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/module.h>
  11 #include <linux/capability.h>
  12 #include <linux/completion.h>
  13 #include <linux/personality.h>
  14 #include <linux/tty.h>
  15 #include <linux/mnt_namespace.h>
  16 #include <linux/key.h>
  17 #include <linux/security.h>
  18 #include <linux/cpu.h>
  19 #include <linux/acct.h>
  20 #include <linux/tsacct_kern.h>
  21 #include <linux/file.h>
  22 #include <linux/binfmts.h>
  23 #include <linux/nsproxy.h>
  24 #include <linux/pid_namespace.h>
  25 #include <linux/ptrace.h>
  26 #include <linux/profile.h>
  27 #include <linux/mount.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/mempolicy.h>
  30 #include <linux/taskstats_kern.h>
  31 #include <linux/delayacct.h>
  32 #include <linux/cpuset.h>
  33 #include <linux/syscalls.h>
  34 #include <linux/signal.h>
  35 #include <linux/posix-timers.h>
  36 #include <linux/cn_proc.h>
  37 #include <linux/mutex.h>
  38 #include <linux/futex.h>
  39 #include <linux/compat.h>
  40 #include <linux/pipe_fs_i.h>
  41 #include <linux/audit.h> /* for audit_free() */
  42 #include <linux/resource.h>
  43 #include <linux/blkdev.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
  52 static void exit_mm(struct task_struct * tsk);
  53
  54 static void __unhash_process(struct task_struct *p)
  55 {
  56         nr_threads--;
  57         detach_pid(p, PIDTYPE_PID);
  58         if (thread_group_leader(p)) {
  59                 detach_pid(p, PIDTYPE_PGID);
  60                 detach_pid(p, PIDTYPE_SID);
  61
  62                 list_del_rcu(&p->tasks);
  63                 __get_cpu_var(process_counts)--;
  64         }
  65         list_del_rcu(&p->thread_group);
  66         remove_parent(p);
  67 }
  68
  69 /*
  70  * This function expects the tasklist_lock write-locked.
  71  */
  72 static void __exit_signal(struct task_struct *tsk)
  73 {
  74         struct signal_struct *sig = tsk->signal;
  75         struct sighand_struct *sighand;
  76
  77         BUG_ON(!sig);
  78         BUG_ON(!atomic_read(&sig->count));
  79
  80         rcu_read_lock();
  81         sighand = rcu_dereference(tsk->sighand);
  82         spin_lock(&sighand->siglock);
  83
  84         posix_cpu_timers_exit(tsk);
  85         if (atomic_dec_and_test(&sig->count))
  86                 posix_cpu_timers_exit_group(tsk);
  87         else {
  88                 /*
  89                  * If there is any task waiting for the group exit
  90                  * then notify it:
  91                  */
  92                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
  93                         wake_up_process(sig->group_exit_task);
  94                         sig->group_exit_task = NULL;
  95                 }
  96                 if (tsk == sig->curr_target)
  97                         sig->curr_target = next_thread(tsk);
  98                 /*
  99                  * Accumulate here the counters for all threads but the
 100                  * group leader as they die, so they can be added into
 101                  * the process-wide totals when those are taken.
 102                  * The group leader stays around as a zombie as long
 103                  * as there are other threads.  When it gets reaped,
 104                  * the exit.c code will add its counts into these totals.
 105                  * We won't ever get here for the group leader, since it
 106                  * will have been the last reference on the signal_struct.
 107                  */
 108                 sig->utime = cputime_add(sig->utime, tsk->utime);
 109                 sig->stime = cputime_add(sig->stime, tsk->stime);
 110                 sig->min_flt += tsk->min_flt;
 111                 sig->maj_flt += tsk->maj_flt;
 112                 sig->nvcsw += tsk->nvcsw;
 113                 sig->nivcsw += tsk->nivcsw;
 114                 sig->sched_time += tsk->sched_time;
 115                 sig = NULL; /* Marker for below. */
 116         }
 117
 118         __unhash_process(tsk);
 119
 120         tsk->signal = NULL;
 121         tsk->sighand = NULL;
 122         spin_unlock(&sighand->siglock);
 123         rcu_read_unlock();
 124
 125         __cleanup_sighand(sighand);
 126         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 127         flush_sigqueue(&tsk->pending);
 128         if (sig) {
 129                 flush_sigqueue(&sig->shared_pending);
 130                 taskstats_tgid_free(sig);
 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  * The caller must hold rcu lock or the tasklist lock.
 189  */
 190 struct pid *session_of_pgrp(struct pid *pgrp)
 191 {
 192         struct task_struct *p;
 193         struct pid *sid = NULL;
 194
 195         p = pid_task(pgrp, PIDTYPE_PGID);
 196         if (p == NULL)
 197                 p = pid_task(pgrp, PIDTYPE_PID);
 198         if (p != NULL)
 199                 sid = task_session(p);
 200
 201         return sid;
 202 }
 203
 204 /*
 205  * Determine if a process group is "orphaned", according to the POSIX
 206  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 207  * by terminal-generated stop signals.  Newly orphaned process groups are
 208  * to receive a SIGHUP and a SIGCONT.
 209  *
 210  * "I ask you, have you ever known what it is to be an orphan?"
 211  */
 212 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 213 {
 214         struct task_struct *p;
 215         int ret = 1;
 216
 217         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 218                 if (p == ignored_task
 219                                 || p->exit_state
 220                                 || is_init(p->real_parent))
 221                         continue;
 222                 if (task_pgrp(p->real_parent) != pgrp &&
 223                     task_session(p->real_parent) == task_session(p)) {
 224                         ret = 0;
 225                         break;
 226                 }
 227         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 228         return ret;     /* (sighing) "Often!" */
 229 }
 230
 231 int is_current_pgrp_orphaned(void)
 232 {
 233         int retval;
 234
 235         read_lock(&tasklist_lock);
 236         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 237         read_unlock(&tasklist_lock);
 238
 239         return retval;
 240 }
 241
 242 static int has_stopped_jobs(struct pid *pgrp)
 243 {
 244         int retval = 0;
 245         struct task_struct *p;
 246
 247         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 248                 if (p->state != TASK_STOPPED)
 249                         continue;
 250                 retval = 1;
 251                 break;
 252         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 253         return retval;
 254 }
 255
 256 /**
 257  * reparent_to_init - Reparent the calling kernel thread to the init task of the pid space that the thread belongs to.
 258  *
 259  * If a kernel thread is launched as a result of a system call, or if
 260  * it ever exits, it should generally reparent itself to init so that
 261  * it is correctly cleaned up on exit.
 262  *
 263  * The various task state such as scheduling policy and priority may have
 264  * been inherited from a user process, so we reset them to sane values here.
 265  *
 266  * NOTE that reparent_to_init() gives the caller full capabilities.
 267  */
 268 static void reparent_to_init(void)
 269 {
 270         write_lock_irq(&tasklist_lock);
 271
 272         ptrace_unlink(current);
 273         /* Reparent to init */
 274         remove_parent(current);
 275         current->parent = child_reaper(current);
 276         current->real_parent = child_reaper(current);
 277         add_parent(current);
 278
 279         /* Set the exit signal to SIGCHLD so we signal init on exit */
 280         current->exit_signal = SIGCHLD;
 281
 282         if (!has_rt_policy(current) && (task_nice(current) < 0))
 283                 set_user_nice(current, 0);
 284         /* cpus_allowed? */
 285         /* rt_priority? */
 286         /* signals? */
 287         security_task_reparent_to_init(current);
 288         memcpy(current->signal->rlim, init_task.signal->rlim,
 289                sizeof(current->signal->rlim));
 290         atomic_inc(&(INIT_USER->__count));
 291         write_unlock_irq(&tasklist_lock);
 292         switch_uid(INIT_USER);
 293 }
 294
 295 void __set_special_pids(pid_t session, pid_t pgrp)
 296 {
 297         struct task_struct *curr = current->group_leader;
 298
 299         if (process_session(curr) != session) {
 300                 detach_pid(curr, PIDTYPE_SID);
 301                 set_signal_session(curr->signal, session);
 302                 attach_pid(curr, PIDTYPE_SID, session);
 303         }
 304         if (process_group(curr) != pgrp) {
 305                 detach_pid(curr, PIDTYPE_PGID);
 306                 curr->signal->pgrp = pgrp;
 307                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 308         }
 309 }
 310
 311 static void set_special_pids(pid_t session, pid_t pgrp)
 312 {
 313         write_lock_irq(&tasklist_lock);
 314         __set_special_pids(session, pgrp);
 315         write_unlock_irq(&tasklist_lock);
 316 }
 317
 318 /*
 319  * Let kernel threads use this to say that they
 320  * allow a certain signal (since daemonize() will
 321  * have disabled all of them by default).
 322  */
 323 int allow_signal(int sig)
 324 {
 325         if (!valid_signal(sig) || sig < 1)
 326                 return -EINVAL;
 327
 328         spin_lock_irq(&current->sighand->siglock);
 329         sigdelset(&current->blocked, sig);
 330         if (!current->mm) {
 331                 /* Kernel threads handle their own signals.
 332                    Let the signal code know it'll be handled, so
 333                    that they don't get converted to SIGKILL or
 334                    just silently dropped */
 335                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 336         }
 337         recalc_sigpending();
 338         spin_unlock_irq(&current->sighand->siglock);
 339         return 0;
 340 }
 341
 342 EXPORT_SYMBOL(allow_signal);
 343
 344 int disallow_signal(int sig)
 345 {
 346         if (!valid_signal(sig) || sig < 1)
 347                 return -EINVAL;
 348
 349         spin_lock_irq(&current->sighand->siglock);
 350         sigaddset(&current->blocked, sig);
 351         recalc_sigpending();
 352         spin_unlock_irq(&current->sighand->siglock);
 353         return 0;
 354 }
 355
 356 EXPORT_SYMBOL(disallow_signal);
 357
 358 /*
 359  *      Put all the gunge required to become a kernel thread without
 360  *      attached user resources in one place where it belongs.
 361  */
 362
 363 void daemonize(const char *name, ...)
 364 {
 365         va_list args;
 366         struct fs_struct *fs;
 367         sigset_t blocked;
 368
 369         va_start(args, name);
 370         vsnprintf(current->comm, sizeof(current->comm), name, args);
 371         va_end(args);
 372
 373         /*
 374          * If we were started as result of loading a module, close all of the
 375          * user space pages.  We don't need them, and if we didn't close them
 376          * they would be locked into memory.
 377          */
 378         exit_mm(current);
 379
 380         set_special_pids(1, 1);
 381         proc_clear_tty(current);
 382
 383         /* Block and flush all signals */
 384         sigfillset(&blocked);
 385         sigprocmask(SIG_BLOCK, &blocked, NULL);
 386         flush_signals(current);
 387
 388         /* Become as one with the init task */
 389
 390         exit_fs(current);       /* current->fs->count--; */
 391         fs = init_task.fs;
 392         current->fs = fs;
 393         atomic_inc(&fs->count);
 394
 395         exit_task_namespaces(current);
 396         current->nsproxy = init_task.nsproxy;
 397         get_task_namespaces(current);
 398
 399         exit_files(current);
 400         current->files = init_task.files;
 401         atomic_inc(&current->files->count);
 402
 403         reparent_to_init();
 404 }
 405
 406 EXPORT_SYMBOL(daemonize);
 407
 408 static void close_files(struct files_struct * files)
 409 {
 410         int i, j;
 411         struct fdtable *fdt;
 412
 413         j = 0;
 414
 415         /*
 416          * It is safe to dereference the fd table without RCU or
 417          * ->file_lock because this is the last reference to the
 418          * files structure.
 419          */
 420         fdt = files_fdtable(files);
 421         for (;;) {
 422                 unsigned long set;
 423                 i = j * __NFDBITS;
 424                 if (i >= fdt->max_fds)
 425                         break;
 426                 set = fdt->open_fds->fds_bits[j++];
 427                 while (set) {
 428                         if (set & 1) {
 429                                 struct file * file = xchg(&fdt->fd[i], NULL);
 430                                 if (file) {
 431                                         filp_close(file, files);
 432                                         cond_resched();
 433                                 }
 434                         }
 435                         i++;
 436                         set >>= 1;
 437                 }
 438         }
 439 }
 440
 441 struct files_struct *get_files_struct(struct task_struct *task)
 442 {
 443         struct files_struct *files;
 444
 445         task_lock(task);
 446         files = task->files;
 447         if (files)
 448                 atomic_inc(&files->count);
 449         task_unlock(task);
 450
 451         return files;
 452 }
 453
 454 void fastcall put_files_struct(struct files_struct *files)
 455 {
 456         struct fdtable *fdt;
 457
 458         if (atomic_dec_and_test(&files->count)) {
 459                 close_files(files);
 460                 /*
 461                  * Free the fd and fdset arrays if we expanded them.
 462                  * If the fdtable was embedded, pass files for freeing
 463                  * at the end of the RCU grace period. Otherwise,
 464                  * you can free files immediately.
 465                  */
 466                 fdt = files_fdtable(files);
 467                 if (fdt != &files->fdtab)
 468                         kmem_cache_free(files_cachep, files);
 469                 free_fdtable(fdt);
 470         }
 471 }
 472
 473 EXPORT_SYMBOL(put_files_struct);
 474
 475 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
 476 {
 477         struct files_struct *old;
 478
 479         old = tsk->files;
 480         task_lock(tsk);
 481         tsk->files = files;
 482         task_unlock(tsk);
 483         put_files_struct(old);
 484 }
 485 EXPORT_SYMBOL(reset_files_struct);
 486
 487 static inline void __exit_files(struct task_struct *tsk)
 488 {
 489         struct files_struct * files = tsk->files;
 490
 491         if (files) {
 492                 task_lock(tsk);
 493                 tsk->files = NULL;
 494                 task_unlock(tsk);
 495                 put_files_struct(files);
 496         }
 497 }
 498
 499 void exit_files(struct task_struct *tsk)
 500 {
 501         __exit_files(tsk);
 502 }
 503
 504 static inline void __put_fs_struct(struct fs_struct *fs)
 505 {
 506         /* No need to hold fs->lock if we are killing it */
 507         if (atomic_dec_and_test(&fs->count)) {
 508                 dput(fs->root);
 509                 mntput(fs->rootmnt);
 510                 dput(fs->pwd);
 511                 mntput(fs->pwdmnt);
 512                 if (fs->altroot) {
 513                         dput(fs->altroot);
 514                         mntput(fs->altrootmnt);
 515                 }
 516                 kmem_cache_free(fs_cachep, fs);
 517         }
 518 }
 519
 520 void put_fs_struct(struct fs_struct *fs)
 521 {
 522         __put_fs_struct(fs);
 523 }
 524
 525 static inline void __exit_fs(struct task_struct *tsk)
 526 {
 527         struct fs_struct * fs = tsk->fs;
 528
 529         if (fs) {
 530                 task_lock(tsk);
 531                 tsk->fs = NULL;
 532                 task_unlock(tsk);
 533                 __put_fs_struct(fs);
 534         }
 535 }
 536
 537 void exit_fs(struct task_struct *tsk)
 538 {
 539         __exit_fs(tsk);
 540 }
 541
 542 EXPORT_SYMBOL_GPL(exit_fs);
 543
 544 /*
 545  * Turn us into a lazy TLB process if we
 546  * aren't already..
 547  */
 548 static void exit_mm(struct task_struct * tsk)
 549 {
 550         struct mm_struct *mm = tsk->mm;
 551
 552         mm_release(tsk, mm);
 553         if (!mm)
 554                 return;
 555         /*
 556          * Serialize with any possible pending coredump.
 557          * We must hold mmap_sem around checking core_waiters
 558          * and clearing tsk->mm.  The core-inducing thread
 559          * will increment core_waiters for each thread in the
 560          * group with ->mm != NULL.
 561          */
 562         down_read(&mm->mmap_sem);
 563         if (mm->core_waiters) {
 564                 up_read(&mm->mmap_sem);
 565                 down_write(&mm->mmap_sem);
 566                 if (!--mm->core_waiters)
 567                         complete(mm->core_startup_done);
 568                 up_write(&mm->mmap_sem);
 569
 570                 wait_for_completion(&mm->core_done);
 571                 down_read(&mm->mmap_sem);
 572         }
 573         atomic_inc(&mm->mm_count);
 574         BUG_ON(mm != tsk->active_mm);
 575         /* more a memory barrier than a real lock */
 576         task_lock(tsk);
 577         tsk->mm = NULL;
 578         up_read(&mm->mmap_sem);
 579         enter_lazy_tlb(mm, current);
 580         task_unlock(tsk);
 581         mmput(mm);
 582 }
 583
 584 static inline void
 585 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 586 {
 587         /*
 588          * Make sure we're not reparenting to ourselves and that
 589          * the parent is not a zombie.
 590          */
 591         BUG_ON(p == reaper || reaper->exit_state);
 592         p->real_parent = reaper;
 593 }
 594
 595 static void
 596 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 597 {
 598         if (p->pdeath_signal)
 599                 /* We already hold the tasklist_lock here.  */
 600                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 601
 602         /* Move the child from its dying parent to the new one.  */
 603         if (unlikely(traced)) {
 604                 /* Preserve ptrace links if someone else is tracing this child.  */
 605                 list_del_init(&p->ptrace_list);
 606                 if (p->parent != p->real_parent)
 607                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 608         } else {
 609                 /* If this child is being traced, then we're the one tracing it
 610                  * anyway, so let go of it.
 611                  */
 612                 p->ptrace = 0;
 613                 remove_parent(p);
 614                 p->parent = p->real_parent;
 615                 add_parent(p);
 616
 617                 if (p->state == TASK_TRACED) {
 618                         /*
 619                          * If it was at a trace stop, turn it into
 620                          * a normal stop since it's no longer being
 621                          * traced.
 622                          */
 623                         ptrace_untrace(p);
 624                 }
 625         }
 626
 627         /* If this is a threaded reparent there is no need to
 628          * notify anyone anything has happened.
 629          */
 630         if (p->real_parent->group_leader == father->group_leader)
 631                 return;
 632
 633         /* We don't want people slaying init.  */
 634         if (p->exit_signal != -1)
 635                 p->exit_signal = SIGCHLD;
 636
 637         /* If we'd notified the old parent about this child's death,
 638          * also notify the new parent.
 639          */
 640         if (!traced && p->exit_state == EXIT_ZOMBIE &&
 641             p->exit_signal != -1 && thread_group_empty(p))
 642                 do_notify_parent(p, p->exit_signal);
 643
 644         /*
 645          * process group orphan check
 646          * Case ii: Our child is in a different pgrp
 647          * than we are, and it was the only connection
 648          * outside, so the child pgrp is now orphaned.
 649          */
 650         if ((task_pgrp(p) != task_pgrp(father)) &&
 651             (task_session(p) == task_session(father))) {
 652                 struct pid *pgrp = task_pgrp(p);
 653
 654                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
 655                     has_stopped_jobs(pgrp)) {
 656                         __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 657                         __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 658                 }
 659         }
 660 }
 661
 662 /*
 663  * When we die, we re-parent all our children.
 664  * Try to give them to another thread in our thread
 665  * group, and if no such member exists, give it to
 666  * the child reaper process (ie "init") in our pid
 667  * space.
 668  */
 669 static void
 670 forget_original_parent(struct task_struct *father, struct list_head *to_release)
 671 {
 672         struct task_struct *p, *reaper = father;
 673         struct list_head *_p, *_n;
 674
 675         do {
 676                 reaper = next_thread(reaper);
 677                 if (reaper == father) {
 678                         reaper = child_reaper(father);
 679                         break;
 680                 }
 681         } while (reaper->exit_state);
 682
 683         /*
 684          * There are only two places where our children can be:
 685          *
 686          * - in our child list
 687          * - in our ptraced child list
 688          *
 689          * Search them and reparent children.
 690          */
 691         list_for_each_safe(_p, _n, &father->children) {
 692                 int ptrace;
 693                 p = list_entry(_p, struct task_struct, sibling);
 694
 695                 ptrace = p->ptrace;
 696
 697                 /* if father isn't the real parent, then ptrace must be enabled */
 698                 BUG_ON(father != p->real_parent && !ptrace);
 699
 700                 if (father == p->real_parent) {
 701                         /* reparent with a reaper, real father it's us */
 702                         choose_new_parent(p, reaper);
 703                         reparent_thread(p, father, 0);
 704                 } else {
 705                         /* reparent ptraced task to its real parent */
 706                         __ptrace_unlink (p);
 707                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 708                             thread_group_empty(p))
 709                                 do_notify_parent(p, p->exit_signal);
 710                 }
 711
 712                 /*
 713                  * if the ptraced child is a zombie with exit_signal == -1
 714                  * we must collect it before we exit, or it will remain
 715                  * zombie forever since we prevented it from self-reap itself
 716                  * while it was being traced by us, to be able to see it in wait4.
 717                  */
 718                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 719                         list_add(&p->ptrace_list, to_release);
 720         }
 721         list_for_each_safe(_p, _n, &father->ptrace_children) {
 722                 p = list_entry(_p, struct task_struct, ptrace_list);
 723                 choose_new_parent(p, reaper);
 724                 reparent_thread(p, father, 1);
 725         }
 726 }
 727
 728 /*
 729  * Send signals to all our closest relatives so that they know
 730  * to properly mourn us..
 731  */
 732 static void exit_notify(struct task_struct *tsk)
 733 {
 734         int state;
 735         struct task_struct *t;
 736         struct list_head ptrace_dead, *_p, *_n;
 737         struct pid *pgrp;
 738
 739         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 740             && !thread_group_empty(tsk)) {
 741                 /*
 742                  * This occurs when there was a race between our exit
 743                  * syscall and a group signal choosing us as the one to
 744                  * wake up.  It could be that we are the only thread
 745                  * alerted to check for pending signals, but another thread
 746                  * should be woken now to take the signal since we will not.
 747                  * Now we'll wake all the threads in the group just to make
 748                  * sure someone gets all the pending signals.
 749                  */
 750                 read_lock(&tasklist_lock);
 751                 spin_lock_irq(&tsk->sighand->siglock);
 752                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 753                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 754                                 recalc_sigpending_tsk(t);
 755                                 if (signal_pending(t))
 756                                         signal_wake_up(t, 0);
 757                         }
 758                 spin_unlock_irq(&tsk->sighand->siglock);
 759                 read_unlock(&tasklist_lock);
 760         }
 761
 762         write_lock_irq(&tasklist_lock);
 763
 764         /*
 765          * This does two things:
 766          *
 767          * A.  Make init inherit all the child processes
 768          * B.  Check to see if any process groups have become orphaned
 769          *      as a result of our exiting, and if they have any stopped
 770          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 771          */
 772
 773         INIT_LIST_HEAD(&ptrace_dead);
 774         forget_original_parent(tsk, &ptrace_dead);
 775         BUG_ON(!list_empty(&tsk->children));
 776         BUG_ON(!list_empty(&tsk->ptrace_children));
 777
 778         /*
 779          * Check to see if any process groups have become orphaned
 780          * as a result of our exiting, and if they have any stopped
 781          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 782          *
 783          * Case i: Our father is in a different pgrp than we are
 784          * and we were the only connection outside, so our pgrp
 785          * is about to become orphaned.
 786          */
 787
 788         t = tsk->real_parent;
 789
 790         pgrp = task_pgrp(tsk);
 791         if ((task_pgrp(t) != pgrp) &&
 792             (task_session(t) == task_session(tsk)) &&
 793             will_become_orphaned_pgrp(pgrp, tsk) &&
 794             has_stopped_jobs(pgrp)) {
 795                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 796                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 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         int group_dead;
 857
 858         profile_task_exit(tsk);
 859
 860         WARN_ON(atomic_read(&tsk->fs_excl));
 861
 862         if (unlikely(in_interrupt()))
 863                 panic("Aiee, killing interrupt handler!");
 864         if (unlikely(!tsk->pid))
 865                 panic("Attempted to kill the idle task!");
 866         if (unlikely(tsk == child_reaper(tsk))) {
 867                 if (tsk->nsproxy->pid_ns != &init_pid_ns)
 868                         tsk->nsproxy->pid_ns->child_reaper = init_pid_ns.child_reaper;
 869                 else
 870                         panic("Attempted to kill init!");
 871         }
 872
 873
 874         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 875                 current->ptrace_message = code;
 876                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 877         }
 878
 879         /*
 880          * We're taking recursive faults here in do_exit. Safest is to just
 881          * leave this task alone and wait for reboot.
 882          */
 883         if (unlikely(tsk->flags & PF_EXITING)) {
 884                 printk(KERN_ALERT
 885                         "Fixing recursive fault but reboot is needed!\n");
 886                 if (tsk->io_context)
 887                         exit_io_context();
 888                 set_current_state(TASK_UNINTERRUPTIBLE);
 889                 schedule();
 890         }
 891
 892         tsk->flags |= PF_EXITING;
 893
 894         if (unlikely(in_atomic()))
 895                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 896                                 current->comm, current->pid,
 897                                 preempt_count());
 898
 899         acct_update_integrals(tsk);
 900         if (tsk->mm) {
 901                 update_hiwater_rss(tsk->mm);
 902                 update_hiwater_vm(tsk->mm);
 903         }
 904         group_dead = atomic_dec_and_test(&tsk->signal->live);
 905         if (group_dead) {
 906                 hrtimer_cancel(&tsk->signal->real_timer);
 907                 exit_itimers(tsk->signal);
 908         }
 909         acct_collect(code, group_dead);
 910         if (unlikely(tsk->robust_list))
 911                 exit_robust_list(tsk);
 912 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 913         if (unlikely(tsk->compat_robust_list))
 914                 compat_exit_robust_list(tsk);
 915 #endif
 916         if (unlikely(tsk->audit_context))
 917                 audit_free(tsk);
 918
 919         taskstats_exit(tsk, group_dead);
 920
 921         exit_mm(tsk);
 922
 923         if (group_dead)
 924                 acct_process();
 925         exit_sem(tsk);
 926         __exit_files(tsk);
 927         __exit_fs(tsk);
 928         exit_thread();
 929         cpuset_exit(tsk);
 930         exit_keys(tsk);
 931
 932         if (group_dead && tsk->signal->leader)
 933                 disassociate_ctty(1);
 934
 935         module_put(task_thread_info(tsk)->exec_domain->module);
 936         if (tsk->binfmt)
 937                 module_put(tsk->binfmt->module);
 938
 939         tsk->exit_code = code;
 940         proc_exit_connector(tsk);
 941         exit_task_namespaces(tsk);
 942         exit_notify(tsk);
 943 #ifdef CONFIG_NUMA
 944         mpol_free(tsk->mempolicy);
 945         tsk->mempolicy = NULL;
 946 #endif
 947         /*
 948          * This must happen late, after the PID is not
 949          * hashed anymore:
 950          */
 951         if (unlikely(!list_empty(&tsk->pi_state_list)))
 952                 exit_pi_state_list(tsk);
 953         if (unlikely(current->pi_state_cache))
 954                 kfree(current->pi_state_cache);
 955         /*
 956          * Make sure we are holding no locks:
 957          */
 958         debug_check_no_locks_held(tsk);
 959
 960         if (tsk->io_context)
 961                 exit_io_context();
 962
 963         if (tsk->splice_pipe)
 964                 __free_pipe_info(tsk->splice_pipe);
 965
 966         preempt_disable();
 967         /* causes final put_task_struct in finish_task_switch(). */
 968         tsk->state = TASK_DEAD;
 969
 970         schedule();
 971         BUG();
 972         /* Avoid "noreturn function does return".  */
 973         for (;;)
 974                 cpu_relax();    /* For when BUG is null */
 975 }
 976
 977 EXPORT_SYMBOL_GPL(do_exit);
 978
 979 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 980 {
 981         if (comp)
 982                 complete(comp);
 983
 984         do_exit(code);
 985 }
 986
 987 EXPORT_SYMBOL(complete_and_exit);
 988
 989 asmlinkage long sys_exit(int error_code)
 990 {
 991         do_exit((error_code&0xff)<<8);
 992 }
 993
 994 /*
 995  * Take down every thread in the group.  This is called by fatal signals
 996  * as well as by sys_exit_group (below).
 997  */
 998 NORET_TYPE void
 999 do_group_exit(int exit_code)
1000 {
1001         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1002
1003         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1004                 exit_code = current->signal->group_exit_code;
1005         else if (!thread_group_empty(current)) {
1006                 struct signal_struct *const sig = current->signal;
1007                 struct sighand_struct *const sighand = current->sighand;
1008                 spin_lock_irq(&sighand->siglock);
1009                 if (sig->flags & SIGNAL_GROUP_EXIT)
1010                         /* Another thread got here before we took the lock.  */
1011                         exit_code = sig->group_exit_code;
1012                 else {
1013                         sig->group_exit_code = exit_code;
1014                         zap_other_threads(current);
1015                 }
1016                 spin_unlock_irq(&sighand->siglock);
1017         }
1018
1019         do_exit(exit_code);
1020         /* NOTREACHED */
1021 }
1022
1023 /*
1024  * this kills every thread in the thread group. Note that any externally
1025  * wait4()-ing process will get the correct exit code - even if this
1026  * thread is not the thread group leader.
1027  */
1028 asmlinkage void sys_exit_group(int error_code)
1029 {
1030         do_group_exit((error_code & 0xff) << 8);
1031 }
1032
1033 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1034 {
1035         int err;
1036
1037         if (pid > 0) {
1038                 if (p->pid != pid)
1039                         return 0;
1040         } else if (!pid) {
1041                 if (process_group(p) != process_group(current))
1042                         return 0;
1043         } else if (pid != -1) {
1044                 if (process_group(p) != -pid)
1045                         return 0;
1046         }
1047
1048         /*
1049          * Do not consider detached threads that are
1050          * not ptraced:
1051          */
1052         if (p->exit_signal == -1 && !p->ptrace)
1053                 return 0;
1054
1055         /* Wait for all children (clone and not) if __WALL is set;
1056          * otherwise, wait for clone children *only* if __WCLONE is
1057          * set; otherwise, wait for non-clone children *only*.  (Note:
1058          * A "clone" child here is one that reports to its parent
1059          * using a signal other than SIGCHLD.) */
1060         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1061             && !(options & __WALL))
1062                 return 0;
1063         /*
1064          * Do not consider thread group leaders that are
1065          * in a non-empty thread group:
1066          */
1067         if (delay_group_leader(p))
1068                 return 2;
1069
1070         err = security_task_wait(p);
1071         if (err)
1072                 return err;
1073
1074         return 1;
1075 }
1076
1077 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1078                                int why, int status,
1079                                struct siginfo __user *infop,
1080                                struct rusage __user *rusagep)
1081 {
1082         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1083
1084         put_task_struct(p);
1085         if (!retval)
1086                 retval = put_user(SIGCHLD, &infop->si_signo);
1087         if (!retval)
1088                 retval = put_user(0, &infop->si_errno);
1089         if (!retval)
1090                 retval = put_user((short)why, &infop->si_code);
1091         if (!retval)
1092                 retval = put_user(pid, &infop->si_pid);
1093         if (!retval)
1094                 retval = put_user(uid, &infop->si_uid);
1095         if (!retval)
1096                 retval = put_user(status, &infop->si_status);
1097         if (!retval)
1098                 retval = pid;
1099         return retval;
1100 }
1101
1102 /*
1103  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1104  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1105  * the lock and this task is uninteresting.  If we return nonzero, we have
1106  * released the lock and the system call should return.
1107  */
1108 static int wait_task_zombie(struct task_struct *p, int noreap,
1109                             struct siginfo __user *infop,
1110                             int __user *stat_addr, struct rusage __user *ru)
1111 {
1112         unsigned long state;
1113         int retval;
1114         int status;
1115
1116         if (unlikely(noreap)) {
1117                 pid_t pid = p->pid;
1118                 uid_t uid = p->uid;
1119                 int exit_code = p->exit_code;
1120                 int why, status;
1121
1122                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1123                         return 0;
1124                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1125                         return 0;
1126                 get_task_struct(p);
1127                 read_unlock(&tasklist_lock);
1128                 if ((exit_code & 0x7f) == 0) {
1129                         why = CLD_EXITED;
1130                         status = exit_code >> 8;
1131                 } else {
1132                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1133                         status = exit_code & 0x7f;
1134                 }
1135                 return wait_noreap_copyout(p, pid, uid, why,
1136                                            status, infop, ru);
1137         }
1138
1139         /*
1140          * Try to move the task's state to DEAD
1141          * only one thread is allowed to do this:
1142          */
1143         state = xchg(&p->exit_state, EXIT_DEAD);
1144         if (state != EXIT_ZOMBIE) {
1145                 BUG_ON(state != EXIT_DEAD);
1146                 return 0;
1147         }
1148         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1149                 /*
1150                  * This can only happen in a race with a ptraced thread
1151                  * dying on another processor.
1152                  */
1153                 return 0;
1154         }
1155
1156         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1157                 struct signal_struct *psig;
1158                 struct signal_struct *sig;
1159
1160                 /*
1161                  * The resource counters for the group leader are in its
1162                  * own task_struct.  Those for dead threads in the group
1163                  * are in its signal_struct, as are those for the child
1164                  * processes it has previously reaped.  All these
1165                  * accumulate in the parent's signal_struct c* fields.
1166                  *
1167                  * We don't bother to take a lock here to protect these
1168                  * p->signal fields, because they are only touched by
1169                  * __exit_signal, which runs with tasklist_lock
1170                  * write-locked anyway, and so is excluded here.  We do
1171                  * need to protect the access to p->parent->signal fields,
1172                  * as other threads in the parent group can be right
1173                  * here reaping other children at the same time.
1174                  */
1175                 spin_lock_irq(&p->parent->sighand->siglock);
1176                 psig = p->parent->signal;
1177                 sig = p->signal;
1178                 psig->cutime =
1179                         cputime_add(psig->cutime,
1180                         cputime_add(p->utime,
1181                         cputime_add(sig->utime,
1182                                     sig->cutime)));
1183                 psig->cstime =
1184                         cputime_add(psig->cstime,
1185                         cputime_add(p->stime,
1186                         cputime_add(sig->stime,
1187                                     sig->cstime)));
1188                 psig->cmin_flt +=
1189                         p->min_flt + sig->min_flt + sig->cmin_flt;
1190                 psig->cmaj_flt +=
1191                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1192                 psig->cnvcsw +=
1193                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1194                 psig->cnivcsw +=
1195                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1196                 spin_unlock_irq(&p->parent->sighand->siglock);
1197         }
1198
1199         /*
1200          * Now we are sure this task is interesting, and no other
1201          * thread can reap it because we set its state to EXIT_DEAD.
1202          */
1203         read_unlock(&tasklist_lock);
1204
1205         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1206         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1207                 ? p->signal->group_exit_code : p->exit_code;
1208         if (!retval && stat_addr)
1209                 retval = put_user(status, stat_addr);
1210         if (!retval && infop)
1211                 retval = put_user(SIGCHLD, &infop->si_signo);
1212         if (!retval && infop)
1213                 retval = put_user(0, &infop->si_errno);
1214         if (!retval && infop) {
1215                 int why;
1216
1217                 if ((status & 0x7f) == 0) {
1218                         why = CLD_EXITED;
1219                         status >>= 8;
1220                 } else {
1221                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1222                         status &= 0x7f;
1223                 }
1224                 retval = put_user((short)why, &infop->si_code);
1225                 if (!retval)
1226                         retval = put_user(status, &infop->si_status);
1227         }
1228         if (!retval && infop)
1229                 retval = put_user(p->pid, &infop->si_pid);
1230         if (!retval && infop)
1231                 retval = put_user(p->uid, &infop->si_uid);
1232         if (retval) {
1233                 // TODO: is this safe?
1234                 p->exit_state = EXIT_ZOMBIE;
1235                 return retval;
1236         }
1237         retval = p->pid;
1238         if (p->real_parent != p->parent) {
1239                 write_lock_irq(&tasklist_lock);
1240                 /* Double-check with lock held.  */
1241                 if (p->real_parent != p->parent) {
1242                         __ptrace_unlink(p);
1243                         // TODO: is this safe?
1244                         p->exit_state = EXIT_ZOMBIE;
1245                         /*
1246                          * If this is not a detached task, notify the parent.
1247                          * If it's still not detached after that, don't release
1248                          * it now.
1249                          */
1250                         if (p->exit_signal != -1) {
1251                                 do_notify_parent(p, p->exit_signal);
1252                                 if (p->exit_signal != -1)
1253                                         p = NULL;
1254                         }
1255                 }
1256                 write_unlock_irq(&tasklist_lock);
1257         }
1258         if (p != NULL)
1259                 release_task(p);
1260         BUG_ON(!retval);
1261         return retval;
1262 }
1263
1264 /*
1265  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1266  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1267  * the lock and this task is uninteresting.  If we return nonzero, we have
1268  * released the lock and the system call should return.
1269  */
1270 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1271                              int noreap, struct siginfo __user *infop,
1272                              int __user *stat_addr, struct rusage __user *ru)
1273 {
1274         int retval, exit_code;
1275
1276         if (!p->exit_code)
1277                 return 0;
1278         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1279             p->signal && p->signal->group_stop_count > 0)
1280                 /*
1281                  * A group stop is in progress and this is the group leader.
1282                  * We won't report until all threads have stopped.
1283                  */
1284                 return 0;
1285
1286         /*
1287          * Now we are pretty sure this task is interesting.
1288          * Make sure it doesn't get reaped out from under us while we
1289          * give up the lock and then examine it below.  We don't want to
1290          * keep holding onto the tasklist_lock while we call getrusage and
1291          * possibly take page faults for user memory.
1292          */
1293         get_task_struct(p);
1294         read_unlock(&tasklist_lock);
1295
1296         if (unlikely(noreap)) {
1297                 pid_t pid = p->pid;
1298                 uid_t uid = p->uid;
1299                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1300
1301                 exit_code = p->exit_code;
1302                 if (unlikely(!exit_code) ||
1303                     unlikely(p->state & TASK_TRACED))
1304                         goto bail_ref;
1305                 return wait_noreap_copyout(p, pid, uid,
1306                                            why, (exit_code << 8) | 0x7f,
1307                                            infop, ru);
1308         }
1309
1310         write_lock_irq(&tasklist_lock);
1311
1312         /*
1313          * This uses xchg to be atomic with the thread resuming and setting
1314          * it.  It must also be done with the write lock held to prevent a
1315          * race with the EXIT_ZOMBIE case.
1316          */
1317         exit_code = xchg(&p->exit_code, 0);
1318         if (unlikely(p->exit_state)) {
1319                 /*
1320                  * The task resumed and then died.  Let the next iteration
1321                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1322                  * already be zero here if it resumed and did _exit(0).
1323                  * The task itself is dead and won't touch exit_code again;
1324                  * other processors in this function are locked out.
1325                  */
1326                 p->exit_code = exit_code;
1327                 exit_code = 0;
1328         }
1329         if (unlikely(exit_code == 0)) {
1330                 /*
1331                  * Another thread in this function got to it first, or it
1332                  * resumed, or it resumed and then died.
1333                  */
1334                 write_unlock_irq(&tasklist_lock);
1335 bail_ref:
1336                 put_task_struct(p);
1337                 /*
1338                  * We are returning to the wait loop without having successfully
1339                  * removed the process and having released the lock. We cannot
1340                  * continue, since the "p" task pointer is potentially stale.
1341                  *
1342                  * Return -EAGAIN, and do_wait() will restart the loop from the
1343                  * beginning. Do _not_ re-acquire the lock.
1344                  */
1345                 return -EAGAIN;
1346         }
1347
1348         /* move to end of parent's list to avoid starvation */
1349         remove_parent(p);
1350         add_parent(p);
1351
1352         write_unlock_irq(&tasklist_lock);
1353
1354         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1355         if (!retval && stat_addr)
1356                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1357         if (!retval && infop)
1358                 retval = put_user(SIGCHLD, &infop->si_signo);
1359         if (!retval && infop)
1360                 retval = put_user(0, &infop->si_errno);
1361         if (!retval && infop)
1362                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1363                                           ? CLD_TRAPPED : CLD_STOPPED),
1364                                   &infop->si_code);
1365         if (!retval && infop)
1366                 retval = put_user(exit_code, &infop->si_status);
1367         if (!retval && infop)
1368                 retval = put_user(p->pid, &infop->si_pid);
1369         if (!retval && infop)
1370                 retval = put_user(p->uid, &infop->si_uid);
1371         if (!retval)
1372                 retval = p->pid;
1373         put_task_struct(p);
1374
1375         BUG_ON(!retval);
1376         return retval;
1377 }
1378
1379 /*
1380  * Handle do_wait work for one task in a live, non-stopped state.
1381  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1382  * the lock and this task is uninteresting.  If we return nonzero, we have
1383  * released the lock and the system call should return.
1384  */
1385 static int wait_task_continued(struct task_struct *p, int noreap,
1386                                struct siginfo __user *infop,
1387                                int __user *stat_addr, struct rusage __user *ru)
1388 {
1389         int retval;
1390         pid_t pid;
1391         uid_t uid;
1392
1393         if (unlikely(!p->signal))
1394                 return 0;
1395
1396         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1397                 return 0;
1398
1399         spin_lock_irq(&p->sighand->siglock);
1400         /* Re-check with the lock held.  */
1401         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1402                 spin_unlock_irq(&p->sighand->siglock);
1403                 return 0;
1404         }
1405         if (!noreap)
1406                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1407         spin_unlock_irq(&p->sighand->siglock);
1408
1409         pid = p->pid;
1410         uid = p->uid;
1411         get_task_struct(p);
1412         read_unlock(&tasklist_lock);
1413
1414         if (!infop) {
1415                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1416                 put_task_struct(p);
1417                 if (!retval && stat_addr)
1418                         retval = put_user(0xffff, stat_addr);
1419                 if (!retval)
1420                         retval = p->pid;
1421         } else {
1422                 retval = wait_noreap_copyout(p, pid, uid,
1423                                              CLD_CONTINUED, SIGCONT,
1424                                              infop, ru);
1425                 BUG_ON(retval == 0);
1426         }
1427
1428         return retval;
1429 }
1430
1431
1432 static inline int my_ptrace_child(struct task_struct *p)
1433 {
1434         if (!(p->ptrace & PT_PTRACED))
1435                 return 0;
1436         if (!(p->ptrace & PT_ATTACHED))
1437                 return 1;
1438         /*
1439          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1440          * we are the attacher.  If we are the real parent, this is a race
1441          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1442          * which we have to switch the parent links, but has already set
1443          * the flags in p->ptrace.
1444          */
1445         return (p->parent != p->real_parent);
1446 }
1447
1448 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1449                     int __user *stat_addr, struct rusage __user *ru)
1450 {
1451         DECLARE_WAITQUEUE(wait, current);
1452         struct task_struct *tsk;
1453         int flag, retval;
1454         int allowed, denied;
1455
1456         add_wait_queue(&current->signal->wait_chldexit,&wait);
1457 repeat:
1458         /*
1459          * We will set this flag if we see any child that might later
1460          * match our criteria, even if we are not able to reap it yet.
1461          */
1462         flag = 0;
1463         allowed = denied = 0;
1464         current->state = TASK_INTERRUPTIBLE;
1465         read_lock(&tasklist_lock);
1466         tsk = current;
1467         do {
1468                 struct task_struct *p;
1469                 struct list_head *_p;
1470                 int ret;
1471
1472                 list_for_each(_p,&tsk->children) {
1473                         p = list_entry(_p, struct task_struct, sibling);
1474
1475                         ret = eligible_child(pid, options, p);
1476                         if (!ret)
1477                                 continue;
1478
1479                         if (unlikely(ret < 0)) {
1480                                 denied = ret;
1481                                 continue;
1482                         }
1483                         allowed = 1;
1484
1485                         switch (p->state) {
1486                         case TASK_TRACED:
1487                                 /*
1488                                  * When we hit the race with PTRACE_ATTACH,
1489                                  * we will not report this child.  But the
1490                                  * race means it has not yet been moved to
1491                                  * our ptrace_children list, so we need to
1492                                  * set the flag here to avoid a spurious ECHILD
1493                                  * when the race happens with the only child.
1494                                  */
1495                                 flag = 1;
1496                                 if (!my_ptrace_child(p))
1497                                         continue;
1498                                 /*FALLTHROUGH*/
1499                         case TASK_STOPPED:
1500                                 /*
1501                                  * It's stopped now, so it might later
1502                                  * continue, exit, or stop again.
1503                                  */
1504                                 flag = 1;
1505                                 if (!(options & WUNTRACED) &&
1506                                     !my_ptrace_child(p))
1507                                         continue;
1508                                 retval = wait_task_stopped(p, ret == 2,
1509                                                            (options & WNOWAIT),
1510                                                            infop,
1511                                                            stat_addr, ru);
1512                                 if (retval == -EAGAIN)
1513                                         goto repeat;
1514                                 if (retval != 0) /* He released the lock.  */
1515                                         goto end;
1516                                 break;
1517                         default:
1518                         // case EXIT_DEAD:
1519                                 if (p->exit_state == EXIT_DEAD)
1520                                         continue;
1521                         // case EXIT_ZOMBIE:
1522                                 if (p->exit_state == EXIT_ZOMBIE) {
1523                                         /*
1524                                          * Eligible but we cannot release
1525                                          * it yet:
1526                                          */
1527                                         if (ret == 2)
1528                                                 goto check_continued;
1529                                         if (!likely(options & WEXITED))
1530                                                 continue;
1531                                         retval = wait_task_zombie(
1532                                                 p, (options & WNOWAIT),
1533                                                 infop, stat_addr, ru);
1534                                         /* He released the lock.  */
1535                                         if (retval != 0)
1536                                                 goto end;
1537                                         break;
1538                                 }
1539 check_continued:
1540                                 /*
1541                                  * It's running now, so it might later
1542                                  * exit, stop, or stop and then continue.
1543                                  */
1544                                 flag = 1;
1545                                 if (!unlikely(options & WCONTINUED))
1546                                         continue;
1547                                 retval = wait_task_continued(
1548                                         p, (options & WNOWAIT),
1549                                         infop, stat_addr, ru);
1550                                 if (retval != 0) /* He released the lock.  */
1551                                         goto end;
1552                                 break;
1553                         }
1554                 }
1555                 if (!flag) {
1556                         list_for_each(_p, &tsk->ptrace_children) {
1557                                 p = list_entry(_p, struct task_struct,
1558                                                 ptrace_list);
1559                                 if (!eligible_child(pid, options, p))
1560                                         continue;
1561                                 flag = 1;
1562                                 break;
1563                         }
1564                 }
1565                 if (options & __WNOTHREAD)
1566                         break;
1567                 tsk = next_thread(tsk);
1568                 BUG_ON(tsk->signal != current->signal);
1569         } while (tsk != current);
1570
1571         read_unlock(&tasklist_lock);
1572         if (flag) {
1573                 retval = 0;
1574                 if (options & WNOHANG)
1575                         goto end;
1576                 retval = -ERESTARTSYS;
1577                 if (signal_pending(current))
1578                         goto end;
1579                 schedule();
1580                 goto repeat;
1581         }
1582         retval = -ECHILD;
1583         if (unlikely(denied) && !allowed)
1584                 retval = denied;
1585 end:
1586         current->state = TASK_RUNNING;
1587         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1588         if (infop) {
1589                 if (retval > 0)
1590                 retval = 0;
1591                 else {
1592                         /*
1593                          * For a WNOHANG return, clear out all the fields
1594                          * we would set so the user can easily tell the
1595                          * difference.
1596                          */
1597                         if (!retval)
1598                                 retval = put_user(0, &infop->si_signo);
1599                         if (!retval)
1600                                 retval = put_user(0, &infop->si_errno);
1601                         if (!retval)
1602                                 retval = put_user(0, &infop->si_code);
1603                         if (!retval)
1604                                 retval = put_user(0, &infop->si_pid);
1605                         if (!retval)
1606                                 retval = put_user(0, &infop->si_uid);
1607                         if (!retval)
1608                                 retval = put_user(0, &infop->si_status);
1609                 }
1610         }
1611         return retval;
1612 }
1613
1614 asmlinkage long sys_waitid(int which, pid_t pid,
1615                            struct siginfo __user *infop, int options,
1616                            struct rusage __user *ru)
1617 {
1618         long ret;
1619
1620         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1621                 return -EINVAL;
1622         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1623                 return -EINVAL;
1624
1625         switch (which) {
1626         case P_ALL:
1627                 pid = -1;
1628                 break;
1629         case P_PID:
1630                 if (pid <= 0)
1631                         return -EINVAL;
1632                 break;
1633         case P_PGID:
1634                 if (pid <= 0)
1635                         return -EINVAL;
1636                 pid = -pid;
1637                 break;
1638         default:
1639                 return -EINVAL;
1640         }
1641
1642         ret = do_wait(pid, options, infop, NULL, ru);
1643
1644         /* avoid REGPARM breakage on x86: */
1645         prevent_tail_call(ret);
1646         return ret;
1647 }
1648
1649 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1650                           int options, struct rusage __user *ru)
1651 {
1652         long ret;
1653
1654         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1655                         __WNOTHREAD|__WCLONE|__WALL))
1656                 return -EINVAL;
1657         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1658
1659         /* avoid REGPARM breakage on x86: */
1660         prevent_tail_call(ret);
1661         return ret;
1662 }
1663
1664 #ifdef __ARCH_WANT_SYS_WAITPID
1665
1666 /*
1667  * sys_waitpid() remains for compatibility. waitpid() should be
1668  * implemented by calling sys_wait4() from libc.a.
1669  */
1670 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1671 {
1672         return sys_wait4(pid, stat_addr, options, NULL);
1673 }
1674
1675 #endif