kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/mm.h>
   8 #include <linux/slab.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/smp_lock.h>
  11 #include <linux/module.h>
  12 #include <linux/capability.h>
  13 #include <linux/completion.h>
  14 #include <linux/personality.h>
  15 #include <linux/tty.h>
  16 #include <linux/namespace.h>
  17 #include <linux/key.h>
  18 #include <linux/security.h>
  19 #include <linux/cpu.h>
  20 #include <linux/acct.h>
  21 #include <linux/file.h>
  22 #include <linux/binfmts.h>
  23 #include <linux/ptrace.h>
  24 #include <linux/profile.h>
  25 #include <linux/mount.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/mempolicy.h>
  28 #include <linux/taskstats_kern.h>
  29 #include <linux/delayacct.h>
  30 #include <linux/cpuset.h>
  31 #include <linux/syscalls.h>
  32 #include <linux/signal.h>
  33 #include <linux/posix-timers.h>
  34 #include <linux/cn_proc.h>
  35 #include <linux/mutex.h>
  36 #include <linux/futex.h>
  37 #include <linux/compat.h>
  38 #include <linux/pipe_fs_i.h>
  39 #include <linux/audit.h> /* for audit_free() */
  40 #include <linux/resource.h>
  41 #include <linux/vs_limit.h>
  42 #include <linux/vs_context.h>
  43 #include <linux/vs_network.h>
  44
  45 #include <asm/uaccess.h>
  46 #include <asm/unistd.h>
  47 #include <asm/pgtable.h>
  48 #include <asm/mmu_context.h>
  49
  50 extern void sem_exit (void);
  51 extern struct task_struct *child_reaper;
  52
  53 static void exit_mm(struct task_struct * tsk);
  54
  55 static void __unhash_process(struct task_struct *p)
  56 {
  57         nr_threads--;
  58         detach_pid(p, PIDTYPE_PID);
  59         if (thread_group_leader(p)) {
  60                 detach_pid(p, PIDTYPE_PGID);
  61                 detach_pid(p, PIDTYPE_SID);
  62
  63                 list_del_rcu(&p->tasks);
  64                 __get_cpu_var(process_counts)--;
  65         }
  66         list_del_rcu(&p->thread_group);
  67         remove_parent(p);
  68 }
  69
  70 /*
  71  * This function expects the tasklist_lock write-locked.
  72  */
  73 static void __exit_signal(struct task_struct *tsk)
  74 {
  75         struct signal_struct *sig = tsk->signal;
  76         struct sighand_struct *sighand;
  77
  78         BUG_ON(!sig);
  79         BUG_ON(!atomic_read(&sig->count));
  80
  81         rcu_read_lock();
  82         sighand = rcu_dereference(tsk->sighand);
  83         spin_lock(&sighand->siglock);
  84
  85         posix_cpu_timers_exit(tsk);
  86         if (atomic_dec_and_test(&sig->count))
  87                 posix_cpu_timers_exit_group(tsk);
  88         else {
  89                 /*
  90                  * If there is any task waiting for the group exit
  91                  * then notify it:
  92                  */
  93                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
  94                         wake_up_process(sig->group_exit_task);
  95                         sig->group_exit_task = NULL;
  96                 }
  97                 if (tsk == sig->curr_target)
  98                         sig->curr_target = next_thread(tsk);
  99                 /*
 100                  * Accumulate here the counters for all threads but the
 101                  * group leader as they die, so they can be added into
 102                  * the process-wide totals when those are taken.
 103                  * The group leader stays around as a zombie as long
 104                  * as there are other threads.  When it gets reaped,
 105                  * the exit.c code will add its counts into these totals.
 106                  * We won't ever get here for the group leader, since it
 107                  * will have been the last reference on the signal_struct.
 108                  */
 109                 sig->utime = cputime_add(sig->utime, tsk->utime);
 110                 sig->stime = cputime_add(sig->stime, tsk->stime);
 111                 sig->min_flt += tsk->min_flt;
 112                 sig->maj_flt += tsk->maj_flt;
 113                 sig->nvcsw += tsk->nvcsw;
 114                 sig->nivcsw += tsk->nivcsw;
 115                 sig->sched_time += tsk->sched_time;
 116                 sig = NULL; /* Marker for below. */
 117         }
 118
 119         __unhash_process(tsk);
 120
 121         tsk->signal = NULL;
 122         tsk->sighand = NULL;
 123         spin_unlock(&sighand->siglock);
 124         rcu_read_unlock();
 125
 126         __cleanup_sighand(sighand);
 127         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 128         flush_sigqueue(&tsk->pending);
 129         if (sig) {
 130                 flush_sigqueue(&sig->shared_pending);
 131                 __cleanup_signal(sig);
 132         }
 133 }
 134
 135 static void delayed_put_task_struct(struct rcu_head *rhp)
 136 {
 137         put_task_struct(container_of(rhp, struct task_struct, rcu));
 138 }
 139
 140 void release_task(struct task_struct * p)
 141 {
 142         struct task_struct *leader;
 143         int zap_leader;
 144 repeat:
 145         atomic_dec(&p->user->processes);
 146         write_lock_irq(&tasklist_lock);
 147         ptrace_unlink(p);
 148         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 149         __exit_signal(p);
 150
 151         /*
 152          * If we are the last non-leader member of the thread
 153          * group, and the leader is zombie, then notify the
 154          * group leader's parent process. (if it wants notification.)
 155          */
 156         zap_leader = 0;
 157         leader = p->group_leader;
 158         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 159                 BUG_ON(leader->exit_signal == -1);
 160                 do_notify_parent(leader, leader->exit_signal);
 161                 /*
 162                  * If we were the last child thread and the leader has
 163                  * exited already, and the leader's parent ignores SIGCHLD,
 164                  * then we are the one who should release the leader.
 165                  *
 166                  * do_notify_parent() will have marked it self-reaping in
 167                  * that case.
 168                  */
 169                 zap_leader = (leader->exit_signal == -1);
 170         }
 171
 172         sched_exit(p);
 173         write_unlock_irq(&tasklist_lock);
 174         proc_flush_task(p);
 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, struct task_struct *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         proc_clear_tty(current);
 400
 401         /* Block and flush all signals */
 402         sigfillset(&blocked);
 403         sigprocmask(SIG_BLOCK, &blocked, NULL);
 404         flush_signals(current);
 405
 406         /* Become as one with the init task */
 407
 408         exit_fs(current);       /* current->fs->count--; */
 409         fs = init_task.fs;
 410         current->fs = fs;
 411         atomic_inc(&fs->count);
 412         exit_namespace(current);
 413         current->namespace = init_task.namespace;
 414         get_namespace(current->namespace);
 415         exit_files(current);
 416         current->files = init_task.files;
 417         atomic_inc(&current->files->count);
 418
 419         reparent_to_init();
 420 }
 421
 422 EXPORT_SYMBOL(daemonize);
 423
 424 static void close_files(struct files_struct * files)
 425 {
 426         int i, j;
 427         struct fdtable *fdt;
 428
 429         j = 0;
 430
 431         /*
 432          * It is safe to dereference the fd table without RCU or
 433          * ->file_lock because this is the last reference to the
 434          * files structure.
 435          */
 436         fdt = files_fdtable(files);
 437         for (;;) {
 438                 unsigned long set;
 439                 i = j * __NFDBITS;
 440                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 441                         break;
 442                 set = fdt->open_fds->fds_bits[j++];
 443                 while (set) {
 444                         if (set & 1) {
 445                                 struct file * file = xchg(&fdt->fd[i], NULL);
 446                                 if (file)
 447                                         filp_close(file, files);
 448                                 vx_openfd_dec(i);
 449                         }
 450                         i++;
 451                         set >>= 1;
 452                         cond_resched();
 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         BUG_ON(mm != tsk->active_mm);
 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
 591 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 592 {
 593         /* check for reaper context */
 594         vxwprintk((p->xid != reaper->xid) && (reaper != child_reaper),
 595                 "rogue reaper: %p[%d,#%u] <> %p[%d,#%u]",
 596                 p, p->pid, p->xid, reaper, reaper->pid, reaper->xid);
 597
 598         /*
 599          * Make sure we're not reparenting to ourselves and that
 600          * the parent is not a zombie.
 601          */
 602         BUG_ON(p == reaper || reaper->exit_state);
 603         p->real_parent = reaper;
 604 }
 605
 606 static void
 607 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 608 {
 609         /* We don't want people slaying init.  */
 610         if (p->exit_signal != -1)
 611                 p->exit_signal = SIGCHLD;
 612
 613         if (p->pdeath_signal)
 614                 /* We already hold the tasklist_lock here.  */
 615                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 616
 617         /* Move the child from its dying parent to the new one.  */
 618         if (unlikely(traced)) {
 619                 /* Preserve ptrace links if someone else is tracing this child.  */
 620                 list_del_init(&p->ptrace_list);
 621                 if (p->parent != p->real_parent)
 622                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 623         } else {
 624                 /* If this child is being traced, then we're the one tracing it
 625                  * anyway, so let go of it.
 626                  */
 627                 p->ptrace = 0;
 628                 remove_parent(p);
 629                 p->parent = p->real_parent;
 630                 add_parent(p);
 631
 632                 /* If we'd notified the old parent about this child's death,
 633                  * also notify the new parent.
 634                  */
 635                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 636                     thread_group_empty(p))
 637                         do_notify_parent(p, p->exit_signal);
 638                 else if (p->state == TASK_TRACED) {
 639                         /*
 640                          * If it was at a trace stop, turn it into
 641                          * a normal stop since it's no longer being
 642                          * traced.
 643                          */
 644                         ptrace_untrace(p);
 645                 }
 646         }
 647
 648         /*
 649          * process group orphan check
 650          * Case ii: Our child is in a different pgrp
 651          * than we are, and it was the only connection
 652          * outside, so the child pgrp is now orphaned.
 653          */
 654         if ((process_group(p) != process_group(father)) &&
 655             (p->signal->session == father->signal->session)) {
 656                 int pgrp = process_group(p);
 657
 658                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 659                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 660                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 661                 }
 662         }
 663 }
 664
 665 /*
 666  * When we die, we re-parent all our children.
 667  * Try to give them to another thread in our thread
 668  * group, and if no such member exists, give it to
 669  * the global child reaper process (ie "init")
 670  */
 671 static void
 672 forget_original_parent(struct task_struct *father, struct list_head *to_release)
 673 {
 674         struct task_struct *p, *reaper = father;
 675         struct list_head *_p, *_n;
 676
 677         do {
 678                 reaper = next_thread(reaper);
 679                 if (reaper == father) {
 680                         reaper = vx_child_reaper(father);
 681                         break;
 682                 }
 683         } while (reaper->exit_state);
 684
 685         /*
 686          * There are only two places where our children can be:
 687          *
 688          * - in our child list
 689          * - in our ptraced child list
 690          *
 691          * Search them and reparent children.
 692          */
 693         list_for_each_safe(_p, _n, &father->children) {
 694                 int ptrace;
 695                 p = list_entry(_p, struct task_struct, sibling);
 696
 697                 ptrace = p->ptrace;
 698
 699                 /* if father isn't the real parent, then ptrace must be enabled */
 700                 BUG_ON(father != p->real_parent && !ptrace);
 701
 702                 if (father == p->real_parent) {
 703                         /* reparent with a reaper, real father it's us */
 704                         choose_new_parent(p, vx_child_reaper(p));
 705                         reparent_thread(p, father, 0);
 706                 } else {
 707                         /* reparent ptraced task to its real parent */
 708                         __ptrace_unlink (p);
 709                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 710                             thread_group_empty(p))
 711                                 do_notify_parent(p, p->exit_signal);
 712                 }
 713
 714                 /*
 715                  * if the ptraced child is a zombie with exit_signal == -1
 716                  * we must collect it before we exit, or it will remain
 717                  * zombie forever since we prevented it from self-reap itself
 718                  * while it was being traced by us, to be able to see it in wait4.
 719                  */
 720                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 721                         list_add(&p->ptrace_list, to_release);
 722         }
 723         list_for_each_safe(_p, _n, &father->ptrace_children) {
 724                 p = list_entry(_p, struct task_struct, ptrace_list);
 725                 choose_new_parent(p, reaper);
 726                 reparent_thread(p, father, 1);
 727         }
 728 }
 729
 730 /*
 731  * Send signals to all our closest relatives so that they know
 732  * to properly mourn us..
 733  */
 734 static void exit_notify(struct task_struct *tsk)
 735 {
 736         int state;
 737         struct task_struct *t;
 738         struct list_head ptrace_dead, *_p, *_n;
 739
 740         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 741             && !thread_group_empty(tsk)) {
 742                 /*
 743                  * This occurs when there was a race between our exit
 744                  * syscall and a group signal choosing us as the one to
 745                  * wake up.  It could be that we are the only thread
 746                  * alerted to check for pending signals, but another thread
 747                  * should be woken now to take the signal since we will not.
 748                  * Now we'll wake all the threads in the group just to make
 749                  * sure someone gets all the pending signals.
 750                  */
 751                 read_lock(&tasklist_lock);
 752                 spin_lock_irq(&tsk->sighand->siglock);
 753                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 754                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 755                                 recalc_sigpending_tsk(t);
 756                                 if (signal_pending(t))
 757                                         signal_wake_up(t, 0);
 758                         }
 759                 spin_unlock_irq(&tsk->sighand->siglock);
 760                 read_unlock(&tasklist_lock);
 761         }
 762
 763         write_lock_irq(&tasklist_lock);
 764
 765         /*
 766          * This does two things:
 767          *
 768          * A.  Make init inherit all the child processes
 769          * B.  Check to see if any process groups have become orphaned
 770          *      as a result of our exiting, and if they have any stopped
 771          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 772          */
 773
 774         INIT_LIST_HEAD(&ptrace_dead);
 775         forget_original_parent(tsk, &ptrace_dead);
 776         BUG_ON(!list_empty(&tsk->children));
 777         BUG_ON(!list_empty(&tsk->ptrace_children));
 778
 779         /*
 780          * Check to see if any process groups have become orphaned
 781          * as a result of our exiting, and if they have any stopped
 782          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 783          *
 784          * Case i: Our father is in a different pgrp than we are
 785          * and we were the only connection outside, so our pgrp
 786          * is about to become orphaned.
 787          */
 788
 789         t = tsk->real_parent;
 790
 791         if ((process_group(t) != process_group(tsk)) &&
 792             (t->signal->session == tsk->signal->session) &&
 793             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 794             has_stopped_jobs(process_group(tsk))) {
 795                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 796                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 797         }
 798
 799         /* Let father know we died
 800          *
 801          * Thread signals are configurable, but you aren't going to use
 802          * that to send signals to arbitary processes.
 803          * That stops right now.
 804          *
 805          * If the parent exec id doesn't match the exec id we saved
 806          * when we started then we know the parent has changed security
 807          * domain.
 808          *
 809          * If our self_exec id doesn't match our parent_exec_id then
 810          * we have changed execution domain as these two values started
 811          * the same after a fork.
 812          *
 813          */
 814
 815         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 816             ( tsk->parent_exec_id != t->self_exec_id  ||
 817               tsk->self_exec_id != tsk->parent_exec_id)
 818             && !capable(CAP_KILL))
 819                 tsk->exit_signal = SIGCHLD;
 820
 821
 822         /* If something other than our normal parent is ptracing us, then
 823          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 824          * only has special meaning to our real parent.
 825          */
 826         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 827                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 828                 do_notify_parent(tsk, signal);
 829         } else if (tsk->ptrace) {
 830                 do_notify_parent(tsk, SIGCHLD);
 831         }
 832
 833         state = EXIT_ZOMBIE;
 834         if (tsk->exit_signal == -1 &&
 835             (likely(tsk->ptrace == 0) ||
 836              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 837                 state = EXIT_DEAD;
 838         tsk->exit_state = state;
 839
 840         write_unlock_irq(&tasklist_lock);
 841
 842         list_for_each_safe(_p, _n, &ptrace_dead) {
 843                 list_del_init(_p);
 844                 t = list_entry(_p, struct task_struct, ptrace_list);
 845                 release_task(t);
 846         }
 847
 848         /* If the process is dead, release it - nobody will wait for it */
 849         if (state == EXIT_DEAD)
 850                 release_task(tsk);
 851 }
 852
 853 fastcall NORET_TYPE void do_exit(long code)
 854 {
 855         struct task_struct *tsk = current;
 856         struct taskstats *tidstats;
 857         int group_dead;
 858         unsigned int mycpu;
 859
 860         profile_task_exit(tsk);
 861
 862         WARN_ON(atomic_read(&tsk->fs_excl));
 863
 864         if (unlikely(in_interrupt()))
 865                 panic("Aiee, killing interrupt handler!");
 866         if (unlikely(!tsk->pid))
 867                 panic("Attempted to kill the idle task!");
 868         if (unlikely(tsk == child_reaper))
 869                 panic("Attempted to kill init!");
 870
 871         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 872                 current->ptrace_message = code;
 873                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 874         }
 875
 876         /*
 877          * We're taking recursive faults here in do_exit. Safest is to just
 878          * leave this task alone and wait for reboot.
 879          */
 880         if (unlikely(tsk->flags & PF_EXITING)) {
 881                 printk(KERN_ALERT
 882                         "Fixing recursive fault but reboot is needed!\n");
 883                 if (tsk->io_context)
 884                         exit_io_context();
 885                 set_current_state(TASK_UNINTERRUPTIBLE);
 886                 schedule();
 887         }
 888
 889         tsk->flags |= PF_EXITING;
 890
 891         if (unlikely(in_atomic()))
 892                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 893                                 current->comm, current->pid,
 894                                 preempt_count());
 895
 896         taskstats_exit_alloc(&tidstats, &mycpu);
 897
 898         acct_update_integrals(tsk);
 899         if (tsk->mm) {
 900                 update_hiwater_rss(tsk->mm);
 901                 update_hiwater_vm(tsk->mm);
 902         }
 903         group_dead = atomic_dec_and_test(&tsk->signal->live);
 904         if (group_dead) {
 905                 hrtimer_cancel(&tsk->signal->real_timer);
 906                 exit_itimers(tsk->signal);
 907         }
 908
 909         if (current->tux_info) {
 910 #ifdef CONFIG_TUX_DEBUG
 911                 printk("Possibly unexpected TUX-thread exit(%ld) at %p?\n",
 912                         code, __builtin_return_address(0));
 913 #endif
 914                 current->tux_exit();
 915         }
 916
 917         acct_collect(code, group_dead);
 918         if (unlikely(tsk->robust_list))
 919                 exit_robust_list(tsk);
 920 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 921         if (unlikely(tsk->compat_robust_list))
 922                 compat_exit_robust_list(tsk);
 923 #endif
 924         if (unlikely(tsk->audit_context))
 925                 audit_free(tsk);
 926         taskstats_exit_send(tsk, tidstats, group_dead, mycpu);
 927         taskstats_exit_free(tidstats);
 928
 929         exit_mm(tsk);
 930
 931         if (group_dead)
 932                 acct_process();
 933         exit_sem(tsk);
 934         __exit_files(tsk);
 935         __exit_fs(tsk);
 936         exit_namespace(tsk);
 937         exit_thread();
 938         cpuset_exit(tsk);
 939         exit_keys(tsk);
 940
 941         if (group_dead && tsk->signal->leader)
 942                 disassociate_ctty(1);
 943
 944         module_put(task_thread_info(tsk)->exec_domain->module);
 945         if (tsk->binfmt)
 946                 module_put(tsk->binfmt->module);
 947
 948         tsk->exit_code = code;
 949         proc_exit_connector(tsk);
 950         /* needs to stay before exit_notify() */
 951         exit_vx_info_early(tsk, code);
 952         exit_notify(tsk);
 953 #ifdef CONFIG_NUMA
 954         mpol_free(tsk->mempolicy);
 955         tsk->mempolicy = NULL;
 956 #endif
 957         /*
 958          * This must happen late, after the PID is not
 959          * hashed anymore:
 960          */
 961         if (unlikely(!list_empty(&tsk->pi_state_list)))
 962                 exit_pi_state_list(tsk);
 963         if (unlikely(current->pi_state_cache))
 964                 kfree(current->pi_state_cache);
 965         /*
 966          * Make sure we are holding no locks:
 967          */
 968         debug_check_no_locks_held(tsk);
 969
 970         if (tsk->io_context)
 971                 exit_io_context();
 972
 973         if (tsk->splice_pipe)
 974                 __free_pipe_info(tsk->splice_pipe);
 975
 976         /* needs to stay after exit_notify() */
 977         exit_vx_info(tsk, code);
 978         exit_nx_info(tsk);
 979
 980         /* PF_DEAD causes final put_task_struct after we schedule. */
 981         preempt_disable();
 982         BUG_ON(tsk->flags & PF_DEAD);
 983         tsk->flags |= PF_DEAD;
 984
 985         schedule();
 986         BUG();
 987         /* Avoid "noreturn function does return".  */
 988         for (;;) ;
 989 }
 990
 991 EXPORT_SYMBOL_GPL(do_exit);
 992
 993 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 994 {
 995         if (comp)
 996                 complete(comp);
 997
 998         do_exit(code);
 999 }
1000
1001 EXPORT_SYMBOL(complete_and_exit);
1002
1003 asmlinkage long sys_exit(int error_code)
1004 {
1005         do_exit((error_code&0xff)<<8);
1006 }
1007
1008 /*
1009  * Take down every thread in the group.  This is called by fatal signals
1010  * as well as by sys_exit_group (below).
1011  */
1012 NORET_TYPE void
1013 do_group_exit(int exit_code)
1014 {
1015         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1016
1017         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1018                 exit_code = current->signal->group_exit_code;
1019         else if (!thread_group_empty(current)) {
1020                 struct signal_struct *const sig = current->signal;
1021                 struct sighand_struct *const sighand = current->sighand;
1022                 spin_lock_irq(&sighand->siglock);
1023                 if (sig->flags & SIGNAL_GROUP_EXIT)
1024                         /* Another thread got here before we took the lock.  */
1025                         exit_code = sig->group_exit_code;
1026                 else {
1027                         sig->group_exit_code = exit_code;
1028                         zap_other_threads(current);
1029                 }
1030                 spin_unlock_irq(&sighand->siglock);
1031         }
1032
1033         do_exit(exit_code);
1034         /* NOTREACHED */
1035 }
1036
1037 /*
1038  * this kills every thread in the thread group. Note that any externally
1039  * wait4()-ing process will get the correct exit code - even if this
1040  * thread is not the thread group leader.
1041  */
1042 asmlinkage void sys_exit_group(int error_code)
1043 {
1044         do_group_exit((error_code & 0xff) << 8);
1045 }
1046
1047 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1048 {
1049         if (pid > 0) {
1050                 if (p->pid != pid)
1051                         return 0;
1052         } else if (!pid) {
1053                 if (process_group(p) != process_group(current))
1054                         return 0;
1055         } else if (pid != -1) {
1056                 if (process_group(p) != -pid)
1057                         return 0;
1058         }
1059
1060         /*
1061          * Do not consider detached threads that are
1062          * not ptraced:
1063          */
1064         if (p->exit_signal == -1 && !p->ptrace)
1065                 return 0;
1066
1067         /* Wait for all children (clone and not) if __WALL is set;
1068          * otherwise, wait for clone children *only* if __WCLONE is
1069          * set; otherwise, wait for non-clone children *only*.  (Note:
1070          * A "clone" child here is one that reports to its parent
1071          * using a signal other than SIGCHLD.) */
1072         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1073             && !(options & __WALL))
1074                 return 0;
1075         /*
1076          * Do not consider thread group leaders that are
1077          * in a non-empty thread group:
1078          */
1079         if (delay_group_leader(p))
1080                 return 2;
1081
1082         if (security_task_wait(p))
1083                 return 0;
1084
1085         return 1;
1086 }
1087
1088 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1089                                int why, int status,
1090                                struct siginfo __user *infop,
1091                                struct rusage __user *rusagep)
1092 {
1093         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1094
1095         put_task_struct(p);
1096         if (!retval)
1097                 retval = put_user(SIGCHLD, &infop->si_signo);
1098         if (!retval)
1099                 retval = put_user(0, &infop->si_errno);
1100         if (!retval)
1101                 retval = put_user((short)why, &infop->si_code);
1102         if (!retval)
1103                 retval = put_user(pid, &infop->si_pid);
1104         if (!retval)
1105                 retval = put_user(uid, &infop->si_uid);
1106         if (!retval)
1107                 retval = put_user(status, &infop->si_status);
1108         if (!retval)
1109                 retval = pid;
1110         return retval;
1111 }
1112
1113 /*
1114  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1115  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1116  * the lock and this task is uninteresting.  If we return nonzero, we have
1117  * released the lock and the system call should return.
1118  */
1119 static int wait_task_zombie(struct task_struct *p, int noreap,
1120                             struct siginfo __user *infop,
1121                             int __user *stat_addr, struct rusage __user *ru)
1122 {
1123         unsigned long state;
1124         int retval;
1125         int status;
1126
1127         if (unlikely(noreap)) {
1128                 pid_t pid = p->pid;
1129                 uid_t uid = p->uid;
1130                 int exit_code = p->exit_code;
1131                 int why, status;
1132
1133                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1134                         return 0;
1135                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1136                         return 0;
1137                 get_task_struct(p);
1138                 read_unlock(&tasklist_lock);
1139                 if ((exit_code & 0x7f) == 0) {
1140                         why = CLD_EXITED;
1141                         status = exit_code >> 8;
1142                 } else {
1143                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1144                         status = exit_code & 0x7f;
1145                 }
1146                 return wait_noreap_copyout(p, pid, uid, why,
1147                                            status, infop, ru);
1148         }
1149
1150         /*
1151          * Try to move the task's state to DEAD
1152          * only one thread is allowed to do this:
1153          */
1154         state = xchg(&p->exit_state, EXIT_DEAD);
1155         if (state != EXIT_ZOMBIE) {
1156                 BUG_ON(state != EXIT_DEAD);
1157                 return 0;
1158         }
1159         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1160                 /*
1161                  * This can only happen in a race with a ptraced thread
1162                  * dying on another processor.
1163                  */
1164                 return 0;
1165         }
1166
1167         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1168                 struct signal_struct *psig;
1169                 struct signal_struct *sig;
1170
1171                 /*
1172                  * The resource counters for the group leader are in its
1173                  * own task_struct.  Those for dead threads in the group
1174                  * are in its signal_struct, as are those for the child
1175                  * processes it has previously reaped.  All these
1176                  * accumulate in the parent's signal_struct c* fields.
1177                  *
1178                  * We don't bother to take a lock here to protect these
1179                  * p->signal fields, because they are only touched by
1180                  * __exit_signal, which runs with tasklist_lock
1181                  * write-locked anyway, and so is excluded here.  We do
1182                  * need to protect the access to p->parent->signal fields,
1183                  * as other threads in the parent group can be right
1184                  * here reaping other children at the same time.
1185                  */
1186                 spin_lock_irq(&p->parent->sighand->siglock);
1187                 psig = p->parent->signal;
1188                 sig = p->signal;
1189                 psig->cutime =
1190                         cputime_add(psig->cutime,
1191                         cputime_add(p->utime,
1192                         cputime_add(sig->utime,
1193                                     sig->cutime)));
1194                 psig->cstime =
1195                         cputime_add(psig->cstime,
1196                         cputime_add(p->stime,
1197                         cputime_add(sig->stime,
1198                                     sig->cstime)));
1199                 psig->cmin_flt +=
1200                         p->min_flt + sig->min_flt + sig->cmin_flt;
1201                 psig->cmaj_flt +=
1202                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1203                 psig->cnvcsw +=
1204                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1205                 psig->cnivcsw +=
1206                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1207                 spin_unlock_irq(&p->parent->sighand->siglock);
1208         }
1209
1210         /*
1211          * Now we are sure this task is interesting, and no other
1212          * thread can reap it because we set its state to EXIT_DEAD.
1213          */
1214         read_unlock(&tasklist_lock);
1215
1216         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1217         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1218                 ? p->signal->group_exit_code : p->exit_code;
1219         if (!retval && stat_addr)
1220                 retval = put_user(status, stat_addr);
1221         if (!retval && infop)
1222                 retval = put_user(SIGCHLD, &infop->si_signo);
1223         if (!retval && infop)
1224                 retval = put_user(0, &infop->si_errno);
1225         if (!retval && infop) {
1226                 int why;
1227
1228                 if ((status & 0x7f) == 0) {
1229                         why = CLD_EXITED;
1230                         status >>= 8;
1231                 } else {
1232                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1233                         status &= 0x7f;
1234                 }
1235                 retval = put_user((short)why, &infop->si_code);
1236                 if (!retval)
1237                         retval = put_user(status, &infop->si_status);
1238         }
1239         if (!retval && infop)
1240                 retval = put_user(p->pid, &infop->si_pid);
1241         if (!retval && infop)
1242                 retval = put_user(p->uid, &infop->si_uid);
1243         if (retval) {
1244                 // TODO: is this safe?
1245                 p->exit_state = EXIT_ZOMBIE;
1246                 return retval;
1247         }
1248         retval = p->pid;
1249         if (p->real_parent != p->parent) {
1250                 write_lock_irq(&tasklist_lock);
1251                 /* Double-check with lock held.  */
1252                 if (p->real_parent != p->parent) {
1253                         __ptrace_unlink(p);
1254                         // TODO: is this safe?
1255                         p->exit_state = EXIT_ZOMBIE;
1256                         /*
1257                          * If this is not a detached task, notify the parent.
1258                          * If it's still not detached after that, don't release
1259                          * it now.
1260                          */
1261                         if (p->exit_signal != -1) {
1262                                 do_notify_parent(p, p->exit_signal);
1263                                 if (p->exit_signal != -1)
1264                                         p = NULL;
1265                         }
1266                 }
1267                 write_unlock_irq(&tasklist_lock);
1268         }
1269         if (p != NULL)
1270                 release_task(p);
1271         BUG_ON(!retval);
1272         return retval;
1273 }
1274
1275 /*
1276  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1277  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1278  * the lock and this task is uninteresting.  If we return nonzero, we have
1279  * released the lock and the system call should return.
1280  */
1281 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1282                              int noreap, struct siginfo __user *infop,
1283                              int __user *stat_addr, struct rusage __user *ru)
1284 {
1285         int retval, exit_code;
1286
1287         if (!p->exit_code)
1288                 return 0;
1289         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1290             p->signal && p->signal->group_stop_count > 0)
1291                 /*
1292                  * A group stop is in progress and this is the group leader.
1293                  * We won't report until all threads have stopped.
1294                  */
1295                 return 0;
1296
1297         /*
1298          * Now we are pretty sure this task is interesting.
1299          * Make sure it doesn't get reaped out from under us while we
1300          * give up the lock and then examine it below.  We don't want to
1301          * keep holding onto the tasklist_lock while we call getrusage and
1302          * possibly take page faults for user memory.
1303          */
1304         get_task_struct(p);
1305         read_unlock(&tasklist_lock);
1306
1307         if (unlikely(noreap)) {
1308                 pid_t pid = p->pid;
1309                 uid_t uid = p->uid;
1310                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1311
1312                 exit_code = p->exit_code;
1313                 if (unlikely(!exit_code) ||
1314                     unlikely(p->state & TASK_TRACED))
1315                         goto bail_ref;
1316                 return wait_noreap_copyout(p, pid, uid,
1317                                            why, (exit_code << 8) | 0x7f,
1318                                            infop, ru);
1319         }
1320
1321         write_lock_irq(&tasklist_lock);
1322
1323         /*
1324          * This uses xchg to be atomic with the thread resuming and setting
1325          * it.  It must also be done with the write lock held to prevent a
1326          * race with the EXIT_ZOMBIE case.
1327          */
1328         exit_code = xchg(&p->exit_code, 0);
1329         if (unlikely(p->exit_state)) {
1330                 /*
1331                  * The task resumed and then died.  Let the next iteration
1332                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1333                  * already be zero here if it resumed and did _exit(0).
1334                  * The task itself is dead and won't touch exit_code again;
1335                  * other processors in this function are locked out.
1336                  */
1337                 p->exit_code = exit_code;
1338                 exit_code = 0;
1339         }
1340         if (unlikely(exit_code == 0)) {
1341                 /*
1342                  * Another thread in this function got to it first, or it
1343                  * resumed, or it resumed and then died.
1344                  */
1345                 write_unlock_irq(&tasklist_lock);
1346 bail_ref:
1347                 put_task_struct(p);
1348                 /*
1349                  * We are returning to the wait loop without having successfully
1350                  * removed the process and having released the lock. We cannot
1351                  * continue, since the "p" task pointer is potentially stale.
1352                  *
1353                  * Return -EAGAIN, and do_wait() will restart the loop from the
1354                  * beginning. Do _not_ re-acquire the lock.
1355                  */
1356                 return -EAGAIN;
1357         }
1358
1359         /* move to end of parent's list to avoid starvation */
1360         remove_parent(p);
1361         add_parent(p);
1362
1363         write_unlock_irq(&tasklist_lock);
1364
1365         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1366         if (!retval && stat_addr)
1367                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1368         if (!retval && infop)
1369                 retval = put_user(SIGCHLD, &infop->si_signo);
1370         if (!retval && infop)
1371                 retval = put_user(0, &infop->si_errno);
1372         if (!retval && infop)
1373                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1374                                           ? CLD_TRAPPED : CLD_STOPPED),
1375                                   &infop->si_code);
1376         if (!retval && infop)
1377                 retval = put_user(exit_code, &infop->si_status);
1378         if (!retval && infop)
1379                 retval = put_user(p->pid, &infop->si_pid);
1380         if (!retval && infop)
1381                 retval = put_user(p->uid, &infop->si_uid);
1382         if (!retval)
1383                 retval = p->pid;
1384         put_task_struct(p);
1385
1386         BUG_ON(!retval);
1387         return retval;
1388 }
1389
1390 /*
1391  * Handle do_wait work for one task in a live, non-stopped state.
1392  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1393  * the lock and this task is uninteresting.  If we return nonzero, we have
1394  * released the lock and the system call should return.
1395  */
1396 static int wait_task_continued(struct task_struct *p, int noreap,
1397                                struct siginfo __user *infop,
1398                                int __user *stat_addr, struct rusage __user *ru)
1399 {
1400         int retval;
1401         pid_t pid;
1402         uid_t uid;
1403
1404         if (unlikely(!p->signal))
1405                 return 0;
1406
1407         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1408                 return 0;
1409
1410         spin_lock_irq(&p->sighand->siglock);
1411         /* Re-check with the lock held.  */
1412         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1413                 spin_unlock_irq(&p->sighand->siglock);
1414                 return 0;
1415         }
1416         if (!noreap)
1417                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1418         spin_unlock_irq(&p->sighand->siglock);
1419
1420         pid = p->pid;
1421         uid = p->uid;
1422         get_task_struct(p);
1423         read_unlock(&tasklist_lock);
1424
1425         if (!infop) {
1426                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1427                 put_task_struct(p);
1428                 if (!retval && stat_addr)
1429                         retval = put_user(0xffff, stat_addr);
1430                 if (!retval)
1431                         retval = p->pid;
1432         } else {
1433                 retval = wait_noreap_copyout(p, pid, uid,
1434                                              CLD_CONTINUED, SIGCONT,
1435                                              infop, ru);
1436                 BUG_ON(retval == 0);
1437         }
1438
1439         return retval;
1440 }
1441
1442
1443 static inline int my_ptrace_child(struct task_struct *p)
1444 {
1445         if (!(p->ptrace & PT_PTRACED))
1446                 return 0;
1447         if (!(p->ptrace & PT_ATTACHED))
1448                 return 1;
1449         /*
1450          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1451          * we are the attacher.  If we are the real parent, this is a race
1452          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1453          * which we have to switch the parent links, but has already set
1454          * the flags in p->ptrace.
1455          */
1456         return (p->parent != p->real_parent);
1457 }
1458
1459 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1460                     int __user *stat_addr, struct rusage __user *ru)
1461 {
1462         DECLARE_WAITQUEUE(wait, current);
1463         struct task_struct *tsk;
1464         int flag, retval;
1465
1466         add_wait_queue(&current->signal->wait_chldexit,&wait);
1467 repeat:
1468         /*
1469          * We will set this flag if we see any child that might later
1470          * match our criteria, even if we are not able to reap it yet.
1471          */
1472         flag = 0;
1473         current->state = TASK_INTERRUPTIBLE;
1474         read_lock(&tasklist_lock);
1475         tsk = current;
1476         do {
1477                 struct task_struct *p;
1478                 struct list_head *_p;
1479                 int ret;
1480
1481                 list_for_each(_p,&tsk->children) {
1482                         p = list_entry(_p, struct task_struct, sibling);
1483
1484                         ret = eligible_child(pid, options, p);
1485                         if (!ret)
1486                                 continue;
1487
1488                         switch (p->state) {
1489                         case TASK_TRACED:
1490                                 /*
1491                                  * When we hit the race with PTRACE_ATTACH,
1492                                  * we will not report this child.  But the
1493                                  * race means it has not yet been moved to
1494                                  * our ptrace_children list, so we need to
1495                                  * set the flag here to avoid a spurious ECHILD
1496                                  * when the race happens with the only child.
1497                                  */
1498                                 flag = 1;
1499                                 if (!my_ptrace_child(p))
1500                                         continue;
1501                                 /*FALLTHROUGH*/
1502                         case TASK_STOPPED:
1503                                 /*
1504                                  * It's stopped now, so it might later
1505                                  * continue, exit, or stop again.
1506                                  */
1507                                 flag = 1;
1508                                 if (!(options & WUNTRACED) &&
1509                                     !my_ptrace_child(p))
1510                                         continue;
1511                                 retval = wait_task_stopped(p, ret == 2,
1512                                                            (options & WNOWAIT),
1513                                                            infop,
1514                                                            stat_addr, ru);
1515                                 if (retval == -EAGAIN)
1516                                         goto repeat;
1517                                 if (retval != 0) /* He released the lock.  */
1518                                         goto end;
1519                                 break;
1520                         default:
1521                         // case EXIT_DEAD:
1522                                 if (p->exit_state == EXIT_DEAD)
1523                                         continue;
1524                         // case EXIT_ZOMBIE:
1525                                 if (p->exit_state == EXIT_ZOMBIE) {
1526                                         /*
1527                                          * Eligible but we cannot release
1528                                          * it yet:
1529                                          */
1530                                         if (ret == 2)
1531                                                 goto check_continued;
1532                                         if (!likely(options & WEXITED))
1533                                                 continue;
1534                                         retval = wait_task_zombie(
1535                                                 p, (options & WNOWAIT),
1536                                                 infop, stat_addr, ru);
1537                                         /* He released the lock.  */
1538                                         if (retval != 0)
1539                                                 goto end;
1540                                         break;
1541                                 }
1542 check_continued:
1543                                 /*
1544                                  * It's running now, so it might later
1545                                  * exit, stop, or stop and then continue.
1546                                  */
1547                                 flag = 1;
1548                                 if (!unlikely(options & WCONTINUED))
1549                                         continue;
1550                                 retval = wait_task_continued(
1551                                         p, (options & WNOWAIT),
1552                                         infop, stat_addr, ru);
1553                                 if (retval != 0) /* He released the lock.  */
1554                                         goto end;
1555                                 break;
1556                         }
1557                 }
1558                 if (!flag) {
1559                         list_for_each(_p, &tsk->ptrace_children) {
1560                                 p = list_entry(_p, struct task_struct,
1561                                                 ptrace_list);
1562                                 if (!eligible_child(pid, options, p))
1563                                         continue;
1564                                 flag = 1;
1565                                 break;
1566                         }
1567                 }
1568                 if (options & __WNOTHREAD)
1569                         break;
1570                 tsk = next_thread(tsk);
1571                 BUG_ON(tsk->signal != current->signal);
1572         } while (tsk != current);
1573
1574         read_unlock(&tasklist_lock);
1575         if (flag) {
1576                 retval = 0;
1577                 if (options & WNOHANG)
1578                         goto end;
1579                 retval = -ERESTARTSYS;
1580                 if (signal_pending(current))
1581                         goto end;
1582                 schedule();
1583                 goto repeat;
1584         }
1585         retval = -ECHILD;
1586 end:
1587         current->state = TASK_RUNNING;
1588         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1589         if (infop) {
1590                 if (retval > 0)
1591                 retval = 0;
1592                 else {
1593                         /*
1594                          * For a WNOHANG return, clear out all the fields
1595                          * we would set so the user can easily tell the
1596                          * difference.
1597                          */
1598                         if (!retval)
1599                                 retval = put_user(0, &infop->si_signo);
1600                         if (!retval)
1601                                 retval = put_user(0, &infop->si_errno);
1602                         if (!retval)
1603                                 retval = put_user(0, &infop->si_code);
1604                         if (!retval)
1605                                 retval = put_user(0, &infop->si_pid);
1606                         if (!retval)
1607                                 retval = put_user(0, &infop->si_uid);
1608                         if (!retval)
1609                                 retval = put_user(0, &infop->si_status);
1610                 }
1611         }
1612         return retval;
1613 }
1614
1615 asmlinkage long sys_waitid(int which, pid_t pid,
1616                            struct siginfo __user *infop, int options,
1617                            struct rusage __user *ru)
1618 {
1619         long ret;
1620
1621         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1622                 return -EINVAL;
1623         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1624                 return -EINVAL;
1625
1626         switch (which) {
1627         case P_ALL:
1628                 pid = -1;
1629                 break;
1630         case P_PID:
1631                 if (pid <= 0)
1632                         return -EINVAL;
1633                 break;
1634         case P_PGID:
1635                 if (pid <= 0)
1636                         return -EINVAL;
1637                 pid = -pid;
1638                 break;
1639         default:
1640                 return -EINVAL;
1641         }
1642
1643         ret = do_wait(pid, options, infop, NULL, ru);
1644
1645         /* avoid REGPARM breakage on x86: */
1646         prevent_tail_call(ret);
1647         return ret;
1648 }
1649
1650 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1651                           int options, struct rusage __user *ru)
1652 {
1653         long ret;
1654
1655         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1656                         __WNOTHREAD|__WCLONE|__WALL))
1657                 return -EINVAL;
1658         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1659
1660         /* avoid REGPARM breakage on x86: */
1661         prevent_tail_call(ret);
1662         return ret;
1663 }
1664
1665 #ifdef __ARCH_WANT_SYS_WAITPID
1666
1667 /*
1668  * sys_waitpid() remains for compatibility. waitpid() should be
1669  * implemented by calling sys_wait4() from libc.a.
1670  */
1671 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1672 {
1673         return sys_wait4(pid, stat_addr, options, NULL);
1674 }
1675
1676 #endif