kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/config.h>
   8 #include <linux/mm.h>
   9 #include <linux/slab.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/module.h>
  13 #include <linux/completion.h>
  14 #include <linux/personality.h>
  15 #include <linux/tty.h>
  16 #include <linux/namespace.h>
  17 #include <linux/key.h>
  18 #include <linux/security.h>
  19 #include <linux/cpu.h>
  20 #include <linux/acct.h>
  21 #include <linux/file.h>
  22 #include <linux/binfmts.h>
  23 #include <linux/ptrace.h>
  24 #include <linux/profile.h>
  25 #include <linux/mount.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/mempolicy.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/vs_limit.h>
  30 #include <linux/vs_network.h>
  31
  32 #include <asm/uaccess.h>
  33 #include <asm/unistd.h>
  34 #include <asm/pgtable.h>
  35 #include <asm/mmu_context.h>
  36
  37 extern void sem_exit (void);
  38 extern struct task_struct *child_reaper;
  39
  40 int getrusage(struct task_struct *, int, struct rusage __user *);
  41
  42 static void __unhash_process(struct task_struct *p)
  43 {
  44         nr_threads--;
  45         /* tasklist_lock is held, is this sufficient? */
  46         if (p->vx_info) {
  47                 atomic_dec(&p->vx_info->cvirt.nr_threads);
  48                 vx_nproc_dec(p);
  49         }
  50         detach_pid(p, PIDTYPE_PID);
  51         detach_pid(p, PIDTYPE_TGID);
  52         if (thread_group_leader(p)) {
  53                 detach_pid(p, PIDTYPE_PGID);
  54                 detach_pid(p, PIDTYPE_SID);
  55                 if (p->pid)
  56                         __get_cpu_var(process_counts)--;
  57         }
  58
  59         REMOVE_LINKS(p);
  60 }
  61
  62 void release_task(struct task_struct * p)
  63 {
  64         int zap_leader;
  65         task_t *leader;
  66         struct dentry *proc_dentry;
  67
  68 repeat:
  69         atomic_dec(&p->user->processes);
  70         spin_lock(&p->proc_lock);
  71         proc_dentry = proc_pid_unhash(p);
  72         write_lock_irq(&tasklist_lock);
  73         if (unlikely(p->ptrace))
  74                 __ptrace_unlink(p);
  75         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  76         __exit_signal(p);
  77         __exit_sighand(p);
  78         __unhash_process(p);
  79
  80         /*
  81          * If we are the last non-leader member of the thread
  82          * group, and the leader is zombie, then notify the
  83          * group leader's parent process. (if it wants notification.)
  84          */
  85         zap_leader = 0;
  86         leader = p->group_leader;
  87         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
  88                 BUG_ON(leader->exit_signal == -1);
  89                 do_notify_parent(leader, leader->exit_signal);
  90                 /*
  91                  * If we were the last child thread and the leader has
  92                  * exited already, and the leader's parent ignores SIGCHLD,
  93                  * then we are the one who should release the leader.
  94                  *
  95                  * do_notify_parent() will have marked it self-reaping in
  96                  * that case.
  97                  */
  98                 zap_leader = (leader->exit_signal == -1);
  99         }
 100
 101         sched_exit(p);
 102         write_unlock_irq(&tasklist_lock);
 103         spin_unlock(&p->proc_lock);
 104         proc_pid_flush(proc_dentry);
 105         release_thread(p);
 106         if (p->vx_info)
 107                 release_vx_info(p->vx_info, p);
 108         if (p->nx_info)
 109                 release_nx_info(p->nx_info, p);
 110         put_task_struct(p);
 111
 112         p = leader;
 113         if (unlikely(zap_leader))
 114                 goto repeat;
 115 }
 116
 117 /* we are using it only for SMP init */
 118
 119 void unhash_process(struct task_struct *p)
 120 {
 121         struct dentry *proc_dentry;
 122
 123         spin_lock(&p->proc_lock);
 124         proc_dentry = proc_pid_unhash(p);
 125         write_lock_irq(&tasklist_lock);
 126         __unhash_process(p);
 127         write_unlock_irq(&tasklist_lock);
 128         spin_unlock(&p->proc_lock);
 129         proc_pid_flush(proc_dentry);
 130 }
 131
 132 /*
 133  * This checks not only the pgrp, but falls back on the pid if no
 134  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 135  * without this...
 136  */
 137 int session_of_pgrp(int pgrp)
 138 {
 139         struct task_struct *p;
 140         int sid = -1;
 141
 142         read_lock(&tasklist_lock);
 143         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 144                 if (p->signal->session > 0) {
 145                         sid = p->signal->session;
 146                         goto out;
 147                 }
 148         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 149         p = find_task_by_pid(pgrp);
 150         if (p)
 151                 sid = p->signal->session;
 152 out:
 153         read_unlock(&tasklist_lock);
 154
 155         return sid;
 156 }
 157
 158 /*
 159  * Determine if a process group is "orphaned", according to the POSIX
 160  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 161  * by terminal-generated stop signals.  Newly orphaned process groups are
 162  * to receive a SIGHUP and a SIGCONT.
 163  *
 164  * "I ask you, have you ever known what it is to be an orphan?"
 165  */
 166 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 167 {
 168         struct task_struct *p;
 169         int ret = 1;
 170
 171         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 172                 if (p == ignored_task
 173                                 || p->exit_state
 174                                 || p->real_parent->pid == 1)
 175                         continue;
 176                 if (process_group(p->real_parent) != pgrp
 177                             && p->real_parent->signal->session == p->signal->session) {
 178                         ret = 0;
 179                         break;
 180                 }
 181         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 182         return ret;     /* (sighing) "Often!" */
 183 }
 184
 185 int is_orphaned_pgrp(int pgrp)
 186 {
 187         int retval;
 188
 189         read_lock(&tasklist_lock);
 190         retval = will_become_orphaned_pgrp(pgrp, NULL);
 191         read_unlock(&tasklist_lock);
 192
 193         return retval;
 194 }
 195
 196 static inline int has_stopped_jobs(int pgrp)
 197 {
 198         int retval = 0;
 199         struct task_struct *p;
 200
 201         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 202                 if (p->state != TASK_STOPPED)
 203                         continue;
 204
 205                 /* If p is stopped by a debugger on a signal that won't
 206                    stop it, then don't count p as stopped.  This isn't
 207                    perfect but it's a good approximation.  */
 208                 if (unlikely (p->ptrace)
 209                     && p->exit_code != SIGSTOP
 210                     && p->exit_code != SIGTSTP
 211                     && p->exit_code != SIGTTOU
 212                     && p->exit_code != SIGTTIN)
 213                         continue;
 214
 215                 retval = 1;
 216                 break;
 217         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 218         return retval;
 219 }
 220
 221 /**
 222  * reparent_to_init() - Reparent the calling kernel thread to the init task.
 223  *
 224  * If a kernel thread is launched as a result of a system call, or if
 225  * it ever exits, it should generally reparent itself to init so that
 226  * it is correctly cleaned up on exit.
 227  *
 228  * The various task state such as scheduling policy and priority may have
 229  * been inherited from a user process, so we reset them to sane values here.
 230  *
 231  * NOTE that reparent_to_init() gives the caller full capabilities.
 232  */
 233 void reparent_to_init(void)
 234 {
 235         write_lock_irq(&tasklist_lock);
 236
 237         ptrace_unlink(current);
 238         /* Reparent to init */
 239         REMOVE_LINKS(current);
 240         /* FIXME handle vchild_reaper/initpid */
 241         current->parent = child_reaper;
 242         current->real_parent = child_reaper;
 243         SET_LINKS(current);
 244
 245         /* Set the exit signal to SIGCHLD so we signal init on exit */
 246         current->exit_signal = SIGCHLD;
 247
 248         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
 249                 set_user_nice(current, 0);
 250         /* cpus_allowed? */
 251         /* rt_priority? */
 252         /* signals? */
 253         security_task_reparent_to_init(current);
 254         memcpy(current->signal->rlim, init_task.signal->rlim,
 255                sizeof(current->signal->rlim));
 256         atomic_inc(&(INIT_USER->__count));
 257         write_unlock_irq(&tasklist_lock);
 258         switch_uid(INIT_USER);
 259 }
 260
 261 void __set_special_pids(pid_t session, pid_t pgrp)
 262 {
 263         struct task_struct *curr = current;
 264
 265         if (curr->signal->session != session) {
 266                 detach_pid(curr, PIDTYPE_SID);
 267                 curr->signal->session = session;
 268                 attach_pid(curr, PIDTYPE_SID, session);
 269         }
 270         if (process_group(curr) != pgrp) {
 271                 detach_pid(curr, PIDTYPE_PGID);
 272                 curr->signal->pgrp = pgrp;
 273                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 274         }
 275 }
 276
 277 void set_special_pids(pid_t session, pid_t pgrp)
 278 {
 279         write_lock_irq(&tasklist_lock);
 280         __set_special_pids(session, pgrp);
 281         write_unlock_irq(&tasklist_lock);
 282 }
 283
 284 /*
 285  * Let kernel threads use this to say that they
 286  * allow a certain signal (since daemonize() will
 287  * have disabled all of them by default).
 288  */
 289 int allow_signal(int sig)
 290 {
 291         if (sig < 1 || sig > _NSIG)
 292                 return -EINVAL;
 293
 294         spin_lock_irq(&current->sighand->siglock);
 295         sigdelset(&current->blocked, sig);
 296         if (!current->mm) {
 297                 /* Kernel threads handle their own signals.
 298                    Let the signal code know it'll be handled, so
 299                    that they don't get converted to SIGKILL or
 300                    just silently dropped */
 301                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 302         }
 303         recalc_sigpending();
 304         spin_unlock_irq(&current->sighand->siglock);
 305         return 0;
 306 }
 307
 308 EXPORT_SYMBOL(allow_signal);
 309
 310 int disallow_signal(int sig)
 311 {
 312         if (sig < 1 || sig > _NSIG)
 313                 return -EINVAL;
 314
 315         spin_lock_irq(&current->sighand->siglock);
 316         sigaddset(&current->blocked, sig);
 317         recalc_sigpending();
 318         spin_unlock_irq(&current->sighand->siglock);
 319         return 0;
 320 }
 321
 322 EXPORT_SYMBOL(disallow_signal);
 323
 324 /*
 325  *      Put all the gunge required to become a kernel thread without
 326  *      attached user resources in one place where it belongs.
 327  */
 328
 329 void daemonize(const char *name, ...)
 330 {
 331         va_list args;
 332         struct fs_struct *fs;
 333         sigset_t blocked;
 334
 335         va_start(args, name);
 336         vsnprintf(current->comm, sizeof(current->comm), name, args);
 337         va_end(args);
 338
 339         /*
 340          * If we were started as result of loading a module, close all of the
 341          * user space pages.  We don't need them, and if we didn't close them
 342          * they would be locked into memory.
 343          */
 344         exit_mm(current);
 345
 346         set_special_pids(1, 1);
 347         down(&tty_sem);
 348         current->signal->tty = NULL;
 349         up(&tty_sem);
 350
 351         /* Block and flush all signals */
 352         sigfillset(&blocked);
 353         sigprocmask(SIG_BLOCK, &blocked, NULL);
 354         flush_signals(current);
 355
 356         /* Become as one with the init task */
 357
 358         exit_fs(current);       /* current->fs->count--; */
 359         fs = init_task.fs;
 360         current->fs = fs;
 361         atomic_inc(&fs->count);
 362         exit_files(current);
 363         current->files = init_task.files;
 364         atomic_inc(&current->files->count);
 365
 366         reparent_to_init();
 367 }
 368
 369 EXPORT_SYMBOL(daemonize);
 370
 371 static inline void close_files(struct files_struct * files)
 372 {
 373         int i, j;
 374
 375         j = 0;
 376         for (;;) {
 377                 unsigned long set;
 378                 i = j * __NFDBITS;
 379                 if (i >= files->max_fdset || i >= files->max_fds)
 380                         break;
 381                 set = files->open_fds->fds_bits[j++];
 382                 while (set) {
 383                         if (set & 1) {
 384                                 struct file * file = xchg(&files->fd[i], NULL);
 385                                 if (file)
 386                                         filp_close(file, files);
 387                                 vx_openfd_dec(i);
 388                         }
 389                         i++;
 390                         set >>= 1;
 391                 }
 392         }
 393 }
 394
 395 struct files_struct *get_files_struct(struct task_struct *task)
 396 {
 397         struct files_struct *files;
 398
 399         task_lock(task);
 400         files = task->files;
 401         if (files)
 402                 atomic_inc(&files->count);
 403         task_unlock(task);
 404
 405         return files;
 406 }
 407
 408 void fastcall put_files_struct(struct files_struct *files)
 409 {
 410         if (atomic_dec_and_test(&files->count)) {
 411                 close_files(files);
 412                 /*
 413                  * Free the fd and fdset arrays if we expanded them.
 414                  */
 415                 if (files->fd != &files->fd_array[0])
 416                         free_fd_array(files->fd, files->max_fds);
 417                 if (files->max_fdset > __FD_SETSIZE) {
 418                         free_fdset(files->open_fds, files->max_fdset);
 419                         free_fdset(files->close_on_exec, files->max_fdset);
 420                 }
 421                 kmem_cache_free(files_cachep, files);
 422         }
 423 }
 424
 425 EXPORT_SYMBOL(put_files_struct);
 426
 427 static inline void __exit_files(struct task_struct *tsk)
 428 {
 429         struct files_struct * files = tsk->files;
 430
 431         if (files) {
 432                 task_lock(tsk);
 433                 tsk->files = NULL;
 434                 task_unlock(tsk);
 435                 put_files_struct(files);
 436         }
 437 }
 438
 439 void exit_files(struct task_struct *tsk)
 440 {
 441         __exit_files(tsk);
 442 }
 443
 444 static inline void __put_fs_struct(struct fs_struct *fs)
 445 {
 446         /* No need to hold fs->lock if we are killing it */
 447         if (atomic_dec_and_test(&fs->count)) {
 448                 dput(fs->root);
 449                 mntput(fs->rootmnt);
 450                 dput(fs->pwd);
 451                 mntput(fs->pwdmnt);
 452                 if (fs->altroot) {
 453                         dput(fs->altroot);
 454                         mntput(fs->altrootmnt);
 455                 }
 456                 kmem_cache_free(fs_cachep, fs);
 457         }
 458 }
 459
 460 void put_fs_struct(struct fs_struct *fs)
 461 {
 462         __put_fs_struct(fs);
 463 }
 464
 465 static inline void __exit_fs(struct task_struct *tsk)
 466 {
 467         struct fs_struct * fs = tsk->fs;
 468
 469         if (fs) {
 470                 task_lock(tsk);
 471                 tsk->fs = NULL;
 472                 task_unlock(tsk);
 473                 __put_fs_struct(fs);
 474         }
 475 }
 476
 477 void exit_fs(struct task_struct *tsk)
 478 {
 479         __exit_fs(tsk);
 480 }
 481
 482 EXPORT_SYMBOL_GPL(exit_fs);
 483
 484 /*
 485  * Turn us into a lazy TLB process if we
 486  * aren't already..
 487  */
 488 void exit_mm(struct task_struct * tsk)
 489 {
 490         struct mm_struct *mm = tsk->mm;
 491
 492         mm_release(tsk, mm);
 493         if (!mm)
 494                 return;
 495         /*
 496          * Serialize with any possible pending coredump.
 497          * We must hold mmap_sem around checking core_waiters
 498          * and clearing tsk->mm.  The core-inducing thread
 499          * will increment core_waiters for each thread in the
 500          * group with ->mm != NULL.
 501          */
 502         down_read(&mm->mmap_sem);
 503         if (mm->core_waiters) {
 504                 up_read(&mm->mmap_sem);
 505                 down_write(&mm->mmap_sem);
 506                 if (!--mm->core_waiters)
 507                         complete(mm->core_startup_done);
 508                 up_write(&mm->mmap_sem);
 509
 510                 wait_for_completion(&mm->core_done);
 511                 down_read(&mm->mmap_sem);
 512         }
 513         atomic_inc(&mm->mm_count);
 514         if (mm != tsk->active_mm) BUG();
 515         /* more a memory barrier than a real lock */
 516         task_lock(tsk);
 517         tsk->mm = NULL;
 518         up_read(&mm->mmap_sem);
 519         enter_lazy_tlb(mm, current);
 520         task_unlock(tsk);
 521         mmput(mm);
 522 }
 523
 524 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 525 {
 526         /*
 527          * Make sure we're not reparenting to ourselves and that
 528          * the parent is not a zombie.
 529          */
 530         BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
 531         p->real_parent = reaper;
 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
 540         if (p->pdeath_signal)
 541                 /* We already hold the tasklist_lock here.  */
 542                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
 543
 544         /* Move the child from its dying parent to the new one.  */
 545         if (unlikely(traced)) {
 546                 /* Preserve ptrace links if someone else is tracing this child.  */
 547                 list_del_init(&p->ptrace_list);
 548                 if (p->parent != p->real_parent)
 549                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 550         } else {
 551                 /* If this child is being traced, then we're the one tracing it
 552                  * anyway, so let go of it.
 553                  */
 554                 p->ptrace = 0;
 555                 list_del_init(&p->sibling);
 556                 p->parent = p->real_parent;
 557                 list_add_tail(&p->sibling, &p->parent->children);
 558
 559                 /* If we'd notified the old parent about this child's death,
 560                  * also notify the new parent.
 561                  */
 562                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 563                     thread_group_empty(p))
 564                         do_notify_parent(p, p->exit_signal);
 565                 else if (p->state == TASK_TRACED) {
 566                         /*
 567                          * If it was at a trace stop, turn it into
 568                          * a normal stop since it's no longer being
 569                          * traced.
 570                          */
 571                         ptrace_untrace(p);
 572                 }
 573         }
 574
 575         /*
 576          * process group orphan check
 577          * Case ii: Our child is in a different pgrp
 578          * than we are, and it was the only connection
 579          * outside, so the child pgrp is now orphaned.
 580          */
 581         if ((process_group(p) != process_group(father)) &&
 582             (p->signal->session == father->signal->session)) {
 583                 int pgrp = process_group(p);
 584
 585                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 586                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
 587                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
 588                 }
 589         }
 590 }
 591
 592 /*
 593  * When we die, we re-parent all our children.
 594  * Try to give them to another thread in our thread
 595  * group, and if no such member exists, give it to
 596  * the global child reaper process (ie "init")
 597  */
 598 static inline void forget_original_parent(struct task_struct * father,
 599                                           struct list_head *to_release)
 600 {
 601         struct task_struct *p, *reaper = father;
 602         struct list_head *_p, *_n;
 603
 604         /* FIXME handle vchild_reaper/initpid */
 605         do {
 606                 reaper = next_thread(reaper);
 607                 if (reaper == father) {
 608                         reaper = child_reaper;
 609                         break;
 610                 }
 611         } while (reaper->exit_state);
 612
 613         /*
 614          * There are only two places where our children can be:
 615          *
 616          * - in our child list
 617          * - in our ptraced child list
 618          *
 619          * Search them and reparent children.
 620          */
 621         list_for_each_safe(_p, _n, &father->children) {
 622                 int ptrace;
 623                 p = list_entry(_p,struct task_struct,sibling);
 624
 625                 ptrace = p->ptrace;
 626
 627                 /* if father isn't the real parent, then ptrace must be enabled */
 628                 BUG_ON(father != p->real_parent && !ptrace);
 629
 630                 if (father == p->real_parent) {
 631                         /* reparent with a reaper, real father it's us */
 632                         choose_new_parent(p, reaper, child_reaper);
 633                         reparent_thread(p, father, 0);
 634                 } else {
 635                         /* reparent ptraced task to its real parent */
 636                         __ptrace_unlink (p);
 637                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 638                             thread_group_empty(p))
 639                                 do_notify_parent(p, p->exit_signal);
 640                 }
 641
 642                 /*
 643                  * if the ptraced child is a zombie with exit_signal == -1
 644                  * we must collect it before we exit, or it will remain
 645                  * zombie forever since we prevented it from self-reap itself
 646                  * while it was being traced by us, to be able to see it in wait4.
 647                  */
 648                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 649                         list_add(&p->ptrace_list, to_release);
 650         }
 651         list_for_each_safe(_p, _n, &father->ptrace_children) {
 652                 p = list_entry(_p,struct task_struct,ptrace_list);
 653                 choose_new_parent(p, reaper, child_reaper);
 654                 reparent_thread(p, father, 1);
 655         }
 656 }
 657
 658 /*
 659  * Send signals to all our closest relatives so that they know
 660  * to properly mourn us..
 661  */
 662 static void exit_notify(struct task_struct *tsk)
 663 {
 664         int state;
 665         struct task_struct *t;
 666         struct list_head ptrace_dead, *_p, *_n;
 667
 668         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 669             && !thread_group_empty(tsk)) {
 670                 /*
 671                  * This occurs when there was a race between our exit
 672                  * syscall and a group signal choosing us as the one to
 673                  * wake up.  It could be that we are the only thread
 674                  * alerted to check for pending signals, but another thread
 675                  * should be woken now to take the signal since we will not.
 676                  * Now we'll wake all the threads in the group just to make
 677                  * sure someone gets all the pending signals.
 678                  */
 679                 read_lock(&tasklist_lock);
 680                 spin_lock_irq(&tsk->sighand->siglock);
 681                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 682                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 683                                 recalc_sigpending_tsk(t);
 684                                 if (signal_pending(t))
 685                                         signal_wake_up(t, 0);
 686                         }
 687                 spin_unlock_irq(&tsk->sighand->siglock);
 688                 read_unlock(&tasklist_lock);
 689         }
 690
 691         write_lock_irq(&tasklist_lock);
 692
 693         /*
 694          * This does two things:
 695          *
 696          * A.  Make init inherit all the child processes
 697          * B.  Check to see if any process groups have become orphaned
 698          *      as a result of our exiting, and if they have any stopped
 699          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 700          */
 701
 702         INIT_LIST_HEAD(&ptrace_dead);
 703         forget_original_parent(tsk, &ptrace_dead);
 704         BUG_ON(!list_empty(&tsk->children));
 705         BUG_ON(!list_empty(&tsk->ptrace_children));
 706
 707         /*
 708          * Check to see if any process groups have become orphaned
 709          * as a result of our exiting, and if they have any stopped
 710          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 711          *
 712          * Case i: Our father is in a different pgrp than we are
 713          * and we were the only connection outside, so our pgrp
 714          * is about to become orphaned.
 715          */
 716
 717         t = tsk->real_parent;
 718
 719         if ((process_group(t) != process_group(tsk)) &&
 720             (t->signal->session == tsk->signal->session) &&
 721             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 722             has_stopped_jobs(process_group(tsk))) {
 723                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
 724                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
 725         }
 726
 727         /* Let father know we died
 728          *
 729          * Thread signals are configurable, but you aren't going to use
 730          * that to send signals to arbitary processes.
 731          * That stops right now.
 732          *
 733          * If the parent exec id doesn't match the exec id we saved
 734          * when we started then we know the parent has changed security
 735          * domain.
 736          *
 737          * If our self_exec id doesn't match our parent_exec_id then
 738          * we have changed execution domain as these two values started
 739          * the same after a fork.
 740          *
 741          */
 742
 743         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 744             ( tsk->parent_exec_id != t->self_exec_id  ||
 745               tsk->self_exec_id != tsk->parent_exec_id)
 746             && !capable(CAP_KILL))
 747                 tsk->exit_signal = SIGCHLD;
 748
 749
 750         /* If something other than our normal parent is ptracing us, then
 751          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 752          * only has special meaning to our real parent.
 753          */
 754         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 755                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 756                 do_notify_parent(tsk, signal);
 757         } else if (tsk->ptrace) {
 758                 do_notify_parent(tsk, SIGCHLD);
 759         }
 760
 761         state = EXIT_ZOMBIE;
 762         if (tsk->exit_signal == -1 &&
 763             (likely(tsk->ptrace == 0) ||
 764              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 765                 state = EXIT_DEAD;
 766         tsk->exit_state = state;
 767
 768         /*
 769          * Clear these here so that update_process_times() won't try to deliver
 770          * itimer, profile or rlimit signals to this task while it is in late exit.
 771          */
 772         tsk->it_virt_value = cputime_zero;
 773         tsk->it_prof_value = cputime_zero;
 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 == EXIT_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 fastcall NORET_TYPE void do_exit(long code)
 793 {
 794         struct task_struct *tsk = current;
 795         int group_dead;
 796
 797         profile_task_exit(tsk);
 798
 799         if (unlikely(in_interrupt()))
 800                 panic("Aiee, killing interrupt handler!");
 801         if (unlikely(!tsk->pid))
 802                 panic("Attempted to kill the idle task!");
 803         if (unlikely(tsk->pid == 1))
 804                 panic("Attempted to kill init!");
 805         if (tsk->io_context)
 806                 exit_io_context();
 807
 808         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 809                 current->ptrace_message = code;
 810                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 811         }
 812
 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         acct_update_integrals();
 822         update_mem_hiwater();
 823         group_dead = atomic_dec_and_test(&tsk->signal->live);
 824         if (group_dead)
 825                 acct_process(code);
 826         exit_mm(tsk);
 827
 828         exit_sem(tsk);
 829         __exit_files(tsk);
 830         __exit_fs(tsk);
 831         exit_namespace(tsk);
 832         exit_thread();
 833         exit_keys(tsk);
 834
 835         if (group_dead && tsk->signal->leader)
 836                 disassociate_ctty(1);
 837
 838         module_put(tsk->thread_info->exec_domain->module);
 839         if (tsk->binfmt)
 840                 module_put(tsk->binfmt->module);
 841
 842         tsk->exit_code = code;
 843         exit_notify(tsk);
 844 #ifdef CONFIG_NUMA
 845         mpol_free(tsk->mempolicy);
 846         tsk->mempolicy = NULL;
 847 #endif
 848
 849         BUG_ON(!(current->flags & PF_DEAD));
 850         schedule();
 851         BUG();
 852         /* Avoid "noreturn function does return".  */
 853         for (;;) ;
 854 }
 855
 856 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 857 {
 858         if (comp)
 859                 complete(comp);
 860
 861         do_exit(code);
 862 }
 863
 864 EXPORT_SYMBOL(complete_and_exit);
 865
 866 asmlinkage long sys_exit(int error_code)
 867 {
 868         do_exit((error_code&0xff)<<8);
 869 }
 870
 871 task_t fastcall *next_thread(const task_t *p)
 872 {
 873         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 874 }
 875
 876 EXPORT_SYMBOL(next_thread);
 877
 878 /*
 879  * Take down every thread in the group.  This is called by fatal signals
 880  * as well as by sys_exit_group (below).
 881  */
 882 NORET_TYPE void
 883 do_group_exit(int exit_code)
 884 {
 885         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 886
 887         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 888                 exit_code = current->signal->group_exit_code;
 889         else if (!thread_group_empty(current)) {
 890                 struct signal_struct *const sig = current->signal;
 891                 struct sighand_struct *const sighand = current->sighand;
 892                 read_lock(&tasklist_lock);
 893                 spin_lock_irq(&sighand->siglock);
 894                 if (sig->flags & SIGNAL_GROUP_EXIT)
 895                         /* Another thread got here before we took the lock.  */
 896                         exit_code = sig->group_exit_code;
 897                 else {
 898                         sig->flags = SIGNAL_GROUP_EXIT;
 899                         sig->group_exit_code = exit_code;
 900                         zap_other_threads(current);
 901                 }
 902                 spin_unlock_irq(&sighand->siglock);
 903                 read_unlock(&tasklist_lock);
 904         }
 905
 906         do_exit(exit_code);
 907         /* NOTREACHED */
 908 }
 909
 910 /*
 911  * this kills every thread in the thread group. Note that any externally
 912  * wait4()-ing process will get the correct exit code - even if this
 913  * thread is not the thread group leader.
 914  */
 915 asmlinkage void sys_exit_group(int error_code)
 916 {
 917         do_group_exit((error_code & 0xff) << 8);
 918 }
 919
 920 static int eligible_child(pid_t pid, int options, task_t *p)
 921 {
 922         if (pid > 0) {
 923                 if (p->pid != pid)
 924                         return 0;
 925         } else if (!pid) {
 926                 if (process_group(p) != process_group(current))
 927                         return 0;
 928         } else if (pid != -1) {
 929                 if (process_group(p) != -pid)
 930                         return 0;
 931         }
 932
 933         /*
 934          * Do not consider detached threads that are
 935          * not ptraced:
 936          */
 937         if (p->exit_signal == -1 && !p->ptrace)
 938                 return 0;
 939
 940         /* Wait for all children (clone and not) if __WALL is set;
 941          * otherwise, wait for clone children *only* if __WCLONE is
 942          * set; otherwise, wait for non-clone children *only*.  (Note:
 943          * A "clone" child here is one that reports to its parent
 944          * using a signal other than SIGCHLD.) */
 945         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 946             && !(options & __WALL))
 947                 return 0;
 948         /*
 949          * Do not consider thread group leaders that are
 950          * in a non-empty thread group:
 951          */
 952         if (current->tgid != p->tgid && delay_group_leader(p))
 953                 return 2;
 954
 955         if (security_task_wait(p))
 956                 return 0;
 957
 958         return 1;
 959 }
 960
 961 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 962                                int why, int status,
 963                                struct siginfo __user *infop,
 964                                struct rusage __user *rusagep)
 965 {
 966         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
 967         put_task_struct(p);
 968         if (!retval)
 969                 retval = put_user(SIGCHLD, &infop->si_signo);
 970         if (!retval)
 971                 retval = put_user(0, &infop->si_errno);
 972         if (!retval)
 973                 retval = put_user((short)why, &infop->si_code);
 974         if (!retval)
 975                 retval = put_user(pid, &infop->si_pid);
 976         if (!retval)
 977                 retval = put_user(uid, &infop->si_uid);
 978         if (!retval)
 979                 retval = put_user(status, &infop->si_status);
 980         if (!retval)
 981                 retval = pid;
 982         return retval;
 983 }
 984
 985 /*
 986  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
 987  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
 988  * the lock and this task is uninteresting.  If we return nonzero, we have
 989  * released the lock and the system call should return.
 990  */
 991 static int wait_task_zombie(task_t *p, int noreap,
 992                             struct siginfo __user *infop,
 993                             int __user *stat_addr, struct rusage __user *ru)
 994 {
 995         unsigned long state;
 996         int retval;
 997         int status;
 998
 999         if (unlikely(noreap)) {
1000                 pid_t pid = p->pid;
1001                 uid_t uid = p->uid;
1002                 int exit_code = p->exit_code;
1003                 int why, status;
1004
1005                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1006                         return 0;
1007                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1008                         return 0;
1009                 get_task_struct(p);
1010                 read_unlock(&tasklist_lock);
1011                 if ((exit_code & 0x7f) == 0) {
1012                         why = CLD_EXITED;
1013                         status = exit_code >> 8;
1014                 } else {
1015                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1016                         status = exit_code & 0x7f;
1017                 }
1018                 return wait_noreap_copyout(p, pid, uid, why,
1019                                            status, infop, ru);
1020         }
1021
1022         /*
1023          * Try to move the task's state to DEAD
1024          * only one thread is allowed to do this:
1025          */
1026         state = xchg(&p->exit_state, EXIT_DEAD);
1027         if (state != EXIT_ZOMBIE) {
1028                 BUG_ON(state != EXIT_DEAD);
1029                 return 0;
1030         }
1031         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1032                 /*
1033                  * This can only happen in a race with a ptraced thread
1034                  * dying on another processor.
1035                  */
1036                 return 0;
1037         }
1038
1039         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1040                 /*
1041                  * The resource counters for the group leader are in its
1042                  * own task_struct.  Those for dead threads in the group
1043                  * are in its signal_struct, as are those for the child
1044                  * processes it has previously reaped.  All these
1045                  * accumulate in the parent's signal_struct c* fields.
1046                  *
1047                  * We don't bother to take a lock here to protect these
1048                  * p->signal fields, because they are only touched by
1049                  * __exit_signal, which runs with tasklist_lock
1050                  * write-locked anyway, and so is excluded here.  We do
1051                  * need to protect the access to p->parent->signal fields,
1052                  * as other threads in the parent group can be right
1053                  * here reaping other children at the same time.
1054                  */
1055                 spin_lock_irq(&p->parent->sighand->siglock);
1056                 p->parent->signal->cutime =
1057                         cputime_add(p->parent->signal->cutime,
1058                         cputime_add(p->utime,
1059                         cputime_add(p->signal->utime,
1060                                     p->signal->cutime)));
1061                 p->parent->signal->cstime =
1062                         cputime_add(p->parent->signal->cstime,
1063                         cputime_add(p->stime,
1064                         cputime_add(p->signal->stime,
1065                                     p->signal->cstime)));
1066                 p->parent->signal->cmin_flt +=
1067                         p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
1068                 p->parent->signal->cmaj_flt +=
1069                         p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
1070                 p->parent->signal->cnvcsw +=
1071                         p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
1072                 p->parent->signal->cnivcsw +=
1073                         p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
1074                 spin_unlock_irq(&p->parent->sighand->siglock);
1075         }
1076
1077         /*
1078          * Now we are sure this task is interesting, and no other
1079          * thread can reap it because we set its state to EXIT_DEAD.
1080          */
1081         read_unlock(&tasklist_lock);
1082
1083         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1084         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1085                 ? p->signal->group_exit_code : p->exit_code;
1086         if (!retval && stat_addr)
1087                 retval = put_user(status, stat_addr);
1088         if (!retval && infop)
1089                 retval = put_user(SIGCHLD, &infop->si_signo);
1090         if (!retval && infop)
1091                 retval = put_user(0, &infop->si_errno);
1092         if (!retval && infop) {
1093                 int why;
1094
1095                 if ((status & 0x7f) == 0) {
1096                         why = CLD_EXITED;
1097                         status >>= 8;
1098                 } else {
1099                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1100                         status &= 0x7f;
1101                 }
1102                 retval = put_user((short)why, &infop->si_code);
1103                 if (!retval)
1104                         retval = put_user(status, &infop->si_status);
1105         }
1106         if (!retval && infop)
1107                 retval = put_user(p->pid, &infop->si_pid);
1108         if (!retval && infop)
1109                 retval = put_user(p->uid, &infop->si_uid);
1110         if (retval) {
1111                 // TODO: is this safe?
1112                 p->exit_state = EXIT_ZOMBIE;
1113                 return retval;
1114         }
1115         retval = p->pid;
1116         if (p->real_parent != p->parent) {
1117                 write_lock_irq(&tasklist_lock);
1118                 /* Double-check with lock held.  */
1119                 if (p->real_parent != p->parent) {
1120                         __ptrace_unlink(p);
1121                         // TODO: is this safe?
1122                         p->exit_state = EXIT_ZOMBIE;
1123                         /*
1124                          * If this is not a detached task, notify the parent.
1125                          * If it's still not detached after that, don't release
1126                          * it now.
1127                          */
1128                         if (p->exit_signal != -1) {
1129                                 do_notify_parent(p, p->exit_signal);
1130                                 if (p->exit_signal != -1)
1131                                         p = NULL;
1132                         }
1133                 }
1134                 write_unlock_irq(&tasklist_lock);
1135         }
1136         if (p != NULL)
1137                 release_task(p);
1138         BUG_ON(!retval);
1139         return retval;
1140 }
1141
1142 /*
1143  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1144  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1145  * the lock and this task is uninteresting.  If we return nonzero, we have
1146  * released the lock and the system call should return.
1147  */
1148 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1149                              struct siginfo __user *infop,
1150                              int __user *stat_addr, struct rusage __user *ru)
1151 {
1152         int retval, exit_code;
1153
1154         if (!p->exit_code)
1155                 return 0;
1156         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1157             p->signal && p->signal->group_stop_count > 0)
1158                 /*
1159                  * A group stop is in progress and this is the group leader.
1160                  * We won't report until all threads have stopped.
1161                  */
1162                 return 0;
1163
1164         /*
1165          * Now we are pretty sure this task is interesting.
1166          * Make sure it doesn't get reaped out from under us while we
1167          * give up the lock and then examine it below.  We don't want to
1168          * keep holding onto the tasklist_lock while we call getrusage and
1169          * possibly take page faults for user memory.
1170          */
1171         get_task_struct(p);
1172         read_unlock(&tasklist_lock);
1173
1174         if (unlikely(noreap)) {
1175                 pid_t pid = p->pid;
1176                 uid_t uid = p->uid;
1177                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1178
1179                 exit_code = p->exit_code;
1180                 if (unlikely(!exit_code) ||
1181                     unlikely(p->state > TASK_STOPPED))
1182                         goto bail_ref;
1183                 return wait_noreap_copyout(p, pid, uid,
1184                                            why, (exit_code << 8) | 0x7f,
1185                                            infop, ru);
1186         }
1187
1188         write_lock_irq(&tasklist_lock);
1189
1190         /*
1191          * This uses xchg to be atomic with the thread resuming and setting
1192          * it.  It must also be done with the write lock held to prevent a
1193          * race with the EXIT_ZOMBIE case.
1194          */
1195         exit_code = xchg(&p->exit_code, 0);
1196         if (unlikely(p->exit_state)) {
1197                 /*
1198                  * The task resumed and then died.  Let the next iteration
1199                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1200                  * already be zero here if it resumed and did _exit(0).
1201                  * The task itself is dead and won't touch exit_code again;
1202                  * other processors in this function are locked out.
1203                  */
1204                 p->exit_code = exit_code;
1205                 exit_code = 0;
1206         }
1207         if (unlikely(exit_code == 0)) {
1208                 /*
1209                  * Another thread in this function got to it first, or it
1210                  * resumed, or it resumed and then died.
1211                  */
1212                 write_unlock_irq(&tasklist_lock);
1213 bail_ref:
1214                 put_task_struct(p);
1215                 /*
1216                  * We are returning to the wait loop without having successfully
1217                  * removed the process and having released the lock. We cannot
1218                  * continue, since the "p" task pointer is potentially stale.
1219                  *
1220                  * Return -EAGAIN, and do_wait() will restart the loop from the
1221                  * beginning. Do _not_ re-acquire the lock.
1222                  */
1223                 return -EAGAIN;
1224         }
1225
1226         /* move to end of parent's list to avoid starvation */
1227         remove_parent(p);
1228         add_parent(p, p->parent);
1229
1230         write_unlock_irq(&tasklist_lock);
1231
1232         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1233         if (!retval && stat_addr)
1234                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1235         if (!retval && infop)
1236                 retval = put_user(SIGCHLD, &infop->si_signo);
1237         if (!retval && infop)
1238                 retval = put_user(0, &infop->si_errno);
1239         if (!retval && infop)
1240                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1241                                           ? CLD_TRAPPED : CLD_STOPPED),
1242                                   &infop->si_code);
1243         if (!retval && infop)
1244                 retval = put_user(exit_code, &infop->si_status);
1245         if (!retval && infop)
1246                 retval = put_user(p->pid, &infop->si_pid);
1247         if (!retval && infop)
1248                 retval = put_user(p->uid, &infop->si_uid);
1249         if (!retval)
1250                 retval = p->pid;
1251         put_task_struct(p);
1252
1253         BUG_ON(!retval);
1254         return retval;
1255 }
1256
1257 /*
1258  * Handle do_wait work for one task in a live, non-stopped state.
1259  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1260  * the lock and this task is uninteresting.  If we return nonzero, we have
1261  * released the lock and the system call should return.
1262  */
1263 static int wait_task_continued(task_t *p, int noreap,
1264                                struct siginfo __user *infop,
1265                                int __user *stat_addr, struct rusage __user *ru)
1266 {
1267         int retval;
1268         pid_t pid;
1269         uid_t uid;
1270
1271         if (unlikely(!p->signal))
1272                 return 0;
1273
1274         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1275                 return 0;
1276
1277         spin_lock_irq(&p->sighand->siglock);
1278         /* Re-check with the lock held.  */
1279         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1280                 spin_unlock_irq(&p->sighand->siglock);
1281                 return 0;
1282         }
1283         if (!noreap)
1284                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1285         spin_unlock_irq(&p->sighand->siglock);
1286
1287         pid = p->pid;
1288         uid = p->uid;
1289         get_task_struct(p);
1290         read_unlock(&tasklist_lock);
1291
1292         if (!infop) {
1293                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1294                 put_task_struct(p);
1295                 if (!retval && stat_addr)
1296                         retval = put_user(0xffff, stat_addr);
1297                 if (!retval)
1298                         retval = p->pid;
1299         } else {
1300                 retval = wait_noreap_copyout(p, pid, uid,
1301                                              CLD_CONTINUED, SIGCONT,
1302                                              infop, ru);
1303                 BUG_ON(retval == 0);
1304         }
1305
1306         return retval;
1307 }
1308
1309
1310 static inline int my_ptrace_child(struct task_struct *p)
1311 {
1312         if (!(p->ptrace & PT_PTRACED))
1313                 return 0;
1314         if (!(p->ptrace & PT_ATTACHED))
1315                 return 1;
1316         /*
1317          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1318          * we are the attacher.  If we are the real parent, this is a race
1319          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1320          * which we have to switch the parent links, but has already set
1321          * the flags in p->ptrace.
1322          */
1323         return (p->parent != p->real_parent);
1324 }
1325
1326 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1327                     int __user *stat_addr, struct rusage __user *ru)
1328 {
1329         DECLARE_WAITQUEUE(wait, current);
1330         struct task_struct *tsk;
1331         int flag, retval;
1332
1333         add_wait_queue(&current->signal->wait_chldexit,&wait);
1334 repeat:
1335         /*
1336          * We will set this flag if we see any child that might later
1337          * match our criteria, even if we are not able to reap it yet.
1338          */
1339         flag = 0;
1340         current->state = TASK_INTERRUPTIBLE;
1341         read_lock(&tasklist_lock);
1342         tsk = current;
1343         do {
1344                 struct task_struct *p;
1345                 struct list_head *_p;
1346                 int ret;
1347
1348                 list_for_each(_p,&tsk->children) {
1349                         p = list_entry(_p,struct task_struct,sibling);
1350
1351                         ret = eligible_child(pid, options, p);
1352                         if (!ret)
1353                                 continue;
1354
1355                         switch (p->state) {
1356                         case TASK_TRACED:
1357                                 if (!my_ptrace_child(p))
1358                                         continue;
1359                                 /*FALLTHROUGH*/
1360                         case TASK_STOPPED:
1361                                 /*
1362                                  * It's stopped now, so it might later
1363                                  * continue, exit, or stop again.
1364                                  */
1365                                 flag = 1;
1366                                 if (!(options & WUNTRACED) &&
1367                                     !my_ptrace_child(p))
1368                                         continue;
1369                                 retval = wait_task_stopped(p, ret == 2,
1370                                                            (options & WNOWAIT),
1371                                                            infop,
1372                                                            stat_addr, ru);
1373                                 if (retval == -EAGAIN)
1374                                         goto repeat;
1375                                 if (retval != 0) /* He released the lock.  */
1376                                         goto end;
1377                                 break;
1378                         default:
1379                         // case EXIT_DEAD:
1380                                 if (p->exit_state == EXIT_DEAD)
1381                                         continue;
1382                         // case EXIT_ZOMBIE:
1383                                 if (p->exit_state == EXIT_ZOMBIE) {
1384                                         /*
1385                                          * Eligible but we cannot release
1386                                          * it yet:
1387                                          */
1388                                         if (ret == 2)
1389                                                 goto check_continued;
1390                                         if (!likely(options & WEXITED))
1391                                                 continue;
1392                                         retval = wait_task_zombie(
1393                                                 p, (options & WNOWAIT),
1394                                                 infop, stat_addr, ru);
1395                                         /* He released the lock.  */
1396                                         if (retval != 0)
1397                                                 goto end;
1398                                         break;
1399                                 }
1400 check_continued:
1401                                 /*
1402                                  * It's running now, so it might later
1403                                  * exit, stop, or stop and then continue.
1404                                  */
1405                                 flag = 1;
1406                                 if (!unlikely(options & WCONTINUED))
1407                                         continue;
1408                                 retval = wait_task_continued(
1409                                         p, (options & WNOWAIT),
1410                                         infop, stat_addr, ru);
1411                                 if (retval != 0) /* He released the lock.  */
1412                                         goto end;
1413                                 break;
1414                         }
1415                 }
1416                 if (!flag) {
1417                         list_for_each(_p, &tsk->ptrace_children) {
1418                                 p = list_entry(_p, struct task_struct,
1419                                                 ptrace_list);
1420                                 if (!eligible_child(pid, options, p))
1421                                         continue;
1422                                 flag = 1;
1423                                 break;
1424                         }
1425                 }
1426                 if (options & __WNOTHREAD)
1427                         break;
1428                 tsk = next_thread(tsk);
1429                 if (tsk->signal != current->signal)
1430                         BUG();
1431         } while (tsk != current);
1432
1433         read_unlock(&tasklist_lock);
1434         if (flag) {
1435                 retval = 0;
1436                 if (options & WNOHANG)
1437                         goto end;
1438                 retval = -ERESTARTSYS;
1439                 if (signal_pending(current))
1440                         goto end;
1441                 schedule();
1442                 goto repeat;
1443         }
1444         retval = -ECHILD;
1445 end:
1446         current->state = TASK_RUNNING;
1447         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1448         if (infop) {
1449                 if (retval > 0)
1450                 retval = 0;
1451                 else {
1452                         /*
1453                          * For a WNOHANG return, clear out all the fields
1454                          * we would set so the user can easily tell the
1455                          * difference.
1456                          */
1457                         if (!retval)
1458                                 retval = put_user(0, &infop->si_signo);
1459                         if (!retval)
1460                                 retval = put_user(0, &infop->si_errno);
1461                         if (!retval)
1462                                 retval = put_user(0, &infop->si_code);
1463                         if (!retval)
1464                                 retval = put_user(0, &infop->si_pid);
1465                         if (!retval)
1466                                 retval = put_user(0, &infop->si_uid);
1467                         if (!retval)
1468                                 retval = put_user(0, &infop->si_status);
1469                 }
1470         }
1471         return retval;
1472 }
1473
1474 asmlinkage long sys_waitid(int which, pid_t pid,
1475                            struct siginfo __user *infop, int options,
1476                            struct rusage __user *ru)
1477 {
1478         long ret;
1479
1480         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1481                 return -EINVAL;
1482         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1483                 return -EINVAL;
1484
1485         switch (which) {
1486         case P_ALL:
1487                 pid = -1;
1488                 break;
1489         case P_PID:
1490                 if (pid <= 0)
1491                         return -EINVAL;
1492                 break;
1493         case P_PGID:
1494                 if (pid <= 0)
1495                         return -EINVAL;
1496                 pid = -pid;
1497                 break;
1498         default:
1499                 return -EINVAL;
1500         }
1501
1502         ret = do_wait(pid, options, infop, NULL, ru);
1503
1504         /* avoid REGPARM breakage on x86: */
1505         prevent_tail_call(ret);
1506         return ret;
1507 }
1508
1509 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1510                           int options, struct rusage __user *ru)
1511 {
1512         long ret;
1513
1514         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1515                         __WNOTHREAD|__WCLONE|__WALL))
1516                 return -EINVAL;
1517         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1518
1519         /* avoid REGPARM breakage on x86: */
1520         prevent_tail_call(ret);
1521         return ret;
1522 }
1523
1524 #ifdef __ARCH_WANT_SYS_WAITPID
1525
1526 /*
1527  * sys_waitpid() remains for compatibility. waitpid() should be
1528  * implemented by calling sys_wait4() from libc.a.
1529  */
1530 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1531 {
1532         return sys_wait4(pid, stat_addr, options, NULL);
1533 }
1534
1535 #endif