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