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