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         /*
 130          * Do this under ->siglock, we can race with another thread
 131          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
 132          */
 133         flush_sigqueue(&tsk->pending);
 134
 135         tsk->signal = NULL;
 136         tsk->sighand = NULL;
 137         spin_unlock(&sighand->siglock);
 138         rcu_read_unlock();
 139
 140         __cleanup_sighand(sighand);
 141         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 142         if (sig) {
 143                 flush_sigqueue(&sig->shared_pending);
 144                 taskstats_tgid_free(sig);
 145                 __cleanup_signal(sig);
 146         }
 147 }
 148
 149 static void delayed_put_task_struct(struct rcu_head *rhp)
 150 {
 151         put_task_struct(container_of(rhp, struct task_struct, rcu));
 152 }
 153
 154 void release_task(struct task_struct * p)
 155 {
 156         struct task_struct *leader;
 157         int zap_leader;
 158 repeat:
 159         atomic_dec(&p->user->processes);
 160         proc_flush_task(p);
 161         write_lock_irq(&tasklist_lock);
 162         ptrace_unlink(p);
 163         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 164         __exit_signal(p);
 165
 166         /*
 167          * If we are the last non-leader member of the thread
 168          * group, and the leader is zombie, then notify the
 169          * group leader's parent process. (if it wants notification.)
 170          */
 171         zap_leader = 0;
 172         leader = p->group_leader;
 173         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 174                 BUG_ON(task_detached(leader));
 175                 do_notify_parent(leader, leader->exit_signal);
 176                 /*
 177                  * If we were the last child thread and the leader has
 178                  * exited already, and the leader's parent ignores SIGCHLD,
 179                  * then we are the one who should release the leader.
 180                  *
 181                  * do_notify_parent() will have marked it self-reaping in
 182                  * that case.
 183                  */
 184                 zap_leader = task_detached(leader);
 185         }
 186
 187         write_unlock_irq(&tasklist_lock);
 188         release_thread(p);
 189         call_rcu(&p->rcu, delayed_put_task_struct);
 190
 191         p = leader;
 192         if (unlikely(zap_leader))
 193                 goto repeat;
 194 }
 195
 196 /*
 197  * This checks not only the pgrp, but falls back on the pid if no
 198  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 199  * without this...
 200  *
 201  * The caller must hold rcu lock or the tasklist lock.
 202  */
 203 struct pid *session_of_pgrp(struct pid *pgrp)
 204 {
 205         struct task_struct *p;
 206         struct pid *sid = NULL;
 207
 208         p = pid_task(pgrp, PIDTYPE_PGID);
 209         if (p == NULL)
 210                 p = pid_task(pgrp, PIDTYPE_PID);
 211         if (p != NULL)
 212                 sid = task_session(p);
 213
 214         return sid;
 215 }
 216
 217 /*
 218  * Determine if a process group is "orphaned", according to the POSIX
 219  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 220  * by terminal-generated stop signals.  Newly orphaned process groups are
 221  * to receive a SIGHUP and a SIGCONT.
 222  *
 223  * "I ask you, have you ever known what it is to be an orphan?"
 224  */
 225 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 226 {
 227         struct task_struct *p;
 228
 229         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 230                 if ((p == ignored_task) ||
 231                     (p->exit_state && thread_group_empty(p)) ||
 232                     is_global_init(p->real_parent))
 233                         continue;
 234
 235                 if (task_pgrp(p->real_parent) != pgrp &&
 236                     task_session(p->real_parent) == task_session(p))
 237                         return 0;
 238         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 239
 240         return 1;
 241 }
 242
 243 int is_current_pgrp_orphaned(void)
 244 {
 245         int retval;
 246
 247         read_lock(&tasklist_lock);
 248         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 249         read_unlock(&tasklist_lock);
 250
 251         return retval;
 252 }
 253
 254 static int has_stopped_jobs(struct pid *pgrp)
 255 {
 256         int retval = 0;
 257         struct task_struct *p;
 258
 259         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 260                 if (!task_is_stopped(p))
 261                         continue;
 262                 retval = 1;
 263                 break;
 264         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 265         return retval;
 266 }
 267
 268 /*
 269  * Check to see if any process groups have become orphaned as
 270  * a result of our exiting, and if they have any stopped jobs,
 271  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
 272  */
 273 static void
 274 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
 275 {
 276         struct pid *pgrp = task_pgrp(tsk);
 277         struct task_struct *ignored_task = tsk;
 278
 279         if (!parent)
 280                  /* exit: our father is in a different pgrp than
 281                   * we are and we were the only connection outside.
 282                   */
 283                 parent = tsk->real_parent;
 284         else
 285                 /* reparent: our child is in a different pgrp than
 286                  * we are, and it was the only connection outside.
 287                  */
 288                 ignored_task = NULL;
 289
 290         if (task_pgrp(parent) != pgrp &&
 291             task_session(parent) == task_session(tsk) &&
 292             will_become_orphaned_pgrp(pgrp, ignored_task) &&
 293             has_stopped_jobs(pgrp)) {
 294                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 295                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 296         }
 297 }
 298
 299 /**
 300  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
 301  *
 302  * If a kernel thread is launched as a result of a system call, or if
 303  * it ever exits, it should generally reparent itself to kthreadd so it
 304  * isn't in the way of other processes and is correctly cleaned up on exit.
 305  *
 306  * The various task state such as scheduling policy and priority may have
 307  * been inherited from a user process, so we reset them to sane values here.
 308  *
 309  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
 310  */
 311 static void reparent_to_kthreadd(void)
 312 {
 313         write_lock_irq(&tasklist_lock);
 314
 315         ptrace_unlink(current);
 316         /* Reparent to init */
 317         remove_parent(current);
 318         current->real_parent = current->parent = kthreadd_task;
 319         add_parent(current);
 320
 321         /* Set the exit signal to SIGCHLD so we signal init on exit */
 322         current->exit_signal = SIGCHLD;
 323
 324         if (task_nice(current) < 0)
 325                 set_user_nice(current, 0);
 326         /* cpus_allowed? */
 327         /* rt_priority? */
 328         /* signals? */
 329         security_task_reparent_to_init(current);
 330         memcpy(current->signal->rlim, init_task.signal->rlim,
 331                sizeof(current->signal->rlim));
 332         atomic_inc(&(INIT_USER->__count));
 333         write_unlock_irq(&tasklist_lock);
 334         switch_uid(INIT_USER);
 335 }
 336
 337 void __set_special_pids(struct pid *pid)
 338 {
 339         struct task_struct *curr = current->group_leader;
 340         pid_t nr = pid_nr(pid);
 341
 342         if (task_session(curr) != pid) {
 343                 change_pid(curr, PIDTYPE_SID, pid);
 344                 set_task_session(curr, nr);
 345         }
 346         if (task_pgrp(curr) != pid) {
 347                 change_pid(curr, PIDTYPE_PGID, pid);
 348                 set_task_pgrp(curr, nr);
 349         }
 350 }
 351
 352 static void set_special_pids(struct pid *pid)
 353 {
 354         write_lock_irq(&tasklist_lock);
 355         __set_special_pids(pid);
 356         write_unlock_irq(&tasklist_lock);
 357 }
 358
 359 /*
 360  * Let kernel threads use this to say that they
 361  * allow a certain signal (since daemonize() will
 362  * have disabled all of them by default).
 363  */
 364 int allow_signal(int sig)
 365 {
 366         if (!valid_signal(sig) || sig < 1)
 367                 return -EINVAL;
 368
 369         spin_lock_irq(&current->sighand->siglock);
 370         sigdelset(&current->blocked, sig);
 371         if (!current->mm) {
 372                 /* Kernel threads handle their own signals.
 373                    Let the signal code know it'll be handled, so
 374                    that they don't get converted to SIGKILL or
 375                    just silently dropped */
 376                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 377         }
 378         recalc_sigpending();
 379         spin_unlock_irq(&current->sighand->siglock);
 380         return 0;
 381 }
 382
 383 EXPORT_SYMBOL(allow_signal);
 384
 385 int disallow_signal(int sig)
 386 {
 387         if (!valid_signal(sig) || sig < 1)
 388                 return -EINVAL;
 389
 390         spin_lock_irq(&current->sighand->siglock);
 391         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
 392         recalc_sigpending();
 393         spin_unlock_irq(&current->sighand->siglock);
 394         return 0;
 395 }
 396
 397 EXPORT_SYMBOL(disallow_signal);
 398
 399 /*
 400  *      Put all the gunge required to become a kernel thread without
 401  *      attached user resources in one place where it belongs.
 402  */
 403
 404 void daemonize(const char *name, ...)
 405 {
 406         va_list args;
 407         struct fs_struct *fs;
 408         sigset_t blocked;
 409
 410         va_start(args, name);
 411         vsnprintf(current->comm, sizeof(current->comm), name, args);
 412         va_end(args);
 413
 414         /*
 415          * If we were started as result of loading a module, close all of the
 416          * user space pages.  We don't need them, and if we didn't close them
 417          * they would be locked into memory.
 418          */
 419         exit_mm(current);
 420         /*
 421          * We don't want to have TIF_FREEZE set if the system-wide hibernation
 422          * or suspend transition begins right now.
 423          */
 424         current->flags |= PF_NOFREEZE;
 425
 426         if (current->nsproxy != &init_nsproxy) {
 427                 get_nsproxy(&init_nsproxy);
 428                 switch_task_namespaces(current, &init_nsproxy);
 429         }
 430         set_special_pids(&init_struct_pid);
 431         proc_clear_tty(current);
 432
 433         /* Block and flush all signals */
 434         sigfillset(&blocked);
 435         sigprocmask(SIG_BLOCK, &blocked, NULL);
 436         flush_signals(current);
 437
 438         /* Become as one with the init task */
 439
 440         exit_fs(current);       /* current->fs->count--; */
 441         fs = init_task.fs;
 442         current->fs = fs;
 443         atomic_inc(&fs->count);
 444
 445         exit_files(current);
 446         current->files = init_task.files;
 447         atomic_inc(&current->files->count);
 448
 449         reparent_to_kthreadd();
 450 }
 451
 452 EXPORT_SYMBOL(daemonize);
 453
 454 static void close_files(struct files_struct * files)
 455 {
 456         int i, j;
 457         struct fdtable *fdt;
 458
 459         j = 0;
 460
 461         /*
 462          * It is safe to dereference the fd table without RCU or
 463          * ->file_lock because this is the last reference to the
 464          * files structure.
 465          */
 466         fdt = files_fdtable(files);
 467         for (;;) {
 468                 unsigned long set;
 469                 i = j * __NFDBITS;
 470                 if (i >= fdt->max_fds)
 471                         break;
 472                 set = fdt->open_fds->fds_bits[j++];
 473                 while (set) {
 474                         if (set & 1) {
 475                                 struct file * file = xchg(&fdt->fd[i], NULL);
 476                                 if (file) {
 477                                         filp_close(file, files);
 478                                         cond_resched();
 479                                 }
 480                         }
 481                         i++;
 482                         set >>= 1;
 483                 }
 484         }
 485 }
 486
 487 struct files_struct *get_files_struct(struct task_struct *task)
 488 {
 489         struct files_struct *files;
 490
 491         task_lock(task);
 492         files = task->files;
 493         if (files)
 494                 atomic_inc(&files->count);
 495         task_unlock(task);
 496
 497         return files;
 498 }
 499
 500 void put_files_struct(struct files_struct *files)
 501 {
 502         struct fdtable *fdt;
 503
 504         if (atomic_dec_and_test(&files->count)) {
 505                 close_files(files);
 506                 /*
 507                  * Free the fd and fdset arrays if we expanded them.
 508                  * If the fdtable was embedded, pass files for freeing
 509                  * at the end of the RCU grace period. Otherwise,
 510                  * you can free files immediately.
 511                  */
 512                 fdt = files_fdtable(files);
 513                 if (fdt != &files->fdtab)
 514                         kmem_cache_free(files_cachep, files);
 515                 free_fdtable(fdt);
 516         }
 517 }
 518
 519 void reset_files_struct(struct files_struct *files)
 520 {
 521         struct task_struct *tsk = current;
 522         struct files_struct *old;
 523
 524         old = tsk->files;
 525         task_lock(tsk);
 526         tsk->files = files;
 527         task_unlock(tsk);
 528         put_files_struct(old);
 529 }
 530
 531 void exit_files(struct task_struct *tsk)
 532 {
 533         struct files_struct * files = tsk->files;
 534
 535         if (files) {
 536                 task_lock(tsk);
 537                 tsk->files = NULL;
 538                 task_unlock(tsk);
 539                 put_files_struct(files);
 540         }
 541 }
 542
 543 void put_fs_struct(struct fs_struct *fs)
 544 {
 545         /* No need to hold fs->lock if we are killing it */
 546         if (atomic_dec_and_test(&fs->count)) {
 547                 path_put(&fs->root);
 548                 path_put(&fs->pwd);
 549                 if (fs->altroot.dentry)
 550                         path_put(&fs->altroot);
 551                 kmem_cache_free(fs_cachep, fs);
 552         }
 553 }
 554
 555 void exit_fs(struct task_struct *tsk)
 556 {
 557         struct fs_struct * fs = tsk->fs;
 558
 559         if (fs) {
 560                 task_lock(tsk);
 561                 tsk->fs = NULL;
 562                 task_unlock(tsk);
 563                 put_fs_struct(fs);
 564         }
 565 }
 566
 567 EXPORT_SYMBOL_GPL(exit_fs);
 568
 569 #ifdef CONFIG_MM_OWNER
 570 /*
 571  * Task p is exiting and it owned mm, lets find a new owner for it
 572  */
 573 static inline int
 574 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
 575 {
 576         /*
 577          * If there are other users of the mm and the owner (us) is exiting
 578          * we need to find a new owner to take on the responsibility.
 579          */
 580         if (!mm)
 581                 return 0;
 582         if (atomic_read(&mm->mm_users) <= 1)
 583                 return 0;
 584         if (mm->owner != p)
 585                 return 0;
 586         return 1;
 587 }
 588
 589 void mm_update_next_owner(struct mm_struct *mm)
 590 {
 591         struct task_struct *c, *g, *p = current;
 592
 593 retry:
 594         if (!mm_need_new_owner(mm, p))
 595                 return;
 596
 597         read_lock(&tasklist_lock);
 598         /*
 599          * Search in the children
 600          */
 601         list_for_each_entry(c, &p->children, sibling) {
 602                 if (c->mm == mm)
 603                         goto assign_new_owner;
 604         }
 605
 606         /*
 607          * Search in the siblings
 608          */
 609         list_for_each_entry(c, &p->parent->children, sibling) {
 610                 if (c->mm == mm)
 611                         goto assign_new_owner;
 612         }
 613
 614         /*
 615          * Search through everything else. We should not get
 616          * here often
 617          */
 618         do_each_thread(g, c) {
 619                 if (c->mm == mm)
 620                         goto assign_new_owner;
 621         } while_each_thread(g, c);
 622
 623         read_unlock(&tasklist_lock);
 624         return;
 625
 626 assign_new_owner:
 627         BUG_ON(c == p);
 628         get_task_struct(c);
 629         /*
 630          * The task_lock protects c->mm from changing.
 631          * We always want mm->owner->mm == mm
 632          */
 633         task_lock(c);
 634         /*
 635          * Delay read_unlock() till we have the task_lock()
 636          * to ensure that c does not slip away underneath us
 637          */
 638         read_unlock(&tasklist_lock);
 639         if (c->mm != mm) {
 640                 task_unlock(c);
 641                 put_task_struct(c);
 642                 goto retry;
 643         }
 644         cgroup_mm_owner_callbacks(mm->owner, c);
 645         mm->owner = c;
 646         task_unlock(c);
 647         put_task_struct(c);
 648 }
 649 #endif /* CONFIG_MM_OWNER */
 650
 651 /*
 652  * Turn us into a lazy TLB process if we
 653  * aren't already..
 654  */
 655 static void exit_mm(struct task_struct * tsk)
 656 {
 657         struct mm_struct *mm = tsk->mm;
 658
 659         mm_release(tsk, mm);
 660         if (!mm)
 661                 return;
 662         /*
 663          * Serialize with any possible pending coredump.
 664          * We must hold mmap_sem around checking core_waiters
 665          * and clearing tsk->mm.  The core-inducing thread
 666          * will increment core_waiters for each thread in the
 667          * group with ->mm != NULL.
 668          */
 669         down_read(&mm->mmap_sem);
 670         if (mm->core_waiters) {
 671                 up_read(&mm->mmap_sem);
 672                 down_write(&mm->mmap_sem);
 673                 if (!--mm->core_waiters)
 674                         complete(mm->core_startup_done);
 675                 up_write(&mm->mmap_sem);
 676
 677                 wait_for_completion(&mm->core_done);
 678                 down_read(&mm->mmap_sem);
 679         }
 680         atomic_inc(&mm->mm_count);
 681         BUG_ON(mm != tsk->active_mm);
 682         /* more a memory barrier than a real lock */
 683         task_lock(tsk);
 684         tsk->mm = NULL;
 685         up_read(&mm->mmap_sem);
 686         enter_lazy_tlb(mm, current);
 687         /* We don't want this task to be frozen prematurely */
 688         clear_freeze_flag(tsk);
 689         task_unlock(tsk);
 690         mm_update_next_owner(mm);
 691         mmput(mm);
 692 }
 693
 694 static void
 695 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 696 {
 697         if (p->pdeath_signal)
 698                 /* We already hold the tasklist_lock here.  */
 699                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 700
 701         /* Move the child from its dying parent to the new one.  */
 702         if (unlikely(traced)) {
 703                 /* Preserve ptrace links if someone else is tracing this child.  */
 704                 list_del_init(&p->ptrace_list);
 705                 if (ptrace_reparented(p))
 706                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 707         } else {
 708                 /* If this child is being traced, then we're the one tracing it
 709                  * anyway, so let go of it.
 710                  */
 711                 p->ptrace = 0;
 712                 remove_parent(p);
 713                 p->parent = p->real_parent;
 714                 add_parent(p);
 715
 716                 if (task_is_traced(p)) {
 717                         /*
 718                          * If it was at a trace stop, turn it into
 719                          * a normal stop since it's no longer being
 720                          * traced.
 721                          */
 722                         ptrace_untrace(p);
 723                 }
 724         }
 725
 726         /* If this is a threaded reparent there is no need to
 727          * notify anyone anything has happened.
 728          */
 729         if (same_thread_group(p->real_parent, father))
 730                 return;
 731
 732         /* We don't want people slaying init.  */
 733         if (!task_detached(p))
 734                 p->exit_signal = SIGCHLD;
 735
 736         /* If we'd notified the old parent about this child's death,
 737          * also notify the new parent.
 738          */
 739         if (!traced && p->exit_state == EXIT_ZOMBIE &&
 740             !task_detached(p) && thread_group_empty(p))
 741                 do_notify_parent(p, p->exit_signal);
 742
 743         kill_orphaned_pgrp(p, father);
 744 }
 745
 746 /*
 747  * When we die, we re-parent all our children.
 748  * Try to give them to another thread in our thread
 749  * group, and if no such member exists, give it to
 750  * the child reaper process (ie "init") in our pid
 751  * space.
 752  */
 753 static void forget_original_parent(struct task_struct *father)
 754 {
 755         struct task_struct *p, *n, *reaper = father;
 756         struct list_head ptrace_dead;
 757
 758         INIT_LIST_HEAD(&ptrace_dead);
 759
 760         write_lock_irq(&tasklist_lock);
 761
 762         do {
 763                 reaper = next_thread(reaper);
 764                 if (reaper == father) {
 765                         reaper = task_child_reaper(father);
 766                         break;
 767                 }
 768         } while (reaper->flags & PF_EXITING);
 769
 770         /*
 771          * There are only two places where our children can be:
 772          *
 773          * - in our child list
 774          * - in our ptraced child list
 775          *
 776          * Search them and reparent children.
 777          */
 778         list_for_each_entry_safe(p, n, &father->children, sibling) {
 779                 int ptrace;
 780
 781                 ptrace = p->ptrace;
 782
 783                 /* if father isn't the real parent, then ptrace must be enabled */
 784                 BUG_ON(father != p->real_parent && !ptrace);
 785
 786                 if (father == p->real_parent) {
 787                         /* reparent with a reaper, real father it's us */
 788                         p->real_parent = reaper;
 789                         reparent_thread(p, father, 0);
 790                 } else {
 791                         /* reparent ptraced task to its real parent */
 792                         __ptrace_unlink (p);
 793                         if (p->exit_state == EXIT_ZOMBIE && !task_detached(p) &&
 794                             thread_group_empty(p))
 795                                 do_notify_parent(p, p->exit_signal);
 796                 }
 797
 798                 /*
 799                  * if the ptraced child is a detached zombie we must collect
 800                  * it before we exit, or it will remain zombie forever since
 801                  * we prevented it from self-reap itself while it was being
 802                  * traced by us, to be able to see it in wait4.
 803                  */
 804                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && task_detached(p)))
 805                         list_add(&p->ptrace_list, &ptrace_dead);
 806         }
 807
 808         list_for_each_entry_safe(p, n, &father->ptrace_children, ptrace_list) {
 809                 p->real_parent = reaper;
 810                 reparent_thread(p, father, 1);
 811         }
 812
 813         write_unlock_irq(&tasklist_lock);
 814         BUG_ON(!list_empty(&father->children));
 815         BUG_ON(!list_empty(&father->ptrace_children));
 816
 817         list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_list) {
 818                 list_del_init(&p->ptrace_list);
 819                 release_task(p);
 820         }
 821
 822 }
 823
 824 /*
 825  * Send signals to all our closest relatives so that they know
 826  * to properly mourn us..
 827  */
 828 static void exit_notify(struct task_struct *tsk, int group_dead)
 829 {
 830         int state;
 831
 832         /*
 833          * This does two things:
 834          *
 835          * A.  Make init inherit all the child processes
 836          * B.  Check to see if any process groups have become orphaned
 837          *      as a result of our exiting, and if they have any stopped
 838          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 839          */
 840         forget_original_parent(tsk);
 841         exit_task_namespaces(tsk);
 842
 843         write_lock_irq(&tasklist_lock);
 844         if (group_dead)
 845                 kill_orphaned_pgrp(tsk->group_leader, NULL);
 846
 847         /* Let father know we died
 848          *
 849          * Thread signals are configurable, but you aren't going to use
 850          * that to send signals to arbitary processes.
 851          * That stops right now.
 852          *
 853          * If the parent exec id doesn't match the exec id we saved
 854          * when we started then we know the parent has changed security
 855          * domain.
 856          *
 857          * If our self_exec id doesn't match our parent_exec_id then
 858          * we have changed execution domain as these two values started
 859          * the same after a fork.
 860          */
 861         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
 862             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
 863              tsk->self_exec_id != tsk->parent_exec_id) &&
 864             !capable(CAP_KILL))
 865                 tsk->exit_signal = SIGCHLD;
 866
 867         /* If something other than our normal parent is ptracing us, then
 868          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 869          * only has special meaning to our real parent.
 870          */
 871         if (!task_detached(tsk) && thread_group_empty(tsk)) {
 872                 int signal = ptrace_reparented(tsk) ?
 873                                 SIGCHLD : tsk->exit_signal;
 874                 do_notify_parent(tsk, signal);
 875         } else if (tsk->ptrace) {
 876                 do_notify_parent(tsk, SIGCHLD);
 877         }
 878
 879         state = EXIT_ZOMBIE;
 880         if (task_detached(tsk) && likely(!tsk->ptrace))
 881                 state = EXIT_DEAD;
 882         tsk->exit_state = state;
 883
 884         /* mt-exec, de_thread() is waiting for us */
 885         if (thread_group_leader(tsk) &&
 886             tsk->signal->notify_count < 0 &&
 887             tsk->signal->group_exit_task)
 888                 wake_up_process(tsk->signal->group_exit_task);
 889
 890         write_unlock_irq(&tasklist_lock);
 891
 892         /* If the process is dead, release it - nobody will wait for it */
 893         if (state == EXIT_DEAD)
 894                 release_task(tsk);
 895 }
 896
 897 #ifdef CONFIG_DEBUG_STACK_USAGE
 898 static void check_stack_usage(void)
 899 {
 900         static DEFINE_SPINLOCK(low_water_lock);
 901         static int lowest_to_date = THREAD_SIZE;
 902         unsigned long *n = end_of_stack(current);
 903         unsigned long free;
 904
 905         while (*n == 0)
 906                 n++;
 907         free = (unsigned long)n - (unsigned long)end_of_stack(current);
 908
 909         if (free >= lowest_to_date)
 910                 return;
 911
 912         spin_lock(&low_water_lock);
 913         if (free < lowest_to_date) {
 914                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
 915                                 "left\n",
 916                                 current->comm, free);
 917                 lowest_to_date = free;
 918         }
 919         spin_unlock(&low_water_lock);
 920 }
 921 #else
 922 static inline void check_stack_usage(void) {}
 923 #endif
 924
 925 static inline void exit_child_reaper(struct task_struct *tsk)
 926 {
 927         if (likely(tsk->group_leader != task_child_reaper(tsk)))
 928                 return;
 929
 930         if (tsk->nsproxy->pid_ns == &init_pid_ns)
 931                 panic("Attempted to kill init!");
 932
 933         /*
 934          * @tsk is the last thread in the 'cgroup-init' and is exiting.
 935          * Terminate all remaining processes in the namespace and reap them
 936          * before exiting @tsk.
 937          *
 938          * Note that @tsk (last thread of cgroup-init) may not necessarily
 939          * be the child-reaper (i.e main thread of cgroup-init) of the
 940          * namespace i.e the child_reaper may have already exited.
 941          *
 942          * Even after a child_reaper exits, we let it inherit orphaned children,
 943          * because, pid_ns->child_reaper remains valid as long as there is
 944          * at least one living sub-thread in the cgroup init.
 945
 946          * This living sub-thread of the cgroup-init will be notified when
 947          * a child inherited by the 'child-reaper' exits (do_notify_parent()
 948          * uses __group_send_sig_info()). Further, when reaping child processes,
 949          * do_wait() iterates over children of all living sub threads.
 950
 951          * i.e even though 'child_reaper' thread is listed as the parent of the
 952          * orphaned children, any living sub-thread in the cgroup-init can
 953          * perform the role of the child_reaper.
 954          */
 955         zap_pid_ns_processes(tsk->nsproxy->pid_ns);
 956 }
 957
 958 NORET_TYPE void do_exit(long code)
 959 {
 960         struct task_struct *tsk = current;
 961         int group_dead;
 962
 963         profile_task_exit(tsk);
 964
 965         WARN_ON(atomic_read(&tsk->fs_excl));
 966
 967         if (unlikely(in_interrupt()))
 968                 panic("Aiee, killing interrupt handler!");
 969         if (unlikely(!tsk->pid))
 970                 panic("Attempted to kill the idle task!");
 971
 972         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 973                 current->ptrace_message = code;
 974                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 975         }
 976
 977         /*
 978          * We're taking recursive faults here in do_exit. Safest is to just
 979          * leave this task alone and wait for reboot.
 980          */
 981         if (unlikely(tsk->flags & PF_EXITING)) {
 982                 printk(KERN_ALERT
 983                         "Fixing recursive fault but reboot is needed!\n");
 984                 /*
 985                  * We can do this unlocked here. The futex code uses
 986                  * this flag just to verify whether the pi state
 987                  * cleanup has been done or not. In the worst case it
 988                  * loops once more. We pretend that the cleanup was
 989                  * done as there is no way to return. Either the
 990                  * OWNER_DIED bit is set by now or we push the blocked
 991                  * task into the wait for ever nirwana as well.
 992                  */
 993                 tsk->flags |= PF_EXITPIDONE;
 994                 if (tsk->io_context)
 995                         exit_io_context();
 996                 set_current_state(TASK_UNINTERRUPTIBLE);
 997                 schedule();
 998         }
 999
1000         exit_signals(tsk);  /* sets PF_EXITING */
1001         /*
1002          * tsk->flags are checked in the futex code to protect against
1003          * an exiting task cleaning up the robust pi futexes.
1004          */
1005         smp_mb();
1006         spin_unlock_wait(&tsk->pi_lock);
1007
1008         if (unlikely(in_atomic()))
1009                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1010                                 current->comm, task_pid_nr(current),
1011                                 preempt_count());
1012
1013         acct_update_integrals(tsk);
1014         if (tsk->mm) {
1015                 update_hiwater_rss(tsk->mm);
1016                 update_hiwater_vm(tsk->mm);
1017         }
1018         group_dead = atomic_dec_and_test(&tsk->signal->live);
1019         if (group_dead) {
1020                 exit_child_reaper(tsk);
1021                 hrtimer_cancel(&tsk->signal->real_timer);
1022                 exit_itimers(tsk->signal);
1023         }
1024         acct_collect(code, group_dead);
1025 #ifdef CONFIG_FUTEX
1026         if (unlikely(tsk->robust_list))
1027                 exit_robust_list(tsk);
1028 #ifdef CONFIG_COMPAT
1029         if (unlikely(tsk->compat_robust_list))
1030                 compat_exit_robust_list(tsk);
1031 #endif
1032 #endif
1033         if (group_dead)
1034                 tty_audit_exit();
1035         if (unlikely(tsk->audit_context))
1036                 audit_free(tsk);
1037
1038         tsk->exit_code = code;
1039         taskstats_exit(tsk, group_dead);
1040
1041         exit_mm(tsk);
1042
1043         if (group_dead)
1044                 acct_process();
1045         exit_sem(tsk);
1046         exit_files(tsk);
1047         exit_fs(tsk);
1048         check_stack_usage();
1049         exit_thread();
1050         cgroup_exit(tsk, 1);
1051         exit_keys(tsk);
1052
1053         if (group_dead && tsk->signal->leader)
1054                 disassociate_ctty(1);
1055
1056         module_put(task_thread_info(tsk)->exec_domain->module);
1057         if (tsk->binfmt)
1058                 module_put(tsk->binfmt->module);
1059
1060         proc_exit_connector(tsk);
1061         exit_notify(tsk, group_dead);
1062 #ifdef CONFIG_NUMA
1063         mpol_put(tsk->mempolicy);
1064         tsk->mempolicy = NULL;
1065 #endif
1066 #ifdef CONFIG_FUTEX
1067         /*
1068          * This must happen late, after the PID is not
1069          * hashed anymore:
1070          */
1071         if (unlikely(!list_empty(&tsk->pi_state_list)))
1072                 exit_pi_state_list(tsk);
1073         if (unlikely(current->pi_state_cache))
1074                 kfree(current->pi_state_cache);
1075 #endif
1076         /*
1077          * Make sure we are holding no locks:
1078          */
1079         debug_check_no_locks_held(tsk);
1080         /*
1081          * We can do this unlocked here. The futex code uses this flag
1082          * just to verify whether the pi state cleanup has been done
1083          * or not. In the worst case it loops once more.
1084          */
1085         tsk->flags |= PF_EXITPIDONE;
1086
1087         if (tsk->io_context)
1088                 exit_io_context();
1089
1090         if (tsk->splice_pipe)
1091                 __free_pipe_info(tsk->splice_pipe);
1092
1093         preempt_disable();
1094         /* causes final put_task_struct in finish_task_switch(). */
1095         tsk->state = TASK_DEAD;
1096
1097         schedule();
1098         BUG();
1099         /* Avoid "noreturn function does return".  */
1100         for (;;)
1101                 cpu_relax();    /* For when BUG is null */
1102 }
1103
1104 EXPORT_SYMBOL_GPL(do_exit);
1105
1106 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1107 {
1108         if (comp)
1109                 complete(comp);
1110
1111         do_exit(code);
1112 }
1113
1114 EXPORT_SYMBOL(complete_and_exit);
1115
1116 asmlinkage long sys_exit(int error_code)
1117 {
1118         do_exit((error_code&0xff)<<8);
1119 }
1120
1121 /*
1122  * Take down every thread in the group.  This is called by fatal signals
1123  * as well as by sys_exit_group (below).
1124  */
1125 NORET_TYPE void
1126 do_group_exit(int exit_code)
1127 {
1128         struct signal_struct *sig = current->signal;
1129
1130         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1131
1132         if (signal_group_exit(sig))
1133                 exit_code = sig->group_exit_code;
1134         else if (!thread_group_empty(current)) {
1135                 struct sighand_struct *const sighand = current->sighand;
1136                 spin_lock_irq(&sighand->siglock);
1137                 if (signal_group_exit(sig))
1138                         /* Another thread got here before we took the lock.  */
1139                         exit_code = sig->group_exit_code;
1140                 else {
1141                         sig->group_exit_code = exit_code;
1142                         sig->flags = SIGNAL_GROUP_EXIT;
1143                         zap_other_threads(current);
1144                 }
1145                 spin_unlock_irq(&sighand->siglock);
1146         }
1147
1148         do_exit(exit_code);
1149         /* NOTREACHED */
1150 }
1151
1152 /*
1153  * this kills every thread in the thread group. Note that any externally
1154  * wait4()-ing process will get the correct exit code - even if this
1155  * thread is not the thread group leader.
1156  */
1157 asmlinkage void sys_exit_group(int error_code)
1158 {
1159         do_group_exit((error_code & 0xff) << 8);
1160 }
1161
1162 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1163 {
1164         struct pid *pid = NULL;
1165         if (type == PIDTYPE_PID)
1166                 pid = task->pids[type].pid;
1167         else if (type < PIDTYPE_MAX)
1168                 pid = task->group_leader->pids[type].pid;
1169         return pid;
1170 }
1171
1172 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1173                           struct task_struct *p)
1174 {
1175         int err;
1176
1177         if (type < PIDTYPE_MAX) {
1178                 if (task_pid_type(p, type) != pid)
1179                         return 0;
1180         }
1181
1182         /*
1183          * Do not consider detached threads that are
1184          * not ptraced:
1185          */
1186         if (task_detached(p) && !p->ptrace)
1187                 return 0;
1188
1189         /* Wait for all children (clone and not) if __WALL is set;
1190          * otherwise, wait for clone children *only* if __WCLONE is
1191          * set; otherwise, wait for non-clone children *only*.  (Note:
1192          * A "clone" child here is one that reports to its parent
1193          * using a signal other than SIGCHLD.) */
1194         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1195             && !(options & __WALL))
1196                 return 0;
1197
1198         err = security_task_wait(p);
1199         if (likely(!err))
1200                 return 1;
1201
1202         if (type != PIDTYPE_PID)
1203                 return 0;
1204         /* This child was explicitly requested, abort */
1205         read_unlock(&tasklist_lock);
1206         return err;
1207 }
1208
1209 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1210                                int why, int status,
1211                                struct siginfo __user *infop,
1212                                struct rusage __user *rusagep)
1213 {
1214         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1215
1216         put_task_struct(p);
1217         if (!retval)
1218                 retval = put_user(SIGCHLD, &infop->si_signo);
1219         if (!retval)
1220                 retval = put_user(0, &infop->si_errno);
1221         if (!retval)
1222                 retval = put_user((short)why, &infop->si_code);
1223         if (!retval)
1224                 retval = put_user(pid, &infop->si_pid);
1225         if (!retval)
1226                 retval = put_user(uid, &infop->si_uid);
1227         if (!retval)
1228                 retval = put_user(status, &infop->si_status);
1229         if (!retval)
1230                 retval = pid;
1231         return retval;
1232 }
1233
1234 /*
1235  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1236  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1237  * the lock and this task is uninteresting.  If we return nonzero, we have
1238  * released the lock and the system call should return.
1239  */
1240 static int wait_task_zombie(struct task_struct *p, int noreap,
1241                             struct siginfo __user *infop,
1242                             int __user *stat_addr, struct rusage __user *ru)
1243 {
1244         unsigned long state;
1245         int retval, status, traced;
1246         pid_t pid = task_pid_vnr(p);
1247
1248         if (unlikely(noreap)) {
1249                 uid_t uid = p->uid;
1250                 int exit_code = p->exit_code;
1251                 int why, status;
1252
1253                 get_task_struct(p);
1254                 read_unlock(&tasklist_lock);
1255                 if ((exit_code & 0x7f) == 0) {
1256                         why = CLD_EXITED;
1257                         status = exit_code >> 8;
1258                 } else {
1259                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1260                         status = exit_code & 0x7f;
1261                 }
1262                 return wait_noreap_copyout(p, pid, uid, why,
1263                                            status, infop, ru);
1264         }
1265
1266         /*
1267          * Try to move the task's state to DEAD
1268          * only one thread is allowed to do this:
1269          */
1270         state = xchg(&p->exit_state, EXIT_DEAD);
1271         if (state != EXIT_ZOMBIE) {
1272                 BUG_ON(state != EXIT_DEAD);
1273                 return 0;
1274         }
1275
1276         traced = ptrace_reparented(p);
1277
1278         if (likely(!traced)) {
1279                 struct signal_struct *psig;
1280                 struct signal_struct *sig;
1281
1282                 /*
1283                  * The resource counters for the group leader are in its
1284                  * own task_struct.  Those for dead threads in the group
1285                  * are in its signal_struct, as are those for the child
1286                  * processes it has previously reaped.  All these
1287                  * accumulate in the parent's signal_struct c* fields.
1288                  *
1289                  * We don't bother to take a lock here to protect these
1290                  * p->signal fields, because they are only touched by
1291                  * __exit_signal, which runs with tasklist_lock
1292                  * write-locked anyway, and so is excluded here.  We do
1293                  * need to protect the access to p->parent->signal fields,
1294                  * as other threads in the parent group can be right
1295                  * here reaping other children at the same time.
1296                  */
1297                 spin_lock_irq(&p->parent->sighand->siglock);
1298                 psig = p->parent->signal;
1299                 sig = p->signal;
1300                 psig->cutime =
1301                         cputime_add(psig->cutime,
1302                         cputime_add(p->utime,
1303                         cputime_add(sig->utime,
1304                                     sig->cutime)));
1305                 psig->cstime =
1306                         cputime_add(psig->cstime,
1307                         cputime_add(p->stime,
1308                         cputime_add(sig->stime,
1309                                     sig->cstime)));
1310                 psig->cgtime =
1311                         cputime_add(psig->cgtime,
1312                         cputime_add(p->gtime,
1313                         cputime_add(sig->gtime,
1314                                     sig->cgtime)));
1315                 psig->cmin_flt +=
1316                         p->min_flt + sig->min_flt + sig->cmin_flt;
1317                 psig->cmaj_flt +=
1318                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1319                 psig->cnvcsw +=
1320                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1321                 psig->cnivcsw +=
1322                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1323                 psig->cinblock +=
1324                         task_io_get_inblock(p) +
1325                         sig->inblock + sig->cinblock;
1326                 psig->coublock +=
1327                         task_io_get_oublock(p) +
1328                         sig->oublock + sig->coublock;
1329                 spin_unlock_irq(&p->parent->sighand->siglock);
1330         }
1331
1332         /*
1333          * Now we are sure this task is interesting, and no other
1334          * thread can reap it because we set its state to EXIT_DEAD.
1335          */
1336         read_unlock(&tasklist_lock);
1337
1338         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1339         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1340                 ? p->signal->group_exit_code : p->exit_code;
1341         if (!retval && stat_addr)
1342                 retval = put_user(status, stat_addr);
1343         if (!retval && infop)
1344                 retval = put_user(SIGCHLD, &infop->si_signo);
1345         if (!retval && infop)
1346                 retval = put_user(0, &infop->si_errno);
1347         if (!retval && infop) {
1348                 int why;
1349
1350                 if ((status & 0x7f) == 0) {
1351                         why = CLD_EXITED;
1352                         status >>= 8;
1353                 } else {
1354                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1355                         status &= 0x7f;
1356                 }
1357                 retval = put_user((short)why, &infop->si_code);
1358                 if (!retval)
1359                         retval = put_user(status, &infop->si_status);
1360         }
1361         if (!retval && infop)
1362                 retval = put_user(pid, &infop->si_pid);
1363         if (!retval && infop)
1364                 retval = put_user(p->uid, &infop->si_uid);
1365         if (!retval)
1366                 retval = pid;
1367
1368         if (traced) {
1369                 write_lock_irq(&tasklist_lock);
1370                 /* We dropped tasklist, ptracer could die and untrace */
1371                 ptrace_unlink(p);
1372                 /*
1373                  * If this is not a detached task, notify the parent.
1374                  * If it's still not detached after that, don't release
1375                  * it now.
1376                  */
1377                 if (!task_detached(p)) {
1378                         do_notify_parent(p, p->exit_signal);
1379                         if (!task_detached(p)) {
1380                                 p->exit_state = EXIT_ZOMBIE;
1381                                 p = NULL;
1382                         }
1383                 }
1384                 write_unlock_irq(&tasklist_lock);
1385         }
1386         if (p != NULL)
1387                 release_task(p);
1388
1389         return retval;
1390 }
1391
1392 /*
1393  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1394  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1395  * the lock and this task is uninteresting.  If we return nonzero, we have
1396  * released the lock and the system call should return.
1397  */
1398 static int wait_task_stopped(struct task_struct *p,
1399                              int noreap, struct siginfo __user *infop,
1400                              int __user *stat_addr, struct rusage __user *ru)
1401 {
1402         int retval, exit_code, why;
1403         uid_t uid = 0; /* unneeded, required by compiler */
1404         pid_t pid;
1405
1406         exit_code = 0;
1407         spin_lock_irq(&p->sighand->siglock);
1408
1409         if (unlikely(!task_is_stopped_or_traced(p)))
1410                 goto unlock_sig;
1411
1412         if (!(p->ptrace & PT_PTRACED) && p->signal->group_stop_count > 0)
1413                 /*
1414                  * A group stop is in progress and this is the group leader.
1415                  * We won't report until all threads have stopped.
1416                  */
1417                 goto unlock_sig;
1418
1419         exit_code = p->exit_code;
1420         if (!exit_code)
1421                 goto unlock_sig;
1422
1423         if (!noreap)
1424                 p->exit_code = 0;
1425
1426         uid = p->uid;
1427 unlock_sig:
1428         spin_unlock_irq(&p->sighand->siglock);
1429         if (!exit_code)
1430                 return 0;
1431
1432         /*
1433          * Now we are pretty sure this task is interesting.
1434          * Make sure it doesn't get reaped out from under us while we
1435          * give up the lock and then examine it below.  We don't want to
1436          * keep holding onto the tasklist_lock while we call getrusage and
1437          * possibly take page faults for user memory.
1438          */
1439         get_task_struct(p);
1440         pid = task_pid_vnr(p);
1441         why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1442         read_unlock(&tasklist_lock);
1443
1444         if (unlikely(noreap))
1445                 return wait_noreap_copyout(p, pid, uid,
1446                                            why, exit_code,
1447                                            infop, ru);
1448
1449         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1450         if (!retval && stat_addr)
1451                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1452         if (!retval && infop)
1453                 retval = put_user(SIGCHLD, &infop->si_signo);
1454         if (!retval && infop)
1455                 retval = put_user(0, &infop->si_errno);
1456         if (!retval && infop)
1457                 retval = put_user((short)why, &infop->si_code);
1458         if (!retval && infop)
1459                 retval = put_user(exit_code, &infop->si_status);
1460         if (!retval && infop)
1461                 retval = put_user(pid, &infop->si_pid);
1462         if (!retval && infop)
1463                 retval = put_user(uid, &infop->si_uid);
1464         if (!retval)
1465                 retval = pid;
1466         put_task_struct(p);
1467
1468         BUG_ON(!retval);
1469         return retval;
1470 }
1471
1472 /*
1473  * Handle do_wait work for one task in a live, non-stopped state.
1474  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1475  * the lock and this task is uninteresting.  If we return nonzero, we have
1476  * released the lock and the system call should return.
1477  */
1478 static int wait_task_continued(struct task_struct *p, int noreap,
1479                                struct siginfo __user *infop,
1480                                int __user *stat_addr, struct rusage __user *ru)
1481 {
1482         int retval;
1483         pid_t pid;
1484         uid_t uid;
1485
1486         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1487                 return 0;
1488
1489         spin_lock_irq(&p->sighand->siglock);
1490         /* Re-check with the lock held.  */
1491         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1492                 spin_unlock_irq(&p->sighand->siglock);
1493                 return 0;
1494         }
1495         if (!noreap)
1496                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1497         spin_unlock_irq(&p->sighand->siglock);
1498
1499         pid = task_pid_vnr(p);
1500         uid = p->uid;
1501         get_task_struct(p);
1502         read_unlock(&tasklist_lock);
1503
1504         if (!infop) {
1505                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1506                 put_task_struct(p);
1507                 if (!retval && stat_addr)
1508                         retval = put_user(0xffff, stat_addr);
1509                 if (!retval)
1510                         retval = pid;
1511         } else {
1512                 retval = wait_noreap_copyout(p, pid, uid,
1513                                              CLD_CONTINUED, SIGCONT,
1514                                              infop, ru);
1515                 BUG_ON(retval == 0);
1516         }
1517
1518         return retval;
1519 }
1520
1521 static long do_wait(enum pid_type type, struct pid *pid, int options,
1522                     struct siginfo __user *infop, int __user *stat_addr,
1523                     struct rusage __user *ru)
1524 {
1525         DECLARE_WAITQUEUE(wait, current);
1526         struct task_struct *tsk;
1527         int flag, retval;
1528
1529         add_wait_queue(&current->signal->wait_chldexit,&wait);
1530 repeat:
1531         /* If there is nothing that can match our critier just get out */
1532         retval = -ECHILD;
1533         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1534                 goto end;
1535
1536         /*
1537          * We will set this flag if we see any child that might later
1538          * match our criteria, even if we are not able to reap it yet.
1539          */
1540         flag = retval = 0;
1541         current->state = TASK_INTERRUPTIBLE;
1542         read_lock(&tasklist_lock);
1543         tsk = current;
1544         do {
1545                 struct task_struct *p;
1546
1547                 list_for_each_entry(p, &tsk->children, sibling) {
1548                         int ret = eligible_child(type, pid, options, p);
1549                         if (!ret)
1550                                 continue;
1551
1552                         if (unlikely(ret < 0)) {
1553                                 retval = ret;
1554                         } else if (task_is_stopped_or_traced(p)) {
1555                                 /*
1556                                  * It's stopped now, so it might later
1557                                  * continue, exit, or stop again.
1558                                  */
1559                                 flag = 1;
1560                                 if (!(p->ptrace & PT_PTRACED) &&
1561                                     !(options & WUNTRACED))
1562                                         continue;
1563
1564                                 retval = wait_task_stopped(p,
1565                                                 (options & WNOWAIT), infop,
1566                                                 stat_addr, ru);
1567                         } else if (p->exit_state == EXIT_ZOMBIE &&
1568                                         !delay_group_leader(p)) {
1569                                 /*
1570                                  * We don't reap group leaders with subthreads.
1571                                  */
1572                                 if (!likely(options & WEXITED))
1573                                         continue;
1574                                 retval = wait_task_zombie(p,
1575                                                 (options & WNOWAIT), infop,
1576                                                 stat_addr, ru);
1577                         } else if (p->exit_state != EXIT_DEAD) {
1578                                 /*
1579                                  * It's running now, so it might later
1580                                  * exit, stop, or stop and then continue.
1581                                  */
1582                                 flag = 1;
1583                                 if (!unlikely(options & WCONTINUED))
1584                                         continue;
1585                                 retval = wait_task_continued(p,
1586                                                 (options & WNOWAIT), infop,
1587                                                 stat_addr, ru);
1588                         }
1589                         if (retval != 0) /* tasklist_lock released */
1590                                 goto end;
1591                 }
1592                 if (!flag) {
1593                         list_for_each_entry(p, &tsk->ptrace_children,
1594                                                                 ptrace_list) {
1595                                 flag = eligible_child(type, pid, options, p);
1596                                 if (!flag)
1597                                         continue;
1598                                 if (likely(flag > 0))
1599                                         break;
1600                                 retval = flag;
1601                                 goto end;
1602                         }
1603                 }
1604                 if (options & __WNOTHREAD)
1605                         break;
1606                 tsk = next_thread(tsk);
1607                 BUG_ON(tsk->signal != current->signal);
1608         } while (tsk != current);
1609         read_unlock(&tasklist_lock);
1610
1611         if (flag) {
1612                 if (options & WNOHANG)
1613                         goto end;
1614                 retval = -ERESTARTSYS;
1615                 if (signal_pending(current))
1616                         goto end;
1617                 schedule();
1618                 goto repeat;
1619         }
1620         retval = -ECHILD;
1621 end:
1622         current->state = TASK_RUNNING;
1623         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1624         if (infop) {
1625                 if (retval > 0)
1626                         retval = 0;
1627                 else {
1628                         /*
1629                          * For a WNOHANG return, clear out all the fields
1630                          * we would set so the user can easily tell the
1631                          * difference.
1632                          */
1633                         if (!retval)
1634                                 retval = put_user(0, &infop->si_signo);
1635                         if (!retval)
1636                                 retval = put_user(0, &infop->si_errno);
1637                         if (!retval)
1638                                 retval = put_user(0, &infop->si_code);
1639                         if (!retval)
1640                                 retval = put_user(0, &infop->si_pid);
1641                         if (!retval)
1642                                 retval = put_user(0, &infop->si_uid);
1643                         if (!retval)
1644                                 retval = put_user(0, &infop->si_status);
1645                 }
1646         }
1647         return retval;
1648 }
1649
1650 asmlinkage long sys_waitid(int which, pid_t upid,
1651                            struct siginfo __user *infop, int options,
1652                            struct rusage __user *ru)
1653 {
1654         struct pid *pid = NULL;
1655         enum pid_type type;
1656         long ret;
1657
1658         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1659                 return -EINVAL;
1660         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1661                 return -EINVAL;
1662
1663         switch (which) {
1664         case P_ALL:
1665                 type = PIDTYPE_MAX;
1666                 break;
1667         case P_PID:
1668                 type = PIDTYPE_PID;
1669                 if (upid <= 0)
1670                         return -EINVAL;
1671                 break;
1672         case P_PGID:
1673                 type = PIDTYPE_PGID;
1674                 if (upid <= 0)
1675                         return -EINVAL;
1676                 break;
1677         default:
1678                 return -EINVAL;
1679         }
1680
1681         if (type < PIDTYPE_MAX)
1682                 pid = find_get_pid(upid);
1683         ret = do_wait(type, pid, options, infop, NULL, ru);
1684         put_pid(pid);
1685
1686         /* avoid REGPARM breakage on x86: */
1687         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1688         return ret;
1689 }
1690
1691 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1692                           int options, struct rusage __user *ru)
1693 {
1694         struct pid *pid = NULL;
1695         enum pid_type type;
1696         long ret;
1697
1698         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1699                         __WNOTHREAD|__WCLONE|__WALL))
1700                 return -EINVAL;
1701
1702         if (upid == -1)
1703                 type = PIDTYPE_MAX;
1704         else if (upid < 0) {
1705                 type = PIDTYPE_PGID;
1706                 pid = find_get_pid(-upid);
1707         } else if (upid == 0) {
1708                 type = PIDTYPE_PGID;
1709                 pid = get_pid(task_pgrp(current));
1710         } else /* upid > 0 */ {
1711                 type = PIDTYPE_PID;
1712                 pid = find_get_pid(upid);
1713         }
1714
1715         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1716         put_pid(pid);
1717
1718         /* avoid REGPARM breakage on x86: */
1719         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1720         return ret;
1721 }
1722
1723 #ifdef __ARCH_WANT_SYS_WAITPID
1724
1725 /*
1726  * sys_waitpid() remains for compatibility. waitpid() should be
1727  * implemented by calling sys_wait4() from libc.a.
1728  */
1729 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1730 {
1731         return sys_wait4(pid, stat_addr, options, NULL);
1732 }
1733
1734 #endif