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/posix-timers.h>
  33 #include <linux/cn_proc.h>
  34 #include <linux/mutex.h>
  35 #include <linux/futex.h>
  36 #include <linux/compat.h>
  37
  38 #include <asm/uaccess.h>
  39 #include <asm/unistd.h>
  40 #include <asm/pgtable.h>
  41 #include <asm/mmu_context.h>
  42
  43 extern void sem_exit (void);
  44 extern struct task_struct *child_reaper;
  45
  46 int getrusage(struct task_struct *, int, struct rusage __user *);
  47
  48 static void exit_mm(struct task_struct * tsk);
  49
  50 static void __unhash_process(struct task_struct *p)
  51 {
  52         nr_threads--;
  53         detach_pid(p, PIDTYPE_PID);
  54         if (thread_group_leader(p)) {
  55                 detach_pid(p, PIDTYPE_PGID);
  56                 detach_pid(p, PIDTYPE_SID);
  57
  58                 list_del_init(&p->tasks);
  59                 __get_cpu_var(process_counts)--;
  60         }
  61         list_del_rcu(&p->thread_group);
  62         remove_parent(p);
  63 }
  64
  65 /*
  66  * This function expects the tasklist_lock write-locked.
  67  */
  68 static void __exit_signal(struct task_struct *tsk)
  69 {
  70         struct signal_struct *sig = tsk->signal;
  71         struct sighand_struct *sighand;
  72
  73         BUG_ON(!sig);
  74         BUG_ON(!atomic_read(&sig->count));
  75
  76         rcu_read_lock();
  77         sighand = rcu_dereference(tsk->sighand);
  78         spin_lock(&sighand->siglock);
  79
  80         posix_cpu_timers_exit(tsk);
  81         if (atomic_dec_and_test(&sig->count))
  82                 posix_cpu_timers_exit_group(tsk);
  83         else {
  84                 /*
  85                  * If there is any task waiting for the group exit
  86                  * then notify it:
  87                  */
  88                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
  89                         wake_up_process(sig->group_exit_task);
  90                         sig->group_exit_task = NULL;
  91                 }
  92                 if (tsk == sig->curr_target)
  93                         sig->curr_target = next_thread(tsk);
  94                 /*
  95                  * Accumulate here the counters for all threads but the
  96                  * group leader as they die, so they can be added into
  97                  * the process-wide totals when those are taken.
  98                  * The group leader stays around as a zombie as long
  99                  * as there are other threads.  When it gets reaped,
 100                  * the exit.c code will add its counts into these totals.
 101                  * We won't ever get here for the group leader, since it
 102                  * will have been the last reference on the signal_struct.
 103                  */
 104                 sig->utime = cputime_add(sig->utime, tsk->utime);
 105                 sig->stime = cputime_add(sig->stime, tsk->stime);
 106                 sig->min_flt += tsk->min_flt;
 107                 sig->maj_flt += tsk->maj_flt;
 108                 sig->nvcsw += tsk->nvcsw;
 109                 sig->nivcsw += tsk->nivcsw;
 110                 sig->sched_time += tsk->sched_time;
 111                 sig = NULL; /* Marker for below. */
 112         }
 113
 114         __unhash_process(tsk);
 115
 116         tsk->signal = NULL;
 117         tsk->sighand = NULL;
 118         spin_unlock(&sighand->siglock);
 119         rcu_read_unlock();
 120
 121         __cleanup_sighand(sighand);
 122         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 123         flush_sigqueue(&tsk->pending);
 124         if (sig) {
 125                 flush_sigqueue(&sig->shared_pending);
 126                 __cleanup_signal(sig);
 127         }
 128 }
 129
 130 static void delayed_put_task_struct(struct rcu_head *rhp)
 131 {
 132         put_task_struct(container_of(rhp, struct task_struct, rcu));
 133 }
 134
 135 void release_task(struct task_struct * p)
 136 {
 137         int zap_leader;
 138         task_t *leader;
 139         struct dentry *proc_dentry;
 140
 141 repeat:
 142         atomic_dec(&p->user->processes);
 143         spin_lock(&p->proc_lock);
 144         proc_dentry = proc_pid_unhash(p);
 145         write_lock_irq(&tasklist_lock);
 146         ptrace_unlink(p);
 147         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 148         __exit_signal(p);
 149
 150         /*
 151          * If we are the last non-leader member of the thread
 152          * group, and the leader is zombie, then notify the
 153          * group leader's parent process. (if it wants notification.)
 154          */
 155         zap_leader = 0;
 156         leader = p->group_leader;
 157         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 158                 BUG_ON(leader->exit_signal == -1);
 159                 do_notify_parent(leader, leader->exit_signal);
 160                 /*
 161                  * If we were the last child thread and the leader has
 162                  * exited already, and the leader's parent ignores SIGCHLD,
 163                  * then we are the one who should release the leader.
 164                  *
 165                  * do_notify_parent() will have marked it self-reaping in
 166                  * that case.
 167                  */
 168                 zap_leader = (leader->exit_signal == -1);
 169         }
 170
 171         sched_exit(p);
 172         write_unlock_irq(&tasklist_lock);
 173         spin_unlock(&p->proc_lock);
 174         proc_pid_flush(proc_dentry);
 175         release_thread(p);
 176         call_rcu(&p->rcu, delayed_put_task_struct);
 177
 178         p = leader;
 179         if (unlikely(zap_leader))
 180                 goto repeat;
 181 }
 182
 183 /*
 184  * This checks not only the pgrp, but falls back on the pid if no
 185  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 186  * without this...
 187  */
 188 int session_of_pgrp(int pgrp)
 189 {
 190         struct task_struct *p;
 191         int sid = -1;
 192
 193         read_lock(&tasklist_lock);
 194         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 195                 if (p->signal->session > 0) {
 196                         sid = p->signal->session;
 197                         goto out;
 198                 }
 199         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 200         p = find_task_by_pid(pgrp);
 201         if (p)
 202                 sid = p->signal->session;
 203 out:
 204         read_unlock(&tasklist_lock);
 205
 206         return sid;
 207 }
 208
 209 /*
 210  * Determine if a process group is "orphaned", according to the POSIX
 211  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 212  * by terminal-generated stop signals.  Newly orphaned process groups are
 213  * to receive a SIGHUP and a SIGCONT.
 214  *
 215  * "I ask you, have you ever known what it is to be an orphan?"
 216  */
 217 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 218 {
 219         struct task_struct *p;
 220         int ret = 1;
 221
 222         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 223                 if (p == ignored_task
 224                                 || p->exit_state
 225                                 || p->real_parent->pid == 1)
 226                         continue;
 227                 if (process_group(p->real_parent) != pgrp
 228                             && p->real_parent->signal->session == p->signal->session) {
 229                         ret = 0;
 230                         break;
 231                 }
 232         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 233         return ret;     /* (sighing) "Often!" */
 234 }
 235
 236 int is_orphaned_pgrp(int pgrp)
 237 {
 238         int retval;
 239
 240         read_lock(&tasklist_lock);
 241         retval = will_become_orphaned_pgrp(pgrp, NULL);
 242         read_unlock(&tasklist_lock);
 243
 244         return retval;
 245 }
 246
 247 static int has_stopped_jobs(int pgrp)
 248 {
 249         int retval = 0;
 250         struct task_struct *p;
 251
 252         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 253                 if (p->state != TASK_STOPPED)
 254                         continue;
 255
 256                 /* If p is stopped by a debugger on a signal that won't
 257                    stop it, then don't count p as stopped.  This isn't
 258                    perfect but it's a good approximation.  */
 259                 if (unlikely (p->ptrace)
 260                     && p->exit_code != SIGSTOP
 261                     && p->exit_code != SIGTSTP
 262                     && p->exit_code != SIGTTOU
 263                     && p->exit_code != SIGTTIN)
 264                         continue;
 265
 266                 retval = 1;
 267                 break;
 268         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 269         return retval;
 270 }
 271
 272 /**
 273  * reparent_to_init - Reparent the calling kernel thread to the init task.
 274  *
 275  * If a kernel thread is launched as a result of a system call, or if
 276  * it ever exits, it should generally reparent itself to init so that
 277  * it is correctly cleaned up on exit.
 278  *
 279  * The various task state such as scheduling policy and priority may have
 280  * been inherited from a user process, so we reset them to sane values here.
 281  *
 282  * NOTE that reparent_to_init() gives the caller full capabilities.
 283  */
 284 static void reparent_to_init(void)
 285 {
 286         write_lock_irq(&tasklist_lock);
 287
 288         ptrace_unlink(current);
 289         /* Reparent to init */
 290         remove_parent(current);
 291         current->parent = child_reaper;
 292         current->real_parent = child_reaper;
 293         add_parent(current);
 294
 295         /* Set the exit signal to SIGCHLD so we signal init on exit */
 296         current->exit_signal = SIGCHLD;
 297
 298         if ((current->policy == SCHED_NORMAL ||
 299                         current->policy == SCHED_BATCH)
 300                                 && (task_nice(current) < 0))
 301                 set_user_nice(current, 0);
 302         /* cpus_allowed? */
 303         /* rt_priority? */
 304         /* signals? */
 305         security_task_reparent_to_init(current);
 306         memcpy(current->signal->rlim, init_task.signal->rlim,
 307                sizeof(current->signal->rlim));
 308         atomic_inc(&(INIT_USER->__count));
 309         write_unlock_irq(&tasklist_lock);
 310         switch_uid(INIT_USER);
 311 }
 312
 313 void __set_special_pids(pid_t session, pid_t pgrp)
 314 {
 315         struct task_struct *curr = current->group_leader;
 316
 317         if (curr->signal->session != session) {
 318                 detach_pid(curr, PIDTYPE_SID);
 319                 curr->signal->session = session;
 320                 attach_pid(curr, PIDTYPE_SID, session);
 321         }
 322         if (process_group(curr) != pgrp) {
 323                 detach_pid(curr, PIDTYPE_PGID);
 324                 curr->signal->pgrp = pgrp;
 325                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 326         }
 327 }
 328
 329 void set_special_pids(pid_t session, pid_t pgrp)
 330 {
 331         write_lock_irq(&tasklist_lock);
 332         __set_special_pids(session, pgrp);
 333         write_unlock_irq(&tasklist_lock);
 334 }
 335
 336 /*
 337  * Let kernel threads use this to say that they
 338  * allow a certain signal (since daemonize() will
 339  * have disabled all of them by default).
 340  */
 341 int allow_signal(int sig)
 342 {
 343         if (!valid_signal(sig) || sig < 1)
 344                 return -EINVAL;
 345
 346         spin_lock_irq(&current->sighand->siglock);
 347         sigdelset(&current->blocked, sig);
 348         if (!current->mm) {
 349                 /* Kernel threads handle their own signals.
 350                    Let the signal code know it'll be handled, so
 351                    that they don't get converted to SIGKILL or
 352                    just silently dropped */
 353                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 354         }
 355         recalc_sigpending();
 356         spin_unlock_irq(&current->sighand->siglock);
 357         return 0;
 358 }
 359
 360 EXPORT_SYMBOL(allow_signal);
 361
 362 int disallow_signal(int sig)
 363 {
 364         if (!valid_signal(sig) || sig < 1)
 365                 return -EINVAL;
 366
 367         spin_lock_irq(&current->sighand->siglock);
 368         sigaddset(&current->blocked, sig);
 369         recalc_sigpending();
 370         spin_unlock_irq(&current->sighand->siglock);
 371         return 0;
 372 }
 373
 374 EXPORT_SYMBOL(disallow_signal);
 375
 376 /*
 377  *      Put all the gunge required to become a kernel thread without
 378  *      attached user resources in one place where it belongs.
 379  */
 380
 381 void daemonize(const char *name, ...)
 382 {
 383         va_list args;
 384         struct fs_struct *fs;
 385         sigset_t blocked;
 386
 387         va_start(args, name);
 388         vsnprintf(current->comm, sizeof(current->comm), name, args);
 389         va_end(args);
 390
 391         /*
 392          * If we were started as result of loading a module, close all of the
 393          * user space pages.  We don't need them, and if we didn't close them
 394          * they would be locked into memory.
 395          */
 396         exit_mm(current);
 397
 398         set_special_pids(1, 1);
 399         mutex_lock(&tty_mutex);
 400         current->signal->tty = NULL;
 401         mutex_unlock(&tty_mutex);
 402
 403         /* Block and flush all signals */
 404         sigfillset(&blocked);
 405         sigprocmask(SIG_BLOCK, &blocked, NULL);
 406         flush_signals(current);
 407
 408         /* Become as one with the init task */
 409
 410         exit_fs(current);       /* current->fs->count--; */
 411         fs = init_task.fs;
 412         current->fs = fs;
 413         atomic_inc(&fs->count);
 414         exit_namespace(current);
 415         current->namespace = init_task.namespace;
 416         get_namespace(current->namespace);
 417         exit_files(current);
 418         current->files = init_task.files;
 419         atomic_inc(&current->files->count);
 420
 421         reparent_to_init();
 422 }
 423
 424 EXPORT_SYMBOL(daemonize);
 425
 426 static void close_files(struct files_struct * files)
 427 {
 428         int i, j;
 429         struct fdtable *fdt;
 430
 431         j = 0;
 432
 433         /*
 434          * It is safe to dereference the fd table without RCU or
 435          * ->file_lock because this is the last reference to the
 436          * files structure.
 437          */
 438         fdt = files_fdtable(files);
 439         for (;;) {
 440                 unsigned long set;
 441                 i = j * __NFDBITS;
 442                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 443                         break;
 444                 set = fdt->open_fds->fds_bits[j++];
 445                 while (set) {
 446                         if (set & 1) {
 447                                 struct file * file = xchg(&fdt->fd[i], NULL);
 448                                 if (file)
 449                                         filp_close(file, files);
 450                         }
 451                         i++;
 452                         set >>= 1;
 453                 }
 454         }
 455 }
 456
 457 struct files_struct *get_files_struct(struct task_struct *task)
 458 {
 459         struct files_struct *files;
 460
 461         task_lock(task);
 462         files = task->files;
 463         if (files)
 464                 atomic_inc(&files->count);
 465         task_unlock(task);
 466
 467         return files;
 468 }
 469
 470 void fastcall put_files_struct(struct files_struct *files)
 471 {
 472         struct fdtable *fdt;
 473
 474         if (atomic_dec_and_test(&files->count)) {
 475                 close_files(files);
 476                 /*
 477                  * Free the fd and fdset arrays if we expanded them.
 478                  * If the fdtable was embedded, pass files for freeing
 479                  * at the end of the RCU grace period. Otherwise,
 480                  * you can free files immediately.
 481                  */
 482                 fdt = files_fdtable(files);
 483                 if (fdt == &files->fdtab)
 484                         fdt->free_files = files;
 485                 else
 486                         kmem_cache_free(files_cachep, files);
 487                 free_fdtable(fdt);
 488         }
 489 }
 490
 491 EXPORT_SYMBOL(put_files_struct);
 492
 493 static inline void __exit_files(struct task_struct *tsk)
 494 {
 495         struct files_struct * files = tsk->files;
 496
 497         if (files) {
 498                 task_lock(tsk);
 499                 tsk->files = NULL;
 500                 task_unlock(tsk);
 501                 put_files_struct(files);
 502         }
 503 }
 504
 505 void exit_files(struct task_struct *tsk)
 506 {
 507         __exit_files(tsk);
 508 }
 509
 510 static inline void __put_fs_struct(struct fs_struct *fs)
 511 {
 512         /* No need to hold fs->lock if we are killing it */
 513         if (atomic_dec_and_test(&fs->count)) {
 514                 dput(fs->root);
 515                 mntput(fs->rootmnt);
 516                 dput(fs->pwd);
 517                 mntput(fs->pwdmnt);
 518                 if (fs->altroot) {
 519                         dput(fs->altroot);
 520                         mntput(fs->altrootmnt);
 521                 }
 522                 kmem_cache_free(fs_cachep, fs);
 523         }
 524 }
 525
 526 void put_fs_struct(struct fs_struct *fs)
 527 {
 528         __put_fs_struct(fs);
 529 }
 530
 531 static inline void __exit_fs(struct task_struct *tsk)
 532 {
 533         struct fs_struct * fs = tsk->fs;
 534
 535         if (fs) {
 536                 task_lock(tsk);
 537                 tsk->fs = NULL;
 538                 task_unlock(tsk);
 539                 __put_fs_struct(fs);
 540         }
 541 }
 542
 543 void exit_fs(struct task_struct *tsk)
 544 {
 545         __exit_fs(tsk);
 546 }
 547
 548 EXPORT_SYMBOL_GPL(exit_fs);
 549
 550 /*
 551  * Turn us into a lazy TLB process if we
 552  * aren't already..
 553  */
 554 static void exit_mm(struct task_struct * tsk)
 555 {
 556         struct mm_struct *mm = tsk->mm;
 557
 558         mm_release(tsk, mm);
 559         if (!mm)
 560                 return;
 561         /*
 562          * Serialize with any possible pending coredump.
 563          * We must hold mmap_sem around checking core_waiters
 564          * and clearing tsk->mm.  The core-inducing thread
 565          * will increment core_waiters for each thread in the
 566          * group with ->mm != NULL.
 567          */
 568         down_read(&mm->mmap_sem);
 569         if (mm->core_waiters) {
 570                 up_read(&mm->mmap_sem);
 571                 down_write(&mm->mmap_sem);
 572                 if (!--mm->core_waiters)
 573                         complete(mm->core_startup_done);
 574                 up_write(&mm->mmap_sem);
 575
 576                 wait_for_completion(&mm->core_done);
 577                 down_read(&mm->mmap_sem);
 578         }
 579         atomic_inc(&mm->mm_count);
 580         if (mm != tsk->active_mm) BUG();
 581         /* more a memory barrier than a real lock */
 582         task_lock(tsk);
 583         tsk->mm = NULL;
 584         up_read(&mm->mmap_sem);
 585         enter_lazy_tlb(mm, current);
 586         task_unlock(tsk);
 587         mmput(mm);
 588 }
 589
 590 static inline void choose_new_parent(task_t *p, task_t *reaper)
 591 {
 592         /*
 593          * Make sure we're not reparenting to ourselves and that
 594          * the parent is not a zombie.
 595          */
 596         BUG_ON(p == reaper || reaper->exit_state);
 597         p->real_parent = reaper;
 598 }
 599
 600 static void reparent_thread(task_t *p, task_t *father, int traced)
 601 {
 602         /* We don't want people slaying init.  */
 603         if (p->exit_signal != -1)
 604                 p->exit_signal = SIGCHLD;
 605
 606         if (p->pdeath_signal)
 607                 /* We already hold the tasklist_lock here.  */
 608                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 609
 610         /* Move the child from its dying parent to the new one.  */
 611         if (unlikely(traced)) {
 612                 /* Preserve ptrace links if someone else is tracing this child.  */
 613                 list_del_init(&p->ptrace_list);
 614                 if (p->parent != p->real_parent)
 615                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 616         } else {
 617                 /* If this child is being traced, then we're the one tracing it
 618                  * anyway, so let go of it.
 619                  */
 620                 p->ptrace = 0;
 621                 remove_parent(p);
 622                 p->parent = p->real_parent;
 623                 add_parent(p);
 624
 625                 /* If we'd notified the old parent about this child's death,
 626                  * also notify the new parent.
 627                  */
 628                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 629                     thread_group_empty(p))
 630                         do_notify_parent(p, p->exit_signal);
 631                 else if (p->state == TASK_TRACED) {
 632                         /*
 633                          * If it was at a trace stop, turn it into
 634                          * a normal stop since it's no longer being
 635                          * traced.
 636                          */
 637                         ptrace_untrace(p);
 638                 }
 639         }
 640
 641         /*
 642          * process group orphan check
 643          * Case ii: Our child is in a different pgrp
 644          * than we are, and it was the only connection
 645          * outside, so the child pgrp is now orphaned.
 646          */
 647         if ((process_group(p) != process_group(father)) &&
 648             (p->signal->session == father->signal->session)) {
 649                 int pgrp = process_group(p);
 650
 651                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 652                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 653                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 654                 }
 655         }
 656 }
 657
 658 /*
 659  * When we die, we re-parent all our children.
 660  * Try to give them to another thread in our thread
 661  * group, and if no such member exists, give it to
 662  * the global child reaper process (ie "init")
 663  */
 664 static void forget_original_parent(struct task_struct * father,
 665                                           struct list_head *to_release)
 666 {
 667         struct task_struct *p, *reaper = father;
 668         struct list_head *_p, *_n;
 669
 670         do {
 671                 reaper = next_thread(reaper);
 672                 if (reaper == father) {
 673                         reaper = child_reaper;
 674                         break;
 675                 }
 676         } while (reaper->exit_state);
 677
 678         /*
 679          * There are only two places where our children can be:
 680          *
 681          * - in our child list
 682          * - in our ptraced child list
 683          *
 684          * Search them and reparent children.
 685          */
 686         list_for_each_safe(_p, _n, &father->children) {
 687                 int ptrace;
 688                 p = list_entry(_p,struct task_struct,sibling);
 689
 690                 ptrace = p->ptrace;
 691
 692                 /* if father isn't the real parent, then ptrace must be enabled */
 693                 BUG_ON(father != p->real_parent && !ptrace);
 694
 695                 if (father == p->real_parent) {
 696                         /* reparent with a reaper, real father it's us */
 697                         choose_new_parent(p, reaper);
 698                         reparent_thread(p, father, 0);
 699                 } else {
 700                         /* reparent ptraced task to its real parent */
 701                         __ptrace_unlink (p);
 702                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 703                             thread_group_empty(p))
 704                                 do_notify_parent(p, p->exit_signal);
 705                 }
 706
 707                 /*
 708                  * if the ptraced child is a zombie with exit_signal == -1
 709                  * we must collect it before we exit, or it will remain
 710                  * zombie forever since we prevented it from self-reap itself
 711                  * while it was being traced by us, to be able to see it in wait4.
 712                  */
 713                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 714                         list_add(&p->ptrace_list, to_release);
 715         }
 716         list_for_each_safe(_p, _n, &father->ptrace_children) {
 717                 p = list_entry(_p,struct task_struct,ptrace_list);
 718                 choose_new_parent(p, reaper);
 719                 reparent_thread(p, father, 1);
 720         }
 721 }
 722
 723 /*
 724  * Send signals to all our closest relatives so that they know
 725  * to properly mourn us..
 726  */
 727 static void exit_notify(struct task_struct *tsk)
 728 {
 729         int state;
 730         struct task_struct *t;
 731         struct list_head ptrace_dead, *_p, *_n;
 732
 733         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 734             && !thread_group_empty(tsk)) {
 735                 /*
 736                  * This occurs when there was a race between our exit
 737                  * syscall and a group signal choosing us as the one to
 738                  * wake up.  It could be that we are the only thread
 739                  * alerted to check for pending signals, but another thread
 740                  * should be woken now to take the signal since we will not.
 741                  * Now we'll wake all the threads in the group just to make
 742                  * sure someone gets all the pending signals.
 743                  */
 744                 read_lock(&tasklist_lock);
 745                 spin_lock_irq(&tsk->sighand->siglock);
 746                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 747                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 748                                 recalc_sigpending_tsk(t);
 749                                 if (signal_pending(t))
 750                                         signal_wake_up(t, 0);
 751                         }
 752                 spin_unlock_irq(&tsk->sighand->siglock);
 753                 read_unlock(&tasklist_lock);
 754         }
 755
 756         write_lock_irq(&tasklist_lock);
 757
 758         /*
 759          * This does two things:
 760          *
 761          * A.  Make init inherit all the child processes
 762          * B.  Check to see if any process groups have become orphaned
 763          *      as a result of our exiting, and if they have any stopped
 764          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 765          */
 766
 767         INIT_LIST_HEAD(&ptrace_dead);
 768         forget_original_parent(tsk, &ptrace_dead);
 769         BUG_ON(!list_empty(&tsk->children));
 770         BUG_ON(!list_empty(&tsk->ptrace_children));
 771
 772         /*
 773          * Check to see if any process groups have become orphaned
 774          * as a result of our exiting, and if they have any stopped
 775          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 776          *
 777          * Case i: Our father is in a different pgrp than we are
 778          * and we were the only connection outside, so our pgrp
 779          * is about to become orphaned.
 780          */
 781
 782         t = tsk->real_parent;
 783
 784         if ((process_group(t) != process_group(tsk)) &&
 785             (t->signal->session == tsk->signal->session) &&
 786             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 787             has_stopped_jobs(process_group(tsk))) {
 788                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 789                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 790         }
 791
 792         /* Let father know we died
 793          *
 794          * Thread signals are configurable, but you aren't going to use
 795          * that to send signals to arbitary processes.
 796          * That stops right now.
 797          *
 798          * If the parent exec id doesn't match the exec id we saved
 799          * when we started then we know the parent has changed security
 800          * domain.
 801          *
 802          * If our self_exec id doesn't match our parent_exec_id then
 803          * we have changed execution domain as these two values started
 804          * the same after a fork.
 805          *
 806          */
 807
 808         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 809             ( tsk->parent_exec_id != t->self_exec_id  ||
 810               tsk->self_exec_id != tsk->parent_exec_id)
 811             && !capable(CAP_KILL))
 812                 tsk->exit_signal = SIGCHLD;
 813
 814
 815         /* If something other than our normal parent is ptracing us, then
 816          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 817          * only has special meaning to our real parent.
 818          */
 819         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 820                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 821                 do_notify_parent(tsk, signal);
 822         } else if (tsk->ptrace) {
 823                 do_notify_parent(tsk, SIGCHLD);
 824         }
 825
 826         state = EXIT_ZOMBIE;
 827         if (tsk->exit_signal == -1 &&
 828             (likely(tsk->ptrace == 0) ||
 829              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 830                 state = EXIT_DEAD;
 831         tsk->exit_state = state;
 832
 833         write_unlock_irq(&tasklist_lock);
 834
 835         list_for_each_safe(_p, _n, &ptrace_dead) {
 836                 list_del_init(_p);
 837                 t = list_entry(_p,struct task_struct,ptrace_list);
 838                 release_task(t);
 839         }
 840
 841         /* If the process is dead, release it - nobody will wait for it */
 842         if (state == EXIT_DEAD)
 843                 release_task(tsk);
 844 }
 845
 846 fastcall NORET_TYPE void do_exit(long code)
 847 {
 848         struct task_struct *tsk = current;
 849         int group_dead;
 850
 851         profile_task_exit(tsk);
 852
 853         WARN_ON(atomic_read(&tsk->fs_excl));
 854
 855         if (unlikely(in_interrupt()))
 856                 panic("Aiee, killing interrupt handler!");
 857         if (unlikely(!tsk->pid))
 858                 panic("Attempted to kill the idle task!");
 859         if (unlikely(tsk == child_reaper))
 860                 panic("Attempted to kill init!");
 861
 862         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 863                 current->ptrace_message = code;
 864                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 865         }
 866
 867         /*
 868          * We're taking recursive faults here in do_exit. Safest is to just
 869          * leave this task alone and wait for reboot.
 870          */
 871         if (unlikely(tsk->flags & PF_EXITING)) {
 872                 printk(KERN_ALERT
 873                         "Fixing recursive fault but reboot is needed!\n");
 874                 if (tsk->io_context)
 875                         exit_io_context();
 876                 set_current_state(TASK_UNINTERRUPTIBLE);
 877                 schedule();
 878         }
 879
 880         tsk->flags |= PF_EXITING;
 881
 882         /*
 883          * Make sure we don't try to process any timer firings
 884          * while we are already exiting.
 885          */
 886         tsk->it_virt_expires = cputime_zero;
 887         tsk->it_prof_expires = cputime_zero;
 888         tsk->it_sched_expires = 0;
 889
 890         if (unlikely(in_atomic()))
 891                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 892                                 current->comm, current->pid,
 893                                 preempt_count());
 894
 895         acct_update_integrals(tsk);
 896         if (tsk->mm) {
 897                 update_hiwater_rss(tsk->mm);
 898                 update_hiwater_vm(tsk->mm);
 899         }
 900         group_dead = atomic_dec_and_test(&tsk->signal->live);
 901         if (group_dead) {
 902                 hrtimer_cancel(&tsk->signal->real_timer);
 903                 exit_itimers(tsk->signal);
 904                 acct_process(code);
 905         }
 906         if (unlikely(tsk->robust_list))
 907                 exit_robust_list(tsk);
 908 #ifdef CONFIG_COMPAT
 909         if (unlikely(tsk->compat_robust_list))
 910                 compat_exit_robust_list(tsk);
 911 #endif
 912         exit_mm(tsk);
 913
 914         exit_sem(tsk);
 915         __exit_files(tsk);
 916         __exit_fs(tsk);
 917         exit_namespace(tsk);
 918         exit_thread();
 919         cpuset_exit(tsk);
 920         exit_keys(tsk);
 921
 922         if (group_dead && tsk->signal->leader)
 923                 disassociate_ctty(1);
 924
 925         module_put(task_thread_info(tsk)->exec_domain->module);
 926         if (tsk->binfmt)
 927                 module_put(tsk->binfmt->module);
 928
 929         tsk->exit_code = code;
 930         proc_exit_connector(tsk);
 931         exit_notify(tsk);
 932 #ifdef CONFIG_NUMA
 933         mpol_free(tsk->mempolicy);
 934         tsk->mempolicy = NULL;
 935 #endif
 936         /*
 937          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
 938          */
 939         mutex_debug_check_no_locks_held(tsk);
 940
 941         if (tsk->io_context)
 942                 exit_io_context();
 943
 944         /* PF_DEAD causes final put_task_struct after we schedule. */
 945         preempt_disable();
 946         BUG_ON(tsk->flags & PF_DEAD);
 947         tsk->flags |= PF_DEAD;
 948
 949         schedule();
 950         BUG();
 951         /* Avoid "noreturn function does return".  */
 952         for (;;) ;
 953 }
 954
 955 EXPORT_SYMBOL_GPL(do_exit);
 956
 957 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 958 {
 959         if (comp)
 960                 complete(comp);
 961
 962         do_exit(code);
 963 }
 964
 965 EXPORT_SYMBOL(complete_and_exit);
 966
 967 asmlinkage long sys_exit(int error_code)
 968 {
 969         do_exit((error_code&0xff)<<8);
 970 }
 971
 972 /*
 973  * Take down every thread in the group.  This is called by fatal signals
 974  * as well as by sys_exit_group (below).
 975  */
 976 NORET_TYPE void
 977 do_group_exit(int exit_code)
 978 {
 979         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 980
 981         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 982                 exit_code = current->signal->group_exit_code;
 983         else if (!thread_group_empty(current)) {
 984                 struct signal_struct *const sig = current->signal;
 985                 struct sighand_struct *const sighand = current->sighand;
 986                 spin_lock_irq(&sighand->siglock);
 987                 if (sig->flags & SIGNAL_GROUP_EXIT)
 988                         /* Another thread got here before we took the lock.  */
 989                         exit_code = sig->group_exit_code;
 990                 else {
 991                         sig->group_exit_code = exit_code;
 992                         zap_other_threads(current);
 993                 }
 994                 spin_unlock_irq(&sighand->siglock);
 995         }
 996
 997         do_exit(exit_code);
 998         /* NOTREACHED */
 999 }
1000
1001 /*
1002  * this kills every thread in the thread group. Note that any externally
1003  * wait4()-ing process will get the correct exit code - even if this
1004  * thread is not the thread group leader.
1005  */
1006 asmlinkage void sys_exit_group(int error_code)
1007 {
1008         do_group_exit((error_code & 0xff) << 8);
1009 }
1010
1011 static int eligible_child(pid_t pid, int options, task_t *p)
1012 {
1013         if (pid > 0) {
1014                 if (p->pid != pid)
1015                         return 0;
1016         } else if (!pid) {
1017                 if (process_group(p) != process_group(current))
1018                         return 0;
1019         } else if (pid != -1) {
1020                 if (process_group(p) != -pid)
1021                         return 0;
1022         }
1023
1024         /*
1025          * Do not consider detached threads that are
1026          * not ptraced:
1027          */
1028         if (p->exit_signal == -1 && !p->ptrace)
1029                 return 0;
1030
1031         /* Wait for all children (clone and not) if __WALL is set;
1032          * otherwise, wait for clone children *only* if __WCLONE is
1033          * set; otherwise, wait for non-clone children *only*.  (Note:
1034          * A "clone" child here is one that reports to its parent
1035          * using a signal other than SIGCHLD.) */
1036         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1037             && !(options & __WALL))
1038                 return 0;
1039         /*
1040          * Do not consider thread group leaders that are
1041          * in a non-empty thread group:
1042          */
1043         if (current->tgid != p->tgid && delay_group_leader(p))
1044                 return 2;
1045
1046         if (security_task_wait(p))
1047                 return 0;
1048
1049         return 1;
1050 }
1051
1052 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
1053                                int why, int status,
1054                                struct siginfo __user *infop,
1055                                struct rusage __user *rusagep)
1056 {
1057         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1058         put_task_struct(p);
1059         if (!retval)
1060                 retval = put_user(SIGCHLD, &infop->si_signo);
1061         if (!retval)
1062                 retval = put_user(0, &infop->si_errno);
1063         if (!retval)
1064                 retval = put_user((short)why, &infop->si_code);
1065         if (!retval)
1066                 retval = put_user(pid, &infop->si_pid);
1067         if (!retval)
1068                 retval = put_user(uid, &infop->si_uid);
1069         if (!retval)
1070                 retval = put_user(status, &infop->si_status);
1071         if (!retval)
1072                 retval = pid;
1073         return retval;
1074 }
1075
1076 /*
1077  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1078  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1079  * the lock and this task is uninteresting.  If we return nonzero, we have
1080  * released the lock and the system call should return.
1081  */
1082 static int wait_task_zombie(task_t *p, int noreap,
1083                             struct siginfo __user *infop,
1084                             int __user *stat_addr, struct rusage __user *ru)
1085 {
1086         unsigned long state;
1087         int retval;
1088         int status;
1089
1090         if (unlikely(noreap)) {
1091                 pid_t pid = p->pid;
1092                 uid_t uid = p->uid;
1093                 int exit_code = p->exit_code;
1094                 int why, status;
1095
1096                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1097                         return 0;
1098                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1099                         return 0;
1100                 get_task_struct(p);
1101                 read_unlock(&tasklist_lock);
1102                 if ((exit_code & 0x7f) == 0) {
1103                         why = CLD_EXITED;
1104                         status = exit_code >> 8;
1105                 } else {
1106                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1107                         status = exit_code & 0x7f;
1108                 }
1109                 return wait_noreap_copyout(p, pid, uid, why,
1110                                            status, infop, ru);
1111         }
1112
1113         /*
1114          * Try to move the task's state to DEAD
1115          * only one thread is allowed to do this:
1116          */
1117         state = xchg(&p->exit_state, EXIT_DEAD);
1118         if (state != EXIT_ZOMBIE) {
1119                 BUG_ON(state != EXIT_DEAD);
1120                 return 0;
1121         }
1122         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1123                 /*
1124                  * This can only happen in a race with a ptraced thread
1125                  * dying on another processor.
1126                  */
1127                 return 0;
1128         }
1129
1130         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1131                 struct signal_struct *psig;
1132                 struct signal_struct *sig;
1133
1134                 /*
1135                  * The resource counters for the group leader are in its
1136                  * own task_struct.  Those for dead threads in the group
1137                  * are in its signal_struct, as are those for the child
1138                  * processes it has previously reaped.  All these
1139                  * accumulate in the parent's signal_struct c* fields.
1140                  *
1141                  * We don't bother to take a lock here to protect these
1142                  * p->signal fields, because they are only touched by
1143                  * __exit_signal, which runs with tasklist_lock
1144                  * write-locked anyway, and so is excluded here.  We do
1145                  * need to protect the access to p->parent->signal fields,
1146                  * as other threads in the parent group can be right
1147                  * here reaping other children at the same time.
1148                  */
1149                 spin_lock_irq(&p->parent->sighand->siglock);
1150                 psig = p->parent->signal;
1151                 sig = p->signal;
1152                 psig->cutime =
1153                         cputime_add(psig->cutime,
1154                         cputime_add(p->utime,
1155                         cputime_add(sig->utime,
1156                                     sig->cutime)));
1157                 psig->cstime =
1158                         cputime_add(psig->cstime,
1159                         cputime_add(p->stime,
1160                         cputime_add(sig->stime,
1161                                     sig->cstime)));
1162                 psig->cmin_flt +=
1163                         p->min_flt + sig->min_flt + sig->cmin_flt;
1164                 psig->cmaj_flt +=
1165                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1166                 psig->cnvcsw +=
1167                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1168                 psig->cnivcsw +=
1169                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1170                 spin_unlock_irq(&p->parent->sighand->siglock);
1171         }
1172
1173         /*
1174          * Now we are sure this task is interesting, and no other
1175          * thread can reap it because we set its state to EXIT_DEAD.
1176          */
1177         read_unlock(&tasklist_lock);
1178
1179         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1180         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1181                 ? p->signal->group_exit_code : p->exit_code;
1182         if (!retval && stat_addr)
1183                 retval = put_user(status, stat_addr);
1184         if (!retval && infop)
1185                 retval = put_user(SIGCHLD, &infop->si_signo);
1186         if (!retval && infop)
1187                 retval = put_user(0, &infop->si_errno);
1188         if (!retval && infop) {
1189                 int why;
1190
1191                 if ((status & 0x7f) == 0) {
1192                         why = CLD_EXITED;
1193                         status >>= 8;
1194                 } else {
1195                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1196                         status &= 0x7f;
1197                 }
1198                 retval = put_user((short)why, &infop->si_code);
1199                 if (!retval)
1200                         retval = put_user(status, &infop->si_status);
1201         }
1202         if (!retval && infop)
1203                 retval = put_user(p->pid, &infop->si_pid);
1204         if (!retval && infop)
1205                 retval = put_user(p->uid, &infop->si_uid);
1206         if (retval) {
1207                 // TODO: is this safe?
1208                 p->exit_state = EXIT_ZOMBIE;
1209                 return retval;
1210         }
1211         retval = p->pid;
1212         if (p->real_parent != p->parent) {
1213                 write_lock_irq(&tasklist_lock);
1214                 /* Double-check with lock held.  */
1215                 if (p->real_parent != p->parent) {
1216                         __ptrace_unlink(p);
1217                         // TODO: is this safe?
1218                         p->exit_state = EXIT_ZOMBIE;
1219                         /*
1220                          * If this is not a detached task, notify the parent.
1221                          * If it's still not detached after that, don't release
1222                          * it now.
1223                          */
1224                         if (p->exit_signal != -1) {
1225                                 do_notify_parent(p, p->exit_signal);
1226                                 if (p->exit_signal != -1)
1227                                         p = NULL;
1228                         }
1229                 }
1230                 write_unlock_irq(&tasklist_lock);
1231         }
1232         if (p != NULL)
1233                 release_task(p);
1234         BUG_ON(!retval);
1235         return retval;
1236 }
1237
1238 /*
1239  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1240  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1241  * the lock and this task is uninteresting.  If we return nonzero, we have
1242  * released the lock and the system call should return.
1243  */
1244 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1245                              struct siginfo __user *infop,
1246                              int __user *stat_addr, struct rusage __user *ru)
1247 {
1248         int retval, exit_code;
1249
1250         if (!p->exit_code)
1251                 return 0;
1252         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1253             p->signal && p->signal->group_stop_count > 0)
1254                 /*
1255                  * A group stop is in progress and this is the group leader.
1256                  * We won't report until all threads have stopped.
1257                  */
1258                 return 0;
1259
1260         /*
1261          * Now we are pretty sure this task is interesting.
1262          * Make sure it doesn't get reaped out from under us while we
1263          * give up the lock and then examine it below.  We don't want to
1264          * keep holding onto the tasklist_lock while we call getrusage and
1265          * possibly take page faults for user memory.
1266          */
1267         get_task_struct(p);
1268         read_unlock(&tasklist_lock);
1269
1270         if (unlikely(noreap)) {
1271                 pid_t pid = p->pid;
1272                 uid_t uid = p->uid;
1273                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1274
1275                 exit_code = p->exit_code;
1276                 if (unlikely(!exit_code) ||
1277                     unlikely(p->state & TASK_TRACED))
1278                         goto bail_ref;
1279                 return wait_noreap_copyout(p, pid, uid,
1280                                            why, (exit_code << 8) | 0x7f,
1281                                            infop, ru);
1282         }
1283
1284         write_lock_irq(&tasklist_lock);
1285
1286         /*
1287          * This uses xchg to be atomic with the thread resuming and setting
1288          * it.  It must also be done with the write lock held to prevent a
1289          * race with the EXIT_ZOMBIE case.
1290          */
1291         exit_code = xchg(&p->exit_code, 0);
1292         if (unlikely(p->exit_state)) {
1293                 /*
1294                  * The task resumed and then died.  Let the next iteration
1295                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1296                  * already be zero here if it resumed and did _exit(0).
1297                  * The task itself is dead and won't touch exit_code again;
1298                  * other processors in this function are locked out.
1299                  */
1300                 p->exit_code = exit_code;
1301                 exit_code = 0;
1302         }
1303         if (unlikely(exit_code == 0)) {
1304                 /*
1305                  * Another thread in this function got to it first, or it
1306                  * resumed, or it resumed and then died.
1307                  */
1308                 write_unlock_irq(&tasklist_lock);
1309 bail_ref:
1310                 put_task_struct(p);
1311                 /*
1312                  * We are returning to the wait loop without having successfully
1313                  * removed the process and having released the lock. We cannot
1314                  * continue, since the "p" task pointer is potentially stale.
1315                  *
1316                  * Return -EAGAIN, and do_wait() will restart the loop from the
1317                  * beginning. Do _not_ re-acquire the lock.
1318                  */
1319                 return -EAGAIN;
1320         }
1321
1322         /* move to end of parent's list to avoid starvation */
1323         remove_parent(p);
1324         add_parent(p);
1325
1326         write_unlock_irq(&tasklist_lock);
1327
1328         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1329         if (!retval && stat_addr)
1330                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1331         if (!retval && infop)
1332                 retval = put_user(SIGCHLD, &infop->si_signo);
1333         if (!retval && infop)
1334                 retval = put_user(0, &infop->si_errno);
1335         if (!retval && infop)
1336                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1337                                           ? CLD_TRAPPED : CLD_STOPPED),
1338                                   &infop->si_code);
1339         if (!retval && infop)
1340                 retval = put_user(exit_code, &infop->si_status);
1341         if (!retval && infop)
1342                 retval = put_user(p->pid, &infop->si_pid);
1343         if (!retval && infop)
1344                 retval = put_user(p->uid, &infop->si_uid);
1345         if (!retval)
1346                 retval = p->pid;
1347         put_task_struct(p);
1348
1349         BUG_ON(!retval);
1350         return retval;
1351 }
1352
1353 /*
1354  * Handle do_wait work for one task in a live, non-stopped state.
1355  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1356  * the lock and this task is uninteresting.  If we return nonzero, we have
1357  * released the lock and the system call should return.
1358  */
1359 static int wait_task_continued(task_t *p, int noreap,
1360                                struct siginfo __user *infop,
1361                                int __user *stat_addr, struct rusage __user *ru)
1362 {
1363         int retval;
1364         pid_t pid;
1365         uid_t uid;
1366
1367         if (unlikely(!p->signal))
1368                 return 0;
1369
1370         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1371                 return 0;
1372
1373         spin_lock_irq(&p->sighand->siglock);
1374         /* Re-check with the lock held.  */
1375         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1376                 spin_unlock_irq(&p->sighand->siglock);
1377                 return 0;
1378         }
1379         if (!noreap)
1380                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1381         spin_unlock_irq(&p->sighand->siglock);
1382
1383         pid = p->pid;
1384         uid = p->uid;
1385         get_task_struct(p);
1386         read_unlock(&tasklist_lock);
1387
1388         if (!infop) {
1389                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1390                 put_task_struct(p);
1391                 if (!retval && stat_addr)
1392                         retval = put_user(0xffff, stat_addr);
1393                 if (!retval)
1394                         retval = p->pid;
1395         } else {
1396                 retval = wait_noreap_copyout(p, pid, uid,
1397                                              CLD_CONTINUED, SIGCONT,
1398                                              infop, ru);
1399                 BUG_ON(retval == 0);
1400         }
1401
1402         return retval;
1403 }
1404
1405
1406 static inline int my_ptrace_child(struct task_struct *p)
1407 {
1408         if (!(p->ptrace & PT_PTRACED))
1409                 return 0;
1410         if (!(p->ptrace & PT_ATTACHED))
1411                 return 1;
1412         /*
1413          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1414          * we are the attacher.  If we are the real parent, this is a race
1415          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1416          * which we have to switch the parent links, but has already set
1417          * the flags in p->ptrace.
1418          */
1419         return (p->parent != p->real_parent);
1420 }
1421
1422 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1423                     int __user *stat_addr, struct rusage __user *ru)
1424 {
1425         DECLARE_WAITQUEUE(wait, current);
1426         struct task_struct *tsk;
1427         int flag, retval;
1428
1429         add_wait_queue(&current->signal->wait_chldexit,&wait);
1430 repeat:
1431         /*
1432          * We will set this flag if we see any child that might later
1433          * match our criteria, even if we are not able to reap it yet.
1434          */
1435         flag = 0;
1436         current->state = TASK_INTERRUPTIBLE;
1437         read_lock(&tasklist_lock);
1438         tsk = current;
1439         do {
1440                 struct task_struct *p;
1441                 struct list_head *_p;
1442                 int ret;
1443
1444                 list_for_each(_p,&tsk->children) {
1445                         p = list_entry(_p,struct task_struct,sibling);
1446
1447                         ret = eligible_child(pid, options, p);
1448                         if (!ret)
1449                                 continue;
1450
1451                         switch (p->state) {
1452                         case TASK_TRACED:
1453                                 /*
1454                                  * When we hit the race with PTRACE_ATTACH,
1455                                  * we will not report this child.  But the
1456                                  * race means it has not yet been moved to
1457                                  * our ptrace_children list, so we need to
1458                                  * set the flag here to avoid a spurious ECHILD
1459                                  * when the race happens with the only child.
1460                                  */
1461                                 flag = 1;
1462                                 if (!my_ptrace_child(p))
1463                                         continue;
1464                                 /*FALLTHROUGH*/
1465                         case TASK_STOPPED:
1466                                 /*
1467                                  * It's stopped now, so it might later
1468                                  * continue, exit, or stop again.
1469                                  */
1470                                 flag = 1;
1471                                 if (!(options & WUNTRACED) &&
1472                                     !my_ptrace_child(p))
1473                                         continue;
1474                                 retval = wait_task_stopped(p, ret == 2,
1475                                                            (options & WNOWAIT),
1476                                                            infop,
1477                                                            stat_addr, ru);
1478                                 if (retval == -EAGAIN)
1479                                         goto repeat;
1480                                 if (retval != 0) /* He released the lock.  */
1481                                         goto end;
1482                                 break;
1483                         default:
1484                         // case EXIT_DEAD:
1485                                 if (p->exit_state == EXIT_DEAD)
1486                                         continue;
1487                         // case EXIT_ZOMBIE:
1488                                 if (p->exit_state == EXIT_ZOMBIE) {
1489                                         /*
1490                                          * Eligible but we cannot release
1491                                          * it yet:
1492                                          */
1493                                         if (ret == 2)
1494                                                 goto check_continued;
1495                                         if (!likely(options & WEXITED))
1496                                                 continue;
1497                                         retval = wait_task_zombie(
1498                                                 p, (options & WNOWAIT),
1499                                                 infop, stat_addr, ru);
1500                                         /* He released the lock.  */
1501                                         if (retval != 0)
1502                                                 goto end;
1503                                         break;
1504                                 }
1505 check_continued:
1506                                 /*
1507                                  * It's running now, so it might later
1508                                  * exit, stop, or stop and then continue.
1509                                  */
1510                                 flag = 1;
1511                                 if (!unlikely(options & WCONTINUED))
1512                                         continue;
1513                                 retval = wait_task_continued(
1514                                         p, (options & WNOWAIT),
1515                                         infop, stat_addr, ru);
1516                                 if (retval != 0) /* He released the lock.  */
1517                                         goto end;
1518                                 break;
1519                         }
1520                 }
1521                 if (!flag) {
1522                         list_for_each(_p, &tsk->ptrace_children) {
1523                                 p = list_entry(_p, struct task_struct,
1524                                                 ptrace_list);
1525                                 if (!eligible_child(pid, options, p))
1526                                         continue;
1527                                 flag = 1;
1528                                 break;
1529                         }
1530                 }
1531                 if (options & __WNOTHREAD)
1532                         break;
1533                 tsk = next_thread(tsk);
1534                 if (tsk->signal != current->signal)
1535                         BUG();
1536         } while (tsk != current);
1537
1538         read_unlock(&tasklist_lock);
1539         if (flag) {
1540                 retval = 0;
1541                 if (options & WNOHANG)
1542                         goto end;
1543                 retval = -ERESTARTSYS;
1544                 if (signal_pending(current))
1545                         goto end;
1546                 schedule();
1547                 goto repeat;
1548         }
1549         retval = -ECHILD;
1550 end:
1551         current->state = TASK_RUNNING;
1552         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1553         if (infop) {
1554                 if (retval > 0)
1555                 retval = 0;
1556                 else {
1557                         /*
1558                          * For a WNOHANG return, clear out all the fields
1559                          * we would set so the user can easily tell the
1560                          * difference.
1561                          */
1562                         if (!retval)
1563                                 retval = put_user(0, &infop->si_signo);
1564                         if (!retval)
1565                                 retval = put_user(0, &infop->si_errno);
1566                         if (!retval)
1567                                 retval = put_user(0, &infop->si_code);
1568                         if (!retval)
1569                                 retval = put_user(0, &infop->si_pid);
1570                         if (!retval)
1571                                 retval = put_user(0, &infop->si_uid);
1572                         if (!retval)
1573                                 retval = put_user(0, &infop->si_status);
1574                 }
1575         }
1576         return retval;
1577 }
1578
1579 asmlinkage long sys_waitid(int which, pid_t pid,
1580                            struct siginfo __user *infop, int options,
1581                            struct rusage __user *ru)
1582 {
1583         long ret;
1584
1585         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1586                 return -EINVAL;
1587         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1588                 return -EINVAL;
1589
1590         switch (which) {
1591         case P_ALL:
1592                 pid = -1;
1593                 break;
1594         case P_PID:
1595                 if (pid <= 0)
1596                         return -EINVAL;
1597                 break;
1598         case P_PGID:
1599                 if (pid <= 0)
1600                         return -EINVAL;
1601                 pid = -pid;
1602                 break;
1603         default:
1604                 return -EINVAL;
1605         }
1606
1607         ret = do_wait(pid, options, infop, NULL, ru);
1608
1609         /* avoid REGPARM breakage on x86: */
1610         prevent_tail_call(ret);
1611         return ret;
1612 }
1613
1614 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1615                           int options, struct rusage __user *ru)
1616 {
1617         long ret;
1618
1619         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1620                         __WNOTHREAD|__WCLONE|__WALL))
1621                 return -EINVAL;
1622         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1623
1624         /* avoid REGPARM breakage on x86: */
1625         prevent_tail_call(ret);
1626         return ret;
1627 }
1628
1629 #ifdef __ARCH_WANT_SYS_WAITPID
1630
1631 /*
1632  * sys_waitpid() remains for compatibility. waitpid() should be
1633  * implemented by calling sys_wait4() from libc.a.
1634  */
1635 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1636 {
1637         return sys_wait4(pid, stat_addr, options, NULL);
1638 }
1639
1640 #endif