kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/config.h>
   8 #include <linux/mm.h>
   9 #include <linux/slab.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/module.h>
  13 #include <linux/capability.h>
  14 #include <linux/completion.h>
  15 #include <linux/personality.h>
  16 #include <linux/tty.h>
  17 #include <linux/namespace.h>
  18 #include <linux/key.h>
  19 #include <linux/security.h>
  20 #include <linux/cpu.h>
  21 #include <linux/acct.h>
  22 #include <linux/file.h>
  23 #include <linux/binfmts.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/profile.h>
  26 #include <linux/mount.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/mempolicy.h>
  29 #include <linux/cpuset.h>
  30 #include <linux/syscalls.h>
  31 #include <linux/signal.h>
  32 #include <linux/cn_proc.h>
  33 #include <linux/mutex.h>
  34 #include <linux/vs_limit.h>
  35 #include <linux/vs_context.h>
  36 #include <linux/vs_network.h>
  37 #include <linux/vs_cvirt.h>
  38
  39 #include <asm/uaccess.h>
  40 #include <asm/unistd.h>
  41 #include <asm/pgtable.h>
  42 #include <asm/mmu_context.h>
  43
  44 extern void sem_exit (void);
  45 extern struct task_struct *child_reaper;
  46
  47 int getrusage(struct task_struct *, int, struct rusage __user *);
  48
  49 static void exit_mm(struct task_struct * tsk);
  50
  51 static void __unhash_process(struct task_struct *p)
  52 {
  53         nr_threads--;
  54         detach_pid(p, PIDTYPE_PID);
  55         detach_pid(p, PIDTYPE_TGID);
  56         if (thread_group_leader(p)) {
  57                 detach_pid(p, PIDTYPE_PGID);
  58                 detach_pid(p, PIDTYPE_SID);
  59                 if (p->pid)
  60                         __get_cpu_var(process_counts)--;
  61         }
  62
  63         REMOVE_LINKS(p);
  64 }
  65
  66 void release_task(struct task_struct * p)
  67 {
  68         int zap_leader;
  69         task_t *leader;
  70         struct dentry *proc_dentry;
  71
  72 repeat:
  73         atomic_dec(&p->user->processes);
  74         spin_lock(&p->proc_lock);
  75         proc_dentry = proc_pid_unhash(p);
  76         write_lock_irq(&tasklist_lock);
  77         if (unlikely(p->ptrace))
  78                 __ptrace_unlink(p);
  79         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  80         __exit_signal(p);
  81         /*
  82          * Note that the fastpath in sys_times depends on __exit_signal having
  83          * updated the counters before a task is removed from the tasklist of
  84          * the process by __unhash_process.
  85          */
  86         __unhash_process(p);
  87
  88         /*
  89          * If we are the last non-leader member of the thread
  90          * group, and the leader is zombie, then notify the
  91          * group leader's parent process. (if it wants notification.)
  92          */
  93         zap_leader = 0;
  94         leader = p->group_leader;
  95         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
  96                 BUG_ON(leader->exit_signal == -1);
  97                 do_notify_parent(leader, leader->exit_signal);
  98                 /*
  99                  * If we were the last child thread and the leader has
 100                  * exited already, and the leader's parent ignores SIGCHLD,
 101                  * then we are the one who should release the leader.
 102                  *
 103                  * do_notify_parent() will have marked it self-reaping in
 104                  * that case.
 105                  */
 106                 zap_leader = (leader->exit_signal == -1);
 107         }
 108
 109         sched_exit(p);
 110         write_unlock_irq(&tasklist_lock);
 111         spin_unlock(&p->proc_lock);
 112         proc_pid_flush(proc_dentry);
 113         release_thread(p);
 114         put_task_struct(p);
 115
 116         p = leader;
 117         if (unlikely(zap_leader))
 118                 goto repeat;
 119 }
 120
 121 /* we are using it only for SMP init */
 122
 123 void unhash_process(struct task_struct *p)
 124 {
 125         struct dentry *proc_dentry;
 126
 127         spin_lock(&p->proc_lock);
 128         proc_dentry = proc_pid_unhash(p);
 129         write_lock_irq(&tasklist_lock);
 130         __unhash_process(p);
 131         write_unlock_irq(&tasklist_lock);
 132         spin_unlock(&p->proc_lock);
 133         proc_pid_flush(proc_dentry);
 134 }
 135
 136 /*
 137  * This checks not only the pgrp, but falls back on the pid if no
 138  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 139  * without this...
 140  */
 141 int session_of_pgrp(int pgrp)
 142 {
 143         struct task_struct *p;
 144         int sid = -1;
 145
 146         read_lock(&tasklist_lock);
 147         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 148                 if (p->signal->session > 0) {
 149                         sid = p->signal->session;
 150                         goto out;
 151                 }
 152         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 153         p = find_task_by_pid(pgrp);
 154         if (p)
 155                 sid = p->signal->session;
 156 out:
 157         read_unlock(&tasklist_lock);
 158
 159         return sid;
 160 }
 161
 162 /*
 163  * Determine if a process group is "orphaned", according to the POSIX
 164  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 165  * by terminal-generated stop signals.  Newly orphaned process groups are
 166  * to receive a SIGHUP and a SIGCONT.
 167  *
 168  * "I ask you, have you ever known what it is to be an orphan?"
 169  */
 170 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 171 {
 172         struct task_struct *p;
 173         int ret = 1;
 174
 175         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 176                 if (p == ignored_task
 177                                 || p->exit_state
 178                                 || p->real_parent->pid == 1)
 179                         continue;
 180                 if (process_group(p->real_parent) != pgrp
 181                             && p->real_parent->signal->session == p->signal->session) {
 182                         ret = 0;
 183                         break;
 184                 }
 185         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 186         return ret;     /* (sighing) "Often!" */
 187 }
 188
 189 int is_orphaned_pgrp(int pgrp)
 190 {
 191         int retval;
 192
 193         read_lock(&tasklist_lock);
 194         retval = will_become_orphaned_pgrp(pgrp, NULL);
 195         read_unlock(&tasklist_lock);
 196
 197         return retval;
 198 }
 199
 200 static int has_stopped_jobs(int pgrp)
 201 {
 202         int retval = 0;
 203         struct task_struct *p;
 204
 205         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 206                 if (p->state != TASK_STOPPED)
 207                         continue;
 208
 209                 /* If p is stopped by a debugger on a signal that won't
 210                    stop it, then don't count p as stopped.  This isn't
 211                    perfect but it's a good approximation.  */
 212                 if (unlikely (p->ptrace)
 213                     && p->exit_code != SIGSTOP
 214                     && p->exit_code != SIGTSTP
 215                     && p->exit_code != SIGTTOU
 216                     && p->exit_code != SIGTTIN)
 217                         continue;
 218
 219                 retval = 1;
 220                 break;
 221         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 222         return retval;
 223 }
 224
 225 /**
 226  * reparent_to_init - Reparent the calling kernel thread to the init task.
 227  *
 228  * If a kernel thread is launched as a result of a system call, or if
 229  * it ever exits, it should generally reparent itself to init so that
 230  * it is correctly cleaned up on exit.
 231  *
 232  * The various task state such as scheduling policy and priority may have
 233  * been inherited from a user process, so we reset them to sane values here.
 234  *
 235  * NOTE that reparent_to_init() gives the caller full capabilities.
 236  */
 237 static void reparent_to_init(void)
 238 {
 239         write_lock_irq(&tasklist_lock);
 240
 241         ptrace_unlink(current);
 242         /* Reparent to init */
 243         REMOVE_LINKS(current);
 244         current->parent = child_reaper;
 245         current->real_parent = child_reaper;
 246         SET_LINKS(current);
 247
 248         /* Set the exit signal to SIGCHLD so we signal init on exit */
 249         current->exit_signal = SIGCHLD;
 250
 251         if ((current->policy == SCHED_NORMAL ||
 252                         current->policy == SCHED_BATCH)
 253                                 && (task_nice(current) < 0))
 254                 set_user_nice(current, 0);
 255         /* cpus_allowed? */
 256         /* rt_priority? */
 257         /* signals? */
 258         security_task_reparent_to_init(current);
 259         memcpy(current->signal->rlim, init_task.signal->rlim,
 260                sizeof(current->signal->rlim));
 261         atomic_inc(&(INIT_USER->__count));
 262         write_unlock_irq(&tasklist_lock);
 263         switch_uid(INIT_USER);
 264 }
 265
 266 void __set_special_pids(pid_t session, pid_t pgrp)
 267 {
 268         struct task_struct *curr = current->group_leader;
 269
 270         if (curr->signal->session != session) {
 271                 detach_pid(curr, PIDTYPE_SID);
 272                 curr->signal->session = session;
 273                 attach_pid(curr, PIDTYPE_SID, session);
 274         }
 275         if (process_group(curr) != pgrp) {
 276                 detach_pid(curr, PIDTYPE_PGID);
 277                 curr->signal->pgrp = pgrp;
 278                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 279         }
 280 }
 281
 282 void set_special_pids(pid_t session, pid_t pgrp)
 283 {
 284         write_lock_irq(&tasklist_lock);
 285         __set_special_pids(session, pgrp);
 286         write_unlock_irq(&tasklist_lock);
 287 }
 288
 289 /*
 290  * Let kernel threads use this to say that they
 291  * allow a certain signal (since daemonize() will
 292  * have disabled all of them by default).
 293  */
 294 int allow_signal(int sig)
 295 {
 296         if (!valid_signal(sig) || sig < 1)
 297                 return -EINVAL;
 298
 299         spin_lock_irq(&current->sighand->siglock);
 300         sigdelset(&current->blocked, sig);
 301         if (!current->mm) {
 302                 /* Kernel threads handle their own signals.
 303                    Let the signal code know it'll be handled, so
 304                    that they don't get converted to SIGKILL or
 305                    just silently dropped */
 306                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 307         }
 308         recalc_sigpending();
 309         spin_unlock_irq(&current->sighand->siglock);
 310         return 0;
 311 }
 312
 313 EXPORT_SYMBOL(allow_signal);
 314
 315 int disallow_signal(int sig)
 316 {
 317         if (!valid_signal(sig) || sig < 1)
 318                 return -EINVAL;
 319
 320         spin_lock_irq(&current->sighand->siglock);
 321         sigaddset(&current->blocked, sig);
 322         recalc_sigpending();
 323         spin_unlock_irq(&current->sighand->siglock);
 324         return 0;
 325 }
 326
 327 EXPORT_SYMBOL(disallow_signal);
 328
 329 /*
 330  *      Put all the gunge required to become a kernel thread without
 331  *      attached user resources in one place where it belongs.
 332  */
 333
 334 void daemonize(const char *name, ...)
 335 {
 336         va_list args;
 337         struct fs_struct *fs;
 338         sigset_t blocked;
 339
 340         va_start(args, name);
 341         vsnprintf(current->comm, sizeof(current->comm), name, args);
 342         va_end(args);
 343
 344         /*
 345          * If we were started as result of loading a module, close all of the
 346          * user space pages.  We don't need them, and if we didn't close them
 347          * they would be locked into memory.
 348          */
 349         exit_mm(current);
 350
 351         set_special_pids(1, 1);
 352         down(&tty_sem);
 353         current->signal->tty = NULL;
 354         up(&tty_sem);
 355
 356         /* Block and flush all signals */
 357         sigfillset(&blocked);
 358         sigprocmask(SIG_BLOCK, &blocked, NULL);
 359         flush_signals(current);
 360
 361         /* Become as one with the init task */
 362
 363         exit_fs(current);       /* current->fs->count--; */
 364         fs = init_task.fs;
 365         current->fs = fs;
 366         atomic_inc(&fs->count);
 367         exit_namespace(current);
 368         current->namespace = init_task.namespace;
 369         get_namespace(current->namespace);
 370         exit_files(current);
 371         current->files = init_task.files;
 372         atomic_inc(&current->files->count);
 373
 374         reparent_to_init();
 375 }
 376
 377 EXPORT_SYMBOL(daemonize);
 378
 379 static void close_files(struct files_struct * files)
 380 {
 381         int i, j;
 382         struct fdtable *fdt;
 383
 384         j = 0;
 385
 386         /*
 387          * It is safe to dereference the fd table without RCU or
 388          * ->file_lock because this is the last reference to the
 389          * files structure.
 390          */
 391         fdt = files_fdtable(files);
 392         for (;;) {
 393                 unsigned long set;
 394                 i = j * __NFDBITS;
 395                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 396                         break;
 397                 set = fdt->open_fds->fds_bits[j++];
 398                 while (set) {
 399                         if (set & 1) {
 400                                 struct file * file = xchg(&fdt->fd[i], NULL);
 401                                 if (file)
 402                                         filp_close(file, files);
 403                                 vx_openfd_dec(i);
 404                         }
 405                         i++;
 406                         set >>= 1;
 407                 }
 408         }
 409 }
 410
 411 struct files_struct *get_files_struct(struct task_struct *task)
 412 {
 413         struct files_struct *files;
 414
 415         task_lock(task);
 416         files = task->files;
 417         if (files)
 418                 atomic_inc(&files->count);
 419         task_unlock(task);
 420
 421         return files;
 422 }
 423
 424 void fastcall put_files_struct(struct files_struct *files)
 425 {
 426         struct fdtable *fdt;
 427
 428         if (atomic_dec_and_test(&files->count)) {
 429                 close_files(files);
 430                 /*
 431                  * Free the fd and fdset arrays if we expanded them.
 432                  * If the fdtable was embedded, pass files for freeing
 433                  * at the end of the RCU grace period. Otherwise,
 434                  * you can free files immediately.
 435                  */
 436                 fdt = files_fdtable(files);
 437                 if (fdt == &files->fdtab)
 438                         fdt->free_files = files;
 439                 else
 440                         kmem_cache_free(files_cachep, files);
 441                 free_fdtable(fdt);
 442         }
 443 }
 444
 445 EXPORT_SYMBOL(put_files_struct);
 446
 447 static inline void __exit_files(struct task_struct *tsk)
 448 {
 449         struct files_struct * files = tsk->files;
 450
 451         if (files) {
 452                 task_lock(tsk);
 453                 tsk->files = NULL;
 454                 task_unlock(tsk);
 455                 put_files_struct(files);
 456         }
 457 }
 458
 459 void exit_files(struct task_struct *tsk)
 460 {
 461         __exit_files(tsk);
 462 }
 463
 464 static inline void __put_fs_struct(struct fs_struct *fs)
 465 {
 466         /* No need to hold fs->lock if we are killing it */
 467         if (atomic_dec_and_test(&fs->count)) {
 468                 dput(fs->root);
 469                 mntput(fs->rootmnt);
 470                 dput(fs->pwd);
 471                 mntput(fs->pwdmnt);
 472                 if (fs->altroot) {
 473                         dput(fs->altroot);
 474                         mntput(fs->altrootmnt);
 475                 }
 476                 kmem_cache_free(fs_cachep, fs);
 477         }
 478 }
 479
 480 void put_fs_struct(struct fs_struct *fs)
 481 {
 482         __put_fs_struct(fs);
 483 }
 484
 485 static inline void __exit_fs(struct task_struct *tsk)
 486 {
 487         struct fs_struct * fs = tsk->fs;
 488
 489         if (fs) {
 490                 task_lock(tsk);
 491                 tsk->fs = NULL;
 492                 task_unlock(tsk);
 493                 __put_fs_struct(fs);
 494         }
 495 }
 496
 497 void exit_fs(struct task_struct *tsk)
 498 {
 499         __exit_fs(tsk);
 500 }
 501
 502 EXPORT_SYMBOL_GPL(exit_fs);
 503
 504 /*
 505  * Turn us into a lazy TLB process if we
 506  * aren't already..
 507  */
 508 static void exit_mm(struct task_struct * tsk)
 509 {
 510         struct mm_struct *mm = tsk->mm;
 511
 512         mm_release(tsk, mm);
 513         if (!mm)
 514                 return;
 515         /*
 516          * Serialize with any possible pending coredump.
 517          * We must hold mmap_sem around checking core_waiters
 518          * and clearing tsk->mm.  The core-inducing thread
 519          * will increment core_waiters for each thread in the
 520          * group with ->mm != NULL.
 521          */
 522         down_read(&mm->mmap_sem);
 523         if (mm->core_waiters) {
 524                 up_read(&mm->mmap_sem);
 525                 down_write(&mm->mmap_sem);
 526                 if (!--mm->core_waiters)
 527                         complete(mm->core_startup_done);
 528                 up_write(&mm->mmap_sem);
 529
 530                 wait_for_completion(&mm->core_done);
 531                 down_read(&mm->mmap_sem);
 532         }
 533         atomic_inc(&mm->mm_count);
 534         if (mm != tsk->active_mm) BUG();
 535         /* more a memory barrier than a real lock */
 536         task_lock(tsk);
 537         tsk->mm = NULL;
 538         up_read(&mm->mmap_sem);
 539         enter_lazy_tlb(mm, current);
 540         task_unlock(tsk);
 541         mmput(mm);
 542 }
 543
 544 static inline void choose_new_parent(task_t *p, task_t *reaper)
 545 {
 546         /* check for reaper context */
 547         vxwprintk((p->xid != reaper->xid) && (reaper != child_reaper),
 548                 "rogue reaper: %p[%d,#%u] <> %p[%d,#%u]",
 549                 p, p->pid, p->xid, reaper, reaper->pid, reaper->xid);
 550
 551         /*
 552          * Make sure we're not reparenting to ourselves and that
 553          * the parent is not a zombie.
 554          */
 555         BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
 556         p->real_parent = reaper;
 557 }
 558
 559 static void reparent_thread(task_t *p, task_t *father, int traced)
 560 {
 561         /* We don't want people slaying init.  */
 562         if (p->exit_signal != -1)
 563                 p->exit_signal = SIGCHLD;
 564
 565         if (p->pdeath_signal)
 566                 /* We already hold the tasklist_lock here.  */
 567                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 568
 569         /* Move the child from its dying parent to the new one.  */
 570         if (unlikely(traced)) {
 571                 /* Preserve ptrace links if someone else is tracing this child.  */
 572                 list_del_init(&p->ptrace_list);
 573                 if (p->parent != p->real_parent)
 574                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 575         } else {
 576                 /* If this child is being traced, then we're the one tracing it
 577                  * anyway, so let go of it.
 578                  */
 579                 p->ptrace = 0;
 580                 list_del_init(&p->sibling);
 581                 p->parent = p->real_parent;
 582                 list_add_tail(&p->sibling, &p->parent->children);
 583
 584                 /* If we'd notified the old parent about this child's death,
 585                  * also notify the new parent.
 586                  */
 587                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 588                     thread_group_empty(p))
 589                         do_notify_parent(p, p->exit_signal);
 590                 else if (p->state == TASK_TRACED) {
 591                         /*
 592                          * If it was at a trace stop, turn it into
 593                          * a normal stop since it's no longer being
 594                          * traced.
 595                          */
 596                         ptrace_untrace(p);
 597                 }
 598         }
 599
 600         /*
 601          * process group orphan check
 602          * Case ii: Our child is in a different pgrp
 603          * than we are, and it was the only connection
 604          * outside, so the child pgrp is now orphaned.
 605          */
 606         if ((process_group(p) != process_group(father)) &&
 607             (p->signal->session == father->signal->session)) {
 608                 int pgrp = process_group(p);
 609
 610                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 611                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 612                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 613                 }
 614         }
 615 }
 616
 617 /*
 618  * When we die, we re-parent all our children.
 619  * Try to give them to another thread in our thread
 620  * group, and if no such member exists, give it to
 621  * the global child reaper process (ie "init")
 622  */
 623 static void forget_original_parent(struct task_struct * father,
 624                                           struct list_head *to_release)
 625 {
 626         struct task_struct *p, *reaper = father;
 627         struct list_head *_p, *_n;
 628
 629         do {
 630                 reaper = next_thread(reaper);
 631                 if (reaper == father) {
 632                         reaper = vx_child_reaper(father);
 633                         break;
 634                 }
 635         } while (reaper->exit_state);
 636
 637         /*
 638          * There are only two places where our children can be:
 639          *
 640          * - in our child list
 641          * - in our ptraced child list
 642          *
 643          * Search them and reparent children.
 644          */
 645         list_for_each_safe(_p, _n, &father->children) {
 646                 int ptrace;
 647                 p = list_entry(_p,struct task_struct,sibling);
 648
 649                 ptrace = p->ptrace;
 650
 651                 /* if father isn't the real parent, then ptrace must be enabled */
 652                 BUG_ON(father != p->real_parent && !ptrace);
 653
 654                 if (father == p->real_parent) {
 655                         /* reparent with a reaper, real father it's us */
 656                         choose_new_parent(p, vx_child_reaper(p));
 657                         reparent_thread(p, father, 0);
 658                 } else {
 659                         /* reparent ptraced task to its real parent */
 660                         __ptrace_unlink (p);
 661                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 662                             thread_group_empty(p))
 663                                 do_notify_parent(p, p->exit_signal);
 664                 }
 665
 666                 /*
 667                  * if the ptraced child is a zombie with exit_signal == -1
 668                  * we must collect it before we exit, or it will remain
 669                  * zombie forever since we prevented it from self-reap itself
 670                  * while it was being traced by us, to be able to see it in wait4.
 671                  */
 672                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 673                         list_add(&p->ptrace_list, to_release);
 674         }
 675         list_for_each_safe(_p, _n, &father->ptrace_children) {
 676                 p = list_entry(_p,struct task_struct,ptrace_list);
 677
 678                 choose_new_parent(p, reaper);
 679                 reparent_thread(p, father, 1);
 680         }
 681 }
 682
 683 /*
 684  * Send signals to all our closest relatives so that they know
 685  * to properly mourn us..
 686  */
 687 static void exit_notify(struct task_struct *tsk)
 688 {
 689         int state;
 690         struct task_struct *t;
 691         struct list_head ptrace_dead, *_p, *_n;
 692
 693         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 694             && !thread_group_empty(tsk)) {
 695                 /*
 696                  * This occurs when there was a race between our exit
 697                  * syscall and a group signal choosing us as the one to
 698                  * wake up.  It could be that we are the only thread
 699                  * alerted to check for pending signals, but another thread
 700                  * should be woken now to take the signal since we will not.
 701                  * Now we'll wake all the threads in the group just to make
 702                  * sure someone gets all the pending signals.
 703                  */
 704                 read_lock(&tasklist_lock);
 705                 spin_lock_irq(&tsk->sighand->siglock);
 706                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 707                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 708                                 recalc_sigpending_tsk(t);
 709                                 if (signal_pending(t))
 710                                         signal_wake_up(t, 0);
 711                         }
 712                 spin_unlock_irq(&tsk->sighand->siglock);
 713                 read_unlock(&tasklist_lock);
 714         }
 715
 716         write_lock_irq(&tasklist_lock);
 717
 718         /*
 719          * This does two things:
 720          *
 721          * A.  Make init inherit all the child processes
 722          * B.  Check to see if any process groups have become orphaned
 723          *      as a result of our exiting, and if they have any stopped
 724          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 725          */
 726
 727         INIT_LIST_HEAD(&ptrace_dead);
 728         forget_original_parent(tsk, &ptrace_dead);
 729         BUG_ON(!list_empty(&tsk->children));
 730         BUG_ON(!list_empty(&tsk->ptrace_children));
 731
 732         /*
 733          * Check to see if any process groups have become orphaned
 734          * as a result of our exiting, and if they have any stopped
 735          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 736          *
 737          * Case i: Our father is in a different pgrp than we are
 738          * and we were the only connection outside, so our pgrp
 739          * is about to become orphaned.
 740          */
 741
 742         t = tsk->real_parent;
 743
 744         if ((process_group(t) != process_group(tsk)) &&
 745             (t->signal->session == tsk->signal->session) &&
 746             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 747             has_stopped_jobs(process_group(tsk))) {
 748                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 749                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 750         }
 751
 752         /* Let father know we died
 753          *
 754          * Thread signals are configurable, but you aren't going to use
 755          * that to send signals to arbitary processes.
 756          * That stops right now.
 757          *
 758          * If the parent exec id doesn't match the exec id we saved
 759          * when we started then we know the parent has changed security
 760          * domain.
 761          *
 762          * If our self_exec id doesn't match our parent_exec_id then
 763          * we have changed execution domain as these two values started
 764          * the same after a fork.
 765          *
 766          */
 767
 768         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 769             ( tsk->parent_exec_id != t->self_exec_id  ||
 770               tsk->self_exec_id != tsk->parent_exec_id)
 771             && !capable(CAP_KILL))
 772                 tsk->exit_signal = SIGCHLD;
 773
 774
 775         /* If something other than our normal parent is ptracing us, then
 776          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 777          * only has special meaning to our real parent.
 778          */
 779         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 780                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 781                 do_notify_parent(tsk, signal);
 782         } else if (tsk->ptrace) {
 783                 do_notify_parent(tsk, SIGCHLD);
 784         }
 785
 786         state = EXIT_ZOMBIE;
 787         if (tsk->exit_signal == -1 &&
 788             (likely(tsk->ptrace == 0) ||
 789              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 790                 state = EXIT_DEAD;
 791         tsk->exit_state = state;
 792
 793         write_unlock_irq(&tasklist_lock);
 794
 795         list_for_each_safe(_p, _n, &ptrace_dead) {
 796                 list_del_init(_p);
 797                 t = list_entry(_p,struct task_struct,ptrace_list);
 798                 release_task(t);
 799         }
 800
 801         /* If the process is dead, release it - nobody will wait for it */
 802         if (state == EXIT_DEAD)
 803                 release_task(tsk);
 804 }
 805
 806 fastcall NORET_TYPE void do_exit(long code)
 807 {
 808         struct task_struct *tsk = current;
 809         int group_dead;
 810
 811         profile_task_exit(tsk);
 812
 813         WARN_ON(atomic_read(&tsk->fs_excl));
 814
 815         if (unlikely(in_interrupt()))
 816                 panic("Aiee, killing interrupt handler!");
 817         if (unlikely(!tsk->pid))
 818                 panic("Attempted to kill the idle task!");
 819         if (unlikely(tsk->pid == 1))
 820                 panic("Attempted to kill init!");
 821         if (tsk->io_context)
 822                 exit_io_context();
 823
 824         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 825                 current->ptrace_message = code;
 826                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 827         }
 828
 829         /*
 830          * We're taking recursive faults here in do_exit. Safest is to just
 831          * leave this task alone and wait for reboot.
 832          */
 833         if (unlikely(tsk->flags & PF_EXITING)) {
 834                 printk(KERN_ALERT
 835                         "Fixing recursive fault but reboot is needed!\n");
 836                 set_current_state(TASK_UNINTERRUPTIBLE);
 837                 schedule();
 838         }
 839
 840         tsk->flags |= PF_EXITING;
 841
 842         if (unlikely(in_atomic()))
 843                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 844                                 current->comm, current->pid,
 845                                 preempt_count());
 846
 847         acct_update_integrals(tsk);
 848         if (tsk->mm) {
 849                 update_hiwater_rss(tsk->mm);
 850                 update_hiwater_vm(tsk->mm);
 851         }
 852         group_dead = atomic_dec_and_test(&tsk->signal->live);
 853         if (group_dead) {
 854                 hrtimer_cancel(&tsk->signal->real_timer);
 855                 exit_itimers(tsk->signal);
 856                 acct_process(code);
 857         }
 858         exit_mm(tsk);
 859
 860         exit_sem(tsk);
 861         __exit_files(tsk);
 862         __exit_fs(tsk);
 863         exit_namespace(tsk);
 864         exit_thread();
 865         cpuset_exit(tsk);
 866         exit_keys(tsk);
 867
 868         if (group_dead && tsk->signal->leader)
 869                 disassociate_ctty(1);
 870
 871         module_put(task_thread_info(tsk)->exec_domain->module);
 872         if (tsk->binfmt)
 873                 module_put(tsk->binfmt->module);
 874
 875         tsk->exit_code = code;
 876         proc_exit_connector(tsk);
 877         /* needs to stay before exit_notify() */
 878         exit_vx_info_early(tsk, code);
 879         exit_notify(tsk);
 880 #ifdef CONFIG_NUMA
 881         mpol_free(tsk->mempolicy);
 882         tsk->mempolicy = NULL;
 883 #endif
 884         /*
 885          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
 886          */
 887         mutex_debug_check_no_locks_held(tsk);
 888
 889         /* needs to stay after exit_notify() */
 890         exit_vx_info(tsk, code);
 891         exit_nx_info(tsk);
 892
 893         /* PF_DEAD causes final put_task_struct after we schedule. */
 894         preempt_disable();
 895         BUG_ON(tsk->flags & PF_DEAD);
 896         tsk->flags |= PF_DEAD;
 897
 898         schedule();
 899         BUG();
 900         /* Avoid "noreturn function does return".  */
 901         for (;;) ;
 902 }
 903
 904 EXPORT_SYMBOL_GPL(do_exit);
 905
 906 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 907 {
 908         if (comp)
 909                 complete(comp);
 910
 911         do_exit(code);
 912 }
 913
 914 EXPORT_SYMBOL(complete_and_exit);
 915
 916 asmlinkage long sys_exit(int error_code)
 917 {
 918         do_exit((error_code&0xff)<<8);
 919 }
 920
 921 task_t fastcall *next_thread(const task_t *p)
 922 {
 923         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 924 }
 925
 926 EXPORT_SYMBOL(next_thread);
 927
 928 /*
 929  * Take down every thread in the group.  This is called by fatal signals
 930  * as well as by sys_exit_group (below).
 931  */
 932 NORET_TYPE void
 933 do_group_exit(int exit_code)
 934 {
 935         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 936
 937         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 938                 exit_code = current->signal->group_exit_code;
 939         else if (!thread_group_empty(current)) {
 940                 struct signal_struct *const sig = current->signal;
 941                 struct sighand_struct *const sighand = current->sighand;
 942                 read_lock(&tasklist_lock);
 943                 spin_lock_irq(&sighand->siglock);
 944                 if (sig->flags & SIGNAL_GROUP_EXIT)
 945                         /* Another thread got here before we took the lock.  */
 946                         exit_code = sig->group_exit_code;
 947                 else {
 948                         sig->group_exit_code = exit_code;
 949                         zap_other_threads(current);
 950                 }
 951                 spin_unlock_irq(&sighand->siglock);
 952                 read_unlock(&tasklist_lock);
 953         }
 954
 955         do_exit(exit_code);
 956         /* NOTREACHED */
 957 }
 958
 959 /*
 960  * this kills every thread in the thread group. Note that any externally
 961  * wait4()-ing process will get the correct exit code - even if this
 962  * thread is not the thread group leader.
 963  */
 964 asmlinkage void sys_exit_group(int error_code)
 965 {
 966         do_group_exit((error_code & 0xff) << 8);
 967 }
 968
 969 static int eligible_child(pid_t pid, int options, task_t *p)
 970 {
 971         if (pid > 0) {
 972                 if (p->pid != pid)
 973                         return 0;
 974         } else if (!pid) {
 975                 if (process_group(p) != process_group(current))
 976                         return 0;
 977         } else if (pid != -1) {
 978                 if (process_group(p) != -pid)
 979                         return 0;
 980         }
 981
 982         /*
 983          * Do not consider detached threads that are
 984          * not ptraced:
 985          */
 986         if (p->exit_signal == -1 && !p->ptrace)
 987                 return 0;
 988
 989         /* Wait for all children (clone and not) if __WALL is set;
 990          * otherwise, wait for clone children *only* if __WCLONE is
 991          * set; otherwise, wait for non-clone children *only*.  (Note:
 992          * A "clone" child here is one that reports to its parent
 993          * using a signal other than SIGCHLD.) */
 994         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 995             && !(options & __WALL))
 996                 return 0;
 997         /*
 998          * Do not consider thread group leaders that are
 999          * in a non-empty thread group:
1000          */
1001         if (current->tgid != p->tgid && delay_group_leader(p))
1002                 return 2;
1003
1004         if (security_task_wait(p))
1005                 return 0;
1006
1007         return 1;
1008 }
1009
1010 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
1011                                int why, int status,
1012                                struct siginfo __user *infop,
1013                                struct rusage __user *rusagep)
1014 {
1015         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1016         put_task_struct(p);
1017         if (!retval)
1018                 retval = put_user(SIGCHLD, &infop->si_signo);
1019         if (!retval)
1020                 retval = put_user(0, &infop->si_errno);
1021         if (!retval)
1022                 retval = put_user((short)why, &infop->si_code);
1023         if (!retval)
1024                 retval = put_user(pid, &infop->si_pid);
1025         if (!retval)
1026                 retval = put_user(uid, &infop->si_uid);
1027         if (!retval)
1028                 retval = put_user(status, &infop->si_status);
1029         if (!retval)
1030                 retval = pid;
1031         return retval;
1032 }
1033
1034 /*
1035  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1036  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1037  * the lock and this task is uninteresting.  If we return nonzero, we have
1038  * released the lock and the system call should return.
1039  */
1040 static int wait_task_zombie(task_t *p, int noreap,
1041                             struct siginfo __user *infop,
1042                             int __user *stat_addr, struct rusage __user *ru)
1043 {
1044         unsigned long state;
1045         int retval;
1046         int status;
1047
1048         if (unlikely(noreap)) {
1049                 pid_t pid = p->pid;
1050                 uid_t uid = p->uid;
1051                 int exit_code = p->exit_code;
1052                 int why, status;
1053
1054                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1055                         return 0;
1056                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1057                         return 0;
1058                 get_task_struct(p);
1059                 read_unlock(&tasklist_lock);
1060                 if ((exit_code & 0x7f) == 0) {
1061                         why = CLD_EXITED;
1062                         status = exit_code >> 8;
1063                 } else {
1064                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1065                         status = exit_code & 0x7f;
1066                 }
1067                 return wait_noreap_copyout(p, pid, uid, why,
1068                                            status, infop, ru);
1069         }
1070
1071         /*
1072          * Try to move the task's state to DEAD
1073          * only one thread is allowed to do this:
1074          */
1075         state = xchg(&p->exit_state, EXIT_DEAD);
1076         if (state != EXIT_ZOMBIE) {
1077                 BUG_ON(state != EXIT_DEAD);
1078                 return 0;
1079         }
1080         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1081                 /*
1082                  * This can only happen in a race with a ptraced thread
1083                  * dying on another processor.
1084                  */
1085                 return 0;
1086         }
1087
1088         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1089                 struct signal_struct *psig;
1090                 struct signal_struct *sig;
1091
1092                 /*
1093                  * The resource counters for the group leader are in its
1094                  * own task_struct.  Those for dead threads in the group
1095                  * are in its signal_struct, as are those for the child
1096                  * processes it has previously reaped.  All these
1097                  * accumulate in the parent's signal_struct c* fields.
1098                  *
1099                  * We don't bother to take a lock here to protect these
1100                  * p->signal fields, because they are only touched by
1101                  * __exit_signal, which runs with tasklist_lock
1102                  * write-locked anyway, and so is excluded here.  We do
1103                  * need to protect the access to p->parent->signal fields,
1104                  * as other threads in the parent group can be right
1105                  * here reaping other children at the same time.
1106                  */
1107                 spin_lock_irq(&p->parent->sighand->siglock);
1108                 psig = p->parent->signal;
1109                 sig = p->signal;
1110                 psig->cutime =
1111                         cputime_add(psig->cutime,
1112                         cputime_add(p->utime,
1113                         cputime_add(sig->utime,
1114                                     sig->cutime)));
1115                 psig->cstime =
1116                         cputime_add(psig->cstime,
1117                         cputime_add(p->stime,
1118                         cputime_add(sig->stime,
1119                                     sig->cstime)));
1120                 psig->cmin_flt +=
1121                         p->min_flt + sig->min_flt + sig->cmin_flt;
1122                 psig->cmaj_flt +=
1123                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1124                 psig->cnvcsw +=
1125                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1126                 psig->cnivcsw +=
1127                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1128                 spin_unlock_irq(&p->parent->sighand->siglock);
1129         }
1130
1131         /*
1132          * Now we are sure this task is interesting, and no other
1133          * thread can reap it because we set its state to EXIT_DEAD.
1134          */
1135         read_unlock(&tasklist_lock);
1136
1137         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1138         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1139                 ? p->signal->group_exit_code : p->exit_code;
1140         if (!retval && stat_addr)
1141                 retval = put_user(status, stat_addr);
1142         if (!retval && infop)
1143                 retval = put_user(SIGCHLD, &infop->si_signo);
1144         if (!retval && infop)
1145                 retval = put_user(0, &infop->si_errno);
1146         if (!retval && infop) {
1147                 int why;
1148
1149                 if ((status & 0x7f) == 0) {
1150                         why = CLD_EXITED;
1151                         status >>= 8;
1152                 } else {
1153                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1154                         status &= 0x7f;
1155                 }
1156                 retval = put_user((short)why, &infop->si_code);
1157                 if (!retval)
1158                         retval = put_user(status, &infop->si_status);
1159         }
1160         if (!retval && infop)
1161                 retval = put_user(p->pid, &infop->si_pid);
1162         if (!retval && infop)
1163                 retval = put_user(p->uid, &infop->si_uid);
1164         if (retval) {
1165                 // TODO: is this safe?
1166                 p->exit_state = EXIT_ZOMBIE;
1167                 return retval;
1168         }
1169         retval = p->pid;
1170         if (p->real_parent != p->parent) {
1171                 write_lock_irq(&tasklist_lock);
1172                 /* Double-check with lock held.  */
1173                 if (p->real_parent != p->parent) {
1174                         __ptrace_unlink(p);
1175                         // TODO: is this safe?
1176                         p->exit_state = EXIT_ZOMBIE;
1177                         /*
1178                          * If this is not a detached task, notify the parent.
1179                          * If it's still not detached after that, don't release
1180                          * it now.
1181                          */
1182                         if (p->exit_signal != -1) {
1183                                 do_notify_parent(p, p->exit_signal);
1184                                 if (p->exit_signal != -1)
1185                                         p = NULL;
1186                         }
1187                 }
1188                 write_unlock_irq(&tasklist_lock);
1189         }
1190         if (p != NULL)
1191                 release_task(p);
1192         BUG_ON(!retval);
1193         return retval;
1194 }
1195
1196 /*
1197  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1198  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1199  * the lock and this task is uninteresting.  If we return nonzero, we have
1200  * released the lock and the system call should return.
1201  */
1202 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1203                              struct siginfo __user *infop,
1204                              int __user *stat_addr, struct rusage __user *ru)
1205 {
1206         int retval, exit_code;
1207
1208         if (!p->exit_code)
1209                 return 0;
1210         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1211             p->signal && p->signal->group_stop_count > 0)
1212                 /*
1213                  * A group stop is in progress and this is the group leader.
1214                  * We won't report until all threads have stopped.
1215                  */
1216                 return 0;
1217
1218         /*
1219          * Now we are pretty sure this task is interesting.
1220          * Make sure it doesn't get reaped out from under us while we
1221          * give up the lock and then examine it below.  We don't want to
1222          * keep holding onto the tasklist_lock while we call getrusage and
1223          * possibly take page faults for user memory.
1224          */
1225         get_task_struct(p);
1226         read_unlock(&tasklist_lock);
1227
1228         if (unlikely(noreap)) {
1229                 pid_t pid = p->pid;
1230                 uid_t uid = p->uid;
1231                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1232
1233                 exit_code = p->exit_code;
1234                 if (unlikely(!exit_code) ||
1235                     unlikely(p->state & TASK_TRACED))
1236                         goto bail_ref;
1237                 return wait_noreap_copyout(p, pid, uid,
1238                                            why, (exit_code << 8) | 0x7f,
1239                                            infop, ru);
1240         }
1241
1242         write_lock_irq(&tasklist_lock);
1243
1244         /*
1245          * This uses xchg to be atomic with the thread resuming and setting
1246          * it.  It must also be done with the write lock held to prevent a
1247          * race with the EXIT_ZOMBIE case.
1248          */
1249         exit_code = xchg(&p->exit_code, 0);
1250         if (unlikely(p->exit_state)) {
1251                 /*
1252                  * The task resumed and then died.  Let the next iteration
1253                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1254                  * already be zero here if it resumed and did _exit(0).
1255                  * The task itself is dead and won't touch exit_code again;
1256                  * other processors in this function are locked out.
1257                  */
1258                 p->exit_code = exit_code;
1259                 exit_code = 0;
1260         }
1261         if (unlikely(exit_code == 0)) {
1262                 /*
1263                  * Another thread in this function got to it first, or it
1264                  * resumed, or it resumed and then died.
1265                  */
1266                 write_unlock_irq(&tasklist_lock);
1267 bail_ref:
1268                 put_task_struct(p);
1269                 /*
1270                  * We are returning to the wait loop without having successfully
1271                  * removed the process and having released the lock. We cannot
1272                  * continue, since the "p" task pointer is potentially stale.
1273                  *
1274                  * Return -EAGAIN, and do_wait() will restart the loop from the
1275                  * beginning. Do _not_ re-acquire the lock.
1276                  */
1277                 return -EAGAIN;
1278         }
1279
1280         /* move to end of parent's list to avoid starvation */
1281         remove_parent(p);
1282         add_parent(p, p->parent);
1283
1284         write_unlock_irq(&tasklist_lock);
1285
1286         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1287         if (!retval && stat_addr)
1288                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1289         if (!retval && infop)
1290                 retval = put_user(SIGCHLD, &infop->si_signo);
1291         if (!retval && infop)
1292                 retval = put_user(0, &infop->si_errno);
1293         if (!retval && infop)
1294                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1295                                           ? CLD_TRAPPED : CLD_STOPPED),
1296                                   &infop->si_code);
1297         if (!retval && infop)
1298                 retval = put_user(exit_code, &infop->si_status);
1299         if (!retval && infop)
1300                 retval = put_user(p->pid, &infop->si_pid);
1301         if (!retval && infop)
1302                 retval = put_user(p->uid, &infop->si_uid);
1303         if (!retval)
1304                 retval = p->pid;
1305         put_task_struct(p);
1306
1307         BUG_ON(!retval);
1308         return retval;
1309 }
1310
1311 /*
1312  * Handle do_wait work for one task in a live, non-stopped state.
1313  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1314  * the lock and this task is uninteresting.  If we return nonzero, we have
1315  * released the lock and the system call should return.
1316  */
1317 static int wait_task_continued(task_t *p, int noreap,
1318                                struct siginfo __user *infop,
1319                                int __user *stat_addr, struct rusage __user *ru)
1320 {
1321         int retval;
1322         pid_t pid;
1323         uid_t uid;
1324
1325         if (unlikely(!p->signal))
1326                 return 0;
1327
1328         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1329                 return 0;
1330
1331         spin_lock_irq(&p->sighand->siglock);
1332         /* Re-check with the lock held.  */
1333         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1334                 spin_unlock_irq(&p->sighand->siglock);
1335                 return 0;
1336         }
1337         if (!noreap)
1338                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1339         spin_unlock_irq(&p->sighand->siglock);
1340
1341         pid = p->pid;
1342         uid = p->uid;
1343         get_task_struct(p);
1344         read_unlock(&tasklist_lock);
1345
1346         if (!infop) {
1347                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1348                 put_task_struct(p);
1349                 if (!retval && stat_addr)
1350                         retval = put_user(0xffff, stat_addr);
1351                 if (!retval)
1352                         retval = p->pid;
1353         } else {
1354                 retval = wait_noreap_copyout(p, pid, uid,
1355                                              CLD_CONTINUED, SIGCONT,
1356                                              infop, ru);
1357                 BUG_ON(retval == 0);
1358         }
1359
1360         return retval;
1361 }
1362
1363
1364 static inline int my_ptrace_child(struct task_struct *p)
1365 {
1366         if (!(p->ptrace & PT_PTRACED))
1367                 return 0;
1368         if (!(p->ptrace & PT_ATTACHED))
1369                 return 1;
1370         /*
1371          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1372          * we are the attacher.  If we are the real parent, this is a race
1373          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1374          * which we have to switch the parent links, but has already set
1375          * the flags in p->ptrace.
1376          */
1377         return (p->parent != p->real_parent);
1378 }
1379
1380 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1381                     int __user *stat_addr, struct rusage __user *ru)
1382 {
1383         DECLARE_WAITQUEUE(wait, current);
1384         struct task_struct *tsk;
1385         int flag, retval;
1386
1387         add_wait_queue(&current->signal->wait_chldexit,&wait);
1388 repeat:
1389         /*
1390          * We will set this flag if we see any child that might later
1391          * match our criteria, even if we are not able to reap it yet.
1392          */
1393         flag = 0;
1394         current->state = TASK_INTERRUPTIBLE;
1395         read_lock(&tasklist_lock);
1396         tsk = current;
1397         do {
1398                 struct task_struct *p;
1399                 struct list_head *_p;
1400                 int ret;
1401
1402                 list_for_each(_p,&tsk->children) {
1403                         p = list_entry(_p,struct task_struct,sibling);
1404
1405                         ret = eligible_child(pid, options, p);
1406                         if (!ret)
1407                                 continue;
1408
1409                         switch (p->state) {
1410                         case TASK_TRACED:
1411                                 /*
1412                                  * When we hit the race with PTRACE_ATTACH,
1413                                  * we will not report this child.  But the
1414                                  * race means it has not yet been moved to
1415                                  * our ptrace_children list, so we need to
1416                                  * set the flag here to avoid a spurious ECHILD
1417                                  * when the race happens with the only child.
1418                                  */
1419                                 flag = 1;
1420                                 if (!my_ptrace_child(p))
1421                                         continue;
1422                                 /*FALLTHROUGH*/
1423                         case TASK_STOPPED:
1424                                 /*
1425                                  * It's stopped now, so it might later
1426                                  * continue, exit, or stop again.
1427                                  */
1428                                 flag = 1;
1429                                 if (!(options & WUNTRACED) &&
1430                                     !my_ptrace_child(p))
1431                                         continue;
1432                                 retval = wait_task_stopped(p, ret == 2,
1433                                                            (options & WNOWAIT),
1434                                                            infop,
1435                                                            stat_addr, ru);
1436                                 if (retval == -EAGAIN)
1437                                         goto repeat;
1438                                 if (retval != 0) /* He released the lock.  */
1439                                         goto end;
1440                                 break;
1441                         default:
1442                         // case EXIT_DEAD:
1443                                 if (p->exit_state == EXIT_DEAD)
1444                                         continue;
1445                         // case EXIT_ZOMBIE:
1446                                 if (p->exit_state == EXIT_ZOMBIE) {
1447                                         /*
1448                                          * Eligible but we cannot release
1449                                          * it yet:
1450                                          */
1451                                         if (ret == 2)
1452                                                 goto check_continued;
1453                                         if (!likely(options & WEXITED))
1454                                                 continue;
1455                                         retval = wait_task_zombie(
1456                                                 p, (options & WNOWAIT),
1457                                                 infop, stat_addr, ru);
1458                                         /* He released the lock.  */
1459                                         if (retval != 0)
1460                                                 goto end;
1461                                         break;
1462                                 }
1463 check_continued:
1464                                 /*
1465                                  * It's running now, so it might later
1466                                  * exit, stop, or stop and then continue.
1467                                  */
1468                                 flag = 1;
1469                                 if (!unlikely(options & WCONTINUED))
1470                                         continue;
1471                                 retval = wait_task_continued(
1472                                         p, (options & WNOWAIT),
1473                                         infop, stat_addr, ru);
1474                                 if (retval != 0) /* He released the lock.  */
1475                                         goto end;
1476                                 break;
1477                         }
1478                 }
1479                 if (!flag) {
1480                         list_for_each(_p, &tsk->ptrace_children) {
1481                                 p = list_entry(_p, struct task_struct,
1482                                                 ptrace_list);
1483                                 if (!eligible_child(pid, options, p))
1484                                         continue;
1485                                 flag = 1;
1486                                 break;
1487                         }
1488                 }
1489                 if (options & __WNOTHREAD)
1490                         break;
1491                 tsk = next_thread(tsk);
1492                 if (tsk->signal != current->signal)
1493                         BUG();
1494         } while (tsk != current);
1495
1496         read_unlock(&tasklist_lock);
1497         if (flag) {
1498                 retval = 0;
1499                 if (options & WNOHANG)
1500                         goto end;
1501                 retval = -ERESTARTSYS;
1502                 if (signal_pending(current))
1503                         goto end;
1504                 schedule();
1505                 goto repeat;
1506         }
1507         retval = -ECHILD;
1508 end:
1509         current->state = TASK_RUNNING;
1510         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1511         if (infop) {
1512                 if (retval > 0)
1513                 retval = 0;
1514                 else {
1515                         /*
1516                          * For a WNOHANG return, clear out all the fields
1517                          * we would set so the user can easily tell the
1518                          * difference.
1519                          */
1520                         if (!retval)
1521                                 retval = put_user(0, &infop->si_signo);
1522                         if (!retval)
1523                                 retval = put_user(0, &infop->si_errno);
1524                         if (!retval)
1525                                 retval = put_user(0, &infop->si_code);
1526                         if (!retval)
1527                                 retval = put_user(0, &infop->si_pid);
1528                         if (!retval)
1529                                 retval = put_user(0, &infop->si_uid);
1530                         if (!retval)
1531                                 retval = put_user(0, &infop->si_status);
1532                 }
1533         }
1534         return retval;
1535 }
1536
1537 asmlinkage long sys_waitid(int which, pid_t pid,
1538                            struct siginfo __user *infop, int options,
1539                            struct rusage __user *ru)
1540 {
1541         long ret;
1542
1543         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1544                 return -EINVAL;
1545         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1546                 return -EINVAL;
1547
1548         switch (which) {
1549         case P_ALL:
1550                 pid = -1;
1551                 break;
1552         case P_PID:
1553                 if (pid <= 0)
1554                         return -EINVAL;
1555                 break;
1556         case P_PGID:
1557                 if (pid <= 0)
1558                         return -EINVAL;
1559                 pid = -pid;
1560                 break;
1561         default:
1562                 return -EINVAL;
1563         }
1564
1565         ret = do_wait(pid, options, infop, NULL, ru);
1566
1567         /* avoid REGPARM breakage on x86: */
1568         prevent_tail_call(ret);
1569         return ret;
1570 }
1571
1572 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1573                           int options, struct rusage __user *ru)
1574 {
1575         long ret;
1576
1577         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1578                         __WNOTHREAD|__WCLONE|__WALL))
1579                 return -EINVAL;
1580         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1581
1582         /* avoid REGPARM breakage on x86: */
1583         prevent_tail_call(ret);
1584         return ret;
1585 }
1586
1587 #ifdef __ARCH_WANT_SYS_WAITPID
1588
1589 /*
1590  * sys_waitpid() remains for compatibility. waitpid() should be
1591  * implemented by calling sys_wait4() from libc.a.
1592  */
1593 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1594 {
1595         return sys_wait4(pid, stat_addr, options, NULL);
1596 }
1597
1598 #endif