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