrace must be enabled */
 627                 BUG_ON(father != p->real_parent && !ptrace);
 628
 629                 if (father == p->real_parent) {
 630                         /* reparent with a reaper, real father it's us */
 631                         choose_new_parent(p, reaper, child_reaper);
 632                         reparent_thread(p, father, 0);
 633                 } else {
 634                         /* reparent ptraced task to its real parent */
 635                         __ptrace_unlink (p);
 636                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 637                             thread_group_empty(p))
 638                                 do_notify_parent(p, p->exit_signal);
 639                 }
 640
 641                 /*
 642                  * if the ptraced child is a zombie with exit_signal == -1
 643                  * we must collect it before we exit, or it will remain
 644                  * zombie forever since we prevented it from self-reap itself
 645                  * while it was being traced by us, to be able to see it in wait4.
 646                  */
 647                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 648                         list_add(&p->ptrace_list, to_release);
 649         }
 650         list_for_each_safe(_p, _n, &father->ptrace_children) {
 651                 p = list_entry(_p,struct task_struct,ptrace_list);
 652                 choose_new_parent(p, reaper, child_reaper);
 653                 reparent_thread(p, father, 1);
 654         }
 655 }
 656
 657 /*
 658  * Send signals to all our closest relatives so that they know
 659  * to properly mourn us..
 660  */
 661 static void exit_notify(struct task_struct *tsk)
 662 {
 663         int state;
 664         struct task_struct *t;
 665         struct list_head ptrace_dead, *_p, *_n;
 666
 667         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 668             && !thread_group_empty(tsk)) {
 669                 /*
 670                  * This occurs when there was a race between our exit
 671                  * syscall and a group signal choosing us as the one to
 672                  * wake up.  It could be that we are the only thread
 673                  * alerted to check for pending signals, but another thread
 674                  * should be woken now to take the signal since we will not.
 675                  * Now we'll wake all the threads in the group just to make
 676                  * sure someone gets all the pending signals.
 677                  */
 678                 read_lock(&tasklist_lock);
 679                 spin_lock_irq(&tsk->sighand->siglock);
 680                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 681                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 682                                 recalc_sigpending_tsk(t);
 683                                 if (signal_pending(t))
 684                                         signal_wake_up(t, 0);
 685                         }
 686                 spin_unlock_irq(&tsk->sighand->siglock);
 687                 read_unlock(&tasklist_lock);
 688         }
 689
 690         write_lock_irq(&tasklist_lock);
 691
 692         /*
 693          * This does two things:
 694          *
 695          * A.  Make init inherit all the child processes
 696          * B.  Check to see if any process groups have become orphaned
 697          *      as a result of our exiting, and if they have any stopped
 698          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 699          */
 700
 701         INIT_LIST_HEAD(&ptrace_dead);
 702         forget_original_parent(tsk, &ptrace_dead);
 703         BUG_ON(!list_empty(&tsk->children));
 704         BUG_ON(!list_empty(&tsk->ptrace_children));
 705
 706         /*
 707          * Check to see if any process groups have become orphaned
 708          * as a result of our exiting, and if they have any stopped
 709          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 710          *
 711          * Case i: Our father is in a different pgrp than we are
 712          * and we were the only connection outside, so our pgrp
 713          * is about to become orphaned.
 714          */
 715
 716         t = tsk->real_parent;
 717
 718         if ((process_group(t) != process_group(tsk)) &&
 719             (t->signal->session == tsk->signal->session) &&
 720             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 721             has_stopped_jobs(process_group(tsk))) {
 722                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
 723                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
 724         }
 725
 726         /* Let father know we died
 727          *
 728          * Thread signals are configurable, but you aren't going to use
 729          * that to send signals to arbitary processes.
 730          * That stops right now.
 731          *
 732          * If the parent exec id doesn't match the exec id we saved
 733          * when we started then we know the parent has changed security
 734          * domain.
 735          *
 736          * If our self_exec id doesn't match our parent_exec_id then
 737          * we have changed execution domain as these two values started
 738          * the same after a fork.
 739          *
 740          */
 741
 742         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 743             ( tsk->parent_exec_id != t->self_exec_id  ||
 744               tsk->self_exec_id != tsk->parent_exec_id)
 745             && !capable(CAP_KILL))
 746                 tsk->exit_signal = SIGCHLD;
 747
 748
 749         /* If something other than our normal parent is ptracing us, then
 750          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 751          * only has special meaning to our real parent.
 752          */
 753         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 754                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 755                 do_notify_parent(tsk, signal);
 756         } else if (tsk->ptrace) {
 757                 do_notify_parent(tsk, SIGCHLD);
 758         }
 759
 760         state = EXIT_ZOMBIE;
 761         if (tsk->exit_signal == -1 &&
 762             (likely(tsk->ptrace == 0) ||
 763              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 764                 state = EXIT_DEAD;
 765         tsk->exit_state = state;
 766
 767         /*
 768          * Clear these here so that update_process_times() won't try to deliver
 769          * itimer, profile or rlimit signals to this task while it is in late exit.
 770          */
 771         tsk->it_virt_value = cputime_zero;
 772         tsk->it_prof_value = cputime_zero;
 773
 774         write_unlock_irq(&tasklist_lock);
 775
 776         list_for_each_safe(_p, _n, &ptrace_dead) {
 777                 list_del_init(_p);
 778                 t = list_entry(_p,struct task_struct,ptrace_list);
 779                 release_task(t);
 780         }
 781
 782         /* If the process is dead, release it - nobody will wait for it */
 783         if (state == EXIT_DEAD)
 784                 release_task(tsk);
 785
 786         /* PF_DEAD causes final put_task_struct after we schedule. */
 787         preempt_disable();
 788         tsk->flags |= PF_DEAD;
 789 }
 790
 791 fastcall NORET_TYPE void do_exit(long code)
 792 {
 793         struct task_struct *tsk = current;
 794         int group_dead;
 795
 796         profile_task_exit(tsk);
 797
 798         if (unlikely(in_interrupt()))
 799                 panic("Aiee, killing interrupt handler!");
 800         if (unlikely(!tsk->pid))
 801                 panic("Attempted to kill the idle task!");
 802         if (unlikely(tsk->pid == 1))
 803                 panic("Attempted to kill init!");
 804         if (tsk->io_context)
 805                 exit_io_context();
 806
 807         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 808                 current->ptrace_message = code;
 809                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 810         }
 811
 812         tsk->flags |= PF_EXITING;
 813         del_timer_sync(&tsk->real_timer);
 814
 815         if (unlikely(in_atomic()))
 816                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 817                                 current->comm, current->pid,
 818                                 preempt_count());
 819
 820         acct_update_integrals();
 821         update_mem_hiwater();
 822         group_dead = atomic_dec_and_test(&tsk->signal->live);
 823         if (group_dead)
 824                 acct_process(code);
 825         exit_mm(tsk);
 826
 827         exit_sem(tsk);
 828         __exit_files(tsk);
 829         __exit_fs(tsk);
 830         exit_namespace(tsk);
 831         exit_thread();
 832         exit_keys(tsk);
 833
 834         if (group_dead && tsk->signal->leader)
 835                 disassociate_ctty(1);
 836
 837         module_put(tsk->thread_info->exec_domain->module);
 838         if (tsk->binfmt)
 839                 module_put(tsk->binfmt->module);
 840
 841         tsk->exit_code = code;
 842         exit_notify(tsk);
 843 #ifdef CONFIG_NUMA
 844         mpol_free(tsk->mempolicy);
 845         tsk->mempolicy = NULL;
 846 #endif
 847
 848         BUG_ON(!(current->flags & PF_DEAD));
 849         schedule();
 850         BUG();
 851         /* Avoid "noreturn function does return".  */
 852         for (;;) ;
 853 }
 854
 855 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 856 {
 857         if (comp)
 858                 complete(comp);
 859
 860         do_exit(code);
 861 }
 862
 863 EXPORT_SYMBOL(complete_and_exit);
 864
 865 asmlinkage long sys_exit(int error_code)
 866 {
 867         do_exit((error_code&0xff)<<8);
 868 }
 869
 870 task_t fastcall *next_thread(const task_t *p)
 871 {
 872         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 873 }
 874
 875 EXPORT_SYMBOL(next_thread);
 876
 877 /*
 878  * Take down every thread in the group.  This is called by fatal signals
 879  * as well as by sys_exit_group (below).
 880  */
 881 NORET_TYPE void
 882 do_group_exit(int exit_code)
 883 {
 884         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 885
 886         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 887                 exit_code = current->signal->group_exit_code;
 888         else if (!thread_group_empty(current)) {
 889                 struct signal_struct *const sig = current->signal;
 890                 struct sighand_struct *const sighand = current->sighand;
 891                 read_lock(&tasklist_lock);
 892                 spin_lock_irq(&sighand->siglock);
 893                 if (sig->flags & SIGNAL_GROUP_EXIT)
 894                         /* Another thread got here before we took the lock.  */
 895                         exit_code = sig->group_exit_code;
 896                 else {
 897                         sig->flags = SIGNAL_GROUP_EXIT;
 898                         sig->group_exit_code = exit_code;
 899                         zap_other_threads(current);
 900                 }
 901                 spin_unlock_irq(&sighand->siglock);
 902                 read_unlock(&tasklist_lock);
 903         }
 904
 905         do_exit(exit_code);
 906         /* NOTREACHED */
 907 }
 908
 909 /*
 910  * this kills every thread in the thread group. Note that any externally
 911  * wait4()-ing process will get the correct exit code - even if this
 912  * thread is not the thread group leader.
 913  */
 914 asmlinkage void sys_exit_group(int error_code)
 915 {
 916         do_group_exit((error_code & 0xff) << 8);
 917 }
 918
 919 static int eligible_child(pid_t pid, int options, task_t *p)
 920 {
 921         if (pid > 0) {
 922                 if (p->pid != pid)
 923                         return 0;
 924         } else if (!pid) {
 925                 if (process_group(p) != process_group(current))
 926                         return 0;
 927         } else if (pid != -1) {
 928                 if (process_group(p) != -pid)
 929                         return 0;
 930         }
 931
 932         /*
 933          * Do not consider detached threads that are
 934          * not ptraced:
 935          */
 936         if (p->exit_signal == -1 && !p->ptrace)
 937                 return 0;
 938
 939         /* Wait for all children (clone and not) if __WALL is set;
 940          * otherwise, wait for clone children *only* if __WCLONE is
 941          * set; otherwise, wait for non-clone children *only*.  (Note:
 942          * A "clone" child here is one that reports to its parent
 943          * using a signal other than SIGCHLD.) */
 944         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 945             && !(options & __WALL))
 946                 return 0;
 947         /*
 948          * Do not consider thread group leaders that are
 949          * in a non-empty thread group:
 950          */
 951         if (current->tgid != p->tgid && delay_group_leader(p))
 952                 return 2;
 953
 954         if (security_task_wait(p))
 955                 return 0;
 956
 957         return 1;
 958 }
 959
 960 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 961                                int why, int status,
 962                                struct siginfo __user *infop,
 963                                struct rusage __user *rusagep)
 964 {
 965         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
 966         put_task_struct(p);
 967         if (!retval)
 968                 retval = put_user(SIGCHLD, &infop->si_signo);
 969         if (!retval)
 970                 retval = put_user(0, &infop->si_errno);
 971         if (!retval)
 972                 retval = put_user((short)why, &infop->si_code);
 973         if (!retval)
 974                 retval = put_user(pid, &infop->si_pid);
 975         if (!retval)
 976                 retval = put_user(uid, &infop->si_uid);
 977         if (!retval)
 978                 retval = put_user(status, &infop->si_status);
 979         if (!retval)
 980                 retval = pid;
 981         return retval;
 982 }
 983
 984 /*
 985  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
 986  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
 987  * the lock and this task is uninteresting.  If we return nonzero, we have
 988  * released the lock and the system call should return.
 989  */
 990 static int wait_task_zombie(task_t *p, int noreap,
 991                             struct siginfo __user *infop,
 992                             int __user *stat_addr, struct rusage __user *ru)
 993 {
 994         unsigned long state;
 995         int retval;
 996         int status;
 997
 998         if (unlikely(noreap)) {
 999                 pid_t pid = p->pid;
1000                 uid_t uid = p->uid;
1001                 int exit_code = p->exit_code;
1002                 int why, status;
1003
1004                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1005                         return 0;
1006                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1007                         return 0;
1008                 get_task_struct(p);
1009                 read_unlock(&tasklist_lock);
1010                 if ((exit_code & 0x7f) == 0) {
1011                         why = CLD_EXITED;
1012                         status = exit_code >> 8;
1013                 } else {
1014                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1015                         status = exit_code & 0x7f;
1016                 }
1017                 return wait_noreap_copyout(p, pid, uid, why,
1018                                            status, infop, ru);
1019         }
1020
1021         /*
1022          * Try to move the task's state to DEAD
1023          * only one thread is allowed to do this:
1024          */
1025         state = xchg(&p->exit_state, EXIT_DEAD);
1026         if (state != EXIT_ZOMBIE) {
1027                 BUG_ON(state != EXIT_DEAD);
1028                 return 0;
1029         }
1030         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1031                 /*
1032                  * This can only happen in a race with a ptraced thread
1033                  * dying on another processor.
1034                  */
1035                 return 0;
1036         }
1037
1038         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1039                 /*
1040                  * The resource counters for the group leader are in its
1041                  * own task_struct.  Those for dead threads in the group
1042                  * are in its signal_struct, as are those for the child
1043                  * processes it has previously reaped.  All these
1044                  * accumulate in the parent's signal_struct c* fields.
1045                  *
1046                  * We don't bother to take a lock here to protect these
1047                  * p->signal fields, because they are only touched by
1048                  * __exit_signal, which runs with tasklist_lock
1049                  * write-locked anyway, and so is excluded here.  We do
1050                  * need to protect the access to p->parent->signal fields,
1051                  * as other threads in the parent group can be right
1052                  * here reaping other children at the same time.
1053                  */
1054                 spin_lock_irq(&p->parent->sighand->siglock);
1055                 p->parent->signal->cutime =
1056                         cputime_add(p->parent->signal->cutime,
1057                         cputime_add(p->utime,
1058                         cputime_add(p->signal->utime,
1059                                     p->signal->cutime)));
1060                 p->parent->signal->cstime =
1061                         cputime_add(p->parent->signal->cstime,
1062                         cputime_add(p->stime,
1063                         cputime_add(p->signal->stime,
1064                                     p->signal->cstime)));
1065                 p->parent->signal->cmin_flt +=
1066                         p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
1067                 p->parent->signal->cmaj_flt +=
1068                         p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
1069                 p->parent->signal->cnvcsw +=
1070                         p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
1071                 p->parent->signal->cnivcsw +=
1072                         p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
1073                 spin_unlock_irq(&p->parent->sighand->siglock);
1074         }
1075
1076         /*
1077          * Now we are sure this task is interesting, and no other
1078          * thread can reap it because we set its state to EXIT_DEAD.
1079          */
1080         read_unlock(&tasklist_lock);
1081
1082         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1083         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1084                 ? p->signal->group_exit_code : p->exit_code;
1085         if (!retval && stat_addr)
1086                 retval = put_user(status, stat_addr);
1087         if (!retval && infop)
1088                 retval = put_user(SIGCHLD, &infop->si_signo);
1089         if (!retval && infop)
1090                 retval = put_user(0, &infop->si_errno);
1091         if (!retval && infop) {
1092                 int why;
1093
1094                 if ((status & 0x7f) == 0) {
1095                         why = CLD_EXITED;
1096                         status >>= 8;
1097                 } else {
1098                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1099                         status &= 0x7f;
1100                 }
1101                 retval = put_user((short)why, &infop->si_code);
1102                 if (!retval)
1103                         retval = put_user(status, &infop->si_status);
1104         }
1105         if (!retval && infop)
1106                 retval = put_user(p->pid, &infop->si_pid);
1107         if (!retval && infop)
1108                 retval = put_user(p->uid, &infop->si_uid);
1109         if (retval) {
1110                 // TODO: is this safe?
1111                 p->exit_state = EXIT_ZOMBIE;
1112                 return retval;
1113         }
1114         retval = p->pid;
1115         if (p->real_parent != p->parent) {
1116                 write_lock_irq(&tasklist_lock);
1117                 /* Double-check with lock held.  */
1118                 if (p->real_parent != p->parent) {
1119                         __ptrace_unlink(p);
1120                         // TODO: is this safe?
1121                         p->exit_state = EXIT_ZOMBIE;
1122                         /*
1123                          * If this is not a detached task, notify the parent.
1124                          * If it's still not detached after that, don't release
1125                          * it now.
1126                          */
1127                         if (p->exit_signal != -1) {
1128                                 do_notify_parent(p, p->exit_signal);
1129                                 if (p->exit_signal != -1)
1130                                         p = NULL;
1131                         }
1132                 }
1133                 write_unlock_irq(&tasklist_lock);
1134         }
1135         if (p != NULL)
1136                 release_task(p);
1137         BUG_ON(!retval);
1138         return retval;
1139 }
1140
1141 /*
1142  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1143  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1144  * the lock and this task is uninteresting.  If we return nonzero, we have
1145  * released the lock and the system call should return.
1146  */
1147 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1148                              struct siginfo __user *infop,
1149                              int __user *stat_addr, struct rusage __user *ru)
1150 {
1151         int retval, exit_code;
1152
1153         if (!p->exit_code)
1154                 return 0;
1155         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1156             p->signal && p->signal->group_stop_count > 0)
1157                 /*
1158                  * A group stop is in progress and this is the group leader.
1159                  * We won't report until all threads have stopped.
1160                  */
1161                 return 0;
1162
1163         /*
1164          * Now we are pretty sure this task is interesting.
1165          * Make sure it doesn't get reaped out from under us while we
1166          * give up the lock and then examine it below.  We don't want to
1167          * keep holding onto the tasklist_lock while we call getrusage and
1168          * possibly take page faults for user memory.
1169          */
1170         get_task_struct(p);
1171         read_unlock(&tasklist_lock);
1172
1173         if (unlikely(noreap)) {
1174                 pid_t pid = p->pid;
1175                 uid_t uid = p->uid;
1176                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1177
1178                 exit_code = p->exit_code;
1179                 if (unlikely(!exit_code) ||
1180                     unlikely(p->state > TASK_STOPPED))
1181                         goto bail_ref;
1182                 return wait_noreap_copyout(p, pid, uid,
1183                                            why, (exit_code << 8) | 0x7f,
1184                                            infop, ru);
1185         }
1186
1187         write_lock_irq(&tasklist_lock);
1188
1189         /*
1190          * This uses xchg to be atomic with the thread resuming and setting
1191          * it.  It must also be done with the write lock held to prevent a
1192          * race with the EXIT_ZOMBIE case.
1193          */
1194         exit_code = xchg(&p->exit_code, 0);
1195         if (unlikely(p->exit_state)) {
1196                 /*
1197                  * The task resumed and then died.  Let the next iteration
1198                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1199                  * already be zero here if it resumed and did _exit(0).
1200                  * The task itself is dead and won't touch exit_code again;
1201                  * other processors in this function are locked out.
1202                  */
1203                 p->exit_code = exit_code;
1204                 exit_code = 0;
1205         }
1206         if (unlikely(exit_code == 0)) {
1207                 /*
1208                  * Another thread in this function got to it first, or it
1209                  * resumed, or it resumed and then died.
1210                  */
1211                 write_unlock_irq(&tasklist_lock);
1212 bail_ref:
1213                 put_task_struct(p);
1214                 /*
1215                  * We are returning to the wait loop without having successfully
1216                  * removed the process and having released the lock. We cannot
1217                  * continue, since the "p" task pointer is potentially stale.
1218                  *
1219                  * Return -EAGAIN, and do_wait() will restart the loop from the
1220                  * beginning. Do _not_ re-acquire the lock.
1221                  */
1222                 return -EAGAIN;
1223         }
1224
1225         /* move to end of parent's list to avoid starvation */
1226         remove_parent(p);
1227         add_parent(p, p->parent);
1228
1229         write_unlock_irq(&tasklist_lock);
1230
1231         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1232         if (!retval && stat_addr)
1233                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1234         if (!retval && infop)
1235                 retval = put_user(SIGCHLD, &infop->si_signo);
1236         if (!retval && infop)
1237                 retval = put_user(0, &infop->si_errno);
1238         if (!retval && infop)
1239                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1240                                           ? CLD_TRAPPED : CLD_STOPPED),
1241                                   &infop->si_code);
1242         if (!retval && infop)
1243                 retval = put_user(exit_code, &infop->si_status);
1244         if (!retval && infop)
1245                 retval = put_user(p->pid, &infop->si_pid);
1246         if (!retval && infop)
1247                 retval = put_user(p->uid, &infop->si_uid);
1248         if (!retval)
1249                 retval = p->pid;
1250         put_task_struct(p);
1251
1252         BUG_ON(!retval);
1253         return retval;
1254 }
1255
1256 /*
1257  * Handle do_wait work for one task in a live, non-stopped state.
1258  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1259  * the lock and this task is uninteresting.  If we return nonzero, we have
1260  * released the lock and the system call should return.
1261  */
1262 static int wait_task_continued(task_t *p, int noreap,
1263                                struct siginfo __user *infop,
1264                                int __user *stat_addr, struct rusage __user *ru)
1265 {
1266         int retval;
1267         pid_t pid;
1268         uid_t uid;
1269
1270         if (unlikely(!p->signal))
1271                 return 0;
1272
1273         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1274                 return 0;
1275
1276         spin_lock_irq(&p->sighand->siglock);
1277         /* Re-check with the lock held.  */
1278         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1279                 spin_unlock_irq(&p->sighand->siglock);
1280                 return 0;
1281         }
1282         if (!noreap)
1283                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1284         spin_unlock_irq(&p->sighand->siglock);
1285
1286         pid = p->pid;
1287         uid = p->uid;
1288         get_task_struct(p);
1289         read_unlock(&tasklist_lock);
1290
1291         if (!infop) {
1292                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1293                 put_task_struct(p);
1294                 if (!retval && stat_addr)
1295                         retval = put_user(0xffff, stat_addr);
1296                 if (!retval)
1297                         retval = p->pid;
1298         } else {
1299                 retval = wait_noreap_copyout(p, pid, uid,
1300                                              CLD_CONTINUED, SIGCONT,
1301                                              infop, ru);
1302                 BUG_ON(retval == 0);
1303         }
1304
1305         return retval;
1306 }
1307
1308
1309 static inline int my_ptrace_child(struct task_struct *p)
1310 {
1311         if (!(p->ptrace & PT_PTRACED))
1312                 return 0;
1313         if (!(p->ptrace & PT_ATTACHED))
1314                 return 1;
1315         /*
1316          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1317          * we are the attacher.  If we are the real parent, this is a race
1318          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1319          * which we have to switch the parent links, but has already set
1320          * the flags in p->ptrace.
1321          */
1322         return (p->parent != p->real_parent);
1323 }
1324
1325 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1326                     int __user *stat_addr, struct rusage __user *ru)
1327 {
1328         DECLARE_WAITQUEUE(wait, current);
1329         struct task_struct *tsk;
1330         int flag, retval;
1331
1332         add_wait_queue(&current->signal->wait_chldexit,&wait);
1333 repeat:
1334         /*
1335          * We will set this flag if we see any child that might later
1336          * match our criteria, even if we are not able to reap it yet.
1337          */
1338         flag = 0;
1339         current->state = TASK_INTERRUPTIBLE;
1340         read_lock(&tasklist_lock);
1341         tsk = current;
1342         do {
1343                 struct task_struct *p;
1344                 struct list_head *_p;
1345                 int ret;
1346
1347                 list_for_each(_p,&tsk->children) {
1348                         p = list_entry(_p,struct task_struct,sibling);
1349
1350                         ret = eligible_child(pid, options, p);
1351                         if (!ret)
1352                                 continue;
1353
1354                         switch (p->state) {
1355                         case TASK_TRACED:
1356                                 if (!my_ptrace_child(p))
1357                                         continue;
1358                                 /*FALLTHROUGH*/
1359                         case TASK_STOPPED:
1360                                 /*
1361                                  * It's stopped now, so it might later
1362                                  * continue, exit, or stop again.
1363                                  */
1364                                 flag = 1;
1365                                 if (!(options & WUNTRACED) &&
1366                                     !my_ptrace_child(p))
1367                                         continue;
1368                                 retval = wait_task_stopped(p, ret == 2,
1369                                                            (options & WNOWAIT),
1370                                                            infop,
1371                                                            stat_addr, ru);
1372                                 if (retval == -EAGAIN)
1373                                         goto repeat;
1374                                 if (retval != 0) /* He released the lock.  */
1375                                         goto end;
1376                                 break;
1377                         default:
1378                         // case EXIT_DEAD:
1379                                 if (p->exit_state == EXIT_DEAD)
1380                                         continue;
1381                         // case EXIT_ZOMBIE:
1382                                 if (p->exit_state == EXIT_ZOMBIE) {
1383                                         /*
1384                                          * Eligible but we cannot release
1385                                          * it yet:
1386                                          */
1387                                         if (ret == 2)
1388                                                 goto check_continued;
1389                                         if (!likely(options & WEXITED))
1390                                                 continue;
1391                                         retval = wait_task_zombie(
1392                                                 p, (options & WNOWAIT),
1393                                                 infop, stat_addr, ru);
1394                                         /* He released the lock.  */
1395                                         if (retval != 0)
1396                                                 goto end;
1397                                         break;
1398                                 }
1399 check_continued:
1400                                 /*
1401                                  * It's running now, so it might later
1402                                  * exit, stop, or stop and then continue.
1403                                  */
1404                                 flag = 1;
1405                                 if (!unlikely(options & WCONTINUED))
1406                                         continue;
1407                                 retval = wait_task_continued(
1408                                         p, (options & WNOWAIT),
1409                                         infop, stat_addr, ru);
1410                                 if (retval != 0) /* He released the lock.  */
1411                                         goto end;
1412                                 break;
1413                         }
1414                 }
1415                 if (!flag) {
1416                         list_for_each(_p, &tsk->ptrace_children) {
1417                                 p = list_entry(_p, struct task_struct,
1418                                                 ptrace_list);
1419                                 if (!eligible_child(pid, options, p))
1420                                         continue;
1421                                 flag = 1;
1422                                 break;
1423                         }
1424                 }
1425                 if (options & __WNOTHREAD)
1426                         break;
1427                 tsk = next_thread(tsk);
1428                 if (tsk->signal != current->signal)
1429                         BUG();
1430         } while (tsk != current);
1431
1432         read_unlock(&tasklist_lock);
1433         if (flag) {
1434                 retval = 0;
1435                 if (options & WNOHANG)
1436                         goto end;
1437                 retval = -ERESTARTSYS;
1438                 if (signal_pending(current))
1439                         goto end;
1440                 schedule();
1441                 goto repeat;
1442         }
1443         retval = -ECHILD;
1444 end:
1445         current->state = TASK_RUNNING;
1446         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1447         if (infop) {
1448                 if (retval > 0)
1449                 retval = 0;
1450                 else {
1451                         /*
1452                          * For a WNOHANG return, clear out all the fields
1453                          * we would set so the user can easily tell the
1454                          * difference.
1455                          */
1456                         if (!retval)
1457                                 retval = put_user(0, &infop->si_signo);
1458                         if (!retval)
1459                                 retval = put_user(0, &infop->si_errno);
1460                         if (!retval)
1461                                 retval = put_user(0, &infop->si_code);
1462                         if (!retval)
1463                                 retval = put_user(0, &infop->si_pid);
1464                         if (!retval)
1465                                 retval = put_user(0, &infop->si_uid);
1466                         if (!retval)
1467                                 retval = put_user(0, &infop->si_status);
1468                 }
1469         }
1470         return retval;
1471 }
1472
1473 asmlinkage long sys_waitid(int which, pid_t pid,
1474                            struct siginfo __user *infop, int options,
1475                            struct rusage __user *ru)
1476 {
1477         long ret;
1478
1479         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1480                 return -EINVAL;
1481         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1482                 return -EINVAL;
1483
1484         switch (which) {
1485         case P_ALL:
1486                 pid = -1;
1487                 break;
1488         case P_PID:
1489                 if (pid <= 0)
1490                         return -EINVAL;
1491                 break;
1492         case P_PGID:
1493                 if (pid <= 0)
1494                         return -EINVAL;
1495                 pid = -pid;
1496                 break;
1497         default:
1498                 return -EINVAL;
1499         }
1500
1501         ret = do_wait(pid, options, infop, NULL, ru);
1502
1503         /* avoid REGPARM breakage on x86: */
1504         prevent_tail_call(ret);
1505         return ret;
1506 }
1507
1508 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1509                           int options, struct rusage __user *ru)
1510 {
1511         long ret;
1512
1513         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1514                         __WNOTHREAD|__WCLONE|__WALL))
1515                 return -EINVAL;
1516         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1517
1518         /* avoid REGPARM breakage on x86: */
1519         prevent_tail_call(ret);
1520         return ret;
1521 }
1522
1523 #ifdef __ARCH_WANT_SYS_WAITPID
1524
1525 /*
1526  * sys_waitpid() remains for compatibility. waitpid() should be
1527  * implemented by calling sys_wait4() from libc.a.
1528  */
1529 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1530 {
1531         return sys_wait4(pid, stat_addr, options, NULL);
1532 }
1533
1534 #endif