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