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#include <linux/writeback.h>
  55#include <linux/shm.h>
  56
  57#include <asm/uaccess.h>
  58#include <asm/unistd.h>
  59#include <asm/pgtable.h>
  60#include <asm/mmu_context.h>
  61
  62static void exit_mm(struct task_struct * tsk);
  63
  64static void __unhash_process(struct task_struct *p, bool group_dead)
  65{
  66        nr_threads--;
  67        detach_pid(p, PIDTYPE_PID);
  68        if (group_dead) {
  69                detach_pid(p, PIDTYPE_PGID);
  70                detach_pid(p, PIDTYPE_SID);
  71
  72                list_del_rcu(&p->tasks);
  73                list_del_init(&p->sibling);
  74                __this_cpu_dec(process_counts);
  75                /*
  76                 * If we are the last child process in a pid namespace to be
  77                 * reaped, notify the reaper sleeping zap_pid_ns_processes().
  78                 */
  79                if (IS_ENABLED(CONFIG_PID_NS)) {
  80                        struct task_struct *parent = p->real_parent;
  81
  82                        if ((task_active_pid_ns(parent)->child_reaper == parent) &&
  83                            list_empty(&parent->children) &&
  84                            (parent->flags & PF_EXITING))
  85                                wake_up_process(parent);
  86                }
  87        }
  88        list_del_rcu(&p->thread_group);
  89}
  90
  91/*
  92 * This function expects the tasklist_lock write-locked.
  93 */
  94static void __exit_signal(struct task_struct *tsk)
  95{
  96        struct signal_struct *sig = tsk->signal;
  97        bool group_dead = thread_group_leader(tsk);
  98        struct sighand_struct *sighand;
  99        struct tty_struct *uninitialized_var(tty);
 100
 101        sighand = rcu_dereference_check(tsk->sighand,
 102                                        lockdep_tasklist_lock_is_held());
 103        spin_lock(&sighand->siglock);
 104
 105        posix_cpu_timers_exit(tsk);
 106        if (group_dead) {
 107                posix_cpu_timers_exit_group(tsk);
 108                tty = sig->tty;
 109                sig->tty = NULL;
 110        } else {
 111                /*
 112                 * This can only happen if the caller is de_thread().
 113                 * FIXME: this is the temporary hack, we should teach
 114                 * posix-cpu-timers to handle this case correctly.
 115                 */
 116                if (unlikely(has_group_leader_pid(tsk)))
 117                        posix_cpu_timers_exit_group(tsk);
 118
 119                /*
 120                 * If there is any task waiting for the group exit
 121                 * then notify it:
 122                 */
 123                if (sig->notify_count > 0 && !--sig->notify_count)
 124                        wake_up_process(sig->group_exit_task);
 125
 126                if (tsk == sig->curr_target)
 127                        sig->curr_target = next_thread(tsk);
 128                /*
 129                 * Accumulate here the counters for all threads but the
 130                 * group leader as they die, so they can be added into
 131                 * the process-wide totals when those are taken.
 132                 * The group leader stays around as a zombie as long
 133                 * as there are other threads.  When it gets reaped,
 134                 * the exit.c code will add its counts into these totals.
 135                 * We won't ever get here for the group leader, since it
 136                 * will have been the last reference on the signal_struct.
 137                 */
 138                sig->utime += tsk->utime;
 139                sig->stime += tsk->stime;
 140                sig->gtime += tsk->gtime;
 141                sig->min_flt += tsk->min_flt;
 142                sig->maj_flt += tsk->maj_flt;
 143                sig->nvcsw += tsk->nvcsw;
 144                sig->nivcsw += tsk->nivcsw;
 145                sig->inblock += task_io_get_inblock(tsk);
 146                sig->oublock += task_io_get_oublock(tsk);
 147                task_io_accounting_add(&sig->ioac, &tsk->ioac);
 148                sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 149        }
 150
 151        sig->nr_threads--;
 152        __unhash_process(tsk, group_dead);
 153
 154        /*
 155         * Do this under ->siglock, we can race with another thread
 156         * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
 157         */
 158        flush_sigqueue(&tsk->pending);
 159        tsk->sighand = NULL;
 160        spin_unlock(&sighand->siglock);
 161
 162        __cleanup_sighand(sighand);
 163        clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 164        if (group_dead) {
 165                flush_sigqueue(&sig->shared_pending);
 166                tty_kref_put(tty);
 167        }
 168}
 169
 170static void delayed_put_task_struct(struct rcu_head *rhp)
 171{
 172        struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
 173
 174        perf_event_delayed_put(tsk);
 175        trace_sched_process_free(tsk);
 176        put_task_struct(tsk);
 177}
 178
 179
 180void release_task(struct task_struct * p)
 181{
 182        struct task_struct *leader;
 183        int zap_leader;
 184repeat:
 185        /* don't need to get the RCU readlock here - the process is dead and
 186         * can't be modifying its own credentials. But shut RCU-lockdep up */
 187        rcu_read_lock();
 188        atomic_dec(&__task_cred(p)->user->processes);
 189        rcu_read_unlock();
 190
 191        proc_flush_task(p);
 192
 193        write_lock_irq(&tasklist_lock);
 194        ptrace_release_task(p);
 195        __exit_signal(p);
 196
 197        /*
 198         * If we are the last non-leader member of the thread
 199         * group, and the leader is zombie, then notify the
 200         * group leader's parent process. (if it wants notification.)
 201         */
 202        zap_leader = 0;
 203        leader = p->group_leader;
 204        if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 205                /*
 206                 * If we were the last child thread and the leader has
 207                 * exited already, and the leader's parent ignores SIGCHLD,
 208                 * then we are the one who should release the leader.
 209                 */
 210                zap_leader = do_notify_parent(leader, leader->exit_signal);
 211                if (zap_leader)
 212                        leader->exit_state = EXIT_DEAD;
 213        }
 214
 215        write_unlock_irq(&tasklist_lock);
 216        release_thread(p);
 217        call_rcu(&p->rcu, delayed_put_task_struct);
 218
 219        p = leader;
 220        if (unlikely(zap_leader))
 221                goto repeat;
 222}
 223
 224/*
 225 * This checks not only the pgrp, but falls back on the pid if no
 226 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 227 * without this...
 228 *
 229 * The caller must hold rcu lock or the tasklist lock.
 230 */
 231struct pid *session_of_pgrp(struct pid *pgrp)
 232{
 233        struct task_struct *p;
 234        struct pid *sid = NULL;
 235
 236        p = pid_task(pgrp, PIDTYPE_PGID);
 237        if (p == NULL)
 238                p = pid_task(pgrp, PIDTYPE_PID);
 239        if (p != NULL)
 240                sid = task_session(p);
 241
 242        return sid;
 243}
 244
 245/*
 246 * Determine if a process group is "orphaned", according to the POSIX
 247 * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 248 * by terminal-generated stop signals.  Newly orphaned process groups are
 249 * to receive a SIGHUP and a SIGCONT.
 250 *
 251 * "I ask you, have you ever known what it is to be an orphan?"
 252 */
 253static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 254{
 255        struct task_struct *p;
 256
 257        do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 258                if ((p == ignored_task) ||
 259                    (p->exit_state && thread_group_empty(p)) ||
 260                    is_global_init(p->real_parent))
 261                        continue;
 262
 263                if (task_pgrp(p->real_parent) != pgrp &&
 264                    task_session(p->real_parent) == task_session(p))
 265                        return 0;
 266        } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 267
 268        return 1;
 269}
 270
 271int is_current_pgrp_orphaned(void)
 272{
 273        int retval;
 274
 275        read_lock(&tasklist_lock);
 276        retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 277        read_unlock(&tasklist_lock);
 278
 279        return retval;
 280}
 281
 282static bool has_stopped_jobs(struct pid *pgrp)
 283{
 284        struct task_struct *p;
 285
 286        do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 287                if (p->signal->flags & SIGNAL_STOP_STOPPED)
 288                        return true;
 289        } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 290
 291        return false;
 292}
 293
 294/*
 295 * Check to see if any process groups have become orphaned as
 296 * a result of our exiting, and if they have any stopped jobs,
 297 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
 298 */
 299static void
 300kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
 301{
 302        struct pid *pgrp = task_pgrp(tsk);
 303        struct task_struct *ignored_task = tsk;
 304
 305        if (!parent)
 306                 /* exit: our father is in a different pgrp than
 307                  * we are and we were the only connection outside.
 308                  */
 309                parent = tsk->real_parent;
 310        else
 311                /* reparent: our child is in a different pgrp than
 312                 * we are, and it was the only connection outside.
 313                 */
 314                ignored_task = NULL;
 315
 316        if (task_pgrp(parent) != pgrp &&
 317            task_session(parent) == task_session(tsk) &&
 318            will_become_orphaned_pgrp(pgrp, ignored_task) &&
 319            has_stopped_jobs(pgrp)) {
 320                __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 321                __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 322        }
 323}
 324
 325/**
 326 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
 327 *
 328 * If a kernel thread is launched as a result of a system call, or if
 329 * it ever exits, it should generally reparent itself to kthreadd so it
 330 * isn't in the way of other processes and is correctly cleaned up on exit.
 331 *
 332 * The various task state such as scheduling policy and priority may have
 333 * been inherited from a user process, so we reset them to sane values here.
 334 *
 335 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
 336 */
 337static void reparent_to_kthreadd(void)
 338{
 339        write_lock_irq(&tasklist_lock);
 340
 341        ptrace_unlink(current);
 342        /* Reparent to init */
 343        current->real_parent = current->parent = kthreadd_task;
 344        list_move_tail(&current->sibling, &current->real_parent->children);
 345
 346        /* Set the exit signal to SIGCHLD so we signal init on exit */
 347        current->exit_signal = SIGCHLD;
 348
 349        if (task_nice(current) < 0)
 350                set_user_nice(current, 0);
 351        /* cpus_allowed? */
 352        /* rt_priority? */
 353        /* signals? */
 354        memcpy(current->signal->rlim, init_task.signal->rlim,
 355               sizeof(current->signal->rlim));
 356
 357        atomic_inc(&init_cred.usage);
 358        commit_creds(&init_cred);
 359        write_unlock_irq(&tasklist_lock);
 360}
 361
 362void __set_special_pids(struct pid *pid)
 363{
 364        struct task_struct *curr = current->group_leader;
 365
 366        if (task_session(curr) != pid)
 367                change_pid(curr, PIDTYPE_SID, pid);
 368
 369        if (task_pgrp(curr) != pid)
 370                change_pid(curr, PIDTYPE_PGID, pid);
 371}
 372
 373static void set_special_pids(struct pid *pid)
 374{
 375        write_lock_irq(&tasklist_lock);
 376        __set_special_pids(pid);
 377        write_unlock_irq(&tasklist_lock);
 378}
 379
 380/*
 381 * Let kernel threads use this to say that they allow a certain signal.
 382 * Must not be used if kthread was cloned with CLONE_SIGHAND.
 383 */
 384int allow_signal(int sig)
 385{
 386        if (!valid_signal(sig) || sig < 1)
 387                return -EINVAL;
 388
 389        spin_lock_irq(&current->sighand->siglock);
 390        /* This is only needed for daemonize()'ed kthreads */
 391        sigdelset(&current->blocked, sig);
 392        /*
 393         * Kernel threads handle their own signals. Let the signal code
 394         * know it'll be handled, so that they don't get converted to
 395         * SIGKILL or just silently dropped.
 396         */
 397        current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 398        recalc_sigpending();
 399        spin_unlock_irq(&current->sighand->siglock);
 400        return 0;
 401}
 402
 403EXPORT_SYMBOL(allow_signal);
 404
 405int disallow_signal(int sig)
 406{
 407        if (!valid_signal(sig) || sig < 1)
 408                return -EINVAL;
 409
 410        spin_lock_irq(&current->sighand->siglock);
 411        current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
 412        recalc_sigpending();
 413        spin_unlock_irq(&current->sighand->siglock);
 414        return 0;
 415}
 416
 417EXPORT_SYMBOL(disallow_signal);
 418
 419/*
 420 *      Put all the gunge required to become a kernel thread without
 421 *      attached user resources in one place where it belongs.
 422 */
 423
 424void daemonize(const char *name, ...)
 425{
 426        va_list args;
 427        sigset_t blocked;
 428
 429        va_start(args, name);
 430        vsnprintf(current->comm, sizeof(current->comm), name, args);
 431        va_end(args);
 432
 433        /*
 434         * If we were started as result of loading a module, close all of the
 435         * user space pages.  We don't need them, and if we didn't close them
 436         * they would be locked into memory.
 437         */
 438        exit_mm(current);
 439        /*
 440         * We don't want to get frozen, in case system-wide hibernation
 441         * or suspend transition begins right now.
 442         */
 443        current->flags |= (PF_NOFREEZE | PF_KTHREAD);
 444
 445        if (current->nsproxy != &init_nsproxy) {
 446                get_nsproxy(&init_nsproxy);
 447                switch_task_namespaces(current, &init_nsproxy);
 448        }
 449        set_special_pids(&init_struct_pid);
 450        proc_clear_tty(current);
 451
 452        /* Block and flush all signals */
 453        sigfillset(&blocked);
 454        sigprocmask(SIG_BLOCK, &blocked, NULL);
 455        flush_signals(current);
 456
 457        /* Become as one with the init task */
 458
 459        daemonize_fs_struct();
 460        exit_files(current);
 461        current->files = init_task.files;
 462        atomic_inc(&current->files->count);
 463
 464        reparent_to_kthreadd();
 465}
 466
 467EXPORT_SYMBOL(daemonize);
 468
 469static void close_files(struct files_struct * files)
 470{
 471        int i, j;
 472        struct fdtable *fdt;
 473
 474        j = 0;
 475
 476        /*
 477         * It is safe to dereference the fd table without RCU or
 478         * ->file_lock because this is the last reference to the
 479         * files structure.  But use RCU to shut RCU-lockdep up.
 480         */
 481        rcu_read_lock();
 482        fdt = files_fdtable(files);
 483        rcu_read_unlock();
 484        for (;;) {
 485                unsigned long set;
 486                i = j * BITS_PER_LONG;
 487                if (i >= fdt->max_fds)
 488                        break;
 489                set = fdt->open_fds[j++];
 490                while (set) {
 491                        if (set & 1) {
 492                                struct file * file = xchg(&fdt->fd[i], NULL);
 493                                if (file) {
 494                                        filp_close(file, files);
 495                                        cond_resched();
 496                                }
 497                        }
 498                        i++;
 499                        set >>= 1;
 500                }
 501        }
 502}
 503
 504struct files_struct *get_files_struct(struct task_struct *task)
 505{
 506        struct files_struct *files;
 507
 508        task_lock(task);
 509        files = task->files;
 510        if (files)
 511                atomic_inc(&files->count);
 512        task_unlock(task);
 513
 514        return files;
 515}
 516
 517void put_files_struct(struct files_struct *files)
 518{
 519        struct fdtable *fdt;
 520
 521        if (atomic_dec_and_test(&files->count)) {
 522                close_files(files);
 523                /*
 524                 * Free the fd and fdset arrays if we expanded them.
 525                 * If the fdtable was embedded, pass files for freeing
 526                 * at the end of the RCU grace period. Otherwise,
 527                 * you can free files immediately.
 528                 */
 529                rcu_read_lock();
 530                fdt = files_fdtable(files);
 531                if (fdt != &files->fdtab)
 532                        kmem_cache_free(files_cachep, files);
 533                free_fdtable(fdt);
 534                rcu_read_unlock();
 535        }
 536}
 537
 538void reset_files_struct(struct files_struct *files)
 539{
 540        struct task_struct *tsk = current;
 541        struct files_struct *old;
 542
 543        old = tsk->files;
 544        task_lock(tsk);
 545        tsk->files = files;
 546        task_unlock(tsk);
 547        put_files_struct(old);
 548}
 549
 550void exit_files(struct task_struct *tsk)
 551{
 552        struct files_struct * files = tsk->files;
 553
 554        if (files) {
 555                task_lock(tsk);
 556                tsk->files = NULL;
 557                task_unlock(tsk);
 558                put_files_struct(files);
 559        }
 560}
 561
 562#ifdef CONFIG_MM_OWNER
 563/*
 564 * A task is exiting.   If it owned this mm, find a new owner for the mm.
 565 */
 566void mm_update_next_owner(struct mm_struct *mm)
 567{
 568        struct task_struct *c, *g, *p = current;
 569
 570retry:
 571        /*
 572         * If the exiting or execing task is not the owner, it's
 573         * someone else's problem.
 574         */
 575        if (mm->owner != p)
 576                return;
 577        /*
 578         * The current owner is exiting/execing and there are no other
 579         * candidates.  Do not leave the mm pointing to a possibly
 580         * freed task structure.
 581         */
 582        if (atomic_read(&mm->mm_users) <= 1) {
 583                mm->owner = NULL;
 584                return;
 585        }
 586
 587        read_lock(&tasklist_lock);
 588        /*
 589         * Search in the children
 590         */
 591        list_for_each_entry(c, &p->children, sibling) {
 592                if (c->mm == mm)
 593                        goto assign_new_owner;
 594        }
 595
 596        /*
 597         * Search in the siblings
 598         */
 599        list_for_each_entry(c, &p->real_parent->children, sibling) {
 600                if (c->mm == mm)
 601                        goto assign_new_owner;
 602        }
 603
 604        /*
 605         * Search through everything else. We should not get
 606         * here often
 607         */
 608        do_each_thread(g, c) {
 609                if (c->mm == mm)
 610                        goto assign_new_owner;
 611        } while_each_thread(g, c);
 612
 613        read_unlock(&tasklist_lock);
 614        /*
 615         * We found no owner yet mm_users > 1: this implies that we are
 616         * most likely racing with swapoff (try_to_unuse()) or /proc or
 617         * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
 618         */
 619        mm->owner = NULL;
 620        return;
 621
 622assign_new_owner:
 623        BUG_ON(c == p);
 624        get_task_struct(c);
 625        /*
 626         * The task_lock protects c->mm from changing.
 627         * We always want mm->owner->mm == mm
 628         */
 629        task_lock(c);
 630        /*
 631         * Delay read_unlock() till we have the task_lock()
 632         * to ensure that c does not slip away underneath us
 633         */
 634        read_unlock(&tasklist_lock);
 635        if (c->mm != mm) {
 636                task_unlock(c);
 637                put_task_struct(c);
 638                goto retry;
 639        }
 640        mm->owner = c;
 641        task_unlock(c);
 642        put_task_struct(c);
 643}
 644#endif /* CONFIG_MM_OWNER */
 645
 646/*
 647 * Turn us into a lazy TLB process if we
 648 * aren't already..
 649 */
 650static void exit_mm(struct task_struct * tsk)
 651{
 652        struct mm_struct *mm = tsk->mm;
 653        struct core_state *core_state;
 654
 655        mm_release(tsk, mm);
 656        if (!mm)
 657                return;
 658        sync_mm_rss(mm);
 659        /*
 660         * Serialize with any possible pending coredump.
 661         * We must hold mmap_sem around checking core_state
 662         * and clearing tsk->mm.  The core-inducing thread
 663         * will increment ->nr_threads for each thread in the
 664         * group with ->mm != NULL.
 665         */
 666        down_read(&mm->mmap_sem);
 667        core_state = mm->core_state;
 668        if (core_state) {
 669                struct core_thread self;
 670                up_read(&mm->mmap_sem);
 671
 672                self.task = tsk;
 673                self.next = xchg(&core_state->dumper.next, &self);
 674                /*
 675                 * Implies mb(), the result of xchg() must be visible
 676                 * to core_state->dumper.
 677                 */
 678                if (atomic_dec_and_test(&core_state->nr_threads))
 679                        complete(&core_state->startup);
 680
 681                for (;;) {
 682                        set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 683                        if (!self.task) /* see coredump_finish() */
 684                                break;
 685                        schedule();
 686                }
 687                __set_task_state(tsk, TASK_RUNNING);
 688                down_read(&mm->mmap_sem);
 689        }
 690        atomic_inc(&mm->mm_count);
 691        BUG_ON(mm != tsk->active_mm);
 692        /* more a memory barrier than a real lock */
 693        task_lock(tsk);
 694        tsk->mm = NULL;
 695        up_read(&mm->mmap_sem);
 696        enter_lazy_tlb(mm, current);
 697        task_unlock(tsk);
 698        mm_update_next_owner(mm);
 699        mmput(mm);
 700}
 701
 702/*
 703 * When we die, we re-parent all our children, and try to:
 704 * 1. give them to another thread in our thread group, if such a member exists
 705 * 2. give it to the first ancestor process which prctl'd itself as a
 706 *    child_subreaper for its children (like a service manager)
 707 * 3. give it to the init process (PID 1) in our pid namespace
 708 */
 709static struct task_struct *find_new_reaper(struct task_struct *father)
 710        __releases(&tasklist_lock)
 711        __acquires(&tasklist_lock)
 712{
 713        struct pid_namespace *pid_ns = task_active_pid_ns(father);
 714        struct task_struct *thread;
 715
 716        thread = father;
 717        while_each_thread(father, thread) {
 718                if (thread->flags & PF_EXITING)
 719                        continue;
 720                if (unlikely(pid_ns->child_reaper == father))
 721                        pid_ns->child_reaper = thread;
 722                return thread;
 723        }
 724
 725        if (unlikely(pid_ns->child_reaper == father)) {
 726                write_unlock_irq(&tasklist_lock);
 727                if (unlikely(pid_ns == &init_pid_ns)) {
 728                        panic("Attempted to kill init! exitcode=0x%08x\n",
 729                                father->signal->group_exit_code ?:
 730                                        father->exit_code);
 731                }
 732
 733                zap_pid_ns_processes(pid_ns);
 734                write_lock_irq(&tasklist_lock);
 735        } else if (father->signal->has_child_subreaper) {
 736                struct task_struct *reaper;
 737
 738                /*
 739                 * Find the first ancestor marked as child_subreaper.
 740                 * Note that the code below checks same_thread_group(reaper,
 741                 * pid_ns->child_reaper).  This is what we need to DTRT in a
 742                 * PID namespace. However we still need the check above, see
 743                 * http://marc.info/?l=linux-kernel&m=131385460420380
 744                 */
 745                for (reaper = father->real_parent;
 746                     reaper != &init_task;
 747                     reaper = reaper->real_parent) {
 748                        if (same_thread_group(reaper, pid_ns->child_reaper))
 749                                break;
 750                        if (!reaper->signal->is_child_subreaper)
 751                                continue;
 752                        thread = reaper;
 753                        do {
 754                                if (!(thread->flags & PF_EXITING))
 755                                        return reaper;
 756                        } while_each_thread(reaper, thread);
 757                }
 758        }
 759
 760        return pid_ns->child_reaper;
 761}
 762
 763/*
 764* Any that need to be release_task'd are put on the @dead list.
 765 */
 766static void reparent_leader(struct task_struct *father, struct task_struct *p,
 767                                struct list_head *dead)
 768{
 769        list_move_tail(&p->sibling, &p->real_parent->children);
 770
 771        if (p->exit_state == EXIT_DEAD)
 772                return;
 773        /*
 774         * If this is a threaded reparent there is no need to
 775         * notify anyone anything has happened.
 776         */
 777        if (same_thread_group(p->real_parent, father))
 778                return;
 779
 780        /* We don't want people slaying init.  */
 781        p->exit_signal = SIGCHLD;
 782
 783        /* If it has exited notify the new parent about this child's death. */
 784        if (!p->ptrace &&
 785            p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
 786                if (do_notify_parent(p, p->exit_signal)) {
 787                        p->exit_state = EXIT_DEAD;
 788                        list_move_tail(&p->sibling, dead);
 789                }
 790        }
 791
 792        kill_orphaned_pgrp(p, father);
 793}
 794
 795static void forget_original_parent(struct task_struct *father)
 796{
 797        struct task_struct *p, *n, *reaper;
 798        LIST_HEAD(dead_children);
 799
 800        write_lock_irq(&tasklist_lock);
 801        /*
 802         * Note that exit_ptrace() and find_new_reaper() might
 803         * drop tasklist_lock and reacquire it.
 804         */
 805        exit_ptrace(father);
 806        reaper = find_new_reaper(father);
 807
 808        list_for_each_entry_safe(p, n, &father->children, sibling) {
 809                struct task_struct *t = p;
 810                do {
 811                        t->real_parent = reaper;
 812                        if (t->parent == father) {
 813                                BUG_ON(t->ptrace);
 814                                t->parent = t->real_parent;
 815                        }
 816                        if (t->pdeath_signal)
 817                                group_send_sig_info(t->pdeath_signal,
 818                                                    SEND_SIG_NOINFO, t);
 819                } while_each_thread(p, t);
 820                reparent_leader(father, p, &dead_children);
 821        }
 822        write_unlock_irq(&tasklist_lock);
 823
 824        BUG_ON(!list_empty(&father->children));
 825
 826        list_for_each_entry_safe(p, n, &dead_children, sibling) {
 827                list_del_init(&p->sibling);
 828                release_task(p);
 829        }
 830}
 831
 832/*
 833 * Send signals to all our closest relatives so that they know
 834 * to properly mourn us..
 835 */
 836static void exit_notify(struct task_struct *tsk, int group_dead)
 837{
 838        bool autoreap;
 839
 840        /*
 841         * This does two things:
 842         *
 843         * A.  Make init inherit all the child processes
 844         * B.  Check to see if any process groups have become orphaned
 845         *      as a result of our exiting, and if they have any stopped
 846         *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 847         */
 848        forget_original_parent(tsk);
 849        exit_task_namespaces(tsk);
 850
 851        write_lock_irq(&tasklist_lock);
 852        if (group_dead)
 853                kill_orphaned_pgrp(tsk->group_leader, NULL);
 854
 855        if (unlikely(tsk->ptrace)) {
 856                int sig = thread_group_leader(tsk) &&
 857                                thread_group_empty(tsk) &&
 858                                !ptrace_reparented(tsk) ?
 859                        tsk->exit_signal : SIGCHLD;
 860                autoreap = do_notify_parent(tsk, sig);
 861        } else if (thread_group_leader(tsk)) {
 862                autoreap = thread_group_empty(tsk) &&
 863                        do_notify_parent(tsk, tsk->exit_signal);
 864        } else {
 865                autoreap = true;
 866        }
 867
 868        tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
 869
 870        /* mt-exec, de_thread() is waiting for group leader */
 871        if (unlikely(tsk->signal->notify_count < 0))
 872                wake_up_process(tsk->signal->group_exit_task);
 873        write_unlock_irq(&tasklist_lock);
 874
 875        /* If the process is dead, release it - nobody will wait for it */
 876        if (autoreap)
 877                release_task(tsk);
 878}
 879
 880#ifdef CONFIG_DEBUG_STACK_USAGE
 881static void check_stack_usage(void)
 882{
 883        static DEFINE_SPINLOCK(low_water_lock);
 884        static int lowest_to_date = THREAD_SIZE;
 885        unsigned long free;
 886
 887        free = stack_not_used(current);
 888
 889        if (free >= lowest_to_date)
 890                return;
 891
 892        spin_lock(&low_water_lock);
 893        if (free < lowest_to_date) {
 894                printk(KERN_WARNING "%s (%d) used greatest stack depth: "
 895                                "%lu bytes left\n",
 896                                current->comm, task_pid_nr(current), free);
 897                lowest_to_date = free;
 898        }
 899        spin_unlock(&low_water_lock);
 900}
 901#else
 902static inline void check_stack_usage(void) {}
 903#endif
 904
 905void do_exit(long code)
 906{
 907        struct task_struct *tsk = current;
 908        int group_dead;
 909
 910        profile_task_exit(tsk);
 911
 912        WARN_ON(blk_needs_flush_plug(tsk));
 913
 914        if (unlikely(in_interrupt()))
 915                panic("Aiee, killing interrupt handler!");
 916        if (unlikely(!tsk->pid))
 917                panic("Attempted to kill the idle task!");
 918
 919        /*
 920         * If do_exit is called because this processes oopsed, it's possible
 921         * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
 922         * continuing. Amongst other possible reasons, this is to prevent
 923         * mm_release()->clear_child_tid() from writing to a user-controlled
 924         * kernel address.
 925         */
 926        set_fs(USER_DS);
 927
 928        ptrace_event(PTRACE_EVENT_EXIT, code);
 929
 930        validate_creds_for_do_exit(tsk);
 931
 932        /*
 933         * We're taking recursive faults here in do_exit. Safest is to just
 934         * leave this task alone and wait for reboot.
 935         */
 936        if (unlikely(tsk->flags & PF_EXITING)) {
 937                printk(KERN_ALERT
 938                        "Fixing recursive fault but reboot is needed!\n");
 939                /*
 940                 * We can do this unlocked here. The futex code uses
 941                 * this flag just to verify whether the pi state
 942                 * cleanup has been done or not. In the worst case it
 943                 * loops once more. We pretend that the cleanup was
 944                 * done as there is no way to return. Either the
 945                 * OWNER_DIED bit is set by now or we push the blocked
 946                 * task into the wait for ever nirwana as well.
 947                 */
 948                tsk->flags |= PF_EXITPIDONE;
 949                set_current_state(TASK_UNINTERRUPTIBLE);
 950                schedule();
 951        }
 952
 953        exit_signals(tsk);  /* sets PF_EXITING */
 954        /*
 955         * tsk->flags are checked in the futex code to protect against
 956         * an exiting task cleaning up the robust pi futexes, and in
 957         * task_work_add() to avoid the race with exit_task_work().
 958         */
 959        smp_mb();
 960        raw_spin_unlock_wait(&tsk->pi_lock);
 961
 962        exit_task_work(tsk);
 963
 964        if (unlikely(in_atomic()))
 965                printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 966                                current->comm, task_pid_nr(current),
 967                                preempt_count());
 968
 969        acct_update_integrals(tsk);
 970        /* sync mm's RSS info before statistics gathering */
 971        if (tsk->mm)
 972                sync_mm_rss(tsk->mm);
 973        group_dead = atomic_dec_and_test(&tsk->signal->live);
 974        if (group_dead) {
 975                hrtimer_cancel(&tsk->signal->real_timer);
 976                exit_itimers(tsk->signal);
 977                if (tsk->mm)
 978                        setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
 979        }
 980        acct_collect(code, group_dead);
 981        if (group_dead)
 982                tty_audit_exit();
 983        audit_free(tsk);
 984
 985        tsk->exit_code = code;
 986        taskstats_exit(tsk, group_dead);
 987
 988        exit_mm(tsk);
 989
 990        if (group_dead)
 991                acct_process();
 992        trace_sched_process_exit(tsk);
 993
 994        exit_sem(tsk);
 995        exit_shm(tsk);
 996        exit_files(tsk);
 997        exit_fs(tsk);
 998        check_stack_usage();
 999        exit_thread();
1000
1001        /*
1002         * Flush inherited counters to the parent - before the parent
1003         * gets woken up by child-exit notifications.
1004         *
1005         * because of cgroup mode, must be called before cgroup_exit()
1006         */
1007        perf_event_exit_task(tsk);
1008
1009        cgroup_exit(tsk, 1);
1010
1011        if (group_dead)
1012                disassociate_ctty(1);
1013
1014        module_put(task_thread_info(tsk)->exec_domain->module);
1015
1016        proc_exit_connector(tsk);
1017
1018        /*
1019         * FIXME: do that only when needed, using sched_exit tracepoint
1020         */
1021        ptrace_put_breakpoints(tsk);
1022
1023        exit_notify(tsk, group_dead);
1024#ifdef CONFIG_NUMA
1025        task_lock(tsk);
1026        mpol_put(tsk->mempolicy);
1027        tsk->mempolicy = NULL;
1028        task_unlock(tsk);
1029#endif
1030#ifdef CONFIG_FUTEX
1031        if (unlikely(current->pi_state_cache))
1032                kfree(current->pi_state_cache);
1033#endif
1034        /*
1035         * Make sure we are holding no locks:
1036         */
1037        debug_check_no_locks_held(tsk);
1038        /*
1039         * We can do this unlocked here. The futex code uses this flag
1040         * just to verify whether the pi state cleanup has been done
1041         * or not. In the worst case it loops once more.
1042         */
1043        tsk->flags |= PF_EXITPIDONE;
1044
1045        if (tsk->io_context)
1046                exit_io_context(tsk);
1047
1048        if (tsk->splice_pipe)
1049                __free_pipe_info(tsk->splice_pipe);
1050
1051        validate_creds_for_do_exit(tsk);
1052
1053        preempt_disable();
1054        if (tsk->nr_dirtied)
1055                __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
1056        exit_rcu();
1057
1058        /*
1059         * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
1060         * when the following two conditions become true.
1061         *   - There is race condition of mmap_sem (It is acquired by
1062         *     exit_mm()), and
1063         *   - SMI occurs before setting TASK_RUNINNG.
1064         *     (or hypervisor of virtual machine switches to other guest)
1065         *  As a result, we may become TASK_RUNNING after becoming TASK_DEAD
1066         *
1067         * To avoid it, we have to wait for releasing tsk->pi_lock which
1068         * is held by try_to_wake_up()
1069         */
1070        smp_mb();
1071        raw_spin_unlock_wait(&tsk->pi_lock);
1072
1073        /* causes final put_task_struct in finish_task_switch(). */
1074        tsk->state = TASK_DEAD;
1075        tsk->flags |= PF_NOFREEZE;      /* tell freezer to ignore us */
1076        schedule();
1077        BUG();
1078        /* Avoid "noreturn function does return".  */
1079        for (;;)
1080                cpu_relax();    /* For when BUG is null */
1081}
1082
1083EXPORT_SYMBOL_GPL(do_exit);
1084
1085void complete_and_exit(struct completion *comp, long code)
1086{
1087        if (comp)
1088                complete(comp);
1089
1090        do_exit(code);
1091}
1092
1093EXPORT_SYMBOL(complete_and_exit);
1094
1095SYSCALL_DEFINE1(exit, int, error_code)
1096{
1097        do_exit((error_code&0xff)<<8);
1098}
1099
1100/*
1101 * Take down every thread in the group.  This is called by fatal signals
1102 * as well as by sys_exit_group (below).
1103 */
1104void
1105do_group_exit(int exit_code)
1106{
1107        struct signal_struct *sig = current->signal;
1108
1109        BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1110
1111        if (signal_group_exit(sig))
1112                exit_code = sig->group_exit_code;
1113        else if (!thread_group_empty(current)) {
1114                struct sighand_struct *const sighand = current->sighand;
1115                spin_lock_irq(&sighand->siglock);
1116                if (signal_group_exit(sig))
1117                        /* Another thread got here before we took the lock.  */
1118                        exit_code = sig->group_exit_code;
1119                else {
1120                        sig->group_exit_code = exit_code;
1121                        sig->flags = SIGNAL_GROUP_EXIT;
1122                        zap_other_threads(current);
1123                }
1124                spin_unlock_irq(&sighand->siglock);
1125        }
1126
1127        do_exit(exit_code);
1128        /* NOTREACHED */
1129}
1130
1131/*
1132 * this kills every thread in the thread group. Note that any externally
1133 * wait4()-ing process will get the correct exit code - even if this
1134 * thread is not the thread group leader.
1135 */
1136SYSCALL_DEFINE1(exit_group, int, error_code)
1137{
1138        do_group_exit((error_code & 0xff) << 8);
1139        /* NOTREACHED */
1140        return 0;
1141}
1142
1143struct wait_opts {
1144        enum pid_type           wo_type;
1145        int                     wo_flags;
1146        struct pid              *wo_pid;
1147
1148        struct siginfo __user   *wo_info;
1149        int __user              *wo_stat;
1150        struct rusage __user    *wo_rusage;
1151
1152        wait_queue_t            child_wait;
1153        int                     notask_error;
1154};
1155
1156static inline
1157struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1158{
1159        if (type != PIDTYPE_PID)
1160                task = task->group_leader;
1161        return task->pids[type].pid;
1162}
1163
1164static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1165{
1166        return  wo->wo_type == PIDTYPE_MAX ||
1167                task_pid_type(p, wo->wo_type) == wo->wo_pid;
1168}
1169
1170static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1171{
1172        if (!eligible_pid(wo, p))
1173                return 0;
1174        /* Wait for all children (clone and not) if __WALL is set;
1175         * otherwise, wait for clone children *only* if __WCLONE is
1176         * set; otherwise, wait for non-clone children *only*.  (Note:
1177         * A "clone" child here is one that reports to its parent
1178         * using a signal other than SIGCHLD.) */
1179        if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1180            && !(wo->wo_flags & __WALL))
1181                return 0;
1182
1183        return 1;
1184}
1185
1186static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1187                                pid_t pid, uid_t uid, int why, int status)
1188{
1189        struct siginfo __user *infop;
1190        int retval = wo->wo_rusage
1191                ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1192
1193        put_task_struct(p);
1194        infop = wo->wo_info;
1195        if (infop) {
1196                if (!retval)
1197                        retval = put_user(SIGCHLD, &infop->si_signo);
1198                if (!retval)
1199                        retval = put_user(0, &infop->si_errno);
1200                if (!retval)
1201                        retval = put_user((short)why, &infop->si_code);
1202                if (!retval)
1203                        retval = put_user(pid, &infop->si_pid);
1204                if (!retval)
1205                        retval = put_user(uid, &infop->si_uid);
1206                if (!retval)
1207                        retval = put_user(status, &infop->si_status);
1208        }
1209        if (!retval)
1210                retval = pid;
1211        return retval;
1212}
1213
1214/*
1215 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1216 * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1217 * the lock and this task is uninteresting.  If we return nonzero, we have
1218 * released the lock and the system call should return.
1219 */
1220static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1221{
1222        unsigned long state;
1223        int retval, status, traced;
1224        pid_t pid = task_pid_vnr(p);
1225        uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1226        struct siginfo __user *infop;
1227
1228        if (!likely(wo->wo_flags & WEXITED))
1229                return 0;
1230
1231        if (unlikely(wo->wo_flags & WNOWAIT)) {
1232                int exit_code = p->exit_code;
1233                int why;
1234
1235                get_task_struct(p);
1236                read_unlock(&tasklist_lock);
1237                if ((exit_code & 0x7f) == 0) {
1238                        why = CLD_EXITED;
1239                        status = exit_code >> 8;
1240                } else {
1241                        why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1242                        status = exit_code & 0x7f;
1243                }
1244                return wait_noreap_copyout(wo, p, pid, uid, why, status);
1245        }
1246
1247        /*
1248         * Try to move the task's state to DEAD
1249         * only one thread is allowed to do this:
1250         */
1251        state = xchg(&p->exit_state, EXIT_DEAD);
1252        if (state != EXIT_ZOMBIE) {
1253                BUG_ON(state != EXIT_DEAD);
1254                return 0;
1255        }
1256
1257        traced = ptrace_reparented(p);
1258        /*
1259         * It can be ptraced but not reparented, check
1260         * thread_group_leader() to filter out sub-threads.
1261         */
1262        if (likely(!traced) && thread_group_leader(p)) {
1263                struct signal_struct *psig;
1264                struct signal_struct *sig;
1265                unsigned long maxrss;
1266                cputime_t tgutime, tgstime;
1267
1268                /*
1269                 * The resource counters for the group leader are in its
1270                 * own task_struct.  Those for dead threads in the group
1271                 * are in its signal_struct, as are those for the child
1272                 * processes it has previously reaped.  All these
1273                 * accumulate in the parent's signal_struct c* fields.
1274                 *
1275                 * We don't bother to take a lock here to protect these
1276                 * p->signal fields, because they are only touched by
1277                 * __exit_signal, which runs with tasklist_lock
1278                 * write-locked anyway, and so is excluded here.  We do
1279                 * need to protect the access to parent->signal fields,
1280                 * as other threads in the parent group can be right
1281                 * here reaping other children at the same time.
1282                 *
1283                 * We use thread_group_times() to get times for the thread
1284                 * group, which consolidates times for all threads in the
1285                 * group including the group leader.
1286                 */
1287                thread_group_times(p, &tgutime, &tgstime);
1288                spin_lock_irq(&p->real_parent->sighand->siglock);
1289                psig = p->real_parent->signal;
1290                sig = p->signal;
1291                psig->cutime += tgutime + sig->cutime;
1292                psig->cstime += tgstime + sig->cstime;
1293                psig->cgtime += p->gtime + sig->gtime + sig->cgtime;
1294                psig->cmin_flt +=
1295                        p->min_flt + sig->min_flt + sig->cmin_flt;
1296                psig->cmaj_flt +=
1297                        p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1298                psig->cnvcsw +=
1299                        p->nvcsw + sig->nvcsw + sig->cnvcsw;
1300                psig->cnivcsw +=
1301                        p->nivcsw + sig->nivcsw + sig->cnivcsw;
1302                psig->cinblock +=
1303                        task_io_get_inblock(p) +
1304                        sig->inblock + sig->cinblock;
1305                psig->coublock +=
1306                        task_io_get_oublock(p) +
1307                        sig->oublock + sig->coublock;
1308                maxrss = max(sig->maxrss, sig->cmaxrss);
1309                if (psig->cmaxrss < maxrss)
1310                        psig->cmaxrss = maxrss;
1311                task_io_accounting_add(&psig->ioac, &p->ioac);
1312                task_io_accounting_add(&psig->ioac, &sig->ioac);
1313                spin_unlock_irq(&p->real_parent->sighand->siglock);
1314        }
1315
1316        /*
1317         * Now we are sure this task is interesting, and no other
1318         * thread can reap it because we set its state to EXIT_DEAD.
1319         */
1320        read_unlock(&tasklist_lock);
1321
1322        retval = wo->wo_rusage
1323                ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1324        status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1325                ? p->signal->group_exit_code : p->exit_code;
1326        if (!retval && wo->wo_stat)
1327                retval = put_user(status, wo->wo_stat);
1328
1329        infop = wo->wo_info;
1330        if (!retval && infop)
1331                retval = put_user(SIGCHLD, &infop->si_signo);
1332        if (!retval && infop)
1333                retval = put_user(0, &infop->si_errno);
1334        if (!retval && infop) {
1335                int why;
1336
1337                if ((status & 0x7f) == 0) {
1338                        why = CLD_EXITED;
1339                        status >>= 8;
1340                } else {
1341                        why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1342                        status &= 0x7f;
1343                }
1344                retval = put_user((short)why, &infop->si_code);
1345                if (!retval)
1346                        retval = put_user(status, &infop->si_status);
1347        }
1348        if (!retval && infop)
1349                retval = put_user(pid, &infop->si_pid);
1350        if (!retval && infop)
1351                retval = put_user(uid, &infop->si_uid);
1352        if (!retval)
1353                retval = pid;
1354
1355        if (traced) {
1356                write_lock_irq(&tasklist_lock);
1357                /* We dropped tasklist, ptracer could die and untrace */
1358                ptrace_unlink(p);
1359                /*
1360                 * If this is not a sub-thread, notify the parent.
1361                 * If parent wants a zombie, don't release it now.
1362                 */
1363                if (thread_group_leader(p) &&
1364                    !do_notify_parent(p, p->exit_signal)) {
1365                        p->exit_state = EXIT_ZOMBIE;
1366                        p = NULL;
1367                }
1368                write_unlock_irq(&tasklist_lock);
1369        }
1370        if (p != NULL)
1371                release_task(p);
1372
1373        return retval;
1374}
1375
1376static int *task_stopped_code(struct task_struct *p, bool ptrace)
1377{
1378        if (ptrace) {
1379                if (task_is_stopped_or_traced(p) &&
1380                    !(p->jobctl & JOBCTL_LISTENING))
1381                        return &p->exit_code;
1382        } else {
1383                if (p->signal->flags & SIGNAL_STOP_STOPPED)
1384                        return &p->signal->group_exit_code;
1385        }
1386        return NULL;
1387}
1388
1389/**
1390 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1391 * @wo: wait options
1392 * @ptrace: is the wait for ptrace
1393 * @p: task to wait for
1394 *
1395 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1396 *
1397 * CONTEXT:
1398 * read_lock(&tasklist_lock), which is released if return value is
1399 * non-zero.  Also, grabs and releases @p->sighand->siglock.
1400 *
1401 * RETURNS:
1402 * 0 if wait condition didn't exist and search for other wait conditions
1403 * should continue.  Non-zero return, -errno on failure and @p's pid on
1404 * success, implies that tasklist_lock is released and wait condition
1405 * search should terminate.
1406 */
1407static int wait_task_stopped(struct wait_opts *wo,
1408                                int ptrace, struct task_struct *p)
1409{
1410        struct siginfo __user *infop;
1411        int retval, exit_code, *p_code, why;
1412        uid_t uid = 0; /* unneeded, required by compiler */
1413        pid_t pid;
1414
1415        /*
1416         * Traditionally we see ptrace'd stopped tasks regardless of options.
1417         */
1418        if (!ptrace && !(wo->wo_flags & WUNTRACED))
1419                return 0;
1420
1421        if (!task_stopped_code(p, ptrace))
1422                return 0;
1423
1424        exit_code = 0;
1425        spin_lock_irq(&p->sighand->siglock);
1426
1427        p_code = task_stopped_code(p, ptrace);
1428        if (unlikely(!p_code))
1429                goto unlock_sig;
1430
1431        exit_code = *p_code;
1432        if (!exit_code)
1433                goto unlock_sig;
1434
1435        if (!unlikely(wo->wo_flags & WNOWAIT))
1436                *p_code = 0;
1437
1438        uid = from_kuid_munged(current_user_ns(), task_uid(p));
1439unlock_sig:
1440        spin_unlock_irq(&p->sighand->siglock);
1441        if (!exit_code)
1442                return 0;
1443
1444        /*
1445         * Now we are pretty sure this task is interesting.
1446         * Make sure it doesn't get reaped out from under us while we
1447         * give up the lock and then examine it below.  We don't want to
1448         * keep holding onto the tasklist_lock while we call getrusage and
1449         * possibly take page faults for user memory.
1450         */
1451        get_task_struct(p);
1452        pid = task_pid_vnr(p);
1453        why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1454        read_unlock(&tasklist_lock);
1455
1456        if (unlikely(wo->wo_flags & WNOWAIT))
1457                return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1458
1459        retval = wo->wo_rusage
1460                ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1461        if (!retval && wo->wo_stat)
1462                retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1463
1464        infop = wo->wo_info;
1465        if (!retval && infop)
1466                retval = put_user(SIGCHLD, &infop->si_signo);
1467        if (!retval && infop)
1468                retval = put_user(0, &infop->si_errno);
1469        if (!retval && infop)
1470                retval = put_user((short)why, &infop->si_code);
1471        if (!retval && infop)
1472                retval = put_user(exit_code, &infop->si_status);
1473        if (!retval && infop)
1474                retval = put_user(pid, &infop->si_pid);
1475        if (!retval && infop)
1476                retval = put_user(uid, &infop->si_uid);
1477        if (!retval)
1478                retval = pid;
1479        put_task_struct(p);
1480
1481        BUG_ON(!retval);
1482        return retval;
1483}
1484
1485/*
1486 * Handle do_wait work for one task in a live, non-stopped state.
1487 * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1488 * the lock and this task is uninteresting.  If we return nonzero, we have
1489 * released the lock and the system call should return.
1490 */
1491static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1492{
1493        int retval;
1494        pid_t pid;
1495        uid_t uid;
1496
1497        if (!unlikely(wo->wo_flags & WCONTINUED))
1498                return 0;
1499
1500        if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1501                return 0;
1502
1503        spin_lock_irq(&p->sighand->siglock);
1504        /* Re-check with the lock held.  */
1505        if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1506                spin_unlock_irq(&p->sighand->siglock);
1507                return 0;
1508        }
1509        if (!unlikely(wo->wo_flags & WNOWAIT))
1510                p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1511        uid = from_kuid_munged(current_user_ns(), task_uid(p));
1512        spin_unlock_irq(&p->sighand->siglock);
1513
1514        pid = task_pid_vnr(p);
1515        get_task_struct(p);
1516        read_unlock(&tasklist_lock);
1517
1518        if (!wo->wo_info) {
1519                retval = wo->wo_rusage
1520                        ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1521                put_task_struct(p);
1522                if (!retval && wo->wo_stat)
1523                        retval = put_user(0xffff, wo->wo_stat);
1524                if (!retval)
1525                        retval = pid;
1526        } else {
1527                retval = wait_noreap_copyout(wo, p, pid, uid,
1528                                             CLD_CONTINUED, SIGCONT);
1529                BUG_ON(retval == 0);
1530        }
1531
1532        return retval;
1533}
1534
1535/*
1536 * Consider @p for a wait by @parent.
1537 *
1538 * -ECHILD should be in ->notask_error before the first call.
1539 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1540 * Returns zero if the search for a child should continue;
1541 * then ->notask_error is 0 if @p is an eligible child,
1542 * or another error from security_task_wait(), or still -ECHILD.
1543 */
1544static int wait_consider_task(struct wait_opts *wo, int ptrace,
1545                                struct task_struct *p)
1546{
1547        int ret = eligible_child(wo, p);
1548        if (!ret)
1549                return ret;
1550
1551        ret = security_task_wait(p);
1552        if (unlikely(ret < 0)) {
1553                /*
1554                 * If we have not yet seen any eligible child,
1555                 * then let this error code replace -ECHILD.
1556                 * A permission error will give the user a clue
1557                 * to look for security policy problems, rather
1558                 * than for mysterious wait bugs.
1559                 */
1560                if (wo->notask_error)
1561                        wo->notask_error = ret;
1562                return 0;
1563        }
1564
1565        /* dead body doesn't have much to contribute */
1566        if (unlikely(p->exit_state == EXIT_DEAD)) {
1567                /*
1568                 * But do not ignore this task until the tracer does
1569                 * wait_task_zombie()->do_notify_parent().
1570                 */
1571                if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
1572                        wo->notask_error = 0;
1573                return 0;
1574        }
1575
1576        /* slay zombie? */
1577        if (p->exit_state == EXIT_ZOMBIE) {
1578                /*
1579                 * A zombie ptracee is only visible to its ptracer.
1580                 * Notification and reaping will be cascaded to the real
1581                 * parent when the ptracer detaches.
1582                 */
1583                if (likely(!ptrace) && unlikely(p->ptrace)) {
1584                        /* it will become visible, clear notask_error */
1585                        wo->notask_error = 0;
1586                        return 0;
1587                }
1588
1589                /* we don't reap group leaders with subthreads */
1590                if (!delay_group_leader(p))
1591                        return wait_task_zombie(wo, p);
1592
1593                /*
1594                 * Allow access to stopped/continued state via zombie by
1595                 * falling through.  Clearing of notask_error is complex.
1596                 *
1597                 * When !@ptrace:
1598                 *
1599                 * If WEXITED is set, notask_error should naturally be
1600                 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1601                 * so, if there are live subthreads, there are events to
1602                 * wait for.  If all subthreads are dead, it's still safe
1603                 * to clear - this function will be called again in finite
1604                 * amount time once all the subthreads are released and
1605                 * will then return without clearing.
1606                 *
1607                 * When @ptrace:
1608                 *
1609                 * Stopped state is per-task and thus can't change once the
1610                 * target task dies.  Only continued and exited can happen.
1611                 * Clear notask_error if WCONTINUED | WEXITED.
1612                 */
1613                if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1614                        wo->notask_error = 0;
1615        } else {
1616                /*
1617                 * If @p is ptraced by a task in its real parent's group,
1618                 * hide group stop/continued state when looking at @p as
1619                 * the real parent; otherwise, a single stop can be
1620                 * reported twice as group and ptrace stops.
1621                 *
1622                 * If a ptracer wants to distinguish the two events for its
1623                 * own children, it should create a separate process which
1624                 * takes the role of real parent.
1625                 */
1626                if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1627                        return 0;
1628
1629                /*
1630                 * @p is alive and it's gonna stop, continue or exit, so
1631                 * there always is something to wait for.
1632                 */
1633                wo->notask_error = 0;
1634        }
1635
1636        /*
1637         * Wait for stopped.  Depending on @ptrace, different stopped state
1638         * is used and the two don't interact with each other.
1639         */
1640        ret = wait_task_stopped(wo, ptrace, p);
1641        if (ret)
1642                return ret;
1643
1644        /*
1645         * Wait for continued.  There's only one continued state and the
1646         * ptracer can consume it which can confuse the real parent.  Don't
1647         * use WCONTINUED from ptracer.  You don't need or want it.
1648         */
1649        return wait_task_continued(wo, p);
1650}
1651
1652/*
1653 * Do the work of do_wait() for one thread in the group, @tsk.
1654 *
1655 * -ECHILD should be in ->notask_error before the first call.
1656 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1657 * Returns zero if the search for a child should continue; then
1658 * ->notask_error is 0 if there were any eligible children,
1659 * or another error from security_task_wait(), or still -ECHILD.
1660 */
1661static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1662{
1663        struct task_struct *p;
1664
1665        list_for_each_entry(p, &tsk->children, sibling) {
1666                int ret = wait_consider_task(wo, 0, p);
1667                if (ret)
1668                        return ret;
1669        }
1670
1671        return 0;
1672}
1673
1674static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1675{
1676        struct task_struct *p;
1677
1678        list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1679                int ret = wait_consider_task(wo, 1, p);
1680                if (ret)
1681                        return ret;
1682        }
1683
1684        return 0;
1685}
1686
1687static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1688                                int sync, void *key)
1689{
1690        struct wait_opts *wo = container_of(wait, struct wait_opts,
1691                                                child_wait);
1692        struct task_struct *p = key;
1693
1694        if (!eligible_pid(wo, p))
1695                return 0;
1696
1697        if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1698                return 0;
1699
1700        return default_wake_function(wait, mode, sync, key);
1701}
1702
1703void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1704{
1705        __wake_up_sync_key(&parent->signal->wait_chldexit,
1706                                TASK_INTERRUPTIBLE, 1, p);
1707}
1708
1709static long do_wait(struct wait_opts *wo)
1710{
1711        struct task_struct *tsk;
1712        int retval;
1713
1714        trace_sched_process_wait(wo->wo_pid);
1715
1716        init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1717        wo->child_wait.private = current;
1718        add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1719repeat:
1720        /*
1721         * If there is nothing that can match our critiera just get out.
1722         * We will clear ->notask_error to zero if we see any child that
1723         * might later match our criteria, even if we are not able to reap
1724         * it yet.
1725         */
1726        wo->notask_error = -ECHILD;
1727        if ((wo->wo_type < PIDTYPE_MAX) &&
1728           (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1729                goto notask;
1730
1731        set_current_state(TASK_INTERRUPTIBLE);
1732        read_lock(&tasklist_lock);
1733        tsk = current;
1734        do {
1735                retval = do_wait_thread(wo, tsk);
1736                if (retval)
1737                        goto end;
1738
1739                retval = ptrace_do_wait(wo, tsk);
1740                if (retval)
1741                        goto end;
1742
1743                if (wo->wo_flags & __WNOTHREAD)
1744                        break;
1745        } while_each_thread(current, tsk);
1746        read_unlock(&tasklist_lock);
1747
1748notask:
1749        retval = wo->notask_error;
1750        if (!retval && !(wo->wo_flags & WNOHANG)) {
1751                retval = -ERESTARTSYS;
1752                if (!signal_pending(current)) {
1753                        schedule();
1754                        goto repeat;
1755                }
1756        }
1757end:
1758        __set_current_state(TASK_RUNNING);
1759        remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1760        return retval;
1761}
1762
1763SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1764                infop, int, options, struct rusage __user *, ru)
1765{
1766        struct wait_opts wo;
1767        struct pid *pid = NULL;
1768        enum pid_type type;
1769        long ret;
1770
1771        if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1772                return -EINVAL;
1773        if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1774                return -EINVAL;
1775
1776        switch (which) {
1777        case P_ALL:
1778                type = PIDTYPE_MAX;
1779                break;
1780        case P_PID:
1781                type = PIDTYPE_PID;
1782                if (upid <= 0)
1783                        return -EINVAL;
1784                break;
1785        case P_PGID:
1786                type = PIDTYPE_PGID;
1787                if (upid <= 0)
1788                        return -EINVAL;
1789                break;
1790        default:
1791                return -EINVAL;
1792        }
1793
1794        if (type < PIDTYPE_MAX)
1795                pid = find_get_pid(upid);
1796
1797        wo.wo_type      = type;
1798        wo.wo_pid       = pid;
1799        wo.wo_flags     = options;
1800        wo.wo_info      = infop;
1801        wo.wo_stat      = NULL;
1802        wo.wo_rusage    = ru;
1803        ret = do_wait(&wo);
1804
1805        if (ret > 0) {
1806                ret = 0;
1807        } else if (infop) {
1808                /*
1809                 * For a WNOHANG return, clear out all the fields
1810                 * we would set so the user can easily tell the
1811                 * difference.
1812                 */
1813                if (!ret)
1814                        ret = put_user(0, &infop->si_signo);
1815                if (!ret)
1816                        ret = put_user(0, &infop->si_errno);
1817                if (!ret)
1818                        ret = put_user(0, &infop->si_code);
1819                if (!ret)
1820                        ret = put_user(0, &infop->si_pid);
1821                if (!ret)
1822                        ret = put_user(0, &infop->si_uid);
1823                if (!ret)
1824                        ret = put_user(0, &infop->si_status);
1825        }
1826
1827        put_pid(pid);
1828
1829        /* avoid REGPARM breakage on x86: */
1830        asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1831        return ret;
1832}
1833
1834SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1835                int, options, struct rusage __user *, ru)
1836{
1837        struct wait_opts wo;
1838        struct pid *pid = NULL;
1839        enum pid_type type;
1840        long ret;
1841
1842        if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1843                        __WNOTHREAD|__WCLONE|__WALL))
1844                return -EINVAL;
1845
1846        if (upid == -1)
1847                type = PIDTYPE_MAX;
1848        else if (upid < 0) {
1849                type = PIDTYPE_PGID;
1850                pid = find_get_pid(-upid);
1851        } else if (upid == 0) {
1852                type = PIDTYPE_PGID;
1853                pid = get_task_pid(current, PIDTYPE_PGID);
1854        } else /* upid > 0 */ {
1855                type = PIDTYPE_PID;
1856                pid = find_get_pid(upid);
1857        }
1858
1859        wo.wo_type      = type;
1860        wo.wo_pid       = pid;
1861        wo.wo_flags     = options | WEXITED;
1862        wo.wo_info      = NULL;
1863        wo.wo_stat      = stat_addr;
1864        wo.wo_rusage    = ru;
1865        ret = do_wait(&wo);
1866        put_pid(pid);
1867
1868        /* avoid REGPARM breakage on x86: */
1869        asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1870        return ret;
1871}
1872
1873#ifdef __ARCH_WANT_SYS_WAITPID
1874
1875/*
1876 * sys_waitpid() remains for compatibility. waitpid() should be
1877 * implemented by calling sys_wait4() from libc.a.
1878 */
1879SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1880{
1881        return sys_wait4(pid, stat_addr, options, NULL);
1882}
1883
1884#endif
1885
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.