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/signalfd.h>
  28 #include <linux/mount.h>
  29 #include <linux/proc_fs.h>
  30 #include <linux/kthread.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/taskstats_kern.h>
  33 #include <linux/delayacct.h>
  34 #include <linux/cpuset.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/signal.h>
  37 #include <linux/posix-timers.h>
  38 #include <linux/cn_proc.h>
  39 #include <linux/mutex.h>
  40 #include <linux/futex.h>
  41 #include <linux/compat.h>
  42 #include <linux/pipe_fs_i.h>
  43 #include <linux/audit.h> /* for audit_free() */
  44 #include <linux/resource.h>
  45 #include <linux/blkdev.h>
  46 #include <linux/task_io_accounting_ops.h>
  47
  48 #include <asm/uaccess.h>
  49 #include <asm/unistd.h>
  50 #include <asm/pgtable.h>
  51 #include <asm/mmu_context.h>
  52
  53 extern void sem_exit (void);
  54
  55 static void exit_mm(struct task_struct * tsk);
  56
  57 static void __unhash_process(struct task_struct *p)
  58 {
  59         nr_threads--;
  60         detach_pid(p, PIDTYPE_PID);
  61         if (thread_group_leader(p)) {
  62                 detach_pid(p, PIDTYPE_PGID);
  63                 detach_pid(p, PIDTYPE_SID);
  64
  65                 list_del_rcu(&p->tasks);
  66                 __get_cpu_var(process_counts)--;
  67         }
  68         list_del_rcu(&p->thread_group);
  69         remove_parent(p);
  70 }
  71
  72 /*
  73  * This function expects the tasklist_lock write-locked.
  74  */
  75 static void __exit_signal(struct task_struct *tsk)
  76 {
  77         struct signal_struct *sig = tsk->signal;
  78         struct sighand_struct *sighand;
  79
  80         BUG_ON(!sig);
  81         BUG_ON(!atomic_read(&sig->count));
  82
  83         rcu_read_lock();
  84         sighand = rcu_dereference(tsk->sighand);
  85         spin_lock(&sighand->siglock);
  86
  87         /*
  88          * Notify that this sighand has been detached. This must
  89          * be called with the tsk->sighand lock held. Also, this
  90          * access tsk->sighand internally, so it must be called
  91          * before tsk->sighand is reset.
  92          */
  93         signalfd_detach_locked(tsk);
  94
  95         posix_cpu_timers_exit(tsk);
  96         if (atomic_dec_and_test(&sig->count))
  97                 posix_cpu_timers_exit_group(tsk);
  98         else {
  99                 /*
 100                  * If there is any task waiting for the group exit
 101                  * then notify it:
 102                  */
 103                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
 104                         wake_up_process(sig->group_exit_task);
 105                         sig->group_exit_task = NULL;
 106                 }
 107                 if (tsk == sig->curr_target)
 108                         sig->curr_target = next_thread(tsk);
 109                 /*
 110                  * Accumulate here the counters for all threads but the
 111                  * group leader as they die, so they can be added into
 112                  * the process-wide totals when those are taken.
 113                  * The group leader stays around as a zombie as long
 114                  * as there are other threads.  When it gets reaped,
 115                  * the exit.c code will add its counts into these totals.
 116                  * We won't ever get here for the group leader, since it
 117                  * will have been the last reference on the signal_struct.
 118                  */
 119                 sig->utime = cputime_add(sig->utime, tsk->utime);
 120                 sig->stime = cputime_add(sig->stime, tsk->stime);
 121                 sig->min_flt += tsk->min_flt;
 122                 sig->maj_flt += tsk->maj_flt;
 123                 sig->nvcsw += tsk->nvcsw;
 124                 sig->nivcsw += tsk->nivcsw;
 125                 sig->inblock += task_io_get_inblock(tsk);
 126                 sig->oublock += task_io_get_oublock(tsk);
 127                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 128                 sig = NULL; /* Marker for below. */
 129         }
 130
 131         __unhash_process(tsk);
 132
 133         tsk->signal = NULL;
 134         tsk->sighand = NULL;
 135         spin_unlock(&sighand->siglock);
 136         rcu_read_unlock();
 137
 138         __cleanup_sighand(sighand);
 139         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 140         flush_sigqueue(&tsk->pending);
 141         if (sig) {
 142                 flush_sigqueue(&sig->shared_pending);
 143                 taskstats_tgid_free(sig);
 144                 __cleanup_signal(sig);
 145         }
 146 }
 147
 148 static void delayed_put_task_struct(struct rcu_head *rhp)
 149 {
 150         put_task_struct(container_of(rhp, struct task_struct, rcu));
 151 }
 152
 153 void release_task(struct task_struct * p)
 154 {
 155         struct task_struct *leader;
 156         int zap_leader;
 157 repeat:
 158         atomic_dec(&p->user->processes);
 159         write_lock_irq(&tasklist_lock);
 160         ptrace_unlink(p);
 161         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 162         __exit_signal(p);
 163
 164         /*
 165          * If we are the last non-leader member of the thread
 166          * group, and the leader is zombie, then notify the
 167          * group leader's parent process. (if it wants notification.)
 168          */
 169         zap_leader = 0;
 170         leader = p->group_leader;
 171         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 172                 BUG_ON(leader->exit_signal == -1);
 173                 do_notify_parent(leader, leader->exit_signal);
 174                 /*
 175                  * If we were the last child thread and the leader has
 176                  * exited already, and the leader's parent ignores SIGCHLD,
 177                  * then we are the one who should release the leader.
 178                  *
 179                  * do_notify_parent() will have marked it self-reaping in
 180                  * that case.
 181                  */
 182                 zap_leader = (leader->exit_signal == -1);
 183         }
 184
 185         write_unlock_irq(&tasklist_lock);
 186         proc_flush_task(p);
 187         release_thread(p);
 188         call_rcu(&p->rcu, delayed_put_task_struct);
 189
 190         p = leader;
 191         if (unlikely(zap_leader))
 192                 goto repeat;
 193 }
 194
 195 /*
 196  * This checks not only the pgrp, but falls back on the pid if no
 197  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 198  * without this...
 199  *
 200  * The caller must hold rcu lock or the tasklist lock.
 201  */
 202 struct pid *session_of_pgrp(struct pid *pgrp)
 203 {
 204         struct task_struct *p;
 205         struct pid *sid = NULL;
 206
 207         p = pid_task(pgrp, PIDTYPE_PGID);
 208         if (p == NULL)
 209                 p = pid_task(pgrp, PIDTYPE_PID);
 210         if (p != NULL)
 211                 sid = task_session(p);
 212
 213         return sid;
 214 }
 215
 216 /*
 217  * Determine if a process group is "orphaned", according to the POSIX
 218  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 219  * by terminal-generated stop signals.  Newly orphaned process groups are
 220  * to receive a SIGHUP and a SIGCONT.
 221  *
 222  * "I ask you, have you ever known what it is to be an orphan?"
 223  */
 224 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 225 {
 226         struct task_struct *p;
 227         int ret = 1;
 228
 229         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 230                 if (p == ignored_task
 231                                 || p->exit_state
 232                                 || is_init(p->real_parent))
 233                         continue;
 234                 if (task_pgrp(p->real_parent) != pgrp &&
 235                     task_session(p->real_parent) == task_session(p)) {
 236                         ret = 0;
 237                         break;
 238                 }
 239         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 240         return ret;     /* (sighing) "Often!" */
 241 }
 242
 243 int is_current_pgrp_orphaned(void)
 244 {
 245         int retval;
 246
 247         read_lock(&tasklist_lock);
 248         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 249         read_unlock(&tasklist_lock);
 250
 251         return retval;
 252 }
 253
 254 static int has_stopped_jobs(struct pid *pgrp)
 255 {
 256         int retval = 0;
 257         struct task_struct *p;
 258
 259         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 260                 if (p->state != TASK_STOPPED)
 261                         continue;
 262                 retval = 1;
 263                 break;
 264         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 265         return retval;
 266 }
 267
 268 /**
 269  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
 270  *
 271  * If a kernel thread is launched as a result of a system call, or if
 272  * it ever exits, it should generally reparent itself to kthreadd so it
 273  * isn't in the way of other processes and is correctly cleaned up on exit.
 274  *
 275  * The various task state such as scheduling policy and priority may have
 276  * been inherited from a user process, so we reset them to sane values here.
 277  *
 278  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
 279  */
 280 static void reparent_to_kthreadd(void)
 281 {
 282         write_lock_irq(&tasklist_lock);
 283
 284         ptrace_unlink(current);
 285         /* Reparent to init */
 286         remove_parent(current);
 287         current->real_parent = current->parent = kthreadd_task;
 288         add_parent(current);
 289
 290         /* Set the exit signal to SIGCHLD so we signal init on exit */
 291         current->exit_signal = SIGCHLD;
 292
 293         if (task_nice(current) < 0)
 294                 set_user_nice(current, 0);
 295         /* cpus_allowed? */
 296         /* rt_priority? */
 297         /* signals? */
 298         security_task_reparent_to_init(current);
 299         memcpy(current->signal->rlim, init_task.signal->rlim,
 300                sizeof(current->signal->rlim));
 301         atomic_inc(&(INIT_USER->__count));
 302         write_unlock_irq(&tasklist_lock);
 303         switch_uid(INIT_USER);
 304 }
 305
 306 void __set_special_pids(pid_t session, pid_t pgrp)
 307 {
 308         struct task_struct *curr = current->group_leader;
 309
 310         if (process_session(curr) != session) {
 311                 detach_pid(curr, PIDTYPE_SID);
 312                 set_signal_session(curr->signal, session);
 313                 attach_pid(curr, PIDTYPE_SID, find_pid(session));
 314         }
 315         if (process_group(curr) != pgrp) {
 316                 detach_pid(curr, PIDTYPE_PGID);
 317                 curr->signal->pgrp = pgrp;
 318                 attach_pid(curr, PIDTYPE_PGID, find_pid(pgrp));
 319         }
 320 }
 321
 322 static void set_special_pids(pid_t session, pid_t pgrp)
 323 {
 324         write_lock_irq(&tasklist_lock);
 325         __set_special_pids(session, pgrp);
 326         write_unlock_irq(&tasklist_lock);
 327 }
 328
 329 /*
 330  * Let kernel threads use this to say that they
 331  * allow a certain signal (since daemonize() will
 332  * have disabled all of them by default).
 333  */
 334 int allow_signal(int sig)
 335 {
 336         if (!valid_signal(sig) || sig < 1)
 337                 return -EINVAL;
 338
 339         spin_lock_irq(&current->sighand->siglock);
 340         sigdelset(&current->blocked, sig);
 341         if (!current->mm) {
 342                 /* Kernel threads handle their own signals.
 343                    Let the signal code know it'll be handled, so
 344                    that they don't get converted to SIGKILL or
 345                    just silently dropped */
 346                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 347         }
 348         recalc_sigpending();
 349         spin_unlock_irq(&current->sighand->siglock);
 350         return 0;
 351 }
 352
 353 EXPORT_SYMBOL(allow_signal);
 354
 355 int disallow_signal(int sig)
 356 {
 357         if (!valid_signal(sig) || sig < 1)
 358                 return -EINVAL;
 359
 360         spin_lock_irq(&current->sighand->siglock);
 361         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
 362         recalc_sigpending();
 363         spin_unlock_irq(&current->sighand->siglock);
 364         return 0;
 365 }
 366
 367 EXPORT_SYMBOL(disallow_signal);
 368
 369 /*
 370  *      Put all the gunge required to become a kernel thread without
 371  *      attached user resources in one place where it belongs.
 372  */
 373
 374 void daemonize(const char *name, ...)
 375 {
 376         va_list args;
 377         struct fs_struct *fs;
 378         sigset_t blocked;
 379
 380         va_start(args, name);
 381         vsnprintf(current->comm, sizeof(current->comm), name, args);
 382         va_end(args);
 383
 384         /*
 385          * If we were started as result of loading a module, close all of the
 386          * user space pages.  We don't need them, and if we didn't close them
 387          * they would be locked into memory.
 388          */
 389         exit_mm(current);
 390
 391         set_special_pids(1, 1);
 392         proc_clear_tty(current);
 393
 394         /* Block and flush all signals */
 395         sigfillset(&blocked);
 396         sigprocmask(SIG_BLOCK, &blocked, NULL);
 397         flush_signals(current);
 398
 399         /* Become as one with the init task */
 400
 401         exit_fs(current);       /* current->fs->count--; */
 402         fs = init_task.fs;
 403         current->fs = fs;
 404         atomic_inc(&fs->count);
 405
 406         exit_task_namespaces(current);
 407         current->nsproxy = init_task.nsproxy;
 408         get_task_namespaces(current);
 409
 410         exit_files(current);
 411         current->files = init_task.files;
 412         atomic_inc(&current->files->count);
 413
 414         reparent_to_kthreadd();
 415 }
 416
 417 EXPORT_SYMBOL(daemonize);
 418
 419 static void close_files(struct files_struct * files)
 420 {
 421         int i, j;
 422         struct fdtable *fdt;
 423
 424         j = 0;
 425
 426         /*
 427          * It is safe to dereference the fd table without RCU or
 428          * ->file_lock because this is the last reference to the
 429          * files structure.
 430          */
 431         fdt = files_fdtable(files);
 432         for (;;) {
 433                 unsigned long set;
 434                 i = j * __NFDBITS;
 435                 if (i >= fdt->max_fds)
 436                         break;
 437                 set = fdt->open_fds->fds_bits[j++];
 438                 while (set) {
 439                         if (set & 1) {
 440                                 struct file * file = xchg(&fdt->fd[i], NULL);
 441                                 if (file) {
 442                                         filp_close(file, files);
 443                                         cond_resched();
 444                                 }
 445                         }
 446                         i++;
 447                         set >>= 1;
 448                 }
 449         }
 450 }
 451
 452 struct files_struct *get_files_struct(struct task_struct *task)
 453 {
 454         struct files_struct *files;
 455
 456         task_lock(task);
 457         files = task->files;
 458         if (files)
 459                 atomic_inc(&files->count);
 460         task_unlock(task);
 461
 462         return files;
 463 }
 464
 465 void fastcall put_files_struct(struct files_struct *files)
 466 {
 467         struct fdtable *fdt;
 468
 469         if (atomic_dec_and_test(&files->count)) {
 470                 close_files(files);
 471                 /*
 472                  * Free the fd and fdset arrays if we expanded them.
 473                  * If the fdtable was embedded, pass files for freeing
 474                  * at the end of the RCU grace period. Otherwise,
 475                  * you can free files immediately.
 476                  */
 477                 fdt = files_fdtable(files);
 478                 if (fdt != &files->fdtab)
 479                         kmem_cache_free(files_cachep, files);
 480                 free_fdtable(fdt);
 481         }
 482 }
 483
 484 EXPORT_SYMBOL(put_files_struct);
 485
 486 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
 487 {
 488         struct files_struct *old;
 489
 490         old = tsk->files;
 491         task_lock(tsk);
 492         tsk->files = files;
 493         task_unlock(tsk);
 494         put_files_struct(old);
 495 }
 496 EXPORT_SYMBOL(reset_files_struct);
 497
 498 static inline void __exit_files(struct task_struct *tsk)
 499 {
 500         struct files_struct * files = tsk->files;
 501
 502         if (files) {
 503                 task_lock(tsk);
 504                 tsk->files = NULL;
 505                 task_unlock(tsk);
 506                 put_files_struct(files);
 507         }
 508 }
 509
 510 void exit_files(struct task_struct *tsk)
 511 {
 512         __exit_files(tsk);
 513 }
 514
 515 static inline void __put_fs_struct(struct fs_struct *fs)
 516 {
 517         /* No need to hold fs->lock if we are killing it */
 518         if (atomic_dec_and_test(&fs->count)) {
 519                 dput(fs->root);
 520                 mntput(fs->rootmnt);
 521                 dput(fs->pwd);
 522                 mntput(fs->pwdmnt);
 523                 if (fs->altroot) {
 524                         dput(fs->altroot);
 525                         mntput(fs->altrootmnt);
 526                 }
 527                 kmem_cache_free(fs_cachep, fs);
 528         }
 529 }
 530
 531 void put_fs_struct(struct fs_struct *fs)
 532 {
 533         __put_fs_struct(fs);
 534 }
 535
 536 static inline void __exit_fs(struct task_struct *tsk)
 537 {
 538         struct fs_struct * fs = tsk->fs;
 539
 540         if (fs) {
 541                 task_lock(tsk);
 542                 tsk->fs = NULL;
 543                 task_unlock(tsk);
 544                 __put_fs_struct(fs);
 545         }
 546 }
 547
 548 void exit_fs(struct task_struct *tsk)
 549 {
 550         __exit_fs(tsk);
 551 }
 552
 553 EXPORT_SYMBOL_GPL(exit_fs);
 554
 555 /*
 556  * Turn us into a lazy TLB process if we
 557  * aren't already..
 558  */
 559 static void exit_mm(struct task_struct * tsk)
 560 {
 561         struct mm_struct *mm = tsk->mm;
 562
 563         mm_release(tsk, mm);
 564         if (!mm)
 565                 return;
 566         /*
 567          * Serialize with any possible pending coredump.
 568          * We must hold mmap_sem around checking core_waiters
 569          * and clearing tsk->mm.  The core-inducing thread
 570          * will increment core_waiters for each thread in the
 571          * group with ->mm != NULL.
 572          */
 573         down_read(&mm->mmap_sem);
 574         if (mm->core_waiters) {
 575                 up_read(&mm->mmap_sem);
 576                 down_write(&mm->mmap_sem);
 577                 if (!--mm->core_waiters)
 578                         complete(mm->core_startup_done);
 579                 up_write(&mm->mmap_sem);
 580
 581                 wait_for_completion(&mm->core_done);
 582                 down_read(&mm->mmap_sem);
 583         }
 584         atomic_inc(&mm->mm_count);
 585         BUG_ON(mm != tsk->active_mm);
 586         /* more a memory barrier than a real lock */
 587         task_lock(tsk);
 588         tsk->mm = NULL;
 589         up_read(&mm->mmap_sem);
 590         enter_lazy_tlb(mm, current);
 591         task_unlock(tsk);
 592         mmput(mm);
 593 }
 594
 595 static inline void
 596 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 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         if (p->pdeath_signal)
 610                 /* We already hold the tasklist_lock here.  */
 611                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 612
 613         /* Move the child from its dying parent to the new one.  */
 614         if (unlikely(traced)) {
 615                 /* Preserve ptrace links if someone else is tracing this child.  */
 616                 list_del_init(&p->ptrace_list);
 617                 if (p->parent != p->real_parent)
 618                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 619         } else {
 620                 /* If this child is being traced, then we're the one tracing it
 621                  * anyway, so let go of it.
 622                  */
 623                 p->ptrace = 0;
 624                 remove_parent(p);
 625                 p->parent = p->real_parent;
 626                 add_parent(p);
 627
 628                 if (p->state == TASK_TRACED) {
 629                         /*
 630                          * If it was at a trace stop, turn it into
 631                          * a normal stop since it's no longer being
 632                          * traced.
 633                          */
 634                         ptrace_untrace(p);
 635                 }
 636         }
 637
 638         /* If this is a threaded reparent there is no need to
 639          * notify anyone anything has happened.
 640          */
 641         if (p->real_parent->group_leader == father->group_leader)
 642                 return;
 643
 644         /* We don't want people slaying init.  */
 645         if (p->exit_signal != -1)
 646                 p->exit_signal = SIGCHLD;
 647
 648         /* If we'd notified the old parent about this child's death,
 649          * also notify the new parent.
 650          */
 651         if (!traced && p->exit_state == EXIT_ZOMBIE &&
 652             p->exit_signal != -1 && thread_group_empty(p))
 653                 do_notify_parent(p, p->exit_signal);
 654
 655         /*
 656          * process group orphan check
 657          * Case ii: Our child is in a different pgrp
 658          * than we are, and it was the only connection
 659          * outside, so the child pgrp is now orphaned.
 660          */
 661         if ((task_pgrp(p) != task_pgrp(father)) &&
 662             (task_session(p) == task_session(father))) {
 663                 struct pid *pgrp = task_pgrp(p);
 664
 665                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
 666                     has_stopped_jobs(pgrp)) {
 667                         __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 668                         __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 669                 }
 670         }
 671 }
 672
 673 /*
 674  * When we die, we re-parent all our children.
 675  * Try to give them to another thread in our thread
 676  * group, and if no such member exists, give it to
 677  * the child reaper process (ie "init") in our pid
 678  * space.
 679  */
 680 static void
 681 forget_original_parent(struct task_struct *father, struct list_head *to_release)
 682 {
 683         struct task_struct *p, *reaper = father;
 684         struct list_head *_p, *_n;
 685
 686         do {
 687                 reaper = next_thread(reaper);
 688                 if (reaper == father) {
 689                         reaper = child_reaper(father);
 690                         break;
 691                 }
 692         } while (reaper->exit_state);
 693
 694         /*
 695          * There are only two places where our children can be:
 696          *
 697          * - in our child list
 698          * - in our ptraced child list
 699          *
 700          * Search them and reparent children.
 701          */
 702         list_for_each_safe(_p, _n, &father->children) {
 703                 int ptrace;
 704                 p = list_entry(_p, struct task_struct, sibling);
 705
 706                 ptrace = p->ptrace;
 707
 708                 /* if father isn't the real parent, then ptrace must be enabled */
 709                 BUG_ON(father != p->real_parent && !ptrace);
 710
 711                 if (father == p->real_parent) {
 712                         /* reparent with a reaper, real father it's us */
 713                         choose_new_parent(p, reaper);
 714                         reparent_thread(p, father, 0);
 715                 } else {
 716                         /* reparent ptraced task to its real parent */
 717                         __ptrace_unlink (p);
 718                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 719                             thread_group_empty(p))
 720                                 do_notify_parent(p, p->exit_signal);
 721                 }
 722
 723                 /*
 724                  * if the ptraced child is a zombie with exit_signal == -1
 725                  * we must collect it before we exit, or it will remain
 726                  * zombie forever since we prevented it from self-reap itself
 727                  * while it was being traced by us, to be able to see it in wait4.
 728                  */
 729                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 730                         list_add(&p->ptrace_list, to_release);
 731         }
 732         list_for_each_safe(_p, _n, &father->ptrace_children) {
 733                 p = list_entry(_p, struct task_struct, ptrace_list);
 734                 choose_new_parent(p, reaper);
 735                 reparent_thread(p, father, 1);
 736         }
 737 }
 738
 739 /*
 740  * Send signals to all our closest relatives so that they know
 741  * to properly mourn us..
 742  */
 743 static void exit_notify(struct task_struct *tsk)
 744 {
 745         int state;
 746         struct task_struct *t;
 747         struct list_head ptrace_dead, *_p, *_n;
 748         struct pid *pgrp;
 749
 750         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 751             && !thread_group_empty(tsk)) {
 752                 /*
 753                  * This occurs when there was a race between our exit
 754                  * syscall and a group signal choosing us as the one to
 755                  * wake up.  It could be that we are the only thread
 756                  * alerted to check for pending signals, but another thread
 757                  * should be woken now to take the signal since we will not.
 758                  * Now we'll wake all the threads in the group just to make
 759                  * sure someone gets all the pending signals.
 760                  */
 761                 read_lock(&tasklist_lock);
 762                 spin_lock_irq(&tsk->sighand->siglock);
 763                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 764                         if (!signal_pending(t) && !(t->flags & PF_EXITING))
 765                                 recalc_sigpending_and_wake(t);
 766                 spin_unlock_irq(&tsk->sighand->siglock);
 767                 read_unlock(&tasklist_lock);
 768         }
 769
 770         write_lock_irq(&tasklist_lock);
 771
 772         /*
 773          * This does two things:
 774          *
 775          * A.  Make init inherit all the child processes
 776          * B.  Check to see if any process groups have become orphaned
 777          *      as a result of our exiting, and if they have any stopped
 778          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 779          */
 780
 781         INIT_LIST_HEAD(&ptrace_dead);
 782         forget_original_parent(tsk, &ptrace_dead);
 783         BUG_ON(!list_empty(&tsk->children));
 784         BUG_ON(!list_empty(&tsk->ptrace_children));
 785
 786         /*
 787          * Check to see if any process groups have become orphaned
 788          * as a result of our exiting, and if they have any stopped
 789          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 790          *
 791          * Case i: Our father is in a different pgrp than we are
 792          * and we were the only connection outside, so our pgrp
 793          * is about to become orphaned.
 794          */
 795
 796         t = tsk->real_parent;
 797
 798         pgrp = task_pgrp(tsk);
 799         if ((task_pgrp(t) != pgrp) &&
 800             (task_session(t) == task_session(tsk)) &&
 801             will_become_orphaned_pgrp(pgrp, tsk) &&
 802             has_stopped_jobs(pgrp)) {
 803                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 804                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 805         }
 806
 807         /* Let father know we died
 808          *
 809          * Thread signals are configurable, but you aren't going to use
 810          * that to send signals to arbitary processes.
 811          * That stops right now.
 812          *
 813          * If the parent exec id doesn't match the exec id we saved
 814          * when we started then we know the parent has changed security
 815          * domain.
 816          *
 817          * If our self_exec id doesn't match our parent_exec_id then
 818          * we have changed execution domain as these two values started
 819          * the same after a fork.
 820          *
 821          */
 822
 823         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 824             ( tsk->parent_exec_id != t->self_exec_id  ||
 825               tsk->self_exec_id != tsk->parent_exec_id)
 826             && !capable(CAP_KILL))
 827                 tsk->exit_signal = SIGCHLD;
 828
 829
 830         /* If something other than our normal parent is ptracing us, then
 831          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 832          * only has special meaning to our real parent.
 833          */
 834         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 835                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 836                 do_notify_parent(tsk, signal);
 837         } else if (tsk->ptrace) {
 838                 do_notify_parent(tsk, SIGCHLD);
 839         }
 840
 841         state = EXIT_ZOMBIE;
 842         if (tsk->exit_signal == -1 &&
 843             (likely(tsk->ptrace == 0) ||
 844              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 845                 state = EXIT_DEAD;
 846         tsk->exit_state = state;
 847
 848         write_unlock_irq(&tasklist_lock);
 849
 850         list_for_each_safe(_p, _n, &ptrace_dead) {
 851                 list_del_init(_p);
 852                 t = list_entry(_p, struct task_struct, ptrace_list);
 853                 release_task(t);
 854         }
 855
 856         /* If the process is dead, release it - nobody will wait for it */
 857         if (state == EXIT_DEAD)
 858                 release_task(tsk);
 859 }
 860
 861 fastcall NORET_TYPE void do_exit(long code)
 862 {
 863         struct task_struct *tsk = current;
 864         int group_dead;
 865
 866         profile_task_exit(tsk);
 867
 868         WARN_ON(atomic_read(&tsk->fs_excl));
 869
 870         if (unlikely(in_interrupt()))
 871                 panic("Aiee, killing interrupt handler!");
 872         if (unlikely(!tsk->pid))
 873                 panic("Attempted to kill the idle task!");
 874         if (unlikely(tsk == child_reaper(tsk))) {
 875                 if (tsk->nsproxy->pid_ns != &init_pid_ns)
 876                         tsk->nsproxy->pid_ns->child_reaper = init_pid_ns.child_reaper;
 877                 else
 878                         panic("Attempted to kill init!");
 879         }
 880
 881
 882         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 883                 current->ptrace_message = code;
 884                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 885         }
 886
 887         /*
 888          * We're taking recursive faults here in do_exit. Safest is to just
 889          * leave this task alone and wait for reboot.
 890          */
 891         if (unlikely(tsk->flags & PF_EXITING)) {
 892                 printk(KERN_ALERT
 893                         "Fixing recursive fault but reboot is needed!\n");
 894                 /*
 895                  * We can do this unlocked here. The futex code uses
 896                  * this flag just to verify whether the pi state
 897                  * cleanup has been done or not. In the worst case it
 898                  * loops once more. We pretend that the cleanup was
 899                  * done as there is no way to return. Either the
 900                  * OWNER_DIED bit is set by now or we push the blocked
 901                  * task into the wait for ever nirwana as well.
 902                  */
 903                 tsk->flags |= PF_EXITPIDONE;
 904                 if (tsk->io_context)
 905                         exit_io_context();
 906                 set_current_state(TASK_UNINTERRUPTIBLE);
 907                 schedule();
 908         }
 909
 910         /*
 911          * tsk->flags are checked in the futex code to protect against
 912          * an exiting task cleaning up the robust pi futexes.
 913          */
 914         spin_lock_irq(&tsk->pi_lock);
 915         tsk->flags |= PF_EXITING;
 916         spin_unlock_irq(&tsk->pi_lock);
 917
 918         if (unlikely(in_atomic()))
 919                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 920                                 current->comm, current->pid,
 921                                 preempt_count());
 922
 923         acct_update_integrals(tsk);
 924         if (tsk->mm) {
 925                 update_hiwater_rss(tsk->mm);
 926                 update_hiwater_vm(tsk->mm);
 927         }
 928         group_dead = atomic_dec_and_test(&tsk->signal->live);
 929         if (group_dead) {
 930                 hrtimer_cancel(&tsk->signal->real_timer);
 931                 exit_itimers(tsk->signal);
 932         }
 933         acct_collect(code, group_dead);
 934         if (unlikely(tsk->robust_list))
 935                 exit_robust_list(tsk);
 936 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 937         if (unlikely(tsk->compat_robust_list))
 938                 compat_exit_robust_list(tsk);
 939 #endif
 940         if (unlikely(tsk->audit_context))
 941                 audit_free(tsk);
 942
 943         taskstats_exit(tsk, group_dead);
 944
 945         exit_mm(tsk);
 946
 947         if (group_dead)
 948                 acct_process();
 949         exit_sem(tsk);
 950         __exit_files(tsk);
 951         __exit_fs(tsk);
 952         exit_thread();
 953         cpuset_exit(tsk);
 954         exit_keys(tsk);
 955
 956         if (group_dead && tsk->signal->leader)
 957                 disassociate_ctty(1);
 958
 959         module_put(task_thread_info(tsk)->exec_domain->module);
 960         if (tsk->binfmt)
 961                 module_put(tsk->binfmt->module);
 962
 963         tsk->exit_code = code;
 964         proc_exit_connector(tsk);
 965         exit_task_namespaces(tsk);
 966         exit_notify(tsk);
 967 #ifdef CONFIG_NUMA
 968         mpol_free(tsk->mempolicy);
 969         tsk->mempolicy = NULL;
 970 #endif
 971         /*
 972          * This must happen late, after the PID is not
 973          * hashed anymore:
 974          */
 975         if (unlikely(!list_empty(&tsk->pi_state_list)))
 976                 exit_pi_state_list(tsk);
 977         if (unlikely(current->pi_state_cache))
 978                 kfree(current->pi_state_cache);
 979         /*
 980          * Make sure we are holding no locks:
 981          */
 982         debug_check_no_locks_held(tsk);
 983         /*
 984          * We can do this unlocked here. The futex code uses this flag
 985          * just to verify whether the pi state cleanup has been done
 986          * or not. In the worst case it loops once more.
 987          */
 988         tsk->flags |= PF_EXITPIDONE;
 989
 990         if (tsk->io_context)
 991                 exit_io_context();
 992
 993         if (tsk->splice_pipe)
 994                 __free_pipe_info(tsk->splice_pipe);
 995
 996         preempt_disable();
 997         /* causes final put_task_struct in finish_task_switch(). */
 998         tsk->state = TASK_DEAD;
 999
1000         schedule();
1001         BUG();
1002         /* Avoid "noreturn function does return".  */
1003         for (;;)
1004                 cpu_relax();    /* For when BUG is null */
1005 }
1006
1007 EXPORT_SYMBOL_GPL(do_exit);
1008
1009 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1010 {
1011         if (comp)
1012                 complete(comp);
1013
1014         do_exit(code);
1015 }
1016
1017 EXPORT_SYMBOL(complete_and_exit);
1018
1019 asmlinkage long sys_exit(int error_code)
1020 {
1021         do_exit((error_code&0xff)<<8);
1022 }
1023
1024 /*
1025  * Take down every thread in the group.  This is called by fatal signals
1026  * as well as by sys_exit_group (below).
1027  */
1028 NORET_TYPE void
1029 do_group_exit(int exit_code)
1030 {
1031         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1032
1033         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1034                 exit_code = current->signal->group_exit_code;
1035         else if (!thread_group_empty(current)) {
1036                 struct signal_struct *const sig = current->signal;
1037                 struct sighand_struct *const sighand = current->sighand;
1038                 spin_lock_irq(&sighand->siglock);
1039                 if (sig->flags & SIGNAL_GROUP_EXIT)
1040                         /* Another thread got here before we took the lock.  */
1041                         exit_code = sig->group_exit_code;
1042                 else {
1043                         sig->group_exit_code = exit_code;
1044                         zap_other_threads(current);
1045                 }
1046                 spin_unlock_irq(&sighand->siglock);
1047         }
1048
1049         do_exit(exit_code);
1050         /* NOTREACHED */
1051 }
1052
1053 /*
1054  * this kills every thread in the thread group. Note that any externally
1055  * wait4()-ing process will get the correct exit code - even if this
1056  * thread is not the thread group leader.
1057  */
1058 asmlinkage void sys_exit_group(int error_code)
1059 {
1060         do_group_exit((error_code & 0xff) << 8);
1061 }
1062
1063 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1064 {
1065         int err;
1066
1067         if (pid > 0) {
1068                 if (p->pid != pid)
1069                         return 0;
1070         } else if (!pid) {
1071                 if (process_group(p) != process_group(current))
1072                         return 0;
1073         } else if (pid != -1) {
1074                 if (process_group(p) != -pid)
1075                         return 0;
1076         }
1077
1078         /*
1079          * Do not consider detached threads that are
1080          * not ptraced:
1081          */
1082         if (p->exit_signal == -1 && !p->ptrace)
1083                 return 0;
1084
1085         /* Wait for all children (clone and not) if __WALL is set;
1086          * otherwise, wait for clone children *only* if __WCLONE is
1087          * set; otherwise, wait for non-clone children *only*.  (Note:
1088          * A "clone" child here is one that reports to its parent
1089          * using a signal other than SIGCHLD.) */
1090         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1091             && !(options & __WALL))
1092                 return 0;
1093         /*
1094          * Do not consider thread group leaders that are
1095          * in a non-empty thread group:
1096          */
1097         if (delay_group_leader(p))
1098                 return 2;
1099
1100         err = security_task_wait(p);
1101         if (err)
1102                 return err;
1103
1104         return 1;
1105 }
1106
1107 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1108                                int why, int status,
1109                                struct siginfo __user *infop,
1110                                struct rusage __user *rusagep)
1111 {
1112         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1113
1114         put_task_struct(p);
1115         if (!retval)
1116                 retval = put_user(SIGCHLD, &infop->si_signo);
1117         if (!retval)
1118                 retval = put_user(0, &infop->si_errno);
1119         if (!retval)
1120                 retval = put_user((short)why, &infop->si_code);
1121         if (!retval)
1122                 retval = put_user(pid, &infop->si_pid);
1123         if (!retval)
1124                 retval = put_user(uid, &infop->si_uid);
1125         if (!retval)
1126                 retval = put_user(status, &infop->si_status);
1127         if (!retval)
1128                 retval = pid;
1129         return retval;
1130 }
1131
1132 /*
1133  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1134  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1135  * the lock and this task is uninteresting.  If we return nonzero, we have
1136  * released the lock and the system call should return.
1137  */
1138 static int wait_task_zombie(struct task_struct *p, int noreap,
1139                             struct siginfo __user *infop,
1140                             int __user *stat_addr, struct rusage __user *ru)
1141 {
1142         unsigned long state;
1143         int retval;
1144         int status;
1145
1146         if (unlikely(noreap)) {
1147                 pid_t pid = p->pid;
1148                 uid_t uid = p->uid;
1149                 int exit_code = p->exit_code;
1150                 int why, status;
1151
1152                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1153                         return 0;
1154                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1155                         return 0;
1156                 get_task_struct(p);
1157                 read_unlock(&tasklist_lock);
1158                 if ((exit_code & 0x7f) == 0) {
1159                         why = CLD_EXITED;
1160                         status = exit_code >> 8;
1161                 } else {
1162                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1163                         status = exit_code & 0x7f;
1164                 }
1165                 return wait_noreap_copyout(p, pid, uid, why,
1166                                            status, infop, ru);
1167         }
1168
1169         /*
1170          * Try to move the task's state to DEAD
1171          * only one thread is allowed to do this:
1172          */
1173         state = xchg(&p->exit_state, EXIT_DEAD);
1174         if (state != EXIT_ZOMBIE) {
1175                 BUG_ON(state != EXIT_DEAD);
1176                 return 0;
1177         }
1178         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1179                 /*
1180                  * This can only happen in a race with a ptraced thread
1181                  * dying on another processor.
1182                  */
1183                 return 0;
1184         }
1185
1186         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1187                 struct signal_struct *psig;
1188                 struct signal_struct *sig;
1189
1190                 /*
1191                  * The resource counters for the group leader are in its
1192                  * own task_struct.  Those for dead threads in the group
1193                  * are in its signal_struct, as are those for the child
1194                  * processes it has previously reaped.  All these
1195                  * accumulate in the parent's signal_struct c* fields.
1196                  *
1197                  * We don't bother to take a lock here to protect these
1198                  * p->signal fields, because they are only touched by
1199                  * __exit_signal, which runs with tasklist_lock
1200                  * write-locked anyway, and so is excluded here.  We do
1201                  * need to protect the access to p->parent->signal fields,
1202                  * as other threads in the parent group can be right
1203                  * here reaping other children at the same time.
1204                  */
1205                 spin_lock_irq(&p->parent->sighand->siglock);
1206                 psig = p->parent->signal;
1207                 sig = p->signal;
1208                 psig->cutime =
1209                         cputime_add(psig->cutime,
1210                         cputime_add(p->utime,
1211                         cputime_add(sig->utime,
1212                                     sig->cutime)));
1213                 psig->cstime =
1214                         cputime_add(psig->cstime,
1215                         cputime_add(p->stime,
1216                         cputime_add(sig->stime,
1217                                     sig->cstime)));
1218                 psig->cmin_flt +=
1219                         p->min_flt + sig->min_flt + sig->cmin_flt;
1220                 psig->cmaj_flt +=
1221                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1222                 psig->cnvcsw +=
1223                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1224                 psig->cnivcsw +=
1225                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1226                 psig->cinblock +=
1227                         task_io_get_inblock(p) +
1228                         sig->inblock + sig->cinblock;
1229                 psig->coublock +=
1230                         task_io_get_oublock(p) +
1231                         sig->oublock + sig->coublock;
1232                 spin_unlock_irq(&p->parent->sighand->siglock);
1233         }
1234
1235         /*
1236          * Now we are sure this task is interesting, and no other
1237          * thread can reap it because we set its state to EXIT_DEAD.
1238          */
1239         read_unlock(&tasklist_lock);
1240
1241         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1242         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1243                 ? p->signal->group_exit_code : p->exit_code;
1244         if (!retval && stat_addr)
1245                 retval = put_user(status, stat_addr);
1246         if (!retval && infop)
1247                 retval = put_user(SIGCHLD, &infop->si_signo);
1248         if (!retval && infop)
1249                 retval = put_user(0, &infop->si_errno);
1250         if (!retval && infop) {
1251                 int why;
1252
1253                 if ((status & 0x7f) == 0) {
1254                         why = CLD_EXITED;
1255                         status >>= 8;
1256                 } else {
1257                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1258                         status &= 0x7f;
1259                 }
1260                 retval = put_user((short)why, &infop->si_code);
1261                 if (!retval)
1262                         retval = put_user(status, &infop->si_status);
1263         }
1264         if (!retval && infop)
1265                 retval = put_user(p->pid, &infop->si_pid);
1266         if (!retval && infop)
1267                 retval = put_user(p->uid, &infop->si_uid);
1268         if (retval) {
1269                 // TODO: is this safe?
1270                 p->exit_state = EXIT_ZOMBIE;
1271                 return retval;
1272         }
1273         retval = p->pid;
1274         if (p->real_parent != p->parent) {
1275                 write_lock_irq(&tasklist_lock);
1276                 /* Double-check with lock held.  */
1277                 if (p->real_parent != p->parent) {
1278                         __ptrace_unlink(p);
1279                         // TODO: is this safe?
1280                         p->exit_state = EXIT_ZOMBIE;
1281                         /*
1282                          * If this is not a detached task, notify the parent.
1283                          * If it's still not detached after that, don't release
1284                          * it now.
1285                          */
1286                         if (p->exit_signal != -1) {
1287                                 do_notify_parent(p, p->exit_signal);
1288                                 if (p->exit_signal != -1)
1289                                         p = NULL;
1290                         }
1291                 }
1292                 write_unlock_irq(&tasklist_lock);
1293         }
1294         if (p != NULL)
1295                 release_task(p);
1296         BUG_ON(!retval);
1297         return retval;
1298 }
1299
1300 /*
1301  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1302  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1303  * the lock and this task is uninteresting.  If we return nonzero, we have
1304  * released the lock and the system call should return.
1305  */
1306 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1307                              int noreap, struct siginfo __user *infop,
1308                              int __user *stat_addr, struct rusage __user *ru)
1309 {
1310         int retval, exit_code;
1311
1312         if (!p->exit_code)
1313                 return 0;
1314         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1315             p->signal && p->signal->group_stop_count > 0)
1316                 /*
1317                  * A group stop is in progress and this is the group leader.
1318                  * We won't report until all threads have stopped.
1319                  */
1320                 return 0;
1321
1322         /*
1323          * Now we are pretty sure this task is interesting.
1324          * Make sure it doesn't get reaped out from under us while we
1325          * give up the lock and then examine it below.  We don't want to
1326          * keep holding onto the tasklist_lock while we call getrusage and
1327          * possibly take page faults for user memory.
1328          */
1329         get_task_struct(p);
1330         read_unlock(&tasklist_lock);
1331
1332         if (unlikely(noreap)) {
1333                 pid_t pid = p->pid;
1334                 uid_t uid = p->uid;
1335                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1336
1337                 exit_code = p->exit_code;
1338                 if (unlikely(!exit_code) ||
1339                     unlikely(p->state & TASK_TRACED))
1340                         goto bail_ref;
1341                 return wait_noreap_copyout(p, pid, uid,
1342                                            why, (exit_code << 8) | 0x7f,
1343                                            infop, ru);
1344         }
1345
1346         write_lock_irq(&tasklist_lock);
1347
1348         /*
1349          * This uses xchg to be atomic with the thread resuming and setting
1350          * it.  It must also be done with the write lock held to prevent a
1351          * race with the EXIT_ZOMBIE case.
1352          */
1353         exit_code = xchg(&p->exit_code, 0);
1354         if (unlikely(p->exit_state)) {
1355                 /*
1356                  * The task resumed and then died.  Let the next iteration
1357                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1358                  * already be zero here if it resumed and did _exit(0).
1359                  * The task itself is dead and won't touch exit_code again;
1360                  * other processors in this function are locked out.
1361                  */
1362                 p->exit_code = exit_code;
1363                 exit_code = 0;
1364         }
1365         if (unlikely(exit_code == 0)) {
1366                 /*
1367                  * Another thread in this function got to it first, or it
1368                  * resumed, or it resumed and then died.
1369                  */
1370                 write_unlock_irq(&tasklist_lock);
1371 bail_ref:
1372                 put_task_struct(p);
1373                 /*
1374                  * We are returning to the wait loop without having successfully
1375                  * removed the process and having released the lock. We cannot
1376                  * continue, since the "p" task pointer is potentially stale.
1377                  *
1378                  * Return -EAGAIN, and do_wait() will restart the loop from the
1379                  * beginning. Do _not_ re-acquire the lock.
1380                  */
1381                 return -EAGAIN;
1382         }
1383
1384         /* move to end of parent's list to avoid starvation */
1385         remove_parent(p);
1386         add_parent(p);
1387
1388         write_unlock_irq(&tasklist_lock);
1389
1390         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1391         if (!retval && stat_addr)
1392                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1393         if (!retval && infop)
1394                 retval = put_user(SIGCHLD, &infop->si_signo);
1395         if (!retval && infop)
1396                 retval = put_user(0, &infop->si_errno);
1397         if (!retval && infop)
1398                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1399                                           ? CLD_TRAPPED : CLD_STOPPED),
1400                                   &infop->si_code);
1401         if (!retval && infop)
1402                 retval = put_user(exit_code, &infop->si_status);
1403         if (!retval && infop)
1404                 retval = put_user(p->pid, &infop->si_pid);
1405         if (!retval && infop)
1406                 retval = put_user(p->uid, &infop->si_uid);
1407         if (!retval)
1408                 retval = p->pid;
1409         put_task_struct(p);
1410
1411         BUG_ON(!retval);
1412         return retval;
1413 }
1414
1415 /*
1416  * Handle do_wait work for one task in a live, non-stopped state.
1417  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1418  * the lock and this task is uninteresting.  If we return nonzero, we have
1419  * released the lock and the system call should return.
1420  */
1421 static int wait_task_continued(struct task_struct *p, int noreap,
1422                                struct siginfo __user *infop,
1423                                int __user *stat_addr, struct rusage __user *ru)
1424 {
1425         int retval;
1426         pid_t pid;
1427         uid_t uid;
1428
1429         if (unlikely(!p->signal))
1430                 return 0;
1431
1432         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1433                 return 0;
1434
1435         spin_lock_irq(&p->sighand->siglock);
1436         /* Re-check with the lock held.  */
1437         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1438                 spin_unlock_irq(&p->sighand->siglock);
1439                 return 0;
1440         }
1441         if (!noreap)
1442                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1443         spin_unlock_irq(&p->sighand->siglock);
1444
1445         pid = p->pid;
1446         uid = p->uid;
1447         get_task_struct(p);
1448         read_unlock(&tasklist_lock);
1449
1450         if (!infop) {
1451                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1452                 put_task_struct(p);
1453                 if (!retval && stat_addr)
1454                         retval = put_user(0xffff, stat_addr);
1455                 if (!retval)
1456                         retval = p->pid;
1457         } else {
1458                 retval = wait_noreap_copyout(p, pid, uid,
1459                                              CLD_CONTINUED, SIGCONT,
1460                                              infop, ru);
1461                 BUG_ON(retval == 0);
1462         }
1463
1464         return retval;
1465 }
1466
1467
1468 static inline int my_ptrace_child(struct task_struct *p)
1469 {
1470         if (!(p->ptrace & PT_PTRACED))
1471                 return 0;
1472         if (!(p->ptrace & PT_ATTACHED))
1473                 return 1;
1474         /*
1475          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1476          * we are the attacher.  If we are the real parent, this is a race
1477          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1478          * which we have to switch the parent links, but has already set
1479          * the flags in p->ptrace.
1480          */
1481         return (p->parent != p->real_parent);
1482 }
1483
1484 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1485                     int __user *stat_addr, struct rusage __user *ru)
1486 {
1487         DECLARE_WAITQUEUE(wait, current);
1488         struct task_struct *tsk;
1489         int flag, retval;
1490         int allowed, denied;
1491
1492         add_wait_queue(&current->signal->wait_chldexit,&wait);
1493 repeat:
1494         /*
1495          * We will set this flag if we see any child that might later
1496          * match our criteria, even if we are not able to reap it yet.
1497          */
1498         flag = 0;
1499         allowed = denied = 0;
1500         current->state = TASK_INTERRUPTIBLE;
1501         read_lock(&tasklist_lock);
1502         tsk = current;
1503         do {
1504                 struct task_struct *p;
1505                 struct list_head *_p;
1506                 int ret;
1507
1508                 list_for_each(_p,&tsk->children) {
1509                         p = list_entry(_p, struct task_struct, sibling);
1510
1511                         ret = eligible_child(pid, options, p);
1512                         if (!ret)
1513                                 continue;
1514
1515                         if (unlikely(ret < 0)) {
1516                                 denied = ret;
1517                                 continue;
1518                         }
1519                         allowed = 1;
1520
1521                         switch (p->state) {
1522                         case TASK_TRACED:
1523                                 /*
1524                                  * When we hit the race with PTRACE_ATTACH,
1525                                  * we will not report this child.  But the
1526                                  * race means it has not yet been moved to
1527                                  * our ptrace_children list, so we need to
1528                                  * set the flag here to avoid a spurious ECHILD
1529                                  * when the race happens with the only child.
1530                                  */
1531                                 flag = 1;
1532                                 if (!my_ptrace_child(p))
1533                                         continue;
1534                                 /*FALLTHROUGH*/
1535                         case TASK_STOPPED:
1536                                 /*
1537                                  * It's stopped now, so it might later
1538                                  * continue, exit, or stop again.
1539                                  */
1540                                 flag = 1;
1541                                 if (!(options & WUNTRACED) &&
1542                                     !my_ptrace_child(p))
1543                                         continue;
1544                                 retval = wait_task_stopped(p, ret == 2,
1545                                                            (options & WNOWAIT),
1546                                                            infop,
1547                                                            stat_addr, ru);
1548                                 if (retval == -EAGAIN)
1549                                         goto repeat;
1550                                 if (retval != 0) /* He released the lock.  */
1551                                         goto end;
1552                                 break;
1553                         default:
1554                         // case EXIT_DEAD:
1555                                 if (p->exit_state == EXIT_DEAD)
1556                                         continue;
1557                         // case EXIT_ZOMBIE:
1558                                 if (p->exit_state == EXIT_ZOMBIE) {
1559                                         /*
1560                                          * Eligible but we cannot release
1561                                          * it yet:
1562                                          */
1563                                         if (ret == 2)
1564                                                 goto check_continued;
1565                                         if (!likely(options & WEXITED))
1566                                                 continue;
1567                                         retval = wait_task_zombie(
1568                                                 p, (options & WNOWAIT),
1569                                                 infop, stat_addr, ru);
1570                                         /* He released the lock.  */
1571                                         if (retval != 0)
1572                                                 goto end;
1573                                         break;
1574                                 }
1575 check_continued:
1576                                 /*
1577                                  * It's running now, so it might later
1578                                  * exit, stop, or stop and then continue.
1579                                  */
1580                                 flag = 1;
1581                                 if (!unlikely(options & WCONTINUED))
1582                                         continue;
1583                                 retval = wait_task_continued(
1584                                         p, (options & WNOWAIT),
1585                                         infop, stat_addr, ru);
1586                                 if (retval != 0) /* He released the lock.  */
1587                                         goto end;
1588                                 break;
1589                         }
1590                 }
1591                 if (!flag) {
1592                         list_for_each(_p, &tsk->ptrace_children) {
1593                                 p = list_entry(_p, struct task_struct,
1594                                                 ptrace_list);
1595                                 if (!eligible_child(pid, options, p))
1596                                         continue;
1597                                 flag = 1;
1598                                 break;
1599                         }
1600                 }
1601                 if (options & __WNOTHREAD)
1602                         break;
1603                 tsk = next_thread(tsk);
1604                 BUG_ON(tsk->signal != current->signal);
1605         } while (tsk != current);
1606
1607         read_unlock(&tasklist_lock);
1608         if (flag) {
1609                 retval = 0;
1610                 if (options & WNOHANG)
1611                         goto end;
1612                 retval = -ERESTARTSYS;
1613                 if (signal_pending(current))
1614                         goto end;
1615                 schedule();
1616                 goto repeat;
1617         }
1618         retval = -ECHILD;
1619         if (unlikely(denied) && !allowed)
1620                 retval = denied;
1621 end:
1622         current->state = TASK_RUNNING;
1623         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1624         if (infop) {
1625                 if (retval > 0)
1626                 retval = 0;
1627                 else {
1628                         /*
1629                          * For a WNOHANG return, clear out all the fields
1630                          * we would set so the user can easily tell the
1631                          * difference.
1632                          */
1633                         if (!retval)
1634                                 retval = put_user(0, &infop->si_signo);
1635                         if (!retval)
1636                                 retval = put_user(0, &infop->si_errno);
1637                         if (!retval)
1638                                 retval = put_user(0, &infop->si_code);
1639                         if (!retval)
1640                                 retval = put_user(0, &infop->si_pid);
1641                         if (!retval)
1642                                 retval = put_user(0, &infop->si_uid);
1643                         if (!retval)
1644                                 retval = put_user(0, &infop->si_status);
1645                 }
1646         }
1647         return retval;
1648 }
1649
1650 asmlinkage long sys_waitid(int which, pid_t pid,
1651                            struct siginfo __user *infop, int options,
1652                            struct rusage __user *ru)
1653 {
1654         long ret;
1655
1656         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1657                 return -EINVAL;
1658         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1659                 return -EINVAL;
1660
1661         switch (which) {
1662         case P_ALL:
1663                 pid = -1;
1664                 break;
1665         case P_PID:
1666                 if (pid <= 0)
1667                         return -EINVAL;
1668                 break;
1669         case P_PGID:
1670                 if (pid <= 0)
1671                         return -EINVAL;
1672                 pid = -pid;
1673                 break;
1674         default:
1675                 return -EINVAL;
1676         }
1677
1678         ret = do_wait(pid, options, infop, NULL, ru);
1679
1680         /* avoid REGPARM breakage on x86: */
1681         prevent_tail_call(ret);
1682         return ret;
1683 }
1684
1685 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1686                           int options, struct rusage __user *ru)
1687 {
1688         long ret;
1689
1690         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1691                         __WNOTHREAD|__WCLONE|__WALL))
1692                 return -EINVAL;
1693         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1694
1695         /* avoid REGPARM breakage on x86: */
1696         prevent_tail_call(ret);
1697         return ret;
1698 }
1699
1700 #ifdef __ARCH_WANT_SYS_WAITPID
1701
1702 /*
1703  * sys_waitpid() remains for compatibility. waitpid() should be
1704  * implemented by calling sys_wait4() from libc.a.
1705  */
1706 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1707 {
1708         return sys_wait4(pid, stat_addr, options, NULL);
1709 }
1710
1711 #endif