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                 if (p->pid)
  58                         __get_cpu_var(process_counts)--;
  59         }
  60
  61         REMOVE_LINKS(p);
  62 }
  63
  64 void release_task(struct task_struct * p)
  65 {
  66         int zap_leader;
  67         task_t *leader;
  68         struct dentry *proc_dentry;
  69
  70 repeat:
  71         atomic_dec(&p->user->processes);
  72         spin_lock(&p->proc_lock);
  73         proc_dentry = proc_pid_unhash(p);
  74         write_lock_irq(&tasklist_lock);
  75         if (unlikely(p->ptrace))
  76                 __ptrace_unlink(p);
  77         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  78         __exit_signal(p);
  79         /*
  80          * Note that the fastpath in sys_times depends on __exit_signal having
  81          * updated the counters before a task is removed from the tasklist of
  82          * the process by __unhash_process.
  83          */
  84         __unhash_process(p);
  85
  86         /*
  87          * If we are the last non-leader member of the thread
  88          * group, and the leader is zombie, then notify the
  89          * group leader's parent process. (if it wants notification.)
  90          */
  91         zap_leader = 0;
  92         leader = p->group_leader;
  93         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
  94                 BUG_ON(leader->exit_signal == -1);
  95                 do_notify_parent(leader, leader->exit_signal);
  96                 /*
  97                  * If we were the last child thread and the leader has
  98                  * exited already, and the leader's parent ignores SIGCHLD,
  99                  * then we are the one who should release the leader.
 100                  *
 101                  * do_notify_parent() will have marked it self-reaping in
 102                  * that case.
 103                  */
 104                 zap_leader = (leader->exit_signal == -1);
 105         }
 106
 107         sched_exit(p);
 108         write_unlock_irq(&tasklist_lock);
 109         spin_unlock(&p->proc_lock);
 110         proc_pid_flush(proc_dentry);
 111         release_thread(p);
 112         put_task_struct(p);
 113
 114         p = leader;
 115         if (unlikely(zap_leader))
 116                 goto repeat;
 117 }
 118
 119 /* we are using it only for SMP init */
 120
 121 void unhash_process(struct task_struct *p)
 122 {
 123         struct dentry *proc_dentry;
 124
 125         spin_lock(&p->proc_lock);
 126         proc_dentry = proc_pid_unhash(p);
 127         write_lock_irq(&tasklist_lock);
 128         __unhash_process(p);
 129         write_unlock_irq(&tasklist_lock);
 130         spin_unlock(&p->proc_lock);
 131         proc_pid_flush(proc_dentry);
 132 }
 133
 134 /*
 135  * This checks not only the pgrp, but falls back on the pid if no
 136  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 137  * without this...
 138  */
 139 int session_of_pgrp(int pgrp)
 140 {
 141         struct task_struct *p;
 142         int sid = -1;
 143
 144         read_lock(&tasklist_lock);
 145         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 146                 if (p->signal->session > 0) {
 147                         sid = p->signal->session;
 148                         goto out;
 149                 }
 150         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 151         p = find_task_by_pid(pgrp);
 152         if (p)
 153                 sid = p->signal->session;
 154 out:
 155         read_unlock(&tasklist_lock);
 156
 157         return sid;
 158 }
 159
 160 /*
 161  * Determine if a process group is "orphaned", according to the POSIX
 162  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 163  * by terminal-generated stop signals.  Newly orphaned process groups are
 164  * to receive a SIGHUP and a SIGCONT.
 165  *
 166  * "I ask you, have you ever known what it is to be an orphan?"
 167  */
 168 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 169 {
 170         struct task_struct *p;
 171         int ret = 1;
 172
 173         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 174                 if (p == ignored_task
 175                                 || p->exit_state
 176                                 || p->real_parent->pid == 1)
 177                         continue;
 178                 if (process_group(p->real_parent) != pgrp
 179                             && p->real_parent->signal->session == p->signal->session) {
 180                         ret = 0;
 181                         break;
 182                 }
 183         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 184         return ret;     /* (sighing) "Often!" */
 185 }
 186
 187 int is_orphaned_pgrp(int pgrp)
 188 {
 189         int retval;
 190
 191         read_lock(&tasklist_lock);
 192         retval = will_become_orphaned_pgrp(pgrp, NULL);
 193         read_unlock(&tasklist_lock);
 194
 195         return retval;
 196 }
 197
 198 static int has_stopped_jobs(int pgrp)
 199 {
 200         int retval = 0;
 201         struct task_struct *p;
 202
 203         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 204                 if (p->state != TASK_STOPPED)
 205                         continue;
 206
 207                 /* If p is stopped by a debugger on a signal that won't
 208                    stop it, then don't count p as stopped.  This isn't
 209                    perfect but it's a good approximation.  */
 210                 if (unlikely (p->ptrace)
 211                     && p->exit_code != SIGSTOP
 212                     && p->exit_code != SIGTSTP
 213                     && p->exit_code != SIGTTOU
 214                     && p->exit_code != SIGTTIN)
 215                         continue;
 216
 217                 retval = 1;
 218                 break;
 219         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 220         return retval;
 221 }
 222
 223 /**
 224  * reparent_to_init - Reparent the calling kernel thread to the init task.
 225  *
 226  * If a kernel thread is launched as a result of a system call, or if
 227  * it ever exits, it should generally reparent itself to init so that
 228  * it is correctly cleaned up on exit.
 229  *
 230  * The various task state such as scheduling policy and priority may have
 231  * been inherited from a user process, so we reset them to sane values here.
 232  *
 233  * NOTE that reparent_to_init() gives the caller full capabilities.
 234  */
 235 static void reparent_to_init(void)
 236 {
 237         write_lock_irq(&tasklist_lock);
 238
 239         ptrace_unlink(current);
 240         /* Reparent to init */
 241         remove_parent(current);
 242         current->parent = child_reaper;
 243         current->real_parent = child_reaper;
 244         add_parent(current);
 245
 246         /* Set the exit signal to SIGCHLD so we signal init on exit */
 247         current->exit_signal = SIGCHLD;
 248
 249         if ((current->policy == SCHED_NORMAL ||
 250                         current->policy == SCHED_BATCH)
 251                                 && (task_nice(current) < 0))
 252                 set_user_nice(current, 0);
 253         /* cpus_allowed? */
 254         /* rt_priority? */
 255         /* signals? */
 256         security_task_reparent_to_init(current);
 257         memcpy(current->signal->rlim, init_task.signal->rlim,
 258                sizeof(current->signal->rlim));
 259         atomic_inc(&(INIT_USER->__count));
 260         write_unlock_irq(&tasklist_lock);
 261         switch_uid(INIT_USER);
 262 }
 263
 264 void __set_special_pids(pid_t session, pid_t pgrp)
 265 {
 266         struct task_struct *curr = current->group_leader;
 267
 268         if (curr->signal->session != session) {
 269                 detach_pid(curr, PIDTYPE_SID);
 270                 curr->signal->session = session;
 271                 attach_pid(curr, PIDTYPE_SID, session);
 272         }
 273         if (process_group(curr) != pgrp) {
 274                 detach_pid(curr, PIDTYPE_PGID);
 275                 curr->signal->pgrp = pgrp;
 276                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 277         }
 278 }
 279
 280 void set_special_pids(pid_t session, pid_t pgrp)
 281 {
 282         write_lock_irq(&tasklist_lock);
 283         __set_special_pids(session, pgrp);
 284         write_unlock_irq(&tasklist_lock);
 285 }
 286
 287 /*
 288  * Let kernel threads use this to say that they
 289  * allow a certain signal (since daemonize() will
 290  * have disabled all of them by default).
 291  */
 292 int allow_signal(int sig)
 293 {
 294         if (!valid_signal(sig) || sig < 1)
 295                 return -EINVAL;
 296
 297         spin_lock_irq(&current->sighand->siglock);
 298         sigdelset(&current->blocked, sig);
 299         if (!current->mm) {
 300                 /* Kernel threads handle their own signals.
 301                    Let the signal code know it'll be handled, so
 302                    that they don't get converted to SIGKILL or
 303                    just silently dropped */
 304                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 305         }
 306         recalc_sigpending();
 307         spin_unlock_irq(&current->sighand->siglock);
 308         return 0;
 309 }
 310
 311 EXPORT_SYMBOL(allow_signal);
 312
 313 int disallow_signal(int sig)
 314 {
 315         if (!valid_signal(sig) || sig < 1)
 316                 return -EINVAL;
 317
 318         spin_lock_irq(&current->sighand->siglock);
 319         sigaddset(&current->blocked, sig);
 320         recalc_sigpending();
 321         spin_unlock_irq(&current->sighand->siglock);
 322         return 0;
 323 }
 324
 325 EXPORT_SYMBOL(disallow_signal);
 326
 327 /*
 328  *      Put all the gunge required to become a kernel thread without
 329  *      attached user resources in one place where it belongs.
 330  */
 331
 332 void daemonize(const char *name, ...)
 333 {
 334         va_list args;
 335         struct fs_struct *fs;
 336         sigset_t blocked;
 337
 338         va_start(args, name);
 339         vsnprintf(current->comm, sizeof(current->comm), name, args);
 340         va_end(args);
 341
 342         /*
 343          * If we were started as result of loading a module, close all of the
 344          * user space pages.  We don't need them, and if we didn't close them
 345          * they would be locked into memory.
 346          */
 347         exit_mm(current);
 348
 349         set_special_pids(1, 1);
 350         mutex_lock(&tty_mutex);
 351         current->signal->tty = NULL;
 352         mutex_unlock(&tty_mutex);
 353
 354         /* Block and flush all signals */
 355         sigfillset(&blocked);
 356         sigprocmask(SIG_BLOCK, &blocked, NULL);
 357         flush_signals(current);
 358
 359         /* Become as one with the init task */
 360
 361         exit_fs(current);       /* current->fs->count--; */
 362         fs = init_task.fs;
 363         current->fs = fs;
 364         atomic_inc(&fs->count);
 365         exit_namespace(current);
 366         current->namespace = init_task.namespace;
 367         get_namespace(current->namespace);
 368         exit_files(current);
 369         current->files = init_task.files;
 370         atomic_inc(&current->files->count);
 371
 372         reparent_to_init();
 373 }
 374
 375 EXPORT_SYMBOL(daemonize);
 376
 377 static void close_files(struct files_struct * files)
 378 {
 379         int i, j;
 380         struct fdtable *fdt;
 381
 382         j = 0;
 383
 384         /*
 385          * It is safe to dereference the fd table without RCU or
 386          * ->file_lock because this is the last reference to the
 387          * files structure.
 388          */
 389         fdt = files_fdtable(files);
 390         for (;;) {
 391                 unsigned long set;
 392                 i = j * __NFDBITS;
 393                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 394                         break;
 395                 set = fdt->open_fds->fds_bits[j++];
 396                 while (set) {
 397                         if (set & 1) {
 398                                 struct file * file = xchg(&fdt->fd[i], NULL);
 399                                 if (file)
 400                                         filp_close(file, files);
 401                         }
 402                         i++;
 403                         set >>= 1;
 404                 }
 405         }
 406 }
 407
 408 struct files_struct *get_files_struct(struct task_struct *task)
 409 {
 410         struct files_struct *files;
 411
 412         task_lock(task);
 413         files = task->files;
 414         if (files)
 415                 atomic_inc(&files->count);
 416         task_unlock(task);
 417
 418         return files;
 419 }
 420
 421 void fastcall put_files_struct(struct files_struct *files)
 422 {
 423         struct fdtable *fdt;
 424
 425         if (atomic_dec_and_test(&files->count)) {
 426                 close_files(files);
 427                 /*
 428                  * Free the fd and fdset arrays if we expanded them.
 429                  * If the fdtable was embedded, pass files for freeing
 430                  * at the end of the RCU grace period. Otherwise,
 431                  * you can free files immediately.
 432                  */
 433                 fdt = files_fdtable(files);
 434                 if (fdt == &files->fdtab)
 435                         fdt->free_files = files;
 436                 else
 437                         kmem_cache_free(files_cachep, files);
 438                 free_fdtable(fdt);
 439         }
 440 }
 441
 442 EXPORT_SYMBOL(put_files_struct);
 443
 444 static inline void __exit_files(struct task_struct *tsk)
 445 {
 446         struct files_struct * files = tsk->files;
 447
 448         if (files) {
 449                 task_lock(tsk);
 450                 tsk->files = NULL;
 451                 task_unlock(tsk);
 452                 put_files_struct(files);
 453         }
 454 }
 455
 456 void exit_files(struct task_struct *tsk)
 457 {
 458         __exit_files(tsk);
 459 }
 460
 461 static inline void __put_fs_struct(struct fs_struct *fs)
 462 {
 463         /* No need to hold fs->lock if we are killing it */
 464         if (atomic_dec_and_test(&fs->count)) {
 465                 dput(fs->root);
 466                 mntput(fs->rootmnt);
 467                 dput(fs->pwd);
 468                 mntput(fs->pwdmnt);
 469                 if (fs->altroot) {
 470                         dput(fs->altroot);
 471                         mntput(fs->altrootmnt);
 472                 }
 473                 kmem_cache_free(fs_cachep, fs);
 474         }
 475 }
 476
 477 void put_fs_struct(struct fs_struct *fs)
 478 {
 479         __put_fs_struct(fs);
 480 }
 481
 482 static inline void __exit_fs(struct task_struct *tsk)
 483 {
 484         struct fs_struct * fs = tsk->fs;
 485
 486         if (fs) {
 487                 task_lock(tsk);
 488                 tsk->fs = NULL;
 489                 task_unlock(tsk);
 490                 __put_fs_struct(fs);
 491         }
 492 }
 493
 494 void exit_fs(struct task_struct *tsk)
 495 {
 496         __exit_fs(tsk);
 497 }
 498
 499 EXPORT_SYMBOL_GPL(exit_fs);
 500
 501 /*
 502  * Turn us into a lazy TLB process if we
 503  * aren't already..
 504  */
 505 static void exit_mm(struct task_struct * tsk)
 506 {
 507         struct mm_struct *mm = tsk->mm;
 508
 509         mm_release(tsk, mm);
 510         if (!mm)
 511                 return;
 512         /*
 513          * Serialize with any possible pending coredump.
 514          * We must hold mmap_sem around checking core_waiters
 515          * and clearing tsk->mm.  The core-inducing thread
 516          * will increment core_waiters for each thread in the
 517          * group with ->mm != NULL.
 518          */
 519         down_read(&mm->mmap_sem);
 520         if (mm->core_waiters) {
 521                 up_read(&mm->mmap_sem);
 522                 down_write(&mm->mmap_sem);
 523                 if (!--mm->core_waiters)
 524                         complete(mm->core_startup_done);
 525                 up_write(&mm->mmap_sem);
 526
 527                 wait_for_completion(&mm->core_done);
 528                 down_read(&mm->mmap_sem);
 529         }
 530         atomic_inc(&mm->mm_count);
 531         if (mm != tsk->active_mm) BUG();
 532         /* more a memory barrier than a real lock */
 533         task_lock(tsk);
 534         tsk->mm = NULL;
 535         up_read(&mm->mmap_sem);
 536         enter_lazy_tlb(mm, current);
 537         task_unlock(tsk);
 538         mmput(mm);
 539 }
 540
 541 static inline void choose_new_parent(task_t *p, task_t *reaper)
 542 {
 543         /*
 544          * Make sure we're not reparenting to ourselves and that
 545          * the parent is not a zombie.
 546          */
 547         BUG_ON(p == reaper || reaper->exit_state);
 548         p->real_parent = reaper;
 549 }
 550
 551 static void reparent_thread(task_t *p, task_t *father, int traced)
 552 {
 553         /* We don't want people slaying init.  */
 554         if (p->exit_signal != -1)
 555                 p->exit_signal = SIGCHLD;
 556
 557         if (p->pdeath_signal)
 558                 /* We already hold the tasklist_lock here.  */
 559                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 560
 561         /* Move the child from its dying parent to the new one.  */
 562         if (unlikely(traced)) {
 563                 /* Preserve ptrace links if someone else is tracing this child.  */
 564                 list_del_init(&p->ptrace_list);
 565                 if (p->parent != p->real_parent)
 566                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 567         } else {
 568                 /* If this child is being traced, then we're the one tracing it
 569                  * anyway, so let go of it.
 570                  */
 571                 p->ptrace = 0;
 572                 list_del_init(&p->sibling);
 573                 p->parent = p->real_parent;
 574                 list_add_tail(&p->sibling, &p->parent->children);
 575
 576                 /* If we'd notified the old parent about this child's death,
 577                  * also notify the new parent.
 578                  */
 579                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 580                     thread_group_empty(p))
 581                         do_notify_parent(p, p->exit_signal);
 582                 else if (p->state == TASK_TRACED) {
 583                         /*
 584                          * If it was at a trace stop, turn it into
 585                          * a normal stop since it's no longer being
 586                          * traced.
 587                          */
 588                         ptrace_untrace(p);
 589                 }
 590         }
 591
 592         /*
 593          * process group orphan check
 594          * Case ii: Our child is in a different pgrp
 595          * than we are, and it was the only connection
 596          * outside, so the child pgrp is now orphaned.
 597          */
 598         if ((process_group(p) != process_group(father)) &&
 599             (p->signal->session == father->signal->session)) {
 600                 int pgrp = process_group(p);
 601
 602                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 603                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 604                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 605                 }
 606         }
 607 }
 608
 609 /*
 610  * When we die, we re-parent all our children.
 611  * Try to give them to another thread in our thread
 612  * group, and if no such member exists, give it to
 613  * the global child reaper process (ie "init")
 614  */
 615 static void forget_original_parent(struct task_struct * father,
 616                                           struct list_head *to_release)
 617 {
 618         struct task_struct *p, *reaper = father;
 619         struct list_head *_p, *_n;
 620
 621         do {
 622                 reaper = next_thread(reaper);
 623                 if (reaper == father) {
 624                         reaper = child_reaper;
 625                         break;
 626                 }
 627         } while (reaper->exit_state);
 628
 629         /*
 630          * There are only two places where our children can be:
 631          *
 632          * - in our child list
 633          * - in our ptraced child list
 634          *
 635          * Search them and reparent children.
 636          */
 637         list_for_each_safe(_p, _n, &father->children) {
 638                 int ptrace;
 639                 p = list_entry(_p,struct task_struct,sibling);
 640
 641                 ptrace = p->ptrace;
 642
 643                 /* if father isn't the real parent, then ptrace must be enabled */
 644                 BUG_ON(father != p->real_parent && !ptrace);
 645
 646                 if (father == p->real_parent) {
 647                         /* reparent with a reaper, real father it's us */
 648                         choose_new_parent(p, reaper);
 649                         reparent_thread(p, father, 0);
 650                 } else {
 651                         /* reparent ptraced task to its real parent */
 652                         __ptrace_unlink (p);
 653                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 654                             thread_group_empty(p))
 655                                 do_notify_parent(p, p->exit_signal);
 656                 }
 657
 658                 /*
 659                  * if the ptraced child is a zombie with exit_signal == -1
 660                  * we must collect it before we exit, or it will remain
 661                  * zombie forever since we prevented it from self-reap itself
 662                  * while it was being traced by us, to be able to see it in wait4.
 663                  */
 664                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 665                         list_add(&p->ptrace_list, to_release);
 666         }
 667         list_for_each_safe(_p, _n, &father->ptrace_children) {
 668                 p = list_entry(_p,struct task_struct,ptrace_list);
 669                 choose_new_parent(p, reaper);
 670                 reparent_thread(p, father, 1);
 671         }
 672 }
 673
 674 /*
 675  * Send signals to all our closest relatives so that they know
 676  * to properly mourn us..
 677  */
 678 static void exit_notify(struct task_struct *tsk)
 679 {
 680         int state;
 681         struct task_struct *t;
 682         struct list_head ptrace_dead, *_p, *_n;
 683
 684         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 685             && !thread_group_empty(tsk)) {
 686                 /*
 687                  * This occurs when there was a race between our exit
 688                  * syscall and a group signal choosing us as the one to
 689                  * wake up.  It could be that we are the only thread
 690                  * alerted to check for pending signals, but another thread
 691                  * should be woken now to take the signal since we will not.
 692                  * Now we'll wake all the threads in the group just to make
 693                  * sure someone gets all the pending signals.
 694                  */
 695                 read_lock(&tasklist_lock);
 696                 spin_lock_irq(&tsk->sighand->siglock);
 697                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 698                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 699                                 recalc_sigpending_tsk(t);
 700                                 if (signal_pending(t))
 701                                         signal_wake_up(t, 0);
 702                         }
 703                 spin_unlock_irq(&tsk->sighand->siglock);
 704                 read_unlock(&tasklist_lock);
 705         }
 706
 707         write_lock_irq(&tasklist_lock);
 708
 709         /*
 710          * This does two things:
 711          *
 712          * A.  Make init inherit all the child processes
 713          * B.  Check to see if any process groups have become orphaned
 714          *      as a result of our exiting, and if they have any stopped
 715          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 716          */
 717
 718         INIT_LIST_HEAD(&ptrace_dead);
 719         forget_original_parent(tsk, &ptrace_dead);
 720         BUG_ON(!list_empty(&tsk->children));
 721         BUG_ON(!list_empty(&tsk->ptrace_children));
 722
 723         /*
 724          * Check to see if any process groups have become orphaned
 725          * as a result of our exiting, and if they have any stopped
 726          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 727          *
 728          * Case i: Our father is in a different pgrp than we are
 729          * and we were the only connection outside, so our pgrp
 730          * is about to become orphaned.
 731          */
 732
 733         t = tsk->real_parent;
 734
 735         if ((process_group(t) != process_group(tsk)) &&
 736             (t->signal->session == tsk->signal->session) &&
 737             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 738             has_stopped_jobs(process_group(tsk))) {
 739                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 740                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 741         }
 742
 743         /* Let father know we died
 744          *
 745          * Thread signals are configurable, but you aren't going to use
 746          * that to send signals to arbitary processes.
 747          * That stops right now.
 748          *
 749          * If the parent exec id doesn't match the exec id we saved
 750          * when we started then we know the parent has changed security
 751          * domain.
 752          *
 753          * If our self_exec id doesn't match our parent_exec_id then
 754          * we have changed execution domain as these two values started
 755          * the same after a fork.
 756          *
 757          */
 758
 759         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 760             ( tsk->parent_exec_id != t->self_exec_id  ||
 761               tsk->self_exec_id != tsk->parent_exec_id)
 762             && !capable(CAP_KILL))
 763                 tsk->exit_signal = SIGCHLD;
 764
 765
 766         /* If something other than our normal parent is ptracing us, then
 767          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 768          * only has special meaning to our real parent.
 769          */
 770         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 771                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 772                 do_notify_parent(tsk, signal);
 773         } else if (tsk->ptrace) {
 774                 do_notify_parent(tsk, SIGCHLD);
 775         }
 776
 777         state = EXIT_ZOMBIE;
 778         if (tsk->exit_signal == -1 &&
 779             (likely(tsk->ptrace == 0) ||
 780              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 781                 state = EXIT_DEAD;
 782         tsk->exit_state = state;
 783
 784         write_unlock_irq(&tasklist_lock);
 785
 786         list_for_each_safe(_p, _n, &ptrace_dead) {
 787                 list_del_init(_p);
 788                 t = list_entry(_p,struct task_struct,ptrace_list);
 789                 release_task(t);
 790         }
 791
 792         /* If the process is dead, release it - nobody will wait for it */
 793         if (state == EXIT_DEAD)
 794                 release_task(tsk);
 795 }
 796
 797 fastcall NORET_TYPE void do_exit(long code)
 798 {
 799         struct task_struct *tsk = current;
 800         int group_dead;
 801
 802         profile_task_exit(tsk);
 803
 804         WARN_ON(atomic_read(&tsk->fs_excl));
 805
 806         if (unlikely(in_interrupt()))
 807                 panic("Aiee, killing interrupt handler!");
 808         if (unlikely(!tsk->pid))
 809                 panic("Attempted to kill the idle task!");
 810         if (unlikely(tsk == child_reaper))
 811                 panic("Attempted to kill init!");
 812
 813         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 814                 current->ptrace_message = code;
 815                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 816         }
 817
 818         /*
 819          * We're taking recursive faults here in do_exit. Safest is to just
 820          * leave this task alone and wait for reboot.
 821          */
 822         if (unlikely(tsk->flags & PF_EXITING)) {
 823                 printk(KERN_ALERT
 824                         "Fixing recursive fault but reboot is needed!\n");
 825                 if (tsk->io_context)
 826                         exit_io_context();
 827                 set_current_state(TASK_UNINTERRUPTIBLE);
 828                 schedule();
 829         }
 830
 831         tsk->flags |= PF_EXITING;
 832
 833         /*
 834          * Make sure we don't try to process any timer firings
 835          * while we are already exiting.
 836          */
 837         tsk->it_virt_expires = cputime_zero;
 838         tsk->it_prof_expires = cputime_zero;
 839         tsk->it_sched_expires = 0;
 840
 841         if (unlikely(in_atomic()))
 842                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 843                                 current->comm, current->pid,
 844                                 preempt_count());
 845
 846         acct_update_integrals(tsk);
 847         if (tsk->mm) {
 848                 update_hiwater_rss(tsk->mm);
 849                 update_hiwater_vm(tsk->mm);
 850         }
 851         group_dead = atomic_dec_and_test(&tsk->signal->live);
 852         if (group_dead) {
 853                 hrtimer_cancel(&tsk->signal->real_timer);
 854                 exit_itimers(tsk->signal);
 855                 acct_process(code);
 856         }
 857         if (unlikely(tsk->robust_list))
 858                 exit_robust_list(tsk);
 859 #ifdef CONFIG_COMPAT
 860         if (unlikely(tsk->compat_robust_list))
 861                 compat_exit_robust_list(tsk);
 862 #endif
 863         exit_mm(tsk);
 864
 865         exit_sem(tsk);
 866         __exit_files(tsk);
 867         __exit_fs(tsk);
 868         exit_namespace(tsk);
 869         exit_thread();
 870         cpuset_exit(tsk);
 871         exit_keys(tsk);
 872
 873         if (group_dead && tsk->signal->leader)
 874                 disassociate_ctty(1);
 875
 876         module_put(task_thread_info(tsk)->exec_domain->module);
 877         if (tsk->binfmt)
 878                 module_put(tsk->binfmt->module);
 879
 880         tsk->exit_code = code;
 881         proc_exit_connector(tsk);
 882         exit_notify(tsk);
 883 #ifdef CONFIG_NUMA
 884         mpol_free(tsk->mempolicy);
 885         tsk->mempolicy = NULL;
 886 #endif
 887         /*
 888          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
 889          */
 890         mutex_debug_check_no_locks_held(tsk);
 891
 892         if (tsk->io_context)
 893                 exit_io_context();
 894
 895         /* PF_DEAD causes final put_task_struct after we schedule. */
 896         preempt_disable();
 897         BUG_ON(tsk->flags & PF_DEAD);
 898         tsk->flags |= PF_DEAD;
 899
 900         schedule();
 901         BUG();
 902         /* Avoid "noreturn function does return".  */
 903         for (;;) ;
 904 }
 905
 906 EXPORT_SYMBOL_GPL(do_exit);
 907
 908 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 909 {
 910         if (comp)
 911                 complete(comp);
 912
 913         do_exit(code);
 914 }
 915
 916 EXPORT_SYMBOL(complete_and_exit);
 917
 918 asmlinkage long sys_exit(int error_code)
 919 {
 920         do_exit((error_code&0xff)<<8);
 921 }
 922
 923 task_t fastcall *next_thread(const task_t *p)
 924 {
 925         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 926 }
 927
 928 EXPORT_SYMBOL(next_thread);
 929
 930 /*
 931  * Take down every thread in the group.  This is called by fatal signals
 932  * as well as by sys_exit_group (below).
 933  */
 934 NORET_TYPE void
 935 do_group_exit(int exit_code)
 936 {
 937         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 938
 939         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 940                 exit_code = current->signal->group_exit_code;
 941         else if (!thread_group_empty(current)) {
 942                 struct signal_struct *const sig = current->signal;
 943                 struct sighand_struct *const sighand = current->sighand;
 944                 read_lock(&tasklist_lock);
 945                 spin_lock_irq(&sighand->siglock);
 946                 if (sig->flags & SIGNAL_GROUP_EXIT)
 947                         /* Another thread got here before we took the lock.  */
 948                         exit_code = sig->group_exit_code;
 949                 else {
 950                         sig->group_exit_code = exit_code;
 951                         zap_other_threads(current);
 952                 }
 953                 spin_unlock_irq(&sighand->siglock);
 954                 read_unlock(&tasklist_lock);
 955         }
 956
 957         do_exit(exit_code);
 958         /* NOTREACHED */
 959 }
 960
 961 /*
 962  * this kills every thread in the thread group. Note that any externally
 963  * wait4()-ing process will get the correct exit code - even if this
 964  * thread is not the thread group leader.
 965  */
 966 asmlinkage void sys_exit_group(int error_code)
 967 {
 968         do_group_exit((error_code & 0xff) << 8);
 969 }
 970
 971 static int eligible_child(pid_t pid, int options, task_t *p)
 972 {
 973         if (pid > 0) {
 974                 if (p->pid != pid)
 975                         return 0;
 976         } else if (!pid) {
 977                 if (process_group(p) != process_group(current))
 978                         return 0;
 979         } else if (pid != -1) {
 980                 if (process_group(p) != -pid)
 981                         return 0;
 982         }
 983
 984         /*
 985          * Do not consider detached threads that are
 986          * not ptraced:
 987          */
 988         if (p->exit_signal == -1 && !p->ptrace)
 989                 return 0;
 990
 991         /* Wait for all children (clone and not) if __WALL is set;
 992          * otherwise, wait for clone children *only* if __WCLONE is
 993          * set; otherwise, wait for non-clone children *only*.  (Note:
 994          * A "clone" child here is one that reports to its parent
 995          * using a signal other than SIGCHLD.) */
 996         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 997             && !(options & __WALL))
 998                 return 0;
 999         /*
1000          * Do not consider thread group leaders that are
1001          * in a non-empty thread group:
1002          */
1003         if (current->tgid != p->tgid && delay_group_leader(p))
1004                 return 2;
1005
1006         if (security_task_wait(p))
1007                 return 0;
1008
1009         return 1;
1010 }
1011
1012 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
1013                                int why, int status,
1014                                struct siginfo __user *infop,
1015                                struct rusage __user *rusagep)
1016 {
1017         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1018         put_task_struct(p);
1019         if (!retval)
1020                 retval = put_user(SIGCHLD, &infop->si_signo);
1021         if (!retval)
1022                 retval = put_user(0, &infop->si_errno);
1023         if (!retval)
1024                 retval = put_user((short)why, &infop->si_code);
1025         if (!retval)
1026                 retval = put_user(pid, &infop->si_pid);
1027         if (!retval)
1028                 retval = put_user(uid, &infop->si_uid);
1029         if (!retval)
1030                 retval = put_user(status, &infop->si_status);
1031         if (!retval)
1032                 retval = pid;
1033         return retval;
1034 }
1035
1036 /*
1037  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1038  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1039  * the lock and this task is uninteresting.  If we return nonzero, we have
1040  * released the lock and the system call should return.
1041  */
1042 static int wait_task_zombie(task_t *p, int noreap,
1043                             struct siginfo __user *infop,
1044                             int __user *stat_addr, struct rusage __user *ru)
1045 {
1046         unsigned long state;
1047         int retval;
1048         int status;
1049
1050         if (unlikely(noreap)) {
1051                 pid_t pid = p->pid;
1052                 uid_t uid = p->uid;
1053                 int exit_code = p->exit_code;
1054                 int why, status;
1055
1056                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1057                         return 0;
1058                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1059                         return 0;
1060                 get_task_struct(p);
1061                 read_unlock(&tasklist_lock);
1062                 if ((exit_code & 0x7f) == 0) {
1063                         why = CLD_EXITED;
1064                         status = exit_code >> 8;
1065                 } else {
1066                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1067                         status = exit_code & 0x7f;
1068                 }
1069                 return wait_noreap_copyout(p, pid, uid, why,
1070                                            status, infop, ru);
1071         }
1072
1073         /*
1074          * Try to move the task's state to DEAD
1075          * only one thread is allowed to do this:
1076          */
1077         state = xchg(&p->exit_state, EXIT_DEAD);
1078         if (state != EXIT_ZOMBIE) {
1079                 BUG_ON(state != EXIT_DEAD);
1080                 return 0;
1081         }
1082         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1083                 /*
1084                  * This can only happen in a race with a ptraced thread
1085                  * dying on another processor.
1086                  */
1087                 return 0;
1088         }
1089
1090         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1091                 struct signal_struct *psig;
1092                 struct signal_struct *sig;
1093
1094                 /*
1095                  * The resource counters for the group leader are in its
1096                  * own task_struct.  Those for dead threads in the group
1097                  * are in its signal_struct, as are those for the child
1098                  * processes it has previously reaped.  All these
1099                  * accumulate in the parent's signal_struct c* fields.
1100                  *
1101                  * We don't bother to take a lock here to protect these
1102                  * p->signal fields, because they are only touched by
1103                  * __exit_signal, which runs with tasklist_lock
1104                  * write-locked anyway, and so is excluded here.  We do
1105                  * need to protect the access to p->parent->signal fields,
1106                  * as other threads in the parent group can be right
1107                  * here reaping other children at the same time.
1108                  */
1109                 spin_lock_irq(&p->parent->sighand->siglock);
1110                 psig = p->parent->signal;
1111                 sig = p->signal;
1112                 psig->cutime =
1113                         cputime_add(psig->cutime,
1114                         cputime_add(p->utime,
1115                         cputime_add(sig->utime,
1116                                     sig->cutime)));
1117                 psig->cstime =
1118                         cputime_add(psig->cstime,
1119                         cputime_add(p->stime,
1120                         cputime_add(sig->stime,
1121                                     sig->cstime)));
1122                 psig->cmin_flt +=
1123                         p->min_flt + sig->min_flt + sig->cmin_flt;
1124                 psig->cmaj_flt +=
1125                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1126                 psig->cnvcsw +=
1127                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1128                 psig->cnivcsw +=
1129                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1130                 spin_unlock_irq(&p->parent->sighand->siglock);
1131         }
1132
1133         /*
1134          * Now we are sure this task is interesting, and no other
1135          * thread can reap it because we set its state to EXIT_DEAD.
1136          */
1137         read_unlock(&tasklist_lock);
1138
1139         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1140         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1141                 ? p->signal->group_exit_code : p->exit_code;
1142         if (!retval && stat_addr)
1143                 retval = put_user(status, stat_addr);
1144         if (!retval && infop)
1145                 retval = put_user(SIGCHLD, &infop->si_signo);
1146         if (!retval && infop)
1147                 retval = put_user(0, &infop->si_errno);
1148         if (!retval && infop) {
1149                 int why;
1150
1151                 if ((status & 0x7f) == 0) {
1152                         why = CLD_EXITED;
1153                         status >>= 8;
1154                 } else {
1155                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1156                         status &= 0x7f;
1157                 }
1158                 retval = put_user((short)why, &infop->si_code);
1159                 if (!retval)
1160                         retval = put_user(status, &infop->si_status);
1161         }
1162         if (!retval && infop)
1163                 retval = put_user(p->pid, &infop->si_pid);
1164         if (!retval && infop)
1165                 retval = put_user(p->uid, &infop->si_uid);
1166         if (retval) {
1167                 // TODO: is this safe?
1168                 p->exit_state = EXIT_ZOMBIE;
1169                 return retval;
1170         }
1171         retval = p->pid;
1172         if (p->real_parent != p->parent) {
1173                 write_lock_irq(&tasklist_lock);
1174                 /* Double-check with lock held.  */
1175                 if (p->real_parent != p->parent) {
1176                         __ptrace_unlink(p);
1177                         // TODO: is this safe?
1178                         p->exit_state = EXIT_ZOMBIE;
1179                         /*
1180                          * If this is not a detached task, notify the parent.
1181                          * If it's still not detached after that, don't release
1182                          * it now.
1183                          */
1184                         if (p->exit_signal != -1) {
1185                                 do_notify_parent(p, p->exit_signal);
1186                                 if (p->exit_signal != -1)
1187                                         p = NULL;
1188                         }
1189                 }
1190                 write_unlock_irq(&tasklist_lock);
1191         }
1192         if (p != NULL)
1193                 release_task(p);
1194         BUG_ON(!retval);
1195         return retval;
1196 }
1197
1198 /*
1199  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1200  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1201  * the lock and this task is uninteresting.  If we return nonzero, we have
1202  * released the lock and the system call should return.
1203  */
1204 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1205                              struct siginfo __user *infop,
1206                              int __user *stat_addr, struct rusage __user *ru)
1207 {
1208         int retval, exit_code;
1209
1210         if (!p->exit_code)
1211                 return 0;
1212         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1213             p->signal && p->signal->group_stop_count > 0)
1214                 /*
1215                  * A group stop is in progress and this is the group leader.
1216                  * We won't report until all threads have stopped.
1217                  */
1218                 return 0;
1219
1220         /*
1221          * Now we are pretty sure this task is interesting.
1222          * Make sure it doesn't get reaped out from under us while we
1223          * give up the lock and then examine it below.  We don't want to
1224          * keep holding onto the tasklist_lock while we call getrusage and
1225          * possibly take page faults for user memory.
1226          */
1227         get_task_struct(p);
1228         read_unlock(&tasklist_lock);
1229
1230         if (unlikely(noreap)) {
1231                 pid_t pid = p->pid;
1232                 uid_t uid = p->uid;
1233                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1234
1235                 exit_code = p->exit_code;
1236                 if (unlikely(!exit_code) ||
1237                     unlikely(p->state & TASK_TRACED))
1238                         goto bail_ref;
1239                 return wait_noreap_copyout(p, pid, uid,
1240                                            why, (exit_code << 8) | 0x7f,
1241                                            infop, ru);
1242         }
1243
1244         write_lock_irq(&tasklist_lock);
1245
1246         /*
1247          * This uses xchg to be atomic with the thread resuming and setting
1248          * it.  It must also be done with the write lock held to prevent a
1249          * race with the EXIT_ZOMBIE case.
1250          */
1251         exit_code = xchg(&p->exit_code, 0);
1252         if (unlikely(p->exit_state)) {
1253                 /*
1254                  * The task resumed and then died.  Let the next iteration
1255                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1256                  * already be zero here if it resumed and did _exit(0).
1257                  * The task itself is dead and won't touch exit_code again;
1258                  * other processors in this function are locked out.
1259                  */
1260                 p->exit_code = exit_code;
1261                 exit_code = 0;
1262         }
1263         if (unlikely(exit_code == 0)) {
1264                 /*
1265                  * Another thread in this function got to it first, or it
1266                  * resumed, or it resumed and then died.
1267                  */
1268                 write_unlock_irq(&tasklist_lock);
1269 bail_ref:
1270                 put_task_struct(p);
1271                 /*
1272                  * We are returning to the wait loop without having successfully
1273                  * removed the process and having released the lock. We cannot
1274                  * continue, since the "p" task pointer is potentially stale.
1275                  *
1276                  * Return -EAGAIN, and do_wait() will restart the loop from the
1277                  * beginning. Do _not_ re-acquire the lock.
1278                  */
1279                 return -EAGAIN;
1280         }
1281
1282         /* move to end of parent's list to avoid starvation */
1283         remove_parent(p);
1284         add_parent(p);
1285
1286         write_unlock_irq(&tasklist_lock);
1287
1288         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1289         if (!retval && stat_addr)
1290                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1291         if (!retval && infop)
1292                 retval = put_user(SIGCHLD, &infop->si_signo);
1293         if (!retval && infop)
1294                 retval = put_user(0, &infop->si_errno);
1295         if (!retval && infop)
1296                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1297                                           ? CLD_TRAPPED : CLD_STOPPED),
1298                                   &infop->si_code);
1299         if (!retval && infop)
1300                 retval = put_user(exit_code, &infop->si_status);
1301         if (!retval && infop)
1302                 retval = put_user(p->pid, &infop->si_pid);
1303         if (!retval && infop)
1304                 retval = put_user(p->uid, &infop->si_uid);
1305         if (!retval)
1306                 retval = p->pid;
1307         put_task_struct(p);
1308
1309         BUG_ON(!retval);
1310         return retval;
1311 }
1312
1313 /*
1314  * Handle do_wait work for one task in a live, non-stopped state.
1315  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1316  * the lock and this task is uninteresting.  If we return nonzero, we have
1317  * released the lock and the system call should return.
1318  */
1319 static int wait_task_continued(task_t *p, int noreap,
1320                                struct siginfo __user *infop,
1321                                int __user *stat_addr, struct rusage __user *ru)
1322 {
1323         int retval;
1324         pid_t pid;
1325         uid_t uid;
1326
1327         if (unlikely(!p->signal))
1328                 return 0;
1329
1330         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1331                 return 0;
1332
1333         spin_lock_irq(&p->sighand->siglock);
1334         /* Re-check with the lock held.  */
1335         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1336                 spin_unlock_irq(&p->sighand->siglock);
1337                 return 0;
1338         }
1339         if (!noreap)
1340                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1341         spin_unlock_irq(&p->sighand->siglock);
1342
1343         pid = p->pid;
1344         uid = p->uid;
1345         get_task_struct(p);
1346         read_unlock(&tasklist_lock);
1347
1348         if (!infop) {
1349                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1350                 put_task_struct(p);
1351                 if (!retval && stat_addr)
1352                         retval = put_user(0xffff, stat_addr);
1353                 if (!retval)
1354                         retval = p->pid;
1355         } else {
1356                 retval = wait_noreap_copyout(p, pid, uid,
1357                                              CLD_CONTINUED, SIGCONT,
1358                                              infop, ru);
1359                 BUG_ON(retval == 0);
1360         }
1361
1362         return retval;
1363 }
1364
1365
1366 static inline int my_ptrace_child(struct task_struct *p)
1367 {
1368         if (!(p->ptrace & PT_PTRACED))
1369                 return 0;
1370         if (!(p->ptrace & PT_ATTACHED))
1371                 return 1;
1372         /*
1373          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1374          * we are the attacher.  If we are the real parent, this is a race
1375          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1376          * which we have to switch the parent links, but has already set
1377          * the flags in p->ptrace.
1378          */
1379         return (p->parent != p->real_parent);
1380 }
1381
1382 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1383                     int __user *stat_addr, struct rusage __user *ru)
1384 {
1385         DECLARE_WAITQUEUE(wait, current);
1386         struct task_struct *tsk;
1387         int flag, retval;
1388
1389         add_wait_queue(&current->signal->wait_chldexit,&wait);
1390 repeat:
1391         /*
1392          * We will set this flag if we see any child that might later
1393          * match our criteria, even if we are not able to reap it yet.
1394          */
1395         flag = 0;
1396         current->state = TASK_INTERRUPTIBLE;
1397         read_lock(&tasklist_lock);
1398         tsk = current;
1399         do {
1400                 struct task_struct *p;
1401                 struct list_head *_p;
1402                 int ret;
1403
1404                 list_for_each(_p,&tsk->children) {
1405                         p = list_entry(_p,struct task_struct,sibling);
1406
1407                         ret = eligible_child(pid, options, p);
1408                         if (!ret)
1409                                 continue;
1410
1411                         switch (p->state) {
1412                         case TASK_TRACED:
1413                                 /*
1414                                  * When we hit the race with PTRACE_ATTACH,
1415                                  * we will not report this child.  But the
1416                                  * race means it has not yet been moved to
1417                                  * our ptrace_children list, so we need to
1418                                  * set the flag here to avoid a spurious ECHILD
1419                                  * when the race happens with the only child.
1420                                  */
1421                                 flag = 1;
1422                                 if (!my_ptrace_child(p))
1423                                         continue;
1424                                 /*FALLTHROUGH*/
1425                         case TASK_STOPPED:
1426                                 /*
1427                                  * It's stopped now, so it might later
1428                                  * continue, exit, or stop again.
1429                                  */
1430                                 flag = 1;
1431                                 if (!(options & WUNTRACED) &&
1432                                     !my_ptrace_child(p))
1433                                         continue;
1434                                 retval = wait_task_stopped(p, ret == 2,
1435                                                            (options & WNOWAIT),
1436                                                            infop,
1437                                                            stat_addr, ru);
1438                                 if (retval == -EAGAIN)
1439                                         goto repeat;
1440                                 if (retval != 0) /* He released the lock.  */
1441                                         goto end;
1442                                 break;
1443                         default:
1444                         // case EXIT_DEAD:
1445                                 if (p->exit_state == EXIT_DEAD)
1446                                         continue;
1447                         // case EXIT_ZOMBIE:
1448                                 if (p->exit_state == EXIT_ZOMBIE) {
1449                                         /*
1450                                          * Eligible but we cannot release
1451                                          * it yet:
1452                                          */
1453                                         if (ret == 2)
1454                                                 goto check_continued;
1455                                         if (!likely(options & WEXITED))
1456                                                 continue;
1457                                         retval = wait_task_zombie(
1458                                                 p, (options & WNOWAIT),
1459                                                 infop, stat_addr, ru);
1460                                         /* He released the lock.  */
1461                                         if (retval != 0)
1462                                                 goto end;
1463                                         break;
1464                                 }
1465 check_continued:
1466                                 /*
1467                                  * It's running now, so it might later
1468                                  * exit, stop, or stop and then continue.
1469                                  */
1470                                 flag = 1;
1471                                 if (!unlikely(options & WCONTINUED))
1472                                         continue;
1473                                 retval = wait_task_continued(
1474                                         p, (options & WNOWAIT),
1475                                         infop, stat_addr, ru);
1476                                 if (retval != 0) /* He released the lock.  */
1477                                         goto end;
1478                                 break;
1479                         }
1480                 }
1481                 if (!flag) {
1482                         list_for_each(_p, &tsk->ptrace_children) {
1483                                 p = list_entry(_p, struct task_struct,
1484                                                 ptrace_list);
1485                                 if (!eligible_child(pid, options, p))
1486                                         continue;
1487                                 flag = 1;
1488                                 break;
1489                         }
1490                 }
1491                 if (options & __WNOTHREAD)
1492                         break;
1493                 tsk = next_thread(tsk);
1494                 if (tsk->signal != current->signal)
1495                         BUG();
1496         } while (tsk != current);
1497
1498         read_unlock(&tasklist_lock);
1499         if (flag) {
1500                 retval = 0;
1501                 if (options & WNOHANG)
1502                         goto end;
1503                 retval = -ERESTARTSYS;
1504                 if (signal_pending(current))
1505                         goto end;
1506                 schedule();
1507                 goto repeat;
1508         }
1509         retval = -ECHILD;
1510 end:
1511         current->state = TASK_RUNNING;
1512         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1513         if (infop) {
1514                 if (retval > 0)
1515                 retval = 0;
1516                 else {
1517                         /*
1518                          * For a WNOHANG return, clear out all the fields
1519                          * we would set so the user can easily tell the
1520                          * difference.
1521                          */
1522                         if (!retval)
1523                                 retval = put_user(0, &infop->si_signo);
1524                         if (!retval)
1525                                 retval = put_user(0, &infop->si_errno);
1526                         if (!retval)
1527                                 retval = put_user(0, &infop->si_code);
1528                         if (!retval)
1529                                 retval = put_user(0, &infop->si_pid);
1530                         if (!retval)
1531                                 retval = put_user(0, &infop->si_uid);
1532                         if (!retval)
1533                                 retval = put_user(0, &infop->si_status);
1534                 }
1535         }
1536         return retval;
1537 }
1538
1539 asmlinkage long sys_waitid(int which, pid_t pid,
1540                            struct siginfo __user *infop, int options,
1541                            struct rusage __user *ru)
1542 {
1543         long ret;
1544
1545         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1546                 return -EINVAL;
1547         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1548                 return -EINVAL;
1549
1550         switch (which) {
1551         case P_ALL:
1552                 pid = -1;
1553                 break;
1554         case P_PID:
1555                 if (pid <= 0)
1556                         return -EINVAL;
1557                 break;
1558         case P_PGID:
1559                 if (pid <= 0)
1560                         return -EINVAL;
1561                 pid = -pid;
1562                 break;
1563         default:
1564                 return -EINVAL;
1565         }
1566
1567         ret = do_wait(pid, options, infop, NULL, ru);
1568
1569         /* avoid REGPARM breakage on x86: */
1570         prevent_tail_call(ret);
1571         return ret;
1572 }
1573
1574 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1575                           int options, struct rusage __user *ru)
1576 {
1577         long ret;
1578
1579         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1580                         __WNOTHREAD|__WCLONE|__WALL))
1581                 return -EINVAL;
1582         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1583
1584         /* avoid REGPARM breakage on x86: */
1585         prevent_tail_call(ret);
1586         return ret;
1587 }
1588
1589 #ifdef __ARCH_WANT_SYS_WAITPID
1590
1591 /*
1592  * sys_waitpid() remains for compatibility. waitpid() should be
1593  * implemented by calling sys_wait4() from libc.a.
1594  */
1595 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1596 {
1597         return sys_wait4(pid, stat_addr, options, NULL);
1598 }
1599
1600 #endif