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