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/vs_limit.h>
  40 #include <linux/vs_context.h>
  41 #include <linux/vs_network.h>
  42 #include <linux/vs_cvirt.h>
  43
  44 #include <asm/uaccess.h>
  45 #include <asm/unistd.h>
  46 #include <asm/pgtable.h>
  47 #include <asm/mmu_context.h>
  48
  49 extern void sem_exit (void);
  50 extern struct task_struct *child_reaper;
  51
  52 int getrusage(struct task_struct *, int, struct rusage __user *);
  53
  54 static void exit_mm(struct task_struct * tsk);
  55
  56 static void __unhash_process(struct task_struct *p)
  57 {
  58         nr_threads--;
  59         detach_pid(p, PIDTYPE_PID);
  60         if (thread_group_leader(p)) {
  61                 detach_pid(p, PIDTYPE_PGID);
  62                 detach_pid(p, PIDTYPE_SID);
  63
  64                 list_del_rcu(&p->tasks);
  65                 __get_cpu_var(process_counts)--;
  66         }
  67         list_del_rcu(&p->thread_group);
  68         remove_parent(p);
  69 }
  70
  71 /*
  72  * This function expects the tasklist_lock write-locked.
  73  */
  74 static void __exit_signal(struct task_struct *tsk)
  75 {
  76         struct signal_struct *sig = tsk->signal;
  77         struct sighand_struct *sighand;
  78
  79         BUG_ON(!sig);
  80         BUG_ON(!atomic_read(&sig->count));
  81
  82         rcu_read_lock();
  83         sighand = rcu_dereference(tsk->sighand);
  84         spin_lock(&sighand->siglock);
  85
  86         posix_cpu_timers_exit(tsk);
  87         if (atomic_dec_and_test(&sig->count))
  88                 posix_cpu_timers_exit_group(tsk);
  89         else {
  90                 /*
  91                  * If there is any task waiting for the group exit
  92                  * then notify it:
  93                  */
  94                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
  95                         wake_up_process(sig->group_exit_task);
  96                         sig->group_exit_task = NULL;
  97                 }
  98                 if (tsk == sig->curr_target)
  99                         sig->curr_target = next_thread(tsk);
 100                 /*
 101                  * Accumulate here the counters for all threads but the
 102                  * group leader as they die, so they can be added into
 103                  * the process-wide totals when those are taken.
 104                  * The group leader stays around as a zombie as long
 105                  * as there are other threads.  When it gets reaped,
 106                  * the exit.c code will add its counts into these totals.
 107                  * We won't ever get here for the group leader, since it
 108                  * will have been the last reference on the signal_struct.
 109                  */
 110                 sig->utime = cputime_add(sig->utime, tsk->utime);
 111                 sig->stime = cputime_add(sig->stime, tsk->stime);
 112                 sig->min_flt += tsk->min_flt;
 113                 sig->maj_flt += tsk->maj_flt;
 114                 sig->nvcsw += tsk->nvcsw;
 115                 sig->nivcsw += tsk->nivcsw;
 116                 sig->sched_time += tsk->sched_time;
 117                 sig = NULL; /* Marker for below. */
 118         }
 119
 120         __unhash_process(tsk);
 121
 122         tsk->signal = NULL;
 123         tsk->sighand = NULL;
 124         spin_unlock(&sighand->siglock);
 125         rcu_read_unlock();
 126
 127         __cleanup_sighand(sighand);
 128         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 129         flush_sigqueue(&tsk->pending);
 130         if (sig) {
 131                 flush_sigqueue(&sig->shared_pending);
 132                 __cleanup_signal(sig);
 133         }
 134 }
 135
 136 static void delayed_put_task_struct(struct rcu_head *rhp)
 137 {
 138         put_task_struct(container_of(rhp, struct task_struct, rcu));
 139 }
 140
 141 void release_task(struct task_struct * p)
 142 {
 143         int zap_leader;
 144         task_t *leader;
 145         struct dentry *proc_dentry;
 146
 147 repeat:
 148         atomic_dec(&p->user->processes);
 149         spin_lock(&p->proc_lock);
 150         proc_dentry = proc_pid_unhash(p);
 151         write_lock_irq(&tasklist_lock);
 152         ptrace_unlink(p);
 153         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 154         __exit_signal(p);
 155
 156         /*
 157          * If we are the last non-leader member of the thread
 158          * group, and the leader is zombie, then notify the
 159          * group leader's parent process. (if it wants notification.)
 160          */
 161         zap_leader = 0;
 162         leader = p->group_leader;
 163         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 164                 BUG_ON(leader->exit_signal == -1);
 165                 do_notify_parent(leader, leader->exit_signal);
 166                 /*
 167                  * If we were the last child thread and the leader has
 168                  * exited already, and the leader's parent ignores SIGCHLD,
 169                  * then we are the one who should release the leader.
 170                  *
 171                  * do_notify_parent() will have marked it self-reaping in
 172                  * that case.
 173                  */
 174                 zap_leader = (leader->exit_signal == -1);
 175         }
 176
 177         sched_exit(p);
 178         write_unlock_irq(&tasklist_lock);
 179         spin_unlock(&p->proc_lock);
 180         proc_pid_flush(proc_dentry);
 181         release_thread(p);
 182         call_rcu(&p->rcu, delayed_put_task_struct);
 183
 184         p = leader;
 185         if (unlikely(zap_leader))
 186                 goto repeat;
 187 }
 188
 189 /*
 190  * This checks not only the pgrp, but falls back on the pid if no
 191  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 192  * without this...
 193  */
 194 int session_of_pgrp(int pgrp)
 195 {
 196         struct task_struct *p;
 197         int sid = -1;
 198
 199         read_lock(&tasklist_lock);
 200         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 201                 if (p->signal->session > 0) {
 202                         sid = p->signal->session;
 203                         goto out;
 204                 }
 205         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 206         p = find_task_by_pid(pgrp);
 207         if (p)
 208                 sid = p->signal->session;
 209 out:
 210         read_unlock(&tasklist_lock);
 211
 212         return sid;
 213 }
 214
 215 /*
 216  * Determine if a process group is "orphaned", according to the POSIX
 217  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 218  * by terminal-generated stop signals.  Newly orphaned process groups are
 219  * to receive a SIGHUP and a SIGCONT.
 220  *
 221  * "I ask you, have you ever known what it is to be an orphan?"
 222  */
 223 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 224 {
 225         struct task_struct *p;
 226         int ret = 1;
 227
 228         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 229                 if (p == ignored_task
 230                                 || p->exit_state
 231                                 || p->real_parent->pid == 1)
 232                         continue;
 233                 if (process_group(p->real_parent) != pgrp
 234                             && p->real_parent->signal->session == p->signal->session) {
 235                         ret = 0;
 236                         break;
 237                 }
 238         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 239         return ret;     /* (sighing) "Often!" */
 240 }
 241
 242 int is_orphaned_pgrp(int pgrp)
 243 {
 244         int retval;
 245
 246         read_lock(&tasklist_lock);
 247         retval = will_become_orphaned_pgrp(pgrp, NULL);
 248         read_unlock(&tasklist_lock);
 249
 250         return retval;
 251 }
 252
 253 static int has_stopped_jobs(int pgrp)
 254 {
 255         int retval = 0;
 256         struct task_struct *p;
 257
 258         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 259                 if (p->state != TASK_STOPPED)
 260                         continue;
 261
 262                 /* If p is stopped by a debugger on a signal that won't
 263                    stop it, then don't count p as stopped.  This isn't
 264                    perfect but it's a good approximation.  */
 265                 if (unlikely (p->ptrace)
 266                     && p->exit_code != SIGSTOP
 267                     && p->exit_code != SIGTSTP
 268                     && p->exit_code != SIGTTOU
 269                     && p->exit_code != SIGTTIN)
 270                         continue;
 271
 272                 retval = 1;
 273                 break;
 274         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 275         return retval;
 276 }
 277
 278 /**
 279  * reparent_to_init - Reparent the calling kernel thread to the init task.
 280  *
 281  * If a kernel thread is launched as a result of a system call, or if
 282  * it ever exits, it should generally reparent itself to init so that
 283  * it is correctly cleaned up on exit.
 284  *
 285  * The various task state such as scheduling policy and priority may have
 286  * been inherited from a user process, so we reset them to sane values here.
 287  *
 288  * NOTE that reparent_to_init() gives the caller full capabilities.
 289  */
 290 static void reparent_to_init(void)
 291 {
 292         write_lock_irq(&tasklist_lock);
 293
 294         ptrace_unlink(current);
 295         /* Reparent to init */
 296         remove_parent(current);
 297         current->parent = child_reaper;
 298         current->real_parent = child_reaper;
 299         add_parent(current);
 300
 301         /* Set the exit signal to SIGCHLD so we signal init on exit */
 302         current->exit_signal = SIGCHLD;
 303
 304         if ((current->policy == SCHED_NORMAL ||
 305                         current->policy == SCHED_BATCH)
 306                                 && (task_nice(current) < 0))
 307                 set_user_nice(current, 0);
 308         /* cpus_allowed? */
 309         /* rt_priority? */
 310         /* signals? */
 311         security_task_reparent_to_init(current);
 312         memcpy(current->signal->rlim, init_task.signal->rlim,
 313                sizeof(current->signal->rlim));
 314         atomic_inc(&(INIT_USER->__count));
 315         write_unlock_irq(&tasklist_lock);
 316         switch_uid(INIT_USER);
 317 }
 318
 319 void __set_special_pids(pid_t session, pid_t pgrp)
 320 {
 321         struct task_struct *curr = current->group_leader;
 322
 323         if (curr->signal->session != session) {
 324                 detach_pid(curr, PIDTYPE_SID);
 325                 curr->signal->session = session;
 326                 attach_pid(curr, PIDTYPE_SID, session);
 327         }
 328         if (process_group(curr) != pgrp) {
 329                 detach_pid(curr, PIDTYPE_PGID);
 330                 curr->signal->pgrp = pgrp;
 331                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 332         }
 333 }
 334
 335 void set_special_pids(pid_t session, pid_t pgrp)
 336 {
 337         write_lock_irq(&tasklist_lock);
 338         __set_special_pids(session, pgrp);
 339         write_unlock_irq(&tasklist_lock);
 340 }
 341
 342 /*
 343  * Let kernel threads use this to say that they
 344  * allow a certain signal (since daemonize() will
 345  * have disabled all of them by default).
 346  */
 347 int allow_signal(int sig)
 348 {
 349         if (!valid_signal(sig) || sig < 1)
 350                 return -EINVAL;
 351
 352         spin_lock_irq(&current->sighand->siglock);
 353         sigdelset(&current->blocked, sig);
 354         if (!current->mm) {
 355                 /* Kernel threads handle their own signals.
 356                    Let the signal code know it'll be handled, so
 357                    that they don't get converted to SIGKILL or
 358                    just silently dropped */
 359                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 360         }
 361         recalc_sigpending();
 362         spin_unlock_irq(&current->sighand->siglock);
 363         return 0;
 364 }
 365
 366 EXPORT_SYMBOL(allow_signal);
 367
 368 int disallow_signal(int sig)
 369 {
 370         if (!valid_signal(sig) || sig < 1)
 371                 return -EINVAL;
 372
 373         spin_lock_irq(&current->sighand->siglock);
 374         sigaddset(&current->blocked, sig);
 375         recalc_sigpending();
 376         spin_unlock_irq(&current->sighand->siglock);
 377         return 0;
 378 }
 379
 380 EXPORT_SYMBOL(disallow_signal);
 381
 382 /*
 383  *      Put all the gunge required to become a kernel thread without
 384  *      attached user resources in one place where it belongs.
 385  */
 386
 387 void daemonize(const char *name, ...)
 388 {
 389         va_list args;
 390         struct fs_struct *fs;
 391         sigset_t blocked;
 392
 393         va_start(args, name);
 394         vsnprintf(current->comm, sizeof(current->comm), name, args);
 395         va_end(args);
 396
 397         /*
 398          * If we were started as result of loading a module, close all of the
 399          * user space pages.  We don't need them, and if we didn't close them
 400          * they would be locked into memory.
 401          */
 402         exit_mm(current);
 403
 404         set_special_pids(1, 1);
 405         mutex_lock(&tty_mutex);
 406         current->signal->tty = NULL;
 407         mutex_unlock(&tty_mutex);
 408
 409         /* Block and flush all signals */
 410         sigfillset(&blocked);
 411         sigprocmask(SIG_BLOCK, &blocked, NULL);
 412         flush_signals(current);
 413
 414         /* Become as one with the init task */
 415
 416         exit_fs(current);       /* current->fs->count--; */
 417         fs = init_task.fs;
 418         current->fs = fs;
 419         atomic_inc(&fs->count);
 420         exit_namespace(current);
 421         current->namespace = init_task.namespace;
 422         get_namespace(current->namespace);
 423         exit_files(current);
 424         current->files = init_task.files;
 425         atomic_inc(&current->files->count);
 426
 427         reparent_to_init();
 428 }
 429
 430 EXPORT_SYMBOL(daemonize);
 431
 432 static void close_files(struct files_struct * files)
 433 {
 434         int i, j;
 435         struct fdtable *fdt;
 436
 437         j = 0;
 438
 439         /*
 440          * It is safe to dereference the fd table without RCU or
 441          * ->file_lock because this is the last reference to the
 442          * files structure.
 443          */
 444         fdt = files_fdtable(files);
 445         for (;;) {
 446                 unsigned long set;
 447                 i = j * __NFDBITS;
 448                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 449                         break;
 450                 set = fdt->open_fds->fds_bits[j++];
 451                 while (set) {
 452                         if (set & 1) {
 453                                 struct file * file = xchg(&fdt->fd[i], NULL);
 454                                 if (file)
 455                                         filp_close(file, files);
 456                                 vx_openfd_dec(i);
 457                         }
 458                         i++;
 459                         set >>= 1;
 460                         cond_resched();
 461                 }
 462         }
 463 }
 464
 465 struct files_struct *get_files_struct(struct task_struct *task)
 466 {
 467         struct files_struct *files;
 468
 469         task_lock(task);
 470         files = task->files;
 471         if (files)
 472                 atomic_inc(&files->count);
 473         task_unlock(task);
 474
 475         return files;
 476 }
 477
 478 void fastcall put_files_struct(struct files_struct *files)
 479 {
 480         struct fdtable *fdt;
 481
 482         if (atomic_dec_and_test(&files->count)) {
 483                 close_files(files);
 484                 /*
 485                  * Free the fd and fdset arrays if we expanded them.
 486                  * If the fdtable was embedded, pass files for freeing
 487                  * at the end of the RCU grace period. Otherwise,
 488                  * you can free files immediately.
 489                  */
 490                 fdt = files_fdtable(files);
 491                 if (fdt == &files->fdtab)
 492                         fdt->free_files = files;
 493                 else
 494                         kmem_cache_free(files_cachep, files);
 495                 free_fdtable(fdt);
 496         }
 497 }
 498
 499 EXPORT_SYMBOL(put_files_struct);
 500
 501 static inline void __exit_files(struct task_struct *tsk)
 502 {
 503         struct files_struct * files = tsk->files;
 504
 505         if (files) {
 506                 task_lock(tsk);
 507                 tsk->files = NULL;
 508                 task_unlock(tsk);
 509                 put_files_struct(files);
 510         }
 511 }
 512
 513 void exit_files(struct task_struct *tsk)
 514 {
 515         __exit_files(tsk);
 516 }
 517
 518 static inline void __put_fs_struct(struct fs_struct *fs)
 519 {
 520         /* No need to hold fs->lock if we are killing it */
 521         if (atomic_dec_and_test(&fs->count)) {
 522                 dput(fs->root);
 523                 mntput(fs->rootmnt);
 524                 dput(fs->pwd);
 525                 mntput(fs->pwdmnt);
 526                 if (fs->altroot) {
 527                         dput(fs->altroot);
 528                         mntput(fs->altrootmnt);
 529                 }
 530                 kmem_cache_free(fs_cachep, fs);
 531         }
 532 }
 533
 534 void put_fs_struct(struct fs_struct *fs)
 535 {
 536         __put_fs_struct(fs);
 537 }
 538
 539 static inline void __exit_fs(struct task_struct *tsk)
 540 {
 541         struct fs_struct * fs = tsk->fs;
 542
 543         if (fs) {
 544                 task_lock(tsk);
 545                 tsk->fs = NULL;
 546                 task_unlock(tsk);
 547                 __put_fs_struct(fs);
 548         }
 549 }
 550
 551 void exit_fs(struct task_struct *tsk)
 552 {
 553         __exit_fs(tsk);
 554 }
 555
 556 EXPORT_SYMBOL_GPL(exit_fs);
 557
 558 /*
 559  * Turn us into a lazy TLB process if we
 560  * aren't already..
 561  */
 562 static void exit_mm(struct task_struct * tsk)
 563 {
 564         struct mm_struct *mm = tsk->mm;
 565
 566         mm_release(tsk, mm);
 567         if (!mm)
 568                 return;
 569         /*
 570          * Serialize with any possible pending coredump.
 571          * We must hold mmap_sem around checking core_waiters
 572          * and clearing tsk->mm.  The core-inducing thread
 573          * will increment core_waiters for each thread in the
 574          * group with ->mm != NULL.
 575          */
 576         down_read(&mm->mmap_sem);
 577         if (mm->core_waiters) {
 578                 up_read(&mm->mmap_sem);
 579                 down_write(&mm->mmap_sem);
 580                 if (!--mm->core_waiters)
 581                         complete(mm->core_startup_done);
 582                 up_write(&mm->mmap_sem);
 583
 584                 wait_for_completion(&mm->core_done);
 585                 down_read(&mm->mmap_sem);
 586         }
 587         atomic_inc(&mm->mm_count);
 588         if (mm != tsk->active_mm) BUG();
 589         /* more a memory barrier than a real lock */
 590         task_lock(tsk);
 591         tsk->mm = NULL;
 592         up_read(&mm->mmap_sem);
 593         enter_lazy_tlb(mm, current);
 594         task_unlock(tsk);
 595         mmput(mm);
 596 }
 597
 598 static inline void choose_new_parent(task_t *p, task_t *reaper)
 599 {
 600         /* check for reaper context */
 601         vxwprintk((p->xid != reaper->xid) && (reaper != child_reaper),
 602                 "rogue reaper: %p[%d,#%u] <> %p[%d,#%u]",
 603                 p, p->pid, p->xid, reaper, reaper->pid, reaper->xid);
 604
 605         /*
 606          * Make sure we're not reparenting to ourselves and that
 607          * the parent is not a zombie.
 608          */
 609         BUG_ON(p == reaper || reaper->exit_state);
 610         p->real_parent = reaper;
 611 }
 612
 613 static void reparent_thread(task_t *p, task_t *father, int traced)
 614 {
 615         /* We don't want people slaying init.  */
 616         if (p->exit_signal != -1)
 617                 p->exit_signal = SIGCHLD;
 618
 619         if (p->pdeath_signal)
 620                 /* We already hold the tasklist_lock here.  */
 621                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 622
 623         /* Move the child from its dying parent to the new one.  */
 624         if (unlikely(traced)) {
 625                 /* Preserve ptrace links if someone else is tracing this child.  */
 626                 list_del_init(&p->ptrace_list);
 627                 if (p->parent != p->real_parent)
 628                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 629         } else {
 630                 /* If this child is being traced, then we're the one tracing it
 631                  * anyway, so let go of it.
 632                  */
 633                 p->ptrace = 0;
 634                 remove_parent(p);
 635                 p->parent = p->real_parent;
 636                 add_parent(p);
 637
 638                 /* If we'd notified the old parent about this child's death,
 639                  * also notify the new parent.
 640                  */
 641                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 642                     thread_group_empty(p))
 643                         do_notify_parent(p, p->exit_signal);
 644                 else if (p->state == TASK_TRACED) {
 645                         /*
 646                          * If it was at a trace stop, turn it into
 647                          * a normal stop since it's no longer being
 648                          * traced.
 649                          */
 650                         ptrace_untrace(p);
 651                 }
 652         }
 653
 654         /*
 655          * process group orphan check
 656          * Case ii: Our child is in a different pgrp
 657          * than we are, and it was the only connection
 658          * outside, so the child pgrp is now orphaned.
 659          */
 660         if ((process_group(p) != process_group(father)) &&
 661             (p->signal->session == father->signal->session)) {
 662                 int pgrp = process_group(p);
 663
 664                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 665                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 666                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 667                 }
 668         }
 669 }
 670
 671 /*
 672  * When we die, we re-parent all our children.
 673  * Try to give them to another thread in our thread
 674  * group, and if no such member exists, give it to
 675  * the global child reaper process (ie "init")
 676  */
 677 static void forget_original_parent(struct task_struct * father,
 678                                           struct list_head *to_release)
 679 {
 680         struct task_struct *p, *reaper = father;
 681         struct list_head *_p, *_n;
 682
 683         do {
 684                 reaper = next_thread(reaper);
 685                 if (reaper == father) {
 686                         reaper = vx_child_reaper(father);
 687                         break;
 688                 }
 689         } while (reaper->exit_state);
 690
 691         /*
 692          * There are only two places where our children can be:
 693          *
 694          * - in our child list
 695          * - in our ptraced child list
 696          *
 697          * Search them and reparent children.
 698          */
 699         list_for_each_safe(_p, _n, &father->children) {
 700                 int ptrace;
 701                 p = list_entry(_p,struct task_struct,sibling);
 702
 703                 ptrace = p->ptrace;
 704
 705                 /* if father isn't the real parent, then ptrace must be enabled */
 706                 BUG_ON(father != p->real_parent && !ptrace);
 707
 708                 if (father == p->real_parent) {
 709                         /* reparent with a reaper, real father it's us */
 710                         choose_new_parent(p, vx_child_reaper(p));
 711                         reparent_thread(p, father, 0);
 712                 } else {
 713                         /* reparent ptraced task to its real parent */
 714                         __ptrace_unlink (p);
 715                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 716                             thread_group_empty(p))
 717                                 do_notify_parent(p, p->exit_signal);
 718                 }
 719
 720                 /*
 721                  * if the ptraced child is a zombie with exit_signal == -1
 722                  * we must collect it before we exit, or it will remain
 723                  * zombie forever since we prevented it from self-reap itself
 724                  * while it was being traced by us, to be able to see it in wait4.
 725                  */
 726                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 727                         list_add(&p->ptrace_list, to_release);
 728         }
 729         list_for_each_safe(_p, _n, &father->ptrace_children) {
 730                 p = list_entry(_p,struct task_struct,ptrace_list);
 731                 choose_new_parent(p, reaper);
 732                 reparent_thread(p, father, 1);
 733         }
 734 }
 735
 736 /*
 737  * Send signals to all our closest relatives so that they know
 738  * to properly mourn us..
 739  */
 740 static void exit_notify(struct task_struct *tsk)
 741 {
 742         int state;
 743         struct task_struct *t;
 744         struct list_head ptrace_dead, *_p, *_n;
 745
 746         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 747             && !thread_group_empty(tsk)) {
 748                 /*
 749                  * This occurs when there was a race between our exit
 750                  * syscall and a group signal choosing us as the one to
 751                  * wake up.  It could be that we are the only thread
 752                  * alerted to check for pending signals, but another thread
 753                  * should be woken now to take the signal since we will not.
 754                  * Now we'll wake all the threads in the group just to make
 755                  * sure someone gets all the pending signals.
 756                  */
 757                 read_lock(&tasklist_lock);
 758                 spin_lock_irq(&tsk->sighand->siglock);
 759                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 760                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 761                                 recalc_sigpending_tsk(t);
 762                                 if (signal_pending(t))
 763                                         signal_wake_up(t, 0);
 764                         }
 765                 spin_unlock_irq(&tsk->sighand->siglock);
 766                 read_unlock(&tasklist_lock);
 767         }
 768
 769         write_lock_irq(&tasklist_lock);
 770
 771         /*
 772          * This does two things:
 773          *
 774          * A.  Make init inherit all the child processes
 775          * B.  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
 780         INIT_LIST_HEAD(&ptrace_dead);
 781         forget_original_parent(tsk, &ptrace_dead);
 782         BUG_ON(!list_empty(&tsk->children));
 783         BUG_ON(!list_empty(&tsk->ptrace_children));
 784
 785         /*
 786          * Check to see if any process groups have become orphaned
 787          * as a result of our exiting, and if they have any stopped
 788          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 789          *
 790          * Case i: Our father is in a different pgrp than we are
 791          * and we were the only connection outside, so our pgrp
 792          * is about to become orphaned.
 793          */
 794
 795         t = tsk->real_parent;
 796
 797         if ((process_group(t) != process_group(tsk)) &&
 798             (t->signal->session == tsk->signal->session) &&
 799             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 800             has_stopped_jobs(process_group(tsk))) {
 801                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 802                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 803         }
 804
 805         /* Let father know we died
 806          *
 807          * Thread signals are configurable, but you aren't going to use
 808          * that to send signals to arbitary processes.
 809          * That stops right now.
 810          *
 811          * If the parent exec id doesn't match the exec id we saved
 812          * when we started then we know the parent has changed security
 813          * domain.
 814          *
 815          * If our self_exec id doesn't match our parent_exec_id then
 816          * we have changed execution domain as these two values started
 817          * the same after a fork.
 818          *
 819          */
 820
 821         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 822             ( tsk->parent_exec_id != t->self_exec_id  ||
 823               tsk->self_exec_id != tsk->parent_exec_id)
 824             && !capable(CAP_KILL))
 825                 tsk->exit_signal = SIGCHLD;
 826
 827
 828         /* If something other than our normal parent is ptracing us, then
 829          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 830          * only has special meaning to our real parent.
 831          */
 832         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 833                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 834                 do_notify_parent(tsk, signal);
 835         } else if (tsk->ptrace) {
 836                 do_notify_parent(tsk, SIGCHLD);
 837         }
 838
 839         state = EXIT_ZOMBIE;
 840         if (tsk->exit_signal == -1 &&
 841             (likely(tsk->ptrace == 0) ||
 842              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 843                 state = EXIT_DEAD;
 844         tsk->exit_state = state;
 845
 846         write_unlock_irq(&tasklist_lock);
 847
 848         list_for_each_safe(_p, _n, &ptrace_dead) {
 849                 list_del_init(_p);
 850                 t = list_entry(_p,struct task_struct,ptrace_list);
 851                 release_task(t);
 852         }
 853
 854         /* If the process is dead, release it - nobody will wait for it */
 855         if (state == EXIT_DEAD)
 856                 release_task(tsk);
 857 }
 858
 859 fastcall NORET_TYPE void do_exit(long code)
 860 {
 861         struct task_struct *tsk = current;
 862         int group_dead;
 863
 864         profile_task_exit(tsk);
 865
 866         WARN_ON(atomic_read(&tsk->fs_excl));
 867
 868         if (unlikely(in_interrupt()))
 869                 panic("Aiee, killing interrupt handler!");
 870         if (unlikely(!tsk->pid))
 871                 panic("Attempted to kill the idle task!");
 872         if (unlikely(tsk == child_reaper))
 873                 panic("Attempted to kill init!");
 874
 875         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 876                 current->ptrace_message = code;
 877                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 878         }
 879
 880         /*
 881          * We're taking recursive faults here in do_exit. Safest is to just
 882          * leave this task alone and wait for reboot.
 883          */
 884         if (unlikely(tsk->flags & PF_EXITING)) {
 885                 printk(KERN_ALERT
 886                         "Fixing recursive fault but reboot is needed!\n");
 887                 if (tsk->io_context)
 888                         exit_io_context();
 889                 set_current_state(TASK_UNINTERRUPTIBLE);
 890                 schedule();
 891         }
 892
 893         tsk->flags |= PF_EXITING;
 894
 895         if (unlikely(in_atomic()))
 896                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 897                                 current->comm, current->pid,
 898                                 preempt_count());
 899
 900         acct_update_integrals(tsk);
 901         if (tsk->mm) {
 902                 update_hiwater_rss(tsk->mm);
 903                 update_hiwater_vm(tsk->mm);
 904         }
 905         group_dead = atomic_dec_and_test(&tsk->signal->live);
 906         if (group_dead) {
 907                 hrtimer_cancel(&tsk->signal->real_timer);
 908                 exit_itimers(tsk->signal);
 909                 acct_process(code);
 910                 if (current->tux_info) {
 911 #ifdef CONFIG_TUX_DEBUG
 912                         printk("Possibly unexpected TUX-thread exit(%ld) at %p?\n",
 913                                 code, __builtin_return_address(0));
 914 #endif
 915                         current->tux_exit();
 916                 }
 917         }
 918         if (unlikely(tsk->robust_list))
 919                 exit_robust_list(tsk);
 920 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 921         if (unlikely(tsk->compat_robust_list))
 922                 compat_exit_robust_list(tsk);
 923 #endif
 924         if (unlikely(tsk->audit_context))
 925                 audit_free(tsk);
 926         exit_mm(tsk);
 927
 928         exit_sem(tsk);
 929         __exit_files(tsk);
 930         __exit_fs(tsk);
 931         exit_namespace(tsk);
 932         exit_vx_info(tsk, code);
 933         exit_nx_info(tsk);
 934         exit_thread();
 935         cpuset_exit(tsk);
 936         exit_keys(tsk);
 937
 938         if (group_dead && tsk->signal->leader)
 939                 disassociate_ctty(1);
 940
 941         module_put(task_thread_info(tsk)->exec_domain->module);
 942         if (tsk->binfmt)
 943                 module_put(tsk->binfmt->module);
 944
 945         tsk->exit_code = code;
 946         proc_exit_connector(tsk);
 947         exit_notify(tsk);
 948 #ifdef CONFIG_NUMA
 949         mpol_free(tsk->mempolicy);
 950         tsk->mempolicy = NULL;
 951 #endif
 952         /*
 953          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
 954          */
 955         mutex_debug_check_no_locks_held(tsk);
 956
 957         if (tsk->io_context)
 958                 exit_io_context();
 959
 960         if (tsk->splice_pipe)
 961                 __free_pipe_info(tsk->splice_pipe);
 962
 963         /* PF_DEAD causes final put_task_struct after we schedule. */
 964         preempt_disable();
 965         BUG_ON(tsk->flags & PF_DEAD);
 966         tsk->flags |= PF_DEAD;
 967
 968         schedule();
 969         BUG();
 970         /* Avoid "noreturn function does return".  */
 971         for (;;) ;
 972 }
 973
 974 EXPORT_SYMBOL_GPL(do_exit);
 975
 976 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 977 {
 978         if (comp)
 979                 complete(comp);
 980
 981         do_exit(code);
 982 }
 983
 984 EXPORT_SYMBOL(complete_and_exit);
 985
 986 asmlinkage long sys_exit(int error_code)
 987 {
 988         do_exit((error_code&0xff)<<8);
 989 }
 990
 991 /*
 992  * Take down every thread in the group.  This is called by fatal signals
 993  * as well as by sys_exit_group (below).
 994  */
 995 NORET_TYPE void
 996 do_group_exit(int exit_code)
 997 {
 998         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 999
1000         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1001                 exit_code = current->signal->group_exit_code;
1002         else if (!thread_group_empty(current)) {
1003                 struct signal_struct *const sig = current->signal;
1004                 struct sighand_struct *const sighand = current->sighand;
1005                 spin_lock_irq(&sighand->siglock);
1006                 if (sig->flags & SIGNAL_GROUP_EXIT)
1007                         /* Another thread got here before we took the lock.  */
1008                         exit_code = sig->group_exit_code;
1009                 else {
1010                         sig->group_exit_code = exit_code;
1011                         zap_other_threads(current);
1012                 }
1013                 spin_unlock_irq(&sighand->siglock);
1014         }
1015
1016         do_exit(exit_code);
1017         /* NOTREACHED */
1018 }
1019
1020 /*
1021  * this kills every thread in the thread group. Note that any externally
1022  * wait4()-ing process will get the correct exit code - even if this
1023  * thread is not the thread group leader.
1024  */
1025 asmlinkage void sys_exit_group(int error_code)
1026 {
1027         do_group_exit((error_code & 0xff) << 8);
1028 }
1029
1030 static int eligible_child(pid_t pid, int options, task_t *p)
1031 {
1032         if (pid > 0) {
1033                 if (p->pid != pid)
1034                         return 0;
1035         } else if (!pid) {
1036                 if (process_group(p) != process_group(current))
1037                         return 0;
1038         } else if (pid != -1) {
1039                 if (process_group(p) != -pid)
1040                         return 0;
1041         }
1042
1043         /*
1044          * Do not consider detached threads that are
1045          * not ptraced:
1046          */
1047         if (p->exit_signal == -1 && !p->ptrace)
1048                 return 0;
1049
1050         /* Wait for all children (clone and not) if __WALL is set;
1051          * otherwise, wait for clone children *only* if __WCLONE is
1052          * set; otherwise, wait for non-clone children *only*.  (Note:
1053          * A "clone" child here is one that reports to its parent
1054          * using a signal other than SIGCHLD.) */
1055         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1056             && !(options & __WALL))
1057                 return 0;
1058         /*
1059          * Do not consider thread group leaders that are
1060          * in a non-empty thread group:
1061          */
1062         if (current->tgid != p->tgid && delay_group_leader(p))
1063                 return 2;
1064
1065         if (security_task_wait(p))
1066                 return 0;
1067
1068         return 1;
1069 }
1070
1071 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
1072                                int why, int status,
1073                                struct siginfo __user *infop,
1074                                struct rusage __user *rusagep)
1075 {
1076         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1077         put_task_struct(p);
1078         if (!retval)
1079                 retval = put_user(SIGCHLD, &infop->si_signo);
1080         if (!retval)
1081                 retval = put_user(0, &infop->si_errno);
1082         if (!retval)
1083                 retval = put_user((short)why, &infop->si_code);
1084         if (!retval)
1085                 retval = put_user(pid, &infop->si_pid);
1086         if (!retval)
1087                 retval = put_user(uid, &infop->si_uid);
1088         if (!retval)
1089                 retval = put_user(status, &infop->si_status);
1090         if (!retval)
1091                 retval = pid;
1092         return retval;
1093 }
1094
1095 /*
1096  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1097  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1098  * the lock and this task is uninteresting.  If we return nonzero, we have
1099  * released the lock and the system call should return.
1100  */
1101 static int wait_task_zombie(task_t *p, int noreap,
1102                             struct siginfo __user *infop,
1103                             int __user *stat_addr, struct rusage __user *ru)
1104 {
1105         unsigned long state;
1106         int retval;
1107         int status;
1108
1109         if (unlikely(noreap)) {
1110                 pid_t pid = p->pid;
1111                 uid_t uid = p->uid;
1112                 int exit_code = p->exit_code;
1113                 int why, status;
1114
1115                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1116                         return 0;
1117                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1118                         return 0;
1119                 get_task_struct(p);
1120                 read_unlock(&tasklist_lock);
1121                 if ((exit_code & 0x7f) == 0) {
1122                         why = CLD_EXITED;
1123                         status = exit_code >> 8;
1124                 } else {
1125                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1126                         status = exit_code & 0x7f;
1127                 }
1128                 return wait_noreap_copyout(p, pid, uid, why,
1129                                            status, infop, ru);
1130         }
1131
1132         /*
1133          * Try to move the task's state to DEAD
1134          * only one thread is allowed to do this:
1135          */
1136         state = xchg(&p->exit_state, EXIT_DEAD);
1137         if (state != EXIT_ZOMBIE) {
1138                 BUG_ON(state != EXIT_DEAD);
1139                 return 0;
1140         }
1141         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1142                 /*
1143                  * This can only happen in a race with a ptraced thread
1144                  * dying on another processor.
1145                  */
1146                 return 0;
1147         }
1148
1149         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1150                 struct signal_struct *psig;
1151                 struct signal_struct *sig;
1152
1153                 /*
1154                  * The resource counters for the group leader are in its
1155                  * own task_struct.  Those for dead threads in the group
1156                  * are in its signal_struct, as are those for the child
1157                  * processes it has previously reaped.  All these
1158                  * accumulate in the parent's signal_struct c* fields.
1159                  *
1160                  * We don't bother to take a lock here to protect these
1161                  * p->signal fields, because they are only touched by
1162                  * __exit_signal, which runs with tasklist_lock
1163                  * write-locked anyway, and so is excluded here.  We do
1164                  * need to protect the access to p->parent->signal fields,
1165                  * as other threads in the parent group can be right
1166                  * here reaping other children at the same time.
1167                  */
1168                 spin_lock_irq(&p->parent->sighand->siglock);
1169                 psig = p->parent->signal;
1170                 sig = p->signal;
1171                 psig->cutime =
1172                         cputime_add(psig->cutime,
1173                         cputime_add(p->utime,
1174                         cputime_add(sig->utime,
1175                                     sig->cutime)));
1176                 psig->cstime =
1177                         cputime_add(psig->cstime,
1178                         cputime_add(p->stime,
1179                         cputime_add(sig->stime,
1180                                     sig->cstime)));
1181                 psig->cmin_flt +=
1182                         p->min_flt + sig->min_flt + sig->cmin_flt;
1183                 psig->cmaj_flt +=
1184                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1185                 psig->cnvcsw +=
1186                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1187                 psig->cnivcsw +=
1188                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1189                 spin_unlock_irq(&p->parent->sighand->siglock);
1190         }
1191
1192         /*
1193          * Now we are sure this task is interesting, and no other
1194          * thread can reap it because we set its state to EXIT_DEAD.
1195          */
1196         read_unlock(&tasklist_lock);
1197
1198         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1199         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1200                 ? p->signal->group_exit_code : p->exit_code;
1201         if (!retval && stat_addr)
1202                 retval = put_user(status, stat_addr);
1203         if (!retval && infop)
1204                 retval = put_user(SIGCHLD, &infop->si_signo);
1205         if (!retval && infop)
1206                 retval = put_user(0, &infop->si_errno);
1207         if (!retval && infop) {
1208                 int why;
1209
1210                 if ((status & 0x7f) == 0) {
1211                         why = CLD_EXITED;
1212                         status >>= 8;
1213                 } else {
1214                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1215                         status &= 0x7f;
1216                 }
1217                 retval = put_user((short)why, &infop->si_code);
1218                 if (!retval)
1219                         retval = put_user(status, &infop->si_status);
1220         }
1221         if (!retval && infop)
1222                 retval = put_user(p->pid, &infop->si_pid);
1223         if (!retval && infop)
1224                 retval = put_user(p->uid, &infop->si_uid);
1225         if (retval) {
1226                 // TODO: is this safe?
1227                 p->exit_state = EXIT_ZOMBIE;
1228                 return retval;
1229         }
1230         retval = p->pid;
1231         if (p->real_parent != p->parent) {
1232                 write_lock_irq(&tasklist_lock);
1233                 /* Double-check with lock held.  */
1234                 if (p->real_parent != p->parent) {
1235                         __ptrace_unlink(p);
1236                         // TODO: is this safe?
1237                         p->exit_state = EXIT_ZOMBIE;
1238                         /*
1239                          * If this is not a detached task, notify the parent.
1240                          * If it's still not detached after that, don't release
1241                          * it now.
1242                          */
1243                         if (p->exit_signal != -1) {
1244                                 do_notify_parent(p, p->exit_signal);
1245                                 if (p->exit_signal != -1)
1246                                         p = NULL;
1247                         }
1248                 }
1249                 write_unlock_irq(&tasklist_lock);
1250         }
1251         if (p != NULL)
1252                 release_task(p);
1253         BUG_ON(!retval);
1254         return retval;
1255 }
1256
1257 /*
1258  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1259  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1260  * the lock and this task is uninteresting.  If we return nonzero, we have
1261  * released the lock and the system call should return.
1262  */
1263 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1264                              struct siginfo __user *infop,
1265                              int __user *stat_addr, struct rusage __user *ru)
1266 {
1267         int retval, exit_code;
1268
1269         if (!p->exit_code)
1270                 return 0;
1271         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1272             p->signal && p->signal->group_stop_count > 0)
1273                 /*
1274                  * A group stop is in progress and this is the group leader.
1275                  * We won't report until all threads have stopped.
1276                  */
1277                 return 0;
1278
1279         /*
1280          * Now we are pretty sure this task is interesting.
1281          * Make sure it doesn't get reaped out from under us while we
1282          * give up the lock and then examine it below.  We don't want to
1283          * keep holding onto the tasklist_lock while we call getrusage and
1284          * possibly take page faults for user memory.
1285          */
1286         get_task_struct(p);
1287         read_unlock(&tasklist_lock);
1288
1289         if (unlikely(noreap)) {
1290                 pid_t pid = p->pid;
1291                 uid_t uid = p->uid;
1292                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1293
1294                 exit_code = p->exit_code;
1295                 if (unlikely(!exit_code) ||
1296                     unlikely(p->state & TASK_TRACED))
1297                         goto bail_ref;
1298                 return wait_noreap_copyout(p, pid, uid,
1299                                            why, (exit_code << 8) | 0x7f,
1300                                            infop, ru);
1301         }
1302
1303         write_lock_irq(&tasklist_lock);
1304
1305         /*
1306          * This uses xchg to be atomic with the thread resuming and setting
1307          * it.  It must also be done with the write lock held to prevent a
1308          * race with the EXIT_ZOMBIE case.
1309          */
1310         exit_code = xchg(&p->exit_code, 0);
1311         if (unlikely(p->exit_state)) {
1312                 /*
1313                  * The task resumed and then died.  Let the next iteration
1314                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1315                  * already be zero here if it resumed and did _exit(0).
1316                  * The task itself is dead and won't touch exit_code again;
1317                  * other processors in this function are locked out.
1318                  */
1319                 p->exit_code = exit_code;
1320                 exit_code = 0;
1321         }
1322         if (unlikely(exit_code == 0)) {
1323                 /*
1324                  * Another thread in this function got to it first, or it
1325                  * resumed, or it resumed and then died.
1326                  */
1327                 write_unlock_irq(&tasklist_lock);
1328 bail_ref:
1329                 put_task_struct(p);
1330                 /*
1331                  * We are returning to the wait loop without having successfully
1332                  * removed the process and having released the lock. We cannot
1333                  * continue, since the "p" task pointer is potentially stale.
1334                  *
1335                  * Return -EAGAIN, and do_wait() will restart the loop from the
1336                  * beginning. Do _not_ re-acquire the lock.
1337                  */
1338                 return -EAGAIN;
1339         }
1340
1341         /* move to end of parent's list to avoid starvation */
1342         remove_parent(p);
1343         add_parent(p);
1344
1345         write_unlock_irq(&tasklist_lock);
1346
1347         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1348         if (!retval && stat_addr)
1349                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1350         if (!retval && infop)
1351                 retval = put_user(SIGCHLD, &infop->si_signo);
1352         if (!retval && infop)
1353                 retval = put_user(0, &infop->si_errno);
1354         if (!retval && infop)
1355                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1356                                           ? CLD_TRAPPED : CLD_STOPPED),
1357                                   &infop->si_code);
1358         if (!retval && infop)
1359                 retval = put_user(exit_code, &infop->si_status);
1360         if (!retval && infop)
1361                 retval = put_user(p->pid, &infop->si_pid);
1362         if (!retval && infop)
1363                 retval = put_user(p->uid, &infop->si_uid);
1364         if (!retval)
1365                 retval = p->pid;
1366         put_task_struct(p);
1367
1368         BUG_ON(!retval);
1369         return retval;
1370 }
1371
1372 /*
1373  * Handle do_wait work for one task in a live, non-stopped state.
1374  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1375  * the lock and this task is uninteresting.  If we return nonzero, we have
1376  * released the lock and the system call should return.
1377  */
1378 static int wait_task_continued(task_t *p, int noreap,
1379                                struct siginfo __user *infop,
1380                                int __user *stat_addr, struct rusage __user *ru)
1381 {
1382         int retval;
1383         pid_t pid;
1384         uid_t uid;
1385
1386         if (unlikely(!p->signal))
1387                 return 0;
1388
1389         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1390                 return 0;
1391
1392         spin_lock_irq(&p->sighand->siglock);
1393         /* Re-check with the lock held.  */
1394         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1395                 spin_unlock_irq(&p->sighand->siglock);
1396                 return 0;
1397         }
1398         if (!noreap)
1399                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1400         spin_unlock_irq(&p->sighand->siglock);
1401
1402         pid = p->pid;
1403         uid = p->uid;
1404         get_task_struct(p);
1405         read_unlock(&tasklist_lock);
1406
1407         if (!infop) {
1408                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1409                 put_task_struct(p);
1410                 if (!retval && stat_addr)
1411                         retval = put_user(0xffff, stat_addr);
1412                 if (!retval)
1413                         retval = p->pid;
1414         } else {
1415                 retval = wait_noreap_copyout(p, pid, uid,
1416                                              CLD_CONTINUED, SIGCONT,
1417                                              infop, ru);
1418                 BUG_ON(retval == 0);
1419         }
1420
1421         return retval;
1422 }
1423
1424
1425 static inline int my_ptrace_child(struct task_struct *p)
1426 {
1427         if (!(p->ptrace & PT_PTRACED))
1428                 return 0;
1429         if (!(p->ptrace & PT_ATTACHED))
1430                 return 1;
1431         /*
1432          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1433          * we are the attacher.  If we are the real parent, this is a race
1434          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1435          * which we have to switch the parent links, but has already set
1436          * the flags in p->ptrace.
1437          */
1438         return (p->parent != p->real_parent);
1439 }
1440
1441 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1442                     int __user *stat_addr, struct rusage __user *ru)
1443 {
1444         DECLARE_WAITQUEUE(wait, current);
1445         struct task_struct *tsk;
1446         int flag, retval;
1447
1448         add_wait_queue(&current->signal->wait_chldexit,&wait);
1449 repeat:
1450         /*
1451          * We will set this flag if we see any child that might later
1452          * match our criteria, even if we are not able to reap it yet.
1453          */
1454         flag = 0;
1455         current->state = TASK_INTERRUPTIBLE;
1456         read_lock(&tasklist_lock);
1457         tsk = current;
1458         do {
1459                 struct task_struct *p;
1460                 struct list_head *_p;
1461                 int ret;
1462
1463                 list_for_each(_p,&tsk->children) {
1464                         p = list_entry(_p,struct task_struct,sibling);
1465
1466                         ret = eligible_child(pid, options, p);
1467                         if (!ret)
1468                                 continue;
1469
1470                         switch (p->state) {
1471                         case TASK_TRACED:
1472                                 /*
1473                                  * When we hit the race with PTRACE_ATTACH,
1474                                  * we will not report this child.  But the
1475                                  * race means it has not yet been moved to
1476                                  * our ptrace_children list, so we need to
1477                                  * set the flag here to avoid a spurious ECHILD
1478                                  * when the race happens with the only child.
1479                                  */
1480                                 flag = 1;
1481                                 if (!my_ptrace_child(p))
1482                                         continue;
1483                                 /*FALLTHROUGH*/
1484                         case TASK_STOPPED:
1485                                 /*
1486                                  * It's stopped now, so it might later
1487                                  * continue, exit, or stop again.
1488                                  */
1489                                 flag = 1;
1490                                 if (!(options & WUNTRACED) &&
1491                                     !my_ptrace_child(p))
1492                                         continue;
1493                                 retval = wait_task_stopped(p, ret == 2,
1494                                                            (options & WNOWAIT),
1495                                                            infop,
1496                                                            stat_addr, ru);
1497                                 if (retval == -EAGAIN)
1498                                         goto repeat;
1499                                 if (retval != 0) /* He released the lock.  */
1500                                         goto end;
1501                                 break;
1502                         default:
1503                         // case EXIT_DEAD:
1504                                 if (p->exit_state == EXIT_DEAD)
1505                                         continue;
1506                         // case EXIT_ZOMBIE:
1507                                 if (p->exit_state == EXIT_ZOMBIE) {
1508                                         /*
1509                                          * Eligible but we cannot release
1510                                          * it yet:
1511                                          */
1512                                         if (ret == 2)
1513                                                 goto check_continued;
1514                                         if (!likely(options & WEXITED))
1515                                                 continue;
1516                                         retval = wait_task_zombie(
1517                                                 p, (options & WNOWAIT),
1518                                                 infop, stat_addr, ru);
1519                                         /* He released the lock.  */
1520                                         if (retval != 0)
1521                                                 goto end;
1522                                         break;
1523                                 }
1524 check_continued:
1525                                 /*
1526                                  * It's running now, so it might later
1527                                  * exit, stop, or stop and then continue.
1528                                  */
1529                                 flag = 1;
1530                                 if (!unlikely(options & WCONTINUED))
1531                                         continue;
1532                                 retval = wait_task_continued(
1533                                         p, (options & WNOWAIT),
1534                                         infop, stat_addr, ru);
1535                                 if (retval != 0) /* He released the lock.  */
1536                                         goto end;
1537                                 break;
1538                         }
1539                 }
1540                 if (!flag) {
1541                         list_for_each(_p, &tsk->ptrace_children) {
1542                                 p = list_entry(_p, struct task_struct,
1543                                                 ptrace_list);
1544                                 if (!eligible_child(pid, options, p))
1545                                         continue;
1546                                 flag = 1;
1547                                 break;
1548                         }
1549                 }
1550                 if (options & __WNOTHREAD)
1551                         break;
1552                 tsk = next_thread(tsk);
1553                 if (tsk->signal != current->signal)
1554                         BUG();
1555         } while (tsk != current);
1556
1557         read_unlock(&tasklist_lock);
1558         if (flag) {
1559                 retval = 0;
1560                 if (options & WNOHANG)
1561                         goto end;
1562                 retval = -ERESTARTSYS;
1563                 if (signal_pending(current))
1564                         goto end;
1565                 schedule();
1566                 goto repeat;
1567         }
1568         retval = -ECHILD;
1569 end:
1570         current->state = TASK_RUNNING;
1571         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1572         if (infop) {
1573                 if (retval > 0)
1574                 retval = 0;
1575                 else {
1576                         /*
1577                          * For a WNOHANG return, clear out all the fields
1578                          * we would set so the user can easily tell the
1579                          * difference.
1580                          */
1581                         if (!retval)
1582                                 retval = put_user(0, &infop->si_signo);
1583                         if (!retval)
1584                                 retval = put_user(0, &infop->si_errno);
1585                         if (!retval)
1586                                 retval = put_user(0, &infop->si_code);
1587                         if (!retval)
1588                                 retval = put_user(0, &infop->si_pid);
1589                         if (!retval)
1590                                 retval = put_user(0, &infop->si_uid);
1591                         if (!retval)
1592                                 retval = put_user(0, &infop->si_status);
1593                 }
1594         }
1595         return retval;
1596 }
1597
1598 asmlinkage long sys_waitid(int which, pid_t pid,
1599                            struct siginfo __user *infop, int options,
1600                            struct rusage __user *ru)
1601 {
1602         long ret;
1603
1604         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1605                 return -EINVAL;
1606         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1607                 return -EINVAL;
1608
1609         switch (which) {
1610         case P_ALL:
1611                 pid = -1;
1612                 break;
1613         case P_PID:
1614                 if (pid <= 0)
1615                         return -EINVAL;
1616                 break;
1617         case P_PGID:
1618                 if (pid <= 0)
1619                         return -EINVAL;
1620                 pid = -pid;
1621                 break;
1622         default:
1623                 return -EINVAL;
1624         }
1625
1626         ret = do_wait(pid, options, infop, NULL, ru);
1627
1628         /* avoid REGPARM breakage on x86: */
1629         prevent_tail_call(ret);
1630         return ret;
1631 }
1632
1633 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1634                           int options, struct rusage __user *ru)
1635 {
1636         long ret;
1637
1638         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1639                         __WNOTHREAD|__WCLONE|__WALL))
1640                 return -EINVAL;
1641         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1642
1643         /* avoid REGPARM breakage on x86: */
1644         prevent_tail_call(ret);
1645         return ret;
1646 }
1647
1648 #ifdef __ARCH_WANT_SYS_WAITPID
1649
1650 /*
1651  * sys_waitpid() remains for compatibility. waitpid() should be
1652  * implemented by calling sys_wait4() from libc.a.
1653  */
1654 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1655 {
1656         return sys_wait4(pid, stat_addr, options, NULL);
1657 }
1658
1659 #endif