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 #include <linux/vs_limit.h>
  34
  35 #include <asm/uaccess.h>
  36 #include <asm/unistd.h>
  37 #include <asm/pgtable.h>
  38 #include <asm/mmu_context.h>
  39
  40 extern void sem_exit (void);
  41 extern struct task_struct *child_reaper;
  42
  43 int getrusage(struct task_struct *, int, struct rusage __user *);
  44
  45 static void __unhash_process(struct task_struct *p)
  46 {
  47         nr_threads--;
  48         /* tasklist_lock is held, is this sufficient? */
  49         if (p->vx_info) {
  50                 atomic_dec(&p->vx_info->cvirt.nr_threads);
  51                 vx_nproc_dec(p);
  52         }
  53         detach_pid(p, PIDTYPE_PID);
  54         detach_pid(p, PIDTYPE_TGID);
  55         if (thread_group_leader(p)) {
  56                 detach_pid(p, PIDTYPE_PGID);
  57                 detach_pid(p, PIDTYPE_SID);
  58                 if (p->pid)
  59                         __get_cpu_var(process_counts)--;
  60         }
  61
  62         REMOVE_LINKS(p);
  63 }
  64
  65 void release_task(struct task_struct * p)
  66 {
  67         int zap_leader;
  68         task_t *leader;
  69         struct dentry *proc_dentry;
  70
  71 repeat:
  72         atomic_dec(&p->user->processes);
  73         spin_lock(&p->proc_lock);
  74         proc_dentry = proc_pid_unhash(p);
  75         write_lock_irq(&tasklist_lock);
  76         if (unlikely(p->ptrace))
  77                 __ptrace_unlink(p);
  78         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  79         __exit_signal(p);
  80         __exit_sighand(p);
  81         __unhash_process(p);
  82
  83         /*
  84          * If we are the last non-leader member of the thread
  85          * group, and the leader is zombie, then notify the
  86          * group leader's parent process. (if it wants notification.)
  87          */
  88         zap_leader = 0;
  89         leader = p->group_leader;
  90         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
  91                 BUG_ON(leader->exit_signal == -1);
  92                 do_notify_parent(leader, leader->exit_signal);
  93                 /*
  94                  * If we were the last child thread and the leader has
  95                  * exited already, and the leader's parent ignores SIGCHLD,
  96                  * then we are the one who should release the leader.
  97                  *
  98                  * do_notify_parent() will have marked it self-reaping in
  99                  * that case.
 100                  */
 101                 zap_leader = (leader->exit_signal == -1);
 102         }
 103
 104         sched_exit(p);
 105         write_unlock_irq(&tasklist_lock);
 106         spin_unlock(&p->proc_lock);
 107         proc_pid_flush(proc_dentry);
 108         release_thread(p);
 109         put_task_struct(p);
 110
 111         p = leader;
 112         if (unlikely(zap_leader))
 113                 goto repeat;
 114 }
 115
 116 /* we are using it only for SMP init */
 117
 118 void unhash_process(struct task_struct *p)
 119 {
 120         struct dentry *proc_dentry;
 121
 122         spin_lock(&p->proc_lock);
 123         proc_dentry = proc_pid_unhash(p);
 124         write_lock_irq(&tasklist_lock);
 125         __unhash_process(p);
 126         write_unlock_irq(&tasklist_lock);
 127         spin_unlock(&p->proc_lock);
 128         proc_pid_flush(proc_dentry);
 129 }
 130
 131 /*
 132  * This checks not only the pgrp, but falls back on the pid if no
 133  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 134  * without this...
 135  */
 136 int session_of_pgrp(int pgrp)
 137 {
 138         struct task_struct *p;
 139         int sid = -1;
 140
 141         read_lock(&tasklist_lock);
 142         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 143                 if (p->signal->session > 0) {
 144                         sid = p->signal->session;
 145                         goto out;
 146                 }
 147         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 148         p = find_task_by_pid(pgrp);
 149         if (p)
 150                 sid = p->signal->session;
 151 out:
 152         read_unlock(&tasklist_lock);
 153
 154         return sid;
 155 }
 156
 157 /*
 158  * Determine if a process group is "orphaned", according to the POSIX
 159  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 160  * by terminal-generated stop signals.  Newly orphaned process groups are
 161  * to receive a SIGHUP and a SIGCONT.
 162  *
 163  * "I ask you, have you ever known what it is to be an orphan?"
 164  */
 165 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 166 {
 167         struct task_struct *p;
 168         int ret = 1;
 169
 170         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 171                 if (p == ignored_task
 172                                 || p->exit_state >= EXIT_ZOMBIE
 173                                 || p->real_parent->pid == 1)
 174                         continue;
 175                 if (process_group(p->real_parent) != pgrp
 176                             && p->real_parent->signal->session == p->signal->session) {
 177                         ret = 0;
 178                         break;
 179                 }
 180         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 181         return ret;     /* (sighing) "Often!" */
 182 }
 183
 184 int is_orphaned_pgrp(int pgrp)
 185 {
 186         int retval;
 187
 188         read_lock(&tasklist_lock);
 189         retval = will_become_orphaned_pgrp(pgrp, NULL);
 190         read_unlock(&tasklist_lock);
 191
 192         return retval;
 193 }
 194
 195 static inline int has_stopped_jobs(int pgrp)
 196 {
 197         int retval = 0;
 198         struct task_struct *p;
 199
 200         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 201                 if (p->state != TASK_STOPPED)
 202                         continue;
 203
 204                 /* If p is stopped by a debugger on a signal that won't
 205                    stop it, then don't count p as stopped.  This isn't
 206                    perfect but it's a good approximation.  */
 207                 if (unlikely (p->ptrace)
 208                     && p->exit_code != SIGSTOP
 209                     && p->exit_code != SIGTSTP
 210                     && p->exit_code != SIGTTOU
 211                     && p->exit_code != SIGTTIN)
 212                         continue;
 213
 214                 retval = 1;
 215                 break;
 216         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 217         return retval;
 218 }
 219
 220 /**
 221  * reparent_to_init() - Reparent the calling kernel thread to the init task.
 222  *
 223  * If a kernel thread is launched as a result of a system call, or if
 224  * it ever exits, it should generally reparent itself to init so that
 225  * it is correctly cleaned up on exit.
 226  *
 227  * The various task state such as scheduling policy and priority may have
 228  * been inherited from a user process, so we reset them to sane values here.
 229  *
 230  * NOTE that reparent_to_init() gives the caller full capabilities.
 231  */
 232 void reparent_to_init(void)
 233 {
 234         write_lock_irq(&tasklist_lock);
 235
 236         ptrace_unlink(current);
 237         /* Reparent to init */
 238         REMOVE_LINKS(current);
 239         /* FIXME handle vchild_reaper/initpid */
 240         current->parent = child_reaper;
 241         current->real_parent = child_reaper;
 242         SET_LINKS(current);
 243
 244         /* Set the exit signal to SIGCHLD so we signal init on exit */
 245         current->exit_signal = SIGCHLD;
 246
 247         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
 248                 set_user_nice(current, 0);
 249         /* cpus_allowed? */
 250         /* rt_priority? */
 251         /* signals? */
 252         security_task_reparent_to_init(current);
 253         memcpy(current->signal->rlim, init_task.signal->rlim,
 254                sizeof(current->signal->rlim));
 255         atomic_inc(&(INIT_USER->__count));
 256         write_unlock_irq(&tasklist_lock);
 257         switch_uid(INIT_USER);
 258 }
 259
 260 void __set_special_pids(pid_t session, pid_t pgrp)
 261 {
 262         struct task_struct *curr = current;
 263
 264         if (curr->signal->session != session) {
 265                 detach_pid(curr, PIDTYPE_SID);
 266                 curr->signal->session = session;
 267                 attach_pid(curr, PIDTYPE_SID, session);
 268         }
 269         if (process_group(curr) != pgrp) {
 270                 detach_pid(curr, PIDTYPE_PGID);
 271                 curr->signal->pgrp = pgrp;
 272                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 273         }
 274 }
 275
 276 void set_special_pids(pid_t session, pid_t pgrp)
 277 {
 278         write_lock_irq(&tasklist_lock);
 279         __set_special_pids(session, pgrp);
 280         write_unlock_irq(&tasklist_lock);
 281 }
 282
 283 /*
 284  * Let kernel threads use this to say that they
 285  * allow a certain signal (since daemonize() will
 286  * have disabled all of them by default).
 287  */
 288 int allow_signal(int sig)
 289 {
 290         if (sig < 1 || sig > _NSIG)
 291                 return -EINVAL;
 292
 293         spin_lock_irq(&current->sighand->siglock);
 294         sigdelset(&current->blocked, sig);
 295         if (!current->mm) {
 296                 /* Kernel threads handle their own signals.
 297                    Let the signal code know it'll be handled, so
 298                    that they don't get converted to SIGKILL or
 299                    just silently dropped */
 300                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 301         }
 302         recalc_sigpending();
 303         spin_unlock_irq(&current->sighand->siglock);
 304         return 0;
 305 }
 306
 307 EXPORT_SYMBOL(allow_signal);
 308
 309 int disallow_signal(int sig)
 310 {
 311         if (sig < 1 || sig > _NSIG)
 312                 return -EINVAL;
 313
 314         spin_lock_irq(&current->sighand->siglock);
 315         sigaddset(&current->blocked, sig);
 316         recalc_sigpending();
 317         spin_unlock_irq(&current->sighand->siglock);
 318         return 0;
 319 }
 320
 321 EXPORT_SYMBOL(disallow_signal);
 322
 323 /*
 324  *      Put all the gunge required to become a kernel thread without
 325  *      attached user resources in one place where it belongs.
 326  */
 327
 328 void daemonize(const char *name, ...)
 329 {
 330         va_list args;
 331         struct fs_struct *fs;
 332         sigset_t blocked;
 333
 334         va_start(args, name);
 335         vsnprintf(current->comm, sizeof(current->comm), name, args);
 336         va_end(args);
 337
 338         /*
 339          * If we were started as result of loading a module, close all of the
 340          * user space pages.  We don't need them, and if we didn't close them
 341          * they would be locked into memory.
 342          */
 343         exit_mm(current);
 344
 345         set_special_pids(1, 1);
 346         down(&tty_sem);
 347         current->signal->tty = NULL;
 348         up(&tty_sem);
 349
 350         /* Block and flush all signals */
 351         sigfillset(&blocked);
 352         sigprocmask(SIG_BLOCK, &blocked, NULL);
 353         flush_signals(current);
 354
 355         /* Become as one with the init task */
 356
 357         exit_fs(current);       /* current->fs->count--; */
 358         fs = init_task.fs;
 359         current->fs = fs;
 360         atomic_inc(&fs->count);
 361         exit_files(current);
 362         current->files = init_task.files;
 363         atomic_inc(&current->files->count);
 364
 365         reparent_to_init();
 366 }
 367
 368 EXPORT_SYMBOL(daemonize);
 369
 370 static inline void close_files(struct files_struct * files)
 371 {
 372         int i, j;
 373
 374         j = 0;
 375         for (;;) {
 376                 unsigned long set;
 377                 i = j * __NFDBITS;
 378                 if (i >= files->max_fdset || i >= files->max_fds)
 379                         break;
 380                 set = files->open_fds->fds_bits[j++];
 381                 while (set) {
 382                         if (set & 1) {
 383                                 struct file * file = xchg(&files->fd[i], NULL);
 384                                 if (file)
 385                                         filp_close(file, files);
 386                                 // vx_openfd_dec(i);
 387                         }
 388                         i++;
 389                         set >>= 1;
 390                 }
 391         }
 392 }
 393
 394 struct files_struct *get_files_struct(struct task_struct *task)
 395 {
 396         struct files_struct *files;
 397
 398         task_lock(task);
 399         files = task->files;
 400         if (files)
 401                 atomic_inc(&files->count);
 402         task_unlock(task);
 403
 404         return files;
 405 }
 406
 407 void fastcall put_files_struct(struct files_struct *files)
 408 {
 409         if (atomic_dec_and_test(&files->count)) {
 410                 close_files(files);
 411                 /*
 412                  * Free the fd and fdset arrays if we expanded them.
 413                  */
 414                 if (files->fd != &files->fd_array[0])
 415                         free_fd_array(files->fd, files->max_fds);
 416                 if (files->max_fdset > __FD_SETSIZE) {
 417                         free_fdset(files->open_fds, files->max_fdset);
 418                         free_fdset(files->close_on_exec, files->max_fdset);
 419                 }
 420                 kmem_cache_free(files_cachep, files);
 421         }
 422 }
 423
 424 EXPORT_SYMBOL(put_files_struct);
 425
 426 static inline void __exit_files(struct task_struct *tsk)
 427 {
 428         struct files_struct * files = tsk->files;
 429
 430         if (files) {
 431                 task_lock(tsk);
 432                 tsk->files = NULL;
 433                 task_unlock(tsk);
 434                 put_files_struct(files);
 435         }
 436 }
 437
 438 void exit_files(struct task_struct *tsk)
 439 {
 440         __exit_files(tsk);
 441 }
 442
 443 static inline void __put_fs_struct(struct fs_struct *fs)
 444 {
 445         /* No need to hold fs->lock if we are killing it */
 446         if (atomic_dec_and_test(&fs->count)) {
 447                 dput(fs->root);
 448                 mntput(fs->rootmnt);
 449                 dput(fs->pwd);
 450                 mntput(fs->pwdmnt);
 451                 if (fs->altroot) {
 452                         dput(fs->altroot);
 453                         mntput(fs->altrootmnt);
 454                 }
 455                 kmem_cache_free(fs_cachep, fs);
 456         }
 457 }
 458
 459 void put_fs_struct(struct fs_struct *fs)
 460 {
 461         __put_fs_struct(fs);
 462 }
 463
 464 static inline void __exit_fs(struct task_struct *tsk)
 465 {
 466         struct fs_struct * fs = tsk->fs;
 467
 468         if (fs) {
 469                 task_lock(tsk);
 470                 tsk->fs = NULL;
 471                 task_unlock(tsk);
 472                 __put_fs_struct(fs);
 473         }
 474 }
 475
 476 void exit_fs(struct task_struct *tsk)
 477 {
 478         __exit_fs(tsk);
 479 }
 480
 481 EXPORT_SYMBOL_GPL(exit_fs);
 482
 483 /*
 484  * Turn us into a lazy TLB process if we
 485  * aren't already..
 486  */
 487 static inline void __exit_mm(struct task_struct * tsk)
 488 {
 489         struct mm_struct *mm = tsk->mm;
 490
 491         mm_release(tsk, mm);
 492         if (!mm)
 493                 return;
 494         /*
 495          * Serialize with any possible pending coredump.
 496          * We must hold mmap_sem around checking core_waiters
 497          * and clearing tsk->mm.  The core-inducing thread
 498          * will increment core_waiters for each thread in the
 499          * group with ->mm != NULL.
 500          */
 501         down_read(&mm->mmap_sem);
 502         if (mm->core_waiters) {
 503                 up_read(&mm->mmap_sem);
 504                 down_write(&mm->mmap_sem);
 505                 if (!--mm->core_waiters)
 506                         complete(mm->core_startup_done);
 507                 up_write(&mm->mmap_sem);
 508
 509                 wait_for_completion(&mm->core_done);
 510                 down_read(&mm->mmap_sem);
 511         }
 512         atomic_inc(&mm->mm_count);
 513         if (mm != tsk->active_mm) BUG();
 514         /* more a memory barrier than a real lock */
 515         task_lock(tsk);
 516         tsk->mm = NULL;
 517         up_read(&mm->mmap_sem);
 518 #ifdef CONFIG_CKRM_RES_MEM
 519         spin_lock(&mm->peertask_lock);
 520         list_del_init(&tsk->mm_peers);
 521         ckrm_mem_evaluate_mm(mm);
 522         spin_unlock(&mm->peertask_lock);
 523 #endif
 524         enter_lazy_tlb(mm, current);
 525         task_unlock(tsk);
 526         mmput(mm);
 527 }
 528
 529 void exit_mm(struct task_struct *tsk)
 530 {
 531         __exit_mm(tsk);
 532 }
 533
 534 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 535 {
 536         /*
 537          * Make sure we're not reparenting to ourselves and that
 538          * the parent is not a zombie.
 539          */
 540         BUG_ON(p == reaper || reaper->state >= EXIT_ZOMBIE || reaper->exit_state >= EXIT_ZOMBIE);
 541         p->real_parent = reaper;
 542         if (p->parent == p->real_parent)
 543                 BUG();
 544 }
 545
 546 static inline void reparent_thread(task_t *p, task_t *father, int traced)
 547 {
 548         /* We don't want people slaying init.  */
 549         if (p->exit_signal != -1)
 550                 p->exit_signal = SIGCHLD;
 551
 552         if (p->pdeath_signal)
 553                 /* We already hold the tasklist_lock here.  */
 554                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
 555
 556         /* Move the child from its dying parent to the new one.  */
 557         if (unlikely(traced)) {
 558                 /* Preserve ptrace links if someone else is tracing this child.  */
 559                 list_del_init(&p->ptrace_list);
 560                 if (p->parent != p->real_parent)
 561                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 562         } else {
 563                 /* If this child is being traced, then we're the one tracing it
 564                  * anyway, so let go of it.
 565                  */
 566                 p->ptrace = 0;
 567                 list_del_init(&p->sibling);
 568                 p->parent = p->real_parent;
 569                 list_add_tail(&p->sibling, &p->parent->children);
 570
 571                 /* If we'd notified the old parent about this child's death,
 572                  * also notify the new parent.
 573                  */
 574                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 575                     thread_group_empty(p))
 576                         do_notify_parent(p, p->exit_signal);
 577                 else if (p->state == TASK_TRACED) {
 578                         /*
 579                          * If it was at a trace stop, turn it into
 580                          * a normal stop since it's no longer being
 581                          * traced.
 582                          */
 583                         ptrace_untrace(p);
 584                 }
 585         }
 586
 587         /*
 588          * process group orphan check
 589          * Case ii: Our child is in a different pgrp
 590          * than we are, and it was the only connection
 591          * outside, so the child pgrp is now orphaned.
 592          */
 593         if ((process_group(p) != process_group(father)) &&
 594             (p->signal->session == father->signal->session)) {
 595                 int pgrp = process_group(p);
 596
 597                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 598                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
 599                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
 600                 }
 601         }
 602 }
 603
 604 /*
 605  * When we die, we re-parent all our children.
 606  * Try to give them to another thread in our thread
 607  * group, and if no such member exists, give it to
 608  * the global child reaper process (ie "init")
 609  */
 610 static inline void forget_original_parent(struct task_struct * father,
 611                                           struct list_head *to_release)
 612 {
 613         struct task_struct *p, *reaper = father;
 614         struct list_head *_p, *_n;
 615
 616         /* FIXME handle vchild_reaper/initpid */
 617         do {
 618                 reaper = next_thread(reaper);
 619                 if (reaper == father) {
 620                         reaper = child_reaper;
 621                         break;
 622                 }
 623         } while (reaper->exit_state >= EXIT_ZOMBIE);
 624
 625         /*
 626          * There are only two places where our children can be:
 627          *
 628          * - in our child list
 629          * - in our ptraced child list
 630          *
 631          * Search them and reparent children.
 632          */
 633         list_for_each_safe(_p, _n, &father->children) {
 634                 int ptrace;
 635                 p = list_entry(_p,struct task_struct,sibling);
 636
 637                 ptrace = p->ptrace;
 638
 639                 /* if father isn't the real parent, then ptrace must be enabled */
 640                 BUG_ON(father != p->real_parent && !ptrace);
 641
 642                 if (father == p->real_parent) {
 643                         /* reparent with a reaper, real father it's us */
 644                         choose_new_parent(p, reaper, child_reaper);
 645                         reparent_thread(p, father, 0);
 646                 } else {
 647                         /* reparent ptraced task to its real parent */
 648                         __ptrace_unlink (p);
 649                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 650                             thread_group_empty(p))
 651                                 do_notify_parent(p, p->exit_signal);
 652                 }
 653
 654                 /*
 655                  * if the ptraced child is a zombie with exit_signal == -1
 656                  * we must collect it before we exit, or it will remain
 657                  * zombie forever since we prevented it from self-reap itself
 658                  * while it was being traced by us, to be able to see it in wait4.
 659                  */
 660                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 661                         list_add(&p->ptrace_list, to_release);
 662         }
 663         list_for_each_safe(_p, _n, &father->ptrace_children) {
 664                 p = list_entry(_p,struct task_struct,ptrace_list);
 665                 choose_new_parent(p, reaper, child_reaper);
 666                 reparent_thread(p, father, 1);
 667         }
 668 }
 669
 670 /*
 671  * Send signals to all our closest relatives so that they know
 672  * to properly mourn us..
 673  */
 674 static void exit_notify(struct task_struct *tsk)
 675 {
 676         int state;
 677         struct task_struct *t;
 678         struct list_head ptrace_dead, *_p, *_n;
 679
 680         ckrm_cb_exit(tsk);
 681
 682         if (signal_pending(tsk) && !tsk->signal->group_exit
 683             && !thread_group_empty(tsk)) {
 684                 /*
 685                  * This occurs when there was a race between our exit
 686                  * syscall and a group signal choosing us as the one to
 687                  * wake up.  It could be that we are the only thread
 688                  * alerted to check for pending signals, but another thread
 689                  * should be woken now to take the signal since we will not.
 690                  * Now we'll wake all the threads in the group just to make
 691                  * sure someone gets all the pending signals.
 692                  */
 693                 read_lock(&tasklist_lock);
 694                 spin_lock_irq(&tsk->sighand->siglock);
 695                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 696                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 697                                 recalc_sigpending_tsk(t);
 698                                 if (signal_pending(t))
 699                                         signal_wake_up(t, 0);
 700                         }
 701                 spin_unlock_irq(&tsk->sighand->siglock);
 702                 read_unlock(&tasklist_lock);
 703         }
 704
 705         write_lock_irq(&tasklist_lock);
 706
 707         /*
 708          * This does two things:
 709          *
 710          * A.  Make init inherit all the child processes
 711          * B.  Check to see if any process groups have become orphaned
 712          *      as a result of our exiting, and if they have any stopped
 713          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 714          */
 715
 716         INIT_LIST_HEAD(&ptrace_dead);
 717         forget_original_parent(tsk, &ptrace_dead);
 718         BUG_ON(!list_empty(&tsk->children));
 719         BUG_ON(!list_empty(&tsk->ptrace_children));
 720
 721         /*
 722          * Check to see if any process groups have become orphaned
 723          * as a result of our exiting, and if they have any stopped
 724          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 725          *
 726          * Case i: Our father is in a different pgrp than we are
 727          * and we were the only connection outside, so our pgrp
 728          * is about to become orphaned.
 729          */
 730
 731         t = tsk->real_parent;
 732
 733         if ((process_group(t) != process_group(tsk)) &&
 734             (t->signal->session == tsk->signal->session) &&
 735             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 736             has_stopped_jobs(process_group(tsk))) {
 737                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
 738                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
 739         }
 740
 741         /* Let father know we died
 742          *
 743          * Thread signals are configurable, but you aren't going to use
 744          * that to send signals to arbitary processes.
 745          * That stops right now.
 746          *
 747          * If the parent exec id doesn't match the exec id we saved
 748          * when we started then we know the parent has changed security
 749          * domain.
 750          *
 751          * If our self_exec id doesn't match our parent_exec_id then
 752          * we have changed execution domain as these two values started
 753          * the same after a fork.
 754          *
 755          */
 756
 757         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 758             ( tsk->parent_exec_id != t->self_exec_id  ||
 759               tsk->self_exec_id != tsk->parent_exec_id)
 760             && !capable(CAP_KILL))
 761                 tsk->exit_signal = SIGCHLD;
 762
 763
 764         /* If something other than our normal parent is ptracing us, then
 765          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 766          * only has special meaning to our real parent.
 767          */
 768         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 769                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 770                 do_notify_parent(tsk, signal);
 771         } else if (tsk->ptrace) {
 772                 do_notify_parent(tsk, SIGCHLD);
 773         }
 774
 775         state = EXIT_ZOMBIE;
 776         if (tsk->exit_signal == -1 && tsk->ptrace == 0)
 777                 state = EXIT_DEAD;
 778         tsk->exit_state = state;
 779
 780         /*
 781          * Clear these here so that update_process_times() won't try to deliver
 782          * itimer, profile or rlimit signals to this task while it is in late exit.
 783          */
 784         tsk->it_virt_value = 0;
 785         tsk->it_prof_value = 0;
 786
 787         write_unlock_irq(&tasklist_lock);
 788
 789         list_for_each_safe(_p, _n, &ptrace_dead) {
 790                 list_del_init(_p);
 791                 t = list_entry(_p,struct task_struct,ptrace_list);
 792                 release_task(t);
 793         }
 794
 795         /* If the process is dead, release it - nobody will wait for it */
 796         if (state == EXIT_DEAD)
 797                 release_task(tsk);
 798
 799         /* PF_DEAD causes final put_task_struct after we schedule. */
 800         preempt_disable();
 801         tsk->flags |= PF_DEAD;
 802 }
 803
 804 fastcall NORET_TYPE void do_exit(long code)
 805 {
 806         struct task_struct *tsk = current;
 807         int group_dead;
 808
 809         profile_task_exit(tsk);
 810
 811         if (unlikely(in_interrupt()))
 812                 panic("Aiee, killing interrupt handler!");
 813         if (unlikely(!tsk->pid))
 814                 panic("Attempted to kill the idle task!");
 815         if (unlikely(tsk->pid == 1))
 816                 panic("Attempted to kill init!");
 817         if (tsk->io_context)
 818                 exit_io_context();
 819         tsk->flags |= PF_EXITING;
 820         del_timer_sync(&tsk->real_timer);
 821
 822         if (unlikely(in_atomic()))
 823                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 824                                 current->comm, current->pid,
 825                                 preempt_count());
 826
 827         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 828                 current->ptrace_message = code;
 829                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 830         }
 831
 832         group_dead = atomic_dec_and_test(&tsk->signal->live);
 833         if (group_dead)
 834                 acct_process(code);
 835         if (current->tux_info) {
 836 #ifdef CONFIG_TUX_DEBUG
 837                 printk("Possibly unexpected TUX-thread exit(%ld) at %p?\n",
 838                         code, __builtin_return_address(0));
 839 #endif
 840                 current->tux_exit();
 841         }
 842         __exit_mm(tsk);
 843
 844         exit_sem(tsk);
 845         __exit_files(tsk);
 846         __exit_fs(tsk);
 847         exit_namespace(tsk);
 848         exit_thread();
 849         exit_keys(tsk);
 850
 851         if (group_dead && tsk->signal->leader)
 852                 disassociate_ctty(1);
 853
 854         module_put(tsk->thread_info->exec_domain->module);
 855         if (tsk->binfmt)
 856                 module_put(tsk->binfmt->module);
 857
 858         tsk->exit_code = code;
 859         exit_notify(tsk);
 860 #ifdef CONFIG_NUMA
 861         mpol_free(tsk->mempolicy);
 862         tsk->mempolicy = NULL;
 863 #endif
 864
 865         BUG_ON(!(current->flags & PF_DEAD));
 866         schedule();
 867         BUG();
 868         /* Avoid "noreturn function does return".  */
 869         for (;;) ;
 870 }
 871
 872 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 873 {
 874         if (comp)
 875                 complete(comp);
 876
 877         do_exit(code);
 878 }
 879
 880 EXPORT_SYMBOL(complete_and_exit);
 881
 882 asmlinkage long sys_exit(int error_code)
 883 {
 884         do_exit((error_code&0xff)<<8);
 885 }
 886
 887 task_t fastcall *next_thread(const task_t *p)
 888 {
 889 #ifdef CONFIG_SMP
 890         if (!p->sighand)
 891                 BUG();
 892         if (!spin_is_locked(&p->sighand->siglock) &&
 893                                 !rwlock_is_locked(&tasklist_lock))
 894                 BUG();
 895 #endif
 896         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 897 }
 898
 899 EXPORT_SYMBOL(next_thread);
 900
 901 /*
 902  * Take down every thread in the group.  This is called by fatal signals
 903  * as well as by sys_exit_group (below).
 904  */
 905 NORET_TYPE void
 906 do_group_exit(int exit_code)
 907 {
 908         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 909
 910         if (current->signal->group_exit)
 911                 exit_code = current->signal->group_exit_code;
 912         else if (!thread_group_empty(current)) {
 913                 struct signal_struct *const sig = current->signal;
 914                 struct sighand_struct *const sighand = current->sighand;
 915                 read_lock(&tasklist_lock);
 916                 spin_lock_irq(&sighand->siglock);
 917                 if (sig->group_exit)
 918                         /* Another thread got here before we took the lock.  */
 919                         exit_code = sig->group_exit_code;
 920                 else {
 921                         sig->group_exit = 1;
 922                         sig->group_exit_code = exit_code;
 923                         zap_other_threads(current);
 924                 }
 925                 spin_unlock_irq(&sighand->siglock);
 926                 read_unlock(&tasklist_lock);
 927         }
 928
 929         do_exit(exit_code);
 930         /* NOTREACHED */
 931 }
 932
 933 /*
 934  * this kills every thread in the thread group. Note that any externally
 935  * wait4()-ing process will get the correct exit code - even if this
 936  * thread is not the thread group leader.
 937  */
 938 asmlinkage void sys_exit_group(int error_code)
 939 {
 940         do_group_exit((error_code & 0xff) << 8);
 941 }
 942
 943 static int eligible_child(pid_t pid, int options, task_t *p)
 944 {
 945         if (pid > 0) {
 946                 if (p->pid != pid)
 947                         return 0;
 948         } else if (!pid) {
 949                 if (process_group(p) != process_group(current))
 950                         return 0;
 951         } else if (pid != -1) {
 952                 if (process_group(p) != -pid)
 953                         return 0;
 954         }
 955
 956         /*
 957          * Do not consider detached threads that are
 958          * not ptraced:
 959          */
 960         if (p->exit_signal == -1 && !p->ptrace)
 961                 return 0;
 962
 963         /* Wait for all children (clone and not) if __WALL is set;
 964          * otherwise, wait for clone children *only* if __WCLONE is
 965          * set; otherwise, wait for non-clone children *only*.  (Note:
 966          * A "clone" child here is one that reports to its parent
 967          * using a signal other than SIGCHLD.) */
 968         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 969             && !(options & __WALL))
 970                 return 0;
 971         /*
 972          * Do not consider thread group leaders that are
 973          * in a non-empty thread group:
 974          */
 975         if (current->tgid != p->tgid && delay_group_leader(p))
 976                 return 2;
 977
 978         if (security_task_wait(p))
 979                 return 0;
 980
 981         return 1;
 982 }
 983
 984 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 985                                int why, int status,
 986                                struct siginfo __user *infop,
 987                                struct rusage __user *rusagep)
 988 {
 989         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
 990         put_task_struct(p);
 991         if (!retval)
 992                 retval = put_user(SIGCHLD, &infop->si_signo);
 993         if (!retval)
 994                 retval = put_user(0, &infop->si_errno);
 995         if (!retval)
 996                 retval = put_user((short)why, &infop->si_code);
 997         if (!retval)
 998                 retval = put_user(pid, &infop->si_pid);
 999         if (!retval)
1000                 retval = put_user(uid, &infop->si_uid);
1001         if (!retval)
1002                 retval = put_user(status, &infop->si_status);
1003         if (!retval)
1004                 retval = pid;
1005         return retval;
1006 }
1007
1008 /*
1009  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1010  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1011  * the lock and this task is uninteresting.  If we return nonzero, we have
1012  * released the lock and the system call should return.
1013  */
1014 static int wait_task_zombie(task_t *p, int noreap,
1015                             struct siginfo __user *infop,
1016                             int __user *stat_addr, struct rusage __user *ru)
1017 {
1018         unsigned long state;
1019         int retval;
1020         int status;
1021
1022         if (unlikely(noreap)) {
1023                 pid_t pid = p->pid;
1024                 uid_t uid = p->uid;
1025                 int exit_code = p->exit_code;
1026                 int why, status;
1027
1028                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1029                         return 0;
1030                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1031                         return 0;
1032                 get_task_struct(p);
1033                 read_unlock(&tasklist_lock);
1034                 if ((exit_code & 0x7f) == 0) {
1035                         why = CLD_EXITED;
1036                         status = exit_code >> 8;
1037                 } else {
1038                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1039                         status = exit_code & 0x7f;
1040                 }
1041                 return wait_noreap_copyout(p, pid, uid, why,
1042                                            status, infop, ru);
1043         }
1044
1045         /*
1046          * Try to move the task's state to DEAD
1047          * only one thread is allowed to do this:
1048          */
1049         state = xchg(&p->exit_state, EXIT_DEAD);
1050         if (state != EXIT_ZOMBIE) {
1051                 BUG_ON(state != EXIT_DEAD);
1052                 return 0;
1053         }
1054         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1055                 /*
1056                  * This can only happen in a race with a ptraced thread
1057                  * dying on another processor.
1058                  */
1059                 return 0;
1060         }
1061
1062         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1063                 /*
1064                  * The resource counters for the group leader are in its
1065                  * own task_struct.  Those for dead threads in the group
1066                  * are in its signal_struct, as are those for the child
1067                  * processes it has previously reaped.  All these
1068                  * accumulate in the parent's signal_struct c* fields.
1069                  *
1070                  * We don't bother to take a lock here to protect these
1071                  * p->signal fields, because they are only touched by
1072                  * __exit_signal, which runs with tasklist_lock
1073                  * write-locked anyway, and so is excluded here.  We do
1074                  * need to protect the access to p->parent->signal fields,
1075                  * as other threads in the parent group can be right
1076                  * here reaping other children at the same time.
1077                  */
1078                 spin_lock_irq(&p->parent->sighand->siglock);
1079                 p->parent->signal->cutime +=
1080                         p->utime + p->signal->utime + p->signal->cutime;
1081                 p->parent->signal->cstime +=
1082                         p->stime + p->signal->stime + p->signal->cstime;
1083                 p->parent->signal->cmin_flt +=
1084                         p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
1085                 p->parent->signal->cmaj_flt +=
1086                         p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
1087                 p->parent->signal->cnvcsw +=
1088                         p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
1089                 p->parent->signal->cnivcsw +=
1090                         p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
1091                 spin_unlock_irq(&p->parent->sighand->siglock);
1092         }
1093
1094         /*
1095          * Now we are sure this task is interesting, and no other
1096          * thread can reap it because we set its state to EXIT_DEAD.
1097          */
1098         read_unlock(&tasklist_lock);
1099
1100         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1101         status = p->signal->group_exit
1102                 ? p->signal->group_exit_code : p->exit_code;
1103         if (!retval && stat_addr)
1104                 retval = put_user(status, stat_addr);
1105         if (!retval && infop)
1106                 retval = put_user(SIGCHLD, &infop->si_signo);
1107         if (!retval && infop)
1108                 retval = put_user(0, &infop->si_errno);
1109         if (!retval && infop) {
1110                 int why;
1111
1112                 if ((status & 0x7f) == 0) {
1113                         why = CLD_EXITED;
1114                         status >>= 8;
1115                 } else {
1116                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1117                         status &= 0x7f;
1118                 }
1119                 retval = put_user((short)why, &infop->si_code);
1120                 if (!retval)
1121                         retval = put_user(status, &infop->si_status);
1122         }
1123         if (!retval && infop)
1124                 retval = put_user(p->pid, &infop->si_pid);
1125         if (!retval && infop)
1126                 retval = put_user(p->uid, &infop->si_uid);
1127         if (retval) {
1128                 // TODO: is this safe?
1129                 p->exit_state = EXIT_ZOMBIE;
1130                 return retval;
1131         }
1132         retval = p->pid;
1133         if (p->real_parent != p->parent) {
1134                 write_lock_irq(&tasklist_lock);
1135                 /* Double-check with lock held.  */
1136                 if (p->real_parent != p->parent) {
1137                         __ptrace_unlink(p);
1138                         // TODO: is this safe?
1139                         p->exit_state = EXIT_ZOMBIE;
1140                         /*
1141                          * If this is not a detached task, notify the parent.
1142                          * If it's still not detached after that, don't release
1143                          * it now.
1144                          */
1145                         if (p->exit_signal != -1) {
1146                                 do_notify_parent(p, p->exit_signal);
1147                                 if (p->exit_signal != -1)
1148                                         p = NULL;
1149                         }
1150                 }
1151                 write_unlock_irq(&tasklist_lock);
1152         }
1153         if (p != NULL)
1154                 release_task(p);
1155         BUG_ON(!retval);
1156         return retval;
1157 }
1158
1159 /*
1160  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1161  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1162  * the lock and this task is uninteresting.  If we return nonzero, we have
1163  * released the lock and the system call should return.
1164  */
1165 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1166                              struct siginfo __user *infop,
1167                              int __user *stat_addr, struct rusage __user *ru)
1168 {
1169         int retval, exit_code;
1170
1171         if (!p->exit_code)
1172                 return 0;
1173         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1174             p->signal && p->signal->group_stop_count > 0)
1175                 /*
1176                  * A group stop is in progress and this is the group leader.
1177                  * We won't report until all threads have stopped.
1178                  */
1179                 return 0;
1180
1181         /*
1182          * Now we are pretty sure this task is interesting.
1183          * Make sure it doesn't get reaped out from under us while we
1184          * give up the lock and then examine it below.  We don't want to
1185          * keep holding onto the tasklist_lock while we call getrusage and
1186          * possibly take page faults for user memory.
1187          */
1188         get_task_struct(p);
1189         read_unlock(&tasklist_lock);
1190
1191         if (unlikely(noreap)) {
1192                 pid_t pid = p->pid;
1193                 uid_t uid = p->uid;
1194                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1195
1196                 exit_code = p->exit_code;
1197                 if (unlikely(!exit_code) ||
1198                     unlikely(p->state > TASK_STOPPED))
1199                         goto bail_ref;
1200                 return wait_noreap_copyout(p, pid, uid,
1201                                            why, (exit_code << 8) | 0x7f,
1202                                            infop, ru);
1203         }
1204
1205         write_lock_irq(&tasklist_lock);
1206
1207         /*
1208          * This uses xchg to be atomic with the thread resuming and setting
1209          * it.  It must also be done with the write lock held to prevent a
1210          * race with the EXIT_ZOMBIE case.
1211          */
1212         exit_code = xchg(&p->exit_code, 0);
1213         if (unlikely(p->exit_state >= EXIT_ZOMBIE)) {
1214                 /*
1215                  * The task resumed and then died.  Let the next iteration
1216                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1217                  * already be zero here if it resumed and did _exit(0).
1218                  * The task itself is dead and won't touch exit_code again;
1219                  * other processors in this function are locked out.
1220                  */
1221                 p->exit_code = exit_code;
1222                 exit_code = 0;
1223         }
1224         if (unlikely(exit_code == 0)) {
1225                 /*
1226                  * Another thread in this function got to it first, or it
1227                  * resumed, or it resumed and then died.
1228                  */
1229                 write_unlock_irq(&tasklist_lock);
1230 bail_ref:
1231                 put_task_struct(p);
1232                 /*
1233                  * We are returning to the wait loop without having successfully
1234                  * removed the process and having released the lock. We cannot
1235                  * continue, since the "p" task pointer is potentially stale.
1236                  *
1237                  * Return -EAGAIN, and do_wait() will restart the loop from the
1238                  * beginning. Do _not_ re-acquire the lock.
1239                  */
1240                 return -EAGAIN;
1241         }
1242
1243         /* move to end of parent's list to avoid starvation */
1244         remove_parent(p);
1245         add_parent(p, p->parent);
1246
1247         write_unlock_irq(&tasklist_lock);
1248
1249         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1250         if (!retval && stat_addr)
1251                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1252         if (!retval && infop)
1253                 retval = put_user(SIGCHLD, &infop->si_signo);
1254         if (!retval && infop)
1255                 retval = put_user(0, &infop->si_errno);
1256         if (!retval && infop)
1257                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1258                                           ? CLD_TRAPPED : CLD_STOPPED),
1259                                   &infop->si_code);
1260         if (!retval && infop)
1261                 retval = put_user(exit_code, &infop->si_status);
1262         if (!retval && infop)
1263                 retval = put_user(p->pid, &infop->si_pid);
1264         if (!retval && infop)
1265                 retval = put_user(p->uid, &infop->si_uid);
1266         if (!retval)
1267                 retval = p->pid;
1268         put_task_struct(p);
1269
1270         BUG_ON(!retval);
1271         return retval;
1272 }
1273
1274 /*
1275  * Handle do_wait work for one task in a live, non-stopped state.
1276  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1277  * the lock and this task is uninteresting.  If we return nonzero, we have
1278  * released the lock and the system call should return.
1279  */
1280 static int wait_task_continued(task_t *p, int noreap,
1281                                struct siginfo __user *infop,
1282                                int __user *stat_addr, struct rusage __user *ru)
1283 {
1284         int retval;
1285         pid_t pid;
1286         uid_t uid;
1287
1288         if (unlikely(!p->signal))
1289                 return 0;
1290
1291         if (p->signal->stop_state >= 0)
1292                 return 0;
1293
1294         spin_lock_irq(&p->sighand->siglock);
1295         if (p->signal->stop_state >= 0) { /* Re-check with the lock held.  */
1296                 spin_unlock_irq(&p->sighand->siglock);
1297                 return 0;
1298         }
1299         if (!noreap)
1300                 p->signal->stop_state = 0;
1301         spin_unlock_irq(&p->sighand->siglock);
1302
1303         pid = p->pid;
1304         uid = p->uid;
1305         get_task_struct(p);
1306         read_unlock(&tasklist_lock);
1307
1308         if (!infop) {
1309                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1310                 put_task_struct(p);
1311                 if (!retval && stat_addr)
1312                         retval = put_user(0xffff, stat_addr);
1313                 if (!retval)
1314                         retval = p->pid;
1315         } else {
1316                 retval = wait_noreap_copyout(p, pid, uid,
1317                                              CLD_CONTINUED, SIGCONT,
1318                                              infop, ru);
1319                 BUG_ON(retval == 0);
1320         }
1321
1322         return retval;
1323 }
1324
1325
1326 static inline int my_ptrace_child(struct task_struct *p)
1327 {
1328         if (!(p->ptrace & PT_PTRACED))
1329                 return 0;
1330         if (!(p->ptrace & PT_ATTACHED))
1331                 return 1;
1332         /*
1333          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1334          * we are the attacher.  If we are the real parent, this is a race
1335          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1336          * which we have to switch the parent links, but has already set
1337          * the flags in p->ptrace.
1338          */
1339         return (p->parent != p->real_parent);
1340 }
1341
1342 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1343                     int __user *stat_addr, struct rusage __user *ru)
1344 {
1345         DECLARE_WAITQUEUE(wait, current);
1346         struct task_struct *tsk;
1347         int flag, retval;
1348
1349         add_wait_queue(&current->wait_chldexit,&wait);
1350 repeat:
1351         /*
1352          * We will set this flag if we see any child that might later
1353          * match our criteria, even if we are not able to reap it yet.
1354          */
1355         flag = 0;
1356         current->state = TASK_INTERRUPTIBLE;
1357         read_lock(&tasklist_lock);
1358         tsk = current;
1359         do {
1360                 struct task_struct *p;
1361                 struct list_head *_p;
1362                 int ret;
1363
1364                 list_for_each(_p,&tsk->children) {
1365                         p = list_entry(_p,struct task_struct,sibling);
1366
1367                         ret = eligible_child(pid, options, p);
1368                         if (!ret)
1369                                 continue;
1370
1371                         switch (p->state) {
1372                         case TASK_TRACED:
1373                                 if (!my_ptrace_child(p))
1374                                         continue;
1375                                 /*FALLTHROUGH*/
1376                         case TASK_STOPPED:
1377                                 /*
1378                                  * It's stopped now, so it might later
1379                                  * continue, exit, or stop again.
1380                                  */
1381                                 flag = 1;
1382                                 if (!(options & WUNTRACED) &&
1383                                     !my_ptrace_child(p))
1384                                         continue;
1385                                 retval = wait_task_stopped(p, ret == 2,
1386                                                            (options & WNOWAIT),
1387                                                            infop,
1388                                                            stat_addr, ru);
1389                                 if (retval == -EAGAIN)
1390                                         goto repeat;
1391                                 if (retval != 0) /* He released the lock.  */
1392                                         goto end;
1393                                 break;
1394                         default:
1395                         // case EXIT_DEAD:
1396                                 if (p->exit_state == EXIT_DEAD)
1397                                         continue;
1398                         // case EXIT_ZOMBIE:
1399                                 if (p->exit_state == EXIT_ZOMBIE) {
1400                                         /*
1401                                          * Eligible but we cannot release
1402                                          * it yet:
1403                                          */
1404                                         if (ret == 2)
1405                                                 goto check_continued;
1406                                         if (!likely(options & WEXITED))
1407                                                 continue;
1408                                         retval = wait_task_zombie(
1409                                                 p, (options & WNOWAIT),
1410                                                 infop, stat_addr, ru);
1411                                         /* He released the lock.  */
1412                                         if (retval != 0)
1413                                                 goto end;
1414                                         break;
1415                                 }
1416 check_continued:
1417                                 /*
1418                                  * It's running now, so it might later
1419                                  * exit, stop, or stop and then continue.
1420                                  */
1421                                 flag = 1;
1422                                 if (!unlikely(options & WCONTINUED))
1423                                         continue;
1424
1425                                 retval = wait_task_continued(
1426                                         p, (options & WNOWAIT),
1427                                         infop, stat_addr, ru);
1428                                 if (retval != 0) /* He released the lock.  */
1429                                         goto end;
1430                                 break;
1431                         }
1432                 }
1433                 if (!flag) {
1434                         list_for_each(_p, &tsk->ptrace_children) {
1435                                 p = list_entry(_p, struct task_struct,
1436                                                 ptrace_list);
1437                                 if (!eligible_child(pid, options, p))
1438                                         continue;
1439                                 flag = 1;
1440                                 break;
1441                         }
1442                 }
1443                 if (options & __WNOTHREAD)
1444                         break;
1445                 tsk = next_thread(tsk);
1446                 if (tsk->signal != current->signal)
1447                         BUG();
1448         } while (tsk != current);
1449
1450         read_unlock(&tasklist_lock);
1451         if (flag) {
1452                 retval = 0;
1453                 if (options & WNOHANG)
1454                         goto end;
1455                 retval = -ERESTARTSYS;
1456                 if (signal_pending(current))
1457                         goto end;
1458                 schedule();
1459                 goto repeat;
1460         }
1461         retval = -ECHILD;
1462 end:
1463         current->state = TASK_RUNNING;
1464         remove_wait_queue(&current->wait_chldexit,&wait);
1465         if (infop) {
1466                 if (retval > 0)
1467                 retval = 0;
1468                 else {
1469                         /*
1470                          * For a WNOHANG return, clear out all the fields
1471                          * we would set so the user can easily tell the
1472                          * difference.
1473                          */
1474                         if (!retval)
1475                                 retval = put_user(0, &infop->si_signo);
1476                         if (!retval)
1477                                 retval = put_user(0, &infop->si_errno);
1478                         if (!retval)
1479                                 retval = put_user(0, &infop->si_code);
1480                         if (!retval)
1481                                 retval = put_user(0, &infop->si_pid);
1482                         if (!retval)
1483                                 retval = put_user(0, &infop->si_uid);
1484                         if (!retval)
1485                                 retval = put_user(0, &infop->si_status);
1486                 }
1487         }
1488         return retval;
1489 }
1490
1491 asmlinkage long sys_waitid(int which, pid_t pid,
1492                            struct siginfo __user *infop, int options,
1493                            struct rusage __user *ru)
1494 {
1495         long ret;
1496
1497         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1498                 return -EINVAL;
1499         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1500                 return -EINVAL;
1501
1502         switch (which) {
1503         case P_ALL:
1504                 pid = -1;
1505                 break;
1506         case P_PID:
1507                 if (pid <= 0)
1508                         return -EINVAL;
1509                 break;
1510         case P_PGID:
1511                 if (pid <= 0)
1512                         return -EINVAL;
1513                 pid = -pid;
1514                 break;
1515         default:
1516                 return -EINVAL;
1517         }
1518
1519         ret = do_wait(pid, options, infop, NULL, ru);
1520
1521         /* avoid REGPARM breakage on x86: */
1522         prevent_tail_call(ret);
1523         return ret;
1524 }
1525
1526 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1527                           int options, struct rusage __user *ru)
1528 {
1529         long ret;
1530
1531         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1532                         __WNOTHREAD|__WCLONE|__WALL))
1533                 return -EINVAL;
1534         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1535
1536         /* avoid REGPARM breakage on x86: */
1537         prevent_tail_call(ret);
1538         return ret;
1539 }
1540
1541 #ifdef __ARCH_WANT_SYS_WAITPID
1542
1543 /*
1544  * sys_waitpid() remains for compatibility. waitpid() should be
1545  * implemented by calling sys_wait4() from libc.a.
1546  */
1547 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1548 {
1549         return sys_wait4(pid, stat_addr, options, NULL);
1550 }
1551
1552 #endif