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