kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/mm.h>
   8 #include <linux/slab.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/smp_lock.h>
  11 #include <linux/module.h>
  12 #include <linux/capability.h>
  13 #include <linux/completion.h>
  14 #include <linux/personality.h>
  15 #include <linux/tty.h>
  16 #include <linux/namespace.h>
  17 #include <linux/key.h>
  18 #include <linux/security.h>
  19 #include <linux/cpu.h>
  20 #include <linux/acct.h>
  21 #include <linux/file.h>
  22 #include <linux/binfmts.h>
  23 #include <linux/ptrace.h>
  24 #include <linux/profile.h>
  25 #include <linux/mount.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/mempolicy.h>
  28 #include <linux/taskstats_kern.h>
  29 #include <linux/delayacct.h>
  30 #include <linux/cpuset.h>
  31 #include <linux/syscalls.h>
  32 #include <linux/signal.h>
  33 #include <linux/posix-timers.h>
  34 #include <linux/cn_proc.h>
  35 #include <linux/mutex.h>
  36 #include <linux/futex.h>
  37 #include <linux/compat.h>
  38 #include <linux/pipe_fs_i.h>
  39 #include <linux/audit.h> /* for audit_free() */
  40 #include <linux/resource.h>
  41 #include <linux/vs_base.h>
  42 #include <linux/vs_context.h>
  43 #include <linux/vs_network.h>
  44 #include <linux/vs_limit.h>
  45
  46 #include <asm/uaccess.h>
  47 #include <asm/unistd.h>
  48 #include <asm/pgtable.h>
  49 #include <asm/mmu_context.h>
  50
  51 extern void sem_exit (void);
  52 extern struct task_struct *child_reaper;
  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         struct task_struct *leader;
 144         int zap_leader;
 145 repeat:
 146         atomic_dec(&p->user->processes);
 147         write_lock_irq(&tasklist_lock);
 148         ptrace_unlink(p);
 149         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 150         __exit_signal(p);
 151
 152         /*
 153          * If we are the last non-leader member of the thread
 154          * group, and the leader is zombie, then notify the
 155          * group leader's parent process. (if it wants notification.)
 156          */
 157         zap_leader = 0;
 158         leader = p->group_leader;
 159         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 160                 BUG_ON(leader->exit_signal == -1);
 161                 do_notify_parent(leader, leader->exit_signal);
 162                 /*
 163                  * If we were the last child thread and the leader has
 164                  * exited already, and the leader's parent ignores SIGCHLD,
 165                  * then we are the one who should release the leader.
 166                  *
 167                  * do_notify_parent() will have marked it self-reaping in
 168                  * that case.
 169                  */
 170                 zap_leader = (leader->exit_signal == -1);
 171         }
 172
 173         sched_exit(p);
 174         write_unlock_irq(&tasklist_lock);
 175         proc_flush_task(p);
 176         release_thread(p);
 177         call_rcu(&p->rcu, delayed_put_task_struct);
 178
 179         p = leader;
 180         if (unlikely(zap_leader))
 181                 goto repeat;
 182 }
 183
 184 /*
 185  * This checks not only the pgrp, but falls back on the pid if no
 186  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 187  * without this...
 188  */
 189 int session_of_pgrp(int pgrp)
 190 {
 191         struct task_struct *p;
 192         int sid = -1;
 193
 194         read_lock(&tasklist_lock);
 195         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 196                 if (p->signal->session > 0) {
 197                         sid = p->signal->session;
 198                         goto out;
 199                 }
 200         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 201         p = find_task_by_pid(pgrp);
 202         if (p)
 203                 sid = p->signal->session;
 204 out:
 205         read_unlock(&tasklist_lock);
 206
 207         return sid;
 208 }
 209
 210 /*
 211  * Determine if a process group is "orphaned", according to the POSIX
 212  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 213  * by terminal-generated stop signals.  Newly orphaned process groups are
 214  * to receive a SIGHUP and a SIGCONT.
 215  *
 216  * "I ask you, have you ever known what it is to be an orphan?"
 217  */
 218 static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
 219 {
 220         struct task_struct *p;
 221         int ret = 1;
 222
 223         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 224                 if (p == ignored_task
 225                                 || p->exit_state
 226                                 || p->real_parent->pid == 1)
 227                         continue;
 228                 if (process_group(p->real_parent) != pgrp
 229                             && p->real_parent->signal->session == p->signal->session) {
 230                         ret = 0;
 231                         break;
 232                 }
 233         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 234         return ret;     /* (sighing) "Often!" */
 235 }
 236
 237 int is_orphaned_pgrp(int pgrp)
 238 {
 239         int retval;
 240
 241         read_lock(&tasklist_lock);
 242         retval = will_become_orphaned_pgrp(pgrp, NULL);
 243         read_unlock(&tasklist_lock);
 244
 245         return retval;
 246 }
 247
 248 static int has_stopped_jobs(int pgrp)
 249 {
 250         int retval = 0;
 251         struct task_struct *p;
 252
 253         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 254                 if (p->state != TASK_STOPPED)
 255                         continue;
 256
 257                 /* If p is stopped by a debugger on a signal that won't
 258                    stop it, then don't count p as stopped.  This isn't
 259                    perfect but it's a good approximation.  */
 260                 if (unlikely (p->ptrace)
 261                     && p->exit_code != SIGSTOP
 262                     && p->exit_code != SIGTSTP
 263                     && p->exit_code != SIGTTOU
 264                     && p->exit_code != SIGTTIN)
 265                         continue;
 266
 267                 retval = 1;
 268                 break;
 269         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 270         return retval;
 271 }
 272
 273 /**
 274  * reparent_to_init - Reparent the calling kernel thread to the init task.
 275  *
 276  * If a kernel thread is launched as a result of a system call, or if
 277  * it ever exits, it should generally reparent itself to init so that
 278  * it is correctly cleaned up on exit.
 279  *
 280  * The various task state such as scheduling policy and priority may have
 281  * been inherited from a user process, so we reset them to sane values here.
 282  *
 283  * NOTE that reparent_to_init() gives the caller full capabilities.
 284  */
 285 static void reparent_to_init(void)
 286 {
 287         write_lock_irq(&tasklist_lock);
 288
 289         ptrace_unlink(current);
 290         /* Reparent to init */
 291         remove_parent(current);
 292         current->parent = child_reaper;
 293         current->real_parent = child_reaper;
 294         add_parent(current);
 295
 296         /* Set the exit signal to SIGCHLD so we signal init on exit */
 297         current->exit_signal = SIGCHLD;
 298
 299         if ((current->policy == SCHED_NORMAL ||
 300                         current->policy == SCHED_BATCH)
 301                                 && (task_nice(current) < 0))
 302                 set_user_nice(current, 0);
 303         /* cpus_allowed? */
 304         /* rt_priority? */
 305         /* signals? */
 306         security_task_reparent_to_init(current);
 307         memcpy(current->signal->rlim, init_task.signal->rlim,
 308                sizeof(current->signal->rlim));
 309         atomic_inc(&(INIT_USER->__count));
 310         write_unlock_irq(&tasklist_lock);
 311         switch_uid(INIT_USER);
 312 }
 313
 314 void __set_special_pids(pid_t session, pid_t pgrp)
 315 {
 316         struct task_struct *curr = current->group_leader;
 317
 318         if (curr->signal->session != session) {
 319                 detach_pid(curr, PIDTYPE_SID);
 320                 curr->signal->session = session;
 321                 attach_pid(curr, PIDTYPE_SID, session);
 322         }
 323         if (process_group(curr) != pgrp) {
 324                 detach_pid(curr, PIDTYPE_PGID);
 325                 curr->signal->pgrp = pgrp;
 326                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 327         }
 328 }
 329
 330 void set_special_pids(pid_t session, pid_t pgrp)
 331 {
 332         write_lock_irq(&tasklist_lock);
 333         __set_special_pids(session, pgrp);
 334         write_unlock_irq(&tasklist_lock);
 335 }
 336
 337 /*
 338  * Let kernel threads use this to say that they
 339  * allow a certain signal (since daemonize() will
 340  * have disabled all of them by default).
 341  */
 342 int allow_signal(int sig)
 343 {
 344         if (!valid_signal(sig) || sig < 1)
 345                 return -EINVAL;
 346
 347         spin_lock_irq(&current->sighand->siglock);
 348         sigdelset(&current->blocked, sig);
 349         if (!current->mm) {
 350                 /* Kernel threads handle their own signals.
 351                    Let the signal code know it'll be handled, so
 352                    that they don't get converted to SIGKILL or
 353                    just silently dropped */
 354                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 355         }
 356         recalc_sigpending();
 357         spin_unlock_irq(&current->sighand->siglock);
 358         return 0;
 359 }
 360
 361 EXPORT_SYMBOL(allow_signal);
 362
 363 int disallow_signal(int sig)
 364 {
 365         if (!valid_signal(sig) || sig < 1)
 366                 return -EINVAL;
 367
 368         spin_lock_irq(&current->sighand->siglock);
 369         sigaddset(&current->blocked, sig);
 370         recalc_sigpending();
 371         spin_unlock_irq(&current->sighand->siglock);
 372         return 0;
 373 }
 374
 375 EXPORT_SYMBOL(disallow_signal);
 376
 377 /*
 378  *      Put all the gunge required to become a kernel thread without
 379  *      attached user resources in one place where it belongs.
 380  */
 381
 382 void daemonize(const char *name, ...)
 383 {
 384         va_list args;
 385         struct fs_struct *fs;
 386         sigset_t blocked;
 387
 388         va_start(args, name);
 389         vsnprintf(current->comm, sizeof(current->comm), name, args);
 390         va_end(args);
 391
 392         /*
 393          * If we were started as result of loading a module, close all of the
 394          * user space pages.  We don't need them, and if we didn't close them
 395          * they would be locked into memory.
 396          */
 397         exit_mm(current);
 398
 399         set_special_pids(1, 1);
 400         proc_clear_tty(current);
 401
 402         /* Block and flush all signals */
 403         sigfillset(&blocked);
 404         sigprocmask(SIG_BLOCK, &blocked, NULL);
 405         flush_signals(current);
 406
 407         /* Become as one with the init task */
 408
 409         exit_fs(current);       /* current->fs->count--; */
 410         fs = init_task.fs;
 411         current->fs = fs;
 412         atomic_inc(&fs->count);
 413         exit_namespace(current);
 414         current->namespace = init_task.namespace;
 415         get_namespace(current->namespace);
 416         exit_files(current);
 417         current->files = init_task.files;
 418         atomic_inc(&current->files->count);
 419
 420         reparent_to_init();
 421 }
 422
 423 EXPORT_SYMBOL(daemonize);
 424
 425 static void close_files(struct files_struct * files)
 426 {
 427         int i, j;
 428         struct fdtable *fdt;
 429
 430         j = 0;
 431
 432         /*
 433          * It is safe to dereference the fd table without RCU or
 434          * ->file_lock because this is the last reference to the
 435          * files structure.
 436          */
 437         fdt = files_fdtable(files);
 438         for (;;) {
 439                 unsigned long set;
 440                 i = j * __NFDBITS;
 441                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 442                         break;
 443                 set = fdt->open_fds->fds_bits[j++];
 444                 while (set) {
 445                         if (set & 1) {
 446                                 struct file * file = xchg(&fdt->fd[i], NULL);
 447                                 if (file)
 448                                         filp_close(file, files);
 449                                 vx_openfd_dec(i);
 450                         }
 451                         i++;
 452                         set >>= 1;
 453                         cond_resched();
 454                 }
 455         }
 456 }
 457
 458 struct files_struct *get_files_struct(struct task_struct *task)
 459 {
 460         struct files_struct *files;
 461
 462         task_lock(task);
 463         files = task->files;
 464         if (files)
 465                 atomic_inc(&files->count);
 466         task_unlock(task);
 467
 468         return files;
 469 }
 470
 471 void fastcall put_files_struct(struct files_struct *files)
 472 {
 473         struct fdtable *fdt;
 474
 475         if (atomic_dec_and_test(&files->count)) {
 476                 close_files(files);
 477                 /*
 478                  * Free the fd and fdset arrays if we expanded them.
 479                  * If the fdtable was embedded, pass files for freeing
 480                  * at the end of the RCU grace period. Otherwise,
 481                  * you can free files immediately.
 482                  */
 483                 fdt = files_fdtable(files);
 484                 if (fdt == &files->fdtab)
 485                         fdt->free_files = files;
 486                 else
 487                         kmem_cache_free(files_cachep, files);
 488                 free_fdtable(fdt);
 489         }
 490 }
 491
 492 EXPORT_SYMBOL(put_files_struct);
 493
 494 static inline void __exit_files(struct task_struct *tsk)
 495 {
 496         struct files_struct * files = tsk->files;
 497
 498         if (files) {
 499                 task_lock(tsk);
 500                 tsk->files = NULL;
 501                 task_unlock(tsk);
 502                 put_files_struct(files);
 503         }
 504 }
 505
 506 void exit_files(struct task_struct *tsk)
 507 {
 508         __exit_files(tsk);
 509 }
 510
 511 static inline void __put_fs_struct(struct fs_struct *fs)
 512 {
 513         /* No need to hold fs->lock if we are killing it */
 514         if (atomic_dec_and_test(&fs->count)) {
 515                 dput(fs->root);
 516                 mntput(fs->rootmnt);
 517                 dput(fs->pwd);
 518                 mntput(fs->pwdmnt);
 519                 if (fs->altroot) {
 520                         dput(fs->altroot);
 521                         mntput(fs->altrootmnt);
 522                 }
 523                 kmem_cache_free(fs_cachep, fs);
 524         }
 525 }
 526
 527 void put_fs_struct(struct fs_struct *fs)
 528 {
 529         __put_fs_struct(fs);
 530 }
 531
 532 static inline void __exit_fs(struct task_struct *tsk)
 533 {
 534         struct fs_struct * fs = tsk->fs;
 535
 536         if (fs) {
 537                 task_lock(tsk);
 538                 tsk->fs = NULL;
 539                 task_unlock(tsk);
 540                 __put_fs_struct(fs);
 541         }
 542 }
 543
 544 void exit_fs(struct task_struct *tsk)
 545 {
 546         __exit_fs(tsk);
 547 }
 548
 549 EXPORT_SYMBOL_GPL(exit_fs);
 550
 551 /*
 552  * Turn us into a lazy TLB process if we
 553  * aren't already..
 554  */
 555 static void exit_mm(struct task_struct * tsk)
 556 {
 557         struct mm_struct *mm = tsk->mm;
 558
 559         mm_release(tsk, mm);
 560         if (!mm)
 561                 return;
 562         /*
 563          * Serialize with any possible pending coredump.
 564          * We must hold mmap_sem around checking core_waiters
 565          * and clearing tsk->mm.  The core-inducing thread
 566          * will increment core_waiters for each thread in the
 567          * group with ->mm != NULL.
 568          */
 569         down_read(&mm->mmap_sem);
 570         if (mm->core_waiters) {
 571                 up_read(&mm->mmap_sem);
 572                 down_write(&mm->mmap_sem);
 573                 if (!--mm->core_waiters)
 574                         complete(mm->core_startup_done);
 575                 up_write(&mm->mmap_sem);
 576
 577                 wait_for_completion(&mm->core_done);
 578                 down_read(&mm->mmap_sem);
 579         }
 580         atomic_inc(&mm->mm_count);
 581         BUG_ON(mm != tsk->active_mm);
 582         /* more a memory barrier than a real lock */
 583         task_lock(tsk);
 584         tsk->mm = NULL;
 585         up_read(&mm->mmap_sem);
 586         enter_lazy_tlb(mm, current);
 587         task_unlock(tsk);
 588         mmput(mm);
 589 }
 590
 591 static inline void
 592 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 593 {
 594         /* check for reaper context */
 595         vxwprintk((p->xid != reaper->xid) && (reaper != child_reaper),
 596                 "rogue reaper: %p[%d,#%u] <> %p[%d,#%u]",
 597                 p, p->pid, p->xid, reaper, reaper->pid, reaper->xid);
 598
 599         /*
 600          * Make sure we're not reparenting to ourselves and that
 601          * the parent is not a zombie.
 602          */
 603         BUG_ON(p == reaper || reaper->exit_state);
 604         p->real_parent = reaper;
 605 }
 606
 607 static void
 608 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 609 {
 610         /* We don't want people slaying init.  */
 611         if (p->exit_signal != -1)
 612                 p->exit_signal = SIGCHLD;
 613
 614         if (p->pdeath_signal)
 615                 /* We already hold the tasklist_lock here.  */
 616                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 617
 618         /* Move the child from its dying parent to the new one.  */
 619         if (unlikely(traced)) {
 620                 /* Preserve ptrace links if someone else is tracing this child.  */
 621                 list_del_init(&p->ptrace_list);
 622                 if (p->parent != p->real_parent)
 623                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 624         } else {
 625                 /* If this child is being traced, then we're the one tracing it
 626                  * anyway, so let go of it.
 627                  */
 628                 p->ptrace = 0;
 629                 remove_parent(p);
 630                 p->parent = p->real_parent;
 631                 add_parent(p);
 632
 633                 /* If we'd notified the old parent about this child's death,
 634                  * also notify the new parent.
 635                  */
 636                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 637                     thread_group_empty(p))
 638                         do_notify_parent(p, p->exit_signal);
 639                 else if (p->state == TASK_TRACED) {
 640                         /*
 641                          * If it was at a trace stop, turn it into
 642                          * a normal stop since it's no longer being
 643                          * traced.
 644                          */
 645                         ptrace_untrace(p);
 646                 }
 647         }
 648
 649         /*
 650          * process group orphan check
 651          * Case ii: Our child is in a different pgrp
 652          * than we are, and it was the only connection
 653          * outside, so the child pgrp is now orphaned.
 654          */
 655         if ((process_group(p) != process_group(father)) &&
 656             (p->signal->session == father->signal->session)) {
 657                 int pgrp = process_group(p);
 658
 659                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 660                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 661                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 662                 }
 663         }
 664 }
 665
 666 /*
 667  * When we die, we re-parent all our children.
 668  * Try to give them to another thread in our thread
 669  * group, and if no such member exists, give it to
 670  * the global child reaper process (ie "init")
 671  */
 672 static void
 673 forget_original_parent(struct task_struct *father, struct list_head *to_release)
 674 {
 675         struct task_struct *p, *reaper = father;
 676         struct list_head *_p, *_n;
 677
 678         do {
 679                 reaper = next_thread(reaper);
 680                 if (reaper == father) {
 681                         reaper = vx_child_reaper(father);
 682                         break;
 683                 }
 684         } while (reaper->exit_state);
 685
 686         /*
 687          * There are only two places where our children can be:
 688          *
 689          * - in our child list
 690          * - in our ptraced child list
 691          *
 692          * Search them and reparent children.
 693          */
 694         list_for_each_safe(_p, _n, &father->children) {
 695                 int ptrace;
 696                 p = list_entry(_p, struct task_struct, sibling);
 697
 698                 ptrace = p->ptrace;
 699
 700                 /* if father isn't the real parent, then ptrace must be enabled */
 701                 BUG_ON(father != p->real_parent && !ptrace);
 702
 703                 if (father == p->real_parent) {
 704                         /* reparent with a reaper, real father it's us */
 705                         choose_new_parent(p, vx_child_reaper(p));
 706                         reparent_thread(p, father, 0);
 707                 } else {
 708                         /* reparent ptraced task to its real parent */
 709                         __ptrace_unlink (p);
 710                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 711                             thread_group_empty(p))
 712                                 do_notify_parent(p, p->exit_signal);
 713                 }
 714
 715                 /*
 716                  * if the ptraced child is a zombie with exit_signal == -1
 717                  * we must collect it before we exit, or it will remain
 718                  * zombie forever since we prevented it from self-reap itself
 719                  * while it was being traced by us, to be able to see it in wait4.
 720                  */
 721                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 722                         list_add(&p->ptrace_list, to_release);
 723         }
 724         list_for_each_safe(_p, _n, &father->ptrace_children) {
 725                 p = list_entry(_p, struct task_struct, ptrace_list);
 726                 choose_new_parent(p, reaper);
 727                 reparent_thread(p, father, 1);
 728         }
 729 }
 730
 731 /*
 732  * Send signals to all our closest relatives so that they know
 733  * to properly mourn us..
 734  */
 735 static void exit_notify(struct task_struct *tsk)
 736 {
 737         int state;
 738         struct task_struct *t;
 739         struct list_head ptrace_dead, *_p, *_n;
 740
 741         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 742             && !thread_group_empty(tsk)) {
 743                 /*
 744                  * This occurs when there was a race between our exit
 745                  * syscall and a group signal choosing us as the one to
 746                  * wake up.  It could be that we are the only thread
 747                  * alerted to check for pending signals, but another thread
 748                  * should be woken now to take the signal since we will not.
 749                  * Now we'll wake all the threads in the group just to make
 750                  * sure someone gets all the pending signals.
 751                  */
 752                 read_lock(&tasklist_lock);
 753                 spin_lock_irq(&tsk->sighand->siglock);
 754                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 755                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 756                                 recalc_sigpending_tsk(t);
 757                                 if (signal_pending(t))
 758                                         signal_wake_up(t, 0);
 759                         }
 760                 spin_unlock_irq(&tsk->sighand->siglock);
 761                 read_unlock(&tasklist_lock);
 762         }
 763
 764         write_lock_irq(&tasklist_lock);
 765
 766         /*
 767          * This does two things:
 768          *
 769          * A.  Make init inherit all the child processes
 770          * B.  Check to see if any process groups have become orphaned
 771          *      as a result of our exiting, and if they have any stopped
 772          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 773          */
 774
 775         INIT_LIST_HEAD(&ptrace_dead);
 776         forget_original_parent(tsk, &ptrace_dead);
 777         BUG_ON(!list_empty(&tsk->children));
 778         BUG_ON(!list_empty(&tsk->ptrace_children));
 779
 780         /*
 781          * Check to see if any process groups have become orphaned
 782          * as a result of our exiting, and if they have any stopped
 783          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 784          *
 785          * Case i: Our father is in a different pgrp than we are
 786          * and we were the only connection outside, so our pgrp
 787          * is about to become orphaned.
 788          */
 789
 790         t = tsk->real_parent;
 791
 792         if ((process_group(t) != process_group(tsk)) &&
 793             (t->signal->session == tsk->signal->session) &&
 794             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 795             has_stopped_jobs(process_group(tsk))) {
 796                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 797                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 798         }
 799
 800         /* Let father know we died
 801          *
 802          * Thread signals are configurable, but you aren't going to use
 803          * that to send signals to arbitary processes.
 804          * That stops right now.
 805          *
 806          * If the parent exec id doesn't match the exec id we saved
 807          * when we started then we know the parent has changed security
 808          * domain.
 809          *
 810          * If our self_exec id doesn't match our parent_exec_id then
 811          * we have changed execution domain as these two values started
 812          * the same after a fork.
 813          *
 814          */
 815
 816         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 817             ( tsk->parent_exec_id != t->self_exec_id  ||
 818               tsk->self_exec_id != tsk->parent_exec_id)
 819             && !capable(CAP_KILL))
 820                 tsk->exit_signal = SIGCHLD;
 821
 822
 823         /* If something other than our normal parent is ptracing us, then
 824          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 825          * only has special meaning to our real parent.
 826          */
 827         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 828                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 829                 do_notify_parent(tsk, signal);
 830         } else if (tsk->ptrace) {
 831                 do_notify_parent(tsk, SIGCHLD);
 832         }
 833
 834         state = EXIT_ZOMBIE;
 835         if (tsk->exit_signal == -1 &&
 836             (likely(tsk->ptrace == 0) ||
 837              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 838                 state = EXIT_DEAD;
 839         tsk->exit_state = state;
 840
 841         write_unlock_irq(&tasklist_lock);
 842
 843         list_for_each_safe(_p, _n, &ptrace_dead) {
 844                 list_del_init(_p);
 845                 t = list_entry(_p, struct task_struct, ptrace_list);
 846                 release_task(t);
 847         }
 848
 849         /* If the process is dead, release it - nobody will wait for it */
 850         if (state == EXIT_DEAD)
 851                 release_task(tsk);
 852 }
 853
 854 fastcall NORET_TYPE void do_exit(long code)
 855 {
 856         struct task_struct *tsk = current;
 857         struct taskstats *tidstats;
 858         int group_dead;
 859         unsigned int mycpu;
 860
 861         profile_task_exit(tsk);
 862
 863         WARN_ON(atomic_read(&tsk->fs_excl));
 864
 865         if (unlikely(in_interrupt()))
 866                 panic("Aiee, killing interrupt handler!");
 867         if (unlikely(!tsk->pid))
 868                 panic("Attempted to kill the idle task!");
 869         if (unlikely(tsk == child_reaper))
 870                 panic("Attempted to kill init!");
 871
 872         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 873                 current->ptrace_message = code;
 874                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 875         }
 876
 877         /*
 878          * We're taking recursive faults here in do_exit. Safest is to just
 879          * leave this task alone and wait for reboot.
 880          */
 881         if (unlikely(tsk->flags & PF_EXITING)) {
 882                 printk(KERN_ALERT
 883                         "Fixing recursive fault but reboot is needed!\n");
 884                 if (tsk->io_context)
 885                         exit_io_context();
 886                 set_current_state(TASK_UNINTERRUPTIBLE);
 887                 schedule();
 888         }
 889
 890         tsk->flags |= PF_EXITING;
 891
 892         if (unlikely(in_atomic()))
 893                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 894                                 current->comm, current->pid,
 895                                 preempt_count());
 896
 897         taskstats_exit_alloc(&tidstats, &mycpu);
 898
 899         acct_update_integrals(tsk);
 900         if (tsk->mm) {
 901                 update_hiwater_rss(tsk->mm);
 902                 update_hiwater_vm(tsk->mm);
 903         }
 904         group_dead = atomic_dec_and_test(&tsk->signal->live);
 905         if (group_dead) {
 906                 hrtimer_cancel(&tsk->signal->real_timer);
 907                 exit_itimers(tsk->signal);
 908         }
 909
 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         acct_collect(code, group_dead);
 919         if (unlikely(tsk->robust_list))
 920                 exit_robust_list(tsk);
 921 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 922         if (unlikely(tsk->compat_robust_list))
 923                 compat_exit_robust_list(tsk);
 924 #endif
 925         if (unlikely(tsk->audit_context))
 926                 audit_free(tsk);
 927         taskstats_exit_send(tsk, tidstats, group_dead, mycpu);
 928         taskstats_exit_free(tidstats);
 929
 930         exit_mm(tsk);
 931
 932         if (group_dead)
 933                 acct_process();
 934         exit_sem(tsk);
 935         __exit_files(tsk);
 936         __exit_fs(tsk);
 937         exit_namespace(tsk);
 938         exit_thread();
 939         cpuset_exit(tsk);
 940         exit_keys(tsk);
 941
 942         if (group_dead && tsk->signal->leader)
 943                 disassociate_ctty(1);
 944
 945         module_put(task_thread_info(tsk)->exec_domain->module);
 946         if (tsk->binfmt)
 947                 module_put(tsk->binfmt->module);
 948
 949         tsk->exit_code = code;
 950         proc_exit_connector(tsk);
 951         /* needs to stay before exit_notify() */
 952         exit_vx_info_early(tsk, code);
 953         exit_notify(tsk);
 954 #ifdef CONFIG_NUMA
 955         mpol_free(tsk->mempolicy);
 956         tsk->mempolicy = NULL;
 957 #endif
 958         /*
 959          * This must happen late, after the PID is not
 960          * hashed anymore:
 961          */
 962         if (unlikely(!list_empty(&tsk->pi_state_list)))
 963                 exit_pi_state_list(tsk);
 964         if (unlikely(current->pi_state_cache))
 965                 kfree(current->pi_state_cache);
 966         /*
 967          * Make sure we are holding no locks:
 968          */
 969         debug_check_no_locks_held(tsk);
 970
 971         if (tsk->io_context)
 972                 exit_io_context();
 973
 974         if (tsk->splice_pipe)
 975                 __free_pipe_info(tsk->splice_pipe);
 976
 977         /* needs to stay after exit_notify() */
 978         exit_vx_info(tsk, code);
 979         exit_nx_info(tsk);
 980
 981         /* PF_DEAD causes final put_task_struct after we schedule. */
 982         preempt_disable();
 983         BUG_ON(tsk->flags & PF_DEAD);
 984         tsk->flags |= PF_DEAD;
 985
 986         schedule();
 987         BUG();
 988         /* Avoid "noreturn function does return".  */
 989         for (;;) ;
 990 }
 991
 992 EXPORT_SYMBOL_GPL(do_exit);
 993
 994 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 995 {
 996         if (comp)
 997                 complete(comp);
 998
 999         do_exit(code);
1000 }
1001
1002 EXPORT_SYMBOL(complete_and_exit);
1003
1004 asmlinkage long sys_exit(int error_code)
1005 {
1006         do_exit((error_code&0xff)<<8);
1007 }
1008
1009 /*
1010  * Take down every thread in the group.  This is called by fatal signals
1011  * as well as by sys_exit_group (below).
1012  */
1013 NORET_TYPE void
1014 do_group_exit(int exit_code)
1015 {
1016         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1017
1018         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1019                 exit_code = current->signal->group_exit_code;
1020         else if (!thread_group_empty(current)) {
1021                 struct signal_struct *const sig = current->signal;
1022                 struct sighand_struct *const sighand = current->sighand;
1023                 spin_lock_irq(&sighand->siglock);
1024                 if (sig->flags & SIGNAL_GROUP_EXIT)
1025                         /* Another thread got here before we took the lock.  */
1026                         exit_code = sig->group_exit_code;
1027                 else {
1028                         sig->group_exit_code = exit_code;
1029                         zap_other_threads(current);
1030                 }
1031                 spin_unlock_irq(&sighand->siglock);
1032         }
1033
1034         do_exit(exit_code);
1035         /* NOTREACHED */
1036 }
1037
1038 /*
1039  * this kills every thread in the thread group. Note that any externally
1040  * wait4()-ing process will get the correct exit code - even if this
1041  * thread is not the thread group leader.
1042  */
1043 asmlinkage void sys_exit_group(int error_code)
1044 {
1045         do_group_exit((error_code & 0xff) << 8);
1046 }
1047
1048 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1049 {
1050         if (pid > 0) {
1051                 if (p->pid != pid)
1052                         return 0;
1053         } else if (!pid) {
1054                 if (process_group(p) != process_group(current))
1055                         return 0;
1056         } else if (pid != -1) {
1057                 if (process_group(p) != -pid)
1058                         return 0;
1059         }
1060
1061         /*
1062          * Do not consider detached threads that are
1063          * not ptraced:
1064          */
1065         if (p->exit_signal == -1 && !p->ptrace)
1066                 return 0;
1067
1068         /* Wait for all children (clone and not) if __WALL is set;
1069          * otherwise, wait for clone children *only* if __WCLONE is
1070          * set; otherwise, wait for non-clone children *only*.  (Note:
1071          * A "clone" child here is one that reports to its parent
1072          * using a signal other than SIGCHLD.) */
1073         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1074             && !(options & __WALL))
1075                 return 0;
1076         /*
1077          * Do not consider thread group leaders that are
1078          * in a non-empty thread group:
1079          */
1080         if (delay_group_leader(p))
1081                 return 2;
1082
1083         if (security_task_wait(p))
1084                 return 0;
1085
1086         return 1;
1087 }
1088
1089 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1090                                int why, int status,
1091                                struct siginfo __user *infop,
1092                                struct rusage __user *rusagep)
1093 {
1094         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1095
1096         put_task_struct(p);
1097         if (!retval)
1098                 retval = put_user(SIGCHLD, &infop->si_signo);
1099         if (!retval)
1100                 retval = put_user(0, &infop->si_errno);
1101         if (!retval)
1102                 retval = put_user((short)why, &infop->si_code);
1103         if (!retval)
1104                 retval = put_user(pid, &infop->si_pid);
1105         if (!retval)
1106                 retval = put_user(uid, &infop->si_uid);
1107         if (!retval)
1108                 retval = put_user(status, &infop->si_status);
1109         if (!retval)
1110                 retval = pid;
1111         return retval;
1112 }
1113
1114 /*
1115  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1116  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1117  * the lock and this task is uninteresting.  If we return nonzero, we have
1118  * released the lock and the system call should return.
1119  */
1120 static int wait_task_zombie(struct task_struct *p, int noreap,
1121                             struct siginfo __user *infop,
1122                             int __user *stat_addr, struct rusage __user *ru)
1123 {
1124         unsigned long state;
1125         int retval;
1126         int status;
1127
1128         if (unlikely(noreap)) {
1129                 pid_t pid = p->pid;
1130                 uid_t uid = p->uid;
1131                 int exit_code = p->exit_code;
1132                 int why, status;
1133
1134                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1135                         return 0;
1136                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1137                         return 0;
1138                 get_task_struct(p);
1139                 read_unlock(&tasklist_lock);
1140                 if ((exit_code & 0x7f) == 0) {
1141                         why = CLD_EXITED;
1142                         status = exit_code >> 8;
1143                 } else {
1144                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1145                         status = exit_code & 0x7f;
1146                 }
1147                 return wait_noreap_copyout(p, pid, uid, why,
1148                                            status, infop, ru);
1149         }
1150
1151         /*
1152          * Try to move the task's state to DEAD
1153          * only one thread is allowed to do this:
1154          */
1155         state = xchg(&p->exit_state, EXIT_DEAD);
1156         if (state != EXIT_ZOMBIE) {
1157                 BUG_ON(state != EXIT_DEAD);
1158                 return 0;
1159         }
1160         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1161                 /*
1162                  * This can only happen in a race with a ptraced thread
1163                  * dying on another processor.
1164                  */
1165                 return 0;
1166         }
1167
1168         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1169                 struct signal_struct *psig;
1170                 struct signal_struct *sig;
1171
1172                 /*
1173                  * The resource counters for the group leader are in its
1174                  * own task_struct.  Those for dead threads in the group
1175                  * are in its signal_struct, as are those for the child
1176                  * processes it has previously reaped.  All these
1177                  * accumulate in the parent's signal_struct c* fields.
1178                  *
1179                  * We don't bother to take a lock here to protect these
1180                  * p->signal fields, because they are only touched by
1181                  * __exit_signal, which runs with tasklist_lock
1182                  * write-locked anyway, and so is excluded here.  We do
1183                  * need to protect the access to p->parent->signal fields,
1184                  * as other threads in the parent group can be right
1185                  * here reaping other children at the same time.
1186                  */
1187                 spin_lock_irq(&p->parent->sighand->siglock);
1188                 psig = p->parent->signal;
1189                 sig = p->signal;
1190                 psig->cutime =
1191                         cputime_add(psig->cutime,
1192                         cputime_add(p->utime,
1193                         cputime_add(sig->utime,
1194                                     sig->cutime)));
1195                 psig->cstime =
1196                         cputime_add(psig->cstime,
1197                         cputime_add(p->stime,
1198                         cputime_add(sig->stime,
1199                                     sig->cstime)));
1200                 psig->cmin_flt +=
1201                         p->min_flt + sig->min_flt + sig->cmin_flt;
1202                 psig->cmaj_flt +=
1203                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1204                 psig->cnvcsw +=
1205                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1206                 psig->cnivcsw +=
1207                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1208                 spin_unlock_irq(&p->parent->sighand->siglock);
1209         }
1210
1211         /*
1212          * Now we are sure this task is interesting, and no other
1213          * thread can reap it because we set its state to EXIT_DEAD.
1214          */
1215         read_unlock(&tasklist_lock);
1216
1217         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1218         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1219                 ? p->signal->group_exit_code : p->exit_code;
1220         if (!retval && stat_addr)
1221                 retval = put_user(status, stat_addr);
1222         if (!retval && infop)
1223                 retval = put_user(SIGCHLD, &infop->si_signo);
1224         if (!retval && infop)
1225                 retval = put_user(0, &infop->si_errno);
1226         if (!retval && infop) {
1227                 int why;
1228
1229                 if ((status & 0x7f) == 0) {
1230                         why = CLD_EXITED;
1231                         status >>= 8;
1232                 } else {
1233                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1234                         status &= 0x7f;
1235                 }
1236                 retval = put_user((short)why, &infop->si_code);
1237                 if (!retval)
1238                         retval = put_user(status, &infop->si_status);
1239         }
1240         if (!retval && infop)
1241                 retval = put_user(p->pid, &infop->si_pid);
1242         if (!retval && infop)
1243                 retval = put_user(p->uid, &infop->si_uid);
1244         if (retval) {
1245                 // TODO: is this safe?
1246                 p->exit_state = EXIT_ZOMBIE;
1247                 return retval;
1248         }
1249         retval = p->pid;
1250         if (p->real_parent != p->parent) {
1251                 write_lock_irq(&tasklist_lock);
1252                 /* Double-check with lock held.  */
1253                 if (p->real_parent != p->parent) {
1254                         __ptrace_unlink(p);
1255                         // TODO: is this safe?
1256                         p->exit_state = EXIT_ZOMBIE;
1257                         /*
1258                          * If this is not a detached task, notify the parent.
1259                          * If it's still not detached after that, don't release
1260                          * it now.
1261                          */
1262                         if (p->exit_signal != -1) {
1263                                 do_notify_parent(p, p->exit_signal);
1264                                 if (p->exit_signal != -1)
1265                                         p = NULL;
1266                         }
1267                 }
1268                 write_unlock_irq(&tasklist_lock);
1269         }
1270         if (p != NULL)
1271                 release_task(p);
1272         BUG_ON(!retval);
1273         return retval;
1274 }
1275
1276 /*
1277  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1278  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1279  * the lock and this task is uninteresting.  If we return nonzero, we have
1280  * released the lock and the system call should return.
1281  */
1282 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1283                              int noreap, struct siginfo __user *infop,
1284                              int __user *stat_addr, struct rusage __user *ru)
1285 {
1286         int retval, exit_code;
1287
1288         if (!p->exit_code)
1289                 return 0;
1290         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1291             p->signal && p->signal->group_stop_count > 0)
1292                 /*
1293                  * A group stop is in progress and this is the group leader.
1294                  * We won't report until all threads have stopped.
1295                  */
1296                 return 0;
1297
1298         /*
1299          * Now we are pretty sure this task is interesting.
1300          * Make sure it doesn't get reaped out from under us while we
1301          * give up the lock and then examine it below.  We don't want to
1302          * keep holding onto the tasklist_lock while we call getrusage and
1303          * possibly take page faults for user memory.
1304          */
1305         get_task_struct(p);
1306         read_unlock(&tasklist_lock);
1307
1308         if (unlikely(noreap)) {
1309                 pid_t pid = p->pid;
1310                 uid_t uid = p->uid;
1311                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1312
1313                 exit_code = p->exit_code;
1314                 if (unlikely(!exit_code) ||
1315                     unlikely(p->state & TASK_TRACED))
1316                         goto bail_ref;
1317                 return wait_noreap_copyout(p, pid, uid,
1318                                            why, (exit_code << 8) | 0x7f,
1319                                            infop, ru);
1320         }
1321
1322         write_lock_irq(&tasklist_lock);
1323
1324         /*
1325          * This uses xchg to be atomic with the thread resuming and setting
1326          * it.  It must also be done with the write lock held to prevent a
1327          * race with the EXIT_ZOMBIE case.
1328          */
1329         exit_code = xchg(&p->exit_code, 0);
1330         if (unlikely(p->exit_state)) {
1331                 /*
1332                  * The task resumed and then died.  Let the next iteration
1333                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1334                  * already be zero here if it resumed and did _exit(0).
1335                  * The task itself is dead and won't touch exit_code again;
1336                  * other processors in this function are locked out.
1337                  */
1338                 p->exit_code = exit_code;
1339                 exit_code = 0;
1340         }
1341         if (unlikely(exit_code == 0)) {
1342                 /*
1343                  * Another thread in this function got to it first, or it
1344                  * resumed, or it resumed and then died.
1345                  */
1346                 write_unlock_irq(&tasklist_lock);
1347 bail_ref:
1348                 put_task_struct(p);
1349                 /*
1350                  * We are returning to the wait loop without having successfully
1351                  * removed the process and having released the lock. We cannot
1352                  * continue, since the "p" task pointer is potentially stale.
1353                  *
1354                  * Return -EAGAIN, and do_wait() will restart the loop from the
1355                  * beginning. Do _not_ re-acquire the lock.
1356                  */
1357                 return -EAGAIN;
1358         }
1359
1360         /* move to end of parent's list to avoid starvation */
1361         remove_parent(p);
1362         add_parent(p);
1363
1364         write_unlock_irq(&tasklist_lock);
1365
1366         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1367         if (!retval && stat_addr)
1368                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1369         if (!retval && infop)
1370                 retval = put_user(SIGCHLD, &infop->si_signo);
1371         if (!retval && infop)
1372                 retval = put_user(0, &infop->si_errno);
1373         if (!retval && infop)
1374                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1375                                           ? CLD_TRAPPED : CLD_STOPPED),
1376                                   &infop->si_code);
1377         if (!retval && infop)
1378                 retval = put_user(exit_code, &infop->si_status);
1379         if (!retval && infop)
1380                 retval = put_user(p->pid, &infop->si_pid);
1381         if (!retval && infop)
1382                 retval = put_user(p->uid, &infop->si_uid);
1383         if (!retval)
1384                 retval = p->pid;
1385         put_task_struct(p);
1386
1387         BUG_ON(!retval);
1388         return retval;
1389 }
1390
1391 /*
1392  * Handle do_wait work for one task in a live, non-stopped state.
1393  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1394  * the lock and this task is uninteresting.  If we return nonzero, we have
1395  * released the lock and the system call should return.
1396  */
1397 static int wait_task_continued(struct task_struct *p, int noreap,
1398                                struct siginfo __user *infop,
1399                                int __user *stat_addr, struct rusage __user *ru)
1400 {
1401         int retval;
1402         pid_t pid;
1403         uid_t uid;
1404
1405         if (unlikely(!p->signal))
1406                 return 0;
1407
1408         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1409                 return 0;
1410
1411         spin_lock_irq(&p->sighand->siglock);
1412         /* Re-check with the lock held.  */
1413         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1414                 spin_unlock_irq(&p->sighand->siglock);
1415                 return 0;
1416         }
1417         if (!noreap)
1418                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1419         spin_unlock_irq(&p->sighand->siglock);
1420
1421         pid = p->pid;
1422         uid = p->uid;
1423         get_task_struct(p);
1424         read_unlock(&tasklist_lock);
1425
1426         if (!infop) {
1427                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1428                 put_task_struct(p);
1429                 if (!retval && stat_addr)
1430                         retval = put_user(0xffff, stat_addr);
1431                 if (!retval)
1432                         retval = p->pid;
1433         } else {
1434                 retval = wait_noreap_copyout(p, pid, uid,
1435                                              CLD_CONTINUED, SIGCONT,
1436                                              infop, ru);
1437                 BUG_ON(retval == 0);
1438         }
1439
1440         return retval;
1441 }
1442
1443
1444 static inline int my_ptrace_child(struct task_struct *p)
1445 {
1446         if (!(p->ptrace & PT_PTRACED))
1447                 return 0;
1448         if (!(p->ptrace & PT_ATTACHED))
1449                 return 1;
1450         /*
1451          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1452          * we are the attacher.  If we are the real parent, this is a race
1453          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1454          * which we have to switch the parent links, but has already set
1455          * the flags in p->ptrace.
1456          */
1457         return (p->parent != p->real_parent);
1458 }
1459
1460 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1461                     int __user *stat_addr, struct rusage __user *ru)
1462 {
1463         DECLARE_WAITQUEUE(wait, current);
1464         struct task_struct *tsk;
1465         int flag, retval;
1466
1467         add_wait_queue(&current->signal->wait_chldexit,&wait);
1468 repeat:
1469         /*
1470          * We will set this flag if we see any child that might later
1471          * match our criteria, even if we are not able to reap it yet.
1472          */
1473         flag = 0;
1474         current->state = TASK_INTERRUPTIBLE;
1475         read_lock(&tasklist_lock);
1476         tsk = current;
1477         do {
1478                 struct task_struct *p;
1479                 struct list_head *_p;
1480                 int ret;
1481
1482                 list_for_each(_p,&tsk->children) {
1483                         p = list_entry(_p, struct task_struct, sibling);
1484
1485                         ret = eligible_child(pid, options, p);
1486                         if (!ret)
1487                                 continue;
1488
1489                         switch (p->state) {
1490                         case TASK_TRACED:
1491                                 /*
1492                                  * When we hit the race with PTRACE_ATTACH,
1493                                  * we will not report this child.  But the
1494                                  * race means it has not yet been moved to
1495                                  * our ptrace_children list, so we need to
1496                                  * set the flag here to avoid a spurious ECHILD
1497                                  * when the race happens with the only child.
1498                                  */
1499                                 flag = 1;
1500                                 if (!my_ptrace_child(p))
1501                                         continue;
1502                                 /*FALLTHROUGH*/
1503                         case TASK_STOPPED:
1504                                 /*
1505                                  * It's stopped now, so it might later
1506                                  * continue, exit, or stop again.
1507                                  */
1508                                 flag = 1;
1509                                 if (!(options & WUNTRACED) &&
1510                                     !my_ptrace_child(p))
1511                                         continue;
1512                                 retval = wait_task_stopped(p, ret == 2,
1513                                                            (options & WNOWAIT),
1514                                                            infop,
1515                                                            stat_addr, ru);
1516                                 if (retval == -EAGAIN)
1517                                         goto repeat;
1518                                 if (retval != 0) /* He released the lock.  */
1519                                         goto end;
1520                                 break;
1521                         default:
1522                         // case EXIT_DEAD:
1523                                 if (p->exit_state == EXIT_DEAD)
1524                                         continue;
1525                         // case EXIT_ZOMBIE:
1526                                 if (p->exit_state == EXIT_ZOMBIE) {
1527                                         /*
1528                                          * Eligible but we cannot release
1529                                          * it yet:
1530                                          */
1531                                         if (ret == 2)
1532                                                 goto check_continued;
1533                                         if (!likely(options & WEXITED))
1534                                                 continue;
1535                                         retval = wait_task_zombie(
1536                                                 p, (options & WNOWAIT),
1537                                                 infop, stat_addr, ru);
1538                                         /* He released the lock.  */
1539                                         if (retval != 0)
1540                                                 goto end;
1541                                         break;
1542                                 }
1543 check_continued:
1544                                 /*
1545                                  * It's running now, so it might later
1546                                  * exit, stop, or stop and then continue.
1547                                  */
1548                                 flag = 1;
1549                                 if (!unlikely(options & WCONTINUED))
1550                                         continue;
1551                                 retval = wait_task_continued(
1552                                         p, (options & WNOWAIT),
1553                                         infop, stat_addr, ru);
1554                                 if (retval != 0) /* He released the lock.  */
1555                                         goto end;
1556                                 break;
1557                         }
1558                 }
1559                 if (!flag) {
1560                         list_for_each(_p, &tsk->ptrace_children) {
1561                                 p = list_entry(_p, struct task_struct,
1562                                                 ptrace_list);
1563                                 if (!eligible_child(pid, options, p))
1564                                         continue;
1565                                 flag = 1;
1566                                 break;
1567                         }
1568                 }
1569                 if (options & __WNOTHREAD)
1570                         break;
1571                 tsk = next_thread(tsk);
1572                 BUG_ON(tsk->signal != current->signal);
1573         } while (tsk != current);
1574
1575         read_unlock(&tasklist_lock);
1576         if (flag) {
1577                 retval = 0;
1578                 if (options & WNOHANG)
1579                         goto end;
1580                 retval = -ERESTARTSYS;
1581                 if (signal_pending(current))
1582                         goto end;
1583                 schedule();
1584                 goto repeat;
1585         }
1586         retval = -ECHILD;
1587 end:
1588         current->state = TASK_RUNNING;
1589         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1590         if (infop) {
1591                 if (retval > 0)
1592                 retval = 0;
1593                 else {
1594                         /*
1595                          * For a WNOHANG return, clear out all the fields
1596                          * we would set so the user can easily tell the
1597                          * difference.
1598                          */
1599                         if (!retval)
1600                                 retval = put_user(0, &infop->si_signo);
1601                         if (!retval)
1602                                 retval = put_user(0, &infop->si_errno);
1603                         if (!retval)
1604                                 retval = put_user(0, &infop->si_code);
1605                         if (!retval)
1606                                 retval = put_user(0, &infop->si_pid);
1607                         if (!retval)
1608                                 retval = put_user(0, &infop->si_uid);
1609                         if (!retval)
1610                                 retval = put_user(0, &infop->si_status);
1611                 }
1612         }
1613         return retval;
1614 }
1615
1616 asmlinkage long sys_waitid(int which, pid_t pid,
1617                            struct siginfo __user *infop, int options,
1618                            struct rusage __user *ru)
1619 {
1620         long ret;
1621
1622         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1623                 return -EINVAL;
1624         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1625                 return -EINVAL;
1626
1627         switch (which) {
1628         case P_ALL:
1629                 pid = -1;
1630                 break;
1631         case P_PID:
1632                 if (pid <= 0)
1633                         return -EINVAL;
1634                 break;
1635         case P_PGID:
1636                 if (pid <= 0)
1637                         return -EINVAL;
1638                 pid = -pid;
1639                 break;
1640         default:
1641                 return -EINVAL;
1642         }
1643
1644         ret = do_wait(pid, options, infop, NULL, ru);
1645
1646         /* avoid REGPARM breakage on x86: */
1647         prevent_tail_call(ret);
1648         return ret;
1649 }
1650
1651 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1652                           int options, struct rusage __user *ru)
1653 {
1654         long ret;
1655
1656         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1657                         __WNOTHREAD|__WCLONE|__WALL))
1658                 return -EINVAL;
1659         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1660
1661         /* avoid REGPARM breakage on x86: */
1662         prevent_tail_call(ret);
1663         return ret;
1664 }
1665
1666 #ifdef __ARCH_WANT_SYS_WAITPID
1667
1668 /*
1669  * sys_waitpid() remains for compatibility. waitpid() should be
1670  * implemented by calling sys_wait4() from libc.a.
1671  */
1672 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1673 {
1674         return sys_wait4(pid, stat_addr, options, NULL);
1675 }
1676
1677 #endif