kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/mm.h>
   8 #include <linux/slab.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/smp_lock.h>
  11 #include <linux/module.h>
  12 #include <linux/capability.h>
  13 #include <linux/completion.h>
  14 #include <linux/personality.h>
  15 #include <linux/tty.h>
  16 #include <linux/mnt_namespace.h>
  17 #include <linux/key.h>
  18 #include <linux/security.h>
  19 #include <linux/cpu.h>
  20 #include <linux/acct.h>
  21 #include <linux/tsacct_kern.h>
  22 #include <linux/file.h>
  23 #include <linux/binfmts.h>
  24 #include <linux/nsproxy.h>
  25 #include <linux/pid_namespace.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/profile.h>
  28 #include <linux/mount.h>
  29 #include <linux/proc_fs.h>
  30 #include <linux/mempolicy.h>
  31 #include <linux/taskstats_kern.h>
  32 #include <linux/delayacct.h>
  33 #include <linux/cpuset.h>
  34 #include <linux/syscalls.h>
  35 #include <linux/signal.h>
  36 #include <linux/posix-timers.h>
  37 #include <linux/cn_proc.h>
  38 #include <linux/mutex.h>
  39 #include <linux/futex.h>
  40 #include <linux/compat.h>
  41 #include <linux/pipe_fs_i.h>
  42 #include <linux/audit.h> /* for audit_free() */
  43 #include <linux/resource.h>
  44 #include <linux/blkdev.h>
  45
  46 #include <asm/uaccess.h>
  47 #include <asm/unistd.h>
  48 #include <asm/pgtable.h>
  49 #include <asm/mmu_context.h>
  50
  51 extern void sem_exit (void);
  52
  53 static void exit_mm(struct task_struct * tsk);
  54
  55 static void __unhash_process(struct task_struct *p)
  56 {
  57         nr_threads--;
  58         detach_pid(p, PIDTYPE_PID);
  59         if (thread_group_leader(p)) {
  60                 detach_pid(p, PIDTYPE_PGID);
  61                 detach_pid(p, PIDTYPE_SID);
  62
  63                 list_del_rcu(&p->tasks);
  64                 __get_cpu_var(process_counts)--;
  65         }
  66         list_del_rcu(&p->thread_group);
  67         remove_parent(p);
  68 }
  69
  70 /*
  71  * This function expects the tasklist_lock write-locked.
  72  */
  73 static void __exit_signal(struct task_struct *tsk)
  74 {
  75         struct signal_struct *sig = tsk->signal;
  76         struct sighand_struct *sighand;
  77
  78         BUG_ON(!sig);
  79         BUG_ON(!atomic_read(&sig->count));
  80
  81         rcu_read_lock();
  82         sighand = rcu_dereference(tsk->sighand);
  83         spin_lock(&sighand->siglock);
  84
  85         posix_cpu_timers_exit(tsk);
  86         if (atomic_dec_and_test(&sig->count))
  87                 posix_cpu_timers_exit_group(tsk);
  88         else {
  89                 /*
  90                  * If there is any task waiting for the group exit
  91                  * then notify it:
  92                  */
  93                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
  94                         wake_up_process(sig->group_exit_task);
  95                         sig->group_exit_task = NULL;
  96                 }
  97                 if (tsk == sig->curr_target)
  98                         sig->curr_target = next_thread(tsk);
  99                 /*
 100                  * Accumulate here the counters for all threads but the
 101                  * group leader as they die, so they can be added into
 102                  * the process-wide totals when those are taken.
 103                  * The group leader stays around as a zombie as long
 104                  * as there are other threads.  When it gets reaped,
 105                  * the exit.c code will add its counts into these totals.
 106                  * We won't ever get here for the group leader, since it
 107                  * will have been the last reference on the signal_struct.
 108                  */
 109                 sig->utime = cputime_add(sig->utime, tsk->utime);
 110                 sig->stime = cputime_add(sig->stime, tsk->stime);
 111                 sig->min_flt += tsk->min_flt;
 112                 sig->maj_flt += tsk->maj_flt;
 113                 sig->nvcsw += tsk->nvcsw;
 114                 sig->nivcsw += tsk->nivcsw;
 115                 sig->sched_time += tsk->sched_time;
 116                 sig = NULL; /* Marker for below. */
 117         }
 118
 119         __unhash_process(tsk);
 120
 121         tsk->signal = NULL;
 122         tsk->sighand = NULL;
 123         spin_unlock(&sighand->siglock);
 124         rcu_read_unlock();
 125
 126         __cleanup_sighand(sighand);
 127         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 128         flush_sigqueue(&tsk->pending);
 129         if (sig) {
 130                 flush_sigqueue(&sig->shared_pending);
 131                 taskstats_tgid_free(sig);
 132                 __cleanup_signal(sig);
 133         }
 134 }
 135
 136 static void delayed_put_task_struct(struct rcu_head *rhp)
 137 {
 138         put_task_struct(container_of(rhp, struct task_struct, rcu));
 139 }
 140
 141 void release_task(struct task_struct * p)
 142 {
 143         struct task_struct *leader;
 144         int zap_leader;
 145 repeat:
 146         atomic_dec(&p->user->processes);
 147         write_lock_irq(&tasklist_lock);
 148         ptrace_unlink(p);
 149         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
 150         __exit_signal(p);
 151
 152         /*
 153          * If we are the last non-leader member of the thread
 154          * group, and the leader is zombie, then notify the
 155          * group leader's parent process. (if it wants notification.)
 156          */
 157         zap_leader = 0;
 158         leader = p->group_leader;
 159         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 160                 BUG_ON(leader->exit_signal == -1);
 161                 do_notify_parent(leader, leader->exit_signal);
 162                 /*
 163                  * If we were the last child thread and the leader has
 164                  * exited already, and the leader's parent ignores SIGCHLD,
 165                  * then we are the one who should release the leader.
 166                  *
 167                  * do_notify_parent() will have marked it self-reaping in
 168                  * that case.
 169                  */
 170                 zap_leader = (leader->exit_signal == -1);
 171         }
 172
 173         sched_exit(p);
 174         write_unlock_irq(&tasklist_lock);
 175         proc_flush_task(p);
 176         release_thread(p);
 177         call_rcu(&p->rcu, delayed_put_task_struct);
 178
 179         p = leader;
 180         if (unlikely(zap_leader))
 181                 goto repeat;
 182 }
 183
 184 /*
 185  * This checks not only the pgrp, but falls back on the pid if no
 186  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 187  * without this...
 188  */
 189 int session_of_pgrp(int pgrp)
 190 {
 191         struct task_struct *p;
 192         int sid = 0;
 193
 194         read_lock(&tasklist_lock);
 195
 196         p = find_task_by_pid_type(PIDTYPE_PGID, pgrp);
 197         if (p == NULL)
 198                 p = find_task_by_pid(pgrp);
 199         if (p != NULL)
 200                 sid = process_session(p);
 201
 202         read_unlock(&tasklist_lock);
 203
 204         return sid;
 205 }
 206
 207 /*
 208  * Determine if a process group is "orphaned", according to the POSIX
 209  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 210  * by terminal-generated stop signals.  Newly orphaned process groups are
 211  * to receive a SIGHUP and a SIGCONT.
 212  *
 213  * "I ask you, have you ever known what it is to be an orphan?"
 214  */
 215 static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
 216 {
 217         struct task_struct *p;
 218         int ret = 1;
 219
 220         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 221                 if (p == ignored_task
 222                                 || p->exit_state
 223                                 || is_init(p->real_parent))
 224                         continue;
 225                 if (process_group(p->real_parent) != pgrp &&
 226                     process_session(p->real_parent) == process_session(p)) {
 227                         ret = 0;
 228                         break;
 229                 }
 230         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 231         return ret;     /* (sighing) "Often!" */
 232 }
 233
 234 int is_orphaned_pgrp(int pgrp)
 235 {
 236         int retval;
 237
 238         read_lock(&tasklist_lock);
 239         retval = will_become_orphaned_pgrp(pgrp, NULL);
 240         read_unlock(&tasklist_lock);
 241
 242         return retval;
 243 }
 244
 245 static int has_stopped_jobs(int pgrp)
 246 {
 247         int retval = 0;
 248         struct task_struct *p;
 249
 250         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 251                 if (p->state != TASK_STOPPED)
 252                         continue;
 253                 retval = 1;
 254                 break;
 255         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 256         return retval;
 257 }
 258
 259 /**
 260  * reparent_to_init - Reparent the calling kernel thread to the init task
 261  * of the pid space that the thread belongs to.
 262  *
 263  * If a kernel thread is launched as a result of a system call, or if
 264  * it ever exits, it should generally reparent itself to init so that
 265  * it is correctly cleaned up on exit.
 266  *
 267  * The various task state such as scheduling policy and priority may have
 268  * been inherited from a user process, so we reset them to sane values here.
 269  *
 270  * NOTE that reparent_to_init() gives the caller full capabilities.
 271  */
 272 static void reparent_to_init(void)
 273 {
 274         write_lock_irq(&tasklist_lock);
 275
 276         ptrace_unlink(current);
 277         /* Reparent to init */
 278         remove_parent(current);
 279         current->parent = child_reaper(current);
 280         current->real_parent = child_reaper(current);
 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 (!has_rt_policy(current) && (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, 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, 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         sigaddset(&current->blocked, sig);
 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         set_special_pids(1, 1);
 385         proc_clear_tty(current);
 386
 387         /* Block and flush all signals */
 388         sigfillset(&blocked);
 389         sigprocmask(SIG_BLOCK, &blocked, NULL);
 390         flush_signals(current);
 391
 392         /* Become as one with the init task */
 393
 394         exit_fs(current);       /* current->fs->count--; */
 395         fs = init_task.fs;
 396         current->fs = fs;
 397         atomic_inc(&fs->count);
 398
 399         exit_task_namespaces(current);
 400         current->nsproxy = init_task.nsproxy;
 401         get_task_namespaces(current);
 402
 403         exit_files(current);
 404         current->files = init_task.files;
 405         atomic_inc(&current->files->count);
 406
 407         reparent_to_init();
 408 }
 409
 410 EXPORT_SYMBOL(daemonize);
 411
 412 static void close_files(struct files_struct * files)
 413 {
 414         int i, j;
 415         struct fdtable *fdt;
 416
 417         j = 0;
 418
 419         /*
 420          * It is safe to dereference the fd table without RCU or
 421          * ->file_lock because this is the last reference to the
 422          * files structure.
 423          */
 424         fdt = files_fdtable(files);
 425         for (;;) {
 426                 unsigned long set;
 427                 i = j * __NFDBITS;
 428                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 429                         break;
 430                 set = fdt->open_fds->fds_bits[j++];
 431                 while (set) {
 432                         if (set & 1) {
 433                                 struct file * file = xchg(&fdt->fd[i], NULL);
 434                                 if (file)
 435                                         filp_close(file, files);
 436                         }
 437                         i++;
 438                         set >>= 1;
 439                 }
 440         }
 441 }
 442
 443 struct files_struct *get_files_struct(struct task_struct *task)
 444 {
 445         struct files_struct *files;
 446
 447         task_lock(task);
 448         files = task->files;
 449         if (files)
 450                 atomic_inc(&files->count);
 451         task_unlock(task);
 452
 453         return files;
 454 }
 455
 456 void fastcall put_files_struct(struct files_struct *files)
 457 {
 458         struct fdtable *fdt;
 459
 460         if (atomic_dec_and_test(&files->count)) {
 461                 close_files(files);
 462                 /*
 463                  * Free the fd and fdset arrays if we expanded them.
 464                  * If the fdtable was embedded, pass files for freeing
 465                  * at the end of the RCU grace period. Otherwise,
 466                  * you can free files immediately.
 467                  */
 468                 fdt = files_fdtable(files);
 469                 if (fdt == &files->fdtab)
 470                         fdt->free_files = files;
 471                 else
 472                         kmem_cache_free(files_cachep, files);
 473                 free_fdtable(fdt);
 474         }
 475 }
 476
 477 EXPORT_SYMBOL(put_files_struct);
 478
 479 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
 480 {
 481         struct files_struct *old;
 482
 483         old = tsk->files;
 484         task_lock(tsk);
 485         tsk->files = files;
 486         task_unlock(tsk);
 487         put_files_struct(old);
 488 }
 489 EXPORT_SYMBOL(reset_files_struct);
 490
 491 static inline void __exit_files(struct task_struct *tsk)
 492 {
 493         struct files_struct * files = tsk->files;
 494
 495         if (files) {
 496                 task_lock(tsk);
 497                 tsk->files = NULL;
 498                 task_unlock(tsk);
 499                 put_files_struct(files);
 500         }
 501 }
 502
 503 void exit_files(struct task_struct *tsk)
 504 {
 505         __exit_files(tsk);
 506 }
 507
 508 static inline void __put_fs_struct(struct fs_struct *fs)
 509 {
 510         /* No need to hold fs->lock if we are killing it */
 511         if (atomic_dec_and_test(&fs->count)) {
 512                 dput(fs->root);
 513                 mntput(fs->rootmnt);
 514                 dput(fs->pwd);
 515                 mntput(fs->pwdmnt);
 516                 if (fs->altroot) {
 517                         dput(fs->altroot);
 518                         mntput(fs->altrootmnt);
 519                 }
 520                 kmem_cache_free(fs_cachep, fs);
 521         }
 522 }
 523
 524 void put_fs_struct(struct fs_struct *fs)
 525 {
 526         __put_fs_struct(fs);
 527 }
 528
 529 static inline void __exit_fs(struct task_struct *tsk)
 530 {
 531         struct fs_struct * fs = tsk->fs;
 532
 533         if (fs) {
 534                 task_lock(tsk);
 535                 tsk->fs = NULL;
 536                 task_unlock(tsk);
 537                 __put_fs_struct(fs);
 538         }
 539 }
 540
 541 void exit_fs(struct task_struct *tsk)
 542 {
 543         __exit_fs(tsk);
 544 }
 545
 546 EXPORT_SYMBOL_GPL(exit_fs);
 547
 548 /*
 549  * Turn us into a lazy TLB process if we
 550  * aren't already..
 551  */
 552 static void exit_mm(struct task_struct * tsk)
 553 {
 554         struct mm_struct *mm = tsk->mm;
 555
 556         mm_release(tsk, mm);
 557         if (!mm)
 558                 return;
 559         /*
 560          * Serialize with any possible pending coredump.
 561          * We must hold mmap_sem around checking core_waiters
 562          * and clearing tsk->mm.  The core-inducing thread
 563          * will increment core_waiters for each thread in the
 564          * group with ->mm != NULL.
 565          */
 566         down_read(&mm->mmap_sem);
 567         if (mm->core_waiters) {
 568                 up_read(&mm->mmap_sem);
 569                 down_write(&mm->mmap_sem);
 570                 if (!--mm->core_waiters)
 571                         complete(mm->core_startup_done);
 572                 up_write(&mm->mmap_sem);
 573
 574                 wait_for_completion(&mm->core_done);
 575                 down_read(&mm->mmap_sem);
 576         }
 577         atomic_inc(&mm->mm_count);
 578         BUG_ON(mm != tsk->active_mm);
 579         /* more a memory barrier than a real lock */
 580         task_lock(tsk);
 581         tsk->mm = NULL;
 582         up_read(&mm->mmap_sem);
 583         enter_lazy_tlb(mm, current);
 584         task_unlock(tsk);
 585         mmput(mm);
 586 }
 587
 588 static inline void
 589 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 590 {
 591         /*
 592          * Make sure we're not reparenting to ourselves and that
 593          * the parent is not a zombie.
 594          */
 595         BUG_ON(p == reaper || reaper->exit_state);
 596         p->real_parent = reaper;
 597 }
 598
 599 static void
 600 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 601 {
 602         /* We don't want people slaying init.  */
 603         if (p->exit_signal != -1)
 604                 p->exit_signal = SIGCHLD;
 605
 606         if (p->pdeath_signal)
 607                 /* We already hold the tasklist_lock here.  */
 608                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 609
 610         /* Move the child from its dying parent to the new one.  */
 611         if (unlikely(traced)) {
 612                 /* Preserve ptrace links if someone else is tracing this child.  */
 613                 list_del_init(&p->ptrace_list);
 614                 if (p->parent != p->real_parent)
 615                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 616         } else {
 617                 /* If this child is being traced, then we're the one tracing it
 618                  * anyway, so let go of it.
 619                  */
 620                 p->ptrace = 0;
 621                 remove_parent(p);
 622                 p->parent = p->real_parent;
 623                 add_parent(p);
 624
 625                 /* If we'd notified the old parent about this child's death,
 626                  * also notify the new parent.
 627                  */
 628                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 629                     thread_group_empty(p))
 630                         do_notify_parent(p, p->exit_signal);
 631                 else if (p->state == TASK_TRACED) {
 632                         /*
 633                          * If it was at a trace stop, turn it into
 634                          * a normal stop since it's no longer being
 635                          * traced.
 636                          */
 637                         ptrace_untrace(p);
 638                 }
 639         }
 640
 641         /*
 642          * process group orphan check
 643          * Case ii: Our child is in a different pgrp
 644          * than we are, and it was the only connection
 645          * outside, so the child pgrp is now orphaned.
 646          */
 647         if ((process_group(p) != process_group(father)) &&
 648             (process_session(p) == process_session(father))) {
 649                 int pgrp = process_group(p);
 650
 651                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
 652                     has_stopped_jobs(pgrp)) {
 653                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 654                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 655                 }
 656         }
 657 }
 658
 659 /*
 660  * When we die, we re-parent all our children.
 661  * Try to give them to another thread in our thread
 662  * group, and if no such member exists, give it to
 663  * the child reaper process (ie "init") in our pid
 664  * space.
 665  */
 666 static void
 667 forget_original_parent(struct task_struct *father, struct list_head *to_release)
 668 {
 669         struct task_struct *p, *reaper = father;
 670         struct list_head *_p, *_n;
 671
 672         do {
 673                 reaper = next_thread(reaper);
 674                 if (reaper == father) {
 675                         reaper = child_reaper(father);
 676                         break;
 677                 }
 678         } while (reaper->exit_state);
 679
 680         /*
 681          * There are only two places where our children can be:
 682          *
 683          * - in our child list
 684          * - in our ptraced child list
 685          *
 686          * Search them and reparent children.
 687          */
 688         list_for_each_safe(_p, _n, &father->children) {
 689                 int ptrace;
 690                 p = list_entry(_p, struct task_struct, sibling);
 691
 692                 ptrace = p->ptrace;
 693
 694                 /* if father isn't the real parent, then ptrace must be enabled */
 695                 BUG_ON(father != p->real_parent && !ptrace);
 696
 697                 if (father == p->real_parent) {
 698                         /* reparent with a reaper, real father it's us */
 699                         choose_new_parent(p, reaper);
 700                         reparent_thread(p, father, 0);
 701                 } else {
 702                         /* reparent ptraced task to its real parent */
 703                         __ptrace_unlink (p);
 704                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 705                             thread_group_empty(p))
 706                                 do_notify_parent(p, p->exit_signal);
 707                 }
 708
 709                 /*
 710                  * if the ptraced child is a zombie with exit_signal == -1
 711                  * we must collect it before we exit, or it will remain
 712                  * zombie forever since we prevented it from self-reap itself
 713                  * while it was being traced by us, to be able to see it in wait4.
 714                  */
 715                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 716                         list_add(&p->ptrace_list, to_release);
 717         }
 718         list_for_each_safe(_p, _n, &father->ptrace_children) {
 719                 p = list_entry(_p, struct task_struct, ptrace_list);
 720                 choose_new_parent(p, reaper);
 721                 reparent_thread(p, father, 1);
 722         }
 723 }
 724
 725 /*
 726  * Send signals to all our closest relatives so that they know
 727  * to properly mourn us..
 728  */
 729 static void exit_notify(struct task_struct *tsk)
 730 {
 731         int state;
 732         struct task_struct *t;
 733         struct list_head ptrace_dead, *_p, *_n;
 734
 735         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 736             && !thread_group_empty(tsk)) {
 737                 /*
 738                  * This occurs when there was a race between our exit
 739                  * syscall and a group signal choosing us as the one to
 740                  * wake up.  It could be that we are the only thread
 741                  * alerted to check for pending signals, but another thread
 742                  * should be woken now to take the signal since we will not.
 743                  * Now we'll wake all the threads in the group just to make
 744                  * sure someone gets all the pending signals.
 745                  */
 746                 read_lock(&tasklist_lock);
 747                 spin_lock_irq(&tsk->sighand->siglock);
 748                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 749                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 750                                 recalc_sigpending_tsk(t);
 751                                 if (signal_pending(t))
 752                                         signal_wake_up(t, 0);
 753                         }
 754                 spin_unlock_irq(&tsk->sighand->siglock);
 755                 read_unlock(&tasklist_lock);
 756         }
 757
 758         write_lock_irq(&tasklist_lock);
 759
 760         /*
 761          * This does two things:
 762          *
 763          * A.  Make init inherit all the child processes
 764          * B.  Check to see if any process groups have become orphaned
 765          *      as a result of our exiting, and if they have any stopped
 766          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 767          */
 768
 769         INIT_LIST_HEAD(&ptrace_dead);
 770         forget_original_parent(tsk, &ptrace_dead);
 771         BUG_ON(!list_empty(&tsk->children));
 772         BUG_ON(!list_empty(&tsk->ptrace_children));
 773
 774         /*
 775          * Check to see if any process groups have become orphaned
 776          * as a result of our exiting, and if they have any stopped
 777          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 778          *
 779          * Case i: Our father is in a different pgrp than we are
 780          * and we were the only connection outside, so our pgrp
 781          * is about to become orphaned.
 782          */
 783
 784         t = tsk->real_parent;
 785
 786         if ((process_group(t) != process_group(tsk)) &&
 787             (process_session(t) == process_session(tsk)) &&
 788             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 789             has_stopped_jobs(process_group(tsk))) {
 790                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 791                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 792         }
 793
 794         /* Let father know we died
 795          *
 796          * Thread signals are configurable, but you aren't going to use
 797          * that to send signals to arbitary processes.
 798          * That stops right now.
 799          *
 800          * If the parent exec id doesn't match the exec id we saved
 801          * when we started then we know the parent has changed security
 802          * domain.
 803          *
 804          * If our self_exec id doesn't match our parent_exec_id then
 805          * we have changed execution domain as these two values started
 806          * the same after a fork.
 807          *
 808          */
 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 &&
 830             (likely(tsk->ptrace == 0) ||
 831              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 832                 state = EXIT_DEAD;
 833         tsk->exit_state = state;
 834
 835         write_unlock_irq(&tasklist_lock);
 836
 837         list_for_each_safe(_p, _n, &ptrace_dead) {
 838                 list_del_init(_p);
 839                 t = list_entry(_p, struct task_struct, ptrace_list);
 840                 release_task(t);
 841         }
 842
 843         /* If the process is dead, release it - nobody will wait for it */
 844         if (state == EXIT_DEAD)
 845                 release_task(tsk);
 846 }
 847
 848 fastcall NORET_TYPE void do_exit(long code)
 849 {
 850         struct task_struct *tsk = current;
 851         int group_dead;
 852
 853         profile_task_exit(tsk);
 854
 855         WARN_ON(atomic_read(&tsk->fs_excl));
 856
 857         if (unlikely(in_interrupt()))
 858                 panic("Aiee, killing interrupt handler!");
 859         if (unlikely(!tsk->pid))
 860                 panic("Attempted to kill the idle task!");
 861         if (unlikely(tsk == child_reaper(tsk))) {
 862                 if (tsk->nsproxy->pid_ns != &init_pid_ns)
 863                         tsk->nsproxy->pid_ns->child_reaper = init_pid_ns.child_reaper;
 864                 else
 865                         panic("Attempted to kill init!");
 866         }
 867
 868
 869         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 870                 current->ptrace_message = code;
 871                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 872         }
 873
 874         /*
 875          * We're taking recursive faults here in do_exit. Safest is to just
 876          * leave this task alone and wait for reboot.
 877          */
 878         if (unlikely(tsk->flags & PF_EXITING)) {
 879                 printk(KERN_ALERT
 880                         "Fixing recursive fault but reboot is needed!\n");
 881                 if (tsk->io_context)
 882                         exit_io_context();
 883                 set_current_state(TASK_UNINTERRUPTIBLE);
 884                 schedule();
 885         }
 886
 887         tsk->flags |= PF_EXITING;
 888
 889         if (unlikely(in_atomic()))
 890                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 891                                 current->comm, current->pid,
 892                                 preempt_count());
 893
 894         acct_update_integrals(tsk);
 895         if (tsk->mm) {
 896                 update_hiwater_rss(tsk->mm);
 897                 update_hiwater_vm(tsk->mm);
 898         }
 899         group_dead = atomic_dec_and_test(&tsk->signal->live);
 900         if (group_dead) {
 901                 hrtimer_cancel(&tsk->signal->real_timer);
 902                 exit_itimers(tsk->signal);
 903         }
 904         acct_collect(code, group_dead);
 905         if (unlikely(tsk->robust_list))
 906                 exit_robust_list(tsk);
 907 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
 908         if (unlikely(tsk->compat_robust_list))
 909                 compat_exit_robust_list(tsk);
 910 #endif
 911         if (unlikely(tsk->audit_context))
 912                 audit_free(tsk);
 913
 914         taskstats_exit(tsk, group_dead);
 915
 916         exit_mm(tsk);
 917
 918         if (group_dead)
 919                 acct_process();
 920         exit_sem(tsk);
 921         __exit_files(tsk);
 922         __exit_fs(tsk);
 923         exit_thread();
 924         cpuset_exit(tsk);
 925         exit_keys(tsk);
 926
 927         if (group_dead && tsk->signal->leader)
 928                 disassociate_ctty(1);
 929
 930         module_put(task_thread_info(tsk)->exec_domain->module);
 931         if (tsk->binfmt)
 932                 module_put(tsk->binfmt->module);
 933
 934         tsk->exit_code = code;
 935         proc_exit_connector(tsk);
 936         exit_notify(tsk);
 937         exit_task_namespaces(tsk);
 938 #ifdef CONFIG_NUMA
 939         mpol_free(tsk->mempolicy);
 940         tsk->mempolicy = NULL;
 941 #endif
 942         /*
 943          * This must happen late, after the PID is not
 944          * hashed anymore:
 945          */
 946         if (unlikely(!list_empty(&tsk->pi_state_list)))
 947                 exit_pi_state_list(tsk);
 948         if (unlikely(current->pi_state_cache))
 949                 kfree(current->pi_state_cache);
 950         /*
 951          * Make sure we are holding no locks:
 952          */
 953         debug_check_no_locks_held(tsk);
 954
 955         if (tsk->io_context)
 956                 exit_io_context();
 957
 958         if (tsk->splice_pipe)
 959                 __free_pipe_info(tsk->splice_pipe);
 960
 961         preempt_disable();
 962         /* causes final put_task_struct in finish_task_switch(). */
 963         tsk->state = TASK_DEAD;
 964
 965         schedule();
 966         BUG();
 967         /* Avoid "noreturn function does return".  */
 968         for (;;)
 969                 cpu_relax();    /* For when BUG is null */
 970 }
 971
 972 EXPORT_SYMBOL_GPL(do_exit);
 973
 974 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 975 {
 976         if (comp)
 977                 complete(comp);
 978
 979         do_exit(code);
 980 }
 981
 982 EXPORT_SYMBOL(complete_and_exit);
 983
 984 asmlinkage long sys_exit(int error_code)
 985 {
 986         do_exit((error_code&0xff)<<8);
 987 }
 988
 989 /*
 990  * Take down every thread in the group.  This is called by fatal signals
 991  * as well as by sys_exit_group (below).
 992  */
 993 NORET_TYPE void
 994 do_group_exit(int exit_code)
 995 {
 996         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 997
 998         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 999                 exit_code = current->signal->group_exit_code;
1000         else if (!thread_group_empty(current)) {
1001                 struct signal_struct *const sig = current->signal;
1002                 struct sighand_struct *const sighand = current->sighand;
1003                 spin_lock_irq(&sighand->siglock);
1004                 if (sig->flags & SIGNAL_GROUP_EXIT)
1005                         /* Another thread got here before we took the lock.  */
1006                         exit_code = sig->group_exit_code;
1007                 else {
1008                         sig->group_exit_code = exit_code;
1009                         zap_other_threads(current);
1010                 }
1011                 spin_unlock_irq(&sighand->siglock);
1012         }
1013
1014         do_exit(exit_code);
1015         /* NOTREACHED */
1016 }
1017
1018 /*
1019  * this kills every thread in the thread group. Note that any externally
1020  * wait4()-ing process will get the correct exit code - even if this
1021  * thread is not the thread group leader.
1022  */
1023 asmlinkage void sys_exit_group(int error_code)
1024 {
1025         do_group_exit((error_code & 0xff) << 8);
1026 }
1027
1028 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1029 {
1030         if (pid > 0) {
1031                 if (p->pid != pid)
1032                         return 0;
1033         } else if (!pid) {
1034                 if (process_group(p) != process_group(current))
1035                         return 0;
1036         } else if (pid != -1) {
1037                 if (process_group(p) != -pid)
1038                         return 0;
1039         }
1040
1041         /*
1042          * Do not consider detached threads that are
1043          * not ptraced:
1044          */
1045         if (p->exit_signal == -1 && !p->ptrace)
1046                 return 0;
1047
1048         /* Wait for all children (clone and not) if __WALL is set;
1049          * otherwise, wait for clone children *only* if __WCLONE is
1050          * set; otherwise, wait for non-clone children *only*.  (Note:
1051          * A "clone" child here is one that reports to its parent
1052          * using a signal other than SIGCHLD.) */
1053         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1054             && !(options & __WALL))
1055                 return 0;
1056         /*
1057          * Do not consider thread group leaders that are
1058          * in a non-empty thread group:
1059          */
1060         if (delay_group_leader(p))
1061                 return 2;
1062
1063         if (security_task_wait(p))
1064                 return 0;
1065
1066         return 1;
1067 }
1068
1069 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1070                                int why, int status,
1071                                struct siginfo __user *infop,
1072                                struct rusage __user *rusagep)
1073 {
1074         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1075
1076         put_task_struct(p);
1077         if (!retval)
1078                 retval = put_user(SIGCHLD, &infop->si_signo);
1079         if (!retval)
1080                 retval = put_user(0, &infop->si_errno);
1081         if (!retval)
1082                 retval = put_user((short)why, &infop->si_code);
1083         if (!retval)
1084                 retval = put_user(pid, &infop->si_pid);
1085         if (!retval)
1086                 retval = put_user(uid, &infop->si_uid);
1087         if (!retval)
1088                 retval = put_user(status, &infop->si_status);
1089         if (!retval)
1090                 retval = pid;
1091         return retval;
1092 }
1093
1094 /*
1095  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1096  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1097  * the lock and this task is uninteresting.  If we return nonzero, we have
1098  * released the lock and the system call should return.
1099  */
1100 static int wait_task_zombie(struct task_struct *p, int noreap,
1101                             struct siginfo __user *infop,
1102                             int __user *stat_addr, struct rusage __user *ru)
1103 {
1104         unsigned long state;
1105         int retval;
1106         int status;
1107
1108         if (unlikely(noreap)) {
1109                 pid_t pid = p->pid;
1110                 uid_t uid = p->uid;
1111                 int exit_code = p->exit_code;
1112                 int why, status;
1113
1114                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1115                         return 0;
1116                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1117                         return 0;
1118                 get_task_struct(p);
1119                 read_unlock(&tasklist_lock);
1120                 if ((exit_code & 0x7f) == 0) {
1121                         why = CLD_EXITED;
1122                         status = exit_code >> 8;
1123                 } else {
1124                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1125                         status = exit_code & 0x7f;
1126                 }
1127                 return wait_noreap_copyout(p, pid, uid, why,
1128                                            status, infop, ru);
1129         }
1130
1131         /*
1132          * Try to move the task's state to DEAD
1133          * only one thread is allowed to do this:
1134          */
1135         state = xchg(&p->exit_state, EXIT_DEAD);
1136         if (state != EXIT_ZOMBIE) {
1137                 BUG_ON(state != EXIT_DEAD);
1138                 return 0;
1139         }
1140         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1141                 /*
1142                  * This can only happen in a race with a ptraced thread
1143                  * dying on another processor.
1144                  */
1145                 return 0;
1146         }
1147
1148         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1149                 struct signal_struct *psig;
1150                 struct signal_struct *sig;
1151
1152                 /*
1153                  * The resource counters for the group leader are in its
1154                  * own task_struct.  Those for dead threads in the group
1155                  * are in its signal_struct, as are those for the child
1156                  * processes it has previously reaped.  All these
1157                  * accumulate in the parent's signal_struct c* fields.
1158                  *
1159                  * We don't bother to take a lock here to protect these
1160                  * p->signal fields, because they are only touched by
1161                  * __exit_signal, which runs with tasklist_lock
1162                  * write-locked anyway, and so is excluded here.  We do
1163                  * need to protect the access to p->parent->signal fields,
1164                  * as other threads in the parent group can be right
1165                  * here reaping other children at the same time.
1166                  */
1167                 spin_lock_irq(&p->parent->sighand->siglock);
1168                 psig = p->parent->signal;
1169                 sig = p->signal;
1170                 psig->cutime =
1171                         cputime_add(psig->cutime,
1172                         cputime_add(p->utime,
1173                         cputime_add(sig->utime,
1174                                     sig->cutime)));
1175                 psig->cstime =
1176                         cputime_add(psig->cstime,
1177                         cputime_add(p->stime,
1178                         cputime_add(sig->stime,
1179                                     sig->cstime)));
1180                 psig->cmin_flt +=
1181                         p->min_flt + sig->min_flt + sig->cmin_flt;
1182                 psig->cmaj_flt +=
1183                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1184                 psig->cnvcsw +=
1185                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1186                 psig->cnivcsw +=
1187                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1188                 spin_unlock_irq(&p->parent->sighand->siglock);
1189         }
1190
1191         /*
1192          * Now we are sure this task is interesting, and no other
1193          * thread can reap it because we set its state to EXIT_DEAD.
1194          */
1195         read_unlock(&tasklist_lock);
1196
1197         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1198         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1199                 ? p->signal->group_exit_code : p->exit_code;
1200         if (!retval && stat_addr)
1201                 retval = put_user(status, stat_addr);
1202         if (!retval && infop)
1203                 retval = put_user(SIGCHLD, &infop->si_signo);
1204         if (!retval && infop)
1205                 retval = put_user(0, &infop->si_errno);
1206         if (!retval && infop) {
1207                 int why;
1208
1209                 if ((status & 0x7f) == 0) {
1210                         why = CLD_EXITED;
1211                         status >>= 8;
1212                 } else {
1213                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1214                         status &= 0x7f;
1215                 }
1216                 retval = put_user((short)why, &infop->si_code);
1217                 if (!retval)
1218                         retval = put_user(status, &infop->si_status);
1219         }
1220         if (!retval && infop)
1221                 retval = put_user(p->pid, &infop->si_pid);
1222         if (!retval && infop)
1223                 retval = put_user(p->uid, &infop->si_uid);
1224         if (retval) {
1225                 // TODO: is this safe?
1226                 p->exit_state = EXIT_ZOMBIE;
1227                 return retval;
1228         }
1229         retval = p->pid;
1230         if (p->real_parent != p->parent) {
1231                 write_lock_irq(&tasklist_lock);
1232                 /* Double-check with lock held.  */
1233                 if (p->real_parent != p->parent) {
1234                         __ptrace_unlink(p);
1235                         // TODO: is this safe?
1236                         p->exit_state = EXIT_ZOMBIE;
1237                         /*
1238                          * If this is not a detached task, notify the parent.
1239                          * If it's still not detached after that, don't release
1240                          * it now.
1241                          */
1242                         if (p->exit_signal != -1) {
1243                                 do_notify_parent(p, p->exit_signal);
1244                                 if (p->exit_signal != -1)
1245                                         p = NULL;
1246                         }
1247                 }
1248                 write_unlock_irq(&tasklist_lock);
1249         }
1250         if (p != NULL)
1251                 release_task(p);
1252         BUG_ON(!retval);
1253         return retval;
1254 }
1255
1256 /*
1257  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1258  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1259  * the lock and this task is uninteresting.  If we return nonzero, we have
1260  * released the lock and the system call should return.
1261  */
1262 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1263                              int noreap, struct siginfo __user *infop,
1264                              int __user *stat_addr, struct rusage __user *ru)
1265 {
1266         int retval, exit_code;
1267
1268         if (!p->exit_code)
1269                 return 0;
1270         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1271             p->signal && p->signal->group_stop_count > 0)
1272                 /*
1273                  * A group stop is in progress and this is the group leader.
1274                  * We won't report until all threads have stopped.
1275                  */
1276                 return 0;
1277
1278         /*
1279          * Now we are pretty sure this task is interesting.
1280          * Make sure it doesn't get reaped out from under us while we
1281          * give up the lock and then examine it below.  We don't want to
1282          * keep holding onto the tasklist_lock while we call getrusage and
1283          * possibly take page faults for user memory.
1284          */
1285         get_task_struct(p);
1286         read_unlock(&tasklist_lock);
1287
1288         if (unlikely(noreap)) {
1289                 pid_t pid = p->pid;
1290                 uid_t uid = p->uid;
1291                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1292
1293                 exit_code = p->exit_code;
1294                 if (unlikely(!exit_code) ||
1295                     unlikely(p->state & TASK_TRACED))
1296                         goto bail_ref;
1297                 return wait_noreap_copyout(p, pid, uid,
1298                                            why, (exit_code << 8) | 0x7f,
1299                                            infop, ru);
1300         }
1301
1302         write_lock_irq(&tasklist_lock);
1303
1304         /*
1305          * This uses xchg to be atomic with the thread resuming and setting
1306          * it.  It must also be done with the write lock held to prevent a
1307          * race with the EXIT_ZOMBIE case.
1308          */
1309         exit_code = xchg(&p->exit_code, 0);
1310         if (unlikely(p->exit_state)) {
1311                 /*
1312                  * The task resumed and then died.  Let the next iteration
1313                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1314                  * already be zero here if it resumed and did _exit(0).
1315                  * The task itself is dead and won't touch exit_code again;
1316                  * other processors in this function are locked out.
1317                  */
1318                 p->exit_code = exit_code;
1319                 exit_code = 0;
1320         }
1321         if (unlikely(exit_code == 0)) {
1322                 /*
1323                  * Another thread in this function got to it first, or it
1324                  * resumed, or it resumed and then died.
1325                  */
1326                 write_unlock_irq(&tasklist_lock);
1327 bail_ref:
1328                 put_task_struct(p);
1329                 /*
1330                  * We are returning to the wait loop without having successfully
1331                  * removed the process and having released the lock. We cannot
1332                  * continue, since the "p" task pointer is potentially stale.
1333                  *
1334                  * Return -EAGAIN, and do_wait() will restart the loop from the
1335                  * beginning. Do _not_ re-acquire the lock.
1336                  */
1337                 return -EAGAIN;
1338         }
1339
1340         /* move to end of parent's list to avoid starvation */
1341         remove_parent(p);
1342         add_parent(p);
1343
1344         write_unlock_irq(&tasklist_lock);
1345
1346         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1347         if (!retval && stat_addr)
1348                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1349         if (!retval && infop)
1350                 retval = put_user(SIGCHLD, &infop->si_signo);
1351         if (!retval && infop)
1352                 retval = put_user(0, &infop->si_errno);
1353         if (!retval && infop)
1354                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1355                                           ? CLD_TRAPPED : CLD_STOPPED),
1356                                   &infop->si_code);
1357         if (!retval && infop)
1358                 retval = put_user(exit_code, &infop->si_status);
1359         if (!retval && infop)
1360                 retval = put_user(p->pid, &infop->si_pid);
1361         if (!retval && infop)
1362                 retval = put_user(p->uid, &infop->si_uid);
1363         if (!retval)
1364                 retval = p->pid;
1365         put_task_struct(p);
1366
1367         BUG_ON(!retval);
1368         return retval;
1369 }
1370
1371 /*
1372  * Handle do_wait work for one task in a live, non-stopped state.
1373  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1374  * the lock and this task is uninteresting.  If we return nonzero, we have
1375  * released the lock and the system call should return.
1376  */
1377 static int wait_task_continued(struct task_struct *p, int noreap,
1378                                struct siginfo __user *infop,
1379                                int __user *stat_addr, struct rusage __user *ru)
1380 {
1381         int retval;
1382         pid_t pid;
1383         uid_t uid;
1384
1385         if (unlikely(!p->signal))
1386                 return 0;
1387
1388         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1389                 return 0;
1390
1391         spin_lock_irq(&p->sighand->siglock);
1392         /* Re-check with the lock held.  */
1393         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1394                 spin_unlock_irq(&p->sighand->siglock);
1395                 return 0;
1396         }
1397         if (!noreap)
1398                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1399         spin_unlock_irq(&p->sighand->siglock);
1400
1401         pid = p->pid;
1402         uid = p->uid;
1403         get_task_struct(p);
1404         read_unlock(&tasklist_lock);
1405
1406         if (!infop) {
1407                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1408                 put_task_struct(p);
1409                 if (!retval && stat_addr)
1410                         retval = put_user(0xffff, stat_addr);
1411                 if (!retval)
1412                         retval = p->pid;
1413         } else {
1414                 retval = wait_noreap_copyout(p, pid, uid,
1415                                              CLD_CONTINUED, SIGCONT,
1416                                              infop, ru);
1417                 BUG_ON(retval == 0);
1418         }
1419
1420         return retval;
1421 }
1422
1423
1424 static inline int my_ptrace_child(struct task_struct *p)
1425 {
1426         if (!(p->ptrace & PT_PTRACED))
1427                 return 0;
1428         if (!(p->ptrace & PT_ATTACHED))
1429                 return 1;
1430         /*
1431          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1432          * we are the attacher.  If we are the real parent, this is a race
1433          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1434          * which we have to switch the parent links, but has already set
1435          * the flags in p->ptrace.
1436          */
1437         return (p->parent != p->real_parent);
1438 }
1439
1440 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1441                     int __user *stat_addr, struct rusage __user *ru)
1442 {
1443         DECLARE_WAITQUEUE(wait, current);
1444         struct task_struct *tsk;
1445         int flag, retval;
1446
1447         add_wait_queue(&current->signal->wait_chldexit,&wait);
1448 repeat:
1449         /*
1450          * We will set this flag if we see any child that might later
1451          * match our criteria, even if we are not able to reap it yet.
1452          */
1453         flag = 0;
1454         current->state = TASK_INTERRUPTIBLE;
1455         read_lock(&tasklist_lock);
1456         tsk = current;
1457         do {
1458                 struct task_struct *p;
1459                 struct list_head *_p;
1460                 int ret;
1461
1462                 list_for_each(_p,&tsk->children) {
1463                         p = list_entry(_p, struct task_struct, sibling);
1464
1465                         ret = eligible_child(pid, options, p);
1466                         if (!ret)
1467                                 continue;
1468
1469                         switch (p->state) {
1470                         case TASK_TRACED:
1471                                 /*
1472                                  * When we hit the race with PTRACE_ATTACH,
1473                                  * we will not report this child.  But the
1474                                  * race means it has not yet been moved to
1475                                  * our ptrace_children list, so we need to
1476                                  * set the flag here to avoid a spurious ECHILD
1477                                  * when the race happens with the only child.
1478                                  */
1479                                 flag = 1;
1480                                 if (!my_ptrace_child(p))
1481                                         continue;
1482                                 /*FALLTHROUGH*/
1483                         case TASK_STOPPED:
1484                                 /*
1485                                  * It's stopped now, so it might later
1486                                  * continue, exit, or stop again.
1487                                  */
1488                                 flag = 1;
1489                                 if (!(options & WUNTRACED) &&
1490                                     !my_ptrace_child(p))
1491                                         continue;
1492                                 retval = wait_task_stopped(p, ret == 2,
1493                                                            (options & WNOWAIT),
1494                                                            infop,
1495                                                            stat_addr, ru);
1496                                 if (retval == -EAGAIN)
1497                                         goto repeat;
1498                                 if (retval != 0) /* He released the lock.  */
1499                                         goto end;
1500                                 break;
1501                         default:
1502                         // case EXIT_DEAD:
1503                                 if (p->exit_state == EXIT_DEAD)
1504                                         continue;
1505                         // case EXIT_ZOMBIE:
1506                                 if (p->exit_state == EXIT_ZOMBIE) {
1507                                         /*
1508                                          * Eligible but we cannot release
1509                                          * it yet:
1510                                          */
1511                                         if (ret == 2)
1512                                                 goto check_continued;
1513                                         if (!likely(options & WEXITED))
1514                                                 continue;
1515                                         retval = wait_task_zombie(
1516                                                 p, (options & WNOWAIT),
1517                                                 infop, stat_addr, ru);
1518                                         /* He released the lock.  */
1519                                         if (retval != 0)
1520                                                 goto end;
1521                                         break;
1522                                 }
1523 check_continued:
1524                                 /*
1525                                  * It's running now, so it might later
1526                                  * exit, stop, or stop and then continue.
1527                                  */
1528                                 flag = 1;
1529                                 if (!unlikely(options & WCONTINUED))
1530                                         continue;
1531                                 retval = wait_task_continued(
1532                                         p, (options & WNOWAIT),
1533                                         infop, stat_addr, ru);
1534                                 if (retval != 0) /* He released the lock.  */
1535                                         goto end;
1536                                 break;
1537                         }
1538                 }
1539                 if (!flag) {
1540                         list_for_each(_p, &tsk->ptrace_children) {
1541                                 p = list_entry(_p, struct task_struct,
1542                                                 ptrace_list);
1543                                 if (!eligible_child(pid, options, p))
1544                                         continue;
1545                                 flag = 1;
1546                                 break;
1547                         }
1548                 }
1549                 if (options & __WNOTHREAD)
1550                         break;
1551                 tsk = next_thread(tsk);
1552                 BUG_ON(tsk->signal != current->signal);
1553         } while (tsk != current);
1554
1555         read_unlock(&tasklist_lock);
1556         if (flag) {
1557                 retval = 0;
1558                 if (options & WNOHANG)
1559                         goto end;
1560                 retval = -ERESTARTSYS;
1561                 if (signal_pending(current))
1562                         goto end;
1563                 schedule();
1564                 goto repeat;
1565         }
1566         retval = -ECHILD;
1567 end:
1568         current->state = TASK_RUNNING;
1569         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1570         if (infop) {
1571                 if (retval > 0)
1572                 retval = 0;
1573                 else {
1574                         /*
1575                          * For a WNOHANG return, clear out all the fields
1576                          * we would set so the user can easily tell the
1577                          * difference.
1578                          */
1579                         if (!retval)
1580                                 retval = put_user(0, &infop->si_signo);
1581                         if (!retval)
1582                                 retval = put_user(0, &infop->si_errno);
1583                         if (!retval)
1584                                 retval = put_user(0, &infop->si_code);
1585                         if (!retval)
1586                                 retval = put_user(0, &infop->si_pid);
1587                         if (!retval)
1588                                 retval = put_user(0, &infop->si_uid);
1589                         if (!retval)
1590                                 retval = put_user(0, &infop->si_status);
1591                 }
1592         }
1593         return retval;
1594 }
1595
1596 asmlinkage long sys_waitid(int which, pid_t pid,
1597                            struct siginfo __user *infop, int options,
1598                            struct rusage __user *ru)
1599 {
1600         long ret;
1601
1602         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1603                 return -EINVAL;
1604         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1605                 return -EINVAL;
1606
1607         switch (which) {
1608         case P_ALL:
1609                 pid = -1;
1610                 break;
1611         case P_PID:
1612                 if (pid <= 0)
1613                         return -EINVAL;
1614                 break;
1615         case P_PGID:
1616                 if (pid <= 0)
1617                         return -EINVAL;
1618                 pid = -pid;
1619                 break;
1620         default:
1621                 return -EINVAL;
1622         }
1623
1624         ret = do_wait(pid, options, infop, NULL, ru);
1625
1626         /* avoid REGPARM breakage on x86: */
1627         prevent_tail_call(ret);
1628         return ret;
1629 }
1630
1631 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1632                           int options, struct rusage __user *ru)
1633 {
1634         long ret;
1635
1636         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1637                         __WNOTHREAD|__WCLONE|__WALL))
1638                 return -EINVAL;
1639         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1640
1641         /* avoid REGPARM breakage on x86: */
1642         prevent_tail_call(ret);
1643         return ret;
1644 }
1645
1646 #ifdef __ARCH_WANT_SYS_WAITPID
1647
1648 /*
1649  * sys_waitpid() remains for compatibility. waitpid() should be
1650  * implemented by calling sys_wait4() from libc.a.
1651  */
1652 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1653 {
1654         return sys_wait4(pid, stat_addr, options, NULL);
1655 }
1656
1657 #endif