kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/config.h>
   8 #include <linux/mm.h>
   9 #include <linux/slab.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/module.h>
  13 #include <linux/capability.h>
  14 #include <linux/completion.h>
  15 #include <linux/personality.h>
  16 #include <linux/tty.h>
  17 #include <linux/namespace.h>
  18 #include <linux/key.h>
  19 #include <linux/security.h>
  20 #include <linux/cpu.h>
  21 #include <linux/acct.h>
  22 #include <linux/file.h>
  23 #include <linux/binfmts.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/profile.h>
  26 #include <linux/mount.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/mempolicy.h>
  29 #include <linux/cpuset.h>
  30 #include <linux/syscalls.h>
  31 #include <linux/signal.h>
  32 #include <linux/posix-timers.h>
  33 #include <linux/cn_proc.h>
  34 #include <linux/mutex.h>
  35 #include <linux/futex.h>
  36 #include <linux/compat.h>
  37 #include <linux/pipe_fs_i.h>
  38 #include <linux/audit.h> /* for audit_free() */
  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 int getrusage(struct task_struct *, int, struct rusage __user *);
  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         int zap_leader;
 140         task_t *leader;
 141         struct dentry *proc_dentry;
 142
 143 repeat:
 144         atomic_dec(&p->user->processes);
 145         spin_lock(&p->proc_lock);
 146         proc_dentry = proc_pid_unhash(p);
 147         write_lock_irq(&tasklist_lock);
 148         ptrace_unlink(p);
 149         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 150         __exit_signal(p);
 151
 152         /*
 153          * If we are the last non-leader member of the thread
 154          * group, and the leader is zombie, then notify the
 155          * group leader's parent process. (if it wants notification.)
 156          */
 157         zap_leader = 0;
 158         leader = p->group_leader;
 159         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 160                 BUG_ON(leader->exit_signal == -1);
 161                 do_notify_parent(leader, leader->exit_signal);
 162                 /*
 163                  * If we were the last child thread and the leader has
 164                  * exited already, and the leader's parent ignores SIGCHLD,
 165                  * then we are the one who should release the leader.
 166                  *
 167                  * do_notify_parent() will have marked it self-reaping in
 168                  * that case.
 169                  */
 170                 zap_leader = (leader->exit_signal == -1);
 171         }
 172
 173         sched_exit(p);
 174         write_unlock_irq(&tasklist_lock);
 175         spin_unlock(&p->proc_lock);
 176         proc_pid_flush(proc_dentry);
 177         release_thread(p);
 178         call_rcu(&p->rcu, delayed_put_task_struct);
 179
 180         p = leader;
 181         if (unlikely(zap_leader))
 182                 goto repeat;
 183 }
 184
 185 /*
 186  * This checks not only the pgrp, but falls back on the pid if no
 187  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 188  * without this...
 189  */
 190 int session_of_pgrp(int pgrp)
 191 {
 192         struct task_struct *p;
 193         int sid = -1;
 194
 195         read_lock(&tasklist_lock);
 196         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 197                 if (p->signal->session > 0) {
 198                         sid = p->signal->session;
 199                         goto out;
 200                 }
 201         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 202         p = find_task_by_pid(pgrp);
 203         if (p)
 204                 sid = p->signal->session;
 205 out:
 206         read_unlock(&tasklist_lock);
 207
 208         return sid;
 209 }
 210
 211 /*
 212  * Determine if a process group is "orphaned", according to the POSIX
 213  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 214  * by terminal-generated stop signals.  Newly orphaned process groups are
 215  * to receive a SIGHUP and a SIGCONT.
 216  *
 217  * "I ask you, have you ever known what it is to be an orphan?"
 218  */
 219 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 220 {
 221         struct task_struct *p;
 222         int ret = 1;
 223
 224         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 225                 if (p == ignored_task
 226                                 || p->exit_state
 227                                 || p->real_parent->pid == 1)
 228                         continue;
 229                 if (process_group(p->real_parent) != pgrp
 230                             && p->real_parent->signal->session == p->signal->session) {
 231                         ret = 0;
 232                         break;
 233                 }
 234         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 235         return ret;     /* (sighing) "Often!" */
 236 }
 237
 238 int is_orphaned_pgrp(int pgrp)
 239 {
 240         int retval;
 241
 242         read_lock(&tasklist_lock);
 243         retval = will_become_orphaned_pgrp(pgrp, NULL);
 244         read_unlock(&tasklist_lock);
 245
 246         return retval;
 247 }
 248
 249 static int has_stopped_jobs(int pgrp)
 250 {
 251         int retval = 0;
 252         struct task_struct *p;
 253
 254         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 255                 if (p->state != TASK_STOPPED)
 256                         continue;
 257
 258                 /* If p is stopped by a debugger on a signal that won't
 259                    stop it, then don't count p as stopped.  This isn't
 260                    perfect but it's a good approximation.  */
 261                 if (unlikely (p->ptrace)
 262                     && p->exit_code != SIGSTOP
 263                     && p->exit_code != SIGTSTP
 264                     && p->exit_code != SIGTTOU
 265                     && p->exit_code != SIGTTIN)
 266                         continue;
 267
 268                 retval = 1;
 269                 break;
 270         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 271         return retval;
 272 }
 273
 274 /**
 275  * reparent_to_init - Reparent the calling kernel thread to the init task.
 276  *
 277  * If a kernel thread is launched as a result of a system call, or if
 278  * it ever exits, it should generally reparent itself to init so that
 279  * it is correctly cleaned up on exit.
 280  *
 281  * The various task state such as scheduling policy and priority may have
 282  * been inherited from a user process, so we reset them to sane values here.
 283  *
 284  * NOTE that reparent_to_init() gives the caller full capabilities.
 285  */
 286 static void reparent_to_init(void)
 287 {
 288         write_lock_irq(&tasklist_lock);
 289
 290         ptrace_unlink(current);
 291         /* Reparent to init */
 292         remove_parent(current);
 293         current->parent = child_reaper;
 294         current->real_parent = child_reaper;
 295         add_parent(current);
 296
 297         /* Set the exit signal to SIGCHLD so we signal init on exit */
 298         current->exit_signal = SIGCHLD;
 299
 300         if ((current->policy == SCHED_NORMAL ||
 301                         current->policy == SCHED_BATCH)
 302                                 && (task_nice(current) < 0))
 303                 set_user_nice(current, 0);
 304         /* cpus_allowed? */
 305         /* rt_priority? */
 306         /* signals? */
 307         security_task_reparent_to_init(current);
 308         memcpy(current->signal->rlim, init_task.signal->rlim,
 309                sizeof(current->signal->rlim));
 310         atomic_inc(&(INIT_USER->__count));
 311         write_unlock_irq(&tasklist_lock);
 312         switch_uid(INIT_USER);
 313 }
 314
 315 void __set_special_pids(pid_t session, pid_t pgrp)
 316 {
 317         struct task_struct *curr = current->group_leader;
 318
 319         if (curr->signal->session != session) {
 320                 detach_pid(curr, PIDTYPE_SID);
 321                 curr->signal->session = session;
 322                 attach_pid(curr, PIDTYPE_SID, session);
 323         }
 324         if (process_group(curr) != pgrp) {
 325                 detach_pid(curr, PIDTYPE_PGID);
 326                 curr->signal->pgrp = pgrp;
 327                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 328         }
 329 }
 330
 331 void set_special_pids(pid_t session, pid_t pgrp)
 332 {
 333         write_lock_irq(&tasklist_lock);
 334         __set_special_pids(session, pgrp);
 335         write_unlock_irq(&tasklist_lock);
 336 }
 337
 338 /*
 339  * Let kernel threads use this to say that they
 340  * allow a certain signal (since daemonize() will
 341  * have disabled all of them by default).
 342  */
 343 int allow_signal(int sig)
 344 {
 345         if (!valid_signal(sig) || sig < 1)
 346                 return -EINVAL;
 347
 348         spin_lock_irq(&current->sighand->siglock);
 349         sigdelset(&current->blocked, sig);
 350         if (!current->mm) {
 351                 /* Kernel threads handle their own signals.
 352                    Let the signal code know it'll be handled, so
 353                    that they don't get converted to SIGKILL or
 354                    just silently dropped */
 355                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 356         }
 357         recalc_sigpending();
 358         spin_unlock_irq(&current->sighand->siglock);
 359         return 0;
 360 }
 361
 362 EXPORT_SYMBOL(allow_signal);
 363
 364 int disallow_signal(int sig)
 365 {
 366         if (!valid_signal(sig) || sig < 1)
 367                 return -EINVAL;
 368
 369         spin_lock_irq(&current->sighand->siglock);
 370         sigaddset(&current->blocked, sig);
 371         recalc_sigpending();
 372         spin_unlock_irq(&current->sighand->siglock);
 373         return 0;
 374 }
 375
 376 EXPORT_SYMBOL(disallow_signal);
 377
 378 /*
 379  *      Put all the gunge required to become a kernel thread without
 380  *      attached user resources in one place where it belongs.
 381  */
 382
 383 void daemonize(const char *name, ...)
 384 {
 385         va_list args;
 386         struct fs_struct *fs;
 387         sigset_t blocked;
 388
 389         va_start(args, name);
 390         vsnprintf(current->comm, sizeof(current->comm), name, args);
 391         va_end(args);
 392
 393         /*
 394          * If we were started as result of loading a module, close all of the
 395          * user space pages.  We don't need them, and if we didn't close them
 396          * they would be locked into memory.
 397          */
 398         exit_mm(current);
 399
 400         set_special_pids(1, 1);
 401         mutex_lock(&tty_mutex);
 402         current->signal->tty = NULL;
 403         mutex_unlock(&tty_mutex);
 404
 405         /* Block and flush all signals */
 406         sigfillset(&blocked);
 407         sigprocmask(SIG_BLOCK, &blocked, NULL);
 408         flush_signals(current);
 409
 410         /* Become as one with the init task */
 411
 412         exit_fs(current);       /* current->fs->count--; */
 413         fs = init_task.fs;
 414         current->fs = fs;
 415         atomic_inc(&fs->count);
 416         exit_namespace(current);
 417         current->namespace = init_task.namespace;
 418         get_namespace(current->namespace);
 419         exit_files(current);
 420         current->files = init_task.files;
 421         atomic_inc(&current->files->count);
 422
 423         reparent_to_init();
 424 }
 425
 426 EXPORT_SYMBOL(daemonize);
 427
 428 static void close_files(struct files_struct * files)
 429 {
 430         int i, j;
 431         struct fdtable *fdt;
 432
 433         j = 0;
 434
 435         /*
 436          * It is safe to dereference the fd table without RCU or
 437          * ->file_lock because this is the last reference to the
 438          * files structure.
 439          */
 440         fdt = files_fdtable(files);
 441         for (;;) {
 442                 unsigned long set;
 443                 i = j * __NFDBITS;
 444                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 445                         break;
 446                 set = fdt->open_fds->fds_bits[j++];
 447                 while (set) {
 448                         if (set & 1) {
 449                                 struct file * file = xchg(&fdt->fd[i], NULL);
 450                                 if (file)
 451                                         filp_close(file, files);
 452                         }
 453                         i++;
 454                         set >>= 1;
 455                 }
 456         }
 457 }
 458
 459 struct files_struct *get_files_struct(struct task_struct *task)
 460 {
 461         struct files_struct *files;
 462
 463         task_lock(task);
 464         files = task->files;
 465         if (files)
 466                 atomic_inc(&files->count);
 467         task_unlock(task);
 468
 469         return files;
 470 }
 471
 472 void fastcall put_files_struct(struct files_struct *files)
 473 {
 474         struct fdtable *fdt;
 475
 476         if (atomic_dec_and_test(&files->count)) {
 477                 close_files(files);
 478                 /*
 479                  * Free the fd and fdset arrays if we expanded them.
 480                  * If the fdtable was embedded, pass files for freeing
 481                  * at the end of the RCU grace period. Otherwise,
 482                  * you can free files immediately.
 483                  */
 484                 fdt = files_fdtable(files);
 485                 if (fdt == &files->fdtab)
 486                         fdt->free_files = files;
 487                 else
 488                         kmem_cache_free(files_cachep, files);
 489                 free_fdtable(fdt);
 490         }
 491 }
 492
 493 EXPORT_SYMBOL(put_files_struct);
 494
 495 static inline void __exit_files(struct task_struct *tsk)
 496 {
 497         struct files_struct * files = tsk->files;
 498
 499         if (files) {
 500                 task_lock(tsk);
 501                 tsk->files = NULL;
 502                 task_unlock(tsk);
 503                 put_files_struct(files);
 504         }
 505 }
 506
 507 void exit_files(struct task_struct *tsk)
 508 {
 509         __exit_files(tsk);
 510 }
 511
 512 static inline void __put_fs_struct(struct fs_struct *fs)
 513 {
 514         /* No need to hold fs->lock if we are killing it */
 515         if (atomic_dec_and_test(&fs->count)) {
 516                 dput(fs->root);
 517                 mntput(fs->rootmnt);
 518                 dput(fs->pwd);
 519                 mntput(fs->pwdmnt);
 520                 if (fs->altroot) {
 521                         dput(fs->altroot);
 522                         mntput(fs->altrootmnt);
 523                 }
 524                 kmem_cache_free(fs_cachep, fs);
 525         }
 526 }
 527
 528 void put_fs_struct(struct fs_struct *fs)
 529 {
 530         __put_fs_struct(fs);
 531 }
 532
 533 static inline void __exit_fs(struct task_struct *tsk)
 534 {
 535         struct fs_struct * fs = tsk->fs;
 536
 537         if (fs) {
 538                 task_lock(tsk);
 539                 tsk->fs = NULL;
 540                 task_unlock(tsk);
 541                 __put_fs_struct(fs);
 542         }
 543 }
 544
 545 void exit_fs(struct task_struct *tsk)
 546 {
 547         __exit_fs(tsk);
 548 }
 549
 550 EXPORT_SYMBOL_GPL(exit_fs);
 551
 552 /*
 553  * Turn us into a lazy TLB process if we
 554  * aren't already..
 555  */
 556 static void exit_mm(struct task_struct * tsk)
 557 {
 558         struct mm_struct *mm = tsk->mm;
 559
 560         mm_release(tsk, mm);
 561         if (!mm)
 562                 return;
 563         /*
 564          * Serialize with any possible pending coredump.
 565          * We must hold mmap_sem around checking core_waiters
 566          * and clearing tsk->mm.  The core-inducing thread
 567          * will increment core_waiters for each thread in the
 568          * group with ->mm != NULL.
 569          */
 570         down_read(&mm->mmap_sem);
 571         if (mm->core_waiters) {
 572                 up_read(&mm->mmap_sem);
 573                 down_write(&mm->mmap_sem);
 574                 if (!--mm->core_waiters)
 575                         complete(mm->core_startup_done);
 576                 up_write(&mm->mmap_sem);
 577
 578                 wait_for_completion(&mm->core_done);
 579                 down_read(&mm->mmap_sem);
 580         }
 581         atomic_inc(&mm->mm_count);
 582         if (mm != tsk->active_mm) BUG();
 583         /* more a memory barrier than a real lock */
 584         task_lock(tsk);
 585         tsk->mm = NULL;
 586         up_read(&mm->mmap_sem);
 587         enter_lazy_tlb(mm, current);
 588         task_unlock(tsk);
 589         mmput(mm);
 590 }
 591
 592 static inline void choose_new_parent(task_t *p, task_t *reaper)
 593 {
 594         /*
 595          * Make sure we're not reparenting to ourselves and that
 596          * the parent is not a zombie.
 597          */
 598         BUG_ON(p == reaper || reaper->exit_state);
 599         p->real_parent = reaper;
 600 }
 601
 602 static void reparent_thread(task_t *p, task_t *father, int traced)
 603 {
 604         /* We don't want people slaying init.  */
 605         if (p->exit_signal != -1)
 606                 p->exit_signal = SIGCHLD;
 607
 608         if (p->pdeath_signal)
 609                 /* We already hold the tasklist_lock here.  */
 610                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 611
 612         /* Move the child from its dying parent to the new one.  */
 613         if (unlikely(traced)) {
 614                 /* Preserve ptrace links if someone else is tracing this child.  */
 615                 list_del_init(&p->ptrace_list);
 616                 if (p->parent != p->real_parent)
 617                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 618         } else {
 619                 /* If this child is being traced, then we're the one tracing it
 620                  * anyway, so let go of it.
 621                  */
 622                 p->ptrace = 0;
 623                 remove_parent(p);
 624                 p->parent = p->real_parent;
 625                 add_parent(p);
 626
 627                 /* If we'd notified the old parent about this child's death,
 628                  * also notify the new parent.
 629                  */
 630                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 631                     thread_group_empty(p))
 632                         do_notify_parent(p, p->exit_signal);
 633                 else if (p->state == TASK_TRACED) {
 634                         /*
 635                          * If it was at a trace stop, turn it into
 636                          * a normal stop since it's no longer being
 637                          * traced.
 638                          */
 639                         ptrace_untrace(p);
 640                 }
 641         }
 642
 643         /*
 644          * process group orphan check
 645          * Case ii: Our child is in a different pgrp
 646          * than we are, and it was the only connection
 647          * outside, so the child pgrp is now orphaned.
 648          */
 649         if ((process_group(p) != process_group(father)) &&
 650             (p->signal->session == father->signal->session)) {
 651                 int pgrp = process_group(p);
 652
 653                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 654                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 655                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 656                 }
 657         }
 658 }
 659
 660 /*
 661  * When we die, we re-parent all our children.
 662  * Try to give them to another thread in our thread
 663  * group, and if no such member exists, give it to
 664  * the global child reaper process (ie "init")
 665  */
 666 static void forget_original_parent(struct task_struct * father,
 667                                           struct list_head *to_release)
 668 {
 669         struct task_struct *p, *reaper = father;
 670         struct list_head *_p, *_n;
 671
 672         do {
 673                 reaper = next_thread(reaper);
 674                 if (reaper == father) {
 675                         reaper = child_reaper;
 676                         break;
 677                 }
 678         } while (reaper->exit_state);
 679
 680         /*
 681          * There are only two places where our children can be:
 682          *
 683          * - in our child list
 684          * - in our ptraced child list
 685          *
 686          * Search them and reparent children.
 687          */
 688         list_for_each_safe(_p, _n, &father->children) {
 689                 int ptrace;
 690                 p = list_entry(_p,struct task_struct,sibling);
 691
 692                 ptrace = p->ptrace;
 693
 694                 /* if father isn't the real parent, then ptrace must be enabled */
 695                 BUG_ON(father != p->real_parent && !ptrace);
 696
 697                 if (father == p->real_parent) {
 698                         /* reparent with a reaper, real father it's us */
 699                         choose_new_parent(p, reaper);
 700                         reparent_thread(p, father, 0);
 701                 } else {
 702                         /* reparent ptraced task to its real parent */
 703                         __ptrace_unlink (p);
 704                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 705                             thread_group_empty(p))
 706                                 do_notify_parent(p, p->exit_signal);
 707                 }
 708
 709                 /*
 710                  * if the ptraced child is a zombie with exit_signal == -1
 711                  * we must collect it before we exit, or it will remain
 712                  * zombie forever since we prevented it from self-reap itself
 713                  * while it was being traced by us, to be able to see it in wait4.
 714                  */
 715                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 716                         list_add(&p->ptrace_list, to_release);
 717         }
 718         list_for_each_safe(_p, _n, &father->ptrace_children) {
 719                 p = list_entry(_p,struct task_struct,ptrace_list);
 720                 choose_new_parent(p, reaper);
 721                 reparent_thread(p, father, 1);
 722         }
 723 }
 724
 725 /*
 726  * Send signals to all our closest relatives so that they know
 727  * to properly mourn us..
 728  */
 729 static void exit_notify(struct task_struct *tsk)
 730 {
 731         int state;
 732         struct task_struct *t;
 733         struct list_head ptrace_dead, *_p, *_n;
 734
 735         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 736             && !thread_group_empty(tsk)) {
 737                 /*
 738                  * This occurs when there was a race between our exit
 739                  * syscall and a group signal choosing us as the one to
 740                  * wake up.  It could be that we are the only thread
 741                  * alerted to check for pending signals, but another thread
 742                  * should be woken now to take the signal since we will not.
 743                  * Now we'll wake all the threads in the group just to make
 744                  * sure someone gets all the pending signals.
 745                  */
 746                 read_lock(&tasklist_lock);
 747                 spin_lock_irq(&tsk->sighand->siglock);
 748                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 749                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 750                                 recalc_sigpending_tsk(t);
 751                                 if (signal_pending(t))
 752                                         signal_wake_up(t, 0);
 753                         }
 754                 spin_unlock_irq(&tsk->sighand->siglock);
 755                 read_unlock(&tasklist_lock);
 756         }
 757
 758         write_lock_irq(&tasklist_lock);
 759
 760         /*
 761          * This does two things:
 762          *
 763          * A.  Make init inherit all the child processes
 764          * B.  Check to see if any process groups have become orphaned
 765          *      as a result of our exiting, and if they have any stopped
 766          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 767          */
 768
 769         INIT_LIST_HEAD(&ptrace_dead);
 770         forget_original_parent(tsk, &ptrace_dead);
 771         BUG_ON(!list_empty(&tsk->children));
 772         BUG_ON(!list_empty(&tsk->ptrace_children));
 773
 774         /*
 775          * Check to see if any process groups have become orphaned
 776          * as a result of our exiting, and if they have any stopped
 777          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 778          *
 779          * Case i: Our father is in a different pgrp than we are
 780          * and we were the only connection outside, so our pgrp
 781          * is about to become orphaned.
 782          */
 783
 784         t = tsk->real_parent;
 785
 786         if ((process_group(t) != process_group(tsk)) &&
 787             (t->signal->session == tsk->signal->session) &&
 788             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 789             has_stopped_jobs(process_group(tsk))) {
 790                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 791                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 792         }
 793
 794         /* Let father know we died
 795          *
 796          * Thread signals are configurable, but you aren't going to use
 797          * that to send signals to arbitary processes.
 798          * That stops right now.
 799          *
 800          * If the parent exec id doesn't match the exec id we saved
 801          * when we started then we know the parent has changed security
 802          * domain.
 803          *
 804          * If our self_exec id doesn't match our parent_exec_id then
 805          * we have changed execution domain as these two values started
 806          * the same after a fork.
 807          *
 808          */
 809
 810         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 811             ( tsk->parent_exec_id != t->self_exec_id  ||
 812               tsk->self_exec_id != tsk->parent_exec_id)
 813             && !capable(CAP_KILL))
 814                 tsk->exit_signal = SIGCHLD;
 815
 816
 817         /* If something other than our normal parent is ptracing us, then
 818          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 819          * only has special meaning to our real parent.
 820          */
 821         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 822                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 823                 do_notify_parent(tsk, signal);
 824         } else if (tsk->ptrace) {
 825                 do_notify_parent(tsk, SIGCHLD);
 826         }
 827
 828         state = EXIT_ZOMBIE;
 829         if (tsk->exit_signal == -1 &&
 830             (likely(tsk->ptrace == 0) ||
 831              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 832                 state = EXIT_DEAD;
 833         tsk->exit_state = state;
 834
 835         write_unlock_irq(&tasklist_lock);
 836
 837         list_for_each_safe(_p, _n, &ptrace_dead) {
 838                 list_del_init(_p);
 839                 t = list_entry(_p,struct task_struct,ptrace_list);
 840                 release_task(t);
 841         }
 842
 843         /* If the process is dead, release it - nobody will wait for it */
 844         if (state == EXIT_DEAD)
 845                 release_task(tsk);
 846 }
 847
 848 fastcall NORET_TYPE void do_exit(long code)
 849 {
 850         struct task_struct *tsk = current;
 851         int group_dead;
 852
 853         profile_task_exit(tsk);
 854
 855         WARN_ON(atomic_read(&tsk->fs_excl));
 856
 857         if (unlikely(in_interrupt()))
 858                 panic("Aiee, killing interrupt handler!");
 859         if (unlikely(!tsk->pid))
 860                 panic("Attempted to kill the idle task!");
 861         if (unlikely(tsk == child_reaper))
 862                 panic("Attempted to kill init!");
 863
 864         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 865                 current->ptrace_message = code;
 866                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 867         }
 868
 869         /*
 870          * We're taking recursive faults here in do_exit. Safest is to just
 871          * leave this task alone and wait for reboot.
 872          */
 873         if (unlikely(tsk->flags & PF_EXITING)) {
 874                 printk(KERN_ALERT
 875                         "Fixing recursive fault but reboot is needed!\n");
 876                 if (tsk->io_context)
 877                         exit_io_context();
 878                 set_current_state(TASK_UNINTERRUPTIBLE);
 879                 schedule();
 880         }
 881
 882         tsk->flags |= PF_EXITING;
 883
 884         if (unlikely(in_atomic()))
 885                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 886                                 current->comm, current->pid,
 887                                 preempt_count());
 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                 acct_process(code);
 899         }
 900         if (unlikely(tsk->robust_list))
 901                 exit_robust_list(tsk);
 902 #ifdef 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         exit_mm(tsk);
 909
 910         exit_sem(tsk);
 911         __exit_files(tsk);
 912         __exit_fs(tsk);
 913         exit_namespace(tsk);
 914         exit_thread();
 915         cpuset_exit(tsk);
 916         exit_keys(tsk);
 917
 918         if (group_dead && tsk->signal->leader)
 919                 disassociate_ctty(1);
 920
 921         module_put(task_thread_info(tsk)->exec_domain->module);
 922         if (tsk->binfmt)
 923                 module_put(tsk->binfmt->module);
 924
 925         tsk->exit_code = code;
 926         proc_exit_connector(tsk);
 927         exit_notify(tsk);
 928 #ifdef CONFIG_NUMA
 929         mpol_free(tsk->mempolicy);
 930         tsk->mempolicy = NULL;
 931 #endif
 932         /*
 933          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
 934          */
 935         mutex_debug_check_no_locks_held(tsk);
 936
 937         if (tsk->io_context)
 938                 exit_io_context();
 939
 940         if (tsk->splice_pipe)
 941                 __free_pipe_info(tsk->splice_pipe);
 942
 943         /* PF_DEAD causes final put_task_struct after we schedule. */
 944         preempt_disable();
 945         BUG_ON(tsk->flags & PF_DEAD);
 946         tsk->flags |= PF_DEAD;
 947
 948         schedule();
 949         BUG();
 950         /* Avoid "noreturn function does return".  */
 951         for (;;) ;
 952 }
 953
 954 EXPORT_SYMBOL_GPL(do_exit);
 955
 956 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 957 {
 958         if (comp)
 959                 complete(comp);
 960
 961         do_exit(code);
 962 }
 963
 964 EXPORT_SYMBOL(complete_and_exit);
 965
 966 asmlinkage long sys_exit(int error_code)
 967 {
 968         do_exit((error_code&0xff)<<8);
 969 }
 970
 971 /*
 972  * Take down every thread in the group.  This is called by fatal signals
 973  * as well as by sys_exit_group (below).
 974  */
 975 NORET_TYPE void
 976 do_group_exit(int exit_code)
 977 {
 978         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 979
 980         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 981                 exit_code = current->signal->group_exit_code;
 982         else if (!thread_group_empty(current)) {
 983                 struct signal_struct *const sig = current->signal;
 984                 struct sighand_struct *const sighand = current->sighand;
 985                 spin_lock_irq(&sighand->siglock);
 986                 if (sig->flags & SIGNAL_GROUP_EXIT)
 987                         /* Another thread got here before we took the lock.  */
 988                         exit_code = sig->group_exit_code;
 989                 else {
 990                         sig->group_exit_code = exit_code;
 991                         zap_other_threads(current);
 992                 }
 993                 spin_unlock_irq(&sighand->siglock);
 994         }
 995
 996         do_exit(exit_code);
 997         /* NOTREACHED */
 998 }
 999
1000 /*
1001  * this kills every thread in the thread group. Note that any externally
1002  * wait4()-ing process will get the correct exit code - even if this
1003  * thread is not the thread group leader.
1004  */
1005 asmlinkage void sys_exit_group(int error_code)
1006 {
1007         do_group_exit((error_code & 0xff) << 8);
1008 }
1009
1010 static int eligible_child(pid_t pid, int options, task_t *p)
1011 {
1012         if (pid > 0) {
1013                 if (p->pid != pid)
1014                         return 0;
1015         } else if (!pid) {
1016                 if (process_group(p) != process_group(current))
1017                         return 0;
1018         } else if (pid != -1) {
1019                 if (process_group(p) != -pid)
1020                         return 0;
1021         }
1022
1023         /*
1024          * Do not consider detached threads that are
1025          * not ptraced:
1026          */
1027         if (p->exit_signal == -1 && !p->ptrace)
1028                 return 0;
1029
1030         /* Wait for all children (clone and not) if __WALL is set;
1031          * otherwise, wait for clone children *only* if __WCLONE is
1032          * set; otherwise, wait for non-clone children *only*.  (Note:
1033          * A "clone" child here is one that reports to its parent
1034          * using a signal other than SIGCHLD.) */
1035         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1036             && !(options & __WALL))
1037                 return 0;
1038         /*
1039          * Do not consider thread group leaders that are
1040          * in a non-empty thread group:
1041          */
1042         if (current->tgid != p->tgid && delay_group_leader(p))
1043                 return 2;
1044
1045         if (security_task_wait(p))
1046                 return 0;
1047
1048         return 1;
1049 }
1050
1051 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
1052                                int why, int status,
1053                                struct siginfo __user *infop,
1054                                struct rusage __user *rusagep)
1055 {
1056         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1057         put_task_struct(p);
1058         if (!retval)
1059                 retval = put_user(SIGCHLD, &infop->si_signo);
1060         if (!retval)
1061                 retval = put_user(0, &infop->si_errno);
1062         if (!retval)
1063                 retval = put_user((short)why, &infop->si_code);
1064         if (!retval)
1065                 retval = put_user(pid, &infop->si_pid);
1066         if (!retval)
1067                 retval = put_user(uid, &infop->si_uid);
1068         if (!retval)
1069                 retval = put_user(status, &infop->si_status);
1070         if (!retval)
1071                 retval = pid;
1072         return retval;
1073 }
1074
1075 /*
1076  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1077  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1078  * the lock and this task is uninteresting.  If we return nonzero, we have
1079  * released the lock and the system call should return.
1080  */
1081 static int wait_task_zombie(task_t *p, int noreap,
1082                             struct siginfo __user *infop,
1083                             int __user *stat_addr, struct rusage __user *ru)
1084 {
1085         unsigned long state;
1086         int retval;
1087         int status;
1088
1089         if (unlikely(noreap)) {
1090                 pid_t pid = p->pid;
1091                 uid_t uid = p->uid;
1092                 int exit_code = p->exit_code;
1093                 int why, status;
1094
1095                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1096                         return 0;
1097                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1098                         return 0;
1099                 get_task_struct(p);
1100                 read_unlock(&tasklist_lock);
1101                 if ((exit_code & 0x7f) == 0) {
1102                         why = CLD_EXITED;
1103                         status = exit_code >> 8;
1104                 } else {
1105                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1106                         status = exit_code & 0x7f;
1107                 }
1108                 return wait_noreap_copyout(p, pid, uid, why,
1109                                            status, infop, ru);
1110         }
1111
1112         /*
1113          * Try to move the task's state to DEAD
1114          * only one thread is allowed to do this:
1115          */
1116         state = xchg(&p->exit_state, EXIT_DEAD);
1117         if (state != EXIT_ZOMBIE) {
1118                 BUG_ON(state != EXIT_DEAD);
1119                 return 0;
1120         }
1121         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1122                 /*
1123                  * This can only happen in a race with a ptraced thread
1124                  * dying on another processor.
1125                  */
1126                 return 0;
1127         }
1128
1129         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1130                 struct signal_struct *psig;
1131                 struct signal_struct *sig;
1132
1133                 /*
1134                  * The resource counters for the group leader are in its
1135                  * own task_struct.  Those for dead threads in the group
1136                  * are in its signal_struct, as are those for the child
1137                  * processes it has previously reaped.  All these
1138                  * accumulate in the parent's signal_struct c* fields.
1139                  *
1140                  * We don't bother to take a lock here to protect these
1141                  * p->signal fields, because they are only touched by
1142                  * __exit_signal, which runs with tasklist_lock
1143                  * write-locked anyway, and so is excluded here.  We do
1144                  * need to protect the access to p->parent->signal fields,
1145                  * as other threads in the parent group can be right
1146                  * here reaping other children at the same time.
1147                  */
1148                 spin_lock_irq(&p->parent->sighand->siglock);
1149                 psig = p->parent->signal;
1150                 sig = p->signal;
1151                 psig->cutime =
1152                         cputime_add(psig->cutime,
1153                         cputime_add(p->utime,
1154                         cputime_add(sig->utime,
1155                                     sig->cutime)));
1156                 psig->cstime =
1157                         cputime_add(psig->cstime,
1158                         cputime_add(p->stime,
1159                         cputime_add(sig->stime,
1160                                     sig->cstime)));
1161                 psig->cmin_flt +=
1162                         p->min_flt + sig->min_flt + sig->cmin_flt;
1163                 psig->cmaj_flt +=
1164                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1165                 psig->cnvcsw +=
1166                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1167                 psig->cnivcsw +=
1168                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1169                 spin_unlock_irq(&p->parent->sighand->siglock);
1170         }
1171
1172         /*
1173          * Now we are sure this task is interesting, and no other
1174          * thread can reap it because we set its state to EXIT_DEAD.
1175          */
1176         read_unlock(&tasklist_lock);
1177
1178         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1179         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1180                 ? p->signal->group_exit_code : p->exit_code;
1181         if (!retval && stat_addr)
1182                 retval = put_user(status, stat_addr);
1183         if (!retval && infop)
1184                 retval = put_user(SIGCHLD, &infop->si_signo);
1185         if (!retval && infop)
1186                 retval = put_user(0, &infop->si_errno);
1187         if (!retval && infop) {
1188                 int why;
1189
1190                 if ((status & 0x7f) == 0) {
1191                         why = CLD_EXITED;
1192                         status >>= 8;
1193                 } else {
1194                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1195                         status &= 0x7f;
1196                 }
1197                 retval = put_user((short)why, &infop->si_code);
1198                 if (!retval)
1199                         retval = put_user(status, &infop->si_status);
1200         }
1201         if (!retval && infop)
1202                 retval = put_user(p->pid, &infop->si_pid);
1203         if (!retval && infop)
1204                 retval = put_user(p->uid, &infop->si_uid);
1205         if (retval) {
1206                 // TODO: is this safe?
1207                 p->exit_state = EXIT_ZOMBIE;
1208                 return retval;
1209         }
1210         retval = p->pid;
1211         if (p->real_parent != p->parent) {
1212                 write_lock_irq(&tasklist_lock);
1213                 /* Double-check with lock held.  */
1214                 if (p->real_parent != p->parent) {
1215                         __ptrace_unlink(p);
1216                         // TODO: is this safe?
1217                         p->exit_state = EXIT_ZOMBIE;
1218                         /*
1219                          * If this is not a detached task, notify the parent.
1220                          * If it's still not detached after that, don't release
1221                          * it now.
1222                          */
1223                         if (p->exit_signal != -1) {
1224                                 do_notify_parent(p, p->exit_signal);
1225                                 if (p->exit_signal != -1)
1226                                         p = NULL;
1227                         }
1228                 }
1229                 write_unlock_irq(&tasklist_lock);
1230         }
1231         if (p != NULL)
1232                 release_task(p);
1233         BUG_ON(!retval);
1234         return retval;
1235 }
1236
1237 /*
1238  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1239  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1240  * the lock and this task is uninteresting.  If we return nonzero, we have
1241  * released the lock and the system call should return.
1242  */
1243 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1244                              struct siginfo __user *infop,
1245                              int __user *stat_addr, struct rusage __user *ru)
1246 {
1247         int retval, exit_code;
1248
1249         if (!p->exit_code)
1250                 return 0;
1251         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1252             p->signal && p->signal->group_stop_count > 0)
1253                 /*
1254                  * A group stop is in progress and this is the group leader.
1255                  * We won't report until all threads have stopped.
1256                  */
1257                 return 0;
1258
1259         /*
1260          * Now we are pretty sure this task is interesting.
1261          * Make sure it doesn't get reaped out from under us while we
1262          * give up the lock and then examine it below.  We don't want to
1263          * keep holding onto the tasklist_lock while we call getrusage and
1264          * possibly take page faults for user memory.
1265          */
1266         get_task_struct(p);
1267         read_unlock(&tasklist_lock);
1268
1269         if (unlikely(noreap)) {
1270                 pid_t pid = p->pid;
1271                 uid_t uid = p->uid;
1272                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1273
1274                 exit_code = p->exit_code;
1275                 if (unlikely(!exit_code) ||
1276                     unlikely(p->state & TASK_TRACED))
1277                         goto bail_ref;
1278                 return wait_noreap_copyout(p, pid, uid,
1279                                            why, (exit_code << 8) | 0x7f,
1280                                            infop, ru);
1281         }
1282
1283         write_lock_irq(&tasklist_lock);
1284
1285         /*
1286          * This uses xchg to be atomic with the thread resuming and setting
1287          * it.  It must also be done with the write lock held to prevent a
1288          * race with the EXIT_ZOMBIE case.
1289          */
1290         exit_code = xchg(&p->exit_code, 0);
1291         if (unlikely(p->exit_state)) {
1292                 /*
1293                  * The task resumed and then died.  Let the next iteration
1294                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1295                  * already be zero here if it resumed and did _exit(0).
1296                  * The task itself is dead and won't touch exit_code again;
1297                  * other processors in this function are locked out.
1298                  */
1299                 p->exit_code = exit_code;
1300                 exit_code = 0;
1301         }
1302         if (unlikely(exit_code == 0)) {
1303                 /*
1304                  * Another thread in this function got to it first, or it
1305                  * resumed, or it resumed and then died.
1306                  */
1307                 write_unlock_irq(&tasklist_lock);
1308 bail_ref:
1309                 put_task_struct(p);
1310                 /*
1311                  * We are returning to the wait loop without having successfully
1312                  * removed the process and having released the lock. We cannot
1313                  * continue, since the "p" task pointer is potentially stale.
1314                  *
1315                  * Return -EAGAIN, and do_wait() will restart the loop from the
1316                  * beginning. Do _not_ re-acquire the lock.
1317                  */
1318                 return -EAGAIN;
1319         }
1320
1321         /* move to end of parent's list to avoid starvation */
1322         remove_parent(p);
1323         add_parent(p);
1324
1325         write_unlock_irq(&tasklist_lock);
1326
1327         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1328         if (!retval && stat_addr)
1329                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1330         if (!retval && infop)
1331                 retval = put_user(SIGCHLD, &infop->si_signo);
1332         if (!retval && infop)
1333                 retval = put_user(0, &infop->si_errno);
1334         if (!retval && infop)
1335                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1336                                           ? CLD_TRAPPED : CLD_STOPPED),
1337                                   &infop->si_code);
1338         if (!retval && infop)
1339                 retval = put_user(exit_code, &infop->si_status);
1340         if (!retval && infop)
1341                 retval = put_user(p->pid, &infop->si_pid);
1342         if (!retval && infop)
1343                 retval = put_user(p->uid, &infop->si_uid);
1344         if (!retval)
1345                 retval = p->pid;
1346         put_task_struct(p);
1347
1348         BUG_ON(!retval);
1349         return retval;
1350 }
1351
1352 /*
1353  * Handle do_wait work for one task in a live, non-stopped state.
1354  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1355  * the lock and this task is uninteresting.  If we return nonzero, we have
1356  * released the lock and the system call should return.
1357  */
1358 static int wait_task_continued(task_t *p, int noreap,
1359                                struct siginfo __user *infop,
1360                                int __user *stat_addr, struct rusage __user *ru)
1361 {
1362         int retval;
1363         pid_t pid;
1364         uid_t uid;
1365
1366         if (unlikely(!p->signal))
1367                 return 0;
1368
1369         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1370                 return 0;
1371
1372         spin_lock_irq(&p->sighand->siglock);
1373         /* Re-check with the lock held.  */
1374         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1375                 spin_unlock_irq(&p->sighand->siglock);
1376                 return 0;
1377         }
1378         if (!noreap)
1379                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1380         spin_unlock_irq(&p->sighand->siglock);
1381
1382         pid = p->pid;
1383         uid = p->uid;
1384         get_task_struct(p);
1385         read_unlock(&tasklist_lock);
1386
1387         if (!infop) {
1388                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1389                 put_task_struct(p);
1390                 if (!retval && stat_addr)
1391                         retval = put_user(0xffff, stat_addr);
1392                 if (!retval)
1393                         retval = p->pid;
1394         } else {
1395                 retval = wait_noreap_copyout(p, pid, uid,
1396                                              CLD_CONTINUED, SIGCONT,
1397                                              infop, ru);
1398                 BUG_ON(retval == 0);
1399         }
1400
1401         return retval;
1402 }
1403
1404
1405 static inline int my_ptrace_child(struct task_struct *p)
1406 {
1407         if (!(p->ptrace & PT_PTRACED))
1408                 return 0;
1409         if (!(p->ptrace & PT_ATTACHED))
1410                 return 1;
1411         /*
1412          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1413          * we are the attacher.  If we are the real parent, this is a race
1414          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1415          * which we have to switch the parent links, but has already set
1416          * the flags in p->ptrace.
1417          */
1418         return (p->parent != p->real_parent);
1419 }
1420
1421 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1422                     int __user *stat_addr, struct rusage __user *ru)
1423 {
1424         DECLARE_WAITQUEUE(wait, current);
1425         struct task_struct *tsk;
1426         int flag, retval;
1427
1428         add_wait_queue(&current->signal->wait_chldexit,&wait);
1429 repeat:
1430         /*
1431          * We will set this flag if we see any child that might later
1432          * match our criteria, even if we are not able to reap it yet.
1433          */
1434         flag = 0;
1435         current->state = TASK_INTERRUPTIBLE;
1436         read_lock(&tasklist_lock);
1437         tsk = current;
1438         do {
1439                 struct task_struct *p;
1440                 struct list_head *_p;
1441                 int ret;
1442
1443                 list_for_each(_p,&tsk->children) {
1444                         p = list_entry(_p,struct task_struct,sibling);
1445
1446                         ret = eligible_child(pid, options, p);
1447                         if (!ret)
1448                                 continue;
1449
1450                         switch (p->state) {
1451                         case TASK_TRACED:
1452                                 /*
1453                                  * When we hit the race with PTRACE_ATTACH,
1454                                  * we will not report this child.  But the
1455                                  * race means it has not yet been moved to
1456                                  * our ptrace_children list, so we need to
1457                                  * set the flag here to avoid a spurious ECHILD
1458                                  * when the race happens with the only child.
1459                                  */
1460                                 flag = 1;
1461                                 if (!my_ptrace_child(p))
1462                                         continue;
1463                                 /*FALLTHROUGH*/
1464                         case TASK_STOPPED:
1465                                 /*
1466                                  * It's stopped now, so it might later
1467                                  * continue, exit, or stop again.
1468                                  */
1469                                 flag = 1;
1470                                 if (!(options & WUNTRACED) &&
1471                                     !my_ptrace_child(p))
1472                                         continue;
1473                                 retval = wait_task_stopped(p, ret == 2,
1474                                                            (options & WNOWAIT),
1475                                                            infop,
1476                                                            stat_addr, ru);
1477                                 if (retval == -EAGAIN)
1478                                         goto repeat;
1479                                 if (retval != 0) /* He released the lock.  */
1480                                         goto end;
1481                                 break;
1482                         default:
1483                         // case EXIT_DEAD:
1484                                 if (p->exit_state == EXIT_DEAD)
1485                                         continue;
1486                         // case EXIT_ZOMBIE:
1487                                 if (p->exit_state == EXIT_ZOMBIE) {
1488                                         /*
1489                                          * Eligible but we cannot release
1490                                          * it yet:
1491                                          */
1492                                         if (ret == 2)
1493                                                 goto check_continued;
1494                                         if (!likely(options & WEXITED))
1495                                                 continue;
1496                                         retval = wait_task_zombie(
1497                                                 p, (options & WNOWAIT),
1498                                                 infop, stat_addr, ru);
1499                                         /* He released the lock.  */
1500                                         if (retval != 0)
1501                                                 goto end;
1502                                         break;
1503                                 }
1504 check_continued:
1505                                 /*
1506                                  * It's running now, so it might later
1507                                  * exit, stop, or stop and then continue.
1508                                  */
1509                                 flag = 1;
1510                                 if (!unlikely(options & WCONTINUED))
1511                                         continue;
1512                                 retval = wait_task_continued(
1513                                         p, (options & WNOWAIT),
1514                                         infop, stat_addr, ru);
1515                                 if (retval != 0) /* He released the lock.  */
1516                                         goto end;
1517                                 break;
1518                         }
1519                 }
1520                 if (!flag) {
1521                         list_for_each(_p, &tsk->ptrace_children) {
1522                                 p = list_entry(_p, struct task_struct,
1523                                                 ptrace_list);
1524                                 if (!eligible_child(pid, options, p))
1525                                         continue;
1526                                 flag = 1;
1527                                 break;
1528                         }
1529                 }
1530                 if (options & __WNOTHREAD)
1531                         break;
1532                 tsk = next_thread(tsk);
1533                 if (tsk->signal != current->signal)
1534                         BUG();
1535         } while (tsk != current);
1536
1537         read_unlock(&tasklist_lock);
1538         if (flag) {
1539                 retval = 0;
1540                 if (options & WNOHANG)
1541                         goto end;
1542                 retval = -ERESTARTSYS;
1543                 if (signal_pending(current))
1544                         goto end;
1545                 schedule();
1546                 goto repeat;
1547         }
1548         retval = -ECHILD;
1549 end:
1550         current->state = TASK_RUNNING;
1551         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1552         if (infop) {
1553                 if (retval > 0)
1554                 retval = 0;
1555                 else {
1556                         /*
1557                          * For a WNOHANG return, clear out all the fields
1558                          * we would set so the user can easily tell the
1559                          * difference.
1560                          */
1561                         if (!retval)
1562                                 retval = put_user(0, &infop->si_signo);
1563                         if (!retval)
1564                                 retval = put_user(0, &infop->si_errno);
1565                         if (!retval)
1566                                 retval = put_user(0, &infop->si_code);
1567                         if (!retval)
1568                                 retval = put_user(0, &infop->si_pid);
1569                         if (!retval)
1570                                 retval = put_user(0, &infop->si_uid);
1571                         if (!retval)
1572                                 retval = put_user(0, &infop->si_status);
1573                 }
1574         }
1575         return retval;
1576 }
1577
1578 asmlinkage long sys_waitid(int which, pid_t pid,
1579                            struct siginfo __user *infop, int options,
1580                            struct rusage __user *ru)
1581 {
1582         long ret;
1583
1584         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1585                 return -EINVAL;
1586         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1587                 return -EINVAL;
1588
1589         switch (which) {
1590         case P_ALL:
1591                 pid = -1;
1592                 break;
1593         case P_PID:
1594                 if (pid <= 0)
1595                         return -EINVAL;
1596                 break;
1597         case P_PGID:
1598                 if (pid <= 0)
1599                         return -EINVAL;
1600                 pid = -pid;
1601                 break;
1602         default:
1603                 return -EINVAL;
1604         }
1605
1606         ret = do_wait(pid, options, infop, NULL, ru);
1607
1608         /* avoid REGPARM breakage on x86: */
1609         prevent_tail_call(ret);
1610         return ret;
1611 }
1612
1613 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1614                           int options, struct rusage __user *ru)
1615 {
1616         long ret;
1617
1618         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1619                         __WNOTHREAD|__WCLONE|__WALL))
1620                 return -EINVAL;
1621         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1622
1623         /* avoid REGPARM breakage on x86: */
1624         prevent_tail_call(ret);
1625         return ret;
1626 }
1627
1628 #ifdef __ARCH_WANT_SYS_WAITPID
1629
1630 /*
1631  * sys_waitpid() remains for compatibility. waitpid() should be
1632  * implemented by calling sys_wait4() from libc.a.
1633  */
1634 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1635 {
1636         return sys_wait4(pid, stat_addr, options, NULL);
1637 }
1638
1639 #endif