kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/mm.h>
   8 #include <linux/slab.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/module.h>
  11 #include <linux/capability.h>
  12 #include <linux/completion.h>
  13 #include <linux/personality.h>
  14 #include <linux/tty.h>
  15 #include <linux/mnt_namespace.h>
  16 #include <linux/key.h>
  17 #include <linux/security.h>
  18 #include <linux/cpu.h>
  19 #include <linux/acct.h>
  20 #include <linux/tsacct_kern.h>
  21 #include <linux/file.h>
  22 #include <linux/binfmts.h>
  23 #include <linux/nsproxy.h>
  24 #include <linux/pid_namespace.h>
  25 #include <linux/ptrace.h>
  26 #include <linux/profile.h>
  27 #include <linux/mount.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/kthread.h>
  30 #include <linux/mempolicy.h>
  31 #include <linux/taskstats_kern.h>
  32 #include <linux/delayacct.h>
  33 #include <linux/freezer.h>
  34 #include <linux/cpuset.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/signal.h>
  37 #include <linux/posix-timers.h>
  38 #include <linux/cn_proc.h>
  39 #include <linux/mutex.h>
  40 #include <linux/futex.h>
  41 #include <linux/compat.h>
  42 #include <linux/pipe_fs_i.h>
  43 #include <linux/audit.h> /* for audit_free() */
  44 #include <linux/resource.h>
  45 #include <linux/blkdev.h>
  46 #include <linux/task_io_accounting_ops.h>
  47
  48 #include <asm/uaccess.h>
  49 #include <asm/unistd.h>
  50 #include <asm/pgtable.h>
  51 #include <asm/mmu_context.h>
  52
  53 extern void sem_exit (void);
  54
  55 static void exit_mm(struct task_struct * tsk);
  56
  57 static void __unhash_process(struct task_struct *p)
  58 {
  59         nr_threads--;
  60         detach_pid(p, PIDTYPE_PID);
  61         if (thread_group_leader(p)) {
  62                 detach_pid(p, PIDTYPE_PGID);
  63                 detach_pid(p, PIDTYPE_SID);
  64
  65                 list_del_rcu(&p->tasks);
  66                 __get_cpu_var(process_counts)--;
  67         }
  68         list_del_rcu(&p->thread_group);
  69         remove_parent(p);
  70 }
  71
  72 /*
  73  * This function expects the tasklist_lock write-locked.
  74  */
  75 static void __exit_signal(struct task_struct *tsk)
  76 {
  77         struct signal_struct *sig = tsk->signal;
  78         struct sighand_struct *sighand;
  79
  80         BUG_ON(!sig);
  81         BUG_ON(!atomic_read(&sig->count));
  82
  83         rcu_read_lock();
  84         sighand = rcu_dereference(tsk->sighand);
  85         spin_lock(&sighand->siglock);
  86
  87         posix_cpu_timers_exit(tsk);
  88         if (atomic_dec_and_test(&sig->count))
  89                 posix_cpu_timers_exit_group(tsk);
  90         else {
  91                 /*
  92                  * If there is any task waiting for the group exit
  93                  * then notify it:
  94                  */
  95                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
  96                         wake_up_process(sig->group_exit_task);
  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->gtime = cputime_add(sig->gtime, tsk->gtime);
 113                 sig->min_flt += tsk->min_flt;
 114                 sig->maj_flt += tsk->maj_flt;
 115                 sig->nvcsw += tsk->nvcsw;
 116                 sig->nivcsw += tsk->nivcsw;
 117                 sig->inblock += task_io_get_inblock(tsk);
 118                 sig->oublock += task_io_get_oublock(tsk);
 119                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 120                 sig = NULL; /* Marker for below. */
 121         }
 122
 123         __unhash_process(tsk);
 124
 125         tsk->signal = NULL;
 126         tsk->sighand = NULL;
 127         spin_unlock(&sighand->siglock);
 128         rcu_read_unlock();
 129
 130         __cleanup_sighand(sighand);
 131         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 132         flush_sigqueue(&tsk->pending);
 133         if (sig) {
 134                 flush_sigqueue(&sig->shared_pending);
 135                 taskstats_tgid_free(sig);
 136                 __cleanup_signal(sig);
 137         }
 138 }
 139
 140 static void delayed_put_task_struct(struct rcu_head *rhp)
 141 {
 142         put_task_struct(container_of(rhp, struct task_struct, rcu));
 143 }
 144
 145 void release_task(struct task_struct * p)
 146 {
 147         struct task_struct *leader;
 148         int zap_leader;
 149 repeat:
 150         atomic_dec(&p->user->processes);
 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         write_unlock_irq(&tasklist_lock);
 178         proc_flush_task(p);
 179         release_thread(p);
 180         call_rcu(&p->rcu, delayed_put_task_struct);
 181
 182         p = leader;
 183         if (unlikely(zap_leader))
 184                 goto repeat;
 185 }
 186
 187 /*
 188  * This checks not only the pgrp, but falls back on the pid if no
 189  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 190  * without this...
 191  *
 192  * The caller must hold rcu lock or the tasklist lock.
 193  */
 194 struct pid *session_of_pgrp(struct pid *pgrp)
 195 {
 196         struct task_struct *p;
 197         struct pid *sid = NULL;
 198
 199         p = pid_task(pgrp, PIDTYPE_PGID);
 200         if (p == NULL)
 201                 p = pid_task(pgrp, PIDTYPE_PID);
 202         if (p != NULL)
 203                 sid = task_session(p);
 204
 205         return sid;
 206 }
 207
 208 /*
 209  * Determine if a process group is "orphaned", according to the POSIX
 210  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 211  * by terminal-generated stop signals.  Newly orphaned process groups are
 212  * to receive a SIGHUP and a SIGCONT.
 213  *
 214  * "I ask you, have you ever known what it is to be an orphan?"
 215  */
 216 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 217 {
 218         struct task_struct *p;
 219         int ret = 1;
 220
 221         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 222                 if (p == ignored_task
 223                                 || p->exit_state
 224                                 || is_init(p->real_parent))
 225                         continue;
 226                 if (task_pgrp(p->real_parent) != pgrp &&
 227                     task_session(p->real_parent) == task_session(p)) {
 228                         ret = 0;
 229                         break;
 230                 }
 231         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 232         return ret;     /* (sighing) "Often!" */
 233 }
 234
 235 int is_current_pgrp_orphaned(void)
 236 {
 237         int retval;
 238
 239         read_lock(&tasklist_lock);
 240         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 241         read_unlock(&tasklist_lock);
 242
 243         return retval;
 244 }
 245
 246 static int has_stopped_jobs(struct pid *pgrp)
 247 {
 248         int retval = 0;
 249         struct task_struct *p;
 250
 251         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 252                 if (p->state != TASK_STOPPED)
 253                         continue;
 254                 retval = 1;
 255                 break;
 256         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 257         return retval;
 258 }
 259
 260 /**
 261  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
 262  *
 263  * If a kernel thread is launched as a result of a system call, or if
 264  * it ever exits, it should generally reparent itself to kthreadd so it
 265  * isn't in the way of other processes and is correctly cleaned up on exit.
 266  *
 267  * The various task state such as scheduling policy and priority may have
 268  * been inherited from a user process, so we reset them to sane values here.
 269  *
 270  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
 271  */
 272 static void reparent_to_kthreadd(void)
 273 {
 274         write_lock_irq(&tasklist_lock);
 275
 276         ptrace_unlink(current);
 277         /* Reparent to init */
 278         remove_parent(current);
 279         current->real_parent = current->parent = kthreadd_task;
 280         add_parent(current);
 281
 282         /* Set the exit signal to SIGCHLD so we signal init on exit */
 283         current->exit_signal = SIGCHLD;
 284
 285         if (task_nice(current) < 0)
 286                 set_user_nice(current, 0);
 287         /* cpus_allowed? */
 288         /* rt_priority? */
 289         /* signals? */
 290         security_task_reparent_to_init(current);
 291         memcpy(current->signal->rlim, init_task.signal->rlim,
 292                sizeof(current->signal->rlim));
 293         atomic_inc(&(INIT_USER->__count));
 294         write_unlock_irq(&tasklist_lock);
 295         switch_uid(INIT_USER);
 296 }
 297
 298 void __set_special_pids(pid_t session, pid_t pgrp)
 299 {
 300         struct task_struct *curr = current->group_leader;
 301
 302         if (process_session(curr) != session) {
 303                 detach_pid(curr, PIDTYPE_SID);
 304                 set_signal_session(curr->signal, session);
 305                 attach_pid(curr, PIDTYPE_SID, find_pid(session));
 306         }
 307         if (process_group(curr) != pgrp) {
 308                 detach_pid(curr, PIDTYPE_PGID);
 309                 curr->signal->pgrp = pgrp;
 310                 attach_pid(curr, PIDTYPE_PGID, find_pid(pgrp));
 311         }
 312 }
 313
 314 static void set_special_pids(pid_t session, pid_t pgrp)
 315 {
 316         write_lock_irq(&tasklist_lock);
 317         __set_special_pids(session, pgrp);
 318         write_unlock_irq(&tasklist_lock);
 319 }
 320
 321 /*
 322  * Let kernel threads use this to say that they
 323  * allow a certain signal (since daemonize() will
 324  * have disabled all of them by default).
 325  */
 326 int allow_signal(int sig)
 327 {
 328         if (!valid_signal(sig) || sig < 1)
 329                 return -EINVAL;
 330
 331         spin_lock_irq(&current->sighand->siglock);
 332         sigdelset(&current->blocked, sig);
 333         if (!current->mm) {
 334                 /* Kernel threads handle their own signals.
 335                    Let the signal code know it'll be handled, so
 336                    that they don't get converted to SIGKILL or
 337                    just silently dropped */
 338                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 339         }
 340         recalc_sigpending();
 341         spin_unlock_irq(&current->sighand->siglock);
 342         return 0;
 343 }
 344
 345 EXPORT_SYMBOL(allow_signal);
 346
 347 int disallow_signal(int sig)
 348 {
 349         if (!valid_signal(sig) || sig < 1)
 350                 return -EINVAL;
 351
 352         spin_lock_irq(&current->sighand->siglock);
 353         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
 354         recalc_sigpending();
 355         spin_unlock_irq(&current->sighand->siglock);
 356         return 0;
 357 }
 358
 359 EXPORT_SYMBOL(disallow_signal);
 360
 361 /*
 362  *      Put all the gunge required to become a kernel thread without
 363  *      attached user resources in one place where it belongs.
 364  */
 365
 366 void daemonize(const char *name, ...)
 367 {
 368         va_list args;
 369         struct fs_struct *fs;
 370         sigset_t blocked;
 371
 372         va_start(args, name);
 373         vsnprintf(current->comm, sizeof(current->comm), name, args);
 374         va_end(args);
 375
 376         /*
 377          * If we were started as result of loading a module, close all of the
 378          * user space pages.  We don't need them, and if we didn't close them
 379          * they would be locked into memory.
 380          */
 381         exit_mm(current);
 382         /*
 383          * We don't want to have TIF_FREEZE set if the system-wide hibernation
 384          * or suspend transition begins right now.
 385          */
 386         current->flags |= PF_NOFREEZE;
 387
 388         set_special_pids(1, 1);
 389         proc_clear_tty(current);
 390
 391         /* Block and flush all signals */
 392         sigfillset(&blocked);
 393         sigprocmask(SIG_BLOCK, &blocked, NULL);
 394         flush_signals(current);
 395
 396         /* Become as one with the init task */
 397
 398         exit_fs(current);       /* current->fs->count--; */
 399         fs = init_task.fs;
 400         current->fs = fs;
 401         atomic_inc(&fs->count);
 402
 403         exit_task_namespaces(current);
 404         current->nsproxy = init_task.nsproxy;
 405         get_task_namespaces(current);
 406
 407         exit_files(current);
 408         current->files = init_task.files;
 409         atomic_inc(&current->files->count);
 410
 411         reparent_to_kthreadd();
 412 }
 413
 414 EXPORT_SYMBOL(daemonize);
 415
 416 static void close_files(struct files_struct * files)
 417 {
 418         int i, j;
 419         struct fdtable *fdt;
 420
 421         j = 0;
 422
 423         /*
 424          * It is safe to dereference the fd table without RCU or
 425          * ->file_lock because this is the last reference to the
 426          * files structure.
 427          */
 428         fdt = files_fdtable(files);
 429         for (;;) {
 430                 unsigned long set;
 431                 i = j * __NFDBITS;
 432                 if (i >= fdt->max_fds)
 433                         break;
 434                 set = fdt->open_fds->fds_bits[j++];
 435                 while (set) {
 436                         if (set & 1) {
 437                                 struct file * file = xchg(&fdt->fd[i], NULL);
 438                                 if (file) {
 439                                         filp_close(file, files);
 440                                         cond_resched();
 441                                 }
 442                         }
 443                         i++;
 444                         set >>= 1;
 445                 }
 446         }
 447 }
 448
 449 struct files_struct *get_files_struct(struct task_struct *task)
 450 {
 451         struct files_struct *files;
 452
 453         task_lock(task);
 454         files = task->files;
 455         if (files)
 456                 atomic_inc(&files->count);
 457         task_unlock(task);
 458
 459         return files;
 460 }
 461
 462 void fastcall put_files_struct(struct files_struct *files)
 463 {
 464         struct fdtable *fdt;
 465
 466         if (atomic_dec_and_test(&files->count)) {
 467                 close_files(files);
 468                 /*
 469                  * Free the fd and fdset arrays if we expanded them.
 470                  * If the fdtable was embedded, pass files for freeing
 471                  * at the end of the RCU grace period. Otherwise,
 472                  * you can free files immediately.
 473                  */
 474                 fdt = files_fdtable(files);
 475                 if (fdt != &files->fdtab)
 476                         kmem_cache_free(files_cachep, files);
 477                 free_fdtable(fdt);
 478         }
 479 }
 480
 481 EXPORT_SYMBOL(put_files_struct);
 482
 483 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
 484 {
 485         struct files_struct *old;
 486
 487         old = tsk->files;
 488         task_lock(tsk);
 489         tsk->files = files;
 490         task_unlock(tsk);
 491         put_files_struct(old);
 492 }
 493 EXPORT_SYMBOL(reset_files_struct);
 494
 495 static inline void __exit_files(struct task_struct *tsk)
 496 {
 497         struct files_struct * files = tsk->files;
 498
 499         if (files) {
 500                 task_lock(tsk);
 501                 tsk->files = NULL;
 502                 task_unlock(tsk);
 503                 put_files_struct(files);
 504         }
 505 }
 506
 507 void exit_files(struct task_struct *tsk)
 508 {
 509         __exit_files(tsk);
 510 }
 511
 512 static inline void __put_fs_struct(struct fs_struct *fs)
 513 {
 514         /* No need to hold fs->lock if we are killing it */
 515         if (atomic_dec_and_test(&fs->count)) {
 516                 dput(fs->root);
 517                 mntput(fs->rootmnt);
 518                 dput(fs->pwd);
 519                 mntput(fs->pwdmnt);
 520                 if (fs->altroot) {
 521                         dput(fs->altroot);
 522                         mntput(fs->altrootmnt);
 523                 }
 524                 kmem_cache_free(fs_cachep, fs);
 525         }
 526 }
 527
 528 void put_fs_struct(struct fs_struct *fs)
 529 {
 530         __put_fs_struct(fs);
 531 }
 532
 533 static inline void __exit_fs(struct task_struct *tsk)
 534 {
 535         struct fs_struct * fs = tsk->fs;
 536
 537         if (fs) {
 538                 task_lock(tsk);
 539                 tsk->fs = NULL;
 540                 task_unlock(tsk);
 541                 __put_fs_struct(fs);
 542         }
 543 }
 544
 545 void exit_fs(struct task_struct *tsk)
 546 {
 547         __exit_fs(tsk);
 548 }
 549
 550 EXPORT_SYMBOL_GPL(exit_fs);
 551
 552 /*
 553  * Turn us into a lazy TLB process if we
 554  * aren't already..
 555  */
 556 static void exit_mm(struct task_struct * tsk)
 557 {
 558         struct mm_struct *mm = tsk->mm;
 559
 560         mm_release(tsk, mm);
 561         if (!mm)
 562                 return;
 563         /*
 564          * Serialize with any possible pending coredump.
 565          * We must hold mmap_sem around checking core_waiters
 566          * and clearing tsk->mm.  The core-inducing thread
 567          * will increment core_waiters for each thread in the
 568          * group with ->mm != NULL.
 569          */
 570         down_read(&mm->mmap_sem);
 571         if (mm->core_waiters) {
 572                 up_read(&mm->mmap_sem);
 573                 down_write(&mm->mmap_sem);
 574                 if (!--mm->core_waiters)
 575                         complete(mm->core_startup_done);
 576                 up_write(&mm->mmap_sem);
 577
 578                 wait_for_completion(&mm->core_done);
 579                 down_read(&mm->mmap_sem);
 580         }
 581         atomic_inc(&mm->mm_count);
 582         BUG_ON(mm != tsk->active_mm);
 583         /* more a memory barrier than a real lock */
 584         task_lock(tsk);
 585         tsk->mm = NULL;
 586         up_read(&mm->mmap_sem);
 587         enter_lazy_tlb(mm, current);
 588         /* We don't want this task to be frozen prematurely */
 589         clear_freeze_flag(tsk);
 590         task_unlock(tsk);
 591         mmput(mm);
 592 }
 593
 594 static void
 595 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 596 {
 597         if (p->pdeath_signal)
 598                 /* We already hold the tasklist_lock here.  */
 599                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 600
 601         /* Move the child from its dying parent to the new one.  */
 602         if (unlikely(traced)) {
 603                 /* Preserve ptrace links if someone else is tracing this child.  */
 604                 list_del_init(&p->ptrace_list);
 605                 if (p->parent != p->real_parent)
 606                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 607         } else {
 608                 /* If this child is being traced, then we're the one tracing it
 609                  * anyway, so let go of it.
 610                  */
 611                 p->ptrace = 0;
 612                 remove_parent(p);
 613                 p->parent = p->real_parent;
 614                 add_parent(p);
 615
 616                 if (p->state == TASK_TRACED) {
 617                         /*
 618                          * If it was at a trace stop, turn it into
 619                          * a normal stop since it's no longer being
 620                          * traced.
 621                          */
 622                         ptrace_untrace(p);
 623                 }
 624         }
 625
 626         /* If this is a threaded reparent there is no need to
 627          * notify anyone anything has happened.
 628          */
 629         if (p->real_parent->group_leader == father->group_leader)
 630                 return;
 631
 632         /* We don't want people slaying init.  */
 633         if (p->exit_signal != -1)
 634                 p->exit_signal = SIGCHLD;
 635
 636         /* If we'd notified the old parent about this child's death,
 637          * also notify the new parent.
 638          */
 639         if (!traced && p->exit_state == EXIT_ZOMBIE &&
 640             p->exit_signal != -1 && thread_group_empty(p))
 641                 do_notify_parent(p, p->exit_signal);
 642
 643         /*
 644          * process group orphan check
 645          * Case ii: Our child is in a different pgrp
 646          * than we are, and it was the only connection
 647          * outside, so the child pgrp is now orphaned.
 648          */
 649         if ((task_pgrp(p) != task_pgrp(father)) &&
 650             (task_session(p) == task_session(father))) {
 651                 struct pid *pgrp = task_pgrp(p);
 652
 653                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
 654                     has_stopped_jobs(pgrp)) {
 655                         __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 656                         __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 657                 }
 658         }
 659 }
 660
 661 /*
 662  * When we die, we re-parent all our children.
 663  * Try to give them to another thread in our thread
 664  * group, and if no such member exists, give it to
 665  * the child reaper process (ie "init") in our pid
 666  * space.
 667  */
 668 static void
 669 forget_original_parent(struct task_struct *father, struct list_head *to_release)
 670 {
 671         struct task_struct *p, *reaper = father;
 672         struct list_head *_p, *_n;
 673
 674         do {
 675                 reaper = next_thread(reaper);
 676                 if (reaper == father) {
 677                         reaper = child_reaper(father);
 678                         break;
 679                 }
 680         } while (reaper->exit_state);
 681
 682         /*
 683          * There are only two places where our children can be:
 684          *
 685          * - in our child list
 686          * - in our ptraced child list
 687          *
 688          * Search them and reparent children.
 689          */
 690         list_for_each_safe(_p, _n, &father->children) {
 691                 int ptrace;
 692                 p = list_entry(_p, struct task_struct, sibling);
 693
 694                 ptrace = p->ptrace;
 695
 696                 /* if father isn't the real parent, then ptrace must be enabled */
 697                 BUG_ON(father != p->real_parent && !ptrace);
 698
 699                 if (father == p->real_parent) {
 700                         /* reparent with a reaper, real father it's us */
 701                         p->real_parent = reaper;
 702                         reparent_thread(p, father, 0);
 703                 } else {
 704                         /* reparent ptraced task to its real parent */
 705                         __ptrace_unlink (p);
 706                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 707                             thread_group_empty(p))
 708                                 do_notify_parent(p, p->exit_signal);
 709                 }
 710
 711                 /*
 712                  * if the ptraced child is a zombie with exit_signal == -1
 713                  * we must collect it before we exit, or it will remain
 714                  * zombie forever since we prevented it from self-reap itself
 715                  * while it was being traced by us, to be able to see it in wait4.
 716                  */
 717                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 718                         list_add(&p->ptrace_list, to_release);
 719         }
 720         list_for_each_safe(_p, _n, &father->ptrace_children) {
 721                 p = list_entry(_p, struct task_struct, ptrace_list);
 722                 p->real_parent = reaper;
 723                 reparent_thread(p, father, 1);
 724         }
 725 }
 726
 727 /*
 728  * Send signals to all our closest relatives so that they know
 729  * to properly mourn us..
 730  */
 731 static void exit_notify(struct task_struct *tsk)
 732 {
 733         int state;
 734         struct task_struct *t;
 735         struct list_head ptrace_dead, *_p, *_n;
 736         struct pid *pgrp;
 737
 738         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 739             && !thread_group_empty(tsk)) {
 740                 /*
 741                  * This occurs when there was a race between our exit
 742                  * syscall and a group signal choosing us as the one to
 743                  * wake up.  It could be that we are the only thread
 744                  * alerted to check for pending signals, but another thread
 745                  * should be woken now to take the signal since we will not.
 746                  * Now we'll wake all the threads in the group just to make
 747                  * sure someone gets all the pending signals.
 748                  */
 749                 spin_lock_irq(&tsk->sighand->siglock);
 750                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 751                         if (!signal_pending(t) && !(t->flags & PF_EXITING))
 752                                 recalc_sigpending_and_wake(t);
 753                 spin_unlock_irq(&tsk->sighand->siglock);
 754         }
 755
 756         write_lock_irq(&tasklist_lock);
 757
 758         /*
 759          * This does two things:
 760          *
 761          * A.  Make init inherit all the child processes
 762          * B.  Check to see if any process groups have become orphaned
 763          *      as a result of our exiting, and if they have any stopped
 764          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 765          */
 766
 767         INIT_LIST_HEAD(&ptrace_dead);
 768         forget_original_parent(tsk, &ptrace_dead);
 769         BUG_ON(!list_empty(&tsk->children));
 770         BUG_ON(!list_empty(&tsk->ptrace_children));
 771
 772         /*
 773          * Check to see if any process groups have become orphaned
 774          * as a result of our exiting, and if they have any stopped
 775          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 776          *
 777          * Case i: Our father is in a different pgrp than we are
 778          * and we were the only connection outside, so our pgrp
 779          * is about to become orphaned.
 780          */
 781         t = tsk->real_parent;
 782
 783         pgrp = task_pgrp(tsk);
 784         if ((task_pgrp(t) != pgrp) &&
 785             (task_session(t) == task_session(tsk)) &&
 786             will_become_orphaned_pgrp(pgrp, tsk) &&
 787             has_stopped_jobs(pgrp)) {
 788                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 789                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 790         }
 791
 792         /* Let father know we died
 793          *
 794          * Thread signals are configurable, but you aren't going to use
 795          * that to send signals to arbitary processes.
 796          * That stops right now.
 797          *
 798          * If the parent exec id doesn't match the exec id we saved
 799          * when we started then we know the parent has changed security
 800          * domain.
 801          *
 802          * If our self_exec id doesn't match our parent_exec_id then
 803          * we have changed execution domain as these two values started
 804          * the same after a fork.
 805          */
 806         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 807             ( tsk->parent_exec_id != t->self_exec_id  ||
 808               tsk->self_exec_id != tsk->parent_exec_id)
 809             && !capable(CAP_KILL))
 810                 tsk->exit_signal = SIGCHLD;
 811
 812
 813         /* If something other than our normal parent is ptracing us, then
 814          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 815          * only has special meaning to our real parent.
 816          */
 817         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 818                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 819                 do_notify_parent(tsk, signal);
 820         } else if (tsk->ptrace) {
 821                 do_notify_parent(tsk, SIGCHLD);
 822         }
 823
 824         state = EXIT_ZOMBIE;
 825         if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
 826                 state = EXIT_DEAD;
 827         tsk->exit_state = state;
 828
 829         if (thread_group_leader(tsk) &&
 830             tsk->signal->notify_count < 0 &&
 831             tsk->signal->group_exit_task)
 832                 wake_up_process(tsk->signal->group_exit_task);
 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 #ifdef CONFIG_DEBUG_STACK_USAGE
 848 static void check_stack_usage(void)
 849 {
 850         static DEFINE_SPINLOCK(low_water_lock);
 851         static int lowest_to_date = THREAD_SIZE;
 852         unsigned long *n = end_of_stack(current);
 853         unsigned long free;
 854
 855         while (*n == 0)
 856                 n++;
 857         free = (unsigned long)n - (unsigned long)end_of_stack(current);
 858
 859         if (free >= lowest_to_date)
 860                 return;
 861
 862         spin_lock(&low_water_lock);
 863         if (free < lowest_to_date) {
 864                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
 865                                 "left\n",
 866                                 current->comm, free);
 867                 lowest_to_date = free;
 868         }
 869         spin_unlock(&low_water_lock);
 870 }
 871 #else
 872 static inline void check_stack_usage(void) {}
 873 #endif
 874
 875 static inline void exit_child_reaper(struct task_struct *tsk)
 876 {
 877         if (likely(tsk->group_leader != child_reaper(tsk)))
 878                 return;
 879
 880         panic("Attempted to kill init!");
 881 }
 882
 883 fastcall NORET_TYPE void do_exit(long code)
 884 {
 885         struct task_struct *tsk = current;
 886         int group_dead;
 887
 888         profile_task_exit(tsk);
 889
 890         WARN_ON(atomic_read(&tsk->fs_excl));
 891
 892         if (unlikely(in_interrupt()))
 893                 panic("Aiee, killing interrupt handler!");
 894         if (unlikely(!tsk->pid))
 895                 panic("Attempted to kill the idle task!");
 896
 897         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 898                 current->ptrace_message = code;
 899                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 900         }
 901
 902         /*
 903          * We're taking recursive faults here in do_exit. Safest is to just
 904          * leave this task alone and wait for reboot.
 905          */
 906         if (unlikely(tsk->flags & PF_EXITING)) {
 907                 printk(KERN_ALERT
 908                         "Fixing recursive fault but reboot is needed!\n");
 909                 /*
 910                  * We can do this unlocked here. The futex code uses
 911                  * this flag just to verify whether the pi state
 912                  * cleanup has been done or not. In the worst case it
 913                  * loops once more. We pretend that the cleanup was
 914                  * done as there is no way to return. Either the
 915                  * OWNER_DIED bit is set by now or we push the blocked
 916                  * task into the wait for ever nirwana as well.
 917                  */
 918                 tsk->flags |= PF_EXITPIDONE;
 919                 if (tsk->io_context)
 920                         exit_io_context();
 921                 set_current_state(TASK_UNINTERRUPTIBLE);
 922                 schedule();
 923         }
 924
 925         tsk->flags |= PF_EXITING;
 926         /*
 927          * tsk->flags are checked in the futex code to protect against
 928          * an exiting task cleaning up the robust pi futexes.
 929          */
 930         smp_mb();
 931         spin_unlock_wait(&tsk->pi_lock);
 932
 933         if (unlikely(in_atomic()))
 934                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 935                                 current->comm, current->pid,
 936                                 preempt_count());
 937
 938         acct_update_integrals(tsk);
 939         if (tsk->mm) {
 940                 update_hiwater_rss(tsk->mm);
 941                 update_hiwater_vm(tsk->mm);
 942         }
 943         group_dead = atomic_dec_and_test(&tsk->signal->live);
 944         if (group_dead) {
 945                 exit_child_reaper(tsk);
 946                 hrtimer_cancel(&tsk->signal->real_timer);
 947                 exit_itimers(tsk->signal);
 948         }
 949         acct_collect(code, group_dead);
 950         if (unlikely(tsk->robust_list))
 951                 exit_robust_list(tsk);
 952 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 953         if (unlikely(tsk->compat_robust_list))
 954                 compat_exit_robust_list(tsk);
 955 #endif
 956         if (group_dead)
 957                 tty_audit_exit();
 958         if (unlikely(tsk->audit_context))
 959                 audit_free(tsk);
 960
 961         tsk->exit_code = code;
 962         taskstats_exit(tsk, group_dead);
 963
 964         exit_mm(tsk);
 965
 966         if (group_dead)
 967                 acct_process();
 968         exit_sem(tsk);
 969         __exit_files(tsk);
 970         __exit_fs(tsk);
 971         check_stack_usage();
 972         exit_thread();
 973         cpuset_exit(tsk);
 974         exit_keys(tsk);
 975
 976         if (group_dead && tsk->signal->leader)
 977                 disassociate_ctty(1);
 978
 979         module_put(task_thread_info(tsk)->exec_domain->module);
 980         if (tsk->binfmt)
 981                 module_put(tsk->binfmt->module);
 982
 983         proc_exit_connector(tsk);
 984         exit_task_namespaces(tsk);
 985         exit_notify(tsk);
 986 #ifdef CONFIG_NUMA
 987         mpol_free(tsk->mempolicy);
 988         tsk->mempolicy = NULL;
 989 #endif
 990         /*
 991          * This must happen late, after the PID is not
 992          * hashed anymore:
 993          */
 994         if (unlikely(!list_empty(&tsk->pi_state_list)))
 995                 exit_pi_state_list(tsk);
 996         if (unlikely(current->pi_state_cache))
 997                 kfree(current->pi_state_cache);
 998         /*
 999          * Make sure we are holding no locks:
1000          */
1001         debug_check_no_locks_held(tsk);
1002         /*
1003          * We can do this unlocked here. The futex code uses this flag
1004          * just to verify whether the pi state cleanup has been done
1005          * or not. In the worst case it loops once more.
1006          */
1007         tsk->flags |= PF_EXITPIDONE;
1008
1009         if (tsk->io_context)
1010                 exit_io_context();
1011
1012         if (tsk->splice_pipe)
1013                 __free_pipe_info(tsk->splice_pipe);
1014
1015         preempt_disable();
1016         /* causes final put_task_struct in finish_task_switch(). */
1017         tsk->state = TASK_DEAD;
1018
1019         schedule();
1020         BUG();
1021         /* Avoid "noreturn function does return".  */
1022         for (;;)
1023                 cpu_relax();    /* For when BUG is null */
1024 }
1025
1026 EXPORT_SYMBOL_GPL(do_exit);
1027
1028 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1029 {
1030         if (comp)
1031                 complete(comp);
1032
1033         do_exit(code);
1034 }
1035
1036 EXPORT_SYMBOL(complete_and_exit);
1037
1038 asmlinkage long sys_exit(int error_code)
1039 {
1040         do_exit((error_code&0xff)<<8);
1041 }
1042
1043 /*
1044  * Take down every thread in the group.  This is called by fatal signals
1045  * as well as by sys_exit_group (below).
1046  */
1047 NORET_TYPE void
1048 do_group_exit(int exit_code)
1049 {
1050         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1051
1052         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1053                 exit_code = current->signal->group_exit_code;
1054         else if (!thread_group_empty(current)) {
1055                 struct signal_struct *const sig = current->signal;
1056                 struct sighand_struct *const sighand = current->sighand;
1057                 spin_lock_irq(&sighand->siglock);
1058                 if (sig->flags & SIGNAL_GROUP_EXIT)
1059                         /* Another thread got here before we took the lock.  */
1060                         exit_code = sig->group_exit_code;
1061                 else {
1062                         sig->group_exit_code = exit_code;
1063                         zap_other_threads(current);
1064                 }
1065                 spin_unlock_irq(&sighand->siglock);
1066         }
1067
1068         do_exit(exit_code);
1069         /* NOTREACHED */
1070 }
1071
1072 /*
1073  * this kills every thread in the thread group. Note that any externally
1074  * wait4()-ing process will get the correct exit code - even if this
1075  * thread is not the thread group leader.
1076  */
1077 asmlinkage void sys_exit_group(int error_code)
1078 {
1079         do_group_exit((error_code & 0xff) << 8);
1080 }
1081
1082 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1083 {
1084         int err;
1085
1086         if (pid > 0) {
1087                 if (p->pid != pid)
1088                         return 0;
1089         } else if (!pid) {
1090                 if (process_group(p) != process_group(current))
1091                         return 0;
1092         } else if (pid != -1) {
1093                 if (process_group(p) != -pid)
1094                         return 0;
1095         }
1096
1097         /*
1098          * Do not consider detached threads that are
1099          * not ptraced:
1100          */
1101         if (p->exit_signal == -1 && !p->ptrace)
1102                 return 0;
1103
1104         /* Wait for all children (clone and not) if __WALL is set;
1105          * otherwise, wait for clone children *only* if __WCLONE is
1106          * set; otherwise, wait for non-clone children *only*.  (Note:
1107          * A "clone" child here is one that reports to its parent
1108          * using a signal other than SIGCHLD.) */
1109         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1110             && !(options & __WALL))
1111                 return 0;
1112         /*
1113          * Do not consider thread group leaders that are
1114          * in a non-empty thread group:
1115          */
1116         if (delay_group_leader(p))
1117                 return 2;
1118
1119         err = security_task_wait(p);
1120         if (err)
1121                 return err;
1122
1123         return 1;
1124 }
1125
1126 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1127                                int why, int status,
1128                                struct siginfo __user *infop,
1129                                struct rusage __user *rusagep)
1130 {
1131         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1132
1133         put_task_struct(p);
1134         if (!retval)
1135                 retval = put_user(SIGCHLD, &infop->si_signo);
1136         if (!retval)
1137                 retval = put_user(0, &infop->si_errno);
1138         if (!retval)
1139                 retval = put_user((short)why, &infop->si_code);
1140         if (!retval)
1141                 retval = put_user(pid, &infop->si_pid);
1142         if (!retval)
1143                 retval = put_user(uid, &infop->si_uid);
1144         if (!retval)
1145                 retval = put_user(status, &infop->si_status);
1146         if (!retval)
1147                 retval = pid;
1148         return retval;
1149 }
1150
1151 /*
1152  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1153  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1154  * the lock and this task is uninteresting.  If we return nonzero, we have
1155  * released the lock and the system call should return.
1156  */
1157 static int wait_task_zombie(struct task_struct *p, int noreap,
1158                             struct siginfo __user *infop,
1159                             int __user *stat_addr, struct rusage __user *ru)
1160 {
1161         unsigned long state;
1162         int retval, status, traced;
1163
1164         if (unlikely(noreap)) {
1165                 pid_t pid = p->pid;
1166                 uid_t uid = p->uid;
1167                 int exit_code = p->exit_code;
1168                 int why, status;
1169
1170                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1171                         return 0;
1172                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1173                         return 0;
1174                 get_task_struct(p);
1175                 read_unlock(&tasklist_lock);
1176                 if ((exit_code & 0x7f) == 0) {
1177                         why = CLD_EXITED;
1178                         status = exit_code >> 8;
1179                 } else {
1180                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1181                         status = exit_code & 0x7f;
1182                 }
1183                 return wait_noreap_copyout(p, pid, uid, why,
1184                                            status, infop, ru);
1185         }
1186
1187         /*
1188          * Try to move the task's state to DEAD
1189          * only one thread is allowed to do this:
1190          */
1191         state = xchg(&p->exit_state, EXIT_DEAD);
1192         if (state != EXIT_ZOMBIE) {
1193                 BUG_ON(state != EXIT_DEAD);
1194                 return 0;
1195         }
1196
1197         /* traced means p->ptrace, but not vice versa */
1198         traced = (p->real_parent != p->parent);
1199
1200         if (likely(!traced)) {
1201                 struct signal_struct *psig;
1202                 struct signal_struct *sig;
1203
1204                 /*
1205                  * The resource counters for the group leader are in its
1206                  * own task_struct.  Those for dead threads in the group
1207                  * are in its signal_struct, as are those for the child
1208                  * processes it has previously reaped.  All these
1209                  * accumulate in the parent's signal_struct c* fields.
1210                  *
1211                  * We don't bother to take a lock here to protect these
1212                  * p->signal fields, because they are only touched by
1213                  * __exit_signal, which runs with tasklist_lock
1214                  * write-locked anyway, and so is excluded here.  We do
1215                  * need to protect the access to p->parent->signal fields,
1216                  * as other threads in the parent group can be right
1217                  * here reaping other children at the same time.
1218                  */
1219                 spin_lock_irq(&p->parent->sighand->siglock);
1220                 psig = p->parent->signal;
1221                 sig = p->signal;
1222                 psig->cutime =
1223                         cputime_add(psig->cutime,
1224                         cputime_add(p->utime,
1225                         cputime_add(sig->utime,
1226                                     sig->cutime)));
1227                 psig->cstime =
1228                         cputime_add(psig->cstime,
1229                         cputime_add(p->stime,
1230                         cputime_add(sig->stime,
1231                                     sig->cstime)));
1232                 psig->cgtime =
1233                         cputime_add(psig->cgtime,
1234                         cputime_add(p->gtime,
1235                         cputime_add(sig->gtime,
1236                                     sig->cgtime)));
1237                 psig->cmin_flt +=
1238                         p->min_flt + sig->min_flt + sig->cmin_flt;
1239                 psig->cmaj_flt +=
1240                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1241                 psig->cnvcsw +=
1242                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1243                 psig->cnivcsw +=
1244                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1245                 psig->cinblock +=
1246                         task_io_get_inblock(p) +
1247                         sig->inblock + sig->cinblock;
1248                 psig->coublock +=
1249                         task_io_get_oublock(p) +
1250                         sig->oublock + sig->coublock;
1251                 spin_unlock_irq(&p->parent->sighand->siglock);
1252         }
1253
1254         /*
1255          * Now we are sure this task is interesting, and no other
1256          * thread can reap it because we set its state to EXIT_DEAD.
1257          */
1258         read_unlock(&tasklist_lock);
1259
1260         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1261         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1262                 ? p->signal->group_exit_code : p->exit_code;
1263         if (!retval && stat_addr)
1264                 retval = put_user(status, stat_addr);
1265         if (!retval && infop)
1266                 retval = put_user(SIGCHLD, &infop->si_signo);
1267         if (!retval && infop)
1268                 retval = put_user(0, &infop->si_errno);
1269         if (!retval && infop) {
1270                 int why;
1271
1272                 if ((status & 0x7f) == 0) {
1273                         why = CLD_EXITED;
1274                         status >>= 8;
1275                 } else {
1276                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1277                         status &= 0x7f;
1278                 }
1279                 retval = put_user((short)why, &infop->si_code);
1280                 if (!retval)
1281                         retval = put_user(status, &infop->si_status);
1282         }
1283         if (!retval && infop)
1284                 retval = put_user(p->pid, &infop->si_pid);
1285         if (!retval && infop)
1286                 retval = put_user(p->uid, &infop->si_uid);
1287         if (!retval)
1288                 retval = p->pid;
1289
1290         if (traced) {
1291                 write_lock_irq(&tasklist_lock);
1292                 /* We dropped tasklist, ptracer could die and untrace */
1293                 ptrace_unlink(p);
1294                 /*
1295                  * If this is not a detached task, notify the parent.
1296                  * If it's still not detached after that, don't release
1297                  * it now.
1298                  */
1299                 if (p->exit_signal != -1) {
1300                         do_notify_parent(p, p->exit_signal);
1301                         if (p->exit_signal != -1) {
1302                                 p->exit_state = EXIT_ZOMBIE;
1303                                 p = NULL;
1304                         }
1305                 }
1306                 write_unlock_irq(&tasklist_lock);
1307         }
1308         if (p != NULL)
1309                 release_task(p);
1310
1311         return retval;
1312 }
1313
1314 /*
1315  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1316  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1317  * the lock and this task is uninteresting.  If we return nonzero, we have
1318  * released the lock and the system call should return.
1319  */
1320 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1321                              int noreap, struct siginfo __user *infop,
1322                              int __user *stat_addr, struct rusage __user *ru)
1323 {
1324         int retval, exit_code;
1325
1326         if (!p->exit_code)
1327                 return 0;
1328         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1329             p->signal->group_stop_count > 0)
1330                 /*
1331                  * A group stop is in progress and this is the group leader.
1332                  * We won't report until all threads have stopped.
1333                  */
1334                 return 0;
1335
1336         /*
1337          * Now we are pretty sure this task is interesting.
1338          * Make sure it doesn't get reaped out from under us while we
1339          * give up the lock and then examine it below.  We don't want to
1340          * keep holding onto the tasklist_lock while we call getrusage and
1341          * possibly take page faults for user memory.
1342          */
1343         get_task_struct(p);
1344         read_unlock(&tasklist_lock);
1345
1346         if (unlikely(noreap)) {
1347                 pid_t pid = p->pid;
1348                 uid_t uid = p->uid;
1349                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1350
1351                 exit_code = p->exit_code;
1352                 if (unlikely(!exit_code) ||
1353                     unlikely(p->state & TASK_TRACED))
1354                         goto bail_ref;
1355                 return wait_noreap_copyout(p, pid, uid,
1356                                            why, (exit_code << 8) | 0x7f,
1357                                            infop, ru);
1358         }
1359
1360         write_lock_irq(&tasklist_lock);
1361
1362         /*
1363          * This uses xchg to be atomic with the thread resuming and setting
1364          * it.  It must also be done with the write lock held to prevent a
1365          * race with the EXIT_ZOMBIE case.
1366          */
1367         exit_code = xchg(&p->exit_code, 0);
1368         if (unlikely(p->exit_state)) {
1369                 /*
1370                  * The task resumed and then died.  Let the next iteration
1371                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1372                  * already be zero here if it resumed and did _exit(0).
1373                  * The task itself is dead and won't touch exit_code again;
1374                  * other processors in this function are locked out.
1375                  */
1376                 p->exit_code = exit_code;
1377                 exit_code = 0;
1378         }
1379         if (unlikely(exit_code == 0)) {
1380                 /*
1381                  * Another thread in this function got to it first, or it
1382                  * resumed, or it resumed and then died.
1383                  */
1384                 write_unlock_irq(&tasklist_lock);
1385 bail_ref:
1386                 put_task_struct(p);
1387                 /*
1388                  * We are returning to the wait loop without having successfully
1389                  * removed the process and having released the lock. We cannot
1390                  * continue, since the "p" task pointer is potentially stale.
1391                  *
1392                  * Return -EAGAIN, and do_wait() will restart the loop from the
1393                  * beginning. Do _not_ re-acquire the lock.
1394                  */
1395                 return -EAGAIN;
1396         }
1397
1398         /* move to end of parent's list to avoid starvation */
1399         remove_parent(p);
1400         add_parent(p);
1401
1402         write_unlock_irq(&tasklist_lock);
1403
1404         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1405         if (!retval && stat_addr)
1406                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1407         if (!retval && infop)
1408                 retval = put_user(SIGCHLD, &infop->si_signo);
1409         if (!retval && infop)
1410                 retval = put_user(0, &infop->si_errno);
1411         if (!retval && infop)
1412                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1413                                           ? CLD_TRAPPED : CLD_STOPPED),
1414                                   &infop->si_code);
1415         if (!retval && infop)
1416                 retval = put_user(exit_code, &infop->si_status);
1417         if (!retval && infop)
1418                 retval = put_user(p->pid, &infop->si_pid);
1419         if (!retval && infop)
1420                 retval = put_user(p->uid, &infop->si_uid);
1421         if (!retval)
1422                 retval = p->pid;
1423         put_task_struct(p);
1424
1425         BUG_ON(!retval);
1426         return retval;
1427 }
1428
1429 /*
1430  * Handle do_wait work for one task in a live, non-stopped state.
1431  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1432  * the lock and this task is uninteresting.  If we return nonzero, we have
1433  * released the lock and the system call should return.
1434  */
1435 static int wait_task_continued(struct task_struct *p, int noreap,
1436                                struct siginfo __user *infop,
1437                                int __user *stat_addr, struct rusage __user *ru)
1438 {
1439         int retval;
1440         pid_t pid;
1441         uid_t uid;
1442
1443         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1444                 return 0;
1445
1446         spin_lock_irq(&p->sighand->siglock);
1447         /* Re-check with the lock held.  */
1448         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1449                 spin_unlock_irq(&p->sighand->siglock);
1450                 return 0;
1451         }
1452         if (!noreap)
1453                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1454         spin_unlock_irq(&p->sighand->siglock);
1455
1456         pid = p->pid;
1457         uid = p->uid;
1458         get_task_struct(p);
1459         read_unlock(&tasklist_lock);
1460
1461         if (!infop) {
1462                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1463                 put_task_struct(p);
1464                 if (!retval && stat_addr)
1465                         retval = put_user(0xffff, stat_addr);
1466                 if (!retval)
1467                         retval = p->pid;
1468         } else {
1469                 retval = wait_noreap_copyout(p, pid, uid,
1470                                              CLD_CONTINUED, SIGCONT,
1471                                              infop, ru);
1472                 BUG_ON(retval == 0);
1473         }
1474
1475         return retval;
1476 }
1477
1478
1479 static inline int my_ptrace_child(struct task_struct *p)
1480 {
1481         if (!(p->ptrace & PT_PTRACED))
1482                 return 0;
1483         if (!(p->ptrace & PT_ATTACHED))
1484                 return 1;
1485         /*
1486          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1487          * we are the attacher.  If we are the real parent, this is a race
1488          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1489          * which we have to switch the parent links, but has already set
1490          * the flags in p->ptrace.
1491          */
1492         return (p->parent != p->real_parent);
1493 }
1494
1495 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1496                     int __user *stat_addr, struct rusage __user *ru)
1497 {
1498         DECLARE_WAITQUEUE(wait, current);
1499         struct task_struct *tsk;
1500         int flag, retval;
1501         int allowed, denied;
1502
1503         add_wait_queue(&current->signal->wait_chldexit,&wait);
1504 repeat:
1505         /*
1506          * We will set this flag if we see any child that might later
1507          * match our criteria, even if we are not able to reap it yet.
1508          */
1509         flag = 0;
1510         allowed = denied = 0;
1511         current->state = TASK_INTERRUPTIBLE;
1512         read_lock(&tasklist_lock);
1513         tsk = current;
1514         do {
1515                 struct task_struct *p;
1516                 struct list_head *_p;
1517                 int ret;
1518
1519                 list_for_each(_p,&tsk->children) {
1520                         p = list_entry(_p, struct task_struct, sibling);
1521
1522                         ret = eligible_child(pid, options, p);
1523                         if (!ret)
1524                                 continue;
1525
1526                         if (unlikely(ret < 0)) {
1527                                 denied = ret;
1528                                 continue;
1529                         }
1530                         allowed = 1;
1531
1532                         switch (p->state) {
1533                         case TASK_TRACED:
1534                                 /*
1535                                  * When we hit the race with PTRACE_ATTACH,
1536                                  * we will not report this child.  But the
1537                                  * race means it has not yet been moved to
1538                                  * our ptrace_children list, so we need to
1539                                  * set the flag here to avoid a spurious ECHILD
1540                                  * when the race happens with the only child.
1541                                  */
1542                                 flag = 1;
1543                                 if (!my_ptrace_child(p))
1544                                         continue;
1545                                 /*FALLTHROUGH*/
1546                         case TASK_STOPPED:
1547                                 /*
1548                                  * It's stopped now, so it might later
1549                                  * continue, exit, or stop again.
1550                                  */
1551                                 flag = 1;
1552                                 if (!(options & WUNTRACED) &&
1553                                     !my_ptrace_child(p))
1554                                         continue;
1555                                 retval = wait_task_stopped(p, ret == 2,
1556                                                            (options & WNOWAIT),
1557                                                            infop,
1558                                                            stat_addr, ru);
1559                                 if (retval == -EAGAIN)
1560                                         goto repeat;
1561                                 if (retval != 0) /* He released the lock.  */
1562                                         goto end;
1563                                 break;
1564                         default:
1565                         // case EXIT_DEAD:
1566                                 if (p->exit_state == EXIT_DEAD)
1567                                         continue;
1568                         // case EXIT_ZOMBIE:
1569                                 if (p->exit_state == EXIT_ZOMBIE) {
1570                                         /*
1571                                          * Eligible but we cannot release
1572                                          * it yet:
1573                                          */
1574                                         if (ret == 2)
1575                                                 goto check_continued;
1576                                         if (!likely(options & WEXITED))
1577                                                 continue;
1578                                         retval = wait_task_zombie(
1579                                                 p, (options & WNOWAIT),
1580                                                 infop, stat_addr, ru);
1581                                         /* He released the lock.  */
1582                                         if (retval != 0)
1583                                                 goto end;
1584                                         break;
1585                                 }
1586 check_continued:
1587                                 /*
1588                                  * It's running now, so it might later
1589                                  * exit, stop, or stop and then continue.
1590                                  */
1591                                 flag = 1;
1592                                 if (!unlikely(options & WCONTINUED))
1593                                         continue;
1594                                 retval = wait_task_continued(
1595                                         p, (options & WNOWAIT),
1596                                         infop, stat_addr, ru);
1597                                 if (retval != 0) /* He released the lock.  */
1598                                         goto end;
1599                                 break;
1600                         }
1601                 }
1602                 if (!flag) {
1603                         list_for_each(_p, &tsk->ptrace_children) {
1604                                 p = list_entry(_p, struct task_struct,
1605                                                 ptrace_list);
1606                                 if (!eligible_child(pid, options, p))
1607                                         continue;
1608                                 flag = 1;
1609                                 break;
1610                         }
1611                 }
1612                 if (options & __WNOTHREAD)
1613                         break;
1614                 tsk = next_thread(tsk);
1615                 BUG_ON(tsk->signal != current->signal);
1616         } while (tsk != current);
1617
1618         read_unlock(&tasklist_lock);
1619         if (flag) {
1620                 retval = 0;
1621                 if (options & WNOHANG)
1622                         goto end;
1623                 retval = -ERESTARTSYS;
1624                 if (signal_pending(current))
1625                         goto end;
1626                 schedule();
1627                 goto repeat;
1628         }
1629         retval = -ECHILD;
1630         if (unlikely(denied) && !allowed)
1631                 retval = denied;
1632 end:
1633         current->state = TASK_RUNNING;
1634         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1635         if (infop) {
1636                 if (retval > 0)
1637                 retval = 0;
1638                 else {
1639                         /*
1640                          * For a WNOHANG return, clear out all the fields
1641                          * we would set so the user can easily tell the
1642                          * difference.
1643                          */
1644                         if (!retval)
1645                                 retval = put_user(0, &infop->si_signo);
1646                         if (!retval)
1647                                 retval = put_user(0, &infop->si_errno);
1648                         if (!retval)
1649                                 retval = put_user(0, &infop->si_code);
1650                         if (!retval)
1651                                 retval = put_user(0, &infop->si_pid);
1652                         if (!retval)
1653                                 retval = put_user(0, &infop->si_uid);
1654                         if (!retval)
1655                                 retval = put_user(0, &infop->si_status);
1656                 }
1657         }
1658         return retval;
1659 }
1660
1661 asmlinkage long sys_waitid(int which, pid_t pid,
1662                            struct siginfo __user *infop, int options,
1663                            struct rusage __user *ru)
1664 {
1665         long ret;
1666
1667         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1668                 return -EINVAL;
1669         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1670                 return -EINVAL;
1671
1672         switch (which) {
1673         case P_ALL:
1674                 pid = -1;
1675                 break;
1676         case P_PID:
1677                 if (pid <= 0)
1678                         return -EINVAL;
1679                 break;
1680         case P_PGID:
1681                 if (pid <= 0)
1682                         return -EINVAL;
1683                 pid = -pid;
1684                 break;
1685         default:
1686                 return -EINVAL;
1687         }
1688
1689         ret = do_wait(pid, options, infop, NULL, ru);
1690
1691         /* avoid REGPARM breakage on x86: */
1692         prevent_tail_call(ret);
1693         return ret;
1694 }
1695
1696 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1697                           int options, struct rusage __user *ru)
1698 {
1699         long ret;
1700
1701         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1702                         __WNOTHREAD|__WCLONE|__WALL))
1703                 return -EINVAL;
1704         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1705
1706         /* avoid REGPARM breakage on x86: */
1707         prevent_tail_call(ret);
1708         return ret;
1709 }
1710
1711 #ifdef __ARCH_WANT_SYS_WAITPID
1712
1713 /*
1714  * sys_waitpid() remains for compatibility. waitpid() should be
1715  * implemented by calling sys_wait4() from libc.a.
1716  */
1717 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1718 {
1719         return sys_wait4(pid, stat_addr, options, NULL);
1720 }
1721
1722 #endif