linux/kernel/exit.c
<<
>>
Prefs
   1/*
   2 *  linux/kernel/exit.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 */
   6
   7#include <linux/mm.h>
   8#include <linux/slab.h>
   9#include <linux/interrupt.h>
  10#include <linux/module.h>
  11#include <linux/capability.h>
  12#include <linux/completion.h>
  13#include <linux/personality.h>
  14#include <linux/tty.h>
  15#include <linux/iocontext.h>
  16#include <linux/key.h>
  17#include <linux/security.h>
  18#include <linux/cpu.h>
  19#include <linux/acct.h>
  20#include <linux/tsacct_kern.h>
  21#include <linux/file.h>
  22#include <linux/fdtable.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/kthread.h>
  31#include <linux/mempolicy.h>
  32#include <linux/taskstats_kern.h>
  33#include <linux/delayacct.h>
  34#include <linux/freezer.h>
  35#include <linux/cgroup.h>
  36#include <linux/syscalls.h>
  37#include <linux/signal.h>
  38#include <linux/posix-timers.h>
  39#include <linux/cn_proc.h>
  40#include <linux/mutex.h>
  41#include <linux/futex.h>
  42#include <linux/pipe_fs_i.h>
  43#include <linux/audit.h> /* for audit_free() */
  44#include <linux/resource.h>
  45#include <linux/blkdev.h>
  46#include <linux/task_io_accounting_ops.h>
  47#include <linux/tracehook.h>
  48#include <linux/fs_struct.h>
  49#include <linux/init_task.h>
  50#include <linux/perf_event.h>
  51#include <trace/events/sched.h>
  52#include <linux/hw_breakpoint.h>
  53#include <linux/oom.h>
  54
  55#include <asm/uaccess.h>
  56#include <asm/unistd.h>
  57#include <asm/pgtable.h>
  58#include <asm/mmu_context.h>
  59
  60static void exit_mm(struct task_struct * tsk);
  61
  62static void __unhash_process(struct task_struct *p, bool group_dead)
  63{
  64        nr_threads--;
  65        detach_pid(p, PIDTYPE_PID);
  66        if (group_dead) {
  67                detach_pid(p, PIDTYPE_PGID);
  68                detach_pid(p, PIDTYPE_SID);
  69
  70                list_del_rcu(&p->tasks);
  71                list_del_init(&p->sibling);
  72                __this_cpu_dec(process_counts);
  73        }
  74        list_del_rcu(&p->thread_group);
  75}
  76
  77/*
  78 * This function expects the tasklist_lock write-locked.
  79 */
  80static void __exit_signal(struct task_struct *tsk)
  81{
  82        struct signal_struct *sig = tsk->signal;
  83        bool group_dead = thread_group_leader(tsk);
  84        struct sighand_struct *sighand;
  85        struct tty_struct *uninitialized_var(tty);
  86
  87        sighand = rcu_dereference_check(tsk->sighand,
  88                                        lockdep_tasklist_lock_is_held());
  89        spin_lock(&sighand->siglock);
  90
  91        posix_cpu_timers_exit(tsk);
  92        if (group_dead) {
  93                posix_cpu_timers_exit_group(tsk);
  94                tty = sig->tty;
  95                sig->tty = NULL;
  96        } else {
  97                /*
  98                 * This can only happen if the caller is de_thread().
  99                 * FIXME: this is the temporary hack, we should teach
 100                 * posix-cpu-timers to handle this case correctly.
 101                 */
 102                if (unlikely(has_group_leader_pid(tsk)))
 103                        posix_cpu_timers_exit_group(tsk);
 104
 105                /*
 106                 * If there is any task waiting for the group exit
 107                 * then notify it:
 108                 */
 109                if (sig->notify_count > 0 && !--sig->notify_count)
 110                        wake_up_process(sig->group_exit_task);
 111
 112                if (tsk == sig->curr_target)
 113                        sig->curr_target = next_thread(tsk);
 114                /*
 115                 * Accumulate here the counters for all threads but the
 116                 * group leader as they die, so they can be added into
 117                 * the process-wide totals when those are taken.
 118                 * The group leader stays around as a zombie as long
 119                 * as there are other threads.  When it gets reaped,
 120                 * the exit.c code will add its counts into these totals.
 121                 * We won't ever get here for the group leader, since it
 122                 * will have been the last reference on the signal_struct.
 123                 */
 124                sig->utime = cputime_add(sig->utime, tsk->utime);
 125                sig->stime = cputime_add(sig->stime, tsk->stime);
 126                sig->gtime = cputime_add(sig->gtime, tsk->gtime);
 127                sig->min_flt += tsk->min_flt;
 128                sig->maj_flt += tsk->maj_flt;
 129                sig->nvcsw += tsk->nvcsw;
 130                sig->nivcsw += tsk->nivcsw;
 131                sig->inblock += task_io_get_inblock(tsk);
 132                sig->oublock += task_io_get_oublock(tsk);
 133                task_io_accounting_add(&sig->ioac, &tsk->ioac);
 134                sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 135        }
 136
 137        sig->nr_threads--;
 138        __unhash_process(tsk, group_dead);
 139
 140        /*
 141         * Do this under ->siglock, we can race with another thread
 142         * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
 143         */
 144        flush_sigqueue(&tsk->pending);
 145        tsk->sighand = NULL;
 146        spin_unlock(&sighand->siglock);
 147
 148        __cleanup_sighand(sighand);
 149        clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 150        if (group_dead) {
 151                flush_sigqueue(&sig->shared_pending);
 152                tty_kref_put(tty);
 153        }
 154}
 155
 156static void delayed_put_task_struct(struct rcu_head *rhp)
 157{
 158        struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
 159
 160        perf_event_delayed_put(tsk);
 161        trace_sched_process_free(tsk);
 162        put_task_struct(tsk);
 163}
 164
 165
 166void release_task(struct task_struct * p)
 167{
 168        struct task_struct *leader;
 169        int zap_leader;
 170repeat:
 171        /* don't need to get the RCU readlock here - the process is dead and
 172         * can't be modifying its own credentials. But shut RCU-lockdep up */
 173        rcu_read_lock();
 174        atomic_dec(&__task_cred(p)->user->processes);
 175        rcu_read_unlock();
 176
 177        proc_flush_task(p);
 178
 179        write_lock_irq(&tasklist_lock);
 180        ptrace_release_task(p);
 181        __exit_signal(p);
 182
 183        /*
 184         * If we are the last non-leader member of the thread
 185         * group, and the leader is zombie, then notify the
 186         * group leader's parent process. (if it wants notification.)
 187         */
 188        zap_leader = 0;
 189        leader = p->group_leader;
 190        if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 191                /*
 192                 * If we were the last child thread and the leader has
 193                 * exited already, and the leader's parent ignores SIGCHLD,
 194                 * then we are the one who should release the leader.
 195                 */
 196                zap_leader = do_notify_parent(leader, leader->exit_signal);
 197                if (zap_leader)
 198                        leader->exit_state = EXIT_DEAD;
 199        }
 200
 201        write_unlock_irq(&tasklist_lock);
 202        release_thread(p);
 203        call_rcu(&p->rcu, delayed_put_task_struct);
 204
 205        p = leader;
 206        if (unlikely(zap_leader))
 207                goto repeat;
 208}
 209
 210/*
 211 * This checks not only the pgrp, but falls back on the pid if no
 212 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 213 * without this...
 214 *
 215 * The caller must hold rcu lock or the tasklist lock.
 216 */
 217struct pid *session_of_pgrp(struct pid *pgrp)
 218{
 219        struct task_struct *p;
 220        struct pid *sid = NULL;
 221
 222        p = pid_task(pgrp, PIDTYPE_PGID);
 223        if (p == NULL)
 224                p = pid_task(pgrp, PIDTYPE_PID);
 225        if (p != NULL)
 226                sid = task_session(p);
 227
 228        return sid;
 229}
 230
 231/*
 232 * Determine if a process group is "orphaned", according to the POSIX
 233 * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 234 * by terminal-generated stop signals.  Newly orphaned process groups are
 235 * to receive a SIGHUP and a SIGCONT.
 236 *
 237 * "I ask you, have you ever known what it is to be an orphan?"
 238 */
 239static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 240{
 241        struct task_struct *p;
 242
 243        do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 244                if ((p == ignored_task) ||
 245                    (p->exit_state && thread_group_empty(p)) ||
 246                    is_global_init(p->real_parent))
 247                        continue;
 248
 249                if (task_pgrp(p->real_parent) != pgrp &&
 250                    task_session(p->real_parent) == task_session(p))
 251                        return 0;
 252        } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 253
 254        return 1;
 255}
 256
 257int is_current_pgrp_orphaned(void)
 258{
 259        int retval;
 260
 261        read_lock(&tasklist_lock);
 262        retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 263        read_unlock(&tasklist_lock);
 264
 265        return retval;
 266}
 267
 268static bool has_stopped_jobs(struct pid *pgrp)
 269{
 270        struct task_struct *p;
 271
 272        do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 273                if (p->signal->flags & SIGNAL_STOP_STOPPED)
 274                        return true;
 275        } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 276
 277        return false;
 278}
 279
 280/*
 281 * Check to see if any process groups have become orphaned as
 282 * a result of our exiting, and if they have any stopped jobs,
 283 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
 284 */
 285static void
 286kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
 287{
 288        struct pid *pgrp = task_pgrp(tsk);
 289        struct task_struct *ignored_task = tsk;
 290
 291        if (!parent)
 292                 /* exit: our father is in a different pgrp than
 293                  * we are and we were the only connection outside.
 294                  */
 295                parent = tsk->real_parent;
 296        else
 297                /* reparent: our child is in a different pgrp than
 298                 * we are, and it was the only connection outside.
 299                 */
 300                ignored_task = NULL;
 301
 302        if (task_pgrp(parent) != pgrp &&
 303            task_session(parent) == task_session(tsk) &&
 304            will_become_orphaned_pgrp(pgrp, ignored_task) &&
 305            has_stopped_jobs(pgrp)) {
 306                __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 307                __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 308        }
 309}
 310
 311/**
 312 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
 313 *
 314 * If a kernel thread is launched as a result of a system call, or if
 315 * it ever exits, it should generally reparent itself to kthreadd so it
 316 * isn't in the way of other processes and is correctly cleaned up on exit.
 317 *
 318 * The various task state such as scheduling policy and priority may have
 319 * been inherited from a user process, so we reset them to sane values here.
 320 *
 321 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
 322 */
 323static void reparent_to_kthreadd(void)
 324{
 325        write_lock_irq(&tasklist_lock);
 326
 327        ptrace_unlink(current);
 328        /* Reparent to init */
 329        current->real_parent = current->parent = kthreadd_task;
 330        list_move_tail(&current->sibling, &current->real_parent->children);
 331
 332        /* Set the exit signal to SIGCHLD so we signal init on exit */
 333        current->exit_signal = SIGCHLD;
 334
 335        if (task_nice(current) < 0)
 336                set_user_nice(current, 0);
 337        /* cpus_allowed? */
 338        /* rt_priority? */
 339        /* signals? */
 340        memcpy(current->signal->rlim, init_task.signal->rlim,
 341               sizeof(current->signal->rlim));
 342
 343        atomic_inc(&init_cred.usage);
 344        commit_creds(&init_cred);
 345        write_unlock_irq(&tasklist_lock);
 346}
 347
 348void __set_special_pids(struct pid *pid)
 349{
 350        struct task_struct *curr = current->group_leader;
 351
 352        if (task_session(curr) != pid)
 353                change_pid(curr, PIDTYPE_SID, pid);
 354
 355        if (task_pgrp(curr) != pid)
 356                change_pid(curr, PIDTYPE_PGID, pid);
 357}
 358
 359static void set_special_pids(struct pid *pid)
 360{
 361        write_lock_irq(&tasklist_lock);
 362        __set_special_pids(pid);
 363        write_unlock_irq(&tasklist_lock);
 364}
 365
 366/*
 367 * Let kernel threads use this to say that they allow a certain signal.
 368 * Must not be used if kthread was cloned with CLONE_SIGHAND.
 369 */
 370int allow_signal(int sig)
 371{
 372        if (!valid_signal(sig) || sig < 1)
 373                return -EINVAL;
 374
 375        spin_lock_irq(&current->sighand->siglock);
 376        /* This is only needed for daemonize()'ed kthreads */
 377        sigdelset(&current->blocked, sig);
 378        /*
 379         * Kernel threads handle their own signals. Let the signal code
 380         * know it'll be handled, so that they don't get converted to
 381         * SIGKILL or just silently dropped.
 382         */
 383        current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 384        recalc_sigpending();
 385        spin_unlock_irq(&current->sighand->siglock);
 386        return 0;
 387}
 388
 389EXPORT_SYMBOL(allow_signal);
 390
 391int disallow_signal(int sig)
 392{
 393        if (!valid_signal(sig) || sig < 1)
 394                return -EINVAL;
 395
 396        spin_lock_irq(&current->sighand->siglock);
 397        current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
 398        recalc_sigpending();
 399        spin_unlock_irq(&current->sighand->siglock);
 400        return 0;
 401}
 402
 403EXPORT_SYMBOL(disallow_signal);
 404
 405/*
 406 *      Put all the gunge required to become a kernel thread without
 407 *      attached user resources in one place where it belongs.
 408 */
 409
 410void daemonize(const char *name, ...)
 411{
 412        va_list args;
 413        sigset_t blocked;
 414
 415        va_start(args, name);
 416        vsnprintf(current->comm, sizeof(current->comm), name, args);
 417        va_end(args);
 418
 419        /*
 420         * If we were started as result of loading a module, close all of the
 421         * user space pages.  We don't need them, and if we didn't close them
 422         * they would be locked into memory.
 423         */
 424        exit_mm(current);
 425        /*
 426         * We don't want to have TIF_FREEZE set if the system-wide hibernation
 427         * or suspend transition begins right now.
 428         */
 429        current->flags |= (PF_NOFREEZE | PF_KTHREAD);
 430
 431        if (current->nsproxy != &init_nsproxy) {
 432                get_nsproxy(&init_nsproxy);
 433                switch_task_namespaces(current, &init_nsproxy);
 434        }
 435        set_special_pids(&init_struct_pid);
 436        proc_clear_tty(current);
 437
 438        /* Block and flush all signals */
 439        sigfillset(&blocked);
 440        sigprocmask(SIG_BLOCK, &blocked, NULL);
 441        flush_signals(current);
 442
 443        /* Become as one with the init task */
 444
 445        daemonize_fs_struct();
 446        exit_files(current);
 447        current->files = init_task.files;
 448        atomic_inc(&current->files->count);
 449
 450        reparent_to_kthreadd();
 451}
 452
 453EXPORT_SYMBOL(daemonize);
 454
 455static void close_files(struct files_struct * files)
 456{
 457        int i, j;
 458        struct fdtable *fdt;
 459
 460        j = 0;
 461
 462        /*
 463         * It is safe to dereference the fd table without RCU or
 464         * ->file_lock because this is the last reference to the
 465         * files structure.  But use RCU to shut RCU-lockdep up.
 466         */
 467        rcu_read_lock();
 468        fdt = files_fdtable(files);
 469        rcu_read_unlock();
 470        for (;;) {
 471                unsigned long set;
 472                i = j * __NFDBITS;
 473                if (i >= fdt->max_fds)
 474                        break;
 475                set = fdt->open_fds->fds_bits[j++];
 476                while (set) {
 477                        if (set & 1) {
 478                                struct file * file = xchg(&fdt->fd[i], NULL);
 479                                if (file) {
 480                                        filp_close(file, files);
 481                                        cond_resched();
 482                                }
 483                        }
 484                        i++;
 485                        set >>= 1;
 486                }
 487        }
 488}
 489
 490struct files_struct *get_files_struct(struct task_struct *task)
 491{
 492        struct files_struct *files;
 493
 494        task_lock(task);
 495        files = task->files;
 496        if (files)
 497                atomic_inc(&files->count);
 498        task_unlock(task);
 499
 500        return files;
 501}
 502
 503void put_files_struct(struct files_struct *files)
 504{
 505        struct fdtable *fdt;
 506
 507        if (atomic_dec_and_test(&files->count)) {
 508                close_files(files);
 509                /*
 510                 * Free the fd and fdset arrays if we expanded them.
 511                 * If the fdtable was embedded, pass files for freeing
 512                 * at the end of the RCU grace period. Otherwise,
 513                 * you can free files immediately.
 514                 */
 515                rcu_read_lock();
 516                fdt = files_fdtable(files);
 517                if (fdt != &files->fdtab)
 518                        kmem_cache_free(files_cachep, files);
 519                free_fdtable(fdt);
 520                rcu_read_unlock();
 521        }
 522}
 523
 524void reset_files_struct(struct files_struct *files)
 525{
 526        struct task_struct *tsk = current;
 527        struct files_struct *old;
 528
 529        old = tsk->files;
 530        task_lock(tsk);
 531        tsk->files = files;
 532        task_unlock(tsk);
 533        put_files_struct(old);
 534}
 535
 536void exit_files(struct task_struct *tsk)
 537{
 538        struct files_struct * files = tsk->files;
 539
 540        if (files) {
 541                task_lock(tsk);
 542                tsk->files = NULL;
 543                task_unlock(tsk);
 544                put_files_struct(files);
 545        }
 546}
 547
 548#ifdef CONFIG_MM_OWNER
 549/*
 550 * A task is exiting.   If it owned this mm, find a new owner for the mm.
 551 */
 552void mm_update_next_owner(struct mm_struct *mm)
 553{
 554        struct task_struct *c, *g, *p = current;
 555
 556retry:
 557        /*
 558         * If the exiting or execing task is not the owner, it's
 559         * someone else's problem.
 560         */
 561        if (mm->owner != p)
 562                return;
 563        /*
 564         * The current owner is exiting/execing and there are no other
 565         * candidates.  Do not leave the mm pointing to a possibly
 566         * freed task structure.
 567         */
 568        if (atomic_read(&mm->mm_users) <= 1) {
 569                mm->owner = NULL;
 570                return;
 571        }
 572
 573        read_lock(&tasklist_lock);
 574        /*
 575         * Search in the children
 576         */
 577        list_for_each_entry(c, &p->children, sibling) {
 578                if (c->mm == mm)
 579                        goto assign_new_owner;
 580        }
 581
 582        /*
 583         * Search in the siblings
 584         */
 585        list_for_each_entry(c, &p->real_parent->children, sibling) {
 586                if (c->mm == mm)
 587                        goto assign_new_owner;
 588        }
 589
 590        /*
 591         * Search through everything else. We should not get
 592         * here often
 593         */
 594        do_each_thread(g, c) {
 595                if (c->mm == mm)
 596                        goto assign_new_owner;
 597        } while_each_thread(g, c);
 598
 599        read_unlock(&tasklist_lock);
 600        /*
 601         * We found no owner yet mm_users > 1: this implies that we are
 602         * most likely racing with swapoff (try_to_unuse()) or /proc or
 603         * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
 604         */
 605        mm->owner = NULL;
 606        return;
 607
 608assign_new_owner:
 609        BUG_ON(c == p);
 610        get_task_struct(c);
 611        /*
 612         * The task_lock protects c->mm from changing.
 613         * We always want mm->owner->mm == mm
 614         */
 615        task_lock(c);
 616        /*
 617         * Delay read_unlock() till we have the task_lock()
 618         * to ensure that c does not slip away underneath us
 619         */
 620        read_unlock(&tasklist_lock);
 621        if (c->mm != mm) {
 622                task_unlock(c);
 623                put_task_struct(c);
 624                goto retry;
 625        }
 626        mm->owner = c;
 627        task_unlock(c);
 628        put_task_struct(c);
 629}
 630#endif /* CONFIG_MM_OWNER */
 631
 632/*
 633 * Turn us into a lazy TLB process if we
 634 * aren't already..
 635 */
 636static void exit_mm(struct task_struct * tsk)
 637{
 638        struct mm_struct *mm = tsk->mm;
 639        struct core_state *core_state;
 640
 641        mm_release(tsk, mm);
 642        if (!mm)
 643                return;
 644        /*
 645         * Serialize with any possible pending coredump.
 646         * We must hold mmap_sem around checking core_state
 647         * and clearing tsk->mm.  The core-inducing thread
 648         * will increment ->nr_threads for each thread in the
 649         * group with ->mm != NULL.
 650         */
 651        down_read(&mm->mmap_sem);
 652        core_state = mm->core_state;
 653        if (core_state) {
 654                struct core_thread self;
 655                up_read(&mm->mmap_sem);
 656
 657                self.task = tsk;
 658                self.next = xchg(&core_state->dumper.next, &self);
 659                /*
 660                 * Implies mb(), the result of xchg() must be visible
 661                 * to core_state->dumper.
 662                 */
 663                if (atomic_dec_and_test(&core_state->nr_threads))
 664                        complete(&core_state->startup);
 665
 666                for (;;) {
 667                        set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 668                        if (!self.task) /* see coredump_finish() */
 669                                break;
 670                        schedule();
 671                }
 672                __set_task_state(tsk, TASK_RUNNING);
 673                down_read(&mm->mmap_sem);
 674        }
 675        atomic_inc(&mm->mm_count);
 676        BUG_ON(mm != tsk->active_mm);
 677        /* more a memory barrier than a real lock */
 678        task_lock(tsk);
 679        tsk->mm = NULL;
 680        up_read(&mm->mmap_sem);
 681        enter_lazy_tlb(mm, current);
 682        /* We don't want this task to be frozen prematurely */
 683        clear_freeze_flag(tsk);
 684        if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
 685                atomic_dec(&mm->oom_disable_count);
 686        task_unlock(tsk);
 687        mm_update_next_owner(mm);
 688        mmput(mm);
 689}
 690
 691/*
 692 * When we die, we re-parent all our children.
 693 * Try to give them to another thread in our thread
 694 * group, and if no such member exists, give it to
 695 * the child reaper process (ie "init") in our pid
 696 * space.
 697 */
 698static struct task_struct *find_new_reaper(struct task_struct *father)
 699        __releases(&tasklist_lock)
 700        __acquires(&tasklist_lock)
 701{
 702        struct pid_namespace *pid_ns = task_active_pid_ns(father);
 703        struct task_struct *thread;
 704
 705        thread = father;
 706        while_each_thread(father, thread) {
 707                if (thread->flags & PF_EXITING)
 708                        continue;
 709                if (unlikely(pid_ns->child_reaper == father))
 710                        pid_ns->child_reaper = thread;
 711                return thread;
 712        }
 713
 714        if (unlikely(pid_ns->child_reaper == father)) {
 715                write_unlock_irq(&tasklist_lock);
 716                if (unlikely(pid_ns == &init_pid_ns))
 717                        panic("Attempted to kill init!");
 718
 719                zap_pid_ns_processes(pid_ns);
 720                write_lock_irq(&tasklist_lock);
 721                /*
 722                 * We can not clear ->child_reaper or leave it alone.
 723                 * There may by stealth EXIT_DEAD tasks on ->children,
 724                 * forget_original_parent() must move them somewhere.
 725                 */
 726                pid_ns->child_reaper = init_pid_ns.child_reaper;
 727        }
 728
 729        return pid_ns->child_reaper;
 730}
 731
 732/*
 733* Any that need to be release_task'd are put on the @dead list.
 734 */
 735static void reparent_leader(struct task_struct *father, struct task_struct *p,
 736                                struct list_head *dead)
 737{
 738        list_move_tail(&p->sibling, &p->real_parent->children);
 739
 740        if (p->exit_state == EXIT_DEAD)
 741                return;
 742        /*
 743         * If this is a threaded reparent there is no need to
 744         * notify anyone anything has happened.
 745         */
 746        if (same_thread_group(p->real_parent, father))
 747                return;
 748
 749        /* We don't want people slaying init.  */
 750        p->exit_signal = SIGCHLD;
 751
 752        /* If it has exited notify the new parent about this child's death. */
 753        if (!p->ptrace &&
 754            p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
 755                if (do_notify_parent(p, p->exit_signal)) {
 756                        p->exit_state = EXIT_DEAD;
 757                        list_move_tail(&p->sibling, dead);
 758                }
 759        }
 760
 761        kill_orphaned_pgrp(p, father);
 762}
 763
 764static void forget_original_parent(struct task_struct *father)
 765{
 766        struct task_struct *p, *n, *reaper;
 767        LIST_HEAD(dead_children);
 768
 769        write_lock_irq(&tasklist_lock);
 770        /*
 771         * Note that exit_ptrace() and find_new_reaper() might
 772         * drop tasklist_lock and reacquire it.
 773         */
 774        exit_ptrace(father);
 775        reaper = find_new_reaper(father);
 776
 777        list_for_each_entry_safe(p, n, &father->children, sibling) {
 778                struct task_struct *t = p;
 779                do {
 780                        t->real_parent = reaper;
 781                        if (t->parent == father) {
 782                                BUG_ON(t->ptrace);
 783                                t->parent = t->real_parent;
 784                        }
 785                        if (t->pdeath_signal)
 786                                group_send_sig_info(t->pdeath_signal,
 787                                                    SEND_SIG_NOINFO, t);
 788                } while_each_thread(p, t);
 789                reparent_leader(father, p, &dead_children);
 790        }
 791        write_unlock_irq(&tasklist_lock);
 792
 793        BUG_ON(!list_empty(&father->children));
 794
 795        list_for_each_entry_safe(p, n, &dead_children, sibling) {
 796                list_del_init(&p->sibling);
 797                release_task(p);
 798        }
 799}
 800
 801/*
 802 * Send signals to all our closest relatives so that they know
 803 * to properly mourn us..
 804 */
 805static void exit_notify(struct task_struct *tsk, int group_dead)
 806{
 807        bool autoreap;
 808
 809        /*
 810         * This does two things:
 811         *
 812         * A.  Make init inherit all the child processes
 813         * B.  Check to see if any process groups have become orphaned
 814         *      as a result of our exiting, and if they have any stopped
 815         *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 816         */
 817        forget_original_parent(tsk);
 818        exit_task_namespaces(tsk);
 819
 820        write_lock_irq(&tasklist_lock);
 821        if (group_dead)
 822                kill_orphaned_pgrp(tsk->group_leader, NULL);
 823
 824        /* Let father know we died
 825         *
 826         * Thread signals are configurable, but you aren't going to use
 827         * that to send signals to arbitrary processes.
 828         * That stops right now.
 829         *
 830         * If the parent exec id doesn't match the exec id we saved
 831         * when we started then we know the parent has changed security
 832         * domain.
 833         *
 834         * If our self_exec id doesn't match our parent_exec_id then
 835         * we have changed execution domain as these two values started
 836         * the same after a fork.
 837         */
 838        if (thread_group_leader(tsk) && tsk->exit_signal != SIGCHLD &&
 839            (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
 840             tsk->self_exec_id != tsk->parent_exec_id))
 841                tsk->exit_signal = SIGCHLD;
 842
 843        if (unlikely(tsk->ptrace)) {
 844                int sig = thread_group_leader(tsk) &&
 845                                thread_group_empty(tsk) &&
 846                                !ptrace_reparented(tsk) ?
 847                        tsk->exit_signal : SIGCHLD;
 848                autoreap = do_notify_parent(tsk, sig);
 849        } else if (thread_group_leader(tsk)) {
 850                autoreap = thread_group_empty(tsk) &&
 851                        do_notify_parent(tsk, tsk->exit_signal);
 852        } else {
 853                autoreap = true;
 854        }
 855
 856        tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
 857
 858        /* mt-exec, de_thread() is waiting for group leader */
 859        if (unlikely(tsk->signal->notify_count < 0))
 860                wake_up_process(tsk->signal->group_exit_task);
 861        write_unlock_irq(&tasklist_lock);
 862
 863        /* If the process is dead, release it - nobody will wait for it */
 864        if (autoreap)
 865                release_task(tsk);
 866}
 867
 868#ifdef CONFIG_DEBUG_STACK_USAGE
 869static void check_stack_usage(void)
 870{
 871        static DEFINE_SPINLOCK(low_water_lock);
 872        static int lowest_to_date = THREAD_SIZE;
 873        unsigned long free;
 874
 875        free = stack_not_used(current);
 876
 877        if (free >= lowest_to_date)
 878                return;
 879
 880        spin_lock(&low_water_lock);
 881        if (free < lowest_to_date) {
 882                printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
 883                                "left\n",
 884                                current->comm, free);
 885                lowest_to_date = free;
 886        }
 887        spin_unlock(&low_water_lock);
 888}
 889#else
 890static inline void check_stack_usage(void) {}
 891#endif
 892
 893NORET_TYPE void do_exit(long code)
 894{
 895        struct task_struct *tsk = current;
 896        int group_dead;
 897
 898        profile_task_exit(tsk);
 899
 900        WARN_ON(blk_needs_flush_plug(tsk));
 901
 902        if (unlikely(in_interrupt()))
 903                panic("Aiee, killing interrupt handler!");
 904        if (unlikely(!tsk->pid))
 905                panic("Attempted to kill the idle task!");
 906
 907        /*
 908         * If do_exit is called because this processes oopsed, it's possible
 909         * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
 910         * continuing. Amongst other possible reasons, this is to prevent
 911         * mm_release()->clear_child_tid() from writing to a user-controlled
 912         * kernel address.
 913         */
 914        set_fs(USER_DS);
 915
 916        ptrace_event(PTRACE_EVENT_EXIT, code);
 917
 918        validate_creds_for_do_exit(tsk);
 919
 920        /*
 921         * We're taking recursive faults here in do_exit. Safest is to just
 922         * leave this task alone and wait for reboot.
 923         */
 924        if (unlikely(tsk->flags & PF_EXITING)) {
 925                printk(KERN_ALERT
 926                        "Fixing recursive fault but reboot is needed!\n");
 927                /*
 928                 * We can do this unlocked here. The futex code uses
 929                 * this flag just to verify whether the pi state
 930                 * cleanup has been done or not. In the worst case it
 931                 * loops once more. We pretend that the cleanup was
 932                 * done as there is no way to return. Either the
 933                 * OWNER_DIED bit is set by now or we push the blocked
 934                 * task into the wait for ever nirwana as well.
 935                 */
 936                tsk->flags |= PF_EXITPIDONE;
 937                set_current_state(TASK_UNINTERRUPTIBLE);
 938                schedule();
 939        }
 940
 941        exit_irq_thread();
 942
 943        exit_signals(tsk);  /* sets PF_EXITING */
 944        /*
 945         * tsk->flags are checked in the futex code to protect against
 946         * an exiting task cleaning up the robust pi futexes.
 947         */
 948        smp_mb();
 949        raw_spin_unlock_wait(&tsk->pi_lock);
 950
 951        if (unlikely(in_atomic()))
 952                printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 953                                current->comm, task_pid_nr(current),
 954                                preempt_count());
 955
 956        acct_update_integrals(tsk);
 957        /* sync mm's RSS info before statistics gathering */
 958        if (tsk->mm)
 959                sync_mm_rss(tsk, tsk->mm);
 960        group_dead = atomic_dec_and_test(&tsk->signal->live);
 961        if (group_dead) {
 962                hrtimer_cancel(&tsk->signal->real_timer);
 963                exit_itimers(tsk->signal);
 964                if (tsk->mm)
 965                        setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
 966        }
 967        acct_collect(code, group_dead);
 968        if (group_dead)
 969                tty_audit_exit();
 970        if (unlikely(tsk->audit_context))
 971                audit_free(tsk);
 972
 973        tsk->exit_code = code;
 974        taskstats_exit(tsk, group_dead);
 975
 976        exit_mm(tsk);
 977
 978        if (group_dead)
 979                acct_process();
 980        trace_sched_process_exit(tsk);
 981
 982        exit_sem(tsk);
 983        exit_shm(tsk);
 984        exit_files(tsk);
 985        exit_fs(tsk);
 986        check_stack_usage();
 987        exit_thread();
 988
 989        /*
 990         * Flush inherited counters to the parent - before the parent
 991         * gets woken up by child-exit notifications.
 992         *
 993         * because of cgroup mode, must be called before cgroup_exit()
 994         */
 995        perf_event_exit_task(tsk);
 996
 997        cgroup_exit(tsk, 1);
 998
 999        if (group_dead)
1000                disassociate_ctty(1);
1001
1002        module_put(task_thread_info(tsk)->exec_domain->module);
1003
1004        proc_exit_connector(tsk);
1005
1006        /*
1007         * FIXME: do that only when needed, using sched_exit tracepoint
1008         */
1009        ptrace_put_breakpoints(tsk);
1010
1011        exit_notify(tsk, group_dead);
1012#ifdef CONFIG_NUMA
1013        task_lock(tsk);
1014        mpol_put(tsk->mempolicy);
1015        tsk->mempolicy = NULL;
1016        task_unlock(tsk);
1017#endif
1018#ifdef CONFIG_FUTEX
1019        if (unlikely(current->pi_state_cache))
1020                kfree(current->pi_state_cache);
1021#endif
1022        /*
1023         * Make sure we are holding no locks:
1024         */
1025        debug_check_no_locks_held(tsk);
1026        /*
1027         * We can do this unlocked here. The futex code uses this flag
1028         * just to verify whether the pi state cleanup has been done
1029         * or not. In the worst case it loops once more.
1030         */
1031        tsk->flags |= PF_EXITPIDONE;
1032
1033        if (tsk->io_context)
1034                exit_io_context(tsk);
1035
1036        if (tsk->splice_pipe)
1037                __free_pipe_info(tsk->splice_pipe);
1038
1039        validate_creds_for_do_exit(tsk);
1040
1041        preempt_disable();
1042        exit_rcu();
1043        /* causes final put_task_struct in finish_task_switch(). */
1044        tsk->state = TASK_DEAD;
1045        schedule();
1046        BUG();
1047        /* Avoid "noreturn function does return".  */
1048        for (;;)
1049                cpu_relax();    /* For when BUG is null */
1050}
1051
1052EXPORT_SYMBOL_GPL(do_exit);
1053
1054NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1055{
1056        if (comp)
1057                complete(comp);
1058
1059        do_exit(code);
1060}
1061
1062EXPORT_SYMBOL(complete_and_exit);
1063
1064SYSCALL_DEFINE1(exit, int, error_code)
1065{
1066        do_exit((error_code&0xff)<<8);
1067}
1068
1069/*
1070 * Take down every thread in the group.  This is called by fatal signals
1071 * as well as by sys_exit_group (below).
1072 */
1073NORET_TYPE void
1074do_group_exit(int exit_code)
1075{
1076        struct signal_struct *sig = current->signal;
1077
1078        BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1079
1080        if (signal_group_exit(sig))
1081                exit_code = sig->group_exit_code;
1082        else if (!thread_group_empty(current)) {
1083                struct sighand_struct *const sighand = current->sighand;
1084                spin_lock_irq(&sighand->siglock);
1085                if (signal_group_exit(sig))
1086                        /* Another thread got here before we took the lock.  */
1087                        exit_code = sig->group_exit_code;
1088                else {
1089                        sig->group_exit_code = exit_code;
1090                        sig->flags = SIGNAL_GROUP_EXIT;
1091                        zap_other_threads(current);
1092                }
1093                spin_unlock_irq(&sighand->siglock);
1094        }
1095
1096        do_exit(exit_code);
1097        /* NOTREACHED */
1098}
1099
1100/*
1101 * this kills every thread in the thread group. Note that any externally
1102 * wait4()-ing process will get the correct exit code - even if this
1103 * thread is not the thread group leader.
1104 */
1105SYSCALL_DEFINE1(exit_group, int, error_code)
1106{
1107        do_group_exit((error_code & 0xff) << 8);
1108        /* NOTREACHED */
1109        return 0;
1110}
1111
1112struct wait_opts {
1113        enum pid_type           wo_type;
1114        int                     wo_flags;
1115        struct pid              *wo_pid;
1116
1117        struct siginfo __user   *wo_info;
1118        int __user              *wo_stat;
1119        struct rusage __user    *wo_rusage;
1120
1121        wait_queue_t            child_wait;
1122        int                     notask_error;
1123};
1124
1125static inline
1126struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1127{
1128        if (type != PIDTYPE_PID)
1129                task = task->group_leader;
1130        return task->pids[type].pid;
1131}
1132
1133static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1134{
1135        return  wo->wo_type == PIDTYPE_MAX ||
1136                task_pid_type(p, wo->wo_type) == wo->wo_pid;
1137}
1138
1139static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1140{
1141        if (!eligible_pid(wo, p))
1142                return 0;
1143        /* Wait for all children (clone and not) if __WALL is set;
1144         * otherwise, wait for clone children *only* if __WCLONE is
1145         * set; otherwise, wait for non-clone children *only*.  (Note:
1146         * A "clone" child here is one that reports to its parent
1147         * using a signal other than SIGCHLD.) */
1148        if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1149            && !(wo->wo_flags & __WALL))
1150                return 0;
1151
1152        return 1;
1153}
1154
1155static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1156                                pid_t pid, uid_t uid, int why, int status)
1157{
1158        struct siginfo __user *infop;
1159        int retval = wo->wo_rusage
1160                ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1161
1162        put_task_struct(p);
1163        infop = wo->wo_info;
1164        if (infop) {
1165                if (!retval)
1166                        retval = put_user(SIGCHLD, &infop->si_signo);
1167                if (!retval)
1168                        retval = put_user(0, &infop->si_errno);
1169                if (!retval)
1170                        retval = put_user((short)why, &infop->si_code);
1171                if (!retval)
1172                        retval = put_user(pid, &infop->si_pid);
1173                if (!retval)
1174                        retval = put_user(uid, &infop->si_uid);
1175                if (!retval)
1176                        retval = put_user(status, &infop->si_status);
1177        }
1178        if (!retval)
1179                retval = pid;
1180        return retval;
1181}
1182
1183/*
1184 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1185 * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1186 * the lock and this task is uninteresting.  If we return nonzero, we have
1187 * released the lock and the system call should return.
1188 */
1189static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1190{
1191        unsigned long state;
1192        int retval, status, traced;
1193        pid_t pid = task_pid_vnr(p);
1194        uid_t uid = __task_cred(p)->uid;
1195        struct siginfo __user *infop;
1196
1197        if (!likely(wo->wo_flags & WEXITED))
1198                return 0;
1199
1200        if (unlikely(wo->wo_flags & WNOWAIT)) {
1201                int exit_code = p->exit_code;
1202                int why;
1203
1204                get_task_struct(p);
1205                read_unlock(&tasklist_lock);
1206                if ((exit_code & 0x7f) == 0) {
1207                        why = CLD_EXITED;
1208                        status = exit_code >> 8;
1209                } else {
1210                        why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1211                        status = exit_code & 0x7f;
1212                }
1213                return wait_noreap_copyout(wo, p, pid, uid, why, status);
1214        }
1215
1216        /*
1217         * Try to move the task's state to DEAD
1218         * only one thread is allowed to do this:
1219         */
1220        state = xchg(&p->exit_state, EXIT_DEAD);
1221        if (state != EXIT_ZOMBIE) {
1222                BUG_ON(state != EXIT_DEAD);
1223                return 0;
1224        }
1225
1226        traced = ptrace_reparented(p);
1227        /*
1228         * It can be ptraced but not reparented, check
1229         * thread_group_leader() to filter out sub-threads.
1230         */
1231        if (likely(!traced) && thread_group_leader(p)) {
1232                struct signal_struct *psig;
1233                struct signal_struct *sig;
1234                unsigned long maxrss;
1235                cputime_t tgutime, tgstime;
1236
1237                /*
1238                 * The resource counters for the group leader are in its
1239                 * own task_struct.  Those for dead threads in the group
1240                 * are in its signal_struct, as are those for the child
1241                 * processes it has previously reaped.  All these
1242                 * accumulate in the parent's signal_struct c* fields.
1243                 *
1244                 * We don't bother to take a lock here to protect these
1245                 * p->signal fields, because they are only touched by
1246                 * __exit_signal, which runs with tasklist_lock
1247                 * write-locked anyway, and so is excluded here.  We do
1248                 * need to protect the access to parent->signal fields,
1249                 * as other threads in the parent group can be right
1250                 * here reaping other children at the same time.
1251                 *
1252                 * We use thread_group_times() to get times for the thread
1253                 * group, which consolidates times for all threads in the
1254                 * group including the group leader.
1255                 */
1256                thread_group_times(p, &tgutime, &tgstime);
1257                spin_lock_irq(&p->real_parent->sighand->siglock);
1258                psig = p->real_parent->signal;
1259                sig = p->signal;
1260                psig->cutime =
1261                        cputime_add(psig->cutime,
1262                        cputime_add(tgutime,
1263                                    sig->cutime));
1264                psig->cstime =
1265                        cputime_add(psig->cstime,
1266                        cputime_add(tgstime,
1267                                    sig->cstime));
1268                psig->cgtime =
1269                        cputime_add(psig->cgtime,
1270                        cputime_add(p->gtime,
1271                        cputime_add(sig->gtime,
1272                                    sig->cgtime)));
1273                psig->cmin_flt +=
1274                        p->min_flt + sig->min_flt + sig->cmin_flt;
1275                psig->cmaj_flt +=
1276                        p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1277                psig->cnvcsw +=
1278                        p->nvcsw + sig->nvcsw + sig->cnvcsw;
1279                psig->cnivcsw +=
1280                        p->nivcsw + sig->nivcsw + sig->cnivcsw;
1281                psig->cinblock +=
1282                        task_io_get_inblock(p) +
1283                        sig->inblock + sig->cinblock;
1284                psig->coublock +=
1285                        task_io_get_oublock(p) +
1286                        sig->oublock + sig->coublock;
1287                maxrss = max(sig->maxrss, sig->cmaxrss);
1288                if (psig->cmaxrss < maxrss)
1289                        psig->cmaxrss = maxrss;
1290                task_io_accounting_add(&psig->ioac, &p->ioac);
1291                task_io_accounting_add(&psig->ioac, &sig->ioac);
1292                spin_unlock_irq(&p->real_parent->sighand->siglock);
1293        }
1294
1295        /*
1296         * Now we are sure this task is interesting, and no other
1297         * thread can reap it because we set its state to EXIT_DEAD.
1298         */
1299        read_unlock(&tasklist_lock);
1300
1301        retval = wo->wo_rusage
1302                ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1303        status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1304                ? p->signal->group_exit_code : p->exit_code;
1305        if (!retval && wo->wo_stat)
1306                retval = put_user(status, wo->wo_stat);
1307
1308        infop = wo->wo_info;
1309        if (!retval && infop)
1310                retval = put_user(SIGCHLD, &infop->si_signo);
1311        if (!retval && infop)
1312                retval = put_user(0, &infop->si_errno);
1313        if (!retval && infop) {
1314                int why;
1315
1316                if ((status & 0x7f) == 0) {
1317                        why = CLD_EXITED;
1318                        status >>= 8;
1319                } else {
1320                        why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1321                        status &= 0x7f;
1322                }
1323                retval = put_user((short)why, &infop->si_code);
1324                if (!retval)
1325                        retval = put_user(status, &infop->si_status);
1326        }
1327        if (!retval && infop)
1328                retval = put_user(pid, &infop->si_pid);
1329        if (!retval && infop)
1330                retval = put_user(uid, &infop->si_uid);
1331        if (!retval)
1332                retval = pid;
1333
1334        if (traced) {
1335                write_lock_irq(&tasklist_lock);
1336                /* We dropped tasklist, ptracer could die and untrace */
1337                ptrace_unlink(p);
1338                /*
1339                 * If this is not a sub-thread, notify the parent.
1340                 * If parent wants a zombie, don't release it now.
1341                 */
1342                if (thread_group_leader(p) &&
1343                    !do_notify_parent(p, p->exit_signal)) {
1344                        p->exit_state = EXIT_ZOMBIE;
1345                        p = NULL;
1346                }
1347                write_unlock_irq(&tasklist_lock);
1348        }
1349        if (p != NULL)
1350                release_task(p);
1351
1352        return retval;
1353}
1354
1355static int *task_stopped_code(struct task_struct *p, bool ptrace)
1356{
1357        if (ptrace) {
1358                if (task_is_stopped_or_traced(p) &&
1359                    !(p->jobctl & JOBCTL_LISTENING))
1360                        return &p->exit_code;
1361        } else {
1362                if (p->signal->flags & SIGNAL_STOP_STOPPED)
1363                        return &p->signal->group_exit_code;
1364        }
1365        return NULL;
1366}
1367
1368/**
1369 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1370 * @wo: wait options
1371 * @ptrace: is the wait for ptrace
1372 * @p: task to wait for
1373 *
1374 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1375 *
1376 * CONTEXT:
1377 * read_lock(&tasklist_lock), which is released if return value is
1378 * non-zero.  Also, grabs and releases @p->sighand->siglock.
1379 *
1380 * RETURNS:
1381 * 0 if wait condition didn't exist and search for other wait conditions
1382 * should continue.  Non-zero return, -errno on failure and @p's pid on
1383 * success, implies that tasklist_lock is released and wait condition
1384 * search should terminate.
1385 */
1386static int wait_task_stopped(struct wait_opts *wo,
1387                                int ptrace, struct task_struct *p)
1388{
1389        struct siginfo __user *infop;
1390        int retval, exit_code, *p_code, why;
1391        uid_t uid = 0; /* unneeded, required by compiler */
1392        pid_t pid;
1393
1394        /*
1395         * Traditionally we see ptrace'd stopped tasks regardless of options.
1396         */
1397        if (!ptrace && !(wo->wo_flags & WUNTRACED))
1398                return 0;
1399
1400        if (!task_stopped_code(p, ptrace))
1401                return 0;
1402
1403        exit_code = 0;
1404        spin_lock_irq(&p->sighand->siglock);
1405
1406        p_code = task_stopped_code(p, ptrace);
1407        if (unlikely(!p_code))
1408                goto unlock_sig;
1409
1410        exit_code = *p_code;
1411        if (!exit_code)
1412                goto unlock_sig;
1413
1414        if (!unlikely(wo->wo_flags & WNOWAIT))
1415                *p_code = 0;
1416
1417        uid = task_uid(p);
1418unlock_sig:
1419        spin_unlock_irq(&p->sighand->siglock);
1420        if (!exit_code)
1421                return 0;
1422
1423        /*
1424         * Now we are pretty sure this task is interesting.
1425         * Make sure it doesn't get reaped out from under us while we
1426         * give up the lock and then examine it below.  We don't want to
1427         * keep holding onto the tasklist_lock while we call getrusage and
1428         * possibly take page faults for user memory.
1429         */
1430        get_task_struct(p);
1431        pid = task_pid_vnr(p);
1432        why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1433        read_unlock(&tasklist_lock);
1434
1435        if (unlikely(wo->wo_flags & WNOWAIT))
1436                return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1437
1438        retval = wo->wo_rusage
1439                ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1440        if (!retval && wo->wo_stat)
1441                retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1442
1443        infop = wo->wo_info;
1444        if (!retval && infop)
1445                retval = put_user(SIGCHLD, &infop->si_signo);
1446        if (!retval && infop)
1447                retval = put_user(0, &infop->si_errno);
1448        if (!retval && infop)
1449                retval = put_user((short)why, &infop->si_code);
1450        if (!retval && infop)
1451                retval = put_user(exit_code, &infop->si_status);
1452        if (!retval && infop)
1453                retval = put_user(pid, &infop->si_pid);
1454        if (!retval && infop)
1455                retval = put_user(uid, &infop->si_uid);
1456        if (!retval)
1457                retval = pid;
1458        put_task_struct(p);
1459
1460        BUG_ON(!retval);
1461        return retval;
1462}
1463
1464/*
1465 * Handle do_wait work for one task in a live, non-stopped state.
1466 * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1467 * the lock and this task is uninteresting.  If we return nonzero, we have
1468 * released the lock and the system call should return.
1469 */
1470static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1471{
1472        int retval;
1473        pid_t pid;
1474        uid_t uid;
1475
1476        if (!unlikely(wo->wo_flags & WCONTINUED))
1477                return 0;
1478
1479        if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1480                return 0;
1481
1482        spin_lock_irq(&p->sighand->siglock);
1483        /* Re-check with the lock held.  */
1484        if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1485                spin_unlock_irq(&p->sighand->siglock);
1486                return 0;
1487        }
1488        if (!unlikely(wo->wo_flags & WNOWAIT))
1489                p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1490        uid = task_uid(p);
1491        spin_unlock_irq(&p->sighand->siglock);
1492
1493        pid = task_pid_vnr(p);
1494        get_task_struct(p);
1495        read_unlock(&tasklist_lock);
1496
1497        if (!wo->wo_info) {
1498                retval = wo->wo_rusage
1499                        ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1500                put_task_struct(p);
1501                if (!retval && wo->wo_stat)
1502                        retval = put_user(0xffff, wo->wo_stat);
1503                if (!retval)
1504                        retval = pid;
1505        } else {
1506                retval = wait_noreap_copyout(wo, p, pid, uid,
1507                                             CLD_CONTINUED, SIGCONT);
1508                BUG_ON(retval == 0);
1509        }
1510
1511        return retval;
1512}
1513
1514/*
1515 * Consider @p for a wait by @parent.
1516 *
1517 * -ECHILD should be in ->notask_error before the first call.
1518 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1519 * Returns zero if the search for a child should continue;
1520 * then ->notask_error is 0 if @p is an eligible child,
1521 * or another error from security_task_wait(), or still -ECHILD.
1522 */
1523static int wait_consider_task(struct wait_opts *wo, int ptrace,
1524                                struct task_struct *p)
1525{
1526        int ret = eligible_child(wo, p);
1527        if (!ret)
1528                return ret;
1529
1530        ret = security_task_wait(p);
1531        if (unlikely(ret < 0)) {
1532                /*
1533                 * If we have not yet seen any eligible child,
1534                 * then let this error code replace -ECHILD.
1535                 * A permission error will give the user a clue
1536                 * to look for security policy problems, rather
1537                 * than for mysterious wait bugs.
1538                 */
1539                if (wo->notask_error)
1540                        wo->notask_error = ret;
1541                return 0;
1542        }
1543
1544        /* dead body doesn't have much to contribute */
1545        if (p->exit_state == EXIT_DEAD)
1546                return 0;
1547
1548        /* slay zombie? */
1549        if (p->exit_state == EXIT_ZOMBIE) {
1550                /*
1551                 * A zombie ptracee is only visible to its ptracer.
1552                 * Notification and reaping will be cascaded to the real
1553                 * parent when the ptracer detaches.
1554                 */
1555                if (likely(!ptrace) && unlikely(p->ptrace)) {
1556                        /* it will become visible, clear notask_error */
1557                        wo->notask_error = 0;
1558                        return 0;
1559                }
1560
1561                /* we don't reap group leaders with subthreads */
1562                if (!delay_group_leader(p))
1563                        return wait_task_zombie(wo, p);
1564
1565                /*
1566                 * Allow access to stopped/continued state via zombie by
1567                 * falling through.  Clearing of notask_error is complex.
1568                 *
1569                 * When !@ptrace:
1570                 *
1571                 * If WEXITED is set, notask_error should naturally be
1572                 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1573                 * so, if there are live subthreads, there are events to
1574                 * wait for.  If all subthreads are dead, it's still safe
1575                 * to clear - this function will be called again in finite
1576                 * amount time once all the subthreads are released and
1577                 * will then return without clearing.
1578                 *
1579                 * When @ptrace:
1580                 *
1581                 * Stopped state is per-task and thus can't change once the
1582                 * target task dies.  Only continued and exited can happen.
1583                 * Clear notask_error if WCONTINUED | WEXITED.
1584                 */
1585                if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1586                        wo->notask_error = 0;
1587        } else {
1588                /*
1589                 * If @p is ptraced by a task in its real parent's group,
1590                 * hide group stop/continued state when looking at @p as
1591                 * the real parent; otherwise, a single stop can be
1592                 * reported twice as group and ptrace stops.
1593                 *
1594                 * If a ptracer wants to distinguish the two events for its
1595                 * own children, it should create a separate process which
1596                 * takes the role of real parent.
1597                 */
1598                if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1599                        return 0;
1600
1601                /*
1602                 * @p is alive and it's gonna stop, continue or exit, so
1603                 * there always is something to wait for.
1604                 */
1605                wo->notask_error = 0;
1606        }
1607
1608        /*
1609         * Wait for stopped.  Depending on @ptrace, different stopped state
1610         * is used and the two don't interact with each other.
1611         */
1612        ret = wait_task_stopped(wo, ptrace, p);
1613        if (ret)
1614                return ret;
1615
1616        /*
1617         * Wait for continued.  There's only one continued state and the
1618         * ptracer can consume it which can confuse the real parent.  Don't
1619         * use WCONTINUED from ptracer.  You don't need or want it.
1620         */
1621        return wait_task_continued(wo, p);
1622}
1623
1624/*
1625 * Do the work of do_wait() for one thread in the group, @tsk.
1626 *
1627 * -ECHILD should be in ->notask_error before the first call.
1628 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1629 * Returns zero if the search for a child should continue; then
1630 * ->notask_error is 0 if there were any eligible children,
1631 * or another error from security_task_wait(), or still -ECHILD.
1632 */
1633static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1634{
1635        struct task_struct *p;
1636
1637        list_for_each_entry(p, &tsk->children, sibling) {
1638                int ret = wait_consider_task(wo, 0, p);
1639                if (ret)
1640                        return ret;
1641        }
1642
1643        return 0;
1644}
1645
1646static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1647{
1648        struct task_struct *p;
1649
1650        list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1651                int ret = wait_consider_task(wo, 1, p);
1652                if (ret)
1653                        return ret;
1654        }
1655
1656        return 0;
1657}
1658
1659static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1660                                int sync, void *key)
1661{
1662        struct wait_opts *wo = container_of(wait, struct wait_opts,
1663                                                child_wait);
1664        struct task_struct *p = key;
1665
1666        if (!eligible_pid(wo, p))
1667                return 0;
1668
1669        if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1670                return 0;
1671
1672        return default_wake_function(wait, mode, sync, key);
1673}
1674
1675void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1676{
1677        __wake_up_sync_key(&parent->signal->wait_chldexit,
1678                                TASK_INTERRUPTIBLE, 1, p);
1679}
1680
1681static long do_wait(struct wait_opts *wo)
1682{
1683        struct task_struct *tsk;
1684        int retval;
1685
1686        trace_sched_process_wait(wo->wo_pid);
1687
1688        init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1689        wo->child_wait.private = current;
1690        add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1691repeat:
1692        /*
1693         * If there is nothing that can match our critiera just get out.
1694         * We will clear ->notask_error to zero if we see any child that
1695         * might later match our criteria, even if we are not able to reap
1696         * it yet.
1697         */
1698        wo->notask_error = -ECHILD;
1699        if ((wo->wo_type < PIDTYPE_MAX) &&
1700           (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1701                goto notask;
1702
1703        set_current_state(TASK_INTERRUPTIBLE);
1704        read_lock(&tasklist_lock);
1705        tsk = current;
1706        do {
1707                retval = do_wait_thread(wo, tsk);
1708                if (retval)
1709                        goto end;
1710
1711                retval = ptrace_do_wait(wo, tsk);
1712                if (retval)
1713                        goto end;
1714
1715                if (wo->wo_flags & __WNOTHREAD)
1716                        break;
1717        } while_each_thread(current, tsk);
1718        read_unlock(&tasklist_lock);
1719
1720notask:
1721        retval = wo->notask_error;
1722        if (!retval && !(wo->wo_flags & WNOHANG)) {
1723                retval = -ERESTARTSYS;
1724                if (!signal_pending(current)) {
1725                        schedule();
1726                        goto repeat;
1727                }
1728        }
1729end:
1730        __set_current_state(TASK_RUNNING);
1731        remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1732        return retval;
1733}
1734
1735SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1736                infop, int, options, struct rusage __user *, ru)
1737{
1738        struct wait_opts wo;
1739        struct pid *pid = NULL;
1740        enum pid_type type;
1741        long ret;
1742
1743        if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1744                return -EINVAL;
1745        if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1746                return -EINVAL;
1747
1748        switch (which) {
1749        case P_ALL:
1750                type = PIDTYPE_MAX;
1751                break;
1752        case P_PID:
1753                type = PIDTYPE_PID;
1754                if (upid <= 0)
1755                        return -EINVAL;
1756                break;
1757        case P_PGID:
1758                type = PIDTYPE_PGID;
1759                if (upid <= 0)
1760                        return -EINVAL;
1761                break;
1762        default:
1763                return -EINVAL;
1764        }
1765
1766        if (type < PIDTYPE_MAX)
1767                pid = find_get_pid(upid);
1768
1769        wo.wo_type      = type;
1770        wo.wo_pid       = pid;
1771        wo.wo_flags     = options;
1772        wo.wo_info      = infop;
1773        wo.wo_stat      = NULL;
1774        wo.wo_rusage    = ru;
1775        ret = do_wait(&wo);
1776
1777        if (ret > 0) {
1778                ret = 0;
1779        } else if (infop) {
1780                /*
1781                 * For a WNOHANG return, clear out all the fields
1782                 * we would set so the user can easily tell the
1783                 * difference.
1784                 */
1785                if (!ret)
1786                        ret = put_user(0, &infop->si_signo);
1787                if (!ret)
1788                        ret = put_user(0, &infop->si_errno);
1789                if (!ret)
1790                        ret = put_user(0, &infop->si_code);
1791                if (!ret)
1792                        ret = put_user(0, &infop->si_pid);
1793                if (!ret)
1794                        ret = put_user(0, &infop->si_uid);
1795                if (!ret)
1796                        ret = put_user(0, &infop->si_status);
1797        }
1798
1799        put_pid(pid);
1800
1801        /* avoid REGPARM breakage on x86: */
1802        asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1803        return ret;
1804}
1805
1806SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1807                int, options, struct rusage __user *, ru)
1808{
1809        struct wait_opts wo;
1810        struct pid *pid = NULL;
1811        enum pid_type type;
1812        long ret;
1813
1814        if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1815                        __WNOTHREAD|__WCLONE|__WALL))
1816                return -EINVAL;
1817
1818        if (upid == -1)
1819                type = PIDTYPE_MAX;
1820        else if (upid < 0) {
1821                type = PIDTYPE_PGID;
1822                pid = find_get_pid(-upid);
1823        } else if (upid == 0) {
1824                type = PIDTYPE_PGID;
1825                pid = get_task_pid(current, PIDTYPE_PGID);
1826        } else /* upid > 0 */ {
1827                type = PIDTYPE_PID;
1828                pid = find_get_pid(upid);
1829        }
1830
1831        wo.wo_type      = type;
1832        wo.wo_pid       = pid;
1833        wo.wo_flags     = options | WEXITED;
1834        wo.wo_info      = NULL;
1835        wo.wo_stat      = stat_addr;
1836        wo.wo_rusage    = ru;
1837        ret = do_wait(&wo);
1838        put_pid(pid);
1839
1840        /* avoid REGPARM breakage on x86: */
1841        asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1842        return ret;
1843}
1844
1845#ifdef __ARCH_WANT_SYS_WAITPID
1846
1847/*
1848 * sys_waitpid() remains for compatibility. waitpid() should be
1849 * implemented by calling sys_wait4() from libc.a.
1850 */
1851SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1852{
1853        return sys_wait4(pid, stat_addr, options, NULL);
1854}
1855
1856#endif
1857
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.