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/ckrm_events.h>
  29 #include <linux/ckrm_tsk.h>
  30 #include <linux/ckrm_mem_inline.h>
  31 #include <linux/syscalls.h>
  32 #include <linux/vs_limit.h>
  33
  34 #include <asm/uaccess.h>
  35 #include <asm/unistd.h>
  36 #include <asm/pgtable.h>
  37 #include <asm/mmu_context.h>
  38
  39 extern void sem_exit (void);
  40 extern struct task_struct *child_reaper;
  41
  42 int getrusage(struct task_struct *, int, struct rusage __user *);
  43
  44 static void __unhash_process(struct task_struct *p)
  45 {
  46         nr_threads--;
  47         /* tasklist_lock is held, is this sufficient? */
  48         if (p->vx_info) {
  49                 atomic_dec(&p->vx_info->cvirt.nr_threads);
  50                 vx_nproc_dec(p);
  51         }
  52         detach_pid(p, PIDTYPE_PID);
  53         detach_pid(p, PIDTYPE_TGID);
  54         if (thread_group_leader(p)) {
  55                 detach_pid(p, PIDTYPE_PGID);
  56                 detach_pid(p, PIDTYPE_SID);
  57                 if (p->pid)
  58                         __get_cpu_var(process_counts)--;
  59         }
  60
  61         REMOVE_LINKS(p);
  62 }
  63
  64 void release_task(struct task_struct * p)
  65 {
  66         int zap_leader;
  67         task_t *leader;
  68         struct dentry *proc_dentry;
  69
  70 repeat:
  71         atomic_dec(&p->user->processes);
  72         spin_lock(&p->proc_lock);
  73         proc_dentry = proc_pid_unhash(p);
  74         write_lock_irq(&tasklist_lock);
  75         if (unlikely(p->ptrace))
  76                 __ptrace_unlink(p);
  77         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  78         __exit_signal(p);
  79         __exit_sighand(p);
  80         __unhash_process(p);
  81
  82         /*
  83          * If we are the last non-leader member of the thread
  84          * group, and the leader is zombie, then notify the
  85          * group leader's parent process. (if it wants notification.)
  86          */
  87         zap_leader = 0;
  88         leader = p->group_leader;
  89         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
  90                 BUG_ON(leader->exit_signal == -1);
  91                 do_notify_parent(leader, leader->exit_signal);
  92                 /*
  93                  * If we were the last child thread and the leader has
  94                  * exited already, and the leader's parent ignores SIGCHLD,
  95                  * then we are the one who should release the leader.
  96                  *
  97                  * do_notify_parent() will have marked it self-reaping in
  98                  * that case.
  99                  */
 100                 zap_leader = (leader->exit_signal == -1);
 101         }
 102
 103         sched_exit(p);
 104         write_unlock_irq(&tasklist_lock);
 105         spin_unlock(&p->proc_lock);
 106         proc_pid_flush(proc_dentry);
 107         release_thread(p);
 108         put_task_struct(p);
 109
 110         p = leader;
 111         if (unlikely(zap_leader))
 112                 goto repeat;
 113 }
 114
 115 /* we are using it only for SMP init */
 116
 117 void unhash_process(struct task_struct *p)
 118 {
 119         struct dentry *proc_dentry;
 120
 121         spin_lock(&p->proc_lock);
 122         proc_dentry = proc_pid_unhash(p);
 123         write_lock_irq(&tasklist_lock);
 124         __unhash_process(p);
 125         write_unlock_irq(&tasklist_lock);
 126         spin_unlock(&p->proc_lock);
 127         proc_pid_flush(proc_dentry);
 128 }
 129
 130 /*
 131  * This checks not only the pgrp, but falls back on the pid if no
 132  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 133  * without this...
 134  */
 135 int session_of_pgrp(int pgrp)
 136 {
 137         struct task_struct *p;
 138         int sid = -1;
 139
 140         read_lock(&tasklist_lock);
 141         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 142                 if (p->signal->session > 0) {
 143                         sid = p->signal->session;
 144                         goto out;
 145                 }
 146         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 147         p = find_task_by_pid(pgrp);
 148         if (p)
 149                 sid = p->signal->session;
 150 out:
 151         read_unlock(&tasklist_lock);
 152
 153         return sid;
 154 }
 155
 156 /*
 157  * Determine if a process group is "orphaned", according to the POSIX
 158  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 159  * by terminal-generated stop signals.  Newly orphaned process groups are
 160  * to receive a SIGHUP and a SIGCONT.
 161  *
 162  * "I ask you, have you ever known what it is to be an orphan?"
 163  */
 164 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 165 {
 166         struct task_struct *p;
 167         int ret = 1;
 168
 169         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 170                 if (p == ignored_task
 171                                 || p->exit_state >= EXIT_ZOMBIE
 172                                 || p->real_parent->pid == 1)
 173                         continue;
 174                 if (process_group(p->real_parent) != pgrp
 175                             && p->real_parent->signal->session == p->signal->session) {
 176                         ret = 0;
 177                         break;
 178                 }
 179         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 180         return ret;     /* (sighing) "Often!" */
 181 }
 182
 183 int is_orphaned_pgrp(int pgrp)
 184 {
 185         int retval;
 186
 187         read_lock(&tasklist_lock);
 188         retval = will_become_orphaned_pgrp(pgrp, NULL);
 189         read_unlock(&tasklist_lock);
 190
 191         return retval;
 192 }
 193
 194 static inline int has_stopped_jobs(int pgrp)
 195 {
 196         int retval = 0;
 197         struct task_struct *p;
 198
 199         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 200                 if (p->state != TASK_STOPPED)
 201                         continue;
 202
 203                 /* If p is stopped by a debugger on a signal that won't
 204                    stop it, then don't count p as stopped.  This isn't
 205                    perfect but it's a good approximation.  */
 206                 if (unlikely (p->ptrace)
 207                     && p->exit_code != SIGSTOP
 208                     && p->exit_code != SIGTSTP
 209                     && p->exit_code != SIGTTOU
 210                     && p->exit_code != SIGTTIN)
 211                         continue;
 212
 213                 retval = 1;
 214                 break;
 215         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 216         return retval;
 217 }
 218
 219 /**
 220  * reparent_to_init() - Reparent the calling kernel thread to the init task.
 221  *
 222  * If a kernel thread is launched as a result of a system call, or if
 223  * it ever exits, it should generally reparent itself to init so that
 224  * it is correctly cleaned up on exit.
 225  *
 226  * The various task state such as scheduling policy and priority may have
 227  * been inherited from a user process, so we reset them to sane values here.
 228  *
 229  * NOTE that reparent_to_init() gives the caller full capabilities.
 230  */
 231 void reparent_to_init(void)
 232 {
 233         write_lock_irq(&tasklist_lock);
 234
 235         ptrace_unlink(current);
 236         /* Reparent to init */
 237         REMOVE_LINKS(current);
 238         /* FIXME handle vchild_reaper/initpid */
 239         current->parent = child_reaper;
 240         current->real_parent = child_reaper;
 241         SET_LINKS(current);
 242
 243         /* Set the exit signal to SIGCHLD so we signal init on exit */
 244         current->exit_signal = SIGCHLD;
 245
 246         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
 247                 set_user_nice(current, 0);
 248         /* cpus_allowed? */
 249         /* rt_priority? */
 250         /* signals? */
 251         security_task_reparent_to_init(current);
 252         memcpy(current->signal->rlim, init_task.signal->rlim,
 253                sizeof(current->signal->rlim));
 254         atomic_inc(&(INIT_USER->__count));
 255         write_unlock_irq(&tasklist_lock);
 256         switch_uid(INIT_USER);
 257 }
 258
 259 void __set_special_pids(pid_t session, pid_t pgrp)
 260 {
 261         struct task_struct *curr = current;
 262
 263         if (curr->signal->session != session) {
 264                 detach_pid(curr, PIDTYPE_SID);
 265                 curr->signal->session = session;
 266                 attach_pid(curr, PIDTYPE_SID, session);
 267         }
 268         if (process_group(curr) != pgrp) {
 269                 detach_pid(curr, PIDTYPE_PGID);
 270                 curr->signal->pgrp = pgrp;
 271                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 272         }
 273 }
 274
 275 void set_special_pids(pid_t session, pid_t pgrp)
 276 {
 277         write_lock_irq(&tasklist_lock);
 278         __set_special_pids(session, pgrp);
 279         write_unlock_irq(&tasklist_lock);
 280 }
 281
 282 /*
 283  * Let kernel threads use this to say that they
 284  * allow a certain signal (since daemonize() will
 285  * have disabled all of them by default).
 286  */
 287 int allow_signal(int sig)
 288 {
 289         if (sig < 1 || sig > _NSIG)
 290                 return -EINVAL;
 291
 292         spin_lock_irq(&current->sighand->siglock);
 293         sigdelset(&current->blocked, sig);
 294         if (!current->mm) {
 295                 /* Kernel threads handle their own signals.
 296                    Let the signal code know it'll be handled, so
 297                    that they don't get converted to SIGKILL or
 298                    just silently dropped */
 299                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 300         }
 301         recalc_sigpending();
 302         spin_unlock_irq(&current->sighand->siglock);
 303         return 0;
 304 }
 305
 306 EXPORT_SYMBOL(allow_signal);
 307
 308 int disallow_signal(int sig)
 309 {
 310         if (sig < 1 || sig > _NSIG)
 311                 return -EINVAL;
 312
 313         spin_lock_irq(&current->sighand->siglock);
 314         sigaddset(&current->blocked, sig);
 315         recalc_sigpending();
 316         spin_unlock_irq(&current->sighand->siglock);
 317         return 0;
 318 }
 319
 320 EXPORT_SYMBOL(disallow_signal);
 321
 322 /*
 323  *      Put all the gunge required to become a kernel thread without
 324  *      attached user resources in one place where it belongs.
 325  */
 326
 327 void daemonize(const char *name, ...)
 328 {
 329         va_list args;
 330         struct fs_struct *fs;
 331         sigset_t blocked;
 332
 333         va_start(args, name);
 334         vsnprintf(current->comm, sizeof(current->comm), name, args);
 335         va_end(args);
 336
 337         /*
 338          * If we were started as result of loading a module, close all of the
 339          * user space pages.  We don't need them, and if we didn't close them
 340          * they would be locked into memory.
 341          */
 342         exit_mm(current);
 343
 344         set_special_pids(1, 1);
 345         down(&tty_sem);
 346         current->signal->tty = NULL;
 347         up(&tty_sem);
 348
 349         /* Block and flush all signals */
 350         sigfillset(&blocked);
 351         sigprocmask(SIG_BLOCK, &blocked, NULL);
 352         flush_signals(current);
 353
 354         /* Become as one with the init task */
 355
 356         exit_fs(current);       /* current->fs->count--; */
 357         fs = init_task.fs;
 358         current->fs = fs;
 359         atomic_inc(&fs->count);
 360         exit_files(current);
 361         current->files = init_task.files;
 362         atomic_inc(&current->files->count);
 363
 364         reparent_to_init();
 365 }
 366
 367 EXPORT_SYMBOL(daemonize);
 368
 369 static inline void close_files(struct files_struct * files)
 370 {
 371         int i, j;
 372
 373         j = 0;
 374         for (;;) {
 375                 unsigned long set;
 376                 i = j * __NFDBITS;
 377                 if (i >= files->max_fdset || i >= files->max_fds)
 378                         break;
 379                 set = files->open_fds->fds_bits[j++];
 380                 while (set) {
 381                         if (set & 1) {
 382                                 struct file * file = xchg(&files->fd[i], NULL);
 383                                 if (file)
 384                                         filp_close(file, files);
 385                                 // vx_openfd_dec(i);
 386                         }
 387                         i++;
 388                         set >>= 1;
 389                 }
 390         }
 391 }
 392
 393 struct files_struct *get_files_struct(struct task_struct *task)
 394 {
 395         struct files_struct *files;
 396
 397         task_lock(task);
 398         files = task->files;
 399         if (files)
 400                 atomic_inc(&files->count);
 401         task_unlock(task);
 402
 403         return files;
 404 }
 405
 406 void fastcall put_files_struct(struct files_struct *files)
 407 {
 408         if (atomic_dec_and_test(&files->count)) {
 409                 close_files(files);
 410                 /*
 411                  * Free the fd and fdset arrays if we expanded them.
 412                  */
 413                 if (files->fd != &files->fd_array[0])
 414                         free_fd_array(files->fd, files->max_fds);
 415                 if (files->max_fdset > __FD_SETSIZE) {
 416                         free_fdset(files->open_fds, files->max_fdset);
 417                         free_fdset(files->close_on_exec, files->max_fdset);
 418                 }
 419                 kmem_cache_free(files_cachep, files);
 420         }
 421 }
 422
 423 EXPORT_SYMBOL(put_files_struct);
 424
 425 static inline void __exit_files(struct task_struct *tsk)
 426 {
 427         struct files_struct * files = tsk->files;
 428
 429         if (files) {
 430                 task_lock(tsk);
 431                 tsk->files = NULL;
 432                 task_unlock(tsk);
 433                 put_files_struct(files);
 434         }
 435 }
 436
 437 void exit_files(struct task_struct *tsk)
 438 {
 439         __exit_files(tsk);
 440 }
 441
 442 static inline void __put_fs_struct(struct fs_struct *fs)
 443 {
 444         /* No need to hold fs->lock if we are killing it */
 445         if (atomic_dec_and_test(&fs->count)) {
 446                 dput(fs->root);
 447                 mntput(fs->rootmnt);
 448                 dput(fs->pwd);
 449                 mntput(fs->pwdmnt);
 450                 if (fs->altroot) {
 451                         dput(fs->altroot);
 452                         mntput(fs->altrootmnt);
 453                 }
 454                 kmem_cache_free(fs_cachep, fs);
 455         }
 456 }
 457
 458 void put_fs_struct(struct fs_struct *fs)
 459 {
 460         __put_fs_struct(fs);
 461 }
 462
 463 static inline void __exit_fs(struct task_struct *tsk)
 464 {
 465         struct fs_struct * fs = tsk->fs;
 466
 467         if (fs) {
 468                 task_lock(tsk);
 469                 tsk->fs = NULL;
 470                 task_unlock(tsk);
 471                 __put_fs_struct(fs);
 472         }
 473 }
 474
 475 void exit_fs(struct task_struct *tsk)
 476 {
 477         __exit_fs(tsk);
 478 }
 479
 480 EXPORT_SYMBOL_GPL(exit_fs);
 481
 482 /*
 483  * Turn us into a lazy TLB process if we
 484  * aren't already..
 485  */
 486 static inline void __exit_mm(struct task_struct * tsk)
 487 {
 488         struct mm_struct *mm = tsk->mm;
 489
 490         mm_release(tsk, mm);
 491         if (!mm)
 492                 return;
 493         /*
 494          * Serialize with any possible pending coredump.
 495          * We must hold mmap_sem around checking core_waiters
 496          * and clearing tsk->mm.  The core-inducing thread
 497          * will increment core_waiters for each thread in the
 498          * group with ->mm != NULL.
 499          */
 500         down_read(&mm->mmap_sem);
 501         if (mm->core_waiters) {
 502                 up_read(&mm->mmap_sem);
 503                 down_write(&mm->mmap_sem);
 504                 if (!--mm->core_waiters)
 505                         complete(mm->core_startup_done);
 506                 up_write(&mm->mmap_sem);
 507
 508                 wait_for_completion(&mm->core_done);
 509                 down_read(&mm->mmap_sem);
 510         }
 511         atomic_inc(&mm->mm_count);
 512         if (mm != tsk->active_mm) BUG();
 513         /* more a memory barrier than a real lock */
 514         task_lock(tsk);
 515         tsk->mm = NULL;
 516         up_read(&mm->mmap_sem);
 517         ckrm_task_mm_clear(tsk, mm);
 518         enter_lazy_tlb(mm, current);
 519         task_unlock(tsk);
 520         mmput(mm);
 521 }
 522
 523 void exit_mm(struct task_struct *tsk)
 524 {
 525         __exit_mm(tsk);
 526 }
 527
 528 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 529 {
 530         /*
 531          * Make sure we're not reparenting to ourselves and that
 532          * the parent is not a zombie.
 533          */
 534         BUG_ON(p == reaper || reaper->state >= EXIT_ZOMBIE || reaper->exit_state >= EXIT_ZOMBIE);
 535         p->real_parent = reaper;
 536         if (p->parent == p->real_parent)
 537                 BUG();
 538 }
 539
 540 static inline void reparent_thread(task_t *p, task_t *father, int traced)
 541 {
 542         /* We don't want people slaying init.  */
 543         if (p->exit_signal != -1)
 544                 p->exit_signal = SIGCHLD;
 545
 546         if (p->pdeath_signal)
 547                 /* We already hold the tasklist_lock here.  */
 548                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
 549
 550         /* Move the child from its dying parent to the new one.  */
 551         if (unlikely(traced)) {
 552                 /* Preserve ptrace links if someone else is tracing this child.  */
 553                 list_del_init(&p->ptrace_list);
 554                 if (p->parent != p->real_parent)
 555                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 556         } else {
 557                 /* If this child is being traced, then we're the one tracing it
 558                  * anyway, so let go of it.
 559                  */
 560                 p->ptrace = 0;
 561                 list_del_init(&p->sibling);
 562                 p->parent = p->real_parent;
 563                 list_add_tail(&p->sibling, &p->parent->children);
 564
 565                 /* If we'd notified the old parent about this child's death,
 566                  * also notify the new parent.
 567                  */
 568                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 569                     thread_group_empty(p))
 570                         do_notify_parent(p, p->exit_signal);
 571                 else if (p->state == TASK_TRACED) {
 572                         /*
 573                          * If it was at a trace stop, turn it into
 574                          * a normal stop since it's no longer being
 575                          * traced.
 576                          */
 577                         ptrace_untrace(p);
 578                 }
 579         }
 580
 581         /*
 582          * process group orphan check
 583          * Case ii: Our child is in a different pgrp
 584          * than we are, and it was the only connection
 585          * outside, so the child pgrp is now orphaned.
 586          */
 587         if ((process_group(p) != process_group(father)) &&
 588             (p->signal->session == father->signal->session)) {
 589                 int pgrp = process_group(p);
 590
 591                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 592                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
 593                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
 594                 }
 595         }
 596 }
 597
 598 /*
 599  * When we die, we re-parent all our children.
 600  * Try to give them to another thread in our thread
 601  * group, and if no such member exists, give it to
 602  * the global child reaper process (ie "init")
 603  */
 604 static inline void forget_original_parent(struct task_struct * father,
 605                                           struct list_head *to_release)
 606 {
 607         struct task_struct *p, *reaper = father;
 608         struct list_head *_p, *_n;
 609
 610         /* FIXME handle vchild_reaper/initpid */
 611         do {
 612                 reaper = next_thread(reaper);
 613                 if (reaper == father) {
 614                         reaper = child_reaper;
 615                         break;
 616                 }
 617         } while (reaper->exit_state >= EXIT_ZOMBIE);
 618
 619         /*
 620          * There are only two places where our children can be:
 621          *
 622          * - in our child list
 623          * - in our ptraced child list
 624          *
 625          * Search them and reparent children.
 626          */
 627         list_for_each_safe(_p, _n, &father->children) {
 628                 int ptrace;
 629                 p = list_entry(_p,struct task_struct,sibling);
 630
 631                 ptrace = p->ptrace;
 632
 633                 /* if father isn't the real parent, then ptrace must be enabled */
 634                 BUG_ON(father != p->real_parent && !ptrace);
 635
 636                 if (father == p->real_parent) {
 637                         /* reparent with a reaper, real father it's us */
 638                         choose_new_parent(p, reaper, child_reaper);
 639                         reparent_thread(p, father, 0);
 640                 } else {
 641                         /* reparent ptraced task to its real parent */
 642                         __ptrace_unlink (p);
 643                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 644                             thread_group_empty(p))
 645                                 do_notify_parent(p, p->exit_signal);
 646                 }
 647
 648                 /*
 649                  * if the ptraced child is a zombie with exit_signal == -1
 650                  * we must collect it before we exit, or it will remain
 651                  * zombie forever since we prevented it from self-reap itself
 652                  * while it was being traced by us, to be able to see it in wait4.
 653                  */
 654                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 655                         list_add(&p->ptrace_list, to_release);
 656         }
 657         list_for_each_safe(_p, _n, &father->ptrace_children) {
 658                 p = list_entry(_p,struct task_struct,ptrace_list);
 659                 choose_new_parent(p, reaper, child_reaper);
 660                 reparent_thread(p, father, 1);
 661         }
 662 }
 663
 664 /*
 665  * Send signals to all our closest relatives so that they know
 666  * to properly mourn us..
 667  */
 668 static void exit_notify(struct task_struct *tsk)
 669 {
 670         int state;
 671         struct task_struct *t;
 672         struct list_head ptrace_dead, *_p, *_n;
 673
 674         ckrm_cb_exit(tsk);
 675
 676         if (signal_pending(tsk) && !tsk->signal->group_exit
 677             && !thread_group_empty(tsk)) {
 678                 /*
 679                  * This occurs when there was a race between our exit
 680                  * syscall and a group signal choosing us as the one to
 681                  * wake up.  It could be that we are the only thread
 682                  * alerted to check for pending signals, but another thread
 683                  * should be woken now to take the signal since we will not.
 684                  * Now we'll wake all the threads in the group just to make
 685                  * sure someone gets all the pending signals.
 686                  */
 687                 read_lock(&tasklist_lock);
 688                 spin_lock_irq(&tsk->sighand->siglock);
 689                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 690                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 691                                 recalc_sigpending_tsk(t);
 692                                 if (signal_pending(t))
 693                                         signal_wake_up(t, 0);
 694                         }
 695                 spin_unlock_irq(&tsk->sighand->siglock);
 696                 read_unlock(&tasklist_lock);
 697         }
 698
 699         write_lock_irq(&tasklist_lock);
 700
 701         /*
 702          * This does two things:
 703          *
 704          * A.  Make init inherit all the child processes
 705          * B.  Check to see if any process groups have become orphaned
 706          *      as a result of our exiting, and if they have any stopped
 707          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 708          */
 709
 710         INIT_LIST_HEAD(&ptrace_dead);
 711         forget_original_parent(tsk, &ptrace_dead);
 712         BUG_ON(!list_empty(&tsk->children));
 713         BUG_ON(!list_empty(&tsk->ptrace_children));
 714
 715         /*
 716          * Check to see if any process groups have become orphaned
 717          * as a result of our exiting, and if they have any stopped
 718          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 719          *
 720          * Case i: Our father is in a different pgrp than we are
 721          * and we were the only connection outside, so our pgrp
 722          * is about to become orphaned.
 723          */
 724
 725         t = tsk->real_parent;
 726
 727         if ((process_group(t) != process_group(tsk)) &&
 728             (t->signal->session == tsk->signal->session) &&
 729             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 730             has_stopped_jobs(process_group(tsk))) {
 731                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
 732                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
 733         }
 734
 735         /* Let father know we died
 736          *
 737          * Thread signals are configurable, but you aren't going to use
 738          * that to send signals to arbitary processes.
 739          * That stops right now.
 740          *
 741          * If the parent exec id doesn't match the exec id we saved
 742          * when we started then we know the parent has changed security
 743          * domain.
 744          *
 745          * If our self_exec id doesn't match our parent_exec_id then
 746          * we have changed execution domain as these two values started
 747          * the same after a fork.
 748          *
 749          */
 750
 751         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 752             ( tsk->parent_exec_id != t->self_exec_id  ||
 753               tsk->self_exec_id != tsk->parent_exec_id)
 754             && !capable(CAP_KILL))
 755                 tsk->exit_signal = SIGCHLD;
 756
 757
 758         /* If something other than our normal parent is ptracing us, then
 759          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 760          * only has special meaning to our real parent.
 761          */
 762         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 763                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 764                 do_notify_parent(tsk, signal);
 765         } else if (tsk->ptrace) {
 766                 do_notify_parent(tsk, SIGCHLD);
 767         }
 768
 769         state = EXIT_ZOMBIE;
 770         if (tsk->exit_signal == -1 && tsk->ptrace == 0)
 771                 state = EXIT_DEAD;
 772         tsk->exit_state = state;
 773
 774         /*
 775          * Clear these here so that update_process_times() won't try to deliver
 776          * itimer, profile or rlimit signals to this task while it is in late exit.
 777          */
 778         tsk->it_virt_value = 0;
 779         tsk->it_prof_value = 0;
 780
 781         write_unlock_irq(&tasklist_lock);
 782
 783         list_for_each_safe(_p, _n, &ptrace_dead) {
 784                 list_del_init(_p);
 785                 t = list_entry(_p,struct task_struct,ptrace_list);
 786                 release_task(t);
 787         }
 788
 789         /* If the process is dead, release it - nobody will wait for it */
 790         if (state == EXIT_DEAD)
 791                 release_task(tsk);
 792
 793         /* PF_DEAD causes final put_task_struct after we schedule. */
 794         preempt_disable();
 795         tsk->flags |= PF_DEAD;
 796 }
 797
 798 fastcall NORET_TYPE void do_exit(long code)
 799 {
 800         struct task_struct *tsk = current;
 801         int group_dead;
 802
 803         profile_task_exit(tsk);
 804
 805         if (unlikely(in_interrupt()))
 806                 panic("Aiee, killing interrupt handler!");
 807         if (unlikely(!tsk->pid))
 808                 panic("Attempted to kill the idle task!");
 809         if (unlikely(tsk->pid == 1))
 810                 panic("Attempted to kill init!");
 811         if (tsk->io_context)
 812                 exit_io_context();
 813         tsk->flags |= PF_EXITING;
 814         del_timer_sync(&tsk->real_timer);
 815
 816         if (unlikely(in_atomic()))
 817                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 818                                 current->comm, current->pid,
 819                                 preempt_count());
 820
 821         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 822                 current->ptrace_message = code;
 823                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 824         }
 825
 826         group_dead = atomic_dec_and_test(&tsk->signal->live);
 827         if (group_dead)
 828                 acct_process(code);
 829         if (current->tux_info) {
 830 #ifdef CONFIG_TUX_DEBUG
 831                 printk("Possibly unexpected TUX-thread exit(%ld) at %p?\n",
 832                         code, __builtin_return_address(0));
 833 #endif
 834                 current->tux_exit();
 835         }
 836         __exit_mm(tsk);
 837
 838         exit_sem(tsk);
 839         __exit_files(tsk);
 840         __exit_fs(tsk);
 841         exit_namespace(tsk);
 842         exit_thread();
 843         exit_keys(tsk);
 844
 845         if (group_dead && tsk->signal->leader)
 846                 disassociate_ctty(1);
 847
 848         module_put(tsk->thread_info->exec_domain->module);
 849         if (tsk->binfmt)
 850                 module_put(tsk->binfmt->module);
 851
 852         tsk->exit_code = code;
 853         exit_notify(tsk);
 854 #ifdef CONFIG_NUMA
 855         mpol_free(tsk->mempolicy);
 856         tsk->mempolicy = NULL;
 857 #endif
 858
 859         BUG_ON(!(current->flags & PF_DEAD));
 860         schedule();
 861         BUG();
 862         /* Avoid "noreturn function does return".  */
 863         for (;;) ;
 864 }
 865
 866 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 867 {
 868         if (comp)
 869                 complete(comp);
 870
 871         do_exit(code);
 872 }
 873
 874 EXPORT_SYMBOL(complete_and_exit);
 875
 876 asmlinkage long sys_exit(int error_code)
 877 {
 878         do_exit((error_code&0xff)<<8);
 879 }
 880
 881 task_t fastcall *next_thread(const task_t *p)
 882 {
 883 #ifdef CONFIG_SMP
 884         if (!p->sighand)
 885                 BUG();
 886         if (!spin_is_locked(&p->sighand->siglock) &&
 887                                 !rwlock_is_locked(&tasklist_lock))
 888                 BUG();
 889 #endif
 890         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 891 }
 892
 893 EXPORT_SYMBOL(next_thread);
 894
 895 /*
 896  * Take down every thread in the group.  This is called by fatal signals
 897  * as well as by sys_exit_group (below).
 898  */
 899 NORET_TYPE void
 900 do_group_exit(int exit_code)
 901 {
 902         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 903
 904         if (current->signal->group_exit)
 905                 exit_code = current->signal->group_exit_code;
 906         else if (!thread_group_empty(current)) {
 907                 struct signal_struct *const sig = current->signal;
 908                 struct sighand_struct *const sighand = current->sighand;
 909                 read_lock(&tasklist_lock);
 910                 spin_lock_irq(&sighand->siglock);
 911                 if (sig->group_exit)
 912                         /* Another thread got here before we took the lock.  */
 913                         exit_code = sig->group_exit_code;
 914                 else {
 915                         sig->group_exit = 1;
 916                         sig->group_exit_code = exit_code;
 917                         zap_other_threads(current);
 918                 }
 919                 spin_unlock_irq(&sighand->siglock);
 920                 read_unlock(&tasklist_lock);
 921         }
 922
 923         do_exit(exit_code);
 924         /* NOTREACHED */
 925 }
 926
 927 /*
 928  * this kills every thread in the thread group. Note that any externally
 929  * wait4()-ing process will get the correct exit code - even if this
 930  * thread is not the thread group leader.
 931  */
 932 asmlinkage void sys_exit_group(int error_code)
 933 {
 934         do_group_exit((error_code & 0xff) << 8);
 935 }
 936
 937 static int eligible_child(pid_t pid, int options, task_t *p)
 938 {
 939         if (pid > 0) {
 940                 if (p->pid != pid)
 941                         return 0;
 942         } else if (!pid) {
 943                 if (process_group(p) != process_group(current))
 944                         return 0;
 945         } else if (pid != -1) {
 946                 if (process_group(p) != -pid)
 947                         return 0;
 948         }
 949
 950         /*
 951          * Do not consider detached threads that are
 952          * not ptraced:
 953          */
 954         if (p->exit_signal == -1 && !p->ptrace)
 955                 return 0;
 956
 957         /* Wait for all children (clone and not) if __WALL is set;
 958          * otherwise, wait for clone children *only* if __WCLONE is
 959          * set; otherwise, wait for non-clone children *only*.  (Note:
 960          * A "clone" child here is one that reports to its parent
 961          * using a signal other than SIGCHLD.) */
 962         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 963             && !(options & __WALL))
 964                 return 0;
 965         /*
 966          * Do not consider thread group leaders that are
 967          * in a non-empty thread group:
 968          */
 969         if (current->tgid != p->tgid && delay_group_leader(p))
 970                 return 2;
 971
 972         if (security_task_wait(p))
 973                 return 0;
 974
 975         return 1;
 976 }
 977
 978 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 979                                int why, int status,
 980                                struct siginfo __user *infop,
 981                                struct rusage __user *rusagep)
 982 {
 983         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
 984         put_task_struct(p);
 985         if (!retval)
 986                 retval = put_user(SIGCHLD, &infop->si_signo);
 987         if (!retval)
 988                 retval = put_user(0, &infop->si_errno);
 989         if (!retval)
 990                 retval = put_user((short)why, &infop->si_code);
 991         if (!retval)
 992                 retval = put_user(pid, &infop->si_pid);
 993         if (!retval)
 994                 retval = put_user(uid, &infop->si_uid);
 995         if (!retval)
 996                 retval = put_user(status, &infop->si_status);
 997         if (!retval)
 998                 retval = pid;
 999         return retval;
1000 }
1001
1002 /*
1003  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1004  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1005  * the lock and this task is uninteresting.  If we return nonzero, we have
1006  * released the lock and the system call should return.
1007  */
1008 static int wait_task_zombie(task_t *p, int noreap,
1009                             struct siginfo __user *infop,
1010                             int __user *stat_addr, struct rusage __user *ru)
1011 {
1012         unsigned long state;
1013         int retval;
1014         int status;
1015
1016         if (unlikely(noreap)) {
1017                 pid_t pid = p->pid;
1018                 uid_t uid = p->uid;
1019                 int exit_code = p->exit_code;
1020                 int why, status;
1021
1022                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1023                         return 0;
1024                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1025                         return 0;
1026                 get_task_struct(p);
1027                 read_unlock(&tasklist_lock);
1028                 if ((exit_code & 0x7f) == 0) {
1029                         why = CLD_EXITED;
1030                         status = exit_code >> 8;
1031                 } else {
1032                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1033                         status = exit_code & 0x7f;
1034                 }
1035                 return wait_noreap_copyout(p, pid, uid, why,
1036                                            status, infop, ru);
1037         }
1038
1039         /*
1040          * Try to move the task's state to DEAD
1041          * only one thread is allowed to do this:
1042          */
1043         state = xchg(&p->exit_state, EXIT_DEAD);
1044         if (state != EXIT_ZOMBIE) {
1045                 BUG_ON(state != EXIT_DEAD);
1046                 return 0;
1047         }
1048         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1049                 /*
1050                  * This can only happen in a race with a ptraced thread
1051                  * dying on another processor.
1052                  */
1053                 return 0;
1054         }
1055
1056         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1057                 /*
1058                  * The resource counters for the group leader are in its
1059                  * own task_struct.  Those for dead threads in the group
1060                  * are in its signal_struct, as are those for the child
1061                  * processes it has previously reaped.  All these
1062                  * accumulate in the parent's signal_struct c* fields.
1063                  *
1064                  * We don't bother to take a lock here to protect these
1065                  * p->signal fields, because they are only touched by
1066                  * __exit_signal, which runs with tasklist_lock
1067                  * write-locked anyway, and so is excluded here.  We do
1068                  * need to protect the access to p->parent->signal fields,
1069                  * as other threads in the parent group can be right
1070                  * here reaping other children at the same time.
1071                  */
1072                 spin_lock_irq(&p->parent->sighand->siglock);
1073                 p->parent->signal->cutime +=
1074                         p->utime + p->signal->utime + p->signal->cutime;
1075                 p->parent->signal->cstime +=
1076                         p->stime + p->signal->stime + p->signal->cstime;
1077                 p->parent->signal->cmin_flt +=
1078                         p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
1079                 p->parent->signal->cmaj_flt +=
1080                         p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
1081                 p->parent->signal->cnvcsw +=
1082                         p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
1083                 p->parent->signal->cnivcsw +=
1084                         p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
1085                 spin_unlock_irq(&p->parent->sighand->siglock);
1086         }
1087
1088         /*
1089          * Now we are sure this task is interesting, and no other
1090          * thread can reap it because we set its state to EXIT_DEAD.
1091          */
1092         read_unlock(&tasklist_lock);
1093
1094         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1095         status = p->signal->group_exit
1096                 ? p->signal->group_exit_code : p->exit_code;
1097         if (!retval && stat_addr)
1098                 retval = put_user(status, stat_addr);
1099         if (!retval && infop)
1100                 retval = put_user(SIGCHLD, &infop->si_signo);
1101         if (!retval && infop)
1102                 retval = put_user(0, &infop->si_errno);
1103         if (!retval && infop) {
1104                 int why;
1105
1106                 if ((status & 0x7f) == 0) {
1107                         why = CLD_EXITED;
1108                         status >>= 8;
1109                 } else {
1110                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1111                         status &= 0x7f;
1112                 }
1113                 retval = put_user((short)why, &infop->si_code);
1114                 if (!retval)
1115                         retval = put_user(status, &infop->si_status);
1116         }
1117         if (!retval && infop)
1118                 retval = put_user(p->pid, &infop->si_pid);
1119         if (!retval && infop)
1120                 retval = put_user(p->uid, &infop->si_uid);
1121         if (retval) {
1122                 // TODO: is this safe?
1123                 p->exit_state = EXIT_ZOMBIE;
1124                 return retval;
1125         }
1126         retval = p->pid;
1127         if (p->real_parent != p->parent) {
1128                 write_lock_irq(&tasklist_lock);
1129                 /* Double-check with lock held.  */
1130                 if (p->real_parent != p->parent) {
1131                         __ptrace_unlink(p);
1132                         // TODO: is this safe?
1133                         p->exit_state = EXIT_ZOMBIE;
1134                         /*
1135                          * If this is not a detached task, notify the parent.
1136                          * If it's still not detached after that, don't release
1137                          * it now.
1138                          */
1139                         if (p->exit_signal != -1) {
1140                                 do_notify_parent(p, p->exit_signal);
1141                                 if (p->exit_signal != -1)
1142                                         p = NULL;
1143                         }
1144                 }
1145                 write_unlock_irq(&tasklist_lock);
1146         }
1147         if (p != NULL)
1148                 release_task(p);
1149         BUG_ON(!retval);
1150         return retval;
1151 }
1152
1153 /*
1154  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1155  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1156  * the lock and this task is uninteresting.  If we return nonzero, we have
1157  * released the lock and the system call should return.
1158  */
1159 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1160                              struct siginfo __user *infop,
1161                              int __user *stat_addr, struct rusage __user *ru)
1162 {
1163         int retval, exit_code;
1164
1165         if (!p->exit_code)
1166                 return 0;
1167         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1168             p->signal && p->signal->group_stop_count > 0)
1169                 /*
1170                  * A group stop is in progress and this is the group leader.
1171                  * We won't report until all threads have stopped.
1172                  */
1173                 return 0;
1174
1175         /*
1176          * Now we are pretty sure this task is interesting.
1177          * Make sure it doesn't get reaped out from under us while we
1178          * give up the lock and then examine it below.  We don't want to
1179          * keep holding onto the tasklist_lock while we call getrusage and
1180          * possibly take page faults for user memory.
1181          */
1182         get_task_struct(p);
1183         read_unlock(&tasklist_lock);
1184
1185         if (unlikely(noreap)) {
1186                 pid_t pid = p->pid;
1187                 uid_t uid = p->uid;
1188                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1189
1190                 exit_code = p->exit_code;
1191                 if (unlikely(!exit_code) ||
1192                     unlikely(p->state > TASK_STOPPED))
1193                         goto bail_ref;
1194                 return wait_noreap_copyout(p, pid, uid,
1195                                            why, (exit_code << 8) | 0x7f,
1196                                            infop, ru);
1197         }
1198
1199         write_lock_irq(&tasklist_lock);
1200
1201         /*
1202          * This uses xchg to be atomic with the thread resuming and setting
1203          * it.  It must also be done with the write lock held to prevent a
1204          * race with the EXIT_ZOMBIE case.
1205          */
1206         exit_code = xchg(&p->exit_code, 0);
1207         if (unlikely(p->exit_state >= EXIT_ZOMBIE)) {
1208                 /*
1209                  * The task resumed and then died.  Let the next iteration
1210                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1211                  * already be zero here if it resumed and did _exit(0).
1212                  * The task itself is dead and won't touch exit_code again;
1213                  * other processors in this function are locked out.
1214                  */
1215                 p->exit_code = exit_code;
1216                 exit_code = 0;
1217         }
1218         if (unlikely(exit_code == 0)) {
1219                 /*
1220                  * Another thread in this function got to it first, or it
1221                  * resumed, or it resumed and then died.
1222                  */
1223                 write_unlock_irq(&tasklist_lock);
1224 bail_ref:
1225                 put_task_struct(p);
1226                 /*
1227                  * We are returning to the wait loop without having successfully
1228                  * removed the process and having released the lock. We cannot
1229                  * continue, since the "p" task pointer is potentially stale.
1230                  *
1231                  * Return -EAGAIN, and do_wait() will restart the loop from the
1232                  * beginning. Do _not_ re-acquire the lock.
1233                  */
1234                 return -EAGAIN;
1235         }
1236
1237         /* move to end of parent's list to avoid starvation */
1238         remove_parent(p);
1239         add_parent(p, p->parent);
1240
1241         write_unlock_irq(&tasklist_lock);
1242
1243         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1244         if (!retval && stat_addr)
1245                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1246         if (!retval && infop)
1247                 retval = put_user(SIGCHLD, &infop->si_signo);
1248         if (!retval && infop)
1249                 retval = put_user(0, &infop->si_errno);
1250         if (!retval && infop)
1251                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1252                                           ? CLD_TRAPPED : CLD_STOPPED),
1253                                   &infop->si_code);
1254         if (!retval && infop)
1255                 retval = put_user(exit_code, &infop->si_status);
1256         if (!retval && infop)
1257                 retval = put_user(p->pid, &infop->si_pid);
1258         if (!retval && infop)
1259                 retval = put_user(p->uid, &infop->si_uid);
1260         if (!retval)
1261                 retval = p->pid;
1262         put_task_struct(p);
1263
1264         BUG_ON(!retval);
1265         return retval;
1266 }
1267
1268 /*
1269  * Handle do_wait work for one task in a live, non-stopped state.
1270  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1271  * the lock and this task is uninteresting.  If we return nonzero, we have
1272  * released the lock and the system call should return.
1273  */
1274 static int wait_task_continued(task_t *p, int noreap,
1275                                struct siginfo __user *infop,
1276                                int __user *stat_addr, struct rusage __user *ru)
1277 {
1278         int retval;
1279         pid_t pid;
1280         uid_t uid;
1281
1282         if (unlikely(!p->signal))
1283                 return 0;
1284
1285         if (p->signal->stop_state >= 0)
1286                 return 0;
1287
1288         spin_lock_irq(&p->sighand->siglock);
1289         if (p->signal->stop_state >= 0) { /* Re-check with the lock held.  */
1290                 spin_unlock_irq(&p->sighand->siglock);
1291                 return 0;
1292         }
1293         if (!noreap)
1294                 p->signal->stop_state = 0;
1295         spin_unlock_irq(&p->sighand->siglock);
1296
1297         pid = p->pid;
1298         uid = p->uid;
1299         get_task_struct(p);
1300         read_unlock(&tasklist_lock);
1301
1302         if (!infop) {
1303                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1304                 put_task_struct(p);
1305                 if (!retval && stat_addr)
1306                         retval = put_user(0xffff, stat_addr);
1307                 if (!retval)
1308                         retval = p->pid;
1309         } else {
1310                 retval = wait_noreap_copyout(p, pid, uid,
1311                                              CLD_CONTINUED, SIGCONT,
1312                                              infop, ru);
1313                 BUG_ON(retval == 0);
1314         }
1315
1316         return retval;
1317 }
1318
1319
1320 static inline int my_ptrace_child(struct task_struct *p)
1321 {
1322         if (!(p->ptrace & PT_PTRACED))
1323                 return 0;
1324         if (!(p->ptrace & PT_ATTACHED))
1325                 return 1;
1326         /*
1327          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1328          * we are the attacher.  If we are the real parent, this is a race
1329          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1330          * which we have to switch the parent links, but has already set
1331          * the flags in p->ptrace.
1332          */
1333         return (p->parent != p->real_parent);
1334 }
1335
1336 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1337                     int __user *stat_addr, struct rusage __user *ru)
1338 {
1339         DECLARE_WAITQUEUE(wait, current);
1340         struct task_struct *tsk;
1341         int flag, retval;
1342
1343         add_wait_queue(&current->wait_chldexit,&wait);
1344 repeat:
1345         /*
1346          * We will set this flag if we see any child that might later
1347          * match our criteria, even if we are not able to reap it yet.
1348          */
1349         flag = 0;
1350         current->state = TASK_INTERRUPTIBLE;
1351         read_lock(&tasklist_lock);
1352         tsk = current;
1353         do {
1354                 struct task_struct *p;
1355                 struct list_head *_p;
1356                 int ret;
1357
1358                 list_for_each(_p,&tsk->children) {
1359                         p = list_entry(_p,struct task_struct,sibling);
1360
1361                         ret = eligible_child(pid, options, p);
1362                         if (!ret)
1363                                 continue;
1364
1365                         switch (p->state) {
1366                         case TASK_TRACED:
1367                                 if (!my_ptrace_child(p))
1368                                         continue;
1369                                 /*FALLTHROUGH*/
1370                         case TASK_STOPPED:
1371                                 /*
1372                                  * It's stopped now, so it might later
1373                                  * continue, exit, or stop again.
1374                                  */
1375                                 flag = 1;
1376                                 if (!(options & WUNTRACED) &&
1377                                     !my_ptrace_child(p))
1378                                         continue;
1379                                 retval = wait_task_stopped(p, ret == 2,
1380                                                            (options & WNOWAIT),
1381                                                            infop,
1382                                                            stat_addr, ru);
1383                                 if (retval == -EAGAIN)
1384                                         goto repeat;
1385                                 if (retval != 0) /* He released the lock.  */
1386                                         goto end;
1387                                 break;
1388                         default:
1389                         // case EXIT_DEAD:
1390                                 if (p->exit_state == EXIT_DEAD)
1391                                         continue;
1392                         // case EXIT_ZOMBIE:
1393                                 if (p->exit_state == EXIT_ZOMBIE) {
1394                                         /*
1395                                          * Eligible but we cannot release
1396                                          * it yet:
1397                                          */
1398                                         if (ret == 2)
1399                                                 goto check_continued;
1400                                         if (!likely(options & WEXITED))
1401                                                 continue;
1402                                         retval = wait_task_zombie(
1403                                                 p, (options & WNOWAIT),
1404                                                 infop, stat_addr, ru);
1405                                         /* He released the lock.  */
1406                                         if (retval != 0)
1407                                                 goto end;
1408                                         break;
1409                                 }
1410 check_continued:
1411                                 /*
1412                                  * It's running now, so it might later
1413                                  * exit, stop, or stop and then continue.
1414                                  */
1415                                 flag = 1;
1416                                 if (!unlikely(options & WCONTINUED))
1417                                         continue;
1418
1419                                 retval = wait_task_continued(
1420                                         p, (options & WNOWAIT),
1421                                         infop, stat_addr, ru);
1422                                 if (retval != 0) /* He released the lock.  */
1423                                         goto end;
1424                                 break;
1425                         }
1426                 }
1427                 if (!flag) {
1428                         list_for_each(_p, &tsk->ptrace_children) {
1429                                 p = list_entry(_p, struct task_struct,
1430                                                 ptrace_list);
1431                                 if (!eligible_child(pid, options, p))
1432                                         continue;
1433                                 flag = 1;
1434                                 break;
1435                         }
1436                 }
1437                 if (options & __WNOTHREAD)
1438                         break;
1439                 tsk = next_thread(tsk);
1440                 if (tsk->signal != current->signal)
1441                         BUG();
1442         } while (tsk != current);
1443
1444         read_unlock(&tasklist_lock);
1445         if (flag) {
1446                 retval = 0;
1447                 if (options & WNOHANG)
1448                         goto end;
1449                 retval = -ERESTARTSYS;
1450                 if (signal_pending(current))
1451                         goto end;
1452                 schedule();
1453                 goto repeat;
1454         }
1455         retval = -ECHILD;
1456 end:
1457         current->state = TASK_RUNNING;
1458         remove_wait_queue(&current->wait_chldexit,&wait);
1459         if (infop) {
1460                 if (retval > 0)
1461                 retval = 0;
1462                 else {
1463                         /*
1464                          * For a WNOHANG return, clear out all the fields
1465                          * we would set so the user can easily tell the
1466                          * difference.
1467                          */
1468                         if (!retval)
1469                                 retval = put_user(0, &infop->si_signo);
1470                         if (!retval)
1471                                 retval = put_user(0, &infop->si_errno);
1472                         if (!retval)
1473                                 retval = put_user(0, &infop->si_code);
1474                         if (!retval)
1475                                 retval = put_user(0, &infop->si_pid);
1476                         if (!retval)
1477                                 retval = put_user(0, &infop->si_uid);
1478                         if (!retval)
1479                                 retval = put_user(0, &infop->si_status);
1480                 }
1481         }
1482         return retval;
1483 }
1484
1485 asmlinkage long sys_waitid(int which, pid_t pid,
1486                            struct siginfo __user *infop, int options,
1487                            struct rusage __user *ru)
1488 {
1489         long ret;
1490
1491         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1492                 return -EINVAL;
1493         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1494                 return -EINVAL;
1495
1496         switch (which) {
1497         case P_ALL:
1498                 pid = -1;
1499                 break;
1500         case P_PID:
1501                 if (pid <= 0)
1502                         return -EINVAL;
1503                 break;
1504         case P_PGID:
1505                 if (pid <= 0)
1506                         return -EINVAL;
1507                 pid = -pid;
1508                 break;
1509         default:
1510                 return -EINVAL;
1511         }
1512
1513         ret = do_wait(pid, options, infop, NULL, ru);
1514
1515         /* avoid REGPARM breakage on x86: */
1516         prevent_tail_call(ret);
1517         return ret;
1518 }
1519
1520 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1521                           int options, struct rusage __user *ru)
1522 {
1523         long ret;
1524
1525         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1526                         __WNOTHREAD|__WCLONE|__WALL))
1527                 return -EINVAL;
1528         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1529
1530         /* avoid REGPARM breakage on x86: */
1531         prevent_tail_call(ret);
1532         return ret;
1533 }
1534
1535 #ifdef __ARCH_WANT_SYS_WAITPID
1536
1537 /*
1538  * sys_waitpid() remains for compatibility. waitpid() should be
1539  * implemented by calling sys_wait4() from libc.a.
1540  */
1541 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1542 {
1543         return sys_wait4(pid, stat_addr, options, NULL);
1544 }
1545
1546 #endif