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/freezer.h>
  35 #include <linux/cpuset.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/signal.h>
  38 #include <linux/posix-timers.h>
  39 #include <linux/cn_proc.h>
  40 #include <linux/mutex.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/pipe_fs_i.h>
  44 #include <linux/audit.h> /* for audit_free() */
  45 #include <linux/resource.h>
  46 #include <linux/blkdev.h>
  47 #include <linux/task_io_accounting_ops.h>
  48 #include <linux/freezer.h>
  49
  50 #include <asm/uaccess.h>
  51 #include <asm/unistd.h>
  52 #include <asm/pgtable.h>
  53 #include <asm/mmu_context.h>
  54
  55 extern void sem_exit (void);
  56
  57 static void exit_mm(struct task_struct * tsk);
  58
  59 static void __unhash_process(struct task_struct *p)
  60 {
  61         nr_threads--;
  62         detach_pid(p, PIDTYPE_PID);
  63         if (thread_group_leader(p)) {
  64                 detach_pid(p, PIDTYPE_PGID);
  65                 detach_pid(p, PIDTYPE_SID);
  66
  67                 list_del_rcu(&p->tasks);
  68                 __get_cpu_var(process_counts)--;
  69         }
  70         list_del_rcu(&p->thread_group);
  71         remove_parent(p);
  72 }
  73
  74 /*
  75  * This function expects the tasklist_lock write-locked.
  76  */
  77 static void __exit_signal(struct task_struct *tsk)
  78 {
  79         struct signal_struct *sig = tsk->signal;
  80         struct sighand_struct *sighand;
  81
  82         BUG_ON(!sig);
  83         BUG_ON(!atomic_read(&sig->count));
  84
  85         rcu_read_lock();
  86         sighand = rcu_dereference(tsk->sighand);
  87         spin_lock(&sighand->siglock);
  88
  89         /*
  90          * Notify that this sighand has been detached. This must
  91          * be called with the tsk->sighand lock held. Also, this
  92          * access tsk->sighand internally, so it must be called
  93          * before tsk->sighand is reset.
  94          */
  95         signalfd_detach_locked(tsk);
  96
  97         posix_cpu_timers_exit(tsk);
  98         if (atomic_dec_and_test(&sig->count))
  99                 posix_cpu_timers_exit_group(tsk);
 100         else {
 101                 /*
 102                  * If there is any task waiting for the group exit
 103                  * then notify it:
 104                  */
 105                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
 106                         wake_up_process(sig->group_exit_task);
 107                         sig->group_exit_task = NULL;
 108                 }
 109                 if (tsk == sig->curr_target)
 110                         sig->curr_target = next_thread(tsk);
 111                 /*
 112                  * Accumulate here the counters for all threads but the
 113                  * group leader as they die, so they can be added into
 114                  * the process-wide totals when those are taken.
 115                  * The group leader stays around as a zombie as long
 116                  * as there are other threads.  When it gets reaped,
 117                  * the exit.c code will add its counts into these totals.
 118                  * We won't ever get here for the group leader, since it
 119                  * will have been the last reference on the signal_struct.
 120                  */
 121                 sig->utime = cputime_add(sig->utime, tsk->utime);
 122                 sig->stime = cputime_add(sig->stime, tsk->stime);
 123                 sig->min_flt += tsk->min_flt;
 124                 sig->maj_flt += tsk->maj_flt;
 125                 sig->nvcsw += tsk->nvcsw;
 126                 sig->nivcsw += tsk->nivcsw;
 127                 sig->inblock += task_io_get_inblock(tsk);
 128                 sig->oublock += task_io_get_oublock(tsk);
 129                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 130                 sig = NULL; /* Marker for below. */
 131         }
 132
 133         __unhash_process(tsk);
 134
 135         tsk->signal = NULL;
 136         tsk->sighand = NULL;
 137         spin_unlock(&sighand->siglock);
 138         rcu_read_unlock();
 139
 140         __cleanup_sighand(sighand);
 141         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 142         flush_sigqueue(&tsk->pending);
 143         if (sig) {
 144                 flush_sigqueue(&sig->shared_pending);
 145                 taskstats_tgid_free(sig);
 146                 __cleanup_signal(sig);
 147         }
 148 }
 149
 150 static void delayed_put_task_struct(struct rcu_head *rhp)
 151 {
 152         put_task_struct(container_of(rhp, struct task_struct, rcu));
 153 }
 154
 155 void release_task(struct task_struct * p)
 156 {
 157         struct task_struct *leader;
 158         int zap_leader;
 159 repeat:
 160         atomic_dec(&p->user->processes);
 161         write_lock_irq(&tasklist_lock);
 162         ptrace_unlink(p);
 163         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 164         __exit_signal(p);
 165
 166         /*
 167          * If we are the last non-leader member of the thread
 168          * group, and the leader is zombie, then notify the
 169          * group leader's parent process. (if it wants notification.)
 170          */
 171         zap_leader = 0;
 172         leader = p->group_leader;
 173         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 174                 BUG_ON(leader->exit_signal == -1);
 175                 do_notify_parent(leader, leader->exit_signal);
 176                 /*
 177                  * If we were the last child thread and the leader has
 178                  * exited already, and the leader's parent ignores SIGCHLD,
 179                  * then we are the one who should release the leader.
 180                  *
 181                  * do_notify_parent() will have marked it self-reaping in
 182                  * that case.
 183                  */
 184                 zap_leader = (leader->exit_signal == -1);
 185         }
 186
 187         write_unlock_irq(&tasklist_lock);
 188         proc_flush_task(p);
 189         release_thread(p);
 190         call_rcu(&p->rcu, delayed_put_task_struct);
 191
 192         p = leader;
 193         if (unlikely(zap_leader))
 194                 goto repeat;
 195 }
 196
 197 /*
 198  * This checks not only the pgrp, but falls back on the pid if no
 199  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 200  * without this...
 201  *
 202  * The caller must hold rcu lock or the tasklist lock.
 203  */
 204 struct pid *session_of_pgrp(struct pid *pgrp)
 205 {
 206         struct task_struct *p;
 207         struct pid *sid = NULL;
 208
 209         p = pid_task(pgrp, PIDTYPE_PGID);
 210         if (p == NULL)
 211                 p = pid_task(pgrp, PIDTYPE_PID);
 212         if (p != NULL)
 213                 sid = task_session(p);
 214
 215         return sid;
 216 }
 217
 218 /*
 219  * Determine if a process group is "orphaned", according to the POSIX
 220  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 221  * by terminal-generated stop signals.  Newly orphaned process groups are
 222  * to receive a SIGHUP and a SIGCONT.
 223  *
 224  * "I ask you, have you ever known what it is to be an orphan?"
 225  */
 226 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 227 {
 228         struct task_struct *p;
 229         int ret = 1;
 230
 231         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 232                 if (p == ignored_task
 233                                 || p->exit_state
 234                                 || is_init(p->real_parent))
 235                         continue;
 236                 if (task_pgrp(p->real_parent) != pgrp &&
 237                     task_session(p->real_parent) == task_session(p)) {
 238                         ret = 0;
 239                         break;
 240                 }
 241         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 242         return ret;     /* (sighing) "Often!" */
 243 }
 244
 245 int is_current_pgrp_orphaned(void)
 246 {
 247         int retval;
 248
 249         read_lock(&tasklist_lock);
 250         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 251         read_unlock(&tasklist_lock);
 252
 253         return retval;
 254 }
 255
 256 static int has_stopped_jobs(struct pid *pgrp)
 257 {
 258         int retval = 0;
 259         struct task_struct *p;
 260
 261         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 262                 if (p->state != TASK_STOPPED)
 263                         continue;
 264                 retval = 1;
 265                 break;
 266         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 267         return retval;
 268 }
 269
 270 /**
 271  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
 272  *
 273  * If a kernel thread is launched as a result of a system call, or if
 274  * it ever exits, it should generally reparent itself to kthreadd so it
 275  * isn't in the way of other processes and is correctly cleaned up on exit.
 276  *
 277  * The various task state such as scheduling policy and priority may have
 278  * been inherited from a user process, so we reset them to sane values here.
 279  *
 280  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
 281  */
 282 static void reparent_to_kthreadd(void)
 283 {
 284         write_lock_irq(&tasklist_lock);
 285
 286         ptrace_unlink(current);
 287         /* Reparent to init */
 288         remove_parent(current);
 289         current->real_parent = current->parent = kthreadd_task;
 290         add_parent(current);
 291
 292         /* Set the exit signal to SIGCHLD so we signal init on exit */
 293         current->exit_signal = SIGCHLD;
 294
 295         if (task_nice(current) < 0)
 296                 set_user_nice(current, 0);
 297         /* cpus_allowed? */
 298         /* rt_priority? */
 299         /* signals? */
 300         security_task_reparent_to_init(current);
 301         memcpy(current->signal->rlim, init_task.signal->rlim,
 302                sizeof(current->signal->rlim));
 303         atomic_inc(&(INIT_USER->__count));
 304         write_unlock_irq(&tasklist_lock);
 305         switch_uid(INIT_USER);
 306 }
 307
 308 void __set_special_pids(pid_t session, pid_t pgrp)
 309 {
 310         struct task_struct *curr = current->group_leader;
 311
 312         if (process_session(curr) != session) {
 313                 detach_pid(curr, PIDTYPE_SID);
 314                 set_signal_session(curr->signal, session);
 315                 attach_pid(curr, PIDTYPE_SID, find_pid(session));
 316         }
 317         if (process_group(curr) != pgrp) {
 318                 detach_pid(curr, PIDTYPE_PGID);
 319                 curr->signal->pgrp = pgrp;
 320                 attach_pid(curr, PIDTYPE_PGID, find_pid(pgrp));
 321         }
 322 }
 323
 324 static void set_special_pids(pid_t session, pid_t pgrp)
 325 {
 326         write_lock_irq(&tasklist_lock);
 327         __set_special_pids(session, pgrp);
 328         write_unlock_irq(&tasklist_lock);
 329 }
 330
 331 /*
 332  * Let kernel threads use this to say that they
 333  * allow a certain signal (since daemonize() will
 334  * have disabled all of them by default).
 335  */
 336 int allow_signal(int sig)
 337 {
 338         if (!valid_signal(sig) || sig < 1)
 339                 return -EINVAL;
 340
 341         spin_lock_irq(&current->sighand->siglock);
 342         sigdelset(&current->blocked, sig);
 343         if (!current->mm) {
 344                 /* Kernel threads handle their own signals.
 345                    Let the signal code know it'll be handled, so
 346                    that they don't get converted to SIGKILL or
 347                    just silently dropped */
 348                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 349         }
 350         recalc_sigpending();
 351         spin_unlock_irq(&current->sighand->siglock);
 352         return 0;
 353 }
 354
 355 EXPORT_SYMBOL(allow_signal);
 356
 357 int disallow_signal(int sig)
 358 {
 359         if (!valid_signal(sig) || sig < 1)
 360                 return -EINVAL;
 361
 362         spin_lock_irq(&current->sighand->siglock);
 363         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
 364         recalc_sigpending();
 365         spin_unlock_irq(&current->sighand->siglock);
 366         return 0;
 367 }
 368
 369 EXPORT_SYMBOL(disallow_signal);
 370
 371 /*
 372  *      Put all the gunge required to become a kernel thread without
 373  *      attached user resources in one place where it belongs.
 374  */
 375
 376 void daemonize(const char *name, ...)
 377 {
 378         va_list args;
 379         struct fs_struct *fs;
 380         sigset_t blocked;
 381
 382         va_start(args, name);
 383         vsnprintf(current->comm, sizeof(current->comm), name, args);
 384         va_end(args);
 385
 386         /*
 387          * If we were started as result of loading a module, close all of the
 388          * user space pages.  We don't need them, and if we didn't close them
 389          * they would be locked into memory.
 390          */
 391         exit_mm(current);
 392         /*
 393          * We don't want to have TIF_FREEZE set if the system-wide hibernation
 394          * or suspend transition begins right now.
 395          */
 396         current->flags |= PF_NOFREEZE;
 397
 398         set_special_pids(1, 1);
 399         proc_clear_tty(current);
 400
 401         /* Block and flush all signals */
 402         sigfillset(&blocked);
 403         sigprocmask(SIG_BLOCK, &blocked, NULL);
 404         flush_signals(current);
 405
 406         /* Become as one with the init task */
 407
 408         exit_fs(current);       /* current->fs->count--; */
 409         fs = init_task.fs;
 410         current->fs = fs;
 411         atomic_inc(&fs->count);
 412
 413         exit_task_namespaces(current);
 414         current->nsproxy = init_task.nsproxy;
 415         get_task_namespaces(current);
 416
 417         exit_files(current);
 418         current->files = init_task.files;
 419         atomic_inc(&current->files->count);
 420
 421         reparent_to_kthreadd();
 422 }
 423
 424 EXPORT_SYMBOL(daemonize);
 425
 426 static void close_files(struct files_struct * files)
 427 {
 428         int i, j;
 429         struct fdtable *fdt;
 430
 431         j = 0;
 432
 433         /*
 434          * It is safe to dereference the fd table without RCU or
 435          * ->file_lock because this is the last reference to the
 436          * files structure.
 437          */
 438         fdt = files_fdtable(files);
 439         for (;;) {
 440                 unsigned long set;
 441                 i = j * __NFDBITS;
 442                 if (i >= fdt->max_fds)
 443                         break;
 444                 set = fdt->open_fds->fds_bits[j++];
 445                 while (set) {
 446                         if (set & 1) {
 447                                 struct file * file = xchg(&fdt->fd[i], NULL);
 448                                 if (file) {
 449                                         filp_close(file, files);
 450                                         cond_resched();
 451                                 }
 452                         }
 453                         i++;
 454                         set >>= 1;
 455                 }
 456         }
 457 }
 458
 459 struct files_struct *get_files_struct(struct task_struct *task)
 460 {
 461         struct files_struct *files;
 462
 463         task_lock(task);
 464         files = task->files;
 465         if (files)
 466                 atomic_inc(&files->count);
 467         task_unlock(task);
 468
 469         return files;
 470 }
 471
 472 void fastcall put_files_struct(struct files_struct *files)
 473 {
 474         struct fdtable *fdt;
 475
 476         if (atomic_dec_and_test(&files->count)) {
 477                 close_files(files);
 478                 /*
 479                  * Free the fd and fdset arrays if we expanded them.
 480                  * If the fdtable was embedded, pass files for freeing
 481                  * at the end of the RCU grace period. Otherwise,
 482                  * you can free files immediately.
 483                  */
 484                 fdt = files_fdtable(files);
 485                 if (fdt != &files->fdtab)
 486                         kmem_cache_free(files_cachep, files);
 487                 free_fdtable(fdt);
 488         }
 489 }
 490
 491 EXPORT_SYMBOL(put_files_struct);
 492
 493 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
 494 {
 495         struct files_struct *old;
 496
 497         old = tsk->files;
 498         task_lock(tsk);
 499         tsk->files = files;
 500         task_unlock(tsk);
 501         put_files_struct(old);
 502 }
 503 EXPORT_SYMBOL(reset_files_struct);
 504
 505 static inline void __exit_files(struct task_struct *tsk)
 506 {
 507         struct files_struct * files = tsk->files;
 508
 509         if (files) {
 510                 task_lock(tsk);
 511                 tsk->files = NULL;
 512                 task_unlock(tsk);
 513                 put_files_struct(files);
 514         }
 515 }
 516
 517 void exit_files(struct task_struct *tsk)
 518 {
 519         __exit_files(tsk);
 520 }
 521
 522 static inline void __put_fs_struct(struct fs_struct *fs)
 523 {
 524         /* No need to hold fs->lock if we are killing it */
 525         if (atomic_dec_and_test(&fs->count)) {
 526                 dput(fs->root);
 527                 mntput(fs->rootmnt);
 528                 dput(fs->pwd);
 529                 mntput(fs->pwdmnt);
 530                 if (fs->altroot) {
 531                         dput(fs->altroot);
 532                         mntput(fs->altrootmnt);
 533                 }
 534                 kmem_cache_free(fs_cachep, fs);
 535         }
 536 }
 537
 538 void put_fs_struct(struct fs_struct *fs)
 539 {
 540         __put_fs_struct(fs);
 541 }
 542
 543 static inline void __exit_fs(struct task_struct *tsk)
 544 {
 545         struct fs_struct * fs = tsk->fs;
 546
 547         if (fs) {
 548                 task_lock(tsk);
 549                 tsk->fs = NULL;
 550                 task_unlock(tsk);
 551                 __put_fs_struct(fs);
 552         }
 553 }
 554
 555 void exit_fs(struct task_struct *tsk)
 556 {
 557         __exit_fs(tsk);
 558 }
 559
 560 EXPORT_SYMBOL_GPL(exit_fs);
 561
 562 /*
 563  * Turn us into a lazy TLB process if we
 564  * aren't already..
 565  */
 566 static void exit_mm(struct task_struct * tsk)
 567 {
 568         struct mm_struct *mm = tsk->mm;
 569
 570         mm_release(tsk, mm);
 571         if (!mm)
 572                 return;
 573         /*
 574          * Serialize with any possible pending coredump.
 575          * We must hold mmap_sem around checking core_waiters
 576          * and clearing tsk->mm.  The core-inducing thread
 577          * will increment core_waiters for each thread in the
 578          * group with ->mm != NULL.
 579          */
 580         down_read(&mm->mmap_sem);
 581         if (mm->core_waiters) {
 582                 up_read(&mm->mmap_sem);
 583                 down_write(&mm->mmap_sem);
 584                 if (!--mm->core_waiters)
 585                         complete(mm->core_startup_done);
 586                 up_write(&mm->mmap_sem);
 587
 588                 wait_for_completion(&mm->core_done);
 589                 down_read(&mm->mmap_sem);
 590         }
 591         atomic_inc(&mm->mm_count);
 592         BUG_ON(mm != tsk->active_mm);
 593         /* more a memory barrier than a real lock */
 594         task_lock(tsk);
 595         tsk->mm = NULL;
 596         up_read(&mm->mmap_sem);
 597         enter_lazy_tlb(mm, current);
 598         /* We don't want this task to be frozen prematurely */
 599         clear_freeze_flag(tsk);
 600         task_unlock(tsk);
 601         mmput(mm);
 602 }
 603
 604 static inline void
 605 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 606 {
 607         /*
 608          * Make sure we're not reparenting to ourselves and that
 609          * the parent is not a zombie.
 610          */
 611         BUG_ON(p == reaper || reaper->exit_state);
 612         p->real_parent = reaper;
 613 }
 614
 615 static void
 616 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 617 {
 618         if (p->pdeath_signal)
 619                 /* We already hold the tasklist_lock here.  */
 620                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 621
 622         /* Move the child from its dying parent to the new one.  */
 623         if (unlikely(traced)) {
 624                 /* Preserve ptrace links if someone else is tracing this child.  */
 625                 list_del_init(&p->ptrace_list);
 626                 if (p->parent != p->real_parent)
 627                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 628         } else {
 629                 /* If this child is being traced, then we're the one tracing it
 630                  * anyway, so let go of it.
 631                  */
 632                 p->ptrace = 0;
 633                 remove_parent(p);
 634                 p->parent = p->real_parent;
 635                 add_parent(p);
 636
 637                 if (p->state == TASK_TRACED) {
 638                         /*
 639                          * If it was at a trace stop, turn it into
 640                          * a normal stop since it's no longer being
 641                          * traced.
 642                          */
 643                         ptrace_untrace(p);
 644                 }
 645         }
 646
 647         /* If this is a threaded reparent there is no need to
 648          * notify anyone anything has happened.
 649          */
 650         if (p->real_parent->group_leader == father->group_leader)
 651                 return;
 652
 653         /* We don't want people slaying init.  */
 654         if (p->exit_signal != -1)
 655                 p->exit_signal = SIGCHLD;
 656
 657         /* If we'd notified the old parent about this child's death,
 658          * also notify the new parent.
 659          */
 660         if (!traced && p->exit_state == EXIT_ZOMBIE &&
 661             p->exit_signal != -1 && thread_group_empty(p))
 662                 do_notify_parent(p, p->exit_signal);
 663
 664         /*
 665          * process group orphan check
 666          * Case ii: Our child is in a different pgrp
 667          * than we are, and it was the only connection
 668          * outside, so the child pgrp is now orphaned.
 669          */
 670         if ((task_pgrp(p) != task_pgrp(father)) &&
 671             (task_session(p) == task_session(father))) {
 672                 struct pid *pgrp = task_pgrp(p);
 673
 674                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
 675                     has_stopped_jobs(pgrp)) {
 676                         __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 677                         __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 678                 }
 679         }
 680 }
 681
 682 /*
 683  * When we die, we re-parent all our children.
 684  * Try to give them to another thread in our thread
 685  * group, and if no such member exists, give it to
 686  * the child reaper process (ie "init") in our pid
 687  * space.
 688  */
 689 static void
 690 forget_original_parent(struct task_struct *father, struct list_head *to_release)
 691 {
 692         struct task_struct *p, *reaper = father;
 693         struct list_head *_p, *_n;
 694
 695         do {
 696                 reaper = next_thread(reaper);
 697                 if (reaper == father) {
 698                         reaper = child_reaper(father);
 699                         break;
 700                 }
 701         } while (reaper->exit_state);
 702
 703         /*
 704          * There are only two places where our children can be:
 705          *
 706          * - in our child list
 707          * - in our ptraced child list
 708          *
 709          * Search them and reparent children.
 710          */
 711         list_for_each_safe(_p, _n, &father->children) {
 712                 int ptrace;
 713                 p = list_entry(_p, struct task_struct, sibling);
 714
 715                 ptrace = p->ptrace;
 716
 717                 /* if father isn't the real parent, then ptrace must be enabled */
 718                 BUG_ON(father != p->real_parent && !ptrace);
 719
 720                 if (father == p->real_parent) {
 721                         /* reparent with a reaper, real father it's us */
 722                         choose_new_parent(p, reaper);
 723                         reparent_thread(p, father, 0);
 724                 } else {
 725                         /* reparent ptraced task to its real parent */
 726                         __ptrace_unlink (p);
 727                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 728                             thread_group_empty(p))
 729                                 do_notify_parent(p, p->exit_signal);
 730                 }
 731
 732                 /*
 733                  * if the ptraced child is a zombie with exit_signal == -1
 734                  * we must collect it before we exit, or it will remain
 735                  * zombie forever since we prevented it from self-reap itself
 736                  * while it was being traced by us, to be able to see it in wait4.
 737                  */
 738                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 739                         list_add(&p->ptrace_list, to_release);
 740         }
 741         list_for_each_safe(_p, _n, &father->ptrace_children) {
 742                 p = list_entry(_p, struct task_struct, ptrace_list);
 743                 choose_new_parent(p, reaper);
 744                 reparent_thread(p, father, 1);
 745         }
 746 }
 747
 748 /*
 749  * Send signals to all our closest relatives so that they know
 750  * to properly mourn us..
 751  */
 752 static void exit_notify(struct task_struct *tsk)
 753 {
 754         int state;
 755         struct task_struct *t;
 756         struct list_head ptrace_dead, *_p, *_n;
 757         struct pid *pgrp;
 758
 759         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 760             && !thread_group_empty(tsk)) {
 761                 /*
 762                  * This occurs when there was a race between our exit
 763                  * syscall and a group signal choosing us as the one to
 764                  * wake up.  It could be that we are the only thread
 765                  * alerted to check for pending signals, but another thread
 766                  * should be woken now to take the signal since we will not.
 767                  * Now we'll wake all the threads in the group just to make
 768                  * sure someone gets all the pending signals.
 769                  */
 770                 read_lock(&tasklist_lock);
 771                 spin_lock_irq(&tsk->sighand->siglock);
 772                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 773                         if (!signal_pending(t) && !(t->flags & PF_EXITING))
 774                                 recalc_sigpending_and_wake(t);
 775                 spin_unlock_irq(&tsk->sighand->siglock);
 776                 read_unlock(&tasklist_lock);
 777         }
 778
 779         write_lock_irq(&tasklist_lock);
 780
 781         /*
 782          * This does two things:
 783          *
 784          * A.  Make init inherit all the child processes
 785          * B.  Check to see if any process groups have become orphaned
 786          *      as a result of our exiting, and if they have any stopped
 787          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 788          */
 789
 790         INIT_LIST_HEAD(&ptrace_dead);
 791         forget_original_parent(tsk, &ptrace_dead);
 792         BUG_ON(!list_empty(&tsk->children));
 793         BUG_ON(!list_empty(&tsk->ptrace_children));
 794
 795         /*
 796          * Check to see if any process groups have become orphaned
 797          * as a result of our exiting, and if they have any stopped
 798          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 799          *
 800          * Case i: Our father is in a different pgrp than we are
 801          * and we were the only connection outside, so our pgrp
 802          * is about to become orphaned.
 803          */
 804
 805         t = tsk->real_parent;
 806
 807         pgrp = task_pgrp(tsk);
 808         if ((task_pgrp(t) != pgrp) &&
 809             (task_session(t) == task_session(tsk)) &&
 810             will_become_orphaned_pgrp(pgrp, tsk) &&
 811             has_stopped_jobs(pgrp)) {
 812                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 813                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 814         }
 815
 816         /* Let father know we died
 817          *
 818          * Thread signals are configurable, but you aren't going to use
 819          * that to send signals to arbitary processes.
 820          * That stops right now.
 821          *
 822          * If the parent exec id doesn't match the exec id we saved
 823          * when we started then we know the parent has changed security
 824          * domain.
 825          *
 826          * If our self_exec id doesn't match our parent_exec_id then
 827          * we have changed execution domain as these two values started
 828          * the same after a fork.
 829          *
 830          */
 831
 832         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 833             ( tsk->parent_exec_id != t->self_exec_id  ||
 834               tsk->self_exec_id != tsk->parent_exec_id)
 835             && !capable(CAP_KILL))
 836                 tsk->exit_signal = SIGCHLD;
 837
 838
 839         /* If something other than our normal parent is ptracing us, then
 840          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 841          * only has special meaning to our real parent.
 842          */
 843         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 844                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 845                 do_notify_parent(tsk, signal);
 846         } else if (tsk->ptrace) {
 847                 do_notify_parent(tsk, SIGCHLD);
 848         }
 849
 850         state = EXIT_ZOMBIE;
 851         if (tsk->exit_signal == -1 &&
 852             (likely(tsk->ptrace == 0) ||
 853              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 854                 state = EXIT_DEAD;
 855         tsk->exit_state = state;
 856
 857         write_unlock_irq(&tasklist_lock);
 858
 859         list_for_each_safe(_p, _n, &ptrace_dead) {
 860                 list_del_init(_p);
 861                 t = list_entry(_p, struct task_struct, ptrace_list);
 862                 release_task(t);
 863         }
 864
 865         /* If the process is dead, release it - nobody will wait for it */
 866         if (state == EXIT_DEAD)
 867                 release_task(tsk);
 868 }
 869
 870 #ifdef CONFIG_DEBUG_STACK_USAGE
 871 static void check_stack_usage(void)
 872 {
 873         static DEFINE_SPINLOCK(low_water_lock);
 874         static int lowest_to_date = THREAD_SIZE;
 875         unsigned long *n = end_of_stack(current);
 876         unsigned long free;
 877
 878         while (*n == 0)
 879                 n++;
 880         free = (unsigned long)n - (unsigned long)end_of_stack(current);
 881
 882         if (free >= lowest_to_date)
 883                 return;
 884
 885         spin_lock(&low_water_lock);
 886         if (free < lowest_to_date) {
 887                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
 888                                 "left\n",
 889                                 current->comm, free);
 890                 lowest_to_date = free;
 891         }
 892         spin_unlock(&low_water_lock);
 893 }
 894 #else
 895 static inline void check_stack_usage(void) {}
 896 #endif
 897
 898 fastcall NORET_TYPE void do_exit(long code)
 899 {
 900         struct task_struct *tsk = current;
 901         int group_dead;
 902
 903         profile_task_exit(tsk);
 904
 905         WARN_ON(atomic_read(&tsk->fs_excl));
 906
 907         if (unlikely(in_interrupt()))
 908                 panic("Aiee, killing interrupt handler!");
 909         if (unlikely(!tsk->pid))
 910                 panic("Attempted to kill the idle task!");
 911         if (unlikely(tsk == child_reaper(tsk))) {
 912                 if (tsk->nsproxy->pid_ns != &init_pid_ns)
 913                         tsk->nsproxy->pid_ns->child_reaper = init_pid_ns.child_reaper;
 914                 else
 915                         panic("Attempted to kill init!");
 916         }
 917
 918
 919         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 920                 current->ptrace_message = code;
 921                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 922         }
 923
 924         /*
 925          * We're taking recursive faults here in do_exit. Safest is to just
 926          * leave this task alone and wait for reboot.
 927          */
 928         if (unlikely(tsk->flags & PF_EXITING)) {
 929                 printk(KERN_ALERT
 930                         "Fixing recursive fault but reboot is needed!\n");
 931                 /*
 932                  * We can do this unlocked here. The futex code uses
 933                  * this flag just to verify whether the pi state
 934                  * cleanup has been done or not. In the worst case it
 935                  * loops once more. We pretend that the cleanup was
 936                  * done as there is no way to return. Either the
 937                  * OWNER_DIED bit is set by now or we push the blocked
 938                  * task into the wait for ever nirwana as well.
 939                  */
 940                 tsk->flags |= PF_EXITPIDONE;
 941                 if (tsk->io_context)
 942                         exit_io_context();
 943                 set_current_state(TASK_UNINTERRUPTIBLE);
 944                 schedule();
 945         }
 946
 947         /*
 948          * tsk->flags are checked in the futex code to protect against
 949          * an exiting task cleaning up the robust pi futexes.
 950          */
 951         spin_lock_irq(&tsk->pi_lock);
 952         tsk->flags |= PF_EXITING;
 953         spin_unlock_irq(&tsk->pi_lock);
 954
 955         if (unlikely(in_atomic()))
 956                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 957                                 current->comm, current->pid,
 958                                 preempt_count());
 959
 960         acct_update_integrals(tsk);
 961         if (tsk->mm) {
 962                 update_hiwater_rss(tsk->mm);
 963                 update_hiwater_vm(tsk->mm);
 964         }
 965         group_dead = atomic_dec_and_test(&tsk->signal->live);
 966         if (group_dead) {
 967                 hrtimer_cancel(&tsk->signal->real_timer);
 968                 exit_itimers(tsk->signal);
 969         }
 970         acct_collect(code, group_dead);
 971         if (unlikely(tsk->robust_list))
 972                 exit_robust_list(tsk);
 973 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 974         if (unlikely(tsk->compat_robust_list))
 975                 compat_exit_robust_list(tsk);
 976 #endif
 977         if (group_dead)
 978                 tty_audit_exit();
 979         if (unlikely(tsk->audit_context))
 980                 audit_free(tsk);
 981
 982         taskstats_exit(tsk, group_dead);
 983
 984         exit_mm(tsk);
 985
 986         if (group_dead)
 987                 acct_process();
 988         exit_sem(tsk);
 989         __exit_files(tsk);
 990         __exit_fs(tsk);
 991         check_stack_usage();
 992         exit_thread();
 993         cpuset_exit(tsk);
 994         exit_keys(tsk);
 995
 996         if (group_dead && tsk->signal->leader)
 997                 disassociate_ctty(1);
 998
 999         module_put(task_thread_info(tsk)->exec_domain->module);
1000         if (tsk->binfmt)
1001                 module_put(tsk->binfmt->module);
1002
1003         tsk->exit_code = code;
1004         proc_exit_connector(tsk);
1005         exit_task_namespaces(tsk);
1006         exit_notify(tsk);
1007 #ifdef CONFIG_NUMA
1008         mpol_free(tsk->mempolicy);
1009         tsk->mempolicy = NULL;
1010 #endif
1011         /*
1012          * This must happen late, after the PID is not
1013          * hashed anymore:
1014          */
1015         if (unlikely(!list_empty(&tsk->pi_state_list)))
1016                 exit_pi_state_list(tsk);
1017         if (unlikely(current->pi_state_cache))
1018                 kfree(current->pi_state_cache);
1019         /*
1020          * Make sure we are holding no locks:
1021          */
1022         debug_check_no_locks_held(tsk);
1023         /*
1024          * We can do this unlocked here. The futex code uses this flag
1025          * just to verify whether the pi state cleanup has been done
1026          * or not. In the worst case it loops once more.
1027          */
1028         tsk->flags |= PF_EXITPIDONE;
1029
1030         if (tsk->io_context)
1031                 exit_io_context();
1032
1033         if (tsk->splice_pipe)
1034                 __free_pipe_info(tsk->splice_pipe);
1035
1036         preempt_disable();
1037         /* causes final put_task_struct in finish_task_switch(). */
1038         tsk->state = TASK_DEAD;
1039
1040         schedule();
1041         BUG();
1042         /* Avoid "noreturn function does return".  */
1043         for (;;)
1044                 cpu_relax();    /* For when BUG is null */
1045 }
1046
1047 EXPORT_SYMBOL_GPL(do_exit);
1048
1049 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1050 {
1051         if (comp)
1052                 complete(comp);
1053
1054         do_exit(code);
1055 }
1056
1057 EXPORT_SYMBOL(complete_and_exit);
1058
1059 asmlinkage long sys_exit(int error_code)
1060 {
1061         do_exit((error_code&0xff)<<8);
1062 }
1063
1064 /*
1065  * Take down every thread in the group.  This is called by fatal signals
1066  * as well as by sys_exit_group (below).
1067  */
1068 NORET_TYPE void
1069 do_group_exit(int exit_code)
1070 {
1071         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1072
1073         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1074                 exit_code = current->signal->group_exit_code;
1075         else if (!thread_group_empty(current)) {
1076                 struct signal_struct *const sig = current->signal;
1077                 struct sighand_struct *const sighand = current->sighand;
1078                 spin_lock_irq(&sighand->siglock);
1079                 if (sig->flags & SIGNAL_GROUP_EXIT)
1080                         /* Another thread got here before we took the lock.  */
1081                         exit_code = sig->group_exit_code;
1082                 else {
1083                         sig->group_exit_code = exit_code;
1084                         zap_other_threads(current);
1085                 }
1086                 spin_unlock_irq(&sighand->siglock);
1087         }
1088
1089         do_exit(exit_code);
1090         /* NOTREACHED */
1091 }
1092
1093 /*
1094  * this kills every thread in the thread group. Note that any externally
1095  * wait4()-ing process will get the correct exit code - even if this
1096  * thread is not the thread group leader.
1097  */
1098 asmlinkage void sys_exit_group(int error_code)
1099 {
1100         do_group_exit((error_code & 0xff) << 8);
1101 }
1102
1103 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1104 {
1105         int err;
1106
1107         if (pid > 0) {
1108                 if (p->pid != pid)
1109                         return 0;
1110         } else if (!pid) {
1111                 if (process_group(p) != process_group(current))
1112                         return 0;
1113         } else if (pid != -1) {
1114                 if (process_group(p) != -pid)
1115                         return 0;
1116         }
1117
1118         /*
1119          * Do not consider detached threads that are
1120          * not ptraced:
1121          */
1122         if (p->exit_signal == -1 && !p->ptrace)
1123                 return 0;
1124
1125         /* Wait for all children (clone and not) if __WALL is set;
1126          * otherwise, wait for clone children *only* if __WCLONE is
1127          * set; otherwise, wait for non-clone children *only*.  (Note:
1128          * A "clone" child here is one that reports to its parent
1129          * using a signal other than SIGCHLD.) */
1130         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1131             && !(options & __WALL))
1132                 return 0;
1133         /*
1134          * Do not consider thread group leaders that are
1135          * in a non-empty thread group:
1136          */
1137         if (delay_group_leader(p))
1138                 return 2;
1139
1140         err = security_task_wait(p);
1141         if (err)
1142                 return err;
1143
1144         return 1;
1145 }
1146
1147 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1148                                int why, int status,
1149                                struct siginfo __user *infop,
1150                                struct rusage __user *rusagep)
1151 {
1152         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1153
1154         put_task_struct(p);
1155         if (!retval)
1156                 retval = put_user(SIGCHLD, &infop->si_signo);
1157         if (!retval)
1158                 retval = put_user(0, &infop->si_errno);
1159         if (!retval)
1160                 retval = put_user((short)why, &infop->si_code);
1161         if (!retval)
1162                 retval = put_user(pid, &infop->si_pid);
1163         if (!retval)
1164                 retval = put_user(uid, &infop->si_uid);
1165         if (!retval)
1166                 retval = put_user(status, &infop->si_status);
1167         if (!retval)
1168                 retval = pid;
1169         return retval;
1170 }
1171
1172 /*
1173  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1174  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1175  * the lock and this task is uninteresting.  If we return nonzero, we have
1176  * released the lock and the system call should return.
1177  */
1178 static int wait_task_zombie(struct task_struct *p, int noreap,
1179                             struct siginfo __user *infop,
1180                             int __user *stat_addr, struct rusage __user *ru)
1181 {
1182         unsigned long state;
1183         int retval;
1184         int status;
1185
1186         if (unlikely(noreap)) {
1187                 pid_t pid = p->pid;
1188                 uid_t uid = p->uid;
1189                 int exit_code = p->exit_code;
1190                 int why, status;
1191
1192                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1193                         return 0;
1194                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1195                         return 0;
1196                 get_task_struct(p);
1197                 read_unlock(&tasklist_lock);
1198                 if ((exit_code & 0x7f) == 0) {
1199                         why = CLD_EXITED;
1200                         status = exit_code >> 8;
1201                 } else {
1202                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1203                         status = exit_code & 0x7f;
1204                 }
1205                 return wait_noreap_copyout(p, pid, uid, why,
1206                                            status, infop, ru);
1207         }
1208
1209         /*
1210          * Try to move the task's state to DEAD
1211          * only one thread is allowed to do this:
1212          */
1213         state = xchg(&p->exit_state, EXIT_DEAD);
1214         if (state != EXIT_ZOMBIE) {
1215                 BUG_ON(state != EXIT_DEAD);
1216                 return 0;
1217         }
1218         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1219                 /*
1220                  * This can only happen in a race with a ptraced thread
1221                  * dying on another processor.
1222                  */
1223                 return 0;
1224         }
1225
1226         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1227                 struct signal_struct *psig;
1228                 struct signal_struct *sig;
1229
1230                 /*
1231                  * The resource counters for the group leader are in its
1232                  * own task_struct.  Those for dead threads in the group
1233                  * are in its signal_struct, as are those for the child
1234                  * processes it has previously reaped.  All these
1235                  * accumulate in the parent's signal_struct c* fields.
1236                  *
1237                  * We don't bother to take a lock here to protect these
1238                  * p->signal fields, because they are only touched by
1239                  * __exit_signal, which runs with tasklist_lock
1240                  * write-locked anyway, and so is excluded here.  We do
1241                  * need to protect the access to p->parent->signal fields,
1242                  * as other threads in the parent group can be right
1243                  * here reaping other children at the same time.
1244                  */
1245                 spin_lock_irq(&p->parent->sighand->siglock);
1246                 psig = p->parent->signal;
1247                 sig = p->signal;
1248                 psig->cutime =
1249                         cputime_add(psig->cutime,
1250                         cputime_add(p->utime,
1251                         cputime_add(sig->utime,
1252                                     sig->cutime)));
1253                 psig->cstime =
1254                         cputime_add(psig->cstime,
1255                         cputime_add(p->stime,
1256                         cputime_add(sig->stime,
1257                                     sig->cstime)));
1258                 psig->cmin_flt +=
1259                         p->min_flt + sig->min_flt + sig->cmin_flt;
1260                 psig->cmaj_flt +=
1261                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1262                 psig->cnvcsw +=
1263                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1264                 psig->cnivcsw +=
1265                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1266                 psig->cinblock +=
1267                         task_io_get_inblock(p) +
1268                         sig->inblock + sig->cinblock;
1269                 psig->coublock +=
1270                         task_io_get_oublock(p) +
1271                         sig->oublock + sig->coublock;
1272                 spin_unlock_irq(&p->parent->sighand->siglock);
1273         }
1274
1275         /*
1276          * Now we are sure this task is interesting, and no other
1277          * thread can reap it because we set its state to EXIT_DEAD.
1278          */
1279         read_unlock(&tasklist_lock);
1280
1281         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1282         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1283                 ? p->signal->group_exit_code : p->exit_code;
1284         if (!retval && stat_addr)
1285                 retval = put_user(status, stat_addr);
1286         if (!retval && infop)
1287                 retval = put_user(SIGCHLD, &infop->si_signo);
1288         if (!retval && infop)
1289                 retval = put_user(0, &infop->si_errno);
1290         if (!retval && infop) {
1291                 int why;
1292
1293                 if ((status & 0x7f) == 0) {
1294                         why = CLD_EXITED;
1295                         status >>= 8;
1296                 } else {
1297                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1298                         status &= 0x7f;
1299                 }
1300                 retval = put_user((short)why, &infop->si_code);
1301                 if (!retval)
1302                         retval = put_user(status, &infop->si_status);
1303         }
1304         if (!retval && infop)
1305                 retval = put_user(p->pid, &infop->si_pid);
1306         if (!retval && infop)
1307                 retval = put_user(p->uid, &infop->si_uid);
1308         if (retval) {
1309                 // TODO: is this safe?
1310                 p->exit_state = EXIT_ZOMBIE;
1311                 return retval;
1312         }
1313         retval = p->pid;
1314         if (p->real_parent != p->parent) {
1315                 write_lock_irq(&tasklist_lock);
1316                 /* Double-check with lock held.  */
1317                 if (p->real_parent != p->parent) {
1318                         __ptrace_unlink(p);
1319                         // TODO: is this safe?
1320                         p->exit_state = EXIT_ZOMBIE;
1321                         /*
1322                          * If this is not a detached task, notify the parent.
1323                          * If it's still not detached after that, don't release
1324                          * it now.
1325                          */
1326                         if (p->exit_signal != -1) {
1327                                 do_notify_parent(p, p->exit_signal);
1328                                 if (p->exit_signal != -1)
1329                                         p = NULL;
1330                         }
1331                 }
1332                 write_unlock_irq(&tasklist_lock);
1333         }
1334         if (p != NULL)
1335                 release_task(p);
1336         BUG_ON(!retval);
1337         return retval;
1338 }
1339
1340 /*
1341  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1342  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1343  * the lock and this task is uninteresting.  If we return nonzero, we have
1344  * released the lock and the system call should return.
1345  */
1346 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1347                              int noreap, struct siginfo __user *infop,
1348                              int __user *stat_addr, struct rusage __user *ru)
1349 {
1350         int retval, exit_code;
1351
1352         if (!p->exit_code)
1353                 return 0;
1354         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1355             p->signal && p->signal->group_stop_count > 0)
1356                 /*
1357                  * A group stop is in progress and this is the group leader.
1358                  * We won't report until all threads have stopped.
1359                  */
1360                 return 0;
1361
1362         /*
1363          * Now we are pretty sure this task is interesting.
1364          * Make sure it doesn't get reaped out from under us while we
1365          * give up the lock and then examine it below.  We don't want to
1366          * keep holding onto the tasklist_lock while we call getrusage and
1367          * possibly take page faults for user memory.
1368          */
1369         get_task_struct(p);
1370         read_unlock(&tasklist_lock);
1371
1372         if (unlikely(noreap)) {
1373                 pid_t pid = p->pid;
1374                 uid_t uid = p->uid;
1375                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1376
1377                 exit_code = p->exit_code;
1378                 if (unlikely(!exit_code) ||
1379                     unlikely(p->state & TASK_TRACED))
1380                         goto bail_ref;
1381                 return wait_noreap_copyout(p, pid, uid,
1382                                            why, (exit_code << 8) | 0x7f,
1383                                            infop, ru);
1384         }
1385
1386         write_lock_irq(&tasklist_lock);
1387
1388         /*
1389          * This uses xchg to be atomic with the thread resuming and setting
1390          * it.  It must also be done with the write lock held to prevent a
1391          * race with the EXIT_ZOMBIE case.
1392          */
1393         exit_code = xchg(&p->exit_code, 0);
1394         if (unlikely(p->exit_state)) {
1395                 /*
1396                  * The task resumed and then died.  Let the next iteration
1397                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1398                  * already be zero here if it resumed and did _exit(0).
1399                  * The task itself is dead and won't touch exit_code again;
1400                  * other processors in this function are locked out.
1401                  */
1402                 p->exit_code = exit_code;
1403                 exit_code = 0;
1404         }
1405         if (unlikely(exit_code == 0)) {
1406                 /*
1407                  * Another thread in this function got to it first, or it
1408                  * resumed, or it resumed and then died.
1409                  */
1410                 write_unlock_irq(&tasklist_lock);
1411 bail_ref:
1412                 put_task_struct(p);
1413                 /*
1414                  * We are returning to the wait loop without having successfully
1415                  * removed the process and having released the lock. We cannot
1416                  * continue, since the "p" task pointer is potentially stale.
1417                  *
1418                  * Return -EAGAIN, and do_wait() will restart the loop from the
1419                  * beginning. Do _not_ re-acquire the lock.
1420                  */
1421                 return -EAGAIN;
1422         }
1423
1424         /* move to end of parent's list to avoid starvation */
1425         remove_parent(p);
1426         add_parent(p);
1427
1428         write_unlock_irq(&tasklist_lock);
1429
1430         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1431         if (!retval && stat_addr)
1432                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1433         if (!retval && infop)
1434                 retval = put_user(SIGCHLD, &infop->si_signo);
1435         if (!retval && infop)
1436                 retval = put_user(0, &infop->si_errno);
1437         if (!retval && infop)
1438                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1439                                           ? CLD_TRAPPED : CLD_STOPPED),
1440                                   &infop->si_code);
1441         if (!retval && infop)
1442                 retval = put_user(exit_code, &infop->si_status);
1443         if (!retval && infop)
1444                 retval = put_user(p->pid, &infop->si_pid);
1445         if (!retval && infop)
1446                 retval = put_user(p->uid, &infop->si_uid);
1447         if (!retval)
1448                 retval = p->pid;
1449         put_task_struct(p);
1450
1451         BUG_ON(!retval);
1452         return retval;
1453 }
1454
1455 /*
1456  * Handle do_wait work for one task in a live, non-stopped state.
1457  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1458  * the lock and this task is uninteresting.  If we return nonzero, we have
1459  * released the lock and the system call should return.
1460  */
1461 static int wait_task_continued(struct task_struct *p, int noreap,
1462                                struct siginfo __user *infop,
1463                                int __user *stat_addr, struct rusage __user *ru)
1464 {
1465         int retval;
1466         pid_t pid;
1467         uid_t uid;
1468
1469         if (unlikely(!p->signal))
1470                 return 0;
1471
1472         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1473                 return 0;
1474
1475         spin_lock_irq(&p->sighand->siglock);
1476         /* Re-check with the lock held.  */
1477         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1478                 spin_unlock_irq(&p->sighand->siglock);
1479                 return 0;
1480         }
1481         if (!noreap)
1482                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1483         spin_unlock_irq(&p->sighand->siglock);
1484
1485         pid = p->pid;
1486         uid = p->uid;
1487         get_task_struct(p);
1488         read_unlock(&tasklist_lock);
1489
1490         if (!infop) {
1491                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1492                 put_task_struct(p);
1493                 if (!retval && stat_addr)
1494                         retval = put_user(0xffff, stat_addr);
1495                 if (!retval)
1496                         retval = p->pid;
1497         } else {
1498                 retval = wait_noreap_copyout(p, pid, uid,
1499                                              CLD_CONTINUED, SIGCONT,
1500                                              infop, ru);
1501                 BUG_ON(retval == 0);
1502         }
1503
1504         return retval;
1505 }
1506
1507
1508 static inline int my_ptrace_child(struct task_struct *p)
1509 {
1510         if (!(p->ptrace & PT_PTRACED))
1511                 return 0;
1512         if (!(p->ptrace & PT_ATTACHED))
1513                 return 1;
1514         /*
1515          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1516          * we are the attacher.  If we are the real parent, this is a race
1517          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1518          * which we have to switch the parent links, but has already set
1519          * the flags in p->ptrace.
1520          */
1521         return (p->parent != p->real_parent);
1522 }
1523
1524 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1525                     int __user *stat_addr, struct rusage __user *ru)
1526 {
1527         DECLARE_WAITQUEUE(wait, current);
1528         struct task_struct *tsk;
1529         int flag, retval;
1530         int allowed, denied;
1531
1532         add_wait_queue(&current->signal->wait_chldexit,&wait);
1533 repeat:
1534         /*
1535          * We will set this flag if we see any child that might later
1536          * match our criteria, even if we are not able to reap it yet.
1537          */
1538         flag = 0;
1539         allowed = denied = 0;
1540         current->state = TASK_INTERRUPTIBLE;
1541         read_lock(&tasklist_lock);
1542         tsk = current;
1543         do {
1544                 struct task_struct *p;
1545                 struct list_head *_p;
1546                 int ret;
1547
1548                 list_for_each(_p,&tsk->children) {
1549                         p = list_entry(_p, struct task_struct, sibling);
1550
1551                         ret = eligible_child(pid, options, p);
1552                         if (!ret)
1553                                 continue;
1554
1555                         if (unlikely(ret < 0)) {
1556                                 denied = ret;
1557                                 continue;
1558                         }
1559                         allowed = 1;
1560
1561                         switch (p->state) {
1562                         case TASK_TRACED:
1563                                 /*
1564                                  * When we hit the race with PTRACE_ATTACH,
1565                                  * we will not report this child.  But the
1566                                  * race means it has not yet been moved to
1567                                  * our ptrace_children list, so we need to
1568                                  * set the flag here to avoid a spurious ECHILD
1569                                  * when the race happens with the only child.
1570                                  */
1571                                 flag = 1;
1572                                 if (!my_ptrace_child(p))
1573                                         continue;
1574                                 /*FALLTHROUGH*/
1575                         case TASK_STOPPED:
1576                                 /*
1577                                  * It's stopped now, so it might later
1578                                  * continue, exit, or stop again.
1579                                  */
1580                                 flag = 1;
1581                                 if (!(options & WUNTRACED) &&
1582                                     !my_ptrace_child(p))
1583                                         continue;
1584                                 retval = wait_task_stopped(p, ret == 2,
1585                                                            (options & WNOWAIT),
1586                                                            infop,
1587                                                            stat_addr, ru);
1588                                 if (retval == -EAGAIN)
1589                                         goto repeat;
1590                                 if (retval != 0) /* He released the lock.  */
1591                                         goto end;
1592                                 break;
1593                         default:
1594                         // case EXIT_DEAD:
1595                                 if (p->exit_state == EXIT_DEAD)
1596                                         continue;
1597                         // case EXIT_ZOMBIE:
1598                                 if (p->exit_state == EXIT_ZOMBIE) {
1599                                         /*
1600                                          * Eligible but we cannot release
1601                                          * it yet:
1602                                          */
1603                                         if (ret == 2)
1604                                                 goto check_continued;
1605                                         if (!likely(options & WEXITED))
1606                                                 continue;
1607                                         retval = wait_task_zombie(
1608                                                 p, (options & WNOWAIT),
1609                                                 infop, stat_addr, ru);
1610                                         /* He released the lock.  */
1611                                         if (retval != 0)
1612                                                 goto end;
1613                                         break;
1614                                 }
1615 check_continued:
1616                                 /*
1617                                  * It's running now, so it might later
1618                                  * exit, stop, or stop and then continue.
1619                                  */
1620                                 flag = 1;
1621                                 if (!unlikely(options & WCONTINUED))
1622                                         continue;
1623                                 retval = wait_task_continued(
1624                                         p, (options & WNOWAIT),
1625                                         infop, stat_addr, ru);
1626                                 if (retval != 0) /* He released the lock.  */
1627                                         goto end;
1628                                 break;
1629                         }
1630                 }
1631                 if (!flag) {
1632                         list_for_each(_p, &tsk->ptrace_children) {
1633                                 p = list_entry(_p, struct task_struct,
1634                                                 ptrace_list);
1635                                 if (!eligible_child(pid, options, p))
1636                                         continue;
1637                                 flag = 1;
1638                                 break;
1639                         }
1640                 }
1641                 if (options & __WNOTHREAD)
1642                         break;
1643                 tsk = next_thread(tsk);
1644                 BUG_ON(tsk->signal != current->signal);
1645         } while (tsk != current);
1646
1647         read_unlock(&tasklist_lock);
1648         if (flag) {
1649                 retval = 0;
1650                 if (options & WNOHANG)
1651                         goto end;
1652                 retval = -ERESTARTSYS;
1653                 if (signal_pending(current))
1654                         goto end;
1655                 schedule();
1656                 goto repeat;
1657         }
1658         retval = -ECHILD;
1659         if (unlikely(denied) && !allowed)
1660                 retval = denied;
1661 end:
1662         current->state = TASK_RUNNING;
1663         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1664         if (infop) {
1665                 if (retval > 0)
1666                 retval = 0;
1667                 else {
1668                         /*
1669                          * For a WNOHANG return, clear out all the fields
1670                          * we would set so the user can easily tell the
1671                          * difference.
1672                          */
1673                         if (!retval)
1674                                 retval = put_user(0, &infop->si_signo);
1675                         if (!retval)
1676                                 retval = put_user(0, &infop->si_errno);
1677                         if (!retval)
1678                                 retval = put_user(0, &infop->si_code);
1679                         if (!retval)
1680                                 retval = put_user(0, &infop->si_pid);
1681                         if (!retval)
1682                                 retval = put_user(0, &infop->si_uid);
1683                         if (!retval)
1684                                 retval = put_user(0, &infop->si_status);
1685                 }
1686         }
1687         return retval;
1688 }
1689
1690 asmlinkage long sys_waitid(int which, pid_t pid,
1691                            struct siginfo __user *infop, int options,
1692                            struct rusage __user *ru)
1693 {
1694         long ret;
1695
1696         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1697                 return -EINVAL;
1698         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1699                 return -EINVAL;
1700
1701         switch (which) {
1702         case P_ALL:
1703                 pid = -1;
1704                 break;
1705         case P_PID:
1706                 if (pid <= 0)
1707                         return -EINVAL;
1708                 break;
1709         case P_PGID:
1710                 if (pid <= 0)
1711                         return -EINVAL;
1712                 pid = -pid;
1713                 break;
1714         default:
1715                 return -EINVAL;
1716         }
1717
1718         ret = do_wait(pid, options, infop, NULL, ru);
1719
1720         /* avoid REGPARM breakage on x86: */
1721         prevent_tail_call(ret);
1722         return ret;
1723 }
1724
1725 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1726                           int options, struct rusage __user *ru)
1727 {
1728         long ret;
1729
1730         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1731                         __WNOTHREAD|__WCLONE|__WALL))
1732                 return -EINVAL;
1733         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1734
1735         /* avoid REGPARM breakage on x86: */
1736         prevent_tail_call(ret);
1737         return ret;
1738 }
1739
1740 #ifdef __ARCH_WANT_SYS_WAITPID
1741
1742 /*
1743  * sys_waitpid() remains for compatibility. waitpid() should be
1744  * implemented by calling sys_wait4() from libc.a.
1745  */
1746 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1747 {
1748         return sys_wait4(pid, stat_addr, options, NULL);
1749 }
1750
1751 #endif