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