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.h>
  29 #include <linux/ckrm_tsk.h>
  30 #include <linux/vs_limit.h>
  31 #include <linux/ckrm_mem.h>
  32 #include <linux/syscalls.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                         }
 386                         i++;
 387                         set >>= 1;
 388                 }
 389         }
 390 }
 391
 392 struct files_struct *get_files_struct(struct task_struct *task)
 393 {
 394         struct files_struct *files;
 395
 396         task_lock(task);
 397         files = task->files;
 398         if (files)
 399                 atomic_inc(&files->count);
 400         task_unlock(task);
 401
 402         return files;
 403 }
 404
 405 void fastcall put_files_struct(struct files_struct *files)
 406 {
 407         if (atomic_dec_and_test(&files->count)) {
 408                 close_files(files);
 409                 /*
 410                  * Free the fd and fdset arrays if we expanded them.
 411                  */
 412                 if (files->fd != &files->fd_array[0])
 413                         free_fd_array(files->fd, files->max_fds);
 414                 if (files->max_fdset > __FD_SETSIZE) {
 415                         free_fdset(files->open_fds, files->max_fdset);
 416                         free_fdset(files->close_on_exec, files->max_fdset);
 417                 }
 418                 kmem_cache_free(files_cachep, files);
 419         }
 420 }
 421
 422 EXPORT_SYMBOL(put_files_struct);
 423
 424 static inline void __exit_files(struct task_struct *tsk)
 425 {
 426         struct files_struct * files = tsk->files;
 427
 428         if (files) {
 429                 task_lock(tsk);
 430                 tsk->files = NULL;
 431                 task_unlock(tsk);
 432                 put_files_struct(files);
 433         }
 434 }
 435
 436 void exit_files(struct task_struct *tsk)
 437 {
 438         __exit_files(tsk);
 439 }
 440
 441 static inline void __put_fs_struct(struct fs_struct *fs)
 442 {
 443         /* No need to hold fs->lock if we are killing it */
 444         if (atomic_dec_and_test(&fs->count)) {
 445                 dput(fs->root);
 446                 mntput(fs->rootmnt);
 447                 dput(fs->pwd);
 448                 mntput(fs->pwdmnt);
 449                 if (fs->altroot) {
 450                         dput(fs->altroot);
 451                         mntput(fs->altrootmnt);
 452                 }
 453                 kmem_cache_free(fs_cachep, fs);
 454         }
 455 }
 456
 457 void put_fs_struct(struct fs_struct *fs)
 458 {
 459         __put_fs_struct(fs);
 460 }
 461
 462 static inline void __exit_fs(struct task_struct *tsk)
 463 {
 464         struct fs_struct * fs = tsk->fs;
 465
 466         if (fs) {
 467                 task_lock(tsk);
 468                 tsk->fs = NULL;
 469                 task_unlock(tsk);
 470                 __put_fs_struct(fs);
 471         }
 472 }
 473
 474 void exit_fs(struct task_struct *tsk)
 475 {
 476         __exit_fs(tsk);
 477 }
 478
 479 EXPORT_SYMBOL_GPL(exit_fs);
 480
 481 /*
 482  * Turn us into a lazy TLB process if we
 483  * aren't already..
 484  */
 485 static inline void __exit_mm(struct task_struct * tsk)
 486 {
 487         struct mm_struct *mm = tsk->mm;
 488
 489         mm_release(tsk, mm);
 490         if (!mm)
 491                 return;
 492         /*
 493          * Serialize with any possible pending coredump.
 494          * We must hold mmap_sem around checking core_waiters
 495          * and clearing tsk->mm.  The core-inducing thread
 496          * will increment core_waiters for each thread in the
 497          * group with ->mm != NULL.
 498          */
 499         down_read(&mm->mmap_sem);
 500         if (mm->core_waiters) {
 501                 up_read(&mm->mmap_sem);
 502                 down_write(&mm->mmap_sem);
 503                 if (!--mm->core_waiters)
 504                         complete(mm->core_startup_done);
 505                 up_write(&mm->mmap_sem);
 506
 507                 wait_for_completion(&mm->core_done);
 508                 down_read(&mm->mmap_sem);
 509         }
 510         atomic_inc(&mm->mm_count);
 511         if (mm != tsk->active_mm) BUG();
 512         /* more a memory barrier than a real lock */
 513         task_lock(tsk);
 514         tsk->mm = NULL;
 515         up_read(&mm->mmap_sem);
 516 #ifdef CONFIG_CKRM_RES_MEM
 517         spin_lock(&mm->peertask_lock);
 518         list_del_init(&tsk->mm_peers);
 519         ckrm_mem_evaluate_mm(mm);
 520         spin_unlock(&mm->peertask_lock);
 521 #endif
 522         enter_lazy_tlb(mm, current);
 523         task_unlock(tsk);
 524         mmput(mm);
 525 }
 526
 527 void exit_mm(struct task_struct *tsk)
 528 {
 529         __exit_mm(tsk);
 530 }
 531
 532 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 533 {
 534         /*
 535          * Make sure we're not reparenting to ourselves and that
 536          * the parent is not a zombie.
 537          */
 538         BUG_ON(p == reaper || reaper->state >= EXIT_ZOMBIE || reaper->exit_state >= EXIT_ZOMBIE);
 539         p->real_parent = reaper;
 540         if (p->parent == p->real_parent)
 541                 BUG();
 542 }
 543
 544 static inline void reparent_thread(task_t *p, task_t *father, int traced)
 545 {
 546         /* We don't want people slaying init.  */
 547         if (p->exit_signal != -1)
 548                 p->exit_signal = SIGCHLD;
 549
 550         if (p->pdeath_signal)
 551                 /* We already hold the tasklist_lock here.  */
 552                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
 553
 554         /* Move the child from its dying parent to the new one.  */
 555         if (unlikely(traced)) {
 556                 /* Preserve ptrace links if someone else is tracing this child.  */
 557                 list_del_init(&p->ptrace_list);
 558                 if (p->parent != p->real_parent)
 559                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 560         } else {
 561                 /* If this child is being traced, then we're the one tracing it
 562                  * anyway, so let go of it.
 563                  */
 564                 p->ptrace = 0;
 565                 list_del_init(&p->sibling);
 566                 p->parent = p->real_parent;
 567                 list_add_tail(&p->sibling, &p->parent->children);
 568
 569                 /* If we'd notified the old parent about this child's death,
 570                  * also notify the new parent.
 571                  */
 572                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 573                     thread_group_empty(p))
 574                         do_notify_parent(p, p->exit_signal);
 575                 else if (p->state == TASK_TRACED) {
 576                         /*
 577                          * If it was at a trace stop, turn it into
 578                          * a normal stop since it's no longer being
 579                          * traced.
 580                          */
 581                         ptrace_untrace(p);
 582                 }
 583         }
 584
 585         /*
 586          * process group orphan check
 587          * Case ii: Our child is in a different pgrp
 588          * than we are, and it was the only connection
 589          * outside, so the child pgrp is now orphaned.
 590          */
 591         if ((process_group(p) != process_group(father)) &&
 592             (p->signal->session == father->signal->session)) {
 593                 int pgrp = process_group(p);
 594
 595                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 596                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
 597                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
 598                 }
 599         }
 600 }
 601
 602 /*
 603  * When we die, we re-parent all our children.
 604  * Try to give them to another thread in our thread
 605  * group, and if no such member exists, give it to
 606  * the global child reaper process (ie "init")
 607  */
 608 static inline void forget_original_parent(struct task_struct * father,
 609                                           struct list_head *to_release)
 610 {
 611         struct task_struct *p, *reaper = father;
 612         struct list_head *_p, *_n;
 613
 614         do {
 615                 reaper = next_thread(reaper);
 616                 if (reaper == father) {
 617                         reaper = child_reaper;
 618                         break;
 619                 }
 620         } while (reaper->exit_state >= EXIT_ZOMBIE);
 621
 622         /*
 623          * There are only two places where our children can be:
 624          *
 625          * - in our child list
 626          * - in our ptraced child list
 627          *
 628          * Search them and reparent children.
 629          */
 630         list_for_each_safe(_p, _n, &father->children) {
 631                 int ptrace;
 632                 p = list_entry(_p,struct task_struct,sibling);
 633
 634                 ptrace = p->ptrace;
 635
 636                 /* if father isn't the real parent, then ptrace must be enabled */
 637                 BUG_ON(father != p->real_parent && !ptrace);
 638
 639                 if (father == p->real_parent) {
 640                         /* reparent with a reaper, real father it's us */
 641                         choose_new_parent(p, reaper, child_reaper);
 642                         reparent_thread(p, father, 0);
 643                 } else {
 644                         /* reparent ptraced task to its real parent */
 645                         __ptrace_unlink (p);
 646                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 647                             thread_group_empty(p))
 648                                 do_notify_parent(p, p->exit_signal);
 649                 }
 650
 651                 /*
 652                  * if the ptraced child is a zombie with exit_signal == -1
 653                  * we must collect it before we exit, or it will remain
 654                  * zombie forever since we prevented it from self-reap itself
 655                  * while it was being traced by us, to be able to see it in wait4.
 656                  */
 657                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 658                         list_add(&p->ptrace_list, to_release);
 659         }
 660         list_for_each_safe(_p, _n, &father->ptrace_children) {
 661                 p = list_entry(_p,struct task_struct,ptrace_list);
 662                 choose_new_parent(p, reaper, child_reaper);
 663                 reparent_thread(p, father, 1);
 664         }
 665 }
 666
 667 /*
 668  * Send signals to all our closest relatives so that they know
 669  * to properly mourn us..
 670  */
 671 static void exit_notify(struct task_struct *tsk)
 672 {
 673         int state;
 674         struct task_struct *t;
 675         struct list_head ptrace_dead, *_p, *_n;
 676
 677         ckrm_cb_exit(tsk);
 678
 679         if (signal_pending(tsk) && !tsk->signal->group_exit
 680             && !thread_group_empty(tsk)) {
 681                 /*
 682                  * This occurs when there was a race between our exit
 683                  * syscall and a group signal choosing us as the one to
 684                  * wake up.  It could be that we are the only thread
 685                  * alerted to check for pending signals, but another thread
 686                  * should be woken now to take the signal since we will not.
 687                  * Now we'll wake all the threads in the group just to make
 688                  * sure someone gets all the pending signals.
 689                  */
 690                 read_lock(&tasklist_lock);
 691                 spin_lock_irq(&tsk->sighand->siglock);
 692                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 693                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 694                                 recalc_sigpending_tsk(t);
 695                                 if (signal_pending(t))
 696                                         signal_wake_up(t, 0);
 697                         }
 698                 spin_unlock_irq(&tsk->sighand->siglock);
 699                 read_unlock(&tasklist_lock);
 700         }
 701
 702         write_lock_irq(&tasklist_lock);
 703
 704         /*
 705          * This does two things:
 706          *
 707          * A.  Make init inherit all the child processes
 708          * B.  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
 713         INIT_LIST_HEAD(&ptrace_dead);
 714         forget_original_parent(tsk, &ptrace_dead);
 715         BUG_ON(!list_empty(&tsk->children));
 716         BUG_ON(!list_empty(&tsk->ptrace_children));
 717
 718         /*
 719          * Check to see if any process groups have become orphaned
 720          * as a result of our exiting, and if they have any stopped
 721          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 722          *
 723          * Case i: Our father is in a different pgrp than we are
 724          * and we were the only connection outside, so our pgrp
 725          * is about to become orphaned.
 726          */
 727
 728         t = tsk->real_parent;
 729
 730         if ((process_group(t) != process_group(tsk)) &&
 731             (t->signal->session == tsk->signal->session) &&
 732             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 733             has_stopped_jobs(process_group(tsk))) {
 734                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
 735                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
 736         }
 737
 738         /* Let father know we died
 739          *
 740          * Thread signals are configurable, but you aren't going to use
 741          * that to send signals to arbitary processes.
 742          * That stops right now.
 743          *
 744          * If the parent exec id doesn't match the exec id we saved
 745          * when we started then we know the parent has changed security
 746          * domain.
 747          *
 748          * If our self_exec id doesn't match our parent_exec_id then
 749          * we have changed execution domain as these two values started
 750          * the same after a fork.
 751          *
 752          */
 753
 754         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 755             ( tsk->parent_exec_id != t->self_exec_id  ||
 756               tsk->self_exec_id != tsk->parent_exec_id)
 757             && !capable(CAP_KILL))
 758                 tsk->exit_signal = SIGCHLD;
 759
 760
 761         /* If something other than our normal parent is ptracing us, then
 762          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 763          * only has special meaning to our real parent.
 764          */
 765         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 766                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 767                 do_notify_parent(tsk, signal);
 768         } else if (tsk->ptrace) {
 769                 do_notify_parent(tsk, SIGCHLD);
 770         }
 771
 772         state = EXIT_ZOMBIE;
 773         if (tsk->exit_signal == -1 && tsk->ptrace == 0)
 774                 state = EXIT_DEAD;
 775         tsk->exit_state = state;
 776
 777         /*
 778          * Clear these here so that update_process_times() won't try to deliver
 779          * itimer, profile or rlimit signals to this task while it is in late exit.
 780          */
 781         tsk->it_virt_value = 0;
 782         tsk->it_prof_value = 0;
 783
 784         write_unlock_irq(&tasklist_lock);
 785
 786         list_for_each_safe(_p, _n, &ptrace_dead) {
 787                 list_del_init(_p);
 788                 t = list_entry(_p,struct task_struct,ptrace_list);
 789                 release_task(t);
 790         }
 791
 792         /* If the process is dead, release it - nobody will wait for it */
 793         if (state == EXIT_DEAD)
 794                 release_task(tsk);
 795
 796         /* PF_DEAD causes final put_task_struct after we schedule. */
 797         preempt_disable();
 798         tsk->flags |= PF_DEAD;
 799 }
 800
 801 fastcall NORET_TYPE void do_exit(long code)
 802 {
 803         struct task_struct *tsk = current;
 804         int group_dead;
 805
 806         profile_task_exit(tsk);
 807
 808         if (unlikely(in_interrupt()))
 809                 panic("Aiee, killing interrupt handler!");
 810         if (unlikely(!tsk->pid))
 811                 panic("Attempted to kill the idle task!");
 812         if (unlikely(tsk->pid == 1))
 813                 panic("Attempted to kill init!");
 814         if (tsk->io_context)
 815                 exit_io_context();
 816         tsk->flags |= PF_EXITING;
 817         del_timer_sync(&tsk->real_timer);
 818
 819         if (unlikely(in_atomic()))
 820                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 821                                 current->comm, current->pid,
 822                                 preempt_count());
 823
 824         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 825                 current->ptrace_message = code;
 826                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 827         }
 828
 829         group_dead = atomic_dec_and_test(&tsk->signal->live);
 830         if (group_dead)
 831                 acct_process(code);
 832         if (current->tux_info) {
 833 #ifdef CONFIG_TUX_DEBUG
 834                 printk("Possibly unexpected TUX-thread exit(%ld) at %p?\n",
 835                         code, __builtin_return_address(0));
 836 #endif
 837                 current->tux_exit();
 838         }
 839         __exit_mm(tsk);
 840
 841         exit_sem(tsk);
 842         __exit_files(tsk);
 843         __exit_fs(tsk);
 844         exit_namespace(tsk);
 845         exit_thread();
 846         exit_keys(tsk);
 847
 848         if (group_dead && tsk->signal->leader)
 849                 disassociate_ctty(1);
 850
 851         module_put(tsk->thread_info->exec_domain->module);
 852         if (tsk->binfmt)
 853                 module_put(tsk->binfmt->module);
 854
 855         tsk->exit_code = code;
 856         exit_notify(tsk);
 857 #ifdef CONFIG_NUMA
 858         mpol_free(tsk->mempolicy);
 859         tsk->mempolicy = NULL;
 860 #endif
 861
 862         BUG_ON(!(current->flags & PF_DEAD));
 863         schedule();
 864         BUG();
 865         /* Avoid "noreturn function does return".  */
 866         for (;;) ;
 867 }
 868
 869 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 870 {
 871         if (comp)
 872                 complete(comp);
 873
 874         do_exit(code);
 875 }
 876
 877 EXPORT_SYMBOL(complete_and_exit);
 878
 879 asmlinkage long sys_exit(int error_code)
 880 {
 881         do_exit((error_code&0xff)<<8);
 882 }
 883
 884 task_t fastcall *next_thread(const task_t *p)
 885 {
 886 #ifdef CONFIG_SMP
 887         if (!p->sighand)
 888                 BUG();
 889         if (!spin_is_locked(&p->sighand->siglock) &&
 890                                 !rwlock_is_locked(&tasklist_lock))
 891                 BUG();
 892 #endif
 893         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 894 }
 895
 896 EXPORT_SYMBOL(next_thread);
 897
 898 /*
 899  * Take down every thread in the group.  This is called by fatal signals
 900  * as well as by sys_exit_group (below).
 901  */
 902 NORET_TYPE void
 903 do_group_exit(int exit_code)
 904 {
 905         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 906
 907         if (current->signal->group_exit)
 908                 exit_code = current->signal->group_exit_code;
 909         else if (!thread_group_empty(current)) {
 910                 struct signal_struct *const sig = current->signal;
 911                 struct sighand_struct *const sighand = current->sighand;
 912                 read_lock(&tasklist_lock);
 913                 spin_lock_irq(&sighand->siglock);
 914                 if (sig->group_exit)
 915                         /* Another thread got here before we took the lock.  */
 916                         exit_code = sig->group_exit_code;
 917                 else {
 918                         sig->group_exit = 1;
 919                         sig->group_exit_code = exit_code;
 920                         zap_other_threads(current);
 921                 }
 922                 spin_unlock_irq(&sighand->siglock);
 923                 read_unlock(&tasklist_lock);
 924         }
 925
 926         do_exit(exit_code);
 927         /* NOTREACHED */
 928 }
 929
 930 /*
 931  * this kills every thread in the thread group. Note that any externally
 932  * wait4()-ing process will get the correct exit code - even if this
 933  * thread is not the thread group leader.
 934  */
 935 asmlinkage void sys_exit_group(int error_code)
 936 {
 937         do_group_exit((error_code & 0xff) << 8);
 938 }
 939
 940 static int eligible_child(pid_t pid, int options, task_t *p)
 941 {
 942         if (pid > 0) {
 943                 if (p->pid != pid)
 944                         return 0;
 945         } else if (!pid) {
 946                 if (process_group(p) != process_group(current))
 947                         return 0;
 948         } else if (pid != -1) {
 949                 if (process_group(p) != -pid)
 950                         return 0;
 951         }
 952
 953         /*
 954          * Do not consider detached threads that are
 955          * not ptraced:
 956          */
 957         if (p->exit_signal == -1 && !p->ptrace)
 958                 return 0;
 959
 960         /* Wait for all children (clone and not) if __WALL is set;
 961          * otherwise, wait for clone children *only* if __WCLONE is
 962          * set; otherwise, wait for non-clone children *only*.  (Note:
 963          * A "clone" child here is one that reports to its parent
 964          * using a signal other than SIGCHLD.) */
 965         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 966             && !(options & __WALL))
 967                 return 0;
 968         /*
 969          * Do not consider thread group leaders that are
 970          * in a non-empty thread group:
 971          */
 972         if (current->tgid != p->tgid && delay_group_leader(p))
 973                 return 2;
 974
 975         if (security_task_wait(p))
 976                 return 0;
 977
 978         return 1;
 979 }
 980
 981 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 982                                int why, int status,
 983                                struct siginfo __user *infop,
 984                                struct rusage __user *rusagep)
 985 {
 986         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
 987         put_task_struct(p);
 988         if (!retval)
 989                 retval = put_user(SIGCHLD, &infop->si_signo);
 990         if (!retval)
 991                 retval = put_user(0, &infop->si_errno);
 992         if (!retval)
 993                 retval = put_user((short)why, &infop->si_code);
 994         if (!retval)
 995                 retval = put_user(pid, &infop->si_pid);
 996         if (!retval)
 997                 retval = put_user(uid, &infop->si_uid);
 998         if (!retval)
 999                 retval = put_user(status, &infop->si_status);
1000         if (!retval)
1001                 retval = pid;
1002         return retval;
1003 }
1004
1005 /*
1006  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1007  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1008  * the lock and this task is uninteresting.  If we return nonzero, we have
1009  * released the lock and the system call should return.
1010  */
1011 static int wait_task_zombie(task_t *p, int noreap,
1012                             struct siginfo __user *infop,
1013                             int __user *stat_addr, struct rusage __user *ru)
1014 {
1015         unsigned long state;
1016         int retval;
1017         int status;
1018
1019         if (unlikely(noreap)) {
1020                 pid_t pid = p->pid;
1021                 uid_t uid = p->uid;
1022                 int exit_code = p->exit_code;
1023                 int why, status;
1024
1025                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1026                         return 0;
1027                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1028                         return 0;
1029                 get_task_struct(p);
1030                 read_unlock(&tasklist_lock);
1031                 if ((exit_code & 0x7f) == 0) {
1032                         why = CLD_EXITED;
1033                         status = exit_code >> 8;
1034                 } else {
1035                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1036                         status = exit_code & 0x7f;
1037                 }
1038                 return wait_noreap_copyout(p, pid, uid, why,
1039                                            status, infop, ru);
1040         }
1041
1042         /*
1043          * Try to move the task's state to DEAD
1044          * only one thread is allowed to do this:
1045          */
1046         state = xchg(&p->exit_state, EXIT_DEAD);
1047         if (state != EXIT_ZOMBIE) {
1048                 BUG_ON(state != EXIT_DEAD);
1049                 return 0;
1050         }
1051         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1052                 /*
1053                  * This can only happen in a race with a ptraced thread
1054                  * dying on another processor.
1055                  */
1056                 return 0;
1057         }
1058
1059         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1060                 /*
1061                  * The resource counters for the group leader are in its
1062                  * own task_struct.  Those for dead threads in the group
1063                  * are in its signal_struct, as are those for the child
1064                  * processes it has previously reaped.  All these
1065                  * accumulate in the parent's signal_struct c* fields.
1066                  *
1067                  * We don't bother to take a lock here to protect these
1068                  * p->signal fields, because they are only touched by
1069                  * __exit_signal, which runs with tasklist_lock
1070                  * write-locked anyway, and so is excluded here.  We do
1071                  * need to protect the access to p->parent->signal fields,
1072                  * as other threads in the parent group can be right
1073                  * here reaping other children at the same time.
1074                  */
1075                 spin_lock_irq(&p->parent->sighand->siglock);
1076                 p->parent->signal->cutime +=
1077                         p->utime + p->signal->utime + p->signal->cutime;
1078                 p->parent->signal->cstime +=
1079                         p->stime + p->signal->stime + p->signal->cstime;
1080                 p->parent->signal->cmin_flt +=
1081                         p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
1082                 p->parent->signal->cmaj_flt +=
1083                         p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
1084                 p->parent->signal->cnvcsw +=
1085                         p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
1086                 p->parent->signal->cnivcsw +=
1087                         p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
1088                 spin_unlock_irq(&p->parent->sighand->siglock);
1089         }
1090
1091         /*
1092          * Now we are sure this task is interesting, and no other
1093          * thread can reap it because we set its state to EXIT_DEAD.
1094          */
1095         read_unlock(&tasklist_lock);
1096
1097         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1098         status = p->signal->group_exit
1099                 ? p->signal->group_exit_code : p->exit_code;
1100         if (!retval && stat_addr)
1101                 retval = put_user(status, stat_addr);
1102         if (!retval && infop)
1103                 retval = put_user(SIGCHLD, &infop->si_signo);
1104         if (!retval && infop)
1105                 retval = put_user(0, &infop->si_errno);
1106         if (!retval && infop) {
1107                 int why;
1108
1109                 if ((status & 0x7f) == 0) {
1110                         why = CLD_EXITED;
1111                         status >>= 8;
1112                 } else {
1113                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1114                         status &= 0x7f;
1115                 }
1116                 retval = put_user((short)why, &infop->si_code);
1117                 if (!retval)
1118                         retval = put_user(status, &infop->si_status);
1119         }
1120         if (!retval && infop)
1121                 retval = put_user(p->pid, &infop->si_pid);
1122         if (!retval && infop)
1123                 retval = put_user(p->uid, &infop->si_uid);
1124         if (retval) {
1125                 // TODO: is this safe?
1126                 p->exit_state = EXIT_ZOMBIE;
1127                 return retval;
1128         }
1129         retval = p->pid;
1130         if (p->real_parent != p->parent) {
1131                 write_lock_irq(&tasklist_lock);
1132                 /* Double-check with lock held.  */
1133                 if (p->real_parent != p->parent) {
1134                         __ptrace_unlink(p);
1135                         // TODO: is this safe?
1136                         p->exit_state = EXIT_ZOMBIE;
1137                         /*
1138                          * If this is not a detached task, notify the parent.
1139                          * If it's still not detached after that, don't release
1140                          * it now.
1141                          */
1142                         if (p->exit_signal != -1) {
1143                                 do_notify_parent(p, p->exit_signal);
1144                                 if (p->exit_signal != -1)
1145                                         p = NULL;
1146                         }
1147                 }
1148                 write_unlock_irq(&tasklist_lock);
1149         }
1150         if (p != NULL)
1151                 release_task(p);
1152         BUG_ON(!retval);
1153         return retval;
1154 }
1155
1156 /*
1157  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1158  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1159  * the lock and this task is uninteresting.  If we return nonzero, we have
1160  * released the lock and the system call should return.
1161  */
1162 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1163                              struct siginfo __user *infop,
1164                              int __user *stat_addr, struct rusage __user *ru)
1165 {
1166         int retval, exit_code;
1167
1168         if (!p->exit_code)
1169                 return 0;
1170         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1171             p->signal && p->signal->group_stop_count > 0)
1172                 /*
1173                  * A group stop is in progress and this is the group leader.
1174                  * We won't report until all threads have stopped.
1175                  */
1176                 return 0;
1177
1178         /*
1179          * Now we are pretty sure this task is interesting.
1180          * Make sure it doesn't get reaped out from under us while we
1181          * give up the lock and then examine it below.  We don't want to
1182          * keep holding onto the tasklist_lock while we call getrusage and
1183          * possibly take page faults for user memory.
1184          */
1185         get_task_struct(p);
1186         read_unlock(&tasklist_lock);
1187
1188         if (unlikely(noreap)) {
1189                 pid_t pid = p->pid;
1190                 uid_t uid = p->uid;
1191                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1192
1193                 exit_code = p->exit_code;
1194                 if (unlikely(!exit_code) ||
1195                     unlikely(p->state > TASK_STOPPED))
1196                         goto bail_ref;
1197                 return wait_noreap_copyout(p, pid, uid,
1198                                            why, (exit_code << 8) | 0x7f,
1199                                            infop, ru);
1200         }
1201
1202         write_lock_irq(&tasklist_lock);
1203
1204         /*
1205          * This uses xchg to be atomic with the thread resuming and setting
1206          * it.  It must also be done with the write lock held to prevent a
1207          * race with the EXIT_ZOMBIE case.
1208          */
1209         exit_code = xchg(&p->exit_code, 0);
1210         if (unlikely(p->exit_state >= EXIT_ZOMBIE)) {
1211                 /*
1212                  * The task resumed and then died.  Let the next iteration
1213                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1214                  * already be zero here if it resumed and did _exit(0).
1215                  * The task itself is dead and won't touch exit_code again;
1216                  * other processors in this function are locked out.
1217                  */
1218                 p->exit_code = exit_code;
1219                 exit_code = 0;
1220         }
1221         if (unlikely(exit_code == 0)) {
1222                 /*
1223                  * Another thread in this function got to it first, or it
1224                  * resumed, or it resumed and then died.
1225                  */
1226                 write_unlock_irq(&tasklist_lock);
1227 bail_ref:
1228                 put_task_struct(p);
1229                 /*
1230                  * We are returning to the wait loop without having successfully
1231                  * removed the process and having released the lock. We cannot
1232                  * continue, since the "p" task pointer is potentially stale.
1233                  *
1234                  * Return -EAGAIN, and do_wait() will restart the loop from the
1235                  * beginning. Do _not_ re-acquire the lock.
1236                  */
1237                 return -EAGAIN;
1238         }
1239
1240         /* move to end of parent's list to avoid starvation */
1241         remove_parent(p);
1242         add_parent(p, p->parent);
1243
1244         write_unlock_irq(&tasklist_lock);
1245
1246         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1247         if (!retval && stat_addr)
1248                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1249         if (!retval && infop)
1250                 retval = put_user(SIGCHLD, &infop->si_signo);
1251         if (!retval && infop)
1252                 retval = put_user(0, &infop->si_errno);
1253         if (!retval && infop)
1254                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1255                                           ? CLD_TRAPPED : CLD_STOPPED),
1256                                   &infop->si_code);
1257         if (!retval && infop)
1258                 retval = put_user(exit_code, &infop->si_status);
1259         if (!retval && infop)
1260                 retval = put_user(p->pid, &infop->si_pid);
1261         if (!retval && infop)
1262                 retval = put_user(p->uid, &infop->si_uid);
1263         if (!retval)
1264                 retval = p->pid;
1265         put_task_struct(p);
1266
1267         BUG_ON(!retval);
1268         return retval;
1269 }
1270
1271 /*
1272  * Handle do_wait work for one task in a live, non-stopped state.
1273  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1274  * the lock and this task is uninteresting.  If we return nonzero, we have
1275  * released the lock and the system call should return.
1276  */
1277 static int wait_task_continued(task_t *p, int noreap,
1278                                struct siginfo __user *infop,
1279                                int __user *stat_addr, struct rusage __user *ru)
1280 {
1281         int retval;
1282         pid_t pid;
1283         uid_t uid;
1284
1285         if (unlikely(!p->signal))
1286                 return 0;
1287
1288         if (p->signal->stop_state >= 0)
1289                 return 0;
1290
1291         spin_lock_irq(&p->sighand->siglock);
1292         if (p->signal->stop_state >= 0) { /* Re-check with the lock held.  */
1293                 spin_unlock_irq(&p->sighand->siglock);
1294                 return 0;
1295         }
1296         if (!noreap)
1297                 p->signal->stop_state = 0;
1298         spin_unlock_irq(&p->sighand->siglock);
1299
1300         pid = p->pid;
1301         uid = p->uid;
1302         get_task_struct(p);
1303         read_unlock(&tasklist_lock);
1304
1305         if (!infop) {
1306                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1307                 put_task_struct(p);
1308                 if (!retval && stat_addr)
1309                         retval = put_user(0xffff, stat_addr);
1310                 if (!retval)
1311                         retval = p->pid;
1312         } else {
1313                 retval = wait_noreap_copyout(p, pid, uid,
1314                                              CLD_CONTINUED, SIGCONT,
1315                                              infop, ru);
1316                 BUG_ON(retval == 0);
1317         }
1318
1319         return retval;
1320 }
1321
1322
1323 static inline int my_ptrace_child(struct task_struct *p)
1324 {
1325         if (!(p->ptrace & PT_PTRACED))
1326                 return 0;
1327         if (!(p->ptrace & PT_ATTACHED))
1328                 return 1;
1329         /*
1330          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1331          * we are the attacher.  If we are the real parent, this is a race
1332          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1333          * which we have to switch the parent links, but has already set
1334          * the flags in p->ptrace.
1335          */
1336         return (p->parent != p->real_parent);
1337 }
1338
1339 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1340                     int __user *stat_addr, struct rusage __user *ru)
1341 {
1342         DECLARE_WAITQUEUE(wait, current);
1343         struct task_struct *tsk;
1344         int flag, retval;
1345
1346         add_wait_queue(&current->wait_chldexit,&wait);
1347 repeat:
1348         /*
1349          * We will set this flag if we see any child that might later
1350          * match our criteria, even if we are not able to reap it yet.
1351          */
1352         flag = 0;
1353         current->state = TASK_INTERRUPTIBLE;
1354         read_lock(&tasklist_lock);
1355         tsk = current;
1356         do {
1357                 struct task_struct *p;
1358                 struct list_head *_p;
1359                 int ret;
1360
1361                 list_for_each(_p,&tsk->children) {
1362                         p = list_entry(_p,struct task_struct,sibling);
1363
1364                         ret = eligible_child(pid, options, p);
1365                         if (!ret)
1366                                 continue;
1367
1368                         switch (p->state) {
1369                         case TASK_TRACED:
1370                                 if (!my_ptrace_child(p))
1371                                         continue;
1372                                 /*FALLTHROUGH*/
1373                         case TASK_STOPPED:
1374                                 /*
1375                                  * It's stopped now, so it might later
1376                                  * continue, exit, or stop again.
1377                                  */
1378                                 flag = 1;
1379                                 if (!(options & WUNTRACED) &&
1380                                     !my_ptrace_child(p))
1381                                         continue;
1382                                 retval = wait_task_stopped(p, ret == 2,
1383                                                            (options & WNOWAIT),
1384                                                            infop,
1385                                                            stat_addr, ru);
1386                                 if (retval == -EAGAIN)
1387                                         goto repeat;
1388                                 if (retval != 0) /* He released the lock.  */
1389                                         goto end;
1390                                 break;
1391                         default:
1392                         // case EXIT_DEAD:
1393                                 if (p->exit_state == EXIT_DEAD)
1394                                         continue;
1395                         // case EXIT_ZOMBIE:
1396                                 if (p->exit_state == EXIT_ZOMBIE) {
1397                                         /*
1398                                          * Eligible but we cannot release
1399                                          * it yet:
1400                                          */
1401                                         if (ret == 2)
1402                                                 goto check_continued;
1403                                         if (!likely(options & WEXITED))
1404                                                 continue;
1405                                         retval = wait_task_zombie(
1406                                                 p, (options & WNOWAIT),
1407                                                 infop, stat_addr, ru);
1408                                         /* He released the lock.  */
1409                                         if (retval != 0)
1410                                                 goto end;
1411                                         break;
1412                                 }
1413 check_continued:
1414                                 /*
1415                                  * It's running now, so it might later
1416                                  * exit, stop, or stop and then continue.
1417                                  */
1418                                 flag = 1;
1419                                 if (!unlikely(options & WCONTINUED))
1420                                         continue;
1421
1422                                 retval = wait_task_continued(
1423                                         p, (options & WNOWAIT),
1424                                         infop, stat_addr, ru);
1425                                 if (retval != 0) /* He released the lock.  */
1426                                         goto end;
1427                                 break;
1428                         }
1429                 }
1430                 if (!flag) {
1431                         list_for_each(_p, &tsk->ptrace_children) {
1432                                 p = list_entry(_p, struct task_struct,
1433                                                 ptrace_list);
1434                                 if (!eligible_child(pid, options, p))
1435                                         continue;
1436                                 flag = 1;
1437                                 break;
1438                         }
1439                 }
1440                 if (options & __WNOTHREAD)
1441                         break;
1442                 tsk = next_thread(tsk);
1443                 if (tsk->signal != current->signal)
1444                         BUG();
1445         } while (tsk != current);
1446
1447         read_unlock(&tasklist_lock);
1448         if (flag) {
1449                 retval = 0;
1450                 if (options & WNOHANG)
1451                         goto end;
1452                 retval = -ERESTARTSYS;
1453                 if (signal_pending(current))
1454                         goto end;
1455                 schedule();
1456                 goto repeat;
1457         }
1458         retval = -ECHILD;
1459 end:
1460         current->state = TASK_RUNNING;
1461         remove_wait_queue(&current->wait_chldexit,&wait);
1462         if (infop) {
1463                 if (retval > 0)
1464                 retval = 0;
1465                 else {
1466                         /*
1467                          * For a WNOHANG return, clear out all the fields
1468                          * we would set so the user can easily tell the
1469                          * difference.
1470                          */
1471                         if (!retval)
1472                                 retval = put_user(0, &infop->si_signo);
1473                         if (!retval)
1474                                 retval = put_user(0, &infop->si_errno);
1475                         if (!retval)
1476                                 retval = put_user(0, &infop->si_code);
1477                         if (!retval)
1478                                 retval = put_user(0, &infop->si_pid);
1479                         if (!retval)
1480                                 retval = put_user(0, &infop->si_uid);
1481                         if (!retval)
1482                                 retval = put_user(0, &infop->si_status);
1483                 }
1484         }
1485         return retval;
1486 }
1487
1488 asmlinkage long sys_waitid(int which, pid_t pid,
1489                            struct siginfo __user *infop, int options,
1490                            struct rusage __user *ru)
1491 {
1492         long ret;
1493
1494         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1495                 return -EINVAL;
1496         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1497                 return -EINVAL;
1498
1499         switch (which) {
1500         case P_ALL:
1501                 pid = -1;
1502                 break;
1503         case P_PID:
1504                 if (pid <= 0)
1505                         return -EINVAL;
1506                 break;
1507         case P_PGID:
1508                 if (pid <= 0)
1509                         return -EINVAL;
1510                 pid = -pid;
1511                 break;
1512         default:
1513                 return -EINVAL;
1514         }
1515
1516         ret = do_wait(pid, options, infop, NULL, ru);
1517
1518         /* avoid REGPARM breakage on x86: */
1519         prevent_tail_call(ret);
1520         return ret;
1521 }
1522
1523 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1524                           int options, struct rusage __user *ru)
1525 {
1526         long ret;
1527
1528         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1529                         __WNOTHREAD|__WCLONE|__WALL))
1530                 return -EINVAL;
1531         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1532
1533         /* avoid REGPARM breakage on x86: */
1534         prevent_tail_call(ret);
1535         return ret;
1536 }
1537
1538 #ifdef __ARCH_WANT_SYS_WAITPID
1539
1540 /*
1541  * sys_waitpid() remains for compatibility. waitpid() should be
1542  * implemented by calling sys_wait4() from libc.a.
1543  */
1544 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1545 {
1546         return sys_wait4(pid, stat_addr, options, NULL);
1547 }
1548
1549 #endif