kernel/exit.c

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