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