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