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/security.h>
  18 #include <linux/cpu.h>
  19 #include <linux/acct.h>
  20 #include <linux/file.h>
  21 #include <linux/binfmts.h>
  22 #include <linux/ptrace.h>
  23 #include <linux/profile.h>
  24 #include <linux/mount.h>
  25 #include <linux/proc_fs.h>
  26 #include <linux/mempolicy.h>
  27 #include <linux/vs_limit.h>
  28
  29 #include <asm/uaccess.h>
  30 #include <asm/unistd.h>
  31 #include <asm/pgtable.h>
  32 #include <asm/mmu_context.h>
  33
  34 extern void sem_exit (void);
  35 extern struct task_struct *child_reaper;
  36
  37 int getrusage(struct task_struct *, int, struct rusage __user *);
  38
  39 static void __unhash_process(struct task_struct *p)
  40 {
  41         nr_threads--;
  42         /* tasklist_lock is held, is this sufficient? */
  43         if (p->vx_info) {
  44                 atomic_dec(&p->vx_info->cvirt.nr_threads);
  45                 vx_nproc_dec(p);
  46         }
  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->state == TASK_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->state >= TASK_ZOMBIE
 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 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         /* FIXME handle vchild_reaper/initpid */
 234         current->parent = child_reaper;
 235         current->real_parent = child_reaper;
 236         SET_LINKS(current);
 237
 238         /* Set the exit signal to SIGCHLD so we signal init on exit */
 239         current->exit_signal = SIGCHLD;
 240
 241         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
 242                 set_user_nice(current, 0);
 243         /* cpus_allowed? */
 244         /* rt_priority? */
 245         /* signals? */
 246         security_task_reparent_to_init(current);
 247         memcpy(current->rlim, init_task.rlim, sizeof(*(current->rlim)));
 248         atomic_inc(&(INIT_USER->__count));
 249         switch_uid(INIT_USER);
 250
 251         write_unlock_irq(&tasklist_lock);
 252 }
 253
 254 void __set_special_pids(pid_t session, pid_t pgrp)
 255 {
 256         struct task_struct *curr = current;
 257
 258         if (curr->signal->session != session) {
 259                 detach_pid(curr, PIDTYPE_SID);
 260                 curr->signal->session = session;
 261                 attach_pid(curr, PIDTYPE_SID, session);
 262         }
 263         if (process_group(curr) != pgrp) {
 264                 detach_pid(curr, PIDTYPE_PGID);
 265                 curr->signal->pgrp = pgrp;
 266                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 267         }
 268 }
 269
 270 void set_special_pids(pid_t session, pid_t pgrp)
 271 {
 272         write_lock_irq(&tasklist_lock);
 273         __set_special_pids(session, pgrp);
 274         write_unlock_irq(&tasklist_lock);
 275 }
 276
 277 /*
 278  * Let kernel threads use this to say that they
 279  * allow a certain signal (since daemonize() will
 280  * have disabled all of them by default).
 281  */
 282 int allow_signal(int sig)
 283 {
 284         if (sig < 1 || sig > _NSIG)
 285                 return -EINVAL;
 286
 287         spin_lock_irq(&current->sighand->siglock);
 288         sigdelset(&current->blocked, sig);
 289         if (!current->mm) {
 290                 /* Kernel threads handle their own signals.
 291                    Let the signal code know it'll be handled, so
 292                    that they don't get converted to SIGKILL or
 293                    just silently dropped */
 294                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 295         }
 296         recalc_sigpending();
 297         spin_unlock_irq(&current->sighand->siglock);
 298         return 0;
 299 }
 300
 301 EXPORT_SYMBOL(allow_signal);
 302
 303 int disallow_signal(int sig)
 304 {
 305         if (sig < 1 || sig > _NSIG)
 306                 return -EINVAL;
 307
 308         spin_lock_irq(&current->sighand->siglock);
 309         sigaddset(&current->blocked, sig);
 310         recalc_sigpending();
 311         spin_unlock_irq(&current->sighand->siglock);
 312         return 0;
 313 }
 314
 315 EXPORT_SYMBOL(disallow_signal);
 316
 317 /*
 318  *      Put all the gunge required to become a kernel thread without
 319  *      attached user resources in one place where it belongs.
 320  */
 321
 322 void daemonize(const char *name, ...)
 323 {
 324         va_list args;
 325         struct fs_struct *fs;
 326         sigset_t blocked;
 327
 328         va_start(args, name);
 329         vsnprintf(current->comm, sizeof(current->comm), name, args);
 330         va_end(args);
 331
 332         /*
 333          * If we were started as result of loading a module, close all of the
 334          * user space pages.  We don't need them, and if we didn't close them
 335          * they would be locked into memory.
 336          */
 337         exit_mm(current);
 338
 339         set_special_pids(1, 1);
 340         current->signal->tty = NULL;
 341
 342         /* Block and flush all signals */
 343         sigfillset(&blocked);
 344         sigprocmask(SIG_BLOCK, &blocked, NULL);
 345         flush_signals(current);
 346
 347         /* Become as one with the init task */
 348
 349         exit_fs(current);       /* current->fs->count--; */
 350         fs = init_task.fs;
 351         current->fs = fs;
 352         atomic_inc(&fs->count);
 353         exit_files(current);
 354         current->files = init_task.files;
 355         atomic_inc(&current->files->count);
 356
 357         reparent_to_init();
 358 }
 359
 360 EXPORT_SYMBOL(daemonize);
 361
 362 static inline void close_files(struct files_struct * files)
 363 {
 364         int i, j;
 365
 366         j = 0;
 367         for (;;) {
 368                 unsigned long set;
 369                 i = j * __NFDBITS;
 370                 if (i >= files->max_fdset || i >= files->max_fds)
 371                         break;
 372                 set = files->open_fds->fds_bits[j++];
 373                 while (set) {
 374                         if (set & 1) {
 375                                 struct file * file = xchg(&files->fd[i], NULL);
 376                                 if (file)
 377                                         filp_close(file, files);
 378                                 // vx_openfd_dec(i);
 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 inline 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 void exit_mm(struct task_struct *tsk)
 516 {
 517         __exit_mm(tsk);
 518 }
 519
 520 EXPORT_SYMBOL(exit_mm);
 521
 522 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 523 {
 524         /*
 525          * Make sure we're not reparenting to ourselves and that
 526          * the parent is not a zombie.
 527          */
 528         BUG_ON(p == reaper || reaper->state >= TASK_ZOMBIE);
 529         p->real_parent = reaper;
 530         if (p->parent == p->real_parent)
 531                 BUG();
 532 }
 533
 534 static inline void reparent_thread(task_t *p, task_t *father, int traced)
 535 {
 536         /* We don't want people slaying init.  */
 537         if (p->exit_signal != -1)
 538                 p->exit_signal = SIGCHLD;
 539         p->self_exec_id++;
 540
 541         if (p->pdeath_signal)
 542                 /* We already hold the tasklist_lock here.  */
 543                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
 544
 545         /* Move the child from its dying parent to the new one.  */
 546         if (unlikely(traced)) {
 547                 /* Preserve ptrace links if someone else is tracing this child.  */
 548                 list_del_init(&p->ptrace_list);
 549                 if (p->parent != p->real_parent)
 550                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 551         } else {
 552                 /* If this child is being traced, then we're the one tracing it
 553                  * anyway, so let go of it.
 554                  */
 555                 p->ptrace = 0;
 556                 list_del_init(&p->sibling);
 557                 p->parent = p->real_parent;
 558                 list_add_tail(&p->sibling, &p->parent->children);
 559
 560                 /* If we'd notified the old parent about this child's death,
 561                  * also notify the new parent.
 562                  */
 563                 if (p->state == TASK_ZOMBIE && p->exit_signal != -1 &&
 564                     thread_group_empty(p))
 565                         do_notify_parent(p, p->exit_signal);
 566                 else if (p->state == TASK_TRACED) {
 567                         /*
 568                          * If it was at a trace stop, turn it into
 569                          * a normal stop since it's no longer being
 570                          * traced.
 571                          */
 572                         p->state = TASK_STOPPED;
 573                 }
 574         }
 575
 576         /*
 577          * process group orphan check
 578          * Case ii: Our child is in a different pgrp
 579          * than we are, and it was the only connection
 580          * outside, so the child pgrp is now orphaned.
 581          */
 582         if ((process_group(p) != process_group(father)) &&
 583             (p->signal->session == father->signal->session)) {
 584                 int pgrp = process_group(p);
 585
 586                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 587                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
 588                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
 589                 }
 590         }
 591 }
 592
 593 /*
 594  * When we die, we re-parent all our children.
 595  * Try to give them to another thread in our thread
 596  * group, and if no such member exists, give it to
 597  * the global child reaper process (ie "init")
 598  */
 599 static inline void forget_original_parent(struct task_struct * father,
 600                                           struct list_head *to_release)
 601 {
 602         struct task_struct *p, *reaper = father;
 603         struct list_head *_p, *_n;
 604
 605         /* FIXME handle vchild_reaper/initpid */
 606         do {
 607                 reaper = next_thread(reaper);
 608                 if (reaper == father) {
 609                         reaper = child_reaper;
 610                         break;
 611                 }
 612         } while (reaper->state >= TASK_ZOMBIE);
 613
 614         /*
 615          * There are only two places where our children can be:
 616          *
 617          * - in our child list
 618          * - in our ptraced child list
 619          *
 620          * Search them and reparent children.
 621          */
 622         list_for_each_safe(_p, _n, &father->children) {
 623                 int ptrace;
 624                 p = list_entry(_p,struct task_struct,sibling);
 625
 626                 ptrace = p->ptrace;
 627
 628                 /* if father isn't the real parent, then ptrace must be enabled */
 629                 BUG_ON(father != p->real_parent && !ptrace);
 630
 631                 if (father == p->real_parent) {
 632                         /* reparent with a reaper, real father it's us */
 633                         choose_new_parent(p, reaper, child_reaper);
 634                         reparent_thread(p, father, 0);
 635                 } else {
 636                         /* reparent ptraced task to its real parent */
 637                         __ptrace_unlink (p);
 638                         if (p->state == TASK_ZOMBIE && p->exit_signal != -1 &&
 639                             thread_group_empty(p))
 640                                 do_notify_parent(p, p->exit_signal);
 641                 }
 642
 643                 /*
 644                  * if the ptraced child is a zombie with exit_signal == -1
 645                  * we must collect it before we exit, or it will remain
 646                  * zombie forever since we prevented it from self-reap itself
 647                  * while it was being traced by us, to be able to see it in wait4.
 648                  */
 649                 if (unlikely(ptrace && p->state == TASK_ZOMBIE && p->exit_signal == -1))
 650                         list_add(&p->ptrace_list, to_release);
 651         }
 652         list_for_each_safe(_p, _n, &father->ptrace_children) {
 653                 p = list_entry(_p,struct task_struct,ptrace_list);
 654                 choose_new_parent(p, reaper, child_reaper);
 655                 reparent_thread(p, father, 1);
 656         }
 657 }
 658
 659 /*
 660  * Send signals to all our closest relatives so that they know
 661  * to properly mourn us..
 662  */
 663 static void exit_notify(struct task_struct *tsk)
 664 {
 665         int state;
 666         struct task_struct *t;
 667         struct list_head ptrace_dead, *_p, *_n;
 668
 669         if (signal_pending(tsk) && !tsk->signal->group_exit
 670             && !thread_group_empty(tsk)) {
 671                 /*
 672                  * This occurs when there was a race between our exit
 673                  * syscall and a group signal choosing us as the one to
 674                  * wake up.  It could be that we are the only thread
 675                  * alerted to check for pending signals, but another thread
 676                  * should be woken now to take the signal since we will not.
 677                  * Now we'll wake all the threads in the group just to make
 678                  * sure someone gets all the pending signals.
 679                  */
 680                 read_lock(&tasklist_lock);
 681                 spin_lock_irq(&tsk->sighand->siglock);
 682                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 683                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 684                                 recalc_sigpending_tsk(t);
 685                                 if (signal_pending(t))
 686                                         signal_wake_up(t, 0);
 687                         }
 688                 spin_unlock_irq(&tsk->sighand->siglock);
 689                 read_unlock(&tasklist_lock);
 690         }
 691
 692         write_lock_irq(&tasklist_lock);
 693
 694         /*
 695          * This does two things:
 696          *
 697          * A.  Make init inherit all the child processes
 698          * B.  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
 703         INIT_LIST_HEAD(&ptrace_dead);
 704         forget_original_parent(tsk, &ptrace_dead);
 705         BUG_ON(!list_empty(&tsk->children));
 706         BUG_ON(!list_empty(&tsk->ptrace_children));
 707
 708         /*
 709          * Check to see if any process groups have become orphaned
 710          * as a result of our exiting, and if they have any stopped
 711          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 712          *
 713          * Case i: Our father is in a different pgrp than we are
 714          * and we were the only connection outside, so our pgrp
 715          * is about to become orphaned.
 716          */
 717
 718         t = tsk->real_parent;
 719
 720         if ((process_group(t) != process_group(tsk)) &&
 721             (t->signal->session == tsk->signal->session) &&
 722             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 723             has_stopped_jobs(process_group(tsk))) {
 724                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
 725                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
 726         }
 727
 728         /* Let father know we died
 729          *
 730          * Thread signals are configurable, but you aren't going to use
 731          * that to send signals to arbitary processes.
 732          * That stops right now.
 733          *
 734          * If the parent exec id doesn't match the exec id we saved
 735          * when we started then we know the parent has changed security
 736          * domain.
 737          *
 738          * If our self_exec id doesn't match our parent_exec_id then
 739          * we have changed execution domain as these two values started
 740          * the same after a fork.
 741          *
 742          */
 743
 744         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 745             ( tsk->parent_exec_id != t->self_exec_id  ||
 746               tsk->self_exec_id != tsk->parent_exec_id)
 747             && !capable(CAP_KILL))
 748                 tsk->exit_signal = SIGCHLD;
 749
 750
 751         /* If something other than our normal parent is ptracing us, then
 752          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 753          * only has special meaning to our real parent.
 754          */
 755         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 756                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 757                 do_notify_parent(tsk, signal);
 758         } else if (tsk->ptrace) {
 759                 do_notify_parent(tsk, SIGCHLD);
 760         }
 761
 762         state = TASK_ZOMBIE;
 763         if (tsk->exit_signal == -1 && tsk->ptrace == 0)
 764                 state = TASK_DEAD;
 765         tsk->state = state;
 766
 767         /*
 768          * Clear these here so that update_process_times() won't try to deliver
 769          * itimer, profile or rlimit signals to this task while it is in late exit.
 770          */
 771         tsk->it_virt_value = 0;
 772         tsk->it_prof_value = 0;
 773         tsk->rlim[RLIMIT_CPU].rlim_cur = RLIM_INFINITY;
 774
 775         write_unlock_irq(&tasklist_lock);
 776
 777         list_for_each_safe(_p, _n, &ptrace_dead) {
 778                 list_del_init(_p);
 779                 t = list_entry(_p,struct task_struct,ptrace_list);
 780                 release_task(t);
 781         }
 782
 783         /* If the process is dead, release it - nobody will wait for it */
 784         if (state == TASK_DEAD)
 785                 release_task(tsk);
 786
 787         /* PF_DEAD causes final put_task_struct after we schedule. */
 788         preempt_disable();
 789         tsk->flags |= PF_DEAD;
 790 }
 791
 792 asmlinkage NORET_TYPE void do_exit(long code)
 793 {
 794         struct task_struct *tsk = current;
 795
 796         profile_task_exit(tsk);
 797
 798         if (unlikely(in_interrupt()))
 799                 panic("Aiee, killing interrupt handler!");
 800         if (unlikely(!tsk->pid))
 801                 panic("Attempted to kill the idle task!");
 802         if (unlikely(tsk->pid == 1))
 803                 panic("Attempted to kill init!");
 804         if (tsk->io_context)
 805                 exit_io_context();
 806         tsk->flags |= PF_EXITING;
 807         del_timer_sync(&tsk->real_timer);
 808
 809         if (unlikely(in_atomic()))
 810                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 811                                 current->comm, current->pid,
 812                                 preempt_count());
 813
 814         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 815                 current->ptrace_message = code;
 816                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 817         }
 818
 819         acct_process(code);
 820         __exit_mm(tsk);
 821
 822         exit_sem(tsk);
 823         __exit_files(tsk);
 824         __exit_fs(tsk);
 825         exit_namespace(tsk);
 826         exit_thread();
 827
 828         if (tsk->signal->leader)
 829                 disassociate_ctty(1);
 830
 831         module_put(tsk->thread_info->exec_domain->module);
 832         if (tsk->binfmt)
 833                 module_put(tsk->binfmt->module);
 834
 835         tsk->exit_code = code;
 836         exit_notify(tsk);
 837 #ifdef CONFIG_NUMA
 838         mpol_free(tsk->mempolicy);
 839         tsk->mempolicy = NULL;
 840 #endif
 841         schedule();
 842         BUG();
 843         /* Avoid "noreturn function does return".  */
 844         for (;;) ;
 845 }
 846
 847 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 848 {
 849         if (comp)
 850                 complete(comp);
 851
 852         do_exit(code);
 853 }
 854
 855 EXPORT_SYMBOL(complete_and_exit);
 856
 857 asmlinkage long sys_exit(int error_code)
 858 {
 859         do_exit((error_code&0xff)<<8);
 860 }
 861
 862 task_t fastcall *next_thread(const task_t *p)
 863 {
 864 #ifdef CONFIG_SMP
 865         if (!p->sighand)
 866                 BUG();
 867         if (!spin_is_locked(&p->sighand->siglock) &&
 868                                 !rwlock_is_locked(&tasklist_lock))
 869                 BUG();
 870 #endif
 871         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 872 }
 873
 874 EXPORT_SYMBOL(next_thread);
 875
 876 /*
 877  * Take down every thread in the group.  This is called by fatal signals
 878  * as well as by sys_exit_group (below).
 879  */
 880 NORET_TYPE void
 881 do_group_exit(int exit_code)
 882 {
 883         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 884
 885         if (current->signal->group_exit)
 886                 exit_code = current->signal->group_exit_code;
 887         else if (!thread_group_empty(current)) {
 888                 struct signal_struct *const sig = current->signal;
 889                 struct sighand_struct *const sighand = current->sighand;
 890                 read_lock(&tasklist_lock);
 891                 spin_lock_irq(&sighand->siglock);
 892                 if (sig->group_exit)
 893                         /* Another thread got here before we took the lock.  */
 894                         exit_code = sig->group_exit_code;
 895                 else {
 896                         sig->group_exit = 1;
 897                         sig->group_exit_code = exit_code;
 898                         zap_other_threads(current);
 899                 }
 900                 spin_unlock_irq(&sighand->siglock);
 901                 read_unlock(&tasklist_lock);
 902         }
 903
 904         do_exit(exit_code);
 905         /* NOTREACHED */
 906 }
 907
 908 /*
 909  * this kills every thread in the thread group. Note that any externally
 910  * wait4()-ing process will get the correct exit code - even if this
 911  * thread is not the thread group leader.
 912  */
 913 asmlinkage void sys_exit_group(int error_code)
 914 {
 915         do_group_exit((error_code & 0xff) << 8);
 916 }
 917
 918 static int eligible_child(pid_t pid, int options, task_t *p)
 919 {
 920         if (pid > 0) {
 921                 if (p->pid != pid)
 922                         return 0;
 923         } else if (!pid) {
 924                 if (process_group(p) != process_group(current))
 925                         return 0;
 926         } else if (pid != -1) {
 927                 if (process_group(p) != -pid)
 928                         return 0;
 929         }
 930
 931         /*
 932          * Do not consider detached threads that are
 933          * not ptraced:
 934          */
 935         if (p->exit_signal == -1 && !p->ptrace)
 936                 return 0;
 937
 938         /* Wait for all children (clone and not) if __WALL is set;
 939          * otherwise, wait for clone children *only* if __WCLONE is
 940          * set; otherwise, wait for non-clone children *only*.  (Note:
 941          * A "clone" child here is one that reports to its parent
 942          * using a signal other than SIGCHLD.) */
 943         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 944             && !(options & __WALL))
 945                 return 0;
 946         /*
 947          * Do not consider thread group leaders that are
 948          * in a non-empty thread group:
 949          */
 950         if (current->tgid != p->tgid && delay_group_leader(p))
 951                 return 2;
 952
 953         if (security_task_wait(p))
 954                 return 0;
 955
 956         return 1;
 957 }
 958
 959 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 960                                int why, int status,
 961                                struct siginfo __user *infop,
 962                                struct rusage __user *rusagep)
 963 {
 964         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
 965         put_task_struct(p);
 966         if (!retval)
 967                 retval = put_user(SIGCHLD, &infop->si_signo);
 968         if (!retval)
 969                 retval = put_user(0, &infop->si_errno);
 970         if (!retval)
 971                 retval = put_user((short)why, &infop->si_code);
 972         if (!retval)
 973                 retval = put_user(pid, &infop->si_pid);
 974         if (!retval)
 975                 retval = put_user(uid, &infop->si_uid);
 976         if (!retval)
 977                 retval = put_user(status, &infop->si_status);
 978         if (!retval)
 979                 retval = pid;
 980         return retval;
 981 }
 982
 983 /*
 984  * Handle sys_wait4 work for one task in state TASK_ZOMBIE.  We hold
 985  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
 986  * the lock and this task is uninteresting.  If we return nonzero, we have
 987  * released the lock and the system call should return.
 988  */
 989 static int wait_task_zombie(task_t *p, int noreap,
 990                             struct siginfo __user *infop,
 991                             int __user *stat_addr, struct rusage __user *ru)
 992 {
 993         unsigned long state;
 994         int retval;
 995         int status;
 996
 997         if (unlikely(noreap)) {
 998                 pid_t pid = p->pid;
 999                 uid_t uid = p->uid;
1000                 int exit_code = p->exit_code;
1001                 int why, status;
1002
1003                 if (unlikely(p->state != TASK_ZOMBIE))
1004                         return 0;
1005                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1006                         return 0;
1007                 get_task_struct(p);
1008                 read_unlock(&tasklist_lock);
1009                 if ((exit_code & 0x7f) == 0) {
1010                         why = CLD_EXITED;
1011                         status = exit_code >> 8;
1012                 } else {
1013                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1014                         status = exit_code & 0x7f;
1015                 }
1016                 return wait_noreap_copyout(p, pid, uid, why,
1017                                            status, infop, ru);
1018         }
1019
1020         /*
1021          * Try to move the task's state to DEAD
1022          * only one thread is allowed to do this:
1023          */
1024         state = xchg(&p->state, TASK_DEAD);
1025         if (state != TASK_ZOMBIE) {
1026                 BUG_ON(state != TASK_DEAD);
1027                 return 0;
1028         }
1029         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1030                 /*
1031                  * This can only happen in a race with a ptraced thread
1032                  * dying on another processor.
1033                  */
1034                 return 0;
1035         }
1036
1037         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1038                 /*
1039                  * The resource counters for the group leader are in its
1040                  * own task_struct.  Those for dead threads in the group
1041                  * are in its signal_struct, as are those for the child
1042                  * processes it has previously reaped.  All these
1043                  * accumulate in the parent's signal_struct c* fields.
1044                  *
1045                  * We don't bother to take a lock here to protect these
1046                  * p->signal fields, because they are only touched by
1047                  * __exit_signal, which runs with tasklist_lock
1048                  * write-locked anyway, and so is excluded here.  We do
1049                  * need to protect the access to p->parent->signal fields,
1050                  * as other threads in the parent group can be right
1051                  * here reaping other children at the same time.
1052                  */
1053                 spin_lock_irq(&p->parent->sighand->siglock);
1054                 p->parent->signal->cutime +=
1055                         p->utime + p->signal->utime + p->signal->cutime;
1056                 p->parent->signal->cstime +=
1057                         p->stime + p->signal->stime + p->signal->cstime;
1058                 p->parent->signal->cmin_flt +=
1059                         p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
1060                 p->parent->signal->cmaj_flt +=
1061                         p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
1062                 p->parent->signal->cnvcsw +=
1063                         p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
1064                 p->parent->signal->cnivcsw +=
1065                         p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
1066                 spin_unlock_irq(&p->parent->sighand->siglock);
1067         }
1068
1069         /*
1070          * Now we are sure this task is interesting, and no other
1071          * thread can reap it because we set its state to TASK_DEAD.
1072          */
1073         read_unlock(&tasklist_lock);
1074
1075         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1076         status = p->signal->group_exit
1077                 ? p->signal->group_exit_code : p->exit_code;
1078         if (!retval && stat_addr)
1079                 retval = put_user(status, stat_addr);
1080         if (!retval && infop)
1081                 retval = put_user(SIGCHLD, &infop->si_signo);
1082         if (!retval && infop)
1083                 retval = put_user(0, &infop->si_errno);
1084         if (!retval && infop) {
1085                 int why;
1086
1087                 if ((status & 0x7f) == 0) {
1088                         why = CLD_EXITED;
1089                         status >>= 8;
1090                 } else {
1091                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1092                         status &= 0x7f;
1093                 }
1094                 retval = put_user((short)why, &infop->si_code);
1095                 if (!retval)
1096                         retval = put_user(status, &infop->si_status);
1097         }
1098         if (!retval && infop)
1099                 retval = put_user(p->pid, &infop->si_pid);
1100         if (!retval && infop)
1101                 retval = put_user(p->uid, &infop->si_uid);
1102         if (retval) {
1103                 p->state = TASK_ZOMBIE;
1104                 return retval;
1105         }
1106         retval = p->pid;
1107         if (p->real_parent != p->parent) {
1108                 write_lock_irq(&tasklist_lock);
1109                 /* Double-check with lock held.  */
1110                 if (p->real_parent != p->parent) {
1111                         __ptrace_unlink(p);
1112                         p->state = TASK_ZOMBIE;
1113                         /*
1114                          * If this is not a detached task, notify the parent.
1115                          * If it's still not detached after that, don't release
1116                          * it now.
1117                          */
1118                         if (p->exit_signal != -1) {
1119                                 do_notify_parent(p, p->exit_signal);
1120                                 if (p->exit_signal != -1)
1121                                         p = NULL;
1122                         }
1123                 }
1124                 write_unlock_irq(&tasklist_lock);
1125         }
1126         if (p != NULL)
1127                 release_task(p);
1128         BUG_ON(!retval);
1129         return retval;
1130 }
1131
1132 /*
1133  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1134  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1135  * the lock and this task is uninteresting.  If we return nonzero, we have
1136  * released the lock and the system call should return.
1137  */
1138 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1139                              struct siginfo __user *infop,
1140                              int __user *stat_addr, struct rusage __user *ru)
1141 {
1142         int retval, exit_code;
1143
1144         if (!p->exit_code)
1145                 return 0;
1146         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1147             p->signal && p->signal->group_stop_count > 0)
1148                 /*
1149                  * A group stop is in progress and this is the group leader.
1150                  * We won't report until all threads have stopped.
1151                  */
1152                 return 0;
1153
1154         /*
1155          * Now we are pretty sure this task is interesting.
1156          * Make sure it doesn't get reaped out from under us while we
1157          * give up the lock and then examine it below.  We don't want to
1158          * keep holding onto the tasklist_lock while we call getrusage and
1159          * possibly take page faults for user memory.
1160          */
1161         get_task_struct(p);
1162         read_unlock(&tasklist_lock);
1163
1164         if (unlikely(noreap)) {
1165                 pid_t pid = p->pid;
1166                 uid_t uid = p->uid;
1167                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1168
1169                 exit_code = p->exit_code;
1170                 if (unlikely(!exit_code) ||
1171                     unlikely(p->state > TASK_STOPPED))
1172                         goto bail_ref;
1173                 return wait_noreap_copyout(p, pid, uid,
1174                                            why, (exit_code << 8) | 0x7f,
1175                                            infop, ru);
1176         }
1177
1178         write_lock_irq(&tasklist_lock);
1179
1180         /*
1181          * This uses xchg to be atomic with the thread resuming and setting
1182          * it.  It must also be done with the write lock held to prevent a
1183          * race with the TASK_ZOMBIE case.
1184          */
1185         exit_code = xchg(&p->exit_code, 0);
1186         if (unlikely(p->state >= TASK_ZOMBIE)) {
1187                 /*
1188                  * The task resumed and then died.  Let the next iteration
1189                  * catch it in TASK_ZOMBIE.  Note that exit_code might
1190                  * already be zero here if it resumed and did _exit(0).
1191                  * The task itself is dead and won't touch exit_code again;
1192                  * other processors in this function are locked out.
1193                  */
1194                 p->exit_code = exit_code;
1195                 exit_code = 0;
1196         }
1197         if (unlikely(exit_code == 0)) {
1198                 /*
1199                  * Another thread in this function got to it first, or it
1200                  * resumed, or it resumed and then died.
1201                  */
1202                 write_unlock_irq(&tasklist_lock);
1203 bail_ref:
1204                 put_task_struct(p);
1205                 read_lock(&tasklist_lock);
1206                 return 0;
1207         }
1208
1209         /* move to end of parent's list to avoid starvation */
1210         remove_parent(p);
1211         add_parent(p, p->parent);
1212
1213         write_unlock_irq(&tasklist_lock);
1214
1215         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1216         if (!retval && stat_addr)
1217                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1218         if (!retval && infop)
1219                 retval = put_user(SIGCHLD, &infop->si_signo);
1220         if (!retval && infop)
1221                 retval = put_user(0, &infop->si_errno);
1222         if (!retval && infop)
1223                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1224                                           ? CLD_TRAPPED : CLD_STOPPED),
1225                                   &infop->si_code);
1226         if (!retval && infop)
1227                 retval = put_user(exit_code, &infop->si_status);
1228         if (!retval && infop)
1229                 retval = put_user(p->pid, &infop->si_pid);
1230         if (!retval && infop)
1231                 retval = put_user(p->uid, &infop->si_uid);
1232         if (!retval)
1233                 retval = p->pid;
1234         put_task_struct(p);
1235
1236         BUG_ON(!retval);
1237         return retval;
1238 }
1239
1240 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1241                     int __user *stat_addr, struct rusage __user *ru)
1242 {
1243         DECLARE_WAITQUEUE(wait, current);
1244         struct task_struct *tsk;
1245         int flag, retval;
1246
1247         add_wait_queue(&current->wait_chldexit,&wait);
1248 repeat:
1249         flag = 0;
1250         current->state = TASK_INTERRUPTIBLE;
1251         read_lock(&tasklist_lock);
1252         tsk = current;
1253         do {
1254                 struct task_struct *p;
1255                 struct list_head *_p;
1256                 int ret;
1257
1258                 list_for_each(_p,&tsk->children) {
1259                         p = list_entry(_p,struct task_struct,sibling);
1260
1261                         ret = eligible_child(pid, options, p);
1262                         if (!ret)
1263                                 continue;
1264                         flag = 1;
1265
1266                         switch (p->state) {
1267                         case TASK_TRACED:
1268                                 if (!(p->ptrace & PT_PTRACED))
1269                                         continue;
1270                                 /*FALLTHROUGH*/
1271                         case TASK_STOPPED:
1272                                 if (!(options & WUNTRACED) &&
1273                                     !(p->ptrace & PT_PTRACED))
1274                                         continue;
1275                                 retval = wait_task_stopped(p, ret == 2,
1276                                                            (options & WNOWAIT),
1277                                                            infop,
1278                                                            stat_addr, ru);
1279                                 if (retval != 0) /* He released the lock.  */
1280                                         goto end;
1281                                 break;
1282                         case TASK_ZOMBIE:
1283                                 /*
1284                                  * Eligible but we cannot release it yet:
1285                                  */
1286                                 if (ret == 2)
1287                                         goto check_continued;
1288                                 if (!likely(options & WEXITED))
1289                                         continue;
1290                                 retval = wait_task_zombie(
1291                                         p, (options & WNOWAIT),
1292                                         infop, stat_addr, ru);
1293                                 if (retval != 0) /* He released the lock.  */
1294                                         goto end;
1295                                 break;
1296                         case TASK_DEAD:
1297                                 continue;
1298                         default:
1299 check_continued:
1300                                 if (!unlikely(options & WCONTINUED))
1301                                         continue;
1302                                 if (unlikely(!p->signal))
1303                                         continue;
1304                                 spin_lock_irq(&p->sighand->siglock);
1305                                 if (p->signal->stop_state < 0) {
1306                                         pid_t pid;
1307                                         uid_t uid;
1308
1309                                         if (!(options & WNOWAIT))
1310                                                 p->signal->stop_state = 0;
1311                                         spin_unlock_irq(&p->sighand->siglock);
1312                                         pid = p->pid;
1313                                         uid = p->uid;
1314                                         get_task_struct(p);
1315                                         read_unlock(&tasklist_lock);
1316                                         retval = wait_noreap_copyout(p, pid,
1317                                                         uid, CLD_CONTINUED,
1318                                                         SIGCONT, infop, ru);
1319                                         BUG_ON(retval == 0);
1320                                         goto end;
1321                                 }
1322                                 spin_unlock_irq(&p->sighand->siglock);
1323                                 break;
1324                         }
1325                 }
1326                 if (!flag) {
1327                         list_for_each(_p, &tsk->ptrace_children) {
1328                                 p = list_entry(_p, struct task_struct,
1329                                                 ptrace_list);
1330                                 if (!eligible_child(pid, options, p))
1331                                         continue;
1332                                 flag = 1;
1333                                 break;
1334                         }
1335                 }
1336                 if (options & __WNOTHREAD)
1337                         break;
1338                 tsk = next_thread(tsk);
1339                 if (tsk->signal != current->signal)
1340                         BUG();
1341         } while (tsk != current);
1342
1343         read_unlock(&tasklist_lock);
1344         if (flag) {
1345                 retval = 0;
1346                 if (options & WNOHANG)
1347                         goto end;
1348                 retval = -ERESTARTSYS;
1349                 if (signal_pending(current))
1350                         goto end;
1351                 schedule();
1352                 goto repeat;
1353         }
1354         retval = -ECHILD;
1355 end:
1356         current->state = TASK_RUNNING;
1357         remove_wait_queue(&current->wait_chldexit,&wait);
1358         if (infop) {
1359                 if (retval > 0)
1360                 retval = 0;
1361                 else {
1362                         /*
1363                          * For a WNOHANG return, clear out all the fields
1364                          * we would set so the user can easily tell the
1365                          * difference.
1366                          */
1367                         if (!retval)
1368                                 retval = put_user(0, &infop->si_signo);
1369                         if (!retval)
1370                                 retval = put_user(0, &infop->si_errno);
1371                         if (!retval)
1372                                 retval = put_user(0, &infop->si_code);
1373                         if (!retval)
1374                                 retval = put_user(0, &infop->si_pid);
1375                         if (!retval)
1376                                 retval = put_user(0, &infop->si_uid);
1377                         if (!retval)
1378                                 retval = put_user(0, &infop->si_status);
1379                 }
1380         }
1381         return retval;
1382 }
1383
1384 asmlinkage long sys_waitid(int which, pid_t pid,
1385                            struct siginfo __user *infop, int options,
1386                            struct rusage __user *ru)
1387 {
1388         long ret;
1389
1390         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1391                 return -EINVAL;
1392         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1393                 return -EINVAL;
1394
1395         switch (which) {
1396         case P_ALL:
1397                 pid = -1;
1398                 break;
1399         case P_PID:
1400                 if (pid <= 0)
1401                         return -EINVAL;
1402                 break;
1403         case P_PGID:
1404                 if (pid <= 0)
1405                         return -EINVAL;
1406                 pid = -pid;
1407                 break;
1408         default:
1409                 return -EINVAL;
1410         }
1411
1412         ret = do_wait(pid, options, infop, NULL, ru);
1413
1414         /* avoid REGPARM breakage on x86: */
1415         prevent_tail_call(ret);
1416         return ret;
1417 }
1418
1419 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1420                           int options, struct rusage __user *ru)
1421 {
1422         long ret;
1423
1424         if (options & ~(WNOHANG|WUNTRACED|__WNOTHREAD|__WCLONE|__WALL))
1425                 return -EINVAL;
1426         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1427
1428         /* avoid REGPARM breakage on x86: */
1429         prevent_tail_call(ret);
1430         return ret;
1431 }
1432
1433 #ifdef __ARCH_WANT_SYS_WAITPID
1434
1435 /*
1436  * sys_waitpid() remains for compatibility. waitpid() should be
1437  * implemented by calling sys_wait4() from libc.a.
1438  */
1439 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1440 {
1441         return sys_wait4(pid, stat_addr, options, NULL);
1442 }
1443
1444 #endif