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/futex.h>
  35 #include <linux/compat.h>
  36
  37 #include <asm/uaccess.h>
  38 #include <asm/unistd.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/mmu_context.h>
  41
  42 extern void sem_exit (void);
  43 extern struct task_struct *child_reaper;
  44
  45 int getrusage(struct task_struct *, int, struct rusage __user *);
  46
  47 static void exit_mm(struct task_struct * tsk);
  48
  49 static void __unhash_process(struct task_struct *p)
  50 {
  51         nr_threads--;
  52         detach_pid(p, PIDTYPE_PID);
  53         detach_pid(p, PIDTYPE_TGID);
  54         if (thread_group_leader(p)) {
  55                 detach_pid(p, PIDTYPE_PGID);
  56                 detach_pid(p, PIDTYPE_SID);
  57
  58                 list_del_init(&p->tasks);
  59                 if (p->pid)
  60                         __get_cpu_var(process_counts)--;
  61         }
  62
  63         remove_parent(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_parent(current);
 244         current->parent = child_reaper;
 245         current->real_parent = child_reaper;
 246         add_parent(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         mutex_lock(&tty_mutex);
 353         current->signal->tty = NULL;
 354         mutex_unlock(&tty_mutex);
 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                         }
 404                         i++;
 405                         set >>= 1;
 406                 }
 407         }
 408 }
 409
 410 struct files_struct *get_files_struct(struct task_struct *task)
 411 {
 412         struct files_struct *files;
 413
 414         task_lock(task);
 415         files = task->files;
 416         if (files)
 417                 atomic_inc(&files->count);
 418         task_unlock(task);
 419
 420         return files;
 421 }
 422
 423 void fastcall put_files_struct(struct files_struct *files)
 424 {
 425         struct fdtable *fdt;
 426
 427         if (atomic_dec_and_test(&files->count)) {
 428                 close_files(files);
 429                 /*
 430                  * Free the fd and fdset arrays if we expanded them.
 431                  * If the fdtable was embedded, pass files for freeing
 432                  * at the end of the RCU grace period. Otherwise,
 433                  * you can free files immediately.
 434                  */
 435                 fdt = files_fdtable(files);
 436                 if (fdt == &files->fdtab)
 437                         fdt->free_files = files;
 438                 else
 439                         kmem_cache_free(files_cachep, files);
 440                 free_fdtable(fdt);
 441         }
 442 }
 443
 444 EXPORT_SYMBOL(put_files_struct);
 445
 446 static inline void __exit_files(struct task_struct *tsk)
 447 {
 448         struct files_struct * files = tsk->files;
 449
 450         if (files) {
 451                 task_lock(tsk);
 452                 tsk->files = NULL;
 453                 task_unlock(tsk);
 454                 put_files_struct(files);
 455         }
 456 }
 457
 458 void exit_files(struct task_struct *tsk)
 459 {
 460         __exit_files(tsk);
 461 }
 462
 463 static inline void __put_fs_struct(struct fs_struct *fs)
 464 {
 465         /* No need to hold fs->lock if we are killing it */
 466         if (atomic_dec_and_test(&fs->count)) {
 467                 dput(fs->root);
 468                 mntput(fs->rootmnt);
 469                 dput(fs->pwd);
 470                 mntput(fs->pwdmnt);
 471                 if (fs->altroot) {
 472                         dput(fs->altroot);
 473                         mntput(fs->altrootmnt);
 474                 }
 475                 kmem_cache_free(fs_cachep, fs);
 476         }
 477 }
 478
 479 void put_fs_struct(struct fs_struct *fs)
 480 {
 481         __put_fs_struct(fs);
 482 }
 483
 484 static inline void __exit_fs(struct task_struct *tsk)
 485 {
 486         struct fs_struct * fs = tsk->fs;
 487
 488         if (fs) {
 489                 task_lock(tsk);
 490                 tsk->fs = NULL;
 491                 task_unlock(tsk);
 492                 __put_fs_struct(fs);
 493         }
 494 }
 495
 496 void exit_fs(struct task_struct *tsk)
 497 {
 498         __exit_fs(tsk);
 499 }
 500
 501 EXPORT_SYMBOL_GPL(exit_fs);
 502
 503 /*
 504  * Turn us into a lazy TLB process if we
 505  * aren't already..
 506  */
 507 static void exit_mm(struct task_struct * tsk)
 508 {
 509         struct mm_struct *mm = tsk->mm;
 510
 511         mm_release(tsk, mm);
 512         if (!mm)
 513                 return;
 514         /*
 515          * Serialize with any possible pending coredump.
 516          * We must hold mmap_sem around checking core_waiters
 517          * and clearing tsk->mm.  The core-inducing thread
 518          * will increment core_waiters for each thread in the
 519          * group with ->mm != NULL.
 520          */
 521         down_read(&mm->mmap_sem);
 522         if (mm->core_waiters) {
 523                 up_read(&mm->mmap_sem);
 524                 down_write(&mm->mmap_sem);
 525                 if (!--mm->core_waiters)
 526                         complete(mm->core_startup_done);
 527                 up_write(&mm->mmap_sem);
 528
 529                 wait_for_completion(&mm->core_done);
 530                 down_read(&mm->mmap_sem);
 531         }
 532         atomic_inc(&mm->mm_count);
 533         if (mm != tsk->active_mm) BUG();
 534         /* more a memory barrier than a real lock */
 535         task_lock(tsk);
 536         tsk->mm = NULL;
 537         up_read(&mm->mmap_sem);
 538         enter_lazy_tlb(mm, current);
 539         task_unlock(tsk);
 540         mmput(mm);
 541 }
 542
 543 static inline void choose_new_parent(task_t *p, task_t *reaper)
 544 {
 545         /*
 546          * Make sure we're not reparenting to ourselves and that
 547          * the parent is not a zombie.
 548          */
 549         BUG_ON(p == reaper || reaper->exit_state);
 550         p->real_parent = reaper;
 551 }
 552
 553 static void reparent_thread(task_t *p, task_t *father, int traced)
 554 {
 555         /* We don't want people slaying init.  */
 556         if (p->exit_signal != -1)
 557                 p->exit_signal = SIGCHLD;
 558
 559         if (p->pdeath_signal)
 560                 /* We already hold the tasklist_lock here.  */
 561                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 562
 563         /* Move the child from its dying parent to the new one.  */
 564         if (unlikely(traced)) {
 565                 /* Preserve ptrace links if someone else is tracing this child.  */
 566                 list_del_init(&p->ptrace_list);
 567                 if (p->parent != p->real_parent)
 568                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 569         } else {
 570                 /* If this child is being traced, then we're the one tracing it
 571                  * anyway, so let go of it.
 572                  */
 573                 p->ptrace = 0;
 574                 list_del_init(&p->sibling);
 575                 p->parent = p->real_parent;
 576                 list_add_tail(&p->sibling, &p->parent->children);
 577
 578                 /* If we'd notified the old parent about this child's death,
 579                  * also notify the new parent.
 580                  */
 581                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 582                     thread_group_empty(p))
 583                         do_notify_parent(p, p->exit_signal);
 584                 else if (p->state == TASK_TRACED) {
 585                         /*
 586                          * If it was at a trace stop, turn it into
 587                          * a normal stop since it's no longer being
 588                          * traced.
 589                          */
 590                         ptrace_untrace(p);
 591                 }
 592         }
 593
 594         /*
 595          * process group orphan check
 596          * Case ii: Our child is in a different pgrp
 597          * than we are, and it was the only connection
 598          * outside, so the child pgrp is now orphaned.
 599          */
 600         if ((process_group(p) != process_group(father)) &&
 601             (p->signal->session == father->signal->session)) {
 602                 int pgrp = process_group(p);
 603
 604                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 605                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 606                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 607                 }
 608         }
 609 }
 610
 611 /*
 612  * When we die, we re-parent all our children.
 613  * Try to give them to another thread in our thread
 614  * group, and if no such member exists, give it to
 615  * the global child reaper process (ie "init")
 616  */
 617 static void forget_original_parent(struct task_struct * father,
 618                                           struct list_head *to_release)
 619 {
 620         struct task_struct *p, *reaper = father;
 621         struct list_head *_p, *_n;
 622
 623         do {
 624                 reaper = next_thread(reaper);
 625                 if (reaper == father) {
 626                         reaper = child_reaper;
 627                         break;
 628                 }
 629         } while (reaper->exit_state);
 630
 631         /*
 632          * There are only two places where our children can be:
 633          *
 634          * - in our child list
 635          * - in our ptraced child list
 636          *
 637          * Search them and reparent children.
 638          */
 639         list_for_each_safe(_p, _n, &father->children) {
 640                 int ptrace;
 641                 p = list_entry(_p,struct task_struct,sibling);
 642
 643                 ptrace = p->ptrace;
 644
 645                 /* if father isn't the real parent, then ptrace must be enabled */
 646                 BUG_ON(father != p->real_parent && !ptrace);
 647
 648                 if (father == p->real_parent) {
 649                         /* reparent with a reaper, real father it's us */
 650                         choose_new_parent(p, reaper);
 651                         reparent_thread(p, father, 0);
 652                 } else {
 653                         /* reparent ptraced task to its real parent */
 654                         __ptrace_unlink (p);
 655                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 656                             thread_group_empty(p))
 657                                 do_notify_parent(p, p->exit_signal);
 658                 }
 659
 660                 /*
 661                  * if the ptraced child is a zombie with exit_signal == -1
 662                  * we must collect it before we exit, or it will remain
 663                  * zombie forever since we prevented it from self-reap itself
 664                  * while it was being traced by us, to be able to see it in wait4.
 665                  */
 666                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 667                         list_add(&p->ptrace_list, to_release);
 668         }
 669         list_for_each_safe(_p, _n, &father->ptrace_children) {
 670                 p = list_entry(_p,struct task_struct,ptrace_list);
 671                 choose_new_parent(p, reaper);
 672                 reparent_thread(p, father, 1);
 673         }
 674 }
 675
 676 /*
 677  * Send signals to all our closest relatives so that they know
 678  * to properly mourn us..
 679  */
 680 static void exit_notify(struct task_struct *tsk)
 681 {
 682         int state;
 683         struct task_struct *t;
 684         struct list_head ptrace_dead, *_p, *_n;
 685
 686         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 687             && !thread_group_empty(tsk)) {
 688                 /*
 689                  * This occurs when there was a race between our exit
 690                  * syscall and a group signal choosing us as the one to
 691                  * wake up.  It could be that we are the only thread
 692                  * alerted to check for pending signals, but another thread
 693                  * should be woken now to take the signal since we will not.
 694                  * Now we'll wake all the threads in the group just to make
 695                  * sure someone gets all the pending signals.
 696                  */
 697                 read_lock(&tasklist_lock);
 698                 spin_lock_irq(&tsk->sighand->siglock);
 699                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 700                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 701                                 recalc_sigpending_tsk(t);
 702                                 if (signal_pending(t))
 703                                         signal_wake_up(t, 0);
 704                         }
 705                 spin_unlock_irq(&tsk->sighand->siglock);
 706                 read_unlock(&tasklist_lock);
 707         }
 708
 709         write_lock_irq(&tasklist_lock);
 710
 711         /*
 712          * This does two things:
 713          *
 714          * A.  Make init inherit all the child processes
 715          * B.  Check to see if any process groups have become orphaned
 716          *      as a result of our exiting, and if they have any stopped
 717          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 718          */
 719
 720         INIT_LIST_HEAD(&ptrace_dead);
 721         forget_original_parent(tsk, &ptrace_dead);
 722         BUG_ON(!list_empty(&tsk->children));
 723         BUG_ON(!list_empty(&tsk->ptrace_children));
 724
 725         /*
 726          * Check to see if any process groups have become orphaned
 727          * as a result of our exiting, and if they have any stopped
 728          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 729          *
 730          * Case i: Our father is in a different pgrp than we are
 731          * and we were the only connection outside, so our pgrp
 732          * is about to become orphaned.
 733          */
 734
 735         t = tsk->real_parent;
 736
 737         if ((process_group(t) != process_group(tsk)) &&
 738             (t->signal->session == tsk->signal->session) &&
 739             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 740             has_stopped_jobs(process_group(tsk))) {
 741                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 742                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 743         }
 744
 745         /* Let father know we died
 746          *
 747          * Thread signals are configurable, but you aren't going to use
 748          * that to send signals to arbitary processes.
 749          * That stops right now.
 750          *
 751          * If the parent exec id doesn't match the exec id we saved
 752          * when we started then we know the parent has changed security
 753          * domain.
 754          *
 755          * If our self_exec id doesn't match our parent_exec_id then
 756          * we have changed execution domain as these two values started
 757          * the same after a fork.
 758          *
 759          */
 760
 761         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 762             ( tsk->parent_exec_id != t->self_exec_id  ||
 763               tsk->self_exec_id != tsk->parent_exec_id)
 764             && !capable(CAP_KILL))
 765                 tsk->exit_signal = SIGCHLD;
 766
 767
 768         /* If something other than our normal parent is ptracing us, then
 769          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 770          * only has special meaning to our real parent.
 771          */
 772         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 773                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 774                 do_notify_parent(tsk, signal);
 775         } else if (tsk->ptrace) {
 776                 do_notify_parent(tsk, SIGCHLD);
 777         }
 778
 779         state = EXIT_ZOMBIE;
 780         if (tsk->exit_signal == -1 &&
 781             (likely(tsk->ptrace == 0) ||
 782              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 783                 state = EXIT_DEAD;
 784         tsk->exit_state = state;
 785
 786         write_unlock_irq(&tasklist_lock);
 787
 788         list_for_each_safe(_p, _n, &ptrace_dead) {
 789                 list_del_init(_p);
 790                 t = list_entry(_p,struct task_struct,ptrace_list);
 791                 release_task(t);
 792         }
 793
 794         /* If the process is dead, release it - nobody will wait for it */
 795         if (state == EXIT_DEAD)
 796                 release_task(tsk);
 797 }
 798
 799 fastcall NORET_TYPE void do_exit(long code)
 800 {
 801         struct task_struct *tsk = current;
 802         int group_dead;
 803
 804         profile_task_exit(tsk);
 805
 806         WARN_ON(atomic_read(&tsk->fs_excl));
 807
 808         if (unlikely(in_interrupt()))
 809                 panic("Aiee, killing interrupt handler!");
 810         if (unlikely(!tsk->pid))
 811                 panic("Attempted to kill the idle task!");
 812         if (unlikely(tsk == child_reaper))
 813                 panic("Attempted to kill init!");
 814
 815         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 816                 current->ptrace_message = code;
 817                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 818         }
 819
 820         /*
 821          * We're taking recursive faults here in do_exit. Safest is to just
 822          * leave this task alone and wait for reboot.
 823          */
 824         if (unlikely(tsk->flags & PF_EXITING)) {
 825                 printk(KERN_ALERT
 826                         "Fixing recursive fault but reboot is needed!\n");
 827                 if (tsk->io_context)
 828                         exit_io_context();
 829                 set_current_state(TASK_UNINTERRUPTIBLE);
 830                 schedule();
 831         }
 832
 833         tsk->flags |= PF_EXITING;
 834
 835         /*
 836          * Make sure we don't try to process any timer firings
 837          * while we are already exiting.
 838          */
 839         tsk->it_virt_expires = cputime_zero;
 840         tsk->it_prof_expires = cputime_zero;
 841         tsk->it_sched_expires = 0;
 842
 843         if (unlikely(in_atomic()))
 844                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 845                                 current->comm, current->pid,
 846                                 preempt_count());
 847
 848         acct_update_integrals(tsk);
 849         if (tsk->mm) {
 850                 update_hiwater_rss(tsk->mm);
 851                 update_hiwater_vm(tsk->mm);
 852         }
 853         group_dead = atomic_dec_and_test(&tsk->signal->live);
 854         if (group_dead) {
 855                 hrtimer_cancel(&tsk->signal->real_timer);
 856                 exit_itimers(tsk->signal);
 857                 acct_process(code);
 858         }
 859         if (unlikely(tsk->robust_list))
 860                 exit_robust_list(tsk);
 861 #ifdef CONFIG_COMPAT
 862         if (unlikely(tsk->compat_robust_list))
 863                 compat_exit_robust_list(tsk);
 864 #endif
 865         exit_mm(tsk);
 866
 867         exit_sem(tsk);
 868         __exit_files(tsk);
 869         __exit_fs(tsk);
 870         exit_namespace(tsk);
 871         exit_thread();
 872         cpuset_exit(tsk);
 873         exit_keys(tsk);
 874
 875         if (group_dead && tsk->signal->leader)
 876                 disassociate_ctty(1);
 877
 878         module_put(task_thread_info(tsk)->exec_domain->module);
 879         if (tsk->binfmt)
 880                 module_put(tsk->binfmt->module);
 881
 882         tsk->exit_code = code;
 883         proc_exit_connector(tsk);
 884         exit_notify(tsk);
 885 #ifdef CONFIG_NUMA
 886         mpol_free(tsk->mempolicy);
 887         tsk->mempolicy = NULL;
 888 #endif
 889         /*
 890          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
 891          */
 892         mutex_debug_check_no_locks_held(tsk);
 893
 894         if (tsk->io_context)
 895                 exit_io_context();
 896
 897         /* PF_DEAD causes final put_task_struct after we schedule. */
 898         preempt_disable();
 899         BUG_ON(tsk->flags & PF_DEAD);
 900         tsk->flags |= PF_DEAD;
 901
 902         schedule();
 903         BUG();
 904         /* Avoid "noreturn function does return".  */
 905         for (;;) ;
 906 }
 907
 908 EXPORT_SYMBOL_GPL(do_exit);
 909
 910 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 911 {
 912         if (comp)
 913                 complete(comp);
 914
 915         do_exit(code);
 916 }
 917
 918 EXPORT_SYMBOL(complete_and_exit);
 919
 920 asmlinkage long sys_exit(int error_code)
 921 {
 922         do_exit((error_code&0xff)<<8);
 923 }
 924
 925 task_t fastcall *next_thread(const task_t *p)
 926 {
 927         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 928 }
 929
 930 EXPORT_SYMBOL(next_thread);
 931
 932 /*
 933  * Take down every thread in the group.  This is called by fatal signals
 934  * as well as by sys_exit_group (below).
 935  */
 936 NORET_TYPE void
 937 do_group_exit(int exit_code)
 938 {
 939         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 940
 941         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 942                 exit_code = current->signal->group_exit_code;
 943         else if (!thread_group_empty(current)) {
 944                 struct signal_struct *const sig = current->signal;
 945                 struct sighand_struct *const sighand = current->sighand;
 946                 read_lock(&tasklist_lock);
 947                 spin_lock_irq(&sighand->siglock);
 948                 if (sig->flags & SIGNAL_GROUP_EXIT)
 949                         /* Another thread got here before we took the lock.  */
 950                         exit_code = sig->group_exit_code;
 951                 else {
 952                         sig->group_exit_code = exit_code;
 953                         zap_other_threads(current);
 954                 }
 955                 spin_unlock_irq(&sighand->siglock);
 956                 read_unlock(&tasklist_lock);
 957         }
 958
 959         do_exit(exit_code);
 960         /* NOTREACHED */
 961 }
 962
 963 /*
 964  * this kills every thread in the thread group. Note that any externally
 965  * wait4()-ing process will get the correct exit code - even if this
 966  * thread is not the thread group leader.
 967  */
 968 asmlinkage void sys_exit_group(int error_code)
 969 {
 970         do_group_exit((error_code & 0xff) << 8);
 971 }
 972
 973 static int eligible_child(pid_t pid, int options, task_t *p)
 974 {
 975         if (pid > 0) {
 976                 if (p->pid != pid)
 977                         return 0;
 978         } else if (!pid) {
 979                 if (process_group(p) != process_group(current))
 980                         return 0;
 981         } else if (pid != -1) {
 982                 if (process_group(p) != -pid)
 983                         return 0;
 984         }
 985
 986         /*
 987          * Do not consider detached threads that are
 988          * not ptraced:
 989          */
 990         if (p->exit_signal == -1 && !p->ptrace)
 991                 return 0;
 992
 993         /* Wait for all children (clone and not) if __WALL is set;
 994          * otherwise, wait for clone children *only* if __WCLONE is
 995          * set; otherwise, wait for non-clone children *only*.  (Note:
 996          * A "clone" child here is one that reports to its parent
 997          * using a signal other than SIGCHLD.) */
 998         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 999             && !(options & __WALL))
1000                 return 0;
1001         /*
1002          * Do not consider thread group leaders that are
1003          * in a non-empty thread group:
1004          */
1005         if (current->tgid != p->tgid && delay_group_leader(p))
1006                 return 2;
1007
1008         if (security_task_wait(p))
1009                 return 0;
1010
1011         return 1;
1012 }
1013
1014 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
1015                                int why, int status,
1016                                struct siginfo __user *infop,
1017                                struct rusage __user *rusagep)
1018 {
1019         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1020         put_task_struct(p);
1021         if (!retval)
1022                 retval = put_user(SIGCHLD, &infop->si_signo);
1023         if (!retval)
1024                 retval = put_user(0, &infop->si_errno);
1025         if (!retval)
1026                 retval = put_user((short)why, &infop->si_code);
1027         if (!retval)
1028                 retval = put_user(pid, &infop->si_pid);
1029         if (!retval)
1030                 retval = put_user(uid, &infop->si_uid);
1031         if (!retval)
1032                 retval = put_user(status, &infop->si_status);
1033         if (!retval)
1034                 retval = pid;
1035         return retval;
1036 }
1037
1038 /*
1039  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1040  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1041  * the lock and this task is uninteresting.  If we return nonzero, we have
1042  * released the lock and the system call should return.
1043  */
1044 static int wait_task_zombie(task_t *p, int noreap,
1045                             struct siginfo __user *infop,
1046                             int __user *stat_addr, struct rusage __user *ru)
1047 {
1048         unsigned long state;
1049         int retval;
1050         int status;
1051
1052         if (unlikely(noreap)) {
1053                 pid_t pid = p->pid;
1054                 uid_t uid = p->uid;
1055                 int exit_code = p->exit_code;
1056                 int why, status;
1057
1058                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1059                         return 0;
1060                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1061                         return 0;
1062                 get_task_struct(p);
1063                 read_unlock(&tasklist_lock);
1064                 if ((exit_code & 0x7f) == 0) {
1065                         why = CLD_EXITED;
1066                         status = exit_code >> 8;
1067                 } else {
1068                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1069                         status = exit_code & 0x7f;
1070                 }
1071                 return wait_noreap_copyout(p, pid, uid, why,
1072                                            status, infop, ru);
1073         }
1074
1075         /*
1076          * Try to move the task's state to DEAD
1077          * only one thread is allowed to do this:
1078          */
1079         state = xchg(&p->exit_state, EXIT_DEAD);
1080         if (state != EXIT_ZOMBIE) {
1081                 BUG_ON(state != EXIT_DEAD);
1082                 return 0;
1083         }
1084         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1085                 /*
1086                  * This can only happen in a race with a ptraced thread
1087                  * dying on another processor.
1088                  */
1089                 return 0;
1090         }
1091
1092         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1093                 struct signal_struct *psig;
1094                 struct signal_struct *sig;
1095
1096                 /*
1097                  * The resource counters for the group leader are in its
1098                  * own task_struct.  Those for dead threads in the group
1099                  * are in its signal_struct, as are those for the child
1100                  * processes it has previously reaped.  All these
1101                  * accumulate in the parent's signal_struct c* fields.
1102                  *
1103                  * We don't bother to take a lock here to protect these
1104                  * p->signal fields, because they are only touched by
1105                  * __exit_signal, which runs with tasklist_lock
1106                  * write-locked anyway, and so is excluded here.  We do
1107                  * need to protect the access to p->parent->signal fields,
1108                  * as other threads in the parent group can be right
1109                  * here reaping other children at the same time.
1110                  */
1111                 spin_lock_irq(&p->parent->sighand->siglock);
1112                 psig = p->parent->signal;
1113                 sig = p->signal;
1114                 psig->cutime =
1115                         cputime_add(psig->cutime,
1116                         cputime_add(p->utime,
1117                         cputime_add(sig->utime,
1118                                     sig->cutime)));
1119                 psig->cstime =
1120                         cputime_add(psig->cstime,
1121                         cputime_add(p->stime,
1122                         cputime_add(sig->stime,
1123                                     sig->cstime)));
1124                 psig->cmin_flt +=
1125                         p->min_flt + sig->min_flt + sig->cmin_flt;
1126                 psig->cmaj_flt +=
1127                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1128                 psig->cnvcsw +=
1129                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1130                 psig->cnivcsw +=
1131                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1132                 spin_unlock_irq(&p->parent->sighand->siglock);
1133         }
1134
1135         /*
1136          * Now we are sure this task is interesting, and no other
1137          * thread can reap it because we set its state to EXIT_DEAD.
1138          */
1139         read_unlock(&tasklist_lock);
1140
1141         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1142         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1143                 ? p->signal->group_exit_code : p->exit_code;
1144         if (!retval && stat_addr)
1145                 retval = put_user(status, stat_addr);
1146         if (!retval && infop)
1147                 retval = put_user(SIGCHLD, &infop->si_signo);
1148         if (!retval && infop)
1149                 retval = put_user(0, &infop->si_errno);
1150         if (!retval && infop) {
1151                 int why;
1152
1153                 if ((status & 0x7f) == 0) {
1154                         why = CLD_EXITED;
1155                         status >>= 8;
1156                 } else {
1157                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1158                         status &= 0x7f;
1159                 }
1160                 retval = put_user((short)why, &infop->si_code);
1161                 if (!retval)
1162                         retval = put_user(status, &infop->si_status);
1163         }
1164         if (!retval && infop)
1165                 retval = put_user(p->pid, &infop->si_pid);
1166         if (!retval && infop)
1167                 retval = put_user(p->uid, &infop->si_uid);
1168         if (retval) {
1169                 // TODO: is this safe?
1170                 p->exit_state = EXIT_ZOMBIE;
1171                 return retval;
1172         }
1173         retval = p->pid;
1174         if (p->real_parent != p->parent) {
1175                 write_lock_irq(&tasklist_lock);
1176                 /* Double-check with lock held.  */
1177                 if (p->real_parent != p->parent) {
1178                         __ptrace_unlink(p);
1179                         // TODO: is this safe?
1180                         p->exit_state = EXIT_ZOMBIE;
1181                         /*
1182                          * If this is not a detached task, notify the parent.
1183                          * If it's still not detached after that, don't release
1184                          * it now.
1185                          */
1186                         if (p->exit_signal != -1) {
1187                                 do_notify_parent(p, p->exit_signal);
1188                                 if (p->exit_signal != -1)
1189                                         p = NULL;
1190                         }
1191                 }
1192                 write_unlock_irq(&tasklist_lock);
1193         }
1194         if (p != NULL)
1195                 release_task(p);
1196         BUG_ON(!retval);
1197         return retval;
1198 }
1199
1200 /*
1201  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1202  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1203  * the lock and this task is uninteresting.  If we return nonzero, we have
1204  * released the lock and the system call should return.
1205  */
1206 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1207                              struct siginfo __user *infop,
1208                              int __user *stat_addr, struct rusage __user *ru)
1209 {
1210         int retval, exit_code;
1211
1212         if (!p->exit_code)
1213                 return 0;
1214         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1215             p->signal && p->signal->group_stop_count > 0)
1216                 /*
1217                  * A group stop is in progress and this is the group leader.
1218                  * We won't report until all threads have stopped.
1219                  */
1220                 return 0;
1221
1222         /*
1223          * Now we are pretty sure this task is interesting.
1224          * Make sure it doesn't get reaped out from under us while we
1225          * give up the lock and then examine it below.  We don't want to
1226          * keep holding onto the tasklist_lock while we call getrusage and
1227          * possibly take page faults for user memory.
1228          */
1229         get_task_struct(p);
1230         read_unlock(&tasklist_lock);
1231
1232         if (unlikely(noreap)) {
1233                 pid_t pid = p->pid;
1234                 uid_t uid = p->uid;
1235                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1236
1237                 exit_code = p->exit_code;
1238                 if (unlikely(!exit_code) ||
1239                     unlikely(p->state & TASK_TRACED))
1240                         goto bail_ref;
1241                 return wait_noreap_copyout(p, pid, uid,
1242                                            why, (exit_code << 8) | 0x7f,
1243                                            infop, ru);
1244         }
1245
1246         write_lock_irq(&tasklist_lock);
1247
1248         /*
1249          * This uses xchg to be atomic with the thread resuming and setting
1250          * it.  It must also be done with the write lock held to prevent a
1251          * race with the EXIT_ZOMBIE case.
1252          */
1253         exit_code = xchg(&p->exit_code, 0);
1254         if (unlikely(p->exit_state)) {
1255                 /*
1256                  * The task resumed and then died.  Let the next iteration
1257                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1258                  * already be zero here if it resumed and did _exit(0).
1259                  * The task itself is dead and won't touch exit_code again;
1260                  * other processors in this function are locked out.
1261                  */
1262                 p->exit_code = exit_code;
1263                 exit_code = 0;
1264         }
1265         if (unlikely(exit_code == 0)) {
1266                 /*
1267                  * Another thread in this function got to it first, or it
1268                  * resumed, or it resumed and then died.
1269                  */
1270                 write_unlock_irq(&tasklist_lock);
1271 bail_ref:
1272                 put_task_struct(p);
1273                 /*
1274                  * We are returning to the wait loop without having successfully
1275                  * removed the process and having released the lock. We cannot
1276                  * continue, since the "p" task pointer is potentially stale.
1277                  *
1278                  * Return -EAGAIN, and do_wait() will restart the loop from the
1279                  * beginning. Do _not_ re-acquire the lock.
1280                  */
1281                 return -EAGAIN;
1282         }
1283
1284         /* move to end of parent's list to avoid starvation */
1285         remove_parent(p);
1286         add_parent(p);
1287
1288         write_unlock_irq(&tasklist_lock);
1289
1290         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1291         if (!retval && stat_addr)
1292                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1293         if (!retval && infop)
1294                 retval = put_user(SIGCHLD, &infop->si_signo);
1295         if (!retval && infop)
1296                 retval = put_user(0, &infop->si_errno);
1297         if (!retval && infop)
1298                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1299                                           ? CLD_TRAPPED : CLD_STOPPED),
1300                                   &infop->si_code);
1301         if (!retval && infop)
1302                 retval = put_user(exit_code, &infop->si_status);
1303         if (!retval && infop)
1304                 retval = put_user(p->pid, &infop->si_pid);
1305         if (!retval && infop)
1306                 retval = put_user(p->uid, &infop->si_uid);
1307         if (!retval)
1308                 retval = p->pid;
1309         put_task_struct(p);
1310
1311         BUG_ON(!retval);
1312         return retval;
1313 }
1314
1315 /*
1316  * Handle do_wait work for one task in a live, non-stopped state.
1317  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1318  * the lock and this task is uninteresting.  If we return nonzero, we have
1319  * released the lock and the system call should return.
1320  */
1321 static int wait_task_continued(task_t *p, int noreap,
1322                                struct siginfo __user *infop,
1323                                int __user *stat_addr, struct rusage __user *ru)
1324 {
1325         int retval;
1326         pid_t pid;
1327         uid_t uid;
1328
1329         if (unlikely(!p->signal))
1330                 return 0;
1331
1332         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1333                 return 0;
1334
1335         spin_lock_irq(&p->sighand->siglock);
1336         /* Re-check with the lock held.  */
1337         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1338                 spin_unlock_irq(&p->sighand->siglock);
1339                 return 0;
1340         }
1341         if (!noreap)
1342                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1343         spin_unlock_irq(&p->sighand->siglock);
1344
1345         pid = p->pid;
1346         uid = p->uid;
1347         get_task_struct(p);
1348         read_unlock(&tasklist_lock);
1349
1350         if (!infop) {
1351                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1352                 put_task_struct(p);
1353                 if (!retval && stat_addr)
1354                         retval = put_user(0xffff, stat_addr);
1355                 if (!retval)
1356                         retval = p->pid;
1357         } else {
1358                 retval = wait_noreap_copyout(p, pid, uid,
1359                                              CLD_CONTINUED, SIGCONT,
1360                                              infop, ru);
1361                 BUG_ON(retval == 0);
1362         }
1363
1364         return retval;
1365 }
1366
1367
1368 static inline int my_ptrace_child(struct task_struct *p)
1369 {
1370         if (!(p->ptrace & PT_PTRACED))
1371                 return 0;
1372         if (!(p->ptrace & PT_ATTACHED))
1373                 return 1;
1374         /*
1375          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1376          * we are the attacher.  If we are the real parent, this is a race
1377          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1378          * which we have to switch the parent links, but has already set
1379          * the flags in p->ptrace.
1380          */
1381         return (p->parent != p->real_parent);
1382 }
1383
1384 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1385                     int __user *stat_addr, struct rusage __user *ru)
1386 {
1387         DECLARE_WAITQUEUE(wait, current);
1388         struct task_struct *tsk;
1389         int flag, retval;
1390
1391         add_wait_queue(&current->signal->wait_chldexit,&wait);
1392 repeat:
1393         /*
1394          * We will set this flag if we see any child that might later
1395          * match our criteria, even if we are not able to reap it yet.
1396          */
1397         flag = 0;
1398         current->state = TASK_INTERRUPTIBLE;
1399         read_lock(&tasklist_lock);
1400         tsk = current;
1401         do {
1402                 struct task_struct *p;
1403                 struct list_head *_p;
1404                 int ret;
1405
1406                 list_for_each(_p,&tsk->children) {
1407                         p = list_entry(_p,struct task_struct,sibling);
1408
1409                         ret = eligible_child(pid, options, p);
1410                         if (!ret)
1411                                 continue;
1412
1413                         switch (p->state) {
1414                         case TASK_TRACED:
1415                                 /*
1416                                  * When we hit the race with PTRACE_ATTACH,
1417                                  * we will not report this child.  But the
1418                                  * race means it has not yet been moved to
1419                                  * our ptrace_children list, so we need to
1420                                  * set the flag here to avoid a spurious ECHILD
1421                                  * when the race happens with the only child.
1422                                  */
1423                                 flag = 1;
1424                                 if (!my_ptrace_child(p))
1425                                         continue;
1426                                 /*FALLTHROUGH*/
1427                         case TASK_STOPPED:
1428                                 /*
1429                                  * It's stopped now, so it might later
1430                                  * continue, exit, or stop again.
1431                                  */
1432                                 flag = 1;
1433                                 if (!(options & WUNTRACED) &&
1434                                     !my_ptrace_child(p))
1435                                         continue;
1436                                 retval = wait_task_stopped(p, ret == 2,
1437                                                            (options & WNOWAIT),
1438                                                            infop,
1439                                                            stat_addr, ru);
1440                                 if (retval == -EAGAIN)
1441                                         goto repeat;
1442                                 if (retval != 0) /* He released the lock.  */
1443                                         goto end;
1444                                 break;
1445                         default:
1446                         // case EXIT_DEAD:
1447                                 if (p->exit_state == EXIT_DEAD)
1448                                         continue;
1449                         // case EXIT_ZOMBIE:
1450                                 if (p->exit_state == EXIT_ZOMBIE) {
1451                                         /*
1452                                          * Eligible but we cannot release
1453                                          * it yet:
1454                                          */
1455                                         if (ret == 2)
1456                                                 goto check_continued;
1457                                         if (!likely(options & WEXITED))
1458                                                 continue;
1459                                         retval = wait_task_zombie(
1460                                                 p, (options & WNOWAIT),
1461                                                 infop, stat_addr, ru);
1462                                         /* He released the lock.  */
1463                                         if (retval != 0)
1464                                                 goto end;
1465                                         break;
1466                                 }
1467 check_continued:
1468                                 /*
1469                                  * It's running now, so it might later
1470                                  * exit, stop, or stop and then continue.
1471                                  */
1472                                 flag = 1;
1473                                 if (!unlikely(options & WCONTINUED))
1474                                         continue;
1475                                 retval = wait_task_continued(
1476                                         p, (options & WNOWAIT),
1477                                         infop, stat_addr, ru);
1478                                 if (retval != 0) /* He released the lock.  */
1479                                         goto end;
1480                                 break;
1481                         }
1482                 }
1483                 if (!flag) {
1484                         list_for_each(_p, &tsk->ptrace_children) {
1485                                 p = list_entry(_p, struct task_struct,
1486                                                 ptrace_list);
1487                                 if (!eligible_child(pid, options, p))
1488                                         continue;
1489                                 flag = 1;
1490                                 break;
1491                         }
1492                 }
1493                 if (options & __WNOTHREAD)
1494                         break;
1495                 tsk = next_thread(tsk);
1496                 if (tsk->signal != current->signal)
1497                         BUG();
1498         } while (tsk != current);
1499
1500         read_unlock(&tasklist_lock);
1501         if (flag) {
1502                 retval = 0;
1503                 if (options & WNOHANG)
1504                         goto end;
1505                 retval = -ERESTARTSYS;
1506                 if (signal_pending(current))
1507                         goto end;
1508                 schedule();
1509                 goto repeat;
1510         }
1511         retval = -ECHILD;
1512 end:
1513         current->state = TASK_RUNNING;
1514         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1515         if (infop) {
1516                 if (retval > 0)
1517                 retval = 0;
1518                 else {
1519                         /*
1520                          * For a WNOHANG return, clear out all the fields
1521                          * we would set so the user can easily tell the
1522                          * difference.
1523                          */
1524                         if (!retval)
1525                                 retval = put_user(0, &infop->si_signo);
1526                         if (!retval)
1527                                 retval = put_user(0, &infop->si_errno);
1528                         if (!retval)
1529                                 retval = put_user(0, &infop->si_code);
1530                         if (!retval)
1531                                 retval = put_user(0, &infop->si_pid);
1532                         if (!retval)
1533                                 retval = put_user(0, &infop->si_uid);
1534                         if (!retval)
1535                                 retval = put_user(0, &infop->si_status);
1536                 }
1537         }
1538         return retval;
1539 }
1540
1541 asmlinkage long sys_waitid(int which, pid_t pid,
1542                            struct siginfo __user *infop, int options,
1543                            struct rusage __user *ru)
1544 {
1545         long ret;
1546
1547         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1548                 return -EINVAL;
1549         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1550                 return -EINVAL;
1551
1552         switch (which) {
1553         case P_ALL:
1554                 pid = -1;
1555                 break;
1556         case P_PID:
1557                 if (pid <= 0)
1558                         return -EINVAL;
1559                 break;
1560         case P_PGID:
1561                 if (pid <= 0)
1562                         return -EINVAL;
1563                 pid = -pid;
1564                 break;
1565         default:
1566                 return -EINVAL;
1567         }
1568
1569         ret = do_wait(pid, options, infop, NULL, ru);
1570
1571         /* avoid REGPARM breakage on x86: */
1572         prevent_tail_call(ret);
1573         return ret;
1574 }
1575
1576 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1577                           int options, struct rusage __user *ru)
1578 {
1579         long ret;
1580
1581         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1582                         __WNOTHREAD|__WCLONE|__WALL))
1583                 return -EINVAL;
1584         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1585
1586         /* avoid REGPARM breakage on x86: */
1587         prevent_tail_call(ret);
1588         return ret;
1589 }
1590
1591 #ifdef __ARCH_WANT_SYS_WAITPID
1592
1593 /*
1594  * sys_waitpid() remains for compatibility. waitpid() should be
1595  * implemented by calling sys_wait4() from libc.a.
1596  */
1597 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1598 {
1599         return sys_wait4(pid, stat_addr, options, NULL);
1600 }
1601
1602 #endif