kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/mm.h>
   8 #include <linux/slab.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/module.h>
  11 #include <linux/capability.h>
  12 #include <linux/completion.h>
  13 #include <linux/personality.h>
  14 #include <linux/tty.h>
  15 #include <linux/mnt_namespace.h>
  16 #include <linux/key.h>
  17 #include <linux/security.h>
  18 #include <linux/cpu.h>
  19 #include <linux/acct.h>
  20 #include <linux/tsacct_kern.h>
  21 #include <linux/file.h>
  22 #include <linux/binfmts.h>
  23 #include <linux/nsproxy.h>
  24 #include <linux/pid_namespace.h>
  25 #include <linux/ptrace.h>
  26 #include <linux/profile.h>
  27 #include <linux/signalfd.h>
  28 #include <linux/mount.h>
  29 #include <linux/proc_fs.h>
  30 #include <linux/kthread.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/taskstats_kern.h>
  33 #include <linux/delayacct.h>
  34 #include <linux/cpuset.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/signal.h>
  37 #include <linux/posix-timers.h>
  38 #include <linux/cn_proc.h>
  39 #include <linux/mutex.h>
  40 #include <linux/futex.h>
  41 #include <linux/compat.h>
  42 #include <linux/pipe_fs_i.h>
  43 #include <linux/audit.h> /* for audit_free() */
  44 #include <linux/resource.h>
  45 #include <linux/blkdev.h>
  46 #include <linux/task_io_accounting_ops.h>
  47
  48 #include <asm/uaccess.h>
  49 #include <asm/unistd.h>
  50 #include <asm/pgtable.h>
  51 #include <asm/mmu_context.h>
  52
  53 extern void sem_exit (void);
  54
  55 static void exit_mm(struct task_struct * tsk);
  56
  57 static void __unhash_process(struct task_struct *p)
  58 {
  59         nr_threads--;
  60         detach_pid(p, PIDTYPE_PID);
  61         if (thread_group_leader(p)) {
  62                 detach_pid(p, PIDTYPE_PGID);
  63                 detach_pid(p, PIDTYPE_SID);
  64
  65                 list_del_rcu(&p->tasks);
  66                 __get_cpu_var(process_counts)--;
  67         }
  68         list_del_rcu(&p->thread_group);
  69         remove_parent(p);
  70 }
  71
  72 /*
  73  * This function expects the tasklist_lock write-locked.
  74  */
  75 static void __exit_signal(struct task_struct *tsk)
  76 {
  77         struct signal_struct *sig = tsk->signal;
  78         struct sighand_struct *sighand;
  79
  80         BUG_ON(!sig);
  81         BUG_ON(!atomic_read(&sig->count));
  82
  83         rcu_read_lock();
  84         sighand = rcu_dereference(tsk->sighand);
  85         spin_lock(&sighand->siglock);
  86
  87         /*
  88          * Notify that this sighand has been detached. This must
  89          * be called with the tsk->sighand lock held. Also, this
  90          * access tsk->sighand internally, so it must be called
  91          * before tsk->sighand is reset.
  92          */
  93         signalfd_detach_locked(tsk);
  94
  95         posix_cpu_timers_exit(tsk);
  96         if (atomic_dec_and_test(&sig->count))
  97                 posix_cpu_timers_exit_group(tsk);
  98         else {
  99                 /*
 100                  * If there is any task waiting for the group exit
 101                  * then notify it:
 102                  */
 103                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
 104                         wake_up_process(sig->group_exit_task);
 105                         sig->group_exit_task = NULL;
 106                 }
 107                 if (tsk == sig->curr_target)
 108                         sig->curr_target = next_thread(tsk);
 109                 /*
 110                  * Accumulate here the counters for all threads but the
 111                  * group leader as they die, so they can be added into
 112                  * the process-wide totals when those are taken.
 113                  * The group leader stays around as a zombie as long
 114                  * as there are other threads.  When it gets reaped,
 115                  * the exit.c code will add its counts into these totals.
 116                  * We won't ever get here for the group leader, since it
 117                  * will have been the last reference on the signal_struct.
 118                  */
 119                 sig->utime = cputime_add(sig->utime, tsk->utime);
 120                 sig->stime = cputime_add(sig->stime, tsk->stime);
 121                 sig->min_flt += tsk->min_flt;
 122                 sig->maj_flt += tsk->maj_flt;
 123                 sig->nvcsw += tsk->nvcsw;
 124                 sig->nivcsw += tsk->nivcsw;
 125                 sig->sched_time += tsk->sched_time;
 126                 sig->inblock += task_io_get_inblock(tsk);
 127                 sig->oublock += task_io_get_oublock(tsk);
 128                 sig = NULL; /* Marker for below. */
 129         }
 130
 131         __unhash_process(tsk);
 132
 133         tsk->signal = NULL;
 134         tsk->sighand = NULL;
 135         spin_unlock(&sighand->siglock);
 136         rcu_read_unlock();
 137
 138         __cleanup_sighand(sighand);
 139         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 140         flush_sigqueue(&tsk->pending);
 141         if (sig) {
 142                 flush_sigqueue(&sig->shared_pending);
 143                 taskstats_tgid_free(sig);
 144                 __cleanup_signal(sig);
 145         }
 146 }
 147
 148 static void delayed_put_task_struct(struct rcu_head *rhp)
 149 {
 150         put_task_struct(container_of(rhp, struct task_struct, rcu));
 151 }
 152
 153 void release_task(struct task_struct * p)
 154 {
 155         struct task_struct *leader;
 156         int zap_leader;
 157 repeat:
 158         atomic_dec(&p->user->processes);
 159         write_lock_irq(&tasklist_lock);
 160         ptrace_unlink(p);
 161         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 162         __exit_signal(p);
 163
 164         /*
 165          * If we are the last non-leader member of the thread
 166          * group, and the leader is zombie, then notify the
 167          * group leader's parent process. (if it wants notification.)
 168          */
 169         zap_leader = 0;
 170         leader = p->group_leader;
 171         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 172                 BUG_ON(leader->exit_signal == -1);
 173                 do_notify_parent(leader, leader->exit_signal);
 174                 /*
 175                  * If we were the last child thread and the leader has
 176                  * exited already, and the leader's parent ignores SIGCHLD,
 177                  * then we are the one who should release the leader.
 178                  *
 179                  * do_notify_parent() will have marked it self-reaping in
 180                  * that case.
 181                  */
 182                 zap_leader = (leader->exit_signal == -1);
 183         }
 184
 185         sched_exit(p);
 186         write_unlock_irq(&tasklist_lock);
 187         proc_flush_task(p);
 188         release_thread(p);
 189         call_rcu(&p->rcu, delayed_put_task_struct);
 190
 191         p = leader;
 192         if (unlikely(zap_leader))
 193                 goto repeat;
 194 }
 195
 196 /*
 197  * This checks not only the pgrp, but falls back on the pid if no
 198  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 199  * without this...
 200  *
 201  * The caller must hold rcu lock or the tasklist lock.
 202  */
 203 struct pid *session_of_pgrp(struct pid *pgrp)
 204 {
 205         struct task_struct *p;
 206         struct pid *sid = NULL;
 207
 208         p = pid_task(pgrp, PIDTYPE_PGID);
 209         if (p == NULL)
 210                 p = pid_task(pgrp, PIDTYPE_PID);
 211         if (p != NULL)
 212                 sid = task_session(p);
 213
 214         return sid;
 215 }
 216
 217 /*
 218  * Determine if a process group is "orphaned", according to the POSIX
 219  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 220  * by terminal-generated stop signals.  Newly orphaned process groups are
 221  * to receive a SIGHUP and a SIGCONT.
 222  *
 223  * "I ask you, have you ever known what it is to be an orphan?"
 224  */
 225 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 226 {
 227         struct task_struct *p;
 228         int ret = 1;
 229
 230         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 231                 if (p == ignored_task
 232                                 || p->exit_state
 233                                 || is_init(p->real_parent))
 234                         continue;
 235                 if (task_pgrp(p->real_parent) != pgrp &&
 236                     task_session(p->real_parent) == task_session(p)) {
 237                         ret = 0;
 238                         break;
 239                 }
 240         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 241         return ret;     /* (sighing) "Often!" */
 242 }
 243
 244 int is_current_pgrp_orphaned(void)
 245 {
 246         int retval;
 247
 248         read_lock(&tasklist_lock);
 249         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 250         read_unlock(&tasklist_lock);
 251
 252         return retval;
 253 }
 254
 255 static int has_stopped_jobs(struct pid *pgrp)
 256 {
 257         int retval = 0;
 258         struct task_struct *p;
 259
 260         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 261                 if (p->state != TASK_STOPPED)
 262                         continue;
 263                 retval = 1;
 264                 break;
 265         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 266         return retval;
 267 }
 268
 269 /**
 270  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
 271  *
 272  * If a kernel thread is launched as a result of a system call, or if
 273  * it ever exits, it should generally reparent itself to kthreadd so it
 274  * isn't in the way of other processes and is correctly cleaned up on exit.
 275  *
 276  * The various task state such as scheduling policy and priority may have
 277  * been inherited from a user process, so we reset them to sane values here.
 278  *
 279  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
 280  */
 281 static void reparent_to_kthreadd(void)
 282 {
 283         write_lock_irq(&tasklist_lock);
 284
 285         ptrace_unlink(current);
 286         /* Reparent to init */
 287         remove_parent(current);
 288         current->real_parent = current->parent = kthreadd_task;
 289         add_parent(current);
 290
 291         /* Set the exit signal to SIGCHLD so we signal init on exit */
 292         current->exit_signal = SIGCHLD;
 293
 294         if (!has_rt_policy(current) && (task_nice(current) < 0))
 295                 set_user_nice(current, 0);
 296         /* cpus_allowed? */
 297         /* rt_priority? */
 298         /* signals? */
 299         security_task_reparent_to_init(current);
 300         memcpy(current->signal->rlim, init_task.signal->rlim,
 301                sizeof(current->signal->rlim));
 302         atomic_inc(&(INIT_USER->__count));
 303         write_unlock_irq(&tasklist_lock);
 304         switch_uid(INIT_USER);
 305 }
 306
 307 void __set_special_pids(pid_t session, pid_t pgrp)
 308 {
 309         struct task_struct *curr = current->group_leader;
 310
 311         if (process_session(curr) != session) {
 312                 detach_pid(curr, PIDTYPE_SID);
 313                 set_signal_session(curr->signal, session);
 314                 attach_pid(curr, PIDTYPE_SID, find_pid(session));
 315         }
 316         if (process_group(curr) != pgrp) {
 317                 detach_pid(curr, PIDTYPE_PGID);
 318                 curr->signal->pgrp = pgrp;
 319                 attach_pid(curr, PIDTYPE_PGID, find_pid(pgrp));
 320         }
 321 }
 322
 323 static void set_special_pids(pid_t session, pid_t pgrp)
 324 {
 325         write_lock_irq(&tasklist_lock);
 326         __set_special_pids(session, pgrp);
 327         write_unlock_irq(&tasklist_lock);
 328 }
 329
 330 /*
 331  * Let kernel threads use this to say that they
 332  * allow a certain signal (since daemonize() will
 333  * have disabled all of them by default).
 334  */
 335 int allow_signal(int sig)
 336 {
 337         if (!valid_signal(sig) || sig < 1)
 338                 return -EINVAL;
 339
 340         spin_lock_irq(&current->sighand->siglock);
 341         sigdelset(&current->blocked, sig);
 342         if (!current->mm) {
 343                 /* Kernel threads handle their own signals.
 344                    Let the signal code know it'll be handled, so
 345                    that they don't get converted to SIGKILL or
 346                    just silently dropped */
 347                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 348         }
 349         recalc_sigpending();
 350         spin_unlock_irq(&current->sighand->siglock);
 351         return 0;
 352 }
 353
 354 EXPORT_SYMBOL(allow_signal);
 355
 356 int disallow_signal(int sig)
 357 {
 358         if (!valid_signal(sig) || sig < 1)
 359                 return -EINVAL;
 360
 361         spin_lock_irq(&current->sighand->siglock);
 362         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
 363         recalc_sigpending();
 364         spin_unlock_irq(&current->sighand->siglock);
 365         return 0;
 366 }
 367
 368 EXPORT_SYMBOL(disallow_signal);
 369
 370 /*
 371  *      Put all the gunge required to become a kernel thread without
 372  *      attached user resources in one place where it belongs.
 373  */
 374
 375 void daemonize(const char *name, ...)
 376 {
 377         va_list args;
 378         struct fs_struct *fs;
 379         sigset_t blocked;
 380
 381         va_start(args, name);
 382         vsnprintf(current->comm, sizeof(current->comm), name, args);
 383         va_end(args);
 384
 385         /*
 386          * If we were started as result of loading a module, close all of the
 387          * user space pages.  We don't need them, and if we didn't close them
 388          * they would be locked into memory.
 389          */
 390         exit_mm(current);
 391
 392         set_special_pids(1, 1);
 393         proc_clear_tty(current);
 394
 395         /* Block and flush all signals */
 396         sigfillset(&blocked);
 397         sigprocmask(SIG_BLOCK, &blocked, NULL);
 398         flush_signals(current);
 399
 400         /* Become as one with the init task */
 401
 402         exit_fs(current);       /* current->fs->count--; */
 403         fs = init_task.fs;
 404         current->fs = fs;
 405         atomic_inc(&fs->count);
 406
 407         exit_task_namespaces(current);
 408         current->nsproxy = init_task.nsproxy;
 409         get_task_namespaces(current);
 410
 411         exit_files(current);
 412         current->files = init_task.files;
 413         atomic_inc(&current->files->count);
 414
 415         reparent_to_kthreadd();
 416 }
 417
 418 EXPORT_SYMBOL(daemonize);
 419
 420 static void close_files(struct files_struct * files)
 421 {
 422         int i, j;
 423         struct fdtable *fdt;
 424
 425         j = 0;
 426
 427         /*
 428          * It is safe to dereference the fd table without RCU or
 429          * ->file_lock because this is the last reference to the
 430          * files structure.
 431          */
 432         fdt = files_fdtable(files);
 433         for (;;) {
 434                 unsigned long set;
 435                 i = j * __NFDBITS;
 436                 if (i >= fdt->max_fds)
 437                         break;
 438                 set = fdt->open_fds->fds_bits[j++];
 439                 while (set) {
 440                         if (set & 1) {
 441                                 struct file * file = xchg(&fdt->fd[i], NULL);
 442                                 if (file) {
 443                                         filp_close(file, files);
 444                                         cond_resched();
 445                                 }
 446                         }
 447                         i++;
 448                         set >>= 1;
 449                 }
 450         }
 451 }
 452
 453 struct files_struct *get_files_struct(struct task_struct *task)
 454 {
 455         struct files_struct *files;
 456
 457         task_lock(task);
 458         files = task->files;
 459         if (files)
 460                 atomic_inc(&files->count);
 461         task_unlock(task);
 462
 463         return files;
 464 }
 465
 466 void fastcall put_files_struct(struct files_struct *files)
 467 {
 468         struct fdtable *fdt;
 469
 470         if (atomic_dec_and_test(&files->count)) {
 471                 close_files(files);
 472                 /*
 473                  * Free the fd and fdset arrays if we expanded them.
 474                  * If the fdtable was embedded, pass files for freeing
 475                  * at the end of the RCU grace period. Otherwise,
 476                  * you can free files immediately.
 477                  */
 478                 fdt = files_fdtable(files);
 479                 if (fdt != &files->fdtab)
 480                         kmem_cache_free(files_cachep, files);
 481                 free_fdtable(fdt);
 482         }
 483 }
 484
 485 EXPORT_SYMBOL(put_files_struct);
 486
 487 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
 488 {
 489         struct files_struct *old;
 490
 491         old = tsk->files;
 492         task_lock(tsk);
 493         tsk->files = files;
 494         task_unlock(tsk);
 495         put_files_struct(old);
 496 }
 497 EXPORT_SYMBOL(reset_files_struct);
 498
 499 static inline void __exit_files(struct task_struct *tsk)
 500 {
 501         struct files_struct * files = tsk->files;
 502
 503         if (files) {
 504                 task_lock(tsk);
 505                 tsk->files = NULL;
 506                 task_unlock(tsk);
 507                 put_files_struct(files);
 508         }
 509 }
 510
 511 void exit_files(struct task_struct *tsk)
 512 {
 513         __exit_files(tsk);
 514 }
 515
 516 static inline void __put_fs_struct(struct fs_struct *fs)
 517 {
 518         /* No need to hold fs->lock if we are killing it */
 519         if (atomic_dec_and_test(&fs->count)) {
 520                 dput(fs->root);
 521                 mntput(fs->rootmnt);
 522                 dput(fs->pwd);
 523                 mntput(fs->pwdmnt);
 524                 if (fs->altroot) {
 525                         dput(fs->altroot);
 526                         mntput(fs->altrootmnt);
 527                 }
 528                 kmem_cache_free(fs_cachep, fs);
 529         }
 530 }
 531
 532 void put_fs_struct(struct fs_struct *fs)
 533 {
 534         __put_fs_struct(fs);
 535 }
 536
 537 static inline void __exit_fs(struct task_struct *tsk)
 538 {
 539         struct fs_struct * fs = tsk->fs;
 540
 541         if (fs) {
 542                 task_lock(tsk);
 543                 tsk->fs = NULL;
 544                 task_unlock(tsk);
 545                 __put_fs_struct(fs);
 546         }
 547 }
 548
 549 void exit_fs(struct task_struct *tsk)
 550 {
 551         __exit_fs(tsk);
 552 }
 553
 554 EXPORT_SYMBOL_GPL(exit_fs);
 555
 556 /*
 557  * Turn us into a lazy TLB process if we
 558  * aren't already..
 559  */
 560 static void exit_mm(struct task_struct * tsk)
 561 {
 562         struct mm_struct *mm = tsk->mm;
 563
 564         mm_release(tsk, mm);
 565         if (!mm)
 566                 return;
 567         /*
 568          * Serialize with any possible pending coredump.
 569          * We must hold mmap_sem around checking core_waiters
 570          * and clearing tsk->mm.  The core-inducing thread
 571          * will increment core_waiters for each thread in the
 572          * group with ->mm != NULL.
 573          */
 574         down_read(&mm->mmap_sem);
 575         if (mm->core_waiters) {
 576                 up_read(&mm->mmap_sem);
 577                 down_write(&mm->mmap_sem);
 578                 if (!--mm->core_waiters)
 579                         complete(mm->core_startup_done);
 580                 up_write(&mm->mmap_sem);
 581
 582                 wait_for_completion(&mm->core_done);
 583                 down_read(&mm->mmap_sem);
 584         }
 585         atomic_inc(&mm->mm_count);
 586         BUG_ON(mm != tsk->active_mm);
 587         /* more a memory barrier than a real lock */
 588         task_lock(tsk);
 589         tsk->mm = NULL;
 590         up_read(&mm->mmap_sem);
 591         enter_lazy_tlb(mm, current);
 592         task_unlock(tsk);
 593         mmput(mm);
 594 }
 595
 596 static inline void
 597 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 598 {
 599         /*
 600          * Make sure we're not reparenting to ourselves and that
 601          * the parent is not a zombie.
 602          */
 603         BUG_ON(p == reaper || reaper->exit_state);
 604         p->real_parent = reaper;
 605 }
 606
 607 static void
 608 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 609 {
 610         if (p->pdeath_signal)
 611                 /* We already hold the tasklist_lock here.  */
 612                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 613
 614         /* Move the child from its dying parent to the new one.  */
 615         if (unlikely(traced)) {
 616                 /* Preserve ptrace links if someone else is tracing this child.  */
 617                 list_del_init(&p->ptrace_list);
 618                 if (p->parent != p->real_parent)
 619                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 620         } else {
 621                 /* If this child is being traced, then we're the one tracing it
 622                  * anyway, so let go of it.
 623                  */
 624                 p->ptrace = 0;
 625                 remove_parent(p);
 626                 p->parent = p->real_parent;
 627                 add_parent(p);
 628
 629                 if (p->state == TASK_TRACED) {
 630                         /*
 631                          * If it was at a trace stop, turn it into
 632                          * a normal stop since it's no longer being
 633                          * traced.
 634                          */
 635                         ptrace_untrace(p);
 636                 }
 637         }
 638
 639         /* If this is a threaded reparent there is no need to
 640          * notify anyone anything has happened.
 641          */
 642         if (p->real_parent->group_leader == father->group_leader)
 643                 return;
 644
 645         /* We don't want people slaying init.  */
 646         if (p->exit_signal != -1)
 647                 p->exit_signal = SIGCHLD;
 648
 649         /* If we'd notified the old parent about this child's death,
 650          * also notify the new parent.
 651          */
 652         if (!traced && p->exit_state == EXIT_ZOMBIE &&
 653             p->exit_signal != -1 && thread_group_empty(p))
 654                 do_notify_parent(p, p->exit_signal);
 655
 656         /*
 657          * process group orphan check
 658          * Case ii: Our child is in a different pgrp
 659          * than we are, and it was the only connection
 660          * outside, so the child pgrp is now orphaned.
 661          */
 662         if ((task_pgrp(p) != task_pgrp(father)) &&
 663             (task_session(p) == task_session(father))) {
 664                 struct pid *pgrp = task_pgrp(p);
 665
 666                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
 667                     has_stopped_jobs(pgrp)) {
 668                         __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 669                         __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 670                 }
 671         }
 672 }
 673
 674 /*
 675  * When we die, we re-parent all our children.
 676  * Try to give them to another thread in our thread
 677  * group, and if no such member exists, give it to
 678  * the child reaper process (ie "init") in our pid
 679  * space.
 680  */
 681 static void
 682 forget_original_parent(struct task_struct *father, struct list_head *to_release)
 683 {
 684         struct task_struct *p, *reaper = father;
 685         struct list_head *_p, *_n;
 686
 687         do {
 688                 reaper = next_thread(reaper);
 689                 if (reaper == father) {
 690                         reaper = child_reaper(father);
 691                         break;
 692                 }
 693         } while (reaper->exit_state);
 694
 695         /*
 696          * There are only two places where our children can be:
 697          *
 698          * - in our child list
 699          * - in our ptraced child list
 700          *
 701          * Search them and reparent children.
 702          */
 703         list_for_each_safe(_p, _n, &father->children) {
 704                 int ptrace;
 705                 p = list_entry(_p, struct task_struct, sibling);
 706
 707                 ptrace = p->ptrace;
 708
 709                 /* if father isn't the real parent, then ptrace must be enabled */
 710                 BUG_ON(father != p->real_parent && !ptrace);
 711
 712                 if (father == p->real_parent) {
 713                         /* reparent with a reaper, real father it's us */
 714                         choose_new_parent(p, reaper);
 715                         reparent_thread(p, father, 0);
 716                 } else {
 717                         /* reparent ptraced task to its real parent */
 718                         __ptrace_unlink (p);
 719                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 720                             thread_group_empty(p))
 721                                 do_notify_parent(p, p->exit_signal);
 722                 }
 723
 724                 /*
 725                  * if the ptraced child is a zombie with exit_signal == -1
 726                  * we must collect it before we exit, or it will remain
 727                  * zombie forever since we prevented it from self-reap itself
 728                  * while it was being traced by us, to be able to see it in wait4.
 729                  */
 730                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 731                         list_add(&p->ptrace_list, to_release);
 732         }
 733         list_for_each_safe(_p, _n, &father->ptrace_children) {
 734                 p = list_entry(_p, struct task_struct, ptrace_list);
 735                 choose_new_parent(p, reaper);
 736                 reparent_thread(p, father, 1);
 737         }
 738 }
 739
 740 /*
 741  * Send signals to all our closest relatives so that they know
 742  * to properly mourn us..
 743  */
 744 static void exit_notify(struct task_struct *tsk)
 745 {
 746         int state;
 747         struct task_struct *t;
 748         struct list_head ptrace_dead, *_p, *_n;
 749         struct pid *pgrp;
 750
 751         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 752             && !thread_group_empty(tsk)) {
 753                 /*
 754                  * This occurs when there was a race between our exit
 755                  * syscall and a group signal choosing us as the one to
 756                  * wake up.  It could be that we are the only thread
 757                  * alerted to check for pending signals, but another thread
 758                  * should be woken now to take the signal since we will not.
 759                  * Now we'll wake all the threads in the group just to make
 760                  * sure someone gets all the pending signals.
 761                  */
 762                 read_lock(&tasklist_lock);
 763                 spin_lock_irq(&tsk->sighand->siglock);
 764                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 765                         if (!signal_pending(t) && !(t->flags & PF_EXITING))
 766                                 recalc_sigpending_and_wake(t);
 767                 spin_unlock_irq(&tsk->sighand->siglock);
 768                 read_unlock(&tasklist_lock);
 769         }
 770
 771         write_lock_irq(&tasklist_lock);
 772
 773         /*
 774          * This does two things:
 775          *
 776          * A.  Make init inherit all the child processes
 777          * B.  Check to see if any process groups have become orphaned
 778          *      as a result of our exiting, and if they have any stopped
 779          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 780          */
 781
 782         INIT_LIST_HEAD(&ptrace_dead);
 783         forget_original_parent(tsk, &ptrace_dead);
 784         BUG_ON(!list_empty(&tsk->children));
 785         BUG_ON(!list_empty(&tsk->ptrace_children));
 786
 787         /*
 788          * Check to see if any process groups have become orphaned
 789          * as a result of our exiting, and if they have any stopped
 790          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 791          *
 792          * Case i: Our father is in a different pgrp than we are
 793          * and we were the only connection outside, so our pgrp
 794          * is about to become orphaned.
 795          */
 796
 797         t = tsk->real_parent;
 798
 799         pgrp = task_pgrp(tsk);
 800         if ((task_pgrp(t) != pgrp) &&
 801             (task_session(t) == task_session(tsk)) &&
 802             will_become_orphaned_pgrp(pgrp, tsk) &&
 803             has_stopped_jobs(pgrp)) {
 804                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 805                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 806         }
 807
 808         /* Let father know we died
 809          *
 810          * Thread signals are configurable, but you aren't going to use
 811          * that to send signals to arbitary processes.
 812          * That stops right now.
 813          *
 814          * If the parent exec id doesn't match the exec id we saved
 815          * when we started then we know the parent has changed security
 816          * domain.
 817          *
 818          * If our self_exec id doesn't match our parent_exec_id then
 819          * we have changed execution domain as these two values started
 820          * the same after a fork.
 821          *
 822          */
 823
 824         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 825             ( tsk->parent_exec_id != t->self_exec_id  ||
 826               tsk->self_exec_id != tsk->parent_exec_id)
 827             && !capable(CAP_KILL))
 828                 tsk->exit_signal = SIGCHLD;
 829
 830
 831         /* If something other than our normal parent is ptracing us, then
 832          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 833          * only has special meaning to our real parent.
 834          */
 835         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 836                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 837                 do_notify_parent(tsk, signal);
 838         } else if (tsk->ptrace) {
 839                 do_notify_parent(tsk, SIGCHLD);
 840         }
 841
 842         state = EXIT_ZOMBIE;
 843         if (tsk->exit_signal == -1 &&
 844             (likely(tsk->ptrace == 0) ||
 845              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 846                 state = EXIT_DEAD;
 847         tsk->exit_state = state;
 848
 849         write_unlock_irq(&tasklist_lock);
 850
 851         list_for_each_safe(_p, _n, &ptrace_dead) {
 852                 list_del_init(_p);
 853                 t = list_entry(_p, struct task_struct, ptrace_list);
 854                 release_task(t);
 855         }
 856
 857         /* If the process is dead, release it - nobody will wait for it */
 858         if (state == EXIT_DEAD)
 859                 release_task(tsk);
 860 }
 861
 862 fastcall NORET_TYPE void do_exit(long code)
 863 {
 864         struct task_struct *tsk = current;
 865         int group_dead;
 866
 867         profile_task_exit(tsk);
 868
 869         WARN_ON(atomic_read(&tsk->fs_excl));
 870
 871         if (unlikely(in_interrupt()))
 872                 panic("Aiee, killing interrupt handler!");
 873         if (unlikely(!tsk->pid))
 874                 panic("Attempted to kill the idle task!");
 875         if (unlikely(tsk == child_reaper(tsk))) {
 876                 if (tsk->nsproxy->pid_ns != &init_pid_ns)
 877                         tsk->nsproxy->pid_ns->child_reaper = init_pid_ns.child_reaper;
 878                 else
 879                         panic("Attempted to kill init!");
 880         }
 881
 882
 883         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 884                 current->ptrace_message = code;
 885                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 886         }
 887
 888         /*
 889          * We're taking recursive faults here in do_exit. Safest is to just
 890          * leave this task alone and wait for reboot.
 891          */
 892         if (unlikely(tsk->flags & PF_EXITING)) {
 893                 printk(KERN_ALERT
 894                         "Fixing recursive fault but reboot is needed!\n");
 895                 /*
 896                  * We can do this unlocked here. The futex code uses
 897                  * this flag just to verify whether the pi state
 898                  * cleanup has been done or not. In the worst case it
 899                  * loops once more. We pretend that the cleanup was
 900                  * done as there is no way to return. Either the
 901                  * OWNER_DIED bit is set by now or we push the blocked
 902                  * task into the wait for ever nirwana as well.
 903                  */
 904                 tsk->flags |= PF_EXITPIDONE;
 905                 if (tsk->io_context)
 906                         exit_io_context();
 907                 set_current_state(TASK_UNINTERRUPTIBLE);
 908                 schedule();
 909         }
 910
 911         /*
 912          * tsk->flags are checked in the futex code to protect against
 913          * an exiting task cleaning up the robust pi futexes.
 914          */
 915         spin_lock_irq(&tsk->pi_lock);
 916         tsk->flags |= PF_EXITING;
 917         spin_unlock_irq(&tsk->pi_lock);
 918
 919         if (unlikely(in_atomic()))
 920                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 921                                 current->comm, current->pid,
 922                                 preempt_count());
 923
 924         acct_update_integrals(tsk);
 925         if (tsk->mm) {
 926                 update_hiwater_rss(tsk->mm);
 927                 update_hiwater_vm(tsk->mm);
 928         }
 929         group_dead = atomic_dec_and_test(&tsk->signal->live);
 930         if (group_dead) {
 931                 hrtimer_cancel(&tsk->signal->real_timer);
 932                 exit_itimers(tsk->signal);
 933         }
 934         acct_collect(code, group_dead);
 935         if (unlikely(tsk->robust_list))
 936                 exit_robust_list(tsk);
 937 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 938         if (unlikely(tsk->compat_robust_list))
 939                 compat_exit_robust_list(tsk);
 940 #endif
 941         if (unlikely(tsk->audit_context))
 942                 audit_free(tsk);
 943
 944         taskstats_exit(tsk, group_dead);
 945
 946         exit_mm(tsk);
 947
 948         if (group_dead)
 949                 acct_process();
 950         exit_sem(tsk);
 951         __exit_files(tsk);
 952         __exit_fs(tsk);
 953         exit_thread();
 954         cpuset_exit(tsk);
 955         exit_keys(tsk);
 956
 957         if (group_dead && tsk->signal->leader)
 958                 disassociate_ctty(1);
 959
 960         module_put(task_thread_info(tsk)->exec_domain->module);
 961         if (tsk->binfmt)
 962                 module_put(tsk->binfmt->module);
 963
 964         tsk->exit_code = code;
 965         proc_exit_connector(tsk);
 966         exit_task_namespaces(tsk);
 967         exit_notify(tsk);
 968 #ifdef CONFIG_NUMA
 969         mpol_free(tsk->mempolicy);
 970         tsk->mempolicy = NULL;
 971 #endif
 972         /*
 973          * This must happen late, after the PID is not
 974          * hashed anymore:
 975          */
 976         if (unlikely(!list_empty(&tsk->pi_state_list)))
 977                 exit_pi_state_list(tsk);
 978         if (unlikely(current->pi_state_cache))
 979                 kfree(current->pi_state_cache);
 980         /*
 981          * Make sure we are holding no locks:
 982          */
 983         debug_check_no_locks_held(tsk);
 984         /*
 985          * We can do this unlocked here. The futex code uses this flag
 986          * just to verify whether the pi state cleanup has been done
 987          * or not. In the worst case it loops once more.
 988          */
 989         tsk->flags |= PF_EXITPIDONE;
 990
 991         if (tsk->io_context)
 992                 exit_io_context();
 993
 994         if (tsk->splice_pipe)
 995                 __free_pipe_info(tsk->splice_pipe);
 996
 997         preempt_disable();
 998         /* causes final put_task_struct in finish_task_switch(). */
 999         tsk->state = TASK_DEAD;
1000
1001         schedule();
1002         BUG();
1003         /* Avoid "noreturn function does return".  */
1004         for (;;)
1005                 cpu_relax();    /* For when BUG is null */
1006 }
1007
1008 EXPORT_SYMBOL_GPL(do_exit);
1009
1010 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1011 {
1012         if (comp)
1013                 complete(comp);
1014
1015         do_exit(code);
1016 }
1017
1018 EXPORT_SYMBOL(complete_and_exit);
1019
1020 asmlinkage long sys_exit(int error_code)
1021 {
1022         do_exit((error_code&0xff)<<8);
1023 }
1024
1025 /*
1026  * Take down every thread in the group.  This is called by fatal signals
1027  * as well as by sys_exit_group (below).
1028  */
1029 NORET_TYPE void
1030 do_group_exit(int exit_code)
1031 {
1032         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1033
1034         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1035                 exit_code = current->signal->group_exit_code;
1036         else if (!thread_group_empty(current)) {
1037                 struct signal_struct *const sig = current->signal;
1038                 struct sighand_struct *const sighand = current->sighand;
1039                 spin_lock_irq(&sighand->siglock);
1040                 if (sig->flags & SIGNAL_GROUP_EXIT)
1041                         /* Another thread got here before we took the lock.  */
1042                         exit_code = sig->group_exit_code;
1043                 else {
1044                         sig->group_exit_code = exit_code;
1045                         zap_other_threads(current);
1046                 }
1047                 spin_unlock_irq(&sighand->siglock);
1048         }
1049
1050         do_exit(exit_code);
1051         /* NOTREACHED */
1052 }
1053
1054 /*
1055  * this kills every thread in the thread group. Note that any externally
1056  * wait4()-ing process will get the correct exit code - even if this
1057  * thread is not the thread group leader.
1058  */
1059 asmlinkage void sys_exit_group(int error_code)
1060 {
1061         do_group_exit((error_code & 0xff) << 8);
1062 }
1063
1064 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1065 {
1066         int err;
1067
1068         if (pid > 0) {
1069                 if (p->pid != pid)
1070                         return 0;
1071         } else if (!pid) {
1072                 if (process_group(p) != process_group(current))
1073                         return 0;
1074         } else if (pid != -1) {
1075                 if (process_group(p) != -pid)
1076                         return 0;
1077         }
1078
1079         /*
1080          * Do not consider detached threads that are
1081          * not ptraced:
1082          */
1083         if (p->exit_signal == -1 && !p->ptrace)
1084                 return 0;
1085
1086         /* Wait for all children (clone and not) if __WALL is set;
1087          * otherwise, wait for clone children *only* if __WCLONE is
1088          * set; otherwise, wait for non-clone children *only*.  (Note:
1089          * A "clone" child here is one that reports to its parent
1090          * using a signal other than SIGCHLD.) */
1091         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1092             && !(options & __WALL))
1093                 return 0;
1094         /*
1095          * Do not consider thread group leaders that are
1096          * in a non-empty thread group:
1097          */
1098         if (delay_group_leader(p))
1099                 return 2;
1100
1101         err = security_task_wait(p);
1102         if (err)
1103                 return err;
1104
1105         return 1;
1106 }
1107
1108 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1109                                int why, int status,
1110                                struct siginfo __user *infop,
1111                                struct rusage __user *rusagep)
1112 {
1113         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1114
1115         put_task_struct(p);
1116         if (!retval)
1117                 retval = put_user(SIGCHLD, &infop->si_signo);
1118         if (!retval)
1119                 retval = put_user(0, &infop->si_errno);
1120         if (!retval)
1121                 retval = put_user((short)why, &infop->si_code);
1122         if (!retval)
1123                 retval = put_user(pid, &infop->si_pid);
1124         if (!retval)
1125                 retval = put_user(uid, &infop->si_uid);
1126         if (!retval)
1127                 retval = put_user(status, &infop->si_status);
1128         if (!retval)
1129                 retval = pid;
1130         return retval;
1131 }
1132
1133 /*
1134  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1135  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1136  * the lock and this task is uninteresting.  If we return nonzero, we have
1137  * released the lock and the system call should return.
1138  */
1139 static int wait_task_zombie(struct task_struct *p, int noreap,
1140                             struct siginfo __user *infop,
1141                             int __user *stat_addr, struct rusage __user *ru)
1142 {
1143         unsigned long state;
1144         int retval;
1145         int status;
1146
1147         if (unlikely(noreap)) {
1148                 pid_t pid = p->pid;
1149                 uid_t uid = p->uid;
1150                 int exit_code = p->exit_code;
1151                 int why, status;
1152
1153                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1154                         return 0;
1155                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1156                         return 0;
1157                 get_task_struct(p);
1158                 read_unlock(&tasklist_lock);
1159                 if ((exit_code & 0x7f) == 0) {
1160                         why = CLD_EXITED;
1161                         status = exit_code >> 8;
1162                 } else {
1163                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1164                         status = exit_code & 0x7f;
1165                 }
1166                 return wait_noreap_copyout(p, pid, uid, why,
1167                                            status, infop, ru);
1168         }
1169
1170         /*
1171          * Try to move the task's state to DEAD
1172          * only one thread is allowed to do this:
1173          */
1174         state = xchg(&p->exit_state, EXIT_DEAD);
1175         if (state != EXIT_ZOMBIE) {
1176                 BUG_ON(state != EXIT_DEAD);
1177                 return 0;
1178         }
1179         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1180                 /*
1181                  * This can only happen in a race with a ptraced thread
1182                  * dying on another processor.
1183                  */
1184                 return 0;
1185         }
1186
1187         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1188                 struct signal_struct *psig;
1189                 struct signal_struct *sig;
1190
1191                 /*
1192                  * The resource counters for the group leader are in its
1193                  * own task_struct.  Those for dead threads in the group
1194                  * are in its signal_struct, as are those for the child
1195                  * processes it has previously reaped.  All these
1196                  * accumulate in the parent's signal_struct c* fields.
1197                  *
1198                  * We don't bother to take a lock here to protect these
1199                  * p->signal fields, because they are only touched by
1200                  * __exit_signal, which runs with tasklist_lock
1201                  * write-locked anyway, and so is excluded here.  We do
1202                  * need to protect the access to p->parent->signal fields,
1203                  * as other threads in the parent group can be right
1204                  * here reaping other children at the same time.
1205                  */
1206                 spin_lock_irq(&p->parent->sighand->siglock);
1207                 psig = p->parent->signal;
1208                 sig = p->signal;
1209                 psig->cutime =
1210                         cputime_add(psig->cutime,
1211                         cputime_add(p->utime,
1212                         cputime_add(sig->utime,
1213                                     sig->cutime)));
1214                 psig->cstime =
1215                         cputime_add(psig->cstime,
1216                         cputime_add(p->stime,
1217                         cputime_add(sig->stime,
1218                                     sig->cstime)));
1219                 psig->cmin_flt +=
1220                         p->min_flt + sig->min_flt + sig->cmin_flt;
1221                 psig->cmaj_flt +=
1222                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1223                 psig->cnvcsw +=
1224                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1225                 psig->cnivcsw +=
1226                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1227                 psig->cinblock +=
1228                         task_io_get_inblock(p) +
1229                         sig->inblock + sig->cinblock;
1230                 psig->coublock +=
1231                         task_io_get_oublock(p) +
1232                         sig->oublock + sig->coublock;
1233                 spin_unlock_irq(&p->parent->sighand->siglock);
1234         }
1235
1236         /*
1237          * Now we are sure this task is interesting, and no other
1238          * thread can reap it because we set its state to EXIT_DEAD.
1239          */
1240         read_unlock(&tasklist_lock);
1241
1242         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1243         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1244                 ? p->signal->group_exit_code : p->exit_code;
1245         if (!retval && stat_addr)
1246                 retval = put_user(status, stat_addr);
1247         if (!retval && infop)
1248                 retval = put_user(SIGCHLD, &infop->si_signo);
1249         if (!retval && infop)
1250                 retval = put_user(0, &infop->si_errno);
1251         if (!retval && infop) {
1252                 int why;
1253
1254                 if ((status & 0x7f) == 0) {
1255                         why = CLD_EXITED;
1256                         status >>= 8;
1257                 } else {
1258                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1259                         status &= 0x7f;
1260                 }
1261                 retval = put_user((short)why, &infop->si_code);
1262                 if (!retval)
1263                         retval = put_user(status, &infop->si_status);
1264         }
1265         if (!retval && infop)
1266                 retval = put_user(p->pid, &infop->si_pid);
1267         if (!retval && infop)
1268                 retval = put_user(p->uid, &infop->si_uid);
1269         if (retval) {
1270                 // TODO: is this safe?
1271                 p->exit_state = EXIT_ZOMBIE;
1272                 return retval;
1273         }
1274         retval = p->pid;
1275         if (p->real_parent != p->parent) {
1276                 write_lock_irq(&tasklist_lock);
1277                 /* Double-check with lock held.  */
1278                 if (p->real_parent != p->parent) {
1279                         __ptrace_unlink(p);
1280                         // TODO: is this safe?
1281                         p->exit_state = EXIT_ZOMBIE;
1282                         /*
1283                          * If this is not a detached task, notify the parent.
1284                          * If it's still not detached after that, don't release
1285                          * it now.
1286                          */
1287                         if (p->exit_signal != -1) {
1288                                 do_notify_parent(p, p->exit_signal);
1289                                 if (p->exit_signal != -1)
1290                                         p = NULL;
1291                         }
1292                 }
1293                 write_unlock_irq(&tasklist_lock);
1294         }
1295         if (p != NULL)
1296                 release_task(p);
1297         BUG_ON(!retval);
1298         return retval;
1299 }
1300
1301 /*
1302  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1303  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1304  * the lock and this task is uninteresting.  If we return nonzero, we have
1305  * released the lock and the system call should return.
1306  */
1307 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1308                              int noreap, struct siginfo __user *infop,
1309                              int __user *stat_addr, struct rusage __user *ru)
1310 {
1311         int retval, exit_code;
1312
1313         if (!p->exit_code)
1314                 return 0;
1315         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1316             p->signal && p->signal->group_stop_count > 0)
1317                 /*
1318                  * A group stop is in progress and this is the group leader.
1319                  * We won't report until all threads have stopped.
1320                  */
1321                 return 0;
1322
1323         /*
1324          * Now we are pretty sure this task is interesting.
1325          * Make sure it doesn't get reaped out from under us while we
1326          * give up the lock and then examine it below.  We don't want to
1327          * keep holding onto the tasklist_lock while we call getrusage and
1328          * possibly take page faults for user memory.
1329          */
1330         get_task_struct(p);
1331         read_unlock(&tasklist_lock);
1332
1333         if (unlikely(noreap)) {
1334                 pid_t pid = p->pid;
1335                 uid_t uid = p->uid;
1336                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1337
1338                 exit_code = p->exit_code;
1339                 if (unlikely(!exit_code) ||
1340                     unlikely(p->state & TASK_TRACED))
1341                         goto bail_ref;
1342                 return wait_noreap_copyout(p, pid, uid,
1343                                            why, (exit_code << 8) | 0x7f,
1344                                            infop, ru);
1345         }
1346
1347         write_lock_irq(&tasklist_lock);
1348
1349         /*
1350          * This uses xchg to be atomic with the thread resuming and setting
1351          * it.  It must also be done with the write lock held to prevent a
1352          * race with the EXIT_ZOMBIE case.
1353          */
1354         exit_code = xchg(&p->exit_code, 0);
1355         if (unlikely(p->exit_state)) {
1356                 /*
1357                  * The task resumed and then died.  Let the next iteration
1358                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1359                  * already be zero here if it resumed and did _exit(0).
1360                  * The task itself is dead and won't touch exit_code again;
1361                  * other processors in this function are locked out.
1362                  */
1363                 p->exit_code = exit_code;
1364                 exit_code = 0;
1365         }
1366         if (unlikely(exit_code == 0)) {
1367                 /*
1368                  * Another thread in this function got to it first, or it
1369                  * resumed, or it resumed and then died.
1370                  */
1371                 write_unlock_irq(&tasklist_lock);
1372 bail_ref:
1373                 put_task_struct(p);
1374                 /*
1375                  * We are returning to the wait loop without having successfully
1376                  * removed the process and having released the lock. We cannot
1377                  * continue, since the "p" task pointer is potentially stale.
1378                  *
1379                  * Return -EAGAIN, and do_wait() will restart the loop from the
1380                  * beginning. Do _not_ re-acquire the lock.
1381                  */
1382                 return -EAGAIN;
1383         }
1384
1385         /* move to end of parent's list to avoid starvation */
1386         remove_parent(p);
1387         add_parent(p);
1388
1389         write_unlock_irq(&tasklist_lock);
1390
1391         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1392         if (!retval && stat_addr)
1393                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1394         if (!retval && infop)
1395                 retval = put_user(SIGCHLD, &infop->si_signo);
1396         if (!retval && infop)
1397                 retval = put_user(0, &infop->si_errno);
1398         if (!retval && infop)
1399                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1400                                           ? CLD_TRAPPED : CLD_STOPPED),
1401                                   &infop->si_code);
1402         if (!retval && infop)
1403                 retval = put_user(exit_code, &infop->si_status);
1404         if (!retval && infop)
1405                 retval = put_user(p->pid, &infop->si_pid);
1406         if (!retval && infop)
1407                 retval = put_user(p->uid, &infop->si_uid);
1408         if (!retval)
1409                 retval = p->pid;
1410         put_task_struct(p);
1411
1412         BUG_ON(!retval);
1413         return retval;
1414 }
1415
1416 /*
1417  * Handle do_wait work for one task in a live, non-stopped state.
1418  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1419  * the lock and this task is uninteresting.  If we return nonzero, we have
1420  * released the lock and the system call should return.
1421  */
1422 static int wait_task_continued(struct task_struct *p, int noreap,
1423                                struct siginfo __user *infop,
1424                                int __user *stat_addr, struct rusage __user *ru)
1425 {
1426         int retval;
1427         pid_t pid;
1428         uid_t uid;
1429
1430         if (unlikely(!p->signal))
1431                 return 0;
1432
1433         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1434                 return 0;
1435
1436         spin_lock_irq(&p->sighand->siglock);
1437         /* Re-check with the lock held.  */
1438         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1439                 spin_unlock_irq(&p->sighand->siglock);
1440                 return 0;
1441         }
1442         if (!noreap)
1443                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1444         spin_unlock_irq(&p->sighand->siglock);
1445
1446         pid = p->pid;
1447         uid = p->uid;
1448         get_task_struct(p);
1449         read_unlock(&tasklist_lock);
1450
1451         if (!infop) {
1452                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1453                 put_task_struct(p);
1454                 if (!retval && stat_addr)
1455                         retval = put_user(0xffff, stat_addr);
1456                 if (!retval)
1457                         retval = p->pid;
1458         } else {
1459                 retval = wait_noreap_copyout(p, pid, uid,
1460                                              CLD_CONTINUED, SIGCONT,
1461                                              infop, ru);
1462                 BUG_ON(retval == 0);
1463         }
1464
1465         return retval;
1466 }
1467
1468
1469 static inline int my_ptrace_child(struct task_struct *p)
1470 {
1471         if (!(p->ptrace & PT_PTRACED))
1472                 return 0;
1473         if (!(p->ptrace & PT_ATTACHED))
1474                 return 1;
1475         /*
1476          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1477          * we are the attacher.  If we are the real parent, this is a race
1478          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1479          * which we have to switch the parent links, but has already set
1480          * the flags in p->ptrace.
1481          */
1482         return (p->parent != p->real_parent);
1483 }
1484
1485 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1486                     int __user *stat_addr, struct rusage __user *ru)
1487 {
1488         DECLARE_WAITQUEUE(wait, current);
1489         struct task_struct *tsk;
1490         int flag, retval;
1491         int allowed, denied;
1492
1493         add_wait_queue(&current->signal->wait_chldexit,&wait);
1494 repeat:
1495         /*
1496          * We will set this flag if we see any child that might later
1497          * match our criteria, even if we are not able to reap it yet.
1498          */
1499         flag = 0;
1500         allowed = denied = 0;
1501         current->state = TASK_INTERRUPTIBLE;
1502         read_lock(&tasklist_lock);
1503         tsk = current;
1504         do {
1505                 struct task_struct *p;
1506                 struct list_head *_p;
1507                 int ret;
1508
1509                 list_for_each(_p,&tsk->children) {
1510                         p = list_entry(_p, struct task_struct, sibling);
1511
1512                         ret = eligible_child(pid, options, p);
1513                         if (!ret)
1514                                 continue;
1515
1516                         if (unlikely(ret < 0)) {
1517                                 denied = ret;
1518                                 continue;
1519                         }
1520                         allowed = 1;
1521
1522                         switch (p->state) {
1523                         case TASK_TRACED:
1524                                 /*
1525                                  * When we hit the race with PTRACE_ATTACH,
1526                                  * we will not report this child.  But the
1527                                  * race means it has not yet been moved to
1528                                  * our ptrace_children list, so we need to
1529                                  * set the flag here to avoid a spurious ECHILD
1530                                  * when the race happens with the only child.
1531                                  */
1532                                 flag = 1;
1533                                 if (!my_ptrace_child(p))
1534                                         continue;
1535                                 /*FALLTHROUGH*/
1536                         case TASK_STOPPED:
1537                                 /*
1538                                  * It's stopped now, so it might later
1539                                  * continue, exit, or stop again.
1540                                  */
1541                                 flag = 1;
1542                                 if (!(options & WUNTRACED) &&
1543                                     !my_ptrace_child(p))
1544                                         continue;
1545                                 retval = wait_task_stopped(p, ret == 2,
1546                                                            (options & WNOWAIT),
1547                                                            infop,
1548                                                            stat_addr, ru);
1549                                 if (retval == -EAGAIN)
1550                                         goto repeat;
1551                                 if (retval != 0) /* He released the lock.  */
1552                                         goto end;
1553                                 break;
1554                         default:
1555                         // case EXIT_DEAD:
1556                                 if (p->exit_state == EXIT_DEAD)
1557                                         continue;
1558                         // case EXIT_ZOMBIE:
1559                                 if (p->exit_state == EXIT_ZOMBIE) {
1560                                         /*
1561                                          * Eligible but we cannot release
1562                                          * it yet:
1563                                          */
1564                                         if (ret == 2)
1565                                                 goto check_continued;
1566                                         if (!likely(options & WEXITED))
1567                                                 continue;
1568                                         retval = wait_task_zombie(
1569                                                 p, (options & WNOWAIT),
1570                                                 infop, stat_addr, ru);
1571                                         /* He released the lock.  */
1572                                         if (retval != 0)
1573                                                 goto end;
1574                                         break;
1575                                 }
1576 check_continued:
1577                                 /*
1578                                  * It's running now, so it might later
1579                                  * exit, stop, or stop and then continue.
1580                                  */
1581                                 flag = 1;
1582                                 if (!unlikely(options & WCONTINUED))
1583                                         continue;
1584                                 retval = wait_task_continued(
1585                                         p, (options & WNOWAIT),
1586                                         infop, stat_addr, ru);
1587                                 if (retval != 0) /* He released the lock.  */
1588                                         goto end;
1589                                 break;
1590                         }
1591                 }
1592                 if (!flag) {
1593                         list_for_each(_p, &tsk->ptrace_children) {
1594                                 p = list_entry(_p, struct task_struct,
1595                                                 ptrace_list);
1596                                 if (!eligible_child(pid, options, p))
1597                                         continue;
1598                                 flag = 1;
1599                                 break;
1600                         }
1601                 }
1602                 if (options & __WNOTHREAD)
1603                         break;
1604                 tsk = next_thread(tsk);
1605                 BUG_ON(tsk->signal != current->signal);
1606         } while (tsk != current);
1607
1608         read_unlock(&tasklist_lock);
1609         if (flag) {
1610                 retval = 0;
1611                 if (options & WNOHANG)
1612                         goto end;
1613                 retval = -ERESTARTSYS;
1614                 if (signal_pending(current))
1615                         goto end;
1616                 schedule();
1617                 goto repeat;
1618         }
1619         retval = -ECHILD;
1620         if (unlikely(denied) && !allowed)
1621                 retval = denied;
1622 end:
1623         current->state = TASK_RUNNING;
1624         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1625         if (infop) {
1626                 if (retval > 0)
1627                 retval = 0;
1628                 else {
1629                         /*
1630                          * For a WNOHANG return, clear out all the fields
1631                          * we would set so the user can easily tell the
1632                          * difference.
1633                          */
1634                         if (!retval)
1635                                 retval = put_user(0, &infop->si_signo);
1636                         if (!retval)
1637                                 retval = put_user(0, &infop->si_errno);
1638                         if (!retval)
1639                                 retval = put_user(0, &infop->si_code);
1640                         if (!retval)
1641                                 retval = put_user(0, &infop->si_pid);
1642                         if (!retval)
1643                                 retval = put_user(0, &infop->si_uid);
1644                         if (!retval)
1645                                 retval = put_user(0, &infop->si_status);
1646                 }
1647         }
1648         return retval;
1649 }
1650
1651 asmlinkage long sys_waitid(int which, pid_t pid,
1652                            struct siginfo __user *infop, int options,
1653                            struct rusage __user *ru)
1654 {
1655         long ret;
1656
1657         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1658                 return -EINVAL;
1659         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1660                 return -EINVAL;
1661
1662         switch (which) {
1663         case P_ALL:
1664                 pid = -1;
1665                 break;
1666         case P_PID:
1667                 if (pid <= 0)
1668                         return -EINVAL;
1669                 break;
1670         case P_PGID:
1671                 if (pid <= 0)
1672                         return -EINVAL;
1673                 pid = -pid;
1674                 break;
1675         default:
1676                 return -EINVAL;
1677         }
1678
1679         ret = do_wait(pid, options, infop, NULL, ru);
1680
1681         /* avoid REGPARM breakage on x86: */
1682         prevent_tail_call(ret);
1683         return ret;
1684 }
1685
1686 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1687                           int options, struct rusage __user *ru)
1688 {
1689         long ret;
1690
1691         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1692                         __WNOTHREAD|__WCLONE|__WALL))
1693                 return -EINVAL;
1694         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1695
1696         /* avoid REGPARM breakage on x86: */
1697         prevent_tail_call(ret);
1698         return ret;
1699 }
1700
1701 #ifdef __ARCH_WANT_SYS_WAITPID
1702
1703 /*
1704  * sys_waitpid() remains for compatibility. waitpid() should be
1705  * implemented by calling sys_wait4() from libc.a.
1706  */
1707 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1708 {
1709         return sys_wait4(pid, stat_addr, options, NULL);
1710 }
1711
1712 #endif