kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/mm.h>
   8 #include <linux/slab.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/module.h>
  11 #include <linux/capability.h>
  12 #include <linux/completion.h>
  13 #include <linux/personality.h>
  14 #include <linux/tty.h>
  15 #include <linux/mnt_namespace.h>
  16 #include <linux/key.h>
  17 #include <linux/security.h>
  18 #include <linux/cpu.h>
  19 #include <linux/acct.h>
  20 #include <linux/tsacct_kern.h>
  21 #include <linux/file.h>
  22 #include <linux/binfmts.h>
  23 #include <linux/nsproxy.h>
  24 #include <linux/pid_namespace.h>
  25 #include <linux/ptrace.h>
  26 #include <linux/profile.h>
  27 #include <linux/mount.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/kthread.h>
  30 #include <linux/mempolicy.h>
  31 #include <linux/taskstats_kern.h>
  32 #include <linux/delayacct.h>
  33 #include <linux/freezer.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         posix_cpu_timers_exit(tsk);
  88         if (atomic_dec_and_test(&sig->count))
  89                 posix_cpu_timers_exit_group(tsk);
  90         else {
  91                 /*
  92                  * If there is any task waiting for the group exit
  93                  * then notify it:
  94                  */
  95                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
  96                         wake_up_process(sig->group_exit_task);
  97                         sig->group_exit_task = NULL;
  98                 }
  99                 if (tsk == sig->curr_target)
 100                         sig->curr_target = next_thread(tsk);
 101                 /*
 102                  * Accumulate here the counters for all threads but the
 103                  * group leader as they die, so they can be added into
 104                  * the process-wide totals when those are taken.
 105                  * The group leader stays around as a zombie as long
 106                  * as there are other threads.  When it gets reaped,
 107                  * the exit.c code will add its counts into these totals.
 108                  * We won't ever get here for the group leader, since it
 109                  * will have been the last reference on the signal_struct.
 110                  */
 111                 sig->utime = cputime_add(sig->utime, tsk->utime);
 112                 sig->stime = cputime_add(sig->stime, tsk->stime);
 113                 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
 114                 sig->min_flt += tsk->min_flt;
 115                 sig->maj_flt += tsk->maj_flt;
 116                 sig->nvcsw += tsk->nvcsw;
 117                 sig->nivcsw += tsk->nivcsw;
 118                 sig->inblock += task_io_get_inblock(tsk);
 119                 sig->oublock += task_io_get_oublock(tsk);
 120                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 121                 sig = NULL; /* Marker for below. */
 122         }
 123
 124         __unhash_process(tsk);
 125
 126         tsk->signal = NULL;
 127         tsk->sighand = NULL;
 128         spin_unlock(&sighand->siglock);
 129         rcu_read_unlock();
 130
 131         __cleanup_sighand(sighand);
 132         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 133         flush_sigqueue(&tsk->pending);
 134         if (sig) {
 135                 flush_sigqueue(&sig->shared_pending);
 136                 taskstats_tgid_free(sig);
 137                 __cleanup_signal(sig);
 138         }
 139 }
 140
 141 static void delayed_put_task_struct(struct rcu_head *rhp)
 142 {
 143         put_task_struct(container_of(rhp, struct task_struct, rcu));
 144 }
 145
 146 void release_task(struct task_struct * p)
 147 {
 148         struct task_struct *leader;
 149         int zap_leader;
 150 repeat:
 151         atomic_dec(&p->user->processes);
 152         write_lock_irq(&tasklist_lock);
 153         ptrace_unlink(p);
 154         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 155         __exit_signal(p);
 156
 157         /*
 158          * If we are the last non-leader member of the thread
 159          * group, and the leader is zombie, then notify the
 160          * group leader's parent process. (if it wants notification.)
 161          */
 162         zap_leader = 0;
 163         leader = p->group_leader;
 164         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 165                 BUG_ON(leader->exit_signal == -1);
 166                 do_notify_parent(leader, leader->exit_signal);
 167                 /*
 168                  * If we were the last child thread and the leader has
 169                  * exited already, and the leader's parent ignores SIGCHLD,
 170                  * then we are the one who should release the leader.
 171                  *
 172                  * do_notify_parent() will have marked it self-reaping in
 173                  * that case.
 174                  */
 175                 zap_leader = (leader->exit_signal == -1);
 176         }
 177
 178         write_unlock_irq(&tasklist_lock);
 179         proc_flush_task(p);
 180         release_thread(p);
 181         call_rcu(&p->rcu, delayed_put_task_struct);
 182
 183         p = leader;
 184         if (unlikely(zap_leader))
 185                 goto repeat;
 186 }
 187
 188 /*
 189  * This checks not only the pgrp, but falls back on the pid if no
 190  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 191  * without this...
 192  *
 193  * The caller must hold rcu lock or the tasklist lock.
 194  */
 195 struct pid *session_of_pgrp(struct pid *pgrp)
 196 {
 197         struct task_struct *p;
 198         struct pid *sid = NULL;
 199
 200         p = pid_task(pgrp, PIDTYPE_PGID);
 201         if (p == NULL)
 202                 p = pid_task(pgrp, PIDTYPE_PID);
 203         if (p != NULL)
 204                 sid = task_session(p);
 205
 206         return sid;
 207 }
 208
 209 /*
 210  * Determine if a process group is "orphaned", according to the POSIX
 211  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 212  * by terminal-generated stop signals.  Newly orphaned process groups are
 213  * to receive a SIGHUP and a SIGCONT.
 214  *
 215  * "I ask you, have you ever known what it is to be an orphan?"
 216  */
 217 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 218 {
 219         struct task_struct *p;
 220         int ret = 1;
 221
 222         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 223                 if (p == ignored_task
 224                                 || p->exit_state
 225                                 || is_init(p->real_parent))
 226                         continue;
 227                 if (task_pgrp(p->real_parent) != pgrp &&
 228                     task_session(p->real_parent) == task_session(p)) {
 229                         ret = 0;
 230                         break;
 231                 }
 232         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 233         return ret;     /* (sighing) "Often!" */
 234 }
 235
 236 int is_current_pgrp_orphaned(void)
 237 {
 238         int retval;
 239
 240         read_lock(&tasklist_lock);
 241         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 242         read_unlock(&tasklist_lock);
 243
 244         return retval;
 245 }
 246
 247 static int has_stopped_jobs(struct pid *pgrp)
 248 {
 249         int retval = 0;
 250         struct task_struct *p;
 251
 252         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 253                 if (p->state != TASK_STOPPED)
 254                         continue;
 255                 retval = 1;
 256                 break;
 257         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 258         return retval;
 259 }
 260
 261 /**
 262  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
 263  *
 264  * If a kernel thread is launched as a result of a system call, or if
 265  * it ever exits, it should generally reparent itself to kthreadd so it
 266  * isn't in the way of other processes and is correctly cleaned up on exit.
 267  *
 268  * The various task state such as scheduling policy and priority may have
 269  * been inherited from a user process, so we reset them to sane values here.
 270  *
 271  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
 272  */
 273 static void reparent_to_kthreadd(void)
 274 {
 275         write_lock_irq(&tasklist_lock);
 276
 277         ptrace_unlink(current);
 278         /* Reparent to init */
 279         remove_parent(current);
 280         current->real_parent = current->parent = kthreadd_task;
 281         add_parent(current);
 282
 283         /* Set the exit signal to SIGCHLD so we signal init on exit */
 284         current->exit_signal = SIGCHLD;
 285
 286         if (task_nice(current) < 0)
 287                 set_user_nice(current, 0);
 288         /* cpus_allowed? */
 289         /* rt_priority? */
 290         /* signals? */
 291         security_task_reparent_to_init(current);
 292         memcpy(current->signal->rlim, init_task.signal->rlim,
 293                sizeof(current->signal->rlim));
 294         atomic_inc(&(INIT_USER->__count));
 295         write_unlock_irq(&tasklist_lock);
 296         switch_uid(INIT_USER);
 297 }
 298
 299 void __set_special_pids(pid_t session, pid_t pgrp)
 300 {
 301         struct task_struct *curr = current->group_leader;
 302
 303         if (process_session(curr) != session) {
 304                 detach_pid(curr, PIDTYPE_SID);
 305                 set_signal_session(curr->signal, session);
 306                 attach_pid(curr, PIDTYPE_SID, find_pid(session));
 307         }
 308         if (process_group(curr) != pgrp) {
 309                 detach_pid(curr, PIDTYPE_PGID);
 310                 curr->signal->pgrp = pgrp;
 311                 attach_pid(curr, PIDTYPE_PGID, find_pid(pgrp));
 312         }
 313 }
 314
 315 static void set_special_pids(pid_t session, pid_t pgrp)
 316 {
 317         write_lock_irq(&tasklist_lock);
 318         __set_special_pids(session, pgrp);
 319         write_unlock_irq(&tasklist_lock);
 320 }
 321
 322 /*
 323  * Let kernel threads use this to say that they
 324  * allow a certain signal (since daemonize() will
 325  * have disabled all of them by default).
 326  */
 327 int allow_signal(int sig)
 328 {
 329         if (!valid_signal(sig) || sig < 1)
 330                 return -EINVAL;
 331
 332         spin_lock_irq(&current->sighand->siglock);
 333         sigdelset(&current->blocked, sig);
 334         if (!current->mm) {
 335                 /* Kernel threads handle their own signals.
 336                    Let the signal code know it'll be handled, so
 337                    that they don't get converted to SIGKILL or
 338                    just silently dropped */
 339                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 340         }
 341         recalc_sigpending();
 342         spin_unlock_irq(&current->sighand->siglock);
 343         return 0;
 344 }
 345
 346 EXPORT_SYMBOL(allow_signal);
 347
 348 int disallow_signal(int sig)
 349 {
 350         if (!valid_signal(sig) || sig < 1)
 351                 return -EINVAL;
 352
 353         spin_lock_irq(&current->sighand->siglock);
 354         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
 355         recalc_sigpending();
 356         spin_unlock_irq(&current->sighand->siglock);
 357         return 0;
 358 }
 359
 360 EXPORT_SYMBOL(disallow_signal);
 361
 362 /*
 363  *      Put all the gunge required to become a kernel thread without
 364  *      attached user resources in one place where it belongs.
 365  */
 366
 367 void daemonize(const char *name, ...)
 368 {
 369         va_list args;
 370         struct fs_struct *fs;
 371         sigset_t blocked;
 372
 373         va_start(args, name);
 374         vsnprintf(current->comm, sizeof(current->comm), name, args);
 375         va_end(args);
 376
 377         /*
 378          * If we were started as result of loading a module, close all of the
 379          * user space pages.  We don't need them, and if we didn't close them
 380          * they would be locked into memory.
 381          */
 382         exit_mm(current);
 383         /*
 384          * We don't want to have TIF_FREEZE set if the system-wide hibernation
 385          * or suspend transition begins right now.
 386          */
 387         current->flags |= PF_NOFREEZE;
 388
 389         set_special_pids(1, 1);
 390         proc_clear_tty(current);
 391
 392         /* Block and flush all signals */
 393         sigfillset(&blocked);
 394         sigprocmask(SIG_BLOCK, &blocked, NULL);
 395         flush_signals(current);
 396
 397         /* Become as one with the init task */
 398
 399         exit_fs(current);       /* current->fs->count--; */
 400         fs = init_task.fs;
 401         current->fs = fs;
 402         atomic_inc(&fs->count);
 403
 404         exit_task_namespaces(current);
 405         current->nsproxy = init_task.nsproxy;
 406         get_task_namespaces(current);
 407
 408         exit_files(current);
 409         current->files = init_task.files;
 410         atomic_inc(&current->files->count);
 411
 412         reparent_to_kthreadd();
 413 }
 414
 415 EXPORT_SYMBOL(daemonize);
 416
 417 static void close_files(struct files_struct * files)
 418 {
 419         int i, j;
 420         struct fdtable *fdt;
 421
 422         j = 0;
 423
 424         /*
 425          * It is safe to dereference the fd table without RCU or
 426          * ->file_lock because this is the last reference to the
 427          * files structure.
 428          */
 429         fdt = files_fdtable(files);
 430         for (;;) {
 431                 unsigned long set;
 432                 i = j * __NFDBITS;
 433                 if (i >= fdt->max_fds)
 434                         break;
 435                 set = fdt->open_fds->fds_bits[j++];
 436                 while (set) {
 437                         if (set & 1) {
 438                                 struct file * file = xchg(&fdt->fd[i], NULL);
 439                                 if (file) {
 440                                         filp_close(file, files);
 441                                         cond_resched();
 442                                 }
 443                         }
 444                         i++;
 445                         set >>= 1;
 446                 }
 447         }
 448 }
 449
 450 struct files_struct *get_files_struct(struct task_struct *task)
 451 {
 452         struct files_struct *files;
 453
 454         task_lock(task);
 455         files = task->files;
 456         if (files)
 457                 atomic_inc(&files->count);
 458         task_unlock(task);
 459
 460         return files;
 461 }
 462
 463 void fastcall put_files_struct(struct files_struct *files)
 464 {
 465         struct fdtable *fdt;
 466
 467         if (atomic_dec_and_test(&files->count)) {
 468                 close_files(files);
 469                 /*
 470                  * Free the fd and fdset arrays if we expanded them.
 471                  * If the fdtable was embedded, pass files for freeing
 472                  * at the end of the RCU grace period. Otherwise,
 473                  * you can free files immediately.
 474                  */
 475                 fdt = files_fdtable(files);
 476                 if (fdt != &files->fdtab)
 477                         kmem_cache_free(files_cachep, files);
 478                 free_fdtable(fdt);
 479         }
 480 }
 481
 482 EXPORT_SYMBOL(put_files_struct);
 483
 484 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
 485 {
 486         struct files_struct *old;
 487
 488         old = tsk->files;
 489         task_lock(tsk);
 490         tsk->files = files;
 491         task_unlock(tsk);
 492         put_files_struct(old);
 493 }
 494 EXPORT_SYMBOL(reset_files_struct);
 495
 496 static inline void __exit_files(struct task_struct *tsk)
 497 {
 498         struct files_struct * files = tsk->files;
 499
 500         if (files) {
 501                 task_lock(tsk);
 502                 tsk->files = NULL;
 503                 task_unlock(tsk);
 504                 put_files_struct(files);
 505         }
 506 }
 507
 508 void exit_files(struct task_struct *tsk)
 509 {
 510         __exit_files(tsk);
 511 }
 512
 513 static inline void __put_fs_struct(struct fs_struct *fs)
 514 {
 515         /* No need to hold fs->lock if we are killing it */
 516         if (atomic_dec_and_test(&fs->count)) {
 517                 dput(fs->root);
 518                 mntput(fs->rootmnt);
 519                 dput(fs->pwd);
 520                 mntput(fs->pwdmnt);
 521                 if (fs->altroot) {
 522                         dput(fs->altroot);
 523                         mntput(fs->altrootmnt);
 524                 }
 525                 kmem_cache_free(fs_cachep, fs);
 526         }
 527 }
 528
 529 void put_fs_struct(struct fs_struct *fs)
 530 {
 531         __put_fs_struct(fs);
 532 }
 533
 534 static inline void __exit_fs(struct task_struct *tsk)
 535 {
 536         struct fs_struct * fs = tsk->fs;
 537
 538         if (fs) {
 539                 task_lock(tsk);
 540                 tsk->fs = NULL;
 541                 task_unlock(tsk);
 542                 __put_fs_struct(fs);
 543         }
 544 }
 545
 546 void exit_fs(struct task_struct *tsk)
 547 {
 548         __exit_fs(tsk);
 549 }
 550
 551 EXPORT_SYMBOL_GPL(exit_fs);
 552
 553 /*
 554  * Turn us into a lazy TLB process if we
 555  * aren't already..
 556  */
 557 static void exit_mm(struct task_struct * tsk)
 558 {
 559         struct mm_struct *mm = tsk->mm;
 560
 561         mm_release(tsk, mm);
 562         if (!mm)
 563                 return;
 564         /*
 565          * Serialize with any possible pending coredump.
 566          * We must hold mmap_sem around checking core_waiters
 567          * and clearing tsk->mm.  The core-inducing thread
 568          * will increment core_waiters for each thread in the
 569          * group with ->mm != NULL.
 570          */
 571         down_read(&mm->mmap_sem);
 572         if (mm->core_waiters) {
 573                 up_read(&mm->mmap_sem);
 574                 down_write(&mm->mmap_sem);
 575                 if (!--mm->core_waiters)
 576                         complete(mm->core_startup_done);
 577                 up_write(&mm->mmap_sem);
 578
 579                 wait_for_completion(&mm->core_done);
 580                 down_read(&mm->mmap_sem);
 581         }
 582         atomic_inc(&mm->mm_count);
 583         BUG_ON(mm != tsk->active_mm);
 584         /* more a memory barrier than a real lock */
 585         task_lock(tsk);
 586         tsk->mm = NULL;
 587         up_read(&mm->mmap_sem);
 588         enter_lazy_tlb(mm, current);
 589         /* We don't want this task to be frozen prematurely */
 590         clear_freeze_flag(tsk);
 591         task_unlock(tsk);
 592         mmput(mm);
 593 }
 594
 595 static void
 596 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 597 {
 598         if (p->pdeath_signal)
 599                 /* We already hold the tasklist_lock here.  */
 600                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 601
 602         /* Move the child from its dying parent to the new one.  */
 603         if (unlikely(traced)) {
 604                 /* Preserve ptrace links if someone else is tracing this child.  */
 605                 list_del_init(&p->ptrace_list);
 606                 if (p->parent != p->real_parent)
 607                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 608         } else {
 609                 /* If this child is being traced, then we're the one tracing it
 610                  * anyway, so let go of it.
 611                  */
 612                 p->ptrace = 0;
 613                 remove_parent(p);
 614                 p->parent = p->real_parent;
 615                 add_parent(p);
 616
 617                 if (p->state == TASK_TRACED) {
 618                         /*
 619                          * If it was at a trace stop, turn it into
 620                          * a normal stop since it's no longer being
 621                          * traced.
 622                          */
 623                         ptrace_untrace(p);
 624                 }
 625         }
 626
 627         /* If this is a threaded reparent there is no need to
 628          * notify anyone anything has happened.
 629          */
 630         if (p->real_parent->group_leader == father->group_leader)
 631                 return;
 632
 633         /* We don't want people slaying init.  */
 634         if (p->exit_signal != -1)
 635                 p->exit_signal = SIGCHLD;
 636
 637         /* If we'd notified the old parent about this child's death,
 638          * also notify the new parent.
 639          */
 640         if (!traced && p->exit_state == EXIT_ZOMBIE &&
 641             p->exit_signal != -1 && thread_group_empty(p))
 642                 do_notify_parent(p, p->exit_signal);
 643
 644         /*
 645          * process group orphan check
 646          * Case ii: Our child is in a different pgrp
 647          * than we are, and it was the only connection
 648          * outside, so the child pgrp is now orphaned.
 649          */
 650         if ((task_pgrp(p) != task_pgrp(father)) &&
 651             (task_session(p) == task_session(father))) {
 652                 struct pid *pgrp = task_pgrp(p);
 653
 654                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
 655                     has_stopped_jobs(pgrp)) {
 656                         __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 657                         __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 658                 }
 659         }
 660 }
 661
 662 /*
 663  * When we die, we re-parent all our children.
 664  * Try to give them to another thread in our thread
 665  * group, and if no such member exists, give it to
 666  * the child reaper process (ie "init") in our pid
 667  * space.
 668  */
 669 static void
 670 forget_original_parent(struct task_struct *father, struct list_head *to_release)
 671 {
 672         struct task_struct *p, *reaper = father;
 673         struct list_head *_p, *_n;
 674
 675         do {
 676                 reaper = next_thread(reaper);
 677                 if (reaper == father) {
 678                         reaper = child_reaper(father);
 679                         break;
 680                 }
 681         } while (reaper->exit_state);
 682
 683         /*
 684          * There are only two places where our children can be:
 685          *
 686          * - in our child list
 687          * - in our ptraced child list
 688          *
 689          * Search them and reparent children.
 690          */
 691         list_for_each_safe(_p, _n, &father->children) {
 692                 int ptrace;
 693                 p = list_entry(_p, struct task_struct, sibling);
 694
 695                 ptrace = p->ptrace;
 696
 697                 /* if father isn't the real parent, then ptrace must be enabled */
 698                 BUG_ON(father != p->real_parent && !ptrace);
 699
 700                 if (father == p->real_parent) {
 701                         /* reparent with a reaper, real father it's us */
 702                         p->real_parent = reaper;
 703                         reparent_thread(p, father, 0);
 704                 } else {
 705                         /* reparent ptraced task to its real parent */
 706                         __ptrace_unlink (p);
 707                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 708                             thread_group_empty(p))
 709                                 do_notify_parent(p, p->exit_signal);
 710                 }
 711
 712                 /*
 713                  * if the ptraced child is a zombie with exit_signal == -1
 714                  * we must collect it before we exit, or it will remain
 715                  * zombie forever since we prevented it from self-reap itself
 716                  * while it was being traced by us, to be able to see it in wait4.
 717                  */
 718                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 719                         list_add(&p->ptrace_list, to_release);
 720         }
 721         list_for_each_safe(_p, _n, &father->ptrace_children) {
 722                 p = list_entry(_p, struct task_struct, ptrace_list);
 723                 p->real_parent = reaper;
 724                 reparent_thread(p, father, 1);
 725         }
 726 }
 727
 728 /*
 729  * Send signals to all our closest relatives so that they know
 730  * to properly mourn us..
 731  */
 732 static void exit_notify(struct task_struct *tsk)
 733 {
 734         int state;
 735         struct task_struct *t;
 736         struct list_head ptrace_dead, *_p, *_n;
 737         struct pid *pgrp;
 738
 739         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 740             && !thread_group_empty(tsk)) {
 741                 /*
 742                  * This occurs when there was a race between our exit
 743                  * syscall and a group signal choosing us as the one to
 744                  * wake up.  It could be that we are the only thread
 745                  * alerted to check for pending signals, but another thread
 746                  * should be woken now to take the signal since we will not.
 747                  * Now we'll wake all the threads in the group just to make
 748                  * sure someone gets all the pending signals.
 749                  */
 750                 read_lock(&tasklist_lock);
 751                 spin_lock_irq(&tsk->sighand->siglock);
 752                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 753                         if (!signal_pending(t) && !(t->flags & PF_EXITING))
 754                                 recalc_sigpending_and_wake(t);
 755                 spin_unlock_irq(&tsk->sighand->siglock);
 756                 read_unlock(&tasklist_lock);
 757         }
 758
 759         write_lock_irq(&tasklist_lock);
 760
 761         /*
 762          * This does two things:
 763          *
 764          * A.  Make init inherit all the child processes
 765          * B.  Check to see if any process groups have become orphaned
 766          *      as a result of our exiting, and if they have any stopped
 767          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 768          */
 769
 770         INIT_LIST_HEAD(&ptrace_dead);
 771         forget_original_parent(tsk, &ptrace_dead);
 772         BUG_ON(!list_empty(&tsk->children));
 773         BUG_ON(!list_empty(&tsk->ptrace_children));
 774
 775         /*
 776          * 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          * Case i: Our father is in a different pgrp than we are
 781          * and we were the only connection outside, so our pgrp
 782          * is about to become orphaned.
 783          */
 784
 785         t = tsk->real_parent;
 786
 787         pgrp = task_pgrp(tsk);
 788         if ((task_pgrp(t) != pgrp) &&
 789             (task_session(t) == task_session(tsk)) &&
 790             will_become_orphaned_pgrp(pgrp, tsk) &&
 791             has_stopped_jobs(pgrp)) {
 792                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 793                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 794         }
 795
 796         /* Let father know we died
 797          *
 798          * Thread signals are configurable, but you aren't going to use
 799          * that to send signals to arbitary processes.
 800          * That stops right now.
 801          *
 802          * If the parent exec id doesn't match the exec id we saved
 803          * when we started then we know the parent has changed security
 804          * domain.
 805          *
 806          * If our self_exec id doesn't match our parent_exec_id then
 807          * we have changed execution domain as these two values started
 808          * the same after a fork.
 809          */
 810         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 811             ( tsk->parent_exec_id != t->self_exec_id  ||
 812               tsk->self_exec_id != tsk->parent_exec_id)
 813             && !capable(CAP_KILL))
 814                 tsk->exit_signal = SIGCHLD;
 815
 816
 817         /* If something other than our normal parent is ptracing us, then
 818          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 819          * only has special meaning to our real parent.
 820          */
 821         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 822                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 823                 do_notify_parent(tsk, signal);
 824         } else if (tsk->ptrace) {
 825                 do_notify_parent(tsk, SIGCHLD);
 826         }
 827
 828         state = EXIT_ZOMBIE;
 829         if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
 830                 state = EXIT_DEAD;
 831         tsk->exit_state = state;
 832
 833         write_unlock_irq(&tasklist_lock);
 834
 835         list_for_each_safe(_p, _n, &ptrace_dead) {
 836                 list_del_init(_p);
 837                 t = list_entry(_p, struct task_struct, ptrace_list);
 838                 release_task(t);
 839         }
 840
 841         /* If the process is dead, release it - nobody will wait for it */
 842         if (state == EXIT_DEAD)
 843                 release_task(tsk);
 844 }
 845
 846 #ifdef CONFIG_DEBUG_STACK_USAGE
 847 static void check_stack_usage(void)
 848 {
 849         static DEFINE_SPINLOCK(low_water_lock);
 850         static int lowest_to_date = THREAD_SIZE;
 851         unsigned long *n = end_of_stack(current);
 852         unsigned long free;
 853
 854         while (*n == 0)
 855                 n++;
 856         free = (unsigned long)n - (unsigned long)end_of_stack(current);
 857
 858         if (free >= lowest_to_date)
 859                 return;
 860
 861         spin_lock(&low_water_lock);
 862         if (free < lowest_to_date) {
 863                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
 864                                 "left\n",
 865                                 current->comm, free);
 866                 lowest_to_date = free;
 867         }
 868         spin_unlock(&low_water_lock);
 869 }
 870 #else
 871 static inline void check_stack_usage(void) {}
 872 #endif
 873
 874 static inline void exit_child_reaper(struct task_struct *tsk)
 875 {
 876         if (likely(tsk->group_leader != child_reaper(tsk)))
 877                 return;
 878
 879         panic("Attempted to kill init!");
 880 }
 881
 882 fastcall NORET_TYPE void do_exit(long code)
 883 {
 884         struct task_struct *tsk = current;
 885         int group_dead;
 886
 887         profile_task_exit(tsk);
 888
 889         WARN_ON(atomic_read(&tsk->fs_excl));
 890
 891         if (unlikely(in_interrupt()))
 892                 panic("Aiee, killing interrupt handler!");
 893         if (unlikely(!tsk->pid))
 894                 panic("Attempted to kill the idle task!");
 895
 896         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 897                 current->ptrace_message = code;
 898                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 899         }
 900
 901         /*
 902          * We're taking recursive faults here in do_exit. Safest is to just
 903          * leave this task alone and wait for reboot.
 904          */
 905         if (unlikely(tsk->flags & PF_EXITING)) {
 906                 printk(KERN_ALERT
 907                         "Fixing recursive fault but reboot is needed!\n");
 908                 /*
 909                  * We can do this unlocked here. The futex code uses
 910                  * this flag just to verify whether the pi state
 911                  * cleanup has been done or not. In the worst case it
 912                  * loops once more. We pretend that the cleanup was
 913                  * done as there is no way to return. Either the
 914                  * OWNER_DIED bit is set by now or we push the blocked
 915                  * task into the wait for ever nirwana as well.
 916                  */
 917                 tsk->flags |= PF_EXITPIDONE;
 918                 if (tsk->io_context)
 919                         exit_io_context();
 920                 set_current_state(TASK_UNINTERRUPTIBLE);
 921                 schedule();
 922         }
 923
 924         tsk->flags |= PF_EXITING;
 925         /*
 926          * tsk->flags are checked in the futex code to protect against
 927          * an exiting task cleaning up the robust pi futexes.
 928          */
 929         smp_mb();
 930         spin_unlock_wait(&tsk->pi_lock);
 931
 932         if (unlikely(in_atomic()))
 933                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 934                                 current->comm, current->pid,
 935                                 preempt_count());
 936
 937         acct_update_integrals(tsk);
 938         if (tsk->mm) {
 939                 update_hiwater_rss(tsk->mm);
 940                 update_hiwater_vm(tsk->mm);
 941         }
 942         group_dead = atomic_dec_and_test(&tsk->signal->live);
 943         if (group_dead) {
 944                 exit_child_reaper(tsk);
 945                 hrtimer_cancel(&tsk->signal->real_timer);
 946                 exit_itimers(tsk->signal);
 947         }
 948         acct_collect(code, group_dead);
 949         if (unlikely(tsk->robust_list))
 950                 exit_robust_list(tsk);
 951 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 952         if (unlikely(tsk->compat_robust_list))
 953                 compat_exit_robust_list(tsk);
 954 #endif
 955         if (group_dead)
 956                 tty_audit_exit();
 957         if (unlikely(tsk->audit_context))
 958                 audit_free(tsk);
 959
 960         tsk->exit_code = code;
 961         taskstats_exit(tsk, group_dead);
 962
 963         exit_mm(tsk);
 964
 965         if (group_dead)
 966                 acct_process();
 967         exit_sem(tsk);
 968         __exit_files(tsk);
 969         __exit_fs(tsk);
 970         check_stack_usage();
 971         exit_thread();
 972         cpuset_exit(tsk);
 973         exit_keys(tsk);
 974
 975         if (group_dead && tsk->signal->leader)
 976                 disassociate_ctty(1);
 977
 978         module_put(task_thread_info(tsk)->exec_domain->module);
 979         if (tsk->binfmt)
 980                 module_put(tsk->binfmt->module);
 981
 982         proc_exit_connector(tsk);
 983         exit_task_namespaces(tsk);
 984         exit_notify(tsk);
 985 #ifdef CONFIG_NUMA
 986         mpol_free(tsk->mempolicy);
 987         tsk->mempolicy = NULL;
 988 #endif
 989         /*
 990          * This must happen late, after the PID is not
 991          * hashed anymore:
 992          */
 993         if (unlikely(!list_empty(&tsk->pi_state_list)))
 994                 exit_pi_state_list(tsk);
 995         if (unlikely(current->pi_state_cache))
 996                 kfree(current->pi_state_cache);
 997         /*
 998          * Make sure we are holding no locks:
 999          */
1000         debug_check_no_locks_held(tsk);
1001         /*
1002          * We can do this unlocked here. The futex code uses this flag
1003          * just to verify whether the pi state cleanup has been done
1004          * or not. In the worst case it loops once more.
1005          */
1006         tsk->flags |= PF_EXITPIDONE;
1007
1008         if (tsk->io_context)
1009                 exit_io_context();
1010
1011         if (tsk->splice_pipe)
1012                 __free_pipe_info(tsk->splice_pipe);
1013
1014         preempt_disable();
1015         /* causes final put_task_struct in finish_task_switch(). */
1016         tsk->state = TASK_DEAD;
1017
1018         schedule();
1019         BUG();
1020         /* Avoid "noreturn function does return".  */
1021         for (;;)
1022                 cpu_relax();    /* For when BUG is null */
1023 }
1024
1025 EXPORT_SYMBOL_GPL(do_exit);
1026
1027 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1028 {
1029         if (comp)
1030                 complete(comp);
1031
1032         do_exit(code);
1033 }
1034
1035 EXPORT_SYMBOL(complete_and_exit);
1036
1037 asmlinkage long sys_exit(int error_code)
1038 {
1039         do_exit((error_code&0xff)<<8);
1040 }
1041
1042 /*
1043  * Take down every thread in the group.  This is called by fatal signals
1044  * as well as by sys_exit_group (below).
1045  */
1046 NORET_TYPE void
1047 do_group_exit(int exit_code)
1048 {
1049         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1050
1051         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1052                 exit_code = current->signal->group_exit_code;
1053         else if (!thread_group_empty(current)) {
1054                 struct signal_struct *const sig = current->signal;
1055                 struct sighand_struct *const sighand = current->sighand;
1056                 spin_lock_irq(&sighand->siglock);
1057                 if (sig->flags & SIGNAL_GROUP_EXIT)
1058                         /* Another thread got here before we took the lock.  */
1059                         exit_code = sig->group_exit_code;
1060                 else {
1061                         sig->group_exit_code = exit_code;
1062                         zap_other_threads(current);
1063                 }
1064                 spin_unlock_irq(&sighand->siglock);
1065         }
1066
1067         do_exit(exit_code);
1068         /* NOTREACHED */
1069 }
1070
1071 /*
1072  * this kills every thread in the thread group. Note that any externally
1073  * wait4()-ing process will get the correct exit code - even if this
1074  * thread is not the thread group leader.
1075  */
1076 asmlinkage void sys_exit_group(int error_code)
1077 {
1078         do_group_exit((error_code & 0xff) << 8);
1079 }
1080
1081 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1082 {
1083         int err;
1084
1085         if (pid > 0) {
1086                 if (p->pid != pid)
1087                         return 0;
1088         } else if (!pid) {
1089                 if (process_group(p) != process_group(current))
1090                         return 0;
1091         } else if (pid != -1) {
1092                 if (process_group(p) != -pid)
1093                         return 0;
1094         }
1095
1096         /*
1097          * Do not consider detached threads that are
1098          * not ptraced:
1099          */
1100         if (p->exit_signal == -1 && !p->ptrace)
1101                 return 0;
1102
1103         /* Wait for all children (clone and not) if __WALL is set;
1104          * otherwise, wait for clone children *only* if __WCLONE is
1105          * set; otherwise, wait for non-clone children *only*.  (Note:
1106          * A "clone" child here is one that reports to its parent
1107          * using a signal other than SIGCHLD.) */
1108         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1109             && !(options & __WALL))
1110                 return 0;
1111         /*
1112          * Do not consider thread group leaders that are
1113          * in a non-empty thread group:
1114          */
1115         if (delay_group_leader(p))
1116                 return 2;
1117
1118         err = security_task_wait(p);
1119         if (err)
1120                 return err;
1121
1122         return 1;
1123 }
1124
1125 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1126                                int why, int status,
1127                                struct siginfo __user *infop,
1128                                struct rusage __user *rusagep)
1129 {
1130         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1131
1132         put_task_struct(p);
1133         if (!retval)
1134                 retval = put_user(SIGCHLD, &infop->si_signo);
1135         if (!retval)
1136                 retval = put_user(0, &infop->si_errno);
1137         if (!retval)
1138                 retval = put_user((short)why, &infop->si_code);
1139         if (!retval)
1140                 retval = put_user(pid, &infop->si_pid);
1141         if (!retval)
1142                 retval = put_user(uid, &infop->si_uid);
1143         if (!retval)
1144                 retval = put_user(status, &infop->si_status);
1145         if (!retval)
1146                 retval = pid;
1147         return retval;
1148 }
1149
1150 /*
1151  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1152  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1153  * the lock and this task is uninteresting.  If we return nonzero, we have
1154  * released the lock and the system call should return.
1155  */
1156 static int wait_task_zombie(struct task_struct *p, int noreap,
1157                             struct siginfo __user *infop,
1158                             int __user *stat_addr, struct rusage __user *ru)
1159 {
1160         unsigned long state;
1161         int retval;
1162         int status;
1163
1164         if (unlikely(noreap)) {
1165                 pid_t pid = p->pid;
1166                 uid_t uid = p->uid;
1167                 int exit_code = p->exit_code;
1168                 int why, status;
1169
1170                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1171                         return 0;
1172                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1173                         return 0;
1174                 get_task_struct(p);
1175                 read_unlock(&tasklist_lock);
1176                 if ((exit_code & 0x7f) == 0) {
1177                         why = CLD_EXITED;
1178                         status = exit_code >> 8;
1179                 } else {
1180                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1181                         status = exit_code & 0x7f;
1182                 }
1183                 return wait_noreap_copyout(p, pid, uid, why,
1184                                            status, infop, ru);
1185         }
1186
1187         /*
1188          * Try to move the task's state to DEAD
1189          * only one thread is allowed to do this:
1190          */
1191         state = xchg(&p->exit_state, EXIT_DEAD);
1192         if (state != EXIT_ZOMBIE) {
1193                 BUG_ON(state != EXIT_DEAD);
1194                 return 0;
1195         }
1196         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1197                 /*
1198                  * This can only happen in a race with a ptraced thread
1199                  * dying on another processor.
1200                  */
1201                 return 0;
1202         }
1203
1204         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1205                 struct signal_struct *psig;
1206                 struct signal_struct *sig;
1207
1208                 /*
1209                  * The resource counters for the group leader are in its
1210                  * own task_struct.  Those for dead threads in the group
1211                  * are in its signal_struct, as are those for the child
1212                  * processes it has previously reaped.  All these
1213                  * accumulate in the parent's signal_struct c* fields.
1214                  *
1215                  * We don't bother to take a lock here to protect these
1216                  * p->signal fields, because they are only touched by
1217                  * __exit_signal, which runs with tasklist_lock
1218                  * write-locked anyway, and so is excluded here.  We do
1219                  * need to protect the access to p->parent->signal fields,
1220                  * as other threads in the parent group can be right
1221                  * here reaping other children at the same time.
1222                  */
1223                 spin_lock_irq(&p->parent->sighand->siglock);
1224                 psig = p->parent->signal;
1225                 sig = p->signal;
1226                 psig->cutime =
1227                         cputime_add(psig->cutime,
1228                         cputime_add(p->utime,
1229                         cputime_add(sig->utime,
1230                                     sig->cutime)));
1231                 psig->cstime =
1232                         cputime_add(psig->cstime,
1233                         cputime_add(p->stime,
1234                         cputime_add(sig->stime,
1235                                     sig->cstime)));
1236                 psig->cgtime =
1237                         cputime_add(psig->cgtime,
1238                         cputime_add(p->gtime,
1239                         cputime_add(sig->gtime,
1240                                     sig->cgtime)));
1241                 psig->cmin_flt +=
1242                         p->min_flt + sig->min_flt + sig->cmin_flt;
1243                 psig->cmaj_flt +=
1244                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1245                 psig->cnvcsw +=
1246                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1247                 psig->cnivcsw +=
1248                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1249                 psig->cinblock +=
1250                         task_io_get_inblock(p) +
1251                         sig->inblock + sig->cinblock;
1252                 psig->coublock +=
1253                         task_io_get_oublock(p) +
1254                         sig->oublock + sig->coublock;
1255                 spin_unlock_irq(&p->parent->sighand->siglock);
1256         }
1257
1258         /*
1259          * Now we are sure this task is interesting, and no other
1260          * thread can reap it because we set its state to EXIT_DEAD.
1261          */
1262         read_unlock(&tasklist_lock);
1263
1264         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1265         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1266                 ? p->signal->group_exit_code : p->exit_code;
1267         if (!retval && stat_addr)
1268                 retval = put_user(status, stat_addr);
1269         if (!retval && infop)
1270                 retval = put_user(SIGCHLD, &infop->si_signo);
1271         if (!retval && infop)
1272                 retval = put_user(0, &infop->si_errno);
1273         if (!retval && infop) {
1274                 int why;
1275
1276                 if ((status & 0x7f) == 0) {
1277                         why = CLD_EXITED;
1278                         status >>= 8;
1279                 } else {
1280                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1281                         status &= 0x7f;
1282                 }
1283                 retval = put_user((short)why, &infop->si_code);
1284                 if (!retval)
1285                         retval = put_user(status, &infop->si_status);
1286         }
1287         if (!retval && infop)
1288                 retval = put_user(p->pid, &infop->si_pid);
1289         if (!retval && infop)
1290                 retval = put_user(p->uid, &infop->si_uid);
1291         if (retval) {
1292                 // TODO: is this safe?
1293                 p->exit_state = EXIT_ZOMBIE;
1294                 return retval;
1295         }
1296         retval = p->pid;
1297         if (p->real_parent != p->parent) {
1298                 write_lock_irq(&tasklist_lock);
1299                 /* Double-check with lock held.  */
1300                 if (p->real_parent != p->parent) {
1301                         __ptrace_unlink(p);
1302                         // TODO: is this safe?
1303                         p->exit_state = EXIT_ZOMBIE;
1304                         /*
1305                          * If this is not a detached task, notify the parent.
1306                          * If it's still not detached after that, don't release
1307                          * it now.
1308                          */
1309                         if (p->exit_signal != -1) {
1310                                 do_notify_parent(p, p->exit_signal);
1311                                 if (p->exit_signal != -1)
1312                                         p = NULL;
1313                         }
1314                 }
1315                 write_unlock_irq(&tasklist_lock);
1316         }
1317         if (p != NULL)
1318                 release_task(p);
1319         BUG_ON(!retval);
1320         return retval;
1321 }
1322
1323 /*
1324  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1325  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1326  * the lock and this task is uninteresting.  If we return nonzero, we have
1327  * released the lock and the system call should return.
1328  */
1329 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1330                              int noreap, struct siginfo __user *infop,
1331                              int __user *stat_addr, struct rusage __user *ru)
1332 {
1333         int retval, exit_code;
1334
1335         if (!p->exit_code)
1336                 return 0;
1337         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1338             p->signal && p->signal->group_stop_count > 0)
1339                 /*
1340                  * A group stop is in progress and this is the group leader.
1341                  * We won't report until all threads have stopped.
1342                  */
1343                 return 0;
1344
1345         /*
1346          * Now we are pretty sure this task is interesting.
1347          * Make sure it doesn't get reaped out from under us while we
1348          * give up the lock and then examine it below.  We don't want to
1349          * keep holding onto the tasklist_lock while we call getrusage and
1350          * possibly take page faults for user memory.
1351          */
1352         get_task_struct(p);
1353         read_unlock(&tasklist_lock);
1354
1355         if (unlikely(noreap)) {
1356                 pid_t pid = p->pid;
1357                 uid_t uid = p->uid;
1358                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1359
1360                 exit_code = p->exit_code;
1361                 if (unlikely(!exit_code) ||
1362                     unlikely(p->state & TASK_TRACED))
1363                         goto bail_ref;
1364                 return wait_noreap_copyout(p, pid, uid,
1365                                            why, (exit_code << 8) | 0x7f,
1366                                            infop, ru);
1367         }
1368
1369         write_lock_irq(&tasklist_lock);
1370
1371         /*
1372          * This uses xchg to be atomic with the thread resuming and setting
1373          * it.  It must also be done with the write lock held to prevent a
1374          * race with the EXIT_ZOMBIE case.
1375          */
1376         exit_code = xchg(&p->exit_code, 0);
1377         if (unlikely(p->exit_state)) {
1378                 /*
1379                  * The task resumed and then died.  Let the next iteration
1380                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1381                  * already be zero here if it resumed and did _exit(0).
1382                  * The task itself is dead and won't touch exit_code again;
1383                  * other processors in this function are locked out.
1384                  */
1385                 p->exit_code = exit_code;
1386                 exit_code = 0;
1387         }
1388         if (unlikely(exit_code == 0)) {
1389                 /*
1390                  * Another thread in this function got to it first, or it
1391                  * resumed, or it resumed and then died.
1392                  */
1393                 write_unlock_irq(&tasklist_lock);
1394 bail_ref:
1395                 put_task_struct(p);
1396                 /*
1397                  * We are returning to the wait loop without having successfully
1398                  * removed the process and having released the lock. We cannot
1399                  * continue, since the "p" task pointer is potentially stale.
1400                  *
1401                  * Return -EAGAIN, and do_wait() will restart the loop from the
1402                  * beginning. Do _not_ re-acquire the lock.
1403                  */
1404                 return -EAGAIN;
1405         }
1406
1407         /* move to end of parent's list to avoid starvation */
1408         remove_parent(p);
1409         add_parent(p);
1410
1411         write_unlock_irq(&tasklist_lock);
1412
1413         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1414         if (!retval && stat_addr)
1415                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1416         if (!retval && infop)
1417                 retval = put_user(SIGCHLD, &infop->si_signo);
1418         if (!retval && infop)
1419                 retval = put_user(0, &infop->si_errno);
1420         if (!retval && infop)
1421                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1422                                           ? CLD_TRAPPED : CLD_STOPPED),
1423                                   &infop->si_code);
1424         if (!retval && infop)
1425                 retval = put_user(exit_code, &infop->si_status);
1426         if (!retval && infop)
1427                 retval = put_user(p->pid, &infop->si_pid);
1428         if (!retval && infop)
1429                 retval = put_user(p->uid, &infop->si_uid);
1430         if (!retval)
1431                 retval = p->pid;
1432         put_task_struct(p);
1433
1434         BUG_ON(!retval);
1435         return retval;
1436 }
1437
1438 /*
1439  * Handle do_wait work for one task in a live, non-stopped state.
1440  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1441  * the lock and this task is uninteresting.  If we return nonzero, we have
1442  * released the lock and the system call should return.
1443  */
1444 static int wait_task_continued(struct task_struct *p, int noreap,
1445                                struct siginfo __user *infop,
1446                                int __user *stat_addr, struct rusage __user *ru)
1447 {
1448         int retval;
1449         pid_t pid;
1450         uid_t uid;
1451
1452         if (unlikely(!p->signal))
1453                 return 0;
1454
1455         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1456                 return 0;
1457
1458         spin_lock_irq(&p->sighand->siglock);
1459         /* Re-check with the lock held.  */
1460         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1461                 spin_unlock_irq(&p->sighand->siglock);
1462                 return 0;
1463         }
1464         if (!noreap)
1465                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1466         spin_unlock_irq(&p->sighand->siglock);
1467
1468         pid = p->pid;
1469         uid = p->uid;
1470         get_task_struct(p);
1471         read_unlock(&tasklist_lock);
1472
1473         if (!infop) {
1474                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1475                 put_task_struct(p);
1476                 if (!retval && stat_addr)
1477                         retval = put_user(0xffff, stat_addr);
1478                 if (!retval)
1479                         retval = p->pid;
1480         } else {
1481                 retval = wait_noreap_copyout(p, pid, uid,
1482                                              CLD_CONTINUED, SIGCONT,
1483                                              infop, ru);
1484                 BUG_ON(retval == 0);
1485         }
1486
1487         return retval;
1488 }
1489
1490
1491 static inline int my_ptrace_child(struct task_struct *p)
1492 {
1493         if (!(p->ptrace & PT_PTRACED))
1494                 return 0;
1495         if (!(p->ptrace & PT_ATTACHED))
1496                 return 1;
1497         /*
1498          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1499          * we are the attacher.  If we are the real parent, this is a race
1500          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1501          * which we have to switch the parent links, but has already set
1502          * the flags in p->ptrace.
1503          */
1504         return (p->parent != p->real_parent);
1505 }
1506
1507 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1508                     int __user *stat_addr, struct rusage __user *ru)
1509 {
1510         DECLARE_WAITQUEUE(wait, current);
1511         struct task_struct *tsk;
1512         int flag, retval;
1513         int allowed, denied;
1514
1515         add_wait_queue(&current->signal->wait_chldexit,&wait);
1516 repeat:
1517         /*
1518          * We will set this flag if we see any child that might later
1519          * match our criteria, even if we are not able to reap it yet.
1520          */
1521         flag = 0;
1522         allowed = denied = 0;
1523         current->state = TASK_INTERRUPTIBLE;
1524         read_lock(&tasklist_lock);
1525         tsk = current;
1526         do {
1527                 struct task_struct *p;
1528                 struct list_head *_p;
1529                 int ret;
1530
1531                 list_for_each(_p,&tsk->children) {
1532                         p = list_entry(_p, struct task_struct, sibling);
1533
1534                         ret = eligible_child(pid, options, p);
1535                         if (!ret)
1536                                 continue;
1537
1538                         if (unlikely(ret < 0)) {
1539                                 denied = ret;
1540                                 continue;
1541                         }
1542                         allowed = 1;
1543
1544                         switch (p->state) {
1545                         case TASK_TRACED:
1546                                 /*
1547                                  * When we hit the race with PTRACE_ATTACH,
1548                                  * we will not report this child.  But the
1549                                  * race means it has not yet been moved to
1550                                  * our ptrace_children list, so we need to
1551                                  * set the flag here to avoid a spurious ECHILD
1552                                  * when the race happens with the only child.
1553                                  */
1554                                 flag = 1;
1555                                 if (!my_ptrace_child(p))
1556                                         continue;
1557                                 /*FALLTHROUGH*/
1558                         case TASK_STOPPED:
1559                                 /*
1560                                  * It's stopped now, so it might later
1561                                  * continue, exit, or stop again.
1562                                  */
1563                                 flag = 1;
1564                                 if (!(options & WUNTRACED) &&
1565                                     !my_ptrace_child(p))
1566                                         continue;
1567                                 retval = wait_task_stopped(p, ret == 2,
1568                                                            (options & WNOWAIT),
1569                                                            infop,
1570                                                            stat_addr, ru);
1571                                 if (retval == -EAGAIN)
1572                                         goto repeat;
1573                                 if (retval != 0) /* He released the lock.  */
1574                                         goto end;
1575                                 break;
1576                         default:
1577                         // case EXIT_DEAD:
1578                                 if (p->exit_state == EXIT_DEAD)
1579                                         continue;
1580                         // case EXIT_ZOMBIE:
1581                                 if (p->exit_state == EXIT_ZOMBIE) {
1582                                         /*
1583                                          * Eligible but we cannot release
1584                                          * it yet:
1585                                          */
1586                                         if (ret == 2)
1587                                                 goto check_continued;
1588                                         if (!likely(options & WEXITED))
1589                                                 continue;
1590                                         retval = wait_task_zombie(
1591                                                 p, (options & WNOWAIT),
1592                                                 infop, stat_addr, ru);
1593                                         /* He released the lock.  */
1594                                         if (retval != 0)
1595                                                 goto end;
1596                                         break;
1597                                 }
1598 check_continued:
1599                                 /*
1600                                  * It's running now, so it might later
1601                                  * exit, stop, or stop and then continue.
1602                                  */
1603                                 flag = 1;
1604                                 if (!unlikely(options & WCONTINUED))
1605                                         continue;
1606                                 retval = wait_task_continued(
1607                                         p, (options & WNOWAIT),
1608                                         infop, stat_addr, ru);
1609                                 if (retval != 0) /* He released the lock.  */
1610                                         goto end;
1611                                 break;
1612                         }
1613                 }
1614                 if (!flag) {
1615                         list_for_each(_p, &tsk->ptrace_children) {
1616                                 p = list_entry(_p, struct task_struct,
1617                                                 ptrace_list);
1618                                 if (!eligible_child(pid, options, p))
1619                                         continue;
1620                                 flag = 1;
1621                                 break;
1622                         }
1623                 }
1624                 if (options & __WNOTHREAD)
1625                         break;
1626                 tsk = next_thread(tsk);
1627                 BUG_ON(tsk->signal != current->signal);
1628         } while (tsk != current);
1629
1630         read_unlock(&tasklist_lock);
1631         if (flag) {
1632                 retval = 0;
1633                 if (options & WNOHANG)
1634                         goto end;
1635                 retval = -ERESTARTSYS;
1636                 if (signal_pending(current))
1637                         goto end;
1638                 schedule();
1639                 goto repeat;
1640         }
1641         retval = -ECHILD;
1642         if (unlikely(denied) && !allowed)
1643                 retval = denied;
1644 end:
1645         current->state = TASK_RUNNING;
1646         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1647         if (infop) {
1648                 if (retval > 0)
1649                 retval = 0;
1650                 else {
1651                         /*
1652                          * For a WNOHANG return, clear out all the fields
1653                          * we would set so the user can easily tell the
1654                          * difference.
1655                          */
1656                         if (!retval)
1657                                 retval = put_user(0, &infop->si_signo);
1658                         if (!retval)
1659                                 retval = put_user(0, &infop->si_errno);
1660                         if (!retval)
1661                                 retval = put_user(0, &infop->si_code);
1662                         if (!retval)
1663                                 retval = put_user(0, &infop->si_pid);
1664                         if (!retval)
1665                                 retval = put_user(0, &infop->si_uid);
1666                         if (!retval)
1667                                 retval = put_user(0, &infop->si_status);
1668                 }
1669         }
1670         return retval;
1671 }
1672
1673 asmlinkage long sys_waitid(int which, pid_t pid,
1674                            struct siginfo __user *infop, int options,
1675                            struct rusage __user *ru)
1676 {
1677         long ret;
1678
1679         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1680                 return -EINVAL;
1681         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1682                 return -EINVAL;
1683
1684         switch (which) {
1685         case P_ALL:
1686                 pid = -1;
1687                 break;
1688         case P_PID:
1689                 if (pid <= 0)
1690                         return -EINVAL;
1691                 break;
1692         case P_PGID:
1693                 if (pid <= 0)
1694                         return -EINVAL;
1695                 pid = -pid;
1696                 break;
1697         default:
1698                 return -EINVAL;
1699         }
1700
1701         ret = do_wait(pid, options, infop, NULL, ru);
1702
1703         /* avoid REGPARM breakage on x86: */
1704         prevent_tail_call(ret);
1705         return ret;
1706 }
1707
1708 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1709                           int options, struct rusage __user *ru)
1710 {
1711         long ret;
1712
1713         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1714                         __WNOTHREAD|__WCLONE|__WALL))
1715                 return -EINVAL;
1716         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1717
1718         /* avoid REGPARM breakage on x86: */
1719         prevent_tail_call(ret);
1720         return ret;
1721 }
1722
1723 #ifdef __ARCH_WANT_SYS_WAITPID
1724
1725 /*
1726  * sys_waitpid() remains for compatibility. waitpid() should be
1727  * implemented by calling sys_wait4() from libc.a.
1728  */
1729 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1730 {
1731         return sys_wait4(pid, stat_addr, options, NULL);
1732 }
1733
1734 #endif