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/fdtable.h>
  23 #include <linux/binfmts.h>
  24 #include <linux/nsproxy.h>
  25 #include <linux/pid_namespace.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/profile.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/freezer.h>
  35 #include <linux/cgroup.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/signal.h>
  38 #include <linux/posix-timers.h>
  39 #include <linux/cn_proc.h>
  40 #include <linux/mutex.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/pipe_fs_i.h>
  44 #include <linux/audit.h> /* for audit_free() */
  45 #include <linux/resource.h>
  46 #include <linux/blkdev.h>
  47 #include <linux/task_io_accounting_ops.h>
  48
  49 #include <asm/uaccess.h>
  50 #include <asm/unistd.h>
  51 #include <asm/pgtable.h>
  52 #include <asm/mmu_context.h>
  53
  54 static void exit_mm(struct task_struct * tsk);
  55
  56 static inline int task_detached(struct task_struct *p)
  57 {
  58         return p->exit_signal == -1;
  59 }
  60
  61 static void __unhash_process(struct task_struct *p)
  62 {
  63         nr_threads--;
  64         detach_pid(p, PIDTYPE_PID);
  65         if (thread_group_leader(p)) {
  66                 detach_pid(p, PIDTYPE_PGID);
  67                 detach_pid(p, PIDTYPE_SID);
  68
  69                 list_del_rcu(&p->tasks);
  70                 __get_cpu_var(process_counts)--;
  71         }
  72         list_del_rcu(&p->thread_group);
  73         remove_parent(p);
  74 }
  75
  76 /*
  77  * This function expects the tasklist_lock write-locked.
  78  */
  79 static void __exit_signal(struct task_struct *tsk)
  80 {
  81         struct signal_struct *sig = tsk->signal;
  82         struct sighand_struct *sighand;
  83
  84         BUG_ON(!sig);
  85         BUG_ON(!atomic_read(&sig->count));
  86
  87         rcu_read_lock();
  88         sighand = rcu_dereference(tsk->sighand);
  89         spin_lock(&sighand->siglock);
  90
  91         posix_cpu_timers_exit(tsk);
  92         if (atomic_dec_and_test(&sig->count))
  93                 posix_cpu_timers_exit_group(tsk);
  94         else {
  95                 /*
  96                  * If there is any task waiting for the group exit
  97                  * then notify it:
  98                  */
  99                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
 100                         wake_up_process(sig->group_exit_task);
 101
 102                 if (tsk == sig->curr_target)
 103                         sig->curr_target = next_thread(tsk);
 104                 /*
 105                  * Accumulate here the counters for all threads but the
 106                  * group leader as they die, so they can be added into
 107                  * the process-wide totals when those are taken.
 108                  * The group leader stays around as a zombie as long
 109                  * as there are other threads.  When it gets reaped,
 110                  * the exit.c code will add its counts into these totals.
 111                  * We won't ever get here for the group leader, since it
 112                  * will have been the last reference on the signal_struct.
 113                  */
 114                 sig->utime = cputime_add(sig->utime, tsk->utime);
 115                 sig->stime = cputime_add(sig->stime, tsk->stime);
 116                 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
 117                 sig->min_flt += tsk->min_flt;
 118                 sig->maj_flt += tsk->maj_flt;
 119                 sig->nvcsw += tsk->nvcsw;
 120                 sig->nivcsw += tsk->nivcsw;
 121                 sig->inblock += task_io_get_inblock(tsk);
 122                 sig->oublock += task_io_get_oublock(tsk);
 123                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 124                 sig = NULL; /* Marker for below. */
 125         }
 126
 127         __unhash_process(tsk);
 128
 129         tsk->signal = NULL;
 130         tsk->sighand = NULL;
 131         spin_unlock(&sighand->siglock);
 132         rcu_read_unlock();
 133
 134         __cleanup_sighand(sighand);
 135         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 136         flush_sigqueue(&tsk->pending);
 137         if (sig) {
 138                 flush_sigqueue(&sig->shared_pending);
 139                 taskstats_tgid_free(sig);
 140                 __cleanup_signal(sig);
 141         }
 142 }
 143
 144 static void delayed_put_task_struct(struct rcu_head *rhp)
 145 {
 146         put_task_struct(container_of(rhp, struct task_struct, rcu));
 147 }
 148
 149 void release_task(struct task_struct * p)
 150 {
 151         struct task_struct *leader;
 152         int zap_leader;
 153 repeat:
 154         atomic_dec(&p->user->processes);
 155         proc_flush_task(p);
 156         write_lock_irq(&tasklist_lock);
 157         ptrace_unlink(p);
 158         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 159         __exit_signal(p);
 160
 161         /*
 162          * If we are the last non-leader member of the thread
 163          * group, and the leader is zombie, then notify the
 164          * group leader's parent process. (if it wants notification.)
 165          */
 166         zap_leader = 0;
 167         leader = p->group_leader;
 168         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 169                 BUG_ON(task_detached(leader));
 170                 do_notify_parent(leader, leader->exit_signal);
 171                 /*
 172                  * If we were the last child thread and the leader has
 173                  * exited already, and the leader's parent ignores SIGCHLD,
 174                  * then we are the one who should release the leader.
 175                  *
 176                  * do_notify_parent() will have marked it self-reaping in
 177                  * that case.
 178                  */
 179                 zap_leader = task_detached(leader);
 180         }
 181
 182         write_unlock_irq(&tasklist_lock);
 183         release_thread(p);
 184         call_rcu(&p->rcu, delayed_put_task_struct);
 185
 186         p = leader;
 187         if (unlikely(zap_leader))
 188                 goto repeat;
 189 }
 190
 191 /*
 192  * This checks not only the pgrp, but falls back on the pid if no
 193  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 194  * without this...
 195  *
 196  * The caller must hold rcu lock or the tasklist lock.
 197  */
 198 struct pid *session_of_pgrp(struct pid *pgrp)
 199 {
 200         struct task_struct *p;
 201         struct pid *sid = NULL;
 202
 203         p = pid_task(pgrp, PIDTYPE_PGID);
 204         if (p == NULL)
 205                 p = pid_task(pgrp, PIDTYPE_PID);
 206         if (p != NULL)
 207                 sid = task_session(p);
 208
 209         return sid;
 210 }
 211
 212 /*
 213  * Determine if a process group is "orphaned", according to the POSIX
 214  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 215  * by terminal-generated stop signals.  Newly orphaned process groups are
 216  * to receive a SIGHUP and a SIGCONT.
 217  *
 218  * "I ask you, have you ever known what it is to be an orphan?"
 219  */
 220 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 221 {
 222         struct task_struct *p;
 223
 224         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 225                 if ((p == ignored_task) ||
 226                     (p->exit_state && thread_group_empty(p)) ||
 227                     is_global_init(p->real_parent))
 228                         continue;
 229
 230                 if (task_pgrp(p->real_parent) != pgrp &&
 231                     task_session(p->real_parent) == task_session(p))
 232                         return 0;
 233         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 234
 235         return 1;
 236 }
 237
 238 int is_current_pgrp_orphaned(void)
 239 {
 240         int retval;
 241
 242         read_lock(&tasklist_lock);
 243         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 244         read_unlock(&tasklist_lock);
 245
 246         return retval;
 247 }
 248
 249 static int has_stopped_jobs(struct pid *pgrp)
 250 {
 251         int retval = 0;
 252         struct task_struct *p;
 253
 254         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 255                 if (!task_is_stopped(p))
 256                         continue;
 257                 retval = 1;
 258                 break;
 259         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 260         return retval;
 261 }
 262
 263 /*
 264  * Check to see if any process groups have become orphaned as
 265  * a result of our exiting, and if they have any stopped jobs,
 266  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
 267  */
 268 static void
 269 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
 270 {
 271         struct pid *pgrp = task_pgrp(tsk);
 272         struct task_struct *ignored_task = tsk;
 273
 274         if (!parent)
 275                  /* exit: our father is in a different pgrp than
 276                   * we are and we were the only connection outside.
 277                   */
 278                 parent = tsk->real_parent;
 279         else
 280                 /* reparent: our child is in a different pgrp than
 281                  * we are, and it was the only connection outside.
 282                  */
 283                 ignored_task = NULL;
 284
 285         if (task_pgrp(parent) != pgrp &&
 286             task_session(parent) == task_session(tsk) &&
 287             will_become_orphaned_pgrp(pgrp, ignored_task) &&
 288             has_stopped_jobs(pgrp)) {
 289                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 290                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 291         }
 292 }
 293
 294 /**
 295  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
 296  *
 297  * If a kernel thread is launched as a result of a system call, or if
 298  * it ever exits, it should generally reparent itself to kthreadd so it
 299  * isn't in the way of other processes and is correctly cleaned up on exit.
 300  *
 301  * The various task state such as scheduling policy and priority may have
 302  * been inherited from a user process, so we reset them to sane values here.
 303  *
 304  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
 305  */
 306 static void reparent_to_kthreadd(void)
 307 {
 308         write_lock_irq(&tasklist_lock);
 309
 310         ptrace_unlink(current);
 311         /* Reparent to init */
 312         remove_parent(current);
 313         current->real_parent = current->parent = kthreadd_task;
 314         add_parent(current);
 315
 316         /* Set the exit signal to SIGCHLD so we signal init on exit */
 317         current->exit_signal = SIGCHLD;
 318
 319         if (task_nice(current) < 0)
 320                 set_user_nice(current, 0);
 321         /* cpus_allowed? */
 322         /* rt_priority? */
 323         /* signals? */
 324         security_task_reparent_to_init(current);
 325         memcpy(current->signal->rlim, init_task.signal->rlim,
 326                sizeof(current->signal->rlim));
 327         atomic_inc(&(INIT_USER->__count));
 328         write_unlock_irq(&tasklist_lock);
 329         switch_uid(INIT_USER);
 330 }
 331
 332 void __set_special_pids(struct pid *pid)
 333 {
 334         struct task_struct *curr = current->group_leader;
 335         pid_t nr = pid_nr(pid);
 336
 337         if (task_session(curr) != pid) {
 338                 change_pid(curr, PIDTYPE_SID, pid);
 339                 set_task_session(curr, nr);
 340         }
 341         if (task_pgrp(curr) != pid) {
 342                 change_pid(curr, PIDTYPE_PGID, pid);
 343                 set_task_pgrp(curr, nr);
 344         }
 345 }
 346
 347 static void set_special_pids(struct pid *pid)
 348 {
 349         write_lock_irq(&tasklist_lock);
 350         __set_special_pids(pid);
 351         write_unlock_irq(&tasklist_lock);
 352 }
 353
 354 /*
 355  * Let kernel threads use this to say that they
 356  * allow a certain signal (since daemonize() will
 357  * have disabled all of them by default).
 358  */
 359 int allow_signal(int sig)
 360 {
 361         if (!valid_signal(sig) || sig < 1)
 362                 return -EINVAL;
 363
 364         spin_lock_irq(&current->sighand->siglock);
 365         sigdelset(&current->blocked, sig);
 366         if (!current->mm) {
 367                 /* Kernel threads handle their own signals.
 368                    Let the signal code know it'll be handled, so
 369                    that they don't get converted to SIGKILL or
 370                    just silently dropped */
 371                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 372         }
 373         recalc_sigpending();
 374         spin_unlock_irq(&current->sighand->siglock);
 375         return 0;
 376 }
 377
 378 EXPORT_SYMBOL(allow_signal);
 379
 380 int disallow_signal(int sig)
 381 {
 382         if (!valid_signal(sig) || sig < 1)
 383                 return -EINVAL;
 384
 385         spin_lock_irq(&current->sighand->siglock);
 386         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
 387         recalc_sigpending();
 388         spin_unlock_irq(&current->sighand->siglock);
 389         return 0;
 390 }
 391
 392 EXPORT_SYMBOL(disallow_signal);
 393
 394 /*
 395  *      Put all the gunge required to become a kernel thread without
 396  *      attached user resources in one place where it belongs.
 397  */
 398
 399 void daemonize(const char *name, ...)
 400 {
 401         va_list args;
 402         struct fs_struct *fs;
 403         sigset_t blocked;
 404
 405         va_start(args, name);
 406         vsnprintf(current->comm, sizeof(current->comm), name, args);
 407         va_end(args);
 408
 409         /*
 410          * If we were started as result of loading a module, close all of the
 411          * user space pages.  We don't need them, and if we didn't close them
 412          * they would be locked into memory.
 413          */
 414         exit_mm(current);
 415         /*
 416          * We don't want to have TIF_FREEZE set if the system-wide hibernation
 417          * or suspend transition begins right now.
 418          */
 419         current->flags |= PF_NOFREEZE;
 420
 421         if (current->nsproxy != &init_nsproxy) {
 422                 get_nsproxy(&init_nsproxy);
 423                 switch_task_namespaces(current, &init_nsproxy);
 424         }
 425         set_special_pids(&init_struct_pid);
 426         proc_clear_tty(current);
 427
 428         /* Block and flush all signals */
 429         sigfillset(&blocked);
 430         sigprocmask(SIG_BLOCK, &blocked, NULL);
 431         flush_signals(current);
 432
 433         /* Become as one with the init task */
 434
 435         exit_fs(current);       /* current->fs->count--; */
 436         fs = init_task.fs;
 437         current->fs = fs;
 438         atomic_inc(&fs->count);
 439
 440         exit_files(current);
 441         current->files = init_task.files;
 442         atomic_inc(&current->files->count);
 443
 444         reparent_to_kthreadd();
 445 }
 446
 447 EXPORT_SYMBOL(daemonize);
 448
 449 static void close_files(struct files_struct * files)
 450 {
 451         int i, j;
 452         struct fdtable *fdt;
 453
 454         j = 0;
 455
 456         /*
 457          * It is safe to dereference the fd table without RCU or
 458          * ->file_lock because this is the last reference to the
 459          * files structure.
 460          */
 461         fdt = files_fdtable(files);
 462         for (;;) {
 463                 unsigned long set;
 464                 i = j * __NFDBITS;
 465                 if (i >= fdt->max_fds)
 466                         break;
 467                 set = fdt->open_fds->fds_bits[j++];
 468                 while (set) {
 469                         if (set & 1) {
 470                                 struct file * file = xchg(&fdt->fd[i], NULL);
 471                                 if (file) {
 472                                         filp_close(file, files);
 473                                         cond_resched();
 474                                 }
 475                         }
 476                         i++;
 477                         set >>= 1;
 478                 }
 479         }
 480 }
 481
 482 struct files_struct *get_files_struct(struct task_struct *task)
 483 {
 484         struct files_struct *files;
 485
 486         task_lock(task);
 487         files = task->files;
 488         if (files)
 489                 atomic_inc(&files->count);
 490         task_unlock(task);
 491
 492         return files;
 493 }
 494
 495 void put_files_struct(struct files_struct *files)
 496 {
 497         struct fdtable *fdt;
 498
 499         if (atomic_dec_and_test(&files->count)) {
 500                 close_files(files);
 501                 /*
 502                  * Free the fd and fdset arrays if we expanded them.
 503                  * If the fdtable was embedded, pass files for freeing
 504                  * at the end of the RCU grace period. Otherwise,
 505                  * you can free files immediately.
 506                  */
 507                 fdt = files_fdtable(files);
 508                 if (fdt != &files->fdtab)
 509                         kmem_cache_free(files_cachep, files);
 510                 free_fdtable(fdt);
 511         }
 512 }
 513
 514 void reset_files_struct(struct files_struct *files)
 515 {
 516         struct task_struct *tsk = current;
 517         struct files_struct *old;
 518
 519         old = tsk->files;
 520         task_lock(tsk);
 521         tsk->files = files;
 522         task_unlock(tsk);
 523         put_files_struct(old);
 524 }
 525
 526 void exit_files(struct task_struct *tsk)
 527 {
 528         struct files_struct * files = tsk->files;
 529
 530         if (files) {
 531                 task_lock(tsk);
 532                 tsk->files = NULL;
 533                 task_unlock(tsk);
 534                 put_files_struct(files);
 535         }
 536 }
 537
 538 void put_fs_struct(struct fs_struct *fs)
 539 {
 540         /* No need to hold fs->lock if we are killing it */
 541         if (atomic_dec_and_test(&fs->count)) {
 542                 path_put(&fs->root);
 543                 path_put(&fs->pwd);
 544                 if (fs->altroot.dentry)
 545                         path_put(&fs->altroot);
 546                 kmem_cache_free(fs_cachep, fs);
 547         }
 548 }
 549
 550 void exit_fs(struct task_struct *tsk)
 551 {
 552         struct fs_struct * fs = tsk->fs;
 553
 554         if (fs) {
 555                 task_lock(tsk);
 556                 tsk->fs = NULL;
 557                 task_unlock(tsk);
 558                 put_fs_struct(fs);
 559         }
 560 }
 561
 562 EXPORT_SYMBOL_GPL(exit_fs);
 563
 564 #ifdef CONFIG_MM_OWNER
 565 /*
 566  * Task p is exiting and it owned mm, lets find a new owner for it
 567  */
 568 static inline int
 569 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
 570 {
 571         /*
 572          * If there are other users of the mm and the owner (us) is exiting
 573          * we need to find a new owner to take on the responsibility.
 574          */
 575         if (!mm)
 576                 return 0;
 577         if (atomic_read(&mm->mm_users) <= 1)
 578                 return 0;
 579         if (mm->owner != p)
 580                 return 0;
 581         return 1;
 582 }
 583
 584 void mm_update_next_owner(struct mm_struct *mm)
 585 {
 586         struct task_struct *c, *g, *p = current;
 587
 588 retry:
 589         if (!mm_need_new_owner(mm, p))
 590                 return;
 591
 592         read_lock(&tasklist_lock);
 593         /*
 594          * Search in the children
 595          */
 596         list_for_each_entry(c, &p->children, sibling) {
 597                 if (c->mm == mm)
 598                         goto assign_new_owner;
 599         }
 600
 601         /*
 602          * Search in the siblings
 603          */
 604         list_for_each_entry(c, &p->parent->children, sibling) {
 605                 if (c->mm == mm)
 606                         goto assign_new_owner;
 607         }
 608
 609         /*
 610          * Search through everything else. We should not get
 611          * here often
 612          */
 613         do_each_thread(g, c) {
 614                 if (c->mm == mm)
 615                         goto assign_new_owner;
 616         } while_each_thread(g, c);
 617
 618         read_unlock(&tasklist_lock);
 619         return;
 620
 621 assign_new_owner:
 622         BUG_ON(c == p);
 623         get_task_struct(c);
 624         /*
 625          * The task_lock protects c->mm from changing.
 626          * We always want mm->owner->mm == mm
 627          */
 628         task_lock(c);
 629         /*
 630          * Delay read_unlock() till we have the task_lock()
 631          * to ensure that c does not slip away underneath us
 632          */
 633         read_unlock(&tasklist_lock);
 634         if (c->mm != mm) {
 635                 task_unlock(c);
 636                 put_task_struct(c);
 637                 goto retry;
 638         }
 639         cgroup_mm_owner_callbacks(mm->owner, c);
 640         mm->owner = c;
 641         task_unlock(c);
 642         put_task_struct(c);
 643 }
 644 #endif /* CONFIG_MM_OWNER */
 645
 646 /*
 647  * Turn us into a lazy TLB process if we
 648  * aren't already..
 649  */
 650 static void exit_mm(struct task_struct * tsk)
 651 {
 652         struct mm_struct *mm = tsk->mm;
 653
 654         mm_release(tsk, mm);
 655         if (!mm)
 656                 return;
 657         /*
 658          * Serialize with any possible pending coredump.
 659          * We must hold mmap_sem around checking core_waiters
 660          * and clearing tsk->mm.  The core-inducing thread
 661          * will increment core_waiters for each thread in the
 662          * group with ->mm != NULL.
 663          */
 664         down_read(&mm->mmap_sem);
 665         if (mm->core_waiters) {
 666                 up_read(&mm->mmap_sem);
 667                 down_write(&mm->mmap_sem);
 668                 if (!--mm->core_waiters)
 669                         complete(mm->core_startup_done);
 670                 up_write(&mm->mmap_sem);
 671
 672                 wait_for_completion(&mm->core_done);
 673                 down_read(&mm->mmap_sem);
 674         }
 675         atomic_inc(&mm->mm_count);
 676         BUG_ON(mm != tsk->active_mm);
 677         /* more a memory barrier than a real lock */
 678         task_lock(tsk);
 679         tsk->mm = NULL;
 680         up_read(&mm->mmap_sem);
 681         enter_lazy_tlb(mm, current);
 682         /* We don't want this task to be frozen prematurely */
 683         clear_freeze_flag(tsk);
 684         task_unlock(tsk);
 685         mm_update_next_owner(mm);
 686         mmput(mm);
 687 }
 688
 689 static void
 690 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 691 {
 692         if (p->pdeath_signal)
 693                 /* We already hold the tasklist_lock here.  */
 694                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 695
 696         /* Move the child from its dying parent to the new one.  */
 697         if (unlikely(traced)) {
 698                 /* Preserve ptrace links if someone else is tracing this child.  */
 699                 list_del_init(&p->ptrace_list);
 700                 if (ptrace_reparented(p))
 701                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 702         } else {
 703                 /* If this child is being traced, then we're the one tracing it
 704                  * anyway, so let go of it.
 705                  */
 706                 p->ptrace = 0;
 707                 remove_parent(p);
 708                 p->parent = p->real_parent;
 709                 add_parent(p);
 710
 711                 if (task_is_traced(p)) {
 712                         /*
 713                          * If it was at a trace stop, turn it into
 714                          * a normal stop since it's no longer being
 715                          * traced.
 716                          */
 717                         ptrace_untrace(p);
 718                 }
 719         }
 720
 721         /* If this is a threaded reparent there is no need to
 722          * notify anyone anything has happened.
 723          */
 724         if (same_thread_group(p->real_parent, father))
 725                 return;
 726
 727         /* We don't want people slaying init.  */
 728         if (!task_detached(p))
 729                 p->exit_signal = SIGCHLD;
 730
 731         /* If we'd notified the old parent about this child's death,
 732          * also notify the new parent.
 733          */
 734         if (!traced && p->exit_state == EXIT_ZOMBIE &&
 735             !task_detached(p) && thread_group_empty(p))
 736                 do_notify_parent(p, p->exit_signal);
 737
 738         kill_orphaned_pgrp(p, father);
 739 }
 740
 741 /*
 742  * When we die, we re-parent all our children.
 743  * Try to give them to another thread in our thread
 744  * group, and if no such member exists, give it to
 745  * the child reaper process (ie "init") in our pid
 746  * space.
 747  */
 748 static void forget_original_parent(struct task_struct *father)
 749 {
 750         struct task_struct *p, *n, *reaper = father;
 751         struct list_head ptrace_dead;
 752
 753         INIT_LIST_HEAD(&ptrace_dead);
 754
 755         write_lock_irq(&tasklist_lock);
 756
 757         do {
 758                 reaper = next_thread(reaper);
 759                 if (reaper == father) {
 760                         reaper = task_child_reaper(father);
 761                         break;
 762                 }
 763         } while (reaper->flags & PF_EXITING);
 764
 765         /*
 766          * There are only two places where our children can be:
 767          *
 768          * - in our child list
 769          * - in our ptraced child list
 770          *
 771          * Search them and reparent children.
 772          */
 773         list_for_each_entry_safe(p, n, &father->children, sibling) {
 774                 int ptrace;
 775
 776                 ptrace = p->ptrace;
 777
 778                 /* if father isn't the real parent, then ptrace must be enabled */
 779                 BUG_ON(father != p->real_parent && !ptrace);
 780
 781                 if (father == p->real_parent) {
 782                         /* reparent with a reaper, real father it's us */
 783                         p->real_parent = reaper;
 784                         reparent_thread(p, father, 0);
 785                 } else {
 786                         /* reparent ptraced task to its real parent */
 787                         __ptrace_unlink (p);
 788                         if (p->exit_state == EXIT_ZOMBIE && !task_detached(p) &&
 789                             thread_group_empty(p))
 790                                 do_notify_parent(p, p->exit_signal);
 791                 }
 792
 793                 /*
 794                  * if the ptraced child is a detached zombie we must collect
 795                  * it before we exit, or it will remain zombie forever since
 796                  * we prevented it from self-reap itself while it was being
 797                  * traced by us, to be able to see it in wait4.
 798                  */
 799                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && task_detached(p)))
 800                         list_add(&p->ptrace_list, &ptrace_dead);
 801         }
 802
 803         list_for_each_entry_safe(p, n, &father->ptrace_children, ptrace_list) {
 804                 p->real_parent = reaper;
 805                 reparent_thread(p, father, 1);
 806         }
 807
 808         write_unlock_irq(&tasklist_lock);
 809         BUG_ON(!list_empty(&father->children));
 810         BUG_ON(!list_empty(&father->ptrace_children));
 811
 812         list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_list) {
 813                 list_del_init(&p->ptrace_list);
 814                 release_task(p);
 815         }
 816
 817 }
 818
 819 /*
 820  * Send signals to all our closest relatives so that they know
 821  * to properly mourn us..
 822  */
 823 static void exit_notify(struct task_struct *tsk, int group_dead)
 824 {
 825         int state;
 826
 827         /*
 828          * This does two things:
 829          *
 830          * A.  Make init inherit all the child processes
 831          * B.  Check to see if any process groups have become orphaned
 832          *      as a result of our exiting, and if they have any stopped
 833          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 834          */
 835         forget_original_parent(tsk);
 836         exit_task_namespaces(tsk);
 837
 838         write_lock_irq(&tasklist_lock);
 839         if (group_dead)
 840                 kill_orphaned_pgrp(tsk->group_leader, NULL);
 841
 842         /* Let father know we died
 843          *
 844          * Thread signals are configurable, but you aren't going to use
 845          * that to send signals to arbitary processes.
 846          * That stops right now.
 847          *
 848          * If the parent exec id doesn't match the exec id we saved
 849          * when we started then we know the parent has changed security
 850          * domain.
 851          *
 852          * If our self_exec id doesn't match our parent_exec_id then
 853          * we have changed execution domain as these two values started
 854          * the same after a fork.
 855          */
 856         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
 857             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
 858              tsk->self_exec_id != tsk->parent_exec_id) &&
 859             !capable(CAP_KILL))
 860                 tsk->exit_signal = SIGCHLD;
 861
 862         /* If something other than our normal parent is ptracing us, then
 863          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 864          * only has special meaning to our real parent.
 865          */
 866         if (!task_detached(tsk) && thread_group_empty(tsk)) {
 867                 int signal = ptrace_reparented(tsk) ?
 868                                 SIGCHLD : tsk->exit_signal;
 869                 do_notify_parent(tsk, signal);
 870         } else if (tsk->ptrace) {
 871                 do_notify_parent(tsk, SIGCHLD);
 872         }
 873
 874         state = EXIT_ZOMBIE;
 875         if (task_detached(tsk) && likely(!tsk->ptrace))
 876                 state = EXIT_DEAD;
 877         tsk->exit_state = state;
 878
 879         /* mt-exec, de_thread() is waiting for us */
 880         if (thread_group_leader(tsk) &&
 881             tsk->signal->notify_count < 0 &&
 882             tsk->signal->group_exit_task)
 883                 wake_up_process(tsk->signal->group_exit_task);
 884
 885         write_unlock_irq(&tasklist_lock);
 886
 887         /* If the process is dead, release it - nobody will wait for it */
 888         if (state == EXIT_DEAD)
 889                 release_task(tsk);
 890 }
 891
 892 #ifdef CONFIG_DEBUG_STACK_USAGE
 893 static void check_stack_usage(void)
 894 {
 895         static DEFINE_SPINLOCK(low_water_lock);
 896         static int lowest_to_date = THREAD_SIZE;
 897         unsigned long *n = end_of_stack(current);
 898         unsigned long free;
 899
 900         while (*n == 0)
 901                 n++;
 902         free = (unsigned long)n - (unsigned long)end_of_stack(current);
 903
 904         if (free >= lowest_to_date)
 905                 return;
 906
 907         spin_lock(&low_water_lock);
 908         if (free < lowest_to_date) {
 909                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
 910                                 "left\n",
 911                                 current->comm, free);
 912                 lowest_to_date = free;
 913         }
 914         spin_unlock(&low_water_lock);
 915 }
 916 #else
 917 static inline void check_stack_usage(void) {}
 918 #endif
 919
 920 static inline void exit_child_reaper(struct task_struct *tsk)
 921 {
 922         if (likely(tsk->group_leader != task_child_reaper(tsk)))
 923                 return;
 924
 925         if (tsk->nsproxy->pid_ns == &init_pid_ns)
 926                 panic("Attempted to kill init!");
 927
 928         /*
 929          * @tsk is the last thread in the 'cgroup-init' and is exiting.
 930          * Terminate all remaining processes in the namespace and reap them
 931          * before exiting @tsk.
 932          *
 933          * Note that @tsk (last thread of cgroup-init) may not necessarily
 934          * be the child-reaper (i.e main thread of cgroup-init) of the
 935          * namespace i.e the child_reaper may have already exited.
 936          *
 937          * Even after a child_reaper exits, we let it inherit orphaned children,
 938          * because, pid_ns->child_reaper remains valid as long as there is
 939          * at least one living sub-thread in the cgroup init.
 940
 941          * This living sub-thread of the cgroup-init will be notified when
 942          * a child inherited by the 'child-reaper' exits (do_notify_parent()
 943          * uses __group_send_sig_info()). Further, when reaping child processes,
 944          * do_wait() iterates over children of all living sub threads.
 945
 946          * i.e even though 'child_reaper' thread is listed as the parent of the
 947          * orphaned children, any living sub-thread in the cgroup-init can
 948          * perform the role of the child_reaper.
 949          */
 950         zap_pid_ns_processes(tsk->nsproxy->pid_ns);
 951 }
 952
 953 NORET_TYPE void do_exit(long code)
 954 {
 955         struct task_struct *tsk = current;
 956         int group_dead;
 957
 958         profile_task_exit(tsk);
 959
 960         WARN_ON(atomic_read(&tsk->fs_excl));
 961
 962         if (unlikely(in_interrupt()))
 963                 panic("Aiee, killing interrupt handler!");
 964         if (unlikely(!tsk->pid))
 965                 panic("Attempted to kill the idle task!");
 966
 967         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 968                 current->ptrace_message = code;
 969                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 970         }
 971
 972         /*
 973          * We're taking recursive faults here in do_exit. Safest is to just
 974          * leave this task alone and wait for reboot.
 975          */
 976         if (unlikely(tsk->flags & PF_EXITING)) {
 977                 printk(KERN_ALERT
 978                         "Fixing recursive fault but reboot is needed!\n");
 979                 /*
 980                  * We can do this unlocked here. The futex code uses
 981                  * this flag just to verify whether the pi state
 982                  * cleanup has been done or not. In the worst case it
 983                  * loops once more. We pretend that the cleanup was
 984                  * done as there is no way to return. Either the
 985                  * OWNER_DIED bit is set by now or we push the blocked
 986                  * task into the wait for ever nirwana as well.
 987                  */
 988                 tsk->flags |= PF_EXITPIDONE;
 989                 if (tsk->io_context)
 990                         exit_io_context();
 991                 set_current_state(TASK_UNINTERRUPTIBLE);
 992                 schedule();
 993         }
 994
 995         exit_signals(tsk);  /* sets PF_EXITING */
 996         /*
 997          * tsk->flags are checked in the futex code to protect against
 998          * an exiting task cleaning up the robust pi futexes.
 999          */
1000         smp_mb();
1001         spin_unlock_wait(&tsk->pi_lock);
1002
1003         if (unlikely(in_atomic()))
1004                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1005                                 current->comm, task_pid_nr(current),
1006                                 preempt_count());
1007
1008         acct_update_integrals(tsk);
1009         if (tsk->mm) {
1010                 update_hiwater_rss(tsk->mm);
1011                 update_hiwater_vm(tsk->mm);
1012         }
1013         group_dead = atomic_dec_and_test(&tsk->signal->live);
1014         if (group_dead) {
1015                 exit_child_reaper(tsk);
1016                 hrtimer_cancel(&tsk->signal->real_timer);
1017                 exit_itimers(tsk->signal);
1018         }
1019         acct_collect(code, group_dead);
1020 #ifdef CONFIG_FUTEX
1021         if (unlikely(tsk->robust_list))
1022                 exit_robust_list(tsk);
1023 #ifdef CONFIG_COMPAT
1024         if (unlikely(tsk->compat_robust_list))
1025                 compat_exit_robust_list(tsk);
1026 #endif
1027 #endif
1028         if (group_dead)
1029                 tty_audit_exit();
1030         if (unlikely(tsk->audit_context))
1031                 audit_free(tsk);
1032
1033         tsk->exit_code = code;
1034         taskstats_exit(tsk, group_dead);
1035
1036         exit_mm(tsk);
1037
1038         if (group_dead)
1039                 acct_process();
1040         exit_sem(tsk);
1041         exit_files(tsk);
1042         exit_fs(tsk);
1043         check_stack_usage();
1044         exit_thread();
1045         cgroup_exit(tsk, 1);
1046         exit_keys(tsk);
1047
1048         if (group_dead && tsk->signal->leader)
1049                 disassociate_ctty(1);
1050
1051         module_put(task_thread_info(tsk)->exec_domain->module);
1052         if (tsk->binfmt)
1053                 module_put(tsk->binfmt->module);
1054
1055         proc_exit_connector(tsk);
1056         exit_notify(tsk, group_dead);
1057 #ifdef CONFIG_NUMA
1058         mpol_put(tsk->mempolicy);
1059         tsk->mempolicy = NULL;
1060 #endif
1061 #ifdef CONFIG_FUTEX
1062         /*
1063          * This must happen late, after the PID is not
1064          * hashed anymore:
1065          */
1066         if (unlikely(!list_empty(&tsk->pi_state_list)))
1067                 exit_pi_state_list(tsk);
1068         if (unlikely(current->pi_state_cache))
1069                 kfree(current->pi_state_cache);
1070 #endif
1071         /*
1072          * Make sure we are holding no locks:
1073          */
1074         debug_check_no_locks_held(tsk);
1075         /*
1076          * We can do this unlocked here. The futex code uses this flag
1077          * just to verify whether the pi state cleanup has been done
1078          * or not. In the worst case it loops once more.
1079          */
1080         tsk->flags |= PF_EXITPIDONE;
1081
1082         if (tsk->io_context)
1083                 exit_io_context();
1084
1085         if (tsk->splice_pipe)
1086                 __free_pipe_info(tsk->splice_pipe);
1087
1088         preempt_disable();
1089         /* causes final put_task_struct in finish_task_switch(). */
1090         tsk->state = TASK_DEAD;
1091
1092         schedule();
1093         BUG();
1094         /* Avoid "noreturn function does return".  */
1095         for (;;)
1096                 cpu_relax();    /* For when BUG is null */
1097 }
1098
1099 EXPORT_SYMBOL_GPL(do_exit);
1100
1101 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1102 {
1103         if (comp)
1104                 complete(comp);
1105
1106         do_exit(code);
1107 }
1108
1109 EXPORT_SYMBOL(complete_and_exit);
1110
1111 asmlinkage long sys_exit(int error_code)
1112 {
1113         do_exit((error_code&0xff)<<8);
1114 }
1115
1116 /*
1117  * Take down every thread in the group.  This is called by fatal signals
1118  * as well as by sys_exit_group (below).
1119  */
1120 NORET_TYPE void
1121 do_group_exit(int exit_code)
1122 {
1123         struct signal_struct *sig = current->signal;
1124
1125         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1126
1127         if (signal_group_exit(sig))
1128                 exit_code = sig->group_exit_code;
1129         else if (!thread_group_empty(current)) {
1130                 struct sighand_struct *const sighand = current->sighand;
1131                 spin_lock_irq(&sighand->siglock);
1132                 if (signal_group_exit(sig))
1133                         /* Another thread got here before we took the lock.  */
1134                         exit_code = sig->group_exit_code;
1135                 else {
1136                         sig->group_exit_code = exit_code;
1137                         sig->flags = SIGNAL_GROUP_EXIT;
1138                         zap_other_threads(current);
1139                 }
1140                 spin_unlock_irq(&sighand->siglock);
1141         }
1142
1143         do_exit(exit_code);
1144         /* NOTREACHED */
1145 }
1146
1147 /*
1148  * this kills every thread in the thread group. Note that any externally
1149  * wait4()-ing process will get the correct exit code - even if this
1150  * thread is not the thread group leader.
1151  */
1152 asmlinkage void sys_exit_group(int error_code)
1153 {
1154         do_group_exit((error_code & 0xff) << 8);
1155 }
1156
1157 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1158 {
1159         struct pid *pid = NULL;
1160         if (type == PIDTYPE_PID)
1161                 pid = task->pids[type].pid;
1162         else if (type < PIDTYPE_MAX)
1163                 pid = task->group_leader->pids[type].pid;
1164         return pid;
1165 }
1166
1167 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1168                           struct task_struct *p)
1169 {
1170         int err;
1171
1172         if (type < PIDTYPE_MAX) {
1173                 if (task_pid_type(p, type) != pid)
1174                         return 0;
1175         }
1176
1177         /*
1178          * Do not consider detached threads that are
1179          * not ptraced:
1180          */
1181         if (task_detached(p) && !p->ptrace)
1182                 return 0;
1183
1184         /* Wait for all children (clone and not) if __WALL is set;
1185          * otherwise, wait for clone children *only* if __WCLONE is
1186          * set; otherwise, wait for non-clone children *only*.  (Note:
1187          * A "clone" child here is one that reports to its parent
1188          * using a signal other than SIGCHLD.) */
1189         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1190             && !(options & __WALL))
1191                 return 0;
1192
1193         err = security_task_wait(p);
1194         if (likely(!err))
1195                 return 1;
1196
1197         if (type != PIDTYPE_PID)
1198                 return 0;
1199         /* This child was explicitly requested, abort */
1200         read_unlock(&tasklist_lock);
1201         return err;
1202 }
1203
1204 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1205                                int why, int status,
1206                                struct siginfo __user *infop,
1207                                struct rusage __user *rusagep)
1208 {
1209         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1210
1211         put_task_struct(p);
1212         if (!retval)
1213                 retval = put_user(SIGCHLD, &infop->si_signo);
1214         if (!retval)
1215                 retval = put_user(0, &infop->si_errno);
1216         if (!retval)
1217                 retval = put_user((short)why, &infop->si_code);
1218         if (!retval)
1219                 retval = put_user(pid, &infop->si_pid);
1220         if (!retval)
1221                 retval = put_user(uid, &infop->si_uid);
1222         if (!retval)
1223                 retval = put_user(status, &infop->si_status);
1224         if (!retval)
1225                 retval = pid;
1226         return retval;
1227 }
1228
1229 /*
1230  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1231  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1232  * the lock and this task is uninteresting.  If we return nonzero, we have
1233  * released the lock and the system call should return.
1234  */
1235 static int wait_task_zombie(struct task_struct *p, int noreap,
1236                             struct siginfo __user *infop,
1237                             int __user *stat_addr, struct rusage __user *ru)
1238 {
1239         unsigned long state;
1240         int retval, status, traced;
1241         pid_t pid = task_pid_vnr(p);
1242
1243         if (unlikely(noreap)) {
1244                 uid_t uid = p->uid;
1245                 int exit_code = p->exit_code;
1246                 int why, status;
1247
1248                 get_task_struct(p);
1249                 read_unlock(&tasklist_lock);
1250                 if ((exit_code & 0x7f) == 0) {
1251                         why = CLD_EXITED;
1252                         status = exit_code >> 8;
1253                 } else {
1254                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1255                         status = exit_code & 0x7f;
1256                 }
1257                 return wait_noreap_copyout(p, pid, uid, why,
1258                                            status, infop, ru);
1259         }
1260
1261         /*
1262          * Try to move the task's state to DEAD
1263          * only one thread is allowed to do this:
1264          */
1265         state = xchg(&p->exit_state, EXIT_DEAD);
1266         if (state != EXIT_ZOMBIE) {
1267                 BUG_ON(state != EXIT_DEAD);
1268                 return 0;
1269         }
1270
1271         traced = ptrace_reparented(p);
1272
1273         if (likely(!traced)) {
1274                 struct signal_struct *psig;
1275                 struct signal_struct *sig;
1276
1277                 /*
1278                  * The resource counters for the group leader are in its
1279                  * own task_struct.  Those for dead threads in the group
1280                  * are in its signal_struct, as are those for the child
1281                  * processes it has previously reaped.  All these
1282                  * accumulate in the parent's signal_struct c* fields.
1283                  *
1284                  * We don't bother to take a lock here to protect these
1285                  * p->signal fields, because they are only touched by
1286                  * __exit_signal, which runs with tasklist_lock
1287                  * write-locked anyway, and so is excluded here.  We do
1288                  * need to protect the access to p->parent->signal fields,
1289                  * as other threads in the parent group can be right
1290                  * here reaping other children at the same time.
1291                  */
1292                 spin_lock_irq(&p->parent->sighand->siglock);
1293                 psig = p->parent->signal;
1294                 sig = p->signal;
1295                 psig->cutime =
1296                         cputime_add(psig->cutime,
1297                         cputime_add(p->utime,
1298                         cputime_add(sig->utime,
1299                                     sig->cutime)));
1300                 psig->cstime =
1301                         cputime_add(psig->cstime,
1302                         cputime_add(p->stime,
1303                         cputime_add(sig->stime,
1304                                     sig->cstime)));
1305                 psig->cgtime =
1306                         cputime_add(psig->cgtime,
1307                         cputime_add(p->gtime,
1308                         cputime_add(sig->gtime,
1309                                     sig->cgtime)));
1310                 psig->cmin_flt +=
1311                         p->min_flt + sig->min_flt + sig->cmin_flt;
1312                 psig->cmaj_flt +=
1313                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1314                 psig->cnvcsw +=
1315                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1316                 psig->cnivcsw +=
1317                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1318                 psig->cinblock +=
1319                         task_io_get_inblock(p) +
1320                         sig->inblock + sig->cinblock;
1321                 psig->coublock +=
1322                         task_io_get_oublock(p) +
1323                         sig->oublock + sig->coublock;
1324                 spin_unlock_irq(&p->parent->sighand->siglock);
1325         }
1326
1327         /*
1328          * Now we are sure this task is interesting, and no other
1329          * thread can reap it because we set its state to EXIT_DEAD.
1330          */
1331         read_unlock(&tasklist_lock);
1332
1333         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1334         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1335                 ? p->signal->group_exit_code : p->exit_code;
1336         if (!retval && stat_addr)
1337                 retval = put_user(status, stat_addr);
1338         if (!retval && infop)
1339                 retval = put_user(SIGCHLD, &infop->si_signo);
1340         if (!retval && infop)
1341                 retval = put_user(0, &infop->si_errno);
1342         if (!retval && infop) {
1343                 int why;
1344
1345                 if ((status & 0x7f) == 0) {
1346                         why = CLD_EXITED;
1347                         status >>= 8;
1348                 } else {
1349                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1350                         status &= 0x7f;
1351                 }
1352                 retval = put_user((short)why, &infop->si_code);
1353                 if (!retval)
1354                         retval = put_user(status, &infop->si_status);
1355         }
1356         if (!retval && infop)
1357                 retval = put_user(pid, &infop->si_pid);
1358         if (!retval && infop)
1359                 retval = put_user(p->uid, &infop->si_uid);
1360         if (!retval)
1361                 retval = pid;
1362
1363         if (traced) {
1364                 write_lock_irq(&tasklist_lock);
1365                 /* We dropped tasklist, ptracer could die and untrace */
1366                 ptrace_unlink(p);
1367                 /*
1368                  * If this is not a detached task, notify the parent.
1369                  * If it's still not detached after that, don't release
1370                  * it now.
1371                  */
1372                 if (!task_detached(p)) {
1373                         do_notify_parent(p, p->exit_signal);
1374                         if (!task_detached(p)) {
1375                                 p->exit_state = EXIT_ZOMBIE;
1376                                 p = NULL;
1377                         }
1378                 }
1379                 write_unlock_irq(&tasklist_lock);
1380         }
1381         if (p != NULL)
1382                 release_task(p);
1383
1384         return retval;
1385 }
1386
1387 /*
1388  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1389  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1390  * the lock and this task is uninteresting.  If we return nonzero, we have
1391  * released the lock and the system call should return.
1392  */
1393 static int wait_task_stopped(struct task_struct *p,
1394                              int noreap, struct siginfo __user *infop,
1395                              int __user *stat_addr, struct rusage __user *ru)
1396 {
1397         int retval, exit_code, why;
1398         uid_t uid = 0; /* unneeded, required by compiler */
1399         pid_t pid;
1400
1401         exit_code = 0;
1402         spin_lock_irq(&p->sighand->siglock);
1403
1404         if (unlikely(!task_is_stopped_or_traced(p)))
1405                 goto unlock_sig;
1406
1407         if (!(p->ptrace & PT_PTRACED) && p->signal->group_stop_count > 0)
1408                 /*
1409                  * A group stop is in progress and this is the group leader.
1410                  * We won't report until all threads have stopped.
1411                  */
1412                 goto unlock_sig;
1413
1414         exit_code = p->exit_code;
1415         if (!exit_code)
1416                 goto unlock_sig;
1417
1418         if (!noreap)
1419                 p->exit_code = 0;
1420
1421         uid = p->uid;
1422 unlock_sig:
1423         spin_unlock_irq(&p->sighand->siglock);
1424         if (!exit_code)
1425                 return 0;
1426
1427         /*
1428          * Now we are pretty sure this task is interesting.
1429          * Make sure it doesn't get reaped out from under us while we
1430          * give up the lock and then examine it below.  We don't want to
1431          * keep holding onto the tasklist_lock while we call getrusage and
1432          * possibly take page faults for user memory.
1433          */
1434         get_task_struct(p);
1435         pid = task_pid_vnr(p);
1436         why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1437         read_unlock(&tasklist_lock);
1438
1439         if (unlikely(noreap))
1440                 return wait_noreap_copyout(p, pid, uid,
1441                                            why, exit_code,
1442                                            infop, ru);
1443
1444         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1445         if (!retval && stat_addr)
1446                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1447         if (!retval && infop)
1448                 retval = put_user(SIGCHLD, &infop->si_signo);
1449         if (!retval && infop)
1450                 retval = put_user(0, &infop->si_errno);
1451         if (!retval && infop)
1452                 retval = put_user((short)why, &infop->si_code);
1453         if (!retval && infop)
1454                 retval = put_user(exit_code, &infop->si_status);
1455         if (!retval && infop)
1456                 retval = put_user(pid, &infop->si_pid);
1457         if (!retval && infop)
1458                 retval = put_user(uid, &infop->si_uid);
1459         if (!retval)
1460                 retval = pid;
1461         put_task_struct(p);
1462
1463         BUG_ON(!retval);
1464         return retval;
1465 }
1466
1467 /*
1468  * Handle do_wait work for one task in a live, non-stopped state.
1469  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1470  * the lock and this task is uninteresting.  If we return nonzero, we have
1471  * released the lock and the system call should return.
1472  */
1473 static int wait_task_continued(struct task_struct *p, int noreap,
1474                                struct siginfo __user *infop,
1475                                int __user *stat_addr, struct rusage __user *ru)
1476 {
1477         int retval;
1478         pid_t pid;
1479         uid_t uid;
1480
1481         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1482                 return 0;
1483
1484         spin_lock_irq(&p->sighand->siglock);
1485         /* Re-check with the lock held.  */
1486         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1487                 spin_unlock_irq(&p->sighand->siglock);
1488                 return 0;
1489         }
1490         if (!noreap)
1491                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1492         spin_unlock_irq(&p->sighand->siglock);
1493
1494         pid = task_pid_vnr(p);
1495         uid = p->uid;
1496         get_task_struct(p);
1497         read_unlock(&tasklist_lock);
1498
1499         if (!infop) {
1500                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1501                 put_task_struct(p);
1502                 if (!retval && stat_addr)
1503                         retval = put_user(0xffff, stat_addr);
1504                 if (!retval)
1505                         retval = pid;
1506         } else {
1507                 retval = wait_noreap_copyout(p, pid, uid,
1508                                              CLD_CONTINUED, SIGCONT,
1509                                              infop, ru);
1510                 BUG_ON(retval == 0);
1511         }
1512
1513         return retval;
1514 }
1515
1516 static long do_wait(enum pid_type type, struct pid *pid, int options,
1517                     struct siginfo __user *infop, int __user *stat_addr,
1518                     struct rusage __user *ru)
1519 {
1520         DECLARE_WAITQUEUE(wait, current);
1521         struct task_struct *tsk;
1522         int flag, retval;
1523
1524         add_wait_queue(&current->signal->wait_chldexit,&wait);
1525 repeat:
1526         /* If there is nothing that can match our critier just get out */
1527         retval = -ECHILD;
1528         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1529                 goto end;
1530
1531         /*
1532          * We will set this flag if we see any child that might later
1533          * match our criteria, even if we are not able to reap it yet.
1534          */
1535         flag = retval = 0;
1536         current->state = TASK_INTERRUPTIBLE;
1537         read_lock(&tasklist_lock);
1538         tsk = current;
1539         do {
1540                 struct task_struct *p;
1541
1542                 list_for_each_entry(p, &tsk->children, sibling) {
1543                         int ret = eligible_child(type, pid, options, p);
1544                         if (!ret)
1545                                 continue;
1546
1547                         if (unlikely(ret < 0)) {
1548                                 retval = ret;
1549                         } else if (task_is_stopped_or_traced(p)) {
1550                                 /*
1551                                  * It's stopped now, so it might later
1552                                  * continue, exit, or stop again.
1553                                  */
1554                                 flag = 1;
1555                                 if (!(p->ptrace & PT_PTRACED) &&
1556                                     !(options & WUNTRACED))
1557                                         continue;
1558
1559                                 retval = wait_task_stopped(p,
1560                                                 (options & WNOWAIT), infop,
1561                                                 stat_addr, ru);
1562                         } else if (p->exit_state == EXIT_ZOMBIE &&
1563                                         !delay_group_leader(p)) {
1564                                 /*
1565                                  * We don't reap group leaders with subthreads.
1566                                  */
1567                                 if (!likely(options & WEXITED))
1568                                         continue;
1569                                 retval = wait_task_zombie(p,
1570                                                 (options & WNOWAIT), infop,
1571                                                 stat_addr, ru);
1572                         } else if (p->exit_state != EXIT_DEAD) {
1573                                 /*
1574                                  * It's running now, so it might later
1575                                  * exit, stop, or stop and then continue.
1576                                  */
1577                                 flag = 1;
1578                                 if (!unlikely(options & WCONTINUED))
1579                                         continue;
1580                                 retval = wait_task_continued(p,
1581                                                 (options & WNOWAIT), infop,
1582                                                 stat_addr, ru);
1583                         }
1584                         if (retval != 0) /* tasklist_lock released */
1585                                 goto end;
1586                 }
1587                 if (!flag) {
1588                         list_for_each_entry(p, &tsk->ptrace_children,
1589                                                                 ptrace_list) {
1590                                 flag = eligible_child(type, pid, options, p);
1591                                 if (!flag)
1592                                         continue;
1593                                 if (likely(flag > 0))
1594                                         break;
1595                                 retval = flag;
1596                                 goto end;
1597                         }
1598                 }
1599                 if (options & __WNOTHREAD)
1600                         break;
1601                 tsk = next_thread(tsk);
1602                 BUG_ON(tsk->signal != current->signal);
1603         } while (tsk != current);
1604         read_unlock(&tasklist_lock);
1605
1606         if (flag) {
1607                 if (options & WNOHANG)
1608                         goto end;
1609                 retval = -ERESTARTSYS;
1610                 if (signal_pending(current))
1611                         goto end;
1612                 schedule();
1613                 goto repeat;
1614         }
1615         retval = -ECHILD;
1616 end:
1617         current->state = TASK_RUNNING;
1618         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1619         if (infop) {
1620                 if (retval > 0)
1621                         retval = 0;
1622                 else {
1623                         /*
1624                          * For a WNOHANG return, clear out all the fields
1625                          * we would set so the user can easily tell the
1626                          * difference.
1627                          */
1628                         if (!retval)
1629                                 retval = put_user(0, &infop->si_signo);
1630                         if (!retval)
1631                                 retval = put_user(0, &infop->si_errno);
1632                         if (!retval)
1633                                 retval = put_user(0, &infop->si_code);
1634                         if (!retval)
1635                                 retval = put_user(0, &infop->si_pid);
1636                         if (!retval)
1637                                 retval = put_user(0, &infop->si_uid);
1638                         if (!retval)
1639                                 retval = put_user(0, &infop->si_status);
1640                 }
1641         }
1642         return retval;
1643 }
1644
1645 asmlinkage long sys_waitid(int which, pid_t upid,
1646                            struct siginfo __user *infop, int options,
1647                            struct rusage __user *ru)
1648 {
1649         struct pid *pid = NULL;
1650         enum pid_type type;
1651         long ret;
1652
1653         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1654                 return -EINVAL;
1655         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1656                 return -EINVAL;
1657
1658         switch (which) {
1659         case P_ALL:
1660                 type = PIDTYPE_MAX;
1661                 break;
1662         case P_PID:
1663                 type = PIDTYPE_PID;
1664                 if (upid <= 0)
1665                         return -EINVAL;
1666                 break;
1667         case P_PGID:
1668                 type = PIDTYPE_PGID;
1669                 if (upid <= 0)
1670                         return -EINVAL;
1671                 break;
1672         default:
1673                 return -EINVAL;
1674         }
1675
1676         if (type < PIDTYPE_MAX)
1677                 pid = find_get_pid(upid);
1678         ret = do_wait(type, pid, options, infop, NULL, ru);
1679         put_pid(pid);
1680
1681         /* avoid REGPARM breakage on x86: */
1682         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1683         return ret;
1684 }
1685
1686 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1687                           int options, struct rusage __user *ru)
1688 {
1689         struct pid *pid = NULL;
1690         enum pid_type type;
1691         long ret;
1692
1693         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1694                         __WNOTHREAD|__WCLONE|__WALL))
1695                 return -EINVAL;
1696
1697         if (upid == -1)
1698                 type = PIDTYPE_MAX;
1699         else if (upid < 0) {
1700                 type = PIDTYPE_PGID;
1701                 pid = find_get_pid(-upid);
1702         } else if (upid == 0) {
1703                 type = PIDTYPE_PGID;
1704                 pid = get_pid(task_pgrp(current));
1705         } else /* upid > 0 */ {
1706                 type = PIDTYPE_PID;
1707                 pid = find_get_pid(upid);
1708         }
1709
1710         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1711         put_pid(pid);
1712
1713         /* avoid REGPARM breakage on x86: */
1714         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1715         return ret;
1716 }
1717
1718 #ifdef __ARCH_WANT_SYS_WAITPID
1719
1720 /*
1721  * sys_waitpid() remains for compatibility. waitpid() should be
1722  * implemented by calling sys_wait4() from libc.a.
1723  */
1724 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1725 {
1726         return sys_wait4(pid, stat_addr, options, NULL);
1727 }
1728
1729 #endif