linux/kernel/exit.c
<<
>>
Prefs
   1/*
   2 *  linux/kernel/exit.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 */
   6
   7#include <linux/mm.h>
   8#include <linux/slab.h>
   9#include <linux/interrupt.h>
  10#include <linux/module.h>
  11#include <linux/capability.h>
  12#include <linux/completion.h>
  13#include <linux/personality.h>
  14#include <linux/tty.h>
  15#include <linux/iocontext.h>
  16#include <linux/key.h>
  17#include <linux/security.h>
  18#include <linux/cpu.h>
  19#include <linux/acct.h>
  20#include <linux/tsacct_kern.h>
  21#include <linux/file.h>
  22#include <linux/fdtable.h>
  23#include <linux/binfmts.h>
  24#include <linux/nsproxy.h>
  25#include <linux/pid_namespace.h>
  26#include <linux/ptrace.h>
  27#include <linux/profile.h>
  28#include <linux/mount.h>
  29#include <linux/proc_fs.h>
  30#include <linux/kthread.h>
  31#include <linux/mempolicy.h>
  32#include <linux/taskstats_kern.h>
  33#include <linux/delayacct.h>
  34#include <linux/freezer.h>
  35#include <linux/cgroup.h>
  36#include <linux/syscalls.h>
  37#include <linux/signal.h>
  38#include <linux/posix-timers.h>
  39#include <linux/cn_proc.h>
  40#include <linux/mutex.h>
  41#include <linux/futex.h>
  42#include <linux/pipe_fs_i.h>
  43#include <linux/audit.h> /* for audit_free() */
  44#include <linux/resource.h>
  45#include <linux/blkdev.h>
  46#include <linux/task_io_accounting_ops.h>
  47#include <linux/tracehook.h>
  48#include <linux/fs_struct.h>
  49#include <linux/init_task.h>
  50#include <linux/perf_event.h>
  51#include <trace/events/sched.h>
  52#include <linux/hw_breakpoint.h>
  53#include <linux/oom.h>
  54#include <linux/writeback.h>
  55#include <linux/shm.h>
  56
  57#include <asm/uaccess.h>
  58#include <asm/unistd.h>
  59#include <asm/pgtable.h>
  60#include <asm/mmu_context.h>
  61
  62static void exit_mm(struct task_struct * tsk);
  63
  64static void __unhash_process(struct task_struct *p, bool group_dead)
  65{
  66        nr_threads--;
  67        detach_pid(p, PIDTYPE_PID);
  68        if (group_dead) {
  69                detach_pid(p, PIDTYPE_PGID);
  70                detach_pid(p, PIDTYPE_SID);
  71
  72                list_del_rcu(&p->tasks);
  73                list_del_init(&p->sibling);
  74                __this_cpu_dec(process_counts);
  75                /*
  76                 * If we are the last child process in a pid namespace to be
  77                 * reaped, notify the reaper sleeping zap_pid_ns_processes().
  78                 */
  79                if (IS_ENABLED(CONFIG_PID_NS)) {
  80                        struct task_struct *parent = p->real_parent;
  81
  82                        if ((task_active_pid_ns(parent)->child_reaper == parent) &&
  83                            list_empty(&parent->children) &&
  84                            (parent->flags & PF_EXITING))
  85                                wake_up_process(parent);
  86                }
  87        }
  88        list_del_rcu(&p->thread_group);
  89}
  90
  91/*
  92 * This function expects the tasklist_lock write-locked.
  93 */
  94static void __exit_signal(struct task_struct *tsk)
  95{
  96        struct signal_struct *sig = tsk->signal;
  97        bool group_dead = thread_group_leader(tsk);
  98        struct sighand_struct *sighand;
  99        struct tty_struct *uninitialized_var(tty);
 100
 101        sighand = rcu_dereference_check(tsk->sighand,
 102                                        lockdep_tasklist_lock_is_held());
 103        spin_lock(&sighand->siglock);
 104
 105        posix_cpu_timers_exit(tsk);
 106        if (group_dead) {
 107                posix_cpu_timers_exit_group(tsk);
 108                tty = sig->tty;
 109                sig->tty = NULL;
 110        } else {
 111                /*
 112                 * This can only happen if the caller is de_thread().
 113                 * FIXME: this is the temporary hack, we should teach
 114                 * posix-cpu-timers to handle this case correctly.
 115                 */
 116                if (unlikely(has_group_leader_pid(tsk)))
 117                        posix_cpu_timers_exit_group(tsk);
 118
 119                /*
 120                 * If there is any task waiting for the group exit
 121                 * then notify it:
 122                 */
 123                if (sig->notify_count > 0 && !--sig->notify_count)
 124                        wake_up_process(sig->group_exit_task);
 125
 126                if (tsk == sig->curr_target)
 127                        sig->curr_target = next_thread(tsk);
 128                /*
 129                 * Accumulate here the counters for all threads but the
 130                 * group leader as they die, so they can be added into
 131                 * the process-wide totals when those are taken.
 132                 * The group leader stays around as a zombie as long
 133                 * as there are other threads.  When it gets reaped,
 134                 * the exit.c code will add its counts into these totals.
 135                 * We won't ever get here for the group leader, since it
 136                 * will have been the last reference on the signal_struct.
 137                 */
 138                sig->utime += tsk->utime;
 139                sig->stime += tsk->stime;
 140                sig->gtime += tsk->gtime;
 141                sig->min_flt += tsk->min_flt;
 142                sig->maj_flt += tsk->maj_flt;
 143                sig->nvcsw += tsk->nvcsw;
 144                sig->nivcsw += tsk->nivcsw;
 145                sig->inblock += task_io_get_inblock(tsk);
 146                sig->oublock += task_io_get_oublock(tsk);
 147                task_io_accounting_add(&sig->ioac, &tsk->ioac);
 148                sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 149        }
 150
 151        sig->nr_threads--;
 152        __unhash_process(tsk, group_dead);
 153
 154        /*
 155         * Do this under ->siglock, we can race with another thread
 156         * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
 157         */
 158        flush_sigqueue(&tsk->pending);
 159        tsk->sighand = NULL;
 160        spin_unlock(&sighand->siglock);
 161
 162        __cleanup_sighand(sighand);
 163        clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
 164        if (group_dead) {
 165                flush_sigqueue(&sig->shared_pending);
 166                tty_kref_put(tty);
 167        }
 168}
 169
 170static void delayed_put_task_struct(struct rcu_head *rhp)
 171{
 172        struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
 173
 174        perf_event_delayed_put(tsk);
 175        trace_sched_process_free(tsk);
 176        put_task_struct(tsk);
 177}
 178
 179
 180void release_task(struct task_struct * p)
 181{
 182        struct task_struct *leader;
 183        int zap_leader;
 184repeat:
 185        /* don't need to get the RCU readlock here - the process is dead and
 186         * can't be modifying its own credentials. But shut RCU-lockdep up */
 187        rcu_read_lock();
 188        atomic_dec(&__task_cred(p)->user->processes);
 189        rcu_read_unlock();
 190
 191        proc_flush_task(p);
 192
 193        write_lock_irq(&tasklist_lock);
 194        ptrace_release_task(p);
 195        __exit_signal(p);
 196
 197        /*
 198         * If we are the last non-leader member of the thread
 199         * group, and the leader is zombie, then notify the
 200         * group leader's parent process. (if it wants notification.)
 201         */
 202        zap_leader = 0;
 203        leader = p->group_leader;
 204        if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
 205                /*
 206                 * If we were the last child thread and the leader has
 207                 * exited already, and the leader's parent ignores SIGCHLD,
 208                 * then we are the one who should release the leader.
 209                 */
 210                zap_leader = do_notify_parent(leader, leader->exit_signal);
 211                if (zap_leader)
 212                        leader->exit_state = EXIT_DEAD;
 213        }
 214
 215        write_unlock_irq(&tasklist_lock);
 216        release_thread(p);
 217        call_rcu(&p->rcu, delayed_put_task_struct);
 218
 219        p = leader;
 220        if (unlikely(zap_leader))
 221                goto repeat;
 222}
 223
 224/*
 225 * This checks not only the pgrp, but falls back on the pid if no
 226 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 227 * without this...
 228 *
 229 * The caller must hold rcu lock or the tasklist lock.
 230 */
 231struct pid *session_of_pgrp(struct pid *pgrp)
 232{
 233        struct task_struct *p;
 234        struct pid *sid = NULL;
 235
 236        p = pid_task(pgrp, PIDTYPE_PGID);
 237        if (p == NULL)
 238                p = pid_task(pgrp, PIDTYPE_PID);
 239        if (p != NULL)
 240                sid = task_session(p);
 241
 242        return sid;
 243}
 244
 245/*
 246 * Determine if a process group is "orphaned", according to the POSIX
 247 * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 248 * by terminal-generated stop signals.  Newly orphaned process groups are
 249 * to receive a SIGHUP and a SIGCONT.
 250 *
 251 * "I ask you, have you ever known what it is to be an orphan?"
 252 */
 253static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
 254{
 255        struct task_struct *p;
 256
 257        do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 258                if ((p == ignored_task) ||
 259                    (p->exit_state && thread_group_empty(p)) ||
 260                    is_global_init(p->real_parent))
 261                        continue;
 262
 263                if (task_pgrp(p->real_parent) != pgrp &&
 264                    task_session(p->real_parent) == task_session(p))
 265                        return 0;
 266        } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 267
 268        return 1;
 269}
 270
 271int is_current_pgrp_orphaned(void)
 272{
 273        int retval;
 274
 275        read_lock(&tasklist_lock);
 276        retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 277        read_unlock(&tasklist_lock);
 278
 279        return retval;
 280}
 281
 282static bool has_stopped_jobs(struct pid *pgrp)
 283{
 284        struct task_struct *p;
 285
 286        do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 287                if (p->signal->flags & SIGNAL_STOP_STOPPED)
 288                        return true;
 289        } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 290
 291        return false;
 292}
 293
 294/*
 295 * Check to see if any process groups have become orphaned as
 296 * a result of our exiting, and if they have any stopped jobs,
 297 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
 298 */
 299static void
 300kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
 301{
 302        struct pid *pgrp = task_pgrp(tsk);
 303        struct task_struct *ignored_task = tsk;
 304
 305        if (!parent)
 306                 /* exit: our father is in a different pgrp than
 307                  * we are and we were the only connection outside.
 308                  */
 309                parent = tsk->real_parent;
 310        else
 311                /* reparent: our child is in a different pgrp than
 312                 * we are, and it was the only connection outside.
 313                 */
 314                ignored_task = NULL;
 315
 316        if (task_pgrp(parent) != pgrp &&
 317            task_session(parent) == task_session(tsk) &&
 318            will_become_orphaned_pgrp(pgrp, ignored_task) &&
 319            has_stopped_jobs(pgrp)) {
 320                __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 321                __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 322        }
 323}
 324
 325/**
 326 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
 327 *
 328 * If a kernel thread is launched as a result of a system call, or if
 329 * it ever exits, it should generally reparent itself to kthreadd so it
 330 * isn't in the way of other processes and is correctly cleaned up on exit.
 331 *
 332 * The various task state such as scheduling policy and priority may have
 333 * been inherited from a user process, so we reset them to sane values here.
 334 *
 335 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
 336 */
 337static void reparent_to_kthreadd(void)
 338{
 339        write_lock_irq(&tasklist_lock);
 340
 341        ptrace_unlink(current);
 342        /* Reparent to init */
 343        current->real_parent = current->parent = kthreadd_task;
 344        list_move_tail(&current->sibling, &current->real_parent->children);
 345
 346        /* Set the exit signal to SIGCHLD so we signal init on exit */
 347        current->exit_signal = SIGCHLD;
 348
 349        if (task_nice(current) < 0)
 350                set_user_nice(current, 0);
 351        /* cpus_allowed? */
 352        /* rt_priority? */
 353        /* signals? */
 354        memcpy(current->signal->rlim, init_task.signal->rlim,
 355               sizeof(current->signal->rlim));
 356
 357        atomic_inc(&init_cred.usage);
 358        commit_creds(&init_cred);
 359        write_unlock_irq(&tasklist_lock);
 360}
 361
 362void __set_special_pids(struct pid *pid)
 363{
 364        struct task_struct *curr = current->group_leader;
 365
 366        if (task_session(curr) != pid)
 367                change_pid(curr, PIDTYPE_SID, pid);
 368
 369        if (task_pgrp(curr) != pid)
 370                change_pid(curr, PIDTYPE_PGID, pid);
 371}
 372
 373static void set_special_pids(struct pid *pid)
 374{
 375        write_lock_irq(&tasklist_lock);
 376        __set_special_pids(pid);
 377        write_unlock_irq(&tasklist_lock);
 378}
 379
 380/*
 381 * Let kernel threads use this to say that they allow a certain signal.
 382 * Must not be used if kthread was cloned with CLONE_SIGHAND.
 383 */
 384int allow_signal(int sig)
 385{
 386        if (!valid_signal(sig) || sig < 1)
 387                return -EINVAL;
 388
 389        spin_lock_irq(&current->sighand->siglock);
 390        /* This is only needed for daemonize()'ed kthreads */
 391        sigdelset(&current->blocked, sig);
 392        /*
 393         * Kernel threads handle their own signals. Let the signal code
 394         * know it'll be handled, so that they don't get converted to
 395         * SIGKILL or just silently dropped.
 396         */
 397        current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 398        recalc_sigpending();
 399        spin_unlock_irq(&current->sighand->siglock);
 400        return 0;
 401}
 402
 403EXPORT_SYMBOL(allow_signal);
 404
 405int disallow_signal(int sig)
 406{
 407        if (!valid_signal(sig) || sig < 1)
 408                return -EINVAL;
 409
 410        spin_lock_irq(&current->sighand->siglock);
 411        current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
 412        recalc_sigpending();
 413        spin_unlock_irq(&current->sighand->siglock);
 414        return 0;
 415}
 416
 417EXPORT_SYMBOL(disallow_signal);
 418
 419/*
 420 *      Put all the gunge required to become a kernel thread without
 421 *      attached user resources in one place where it belongs.
 422 */
 423
 424void daemonize(const char *name, ...)
 425{
 426        va_list args;
 427        sigset_t blocked;
 428
 429        va_start(args, name);
 430        vsnprintf(current->comm, sizeof(current->comm), name, args);
 431        va_end(args);
 432
 433        /*
 434         * If we were started as result of loading a module, close all of the
 435         * user space pages.  We don't need them, and if we didn't close them
 436         * they would be locked into memory.
 437         */
 438        exit_mm(current);
 439        /*
 440         * We don't want to get frozen, in case system-wide hibernation
 441         * or suspend transition begins right now.
 442         */
 443        current->flags |= (PF_NOFREEZE | PF_KTHREAD);
 444
 445        if (current->nsproxy != &init_nsproxy) {
 446                get_nsproxy(&init_nsproxy);
 447                switch_task_namespaces(current, &init_nsproxy);
 448        }
 449        set_special_pids(&init_struct_pid);
 450        proc_clear_tty(current);
 451
 452        /* Block and flush all signals */
 453        sigfillset(&blocked);
 454        sigprocmask(SIG_BLOCK, &blocked, NULL);
 455        flush_signals(current);
 456
 457        /* Become as one with the init task */
 458
 459        daemonize_fs_struct();
 460        daemonize_descriptors();
 461
 462        reparent_to_kthreadd();
 463}
 464
 465EXPORT_SYMBOL(daemonize);
 466
 467#ifdef CONFIG_MM_OWNER
 468/*
 469 * A task is exiting.   If it owned this mm, find a new owner for the mm.
 470 */
 471void mm_update_next_owner(struct mm_struct *mm)
 472{
 473        struct task_struct *c, *g, *p = current;
 474
 475retry:
 476        /*
 477         * If the exiting or execing task is not the owner, it's
 478         * someone else's problem.
 479         */
 480        if (mm->owner != p)
 481                return;
 482        /*
 483         * The current owner is exiting/execing and there are no other
 484         * candidates.  Do not leave the mm pointing to a possibly
 485         * freed task structure.
 486         */
 487        if (atomic_read(&mm->mm_users) <= 1) {
 488                mm->owner = NULL;
 489                return;
 490        }
 491
 492        read_lock(&tasklist_lock);
 493        /*
 494         * Search in the children
 495         */
 496        list_for_each_entry(c, &p->children, sibling) {
 497                if (c->mm == mm)
 498                        goto assign_new_owner;
 499        }
 500
 501        /*
 502         * Search in the siblings
 503         */
 504        list_for_each_entry(c, &p->real_parent->children, sibling) {
 505                if (c->mm == mm)
 506                        goto assign_new_owner;
 507        }
 508
 509        /*
 510         * Search through everything else. We should not get
 511         * here often
 512         */
 513        do_each_thread(g, c) {
 514                if (c->mm == mm)
 515                        goto assign_new_owner;
 516        } while_each_thread(g, c);
 517
 518        read_unlock(&tasklist_lock);
 519        /*
 520         * We found no owner yet mm_users > 1: this implies that we are
 521         * most likely racing with swapoff (try_to_unuse()) or /proc or
 522         * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
 523         */
 524        mm->owner = NULL;
 525        return;
 526
 527assign_new_owner:
 528        BUG_ON(c == p);
 529        get_task_struct(c);
 530        /*
 531         * The task_lock protects c->mm from changing.
 532         * We always want mm->owner->mm == mm
 533         */
 534        task_lock(c);
 535        /*
 536         * Delay read_unlock() till we have the task_lock()
 537         * to ensure that c does not slip away underneath us
 538         */
 539        read_unlock(&tasklist_lock);
 540        if (c->mm != mm) {
 541                task_unlock(c);
 542                put_task_struct(c);
 543                goto retry;
 544        }
 545        mm->owner = c;
 546        task_unlock(c);
 547        put_task_struct(c);
 548}
 549#endif /* CONFIG_MM_OWNER */
 550
 551/*
 552 * Turn us into a lazy TLB process if we
 553 * aren't already..
 554 */
 555static void exit_mm(struct task_struct * tsk)
 556{
 557        struct mm_struct *mm = tsk->mm;
 558        struct core_state *core_state;
 559
 560        mm_release(tsk, mm);
 561        if (!mm)
 562                return;
 563        sync_mm_rss(mm);
 564        /*
 565         * Serialize with any possible pending coredump.
 566         * We must hold mmap_sem around checking core_state
 567         * and clearing tsk->mm.  The core-inducing thread
 568         * will increment ->nr_threads for each thread in the
 569         * group with ->mm != NULL.
 570         */
 571        down_read(&mm->mmap_sem);
 572        core_state = mm->core_state;
 573        if (core_state) {
 574                struct core_thread self;
 575                up_read(&mm->mmap_sem);
 576
 577                self.task = tsk;
 578                self.next = xchg(&core_state->dumper.next, &self);
 579                /*
 580                 * Implies mb(), the result of xchg() must be visible
 581                 * to core_state->dumper.
 582                 */
 583                if (atomic_dec_and_test(&core_state->nr_threads))
 584                        complete(&core_state->startup);
 585
 586                for (;;) {
 587                        set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 588                        if (!self.task) /* see coredump_finish() */
 589                                break;
 590                        schedule();
 591                }
 592                __set_task_state(tsk, TASK_RUNNING);
 593                down_read(&mm->mmap_sem);
 594        }
 595        atomic_inc(&mm->mm_count);
 596        BUG_ON(mm != tsk->active_mm);
 597        /* more a memory barrier than a real lock */
 598        task_lock(tsk);
 599        tsk->mm = NULL;
 600        up_read(&mm->mmap_sem);
 601        enter_lazy_tlb(mm, current);
 602        task_unlock(tsk);
 603        mm_update_next_owner(mm);
 604        mmput(mm);
 605}
 606
 607/*
 608 * When we die, we re-parent all our children, and try to:
 609 * 1. give them to another thread in our thread group, if such a member exists
 610 * 2. give it to the first ancestor process which prctl'd itself as a
 611 *    child_subreaper for its children (like a service manager)
 612 * 3. give it to the init process (PID 1) in our pid namespace
 613 */
 614static struct task_struct *find_new_reaper(struct task_struct *father)
 615        __releases(&tasklist_lock)
 616        __acquires(&tasklist_lock)
 617{
 618        struct pid_namespace *pid_ns = task_active_pid_ns(father);
 619        struct task_struct *thread;
 620
 621        thread = father;
 622        while_each_thread(father, thread) {
 623                if (thread->flags & PF_EXITING)
 624                        continue;
 625                if (unlikely(pid_ns->child_reaper == father))
 626                        pid_ns->child_reaper = thread;
 627                return thread;
 628        }
 629
 630        if (unlikely(pid_ns->child_reaper == father)) {
 631                write_unlock_irq(&tasklist_lock);
 632                if (unlikely(pid_ns == &init_pid_ns)) {
 633                        panic("Attempted to kill init! exitcode=0x%08x\n",
 634                                father->signal->group_exit_code ?:
 635                                        father->exit_code);
 636                }
 637
 638                zap_pid_ns_processes(pid_ns);
 639                write_lock_irq(&tasklist_lock);
 640        } else if (father->signal->has_child_subreaper) {
 641                struct task_struct *reaper;
 642
 643                /*
 644                 * Find the first ancestor marked as child_subreaper.
 645                 * Note that the code below checks same_thread_group(reaper,
 646                 * pid_ns->child_reaper).  This is what we need to DTRT in a
 647                 * PID namespace. However we still need the check above, see
 648                 * http://marc.info/?l=linux-kernel&m=131385460420380
 649                 */
 650                for (reaper = father->real_parent;
 651                     reaper != &init_task;
 652                     reaper = reaper->real_parent) {
 653                        if (same_thread_group(reaper, pid_ns->child_reaper))
 654                                break;
 655                        if (!reaper->signal->is_child_subreaper)
 656                                continue;
 657                        thread = reaper;
 658                        do {
 659                                if (!(thread->flags & PF_EXITING))
 660                                        return reaper;
 661                        } while_each_thread(reaper, thread);
 662                }
 663        }
 664
 665        return pid_ns->child_reaper;
 666}
 667
 668/*
 669* Any that need to be release_task'd are put on the @dead list.
 670 */
 671static void reparent_leader(struct task_struct *father, struct task_struct *p,
 672                                struct list_head *dead)
 673{
 674        list_move_tail(&p->sibling, &p->real_parent->children);
 675
 676        if (p->exit_state == EXIT_DEAD)
 677                return;
 678        /*
 679         * If this is a threaded reparent there is no need to
 680         * notify anyone anything has happened.
 681         */
 682        if (same_thread_group(p->real_parent, father))
 683                return;
 684
 685        /* We don't want people slaying init.  */
 686        p->exit_signal = SIGCHLD;
 687
 688        /* If it has exited notify the new parent about this child's death. */
 689        if (!p->ptrace &&
 690            p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
 691                if (do_notify_parent(p, p->exit_signal)) {
 692                        p->exit_state = EXIT_DEAD;
 693                        list_move_tail(&p->sibling, dead);
 694                }
 695        }
 696
 697        kill_orphaned_pgrp(p, father);
 698}
 699
 700static void forget_original_parent(struct task_struct *father)
 701{
 702        struct task_struct *p, *n, *reaper;
 703        LIST_HEAD(dead_children);
 704
 705        write_lock_irq(&tasklist_lock);
 706        /*
 707         * Note that exit_ptrace() and find_new_reaper() might
 708         * drop tasklist_lock and reacquire it.
 709         */
 710        exit_ptrace(father);
 711        reaper = find_new_reaper(father);
 712
 713        list_for_each_entry_safe(p, n, &father->children, sibling) {
 714                struct task_struct *t = p;
 715                do {
 716                        t->real_parent = reaper;
 717                        if (t->parent == father) {
 718                                BUG_ON(t->ptrace);
 719                                t->parent = t->real_parent;
 720                        }
 721                        if (t->pdeath_signal)
 722                                group_send_sig_info(t->pdeath_signal,
 723                                                    SEND_SIG_NOINFO, t);
 724                } while_each_thread(p, t);
 725                reparent_leader(father, p, &dead_children);
 726        }
 727        write_unlock_irq(&tasklist_lock);
 728
 729        BUG_ON(!list_empty(&father->children));
 730
 731        list_for_each_entry_safe(p, n, &dead_children, sibling) {
 732                list_del_init(&p->sibling);
 733                release_task(p);
 734        }
 735}
 736
 737/*
 738 * Send signals to all our closest relatives so that they know
 739 * to properly mourn us..
 740 */
 741static void exit_notify(struct task_struct *tsk, int group_dead)
 742{
 743        bool autoreap;
 744
 745        /*
 746         * This does two things:
 747         *
 748         * A.  Make init inherit all the child processes
 749         * B.  Check to see if any process groups have become orphaned
 750         *      as a result of our exiting, and if they have any stopped
 751         *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 752         */
 753        forget_original_parent(tsk);
 754        exit_task_namespaces(tsk);
 755
 756        write_lock_irq(&tasklist_lock);
 757        if (group_dead)
 758                kill_orphaned_pgrp(tsk->group_leader, NULL);
 759
 760        if (unlikely(tsk->ptrace)) {
 761                int sig = thread_group_leader(tsk) &&
 762                                thread_group_empty(tsk) &&
 763                                !ptrace_reparented(tsk) ?
 764                        tsk->exit_signal : SIGCHLD;
 765                autoreap = do_notify_parent(tsk, sig);
 766        } else if (thread_group_leader(tsk)) {
 767                autoreap = thread_group_empty(tsk) &&
 768                        do_notify_parent(tsk, tsk->exit_signal);
 769        } else {
 770                autoreap = true;
 771        }
 772
 773        tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
 774
 775        /* mt-exec, de_thread() is waiting for group leader */
 776        if (unlikely(tsk->signal->notify_count < 0))
 777                wake_up_process(tsk->signal->group_exit_task);
 778        write_unlock_irq(&tasklist_lock);
 779
 780        /* If the process is dead, release it - nobody will wait for it */
 781        if (autoreap)
 782                release_task(tsk);
 783}
 784
 785#ifdef CONFIG_DEBUG_STACK_USAGE
 786static void check_stack_usage(void)
 787{
 788        static DEFINE_SPINLOCK(low_water_lock);
 789        static int lowest_to_date = THREAD_SIZE;
 790        unsigned long free;
 791
 792        free = stack_not_used(current);
 793
 794        if (free >= lowest_to_date)
 795                return;
 796
 797        spin_lock(&low_water_lock);
 798        if (free < lowest_to_date) {
 799                printk(KERN_WARNING "%s (%d) used greatest stack depth: "
 800                                "%lu bytes left\n",
 801                                current->comm, task_pid_nr(current), free);
 802                lowest_to_date = free;
 803        }
 804        spin_unlock(&low_water_lock);
 805}
 806#else
 807static inline void check_stack_usage(void) {}
 808#endif
 809
 810void do_exit(long code)
 811{
 812        struct task_struct *tsk = current;
 813        int group_dead;
 814
 815        profile_task_exit(tsk);
 816
 817        WARN_ON(blk_needs_flush_plug(tsk));
 818
 819        if (unlikely(in_interrupt()))
 820                panic("Aiee, killing interrupt handler!");
 821        if (unlikely(!tsk->pid))
 822                panic("Attempted to kill the idle task!");
 823
 824        /*
 825         * If do_exit is called because this processes oopsed, it's possible
 826         * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
 827         * continuing. Amongst other possible reasons, this is to prevent
 828         * mm_release()->clear_child_tid() from writing to a user-controlled
 829         * kernel address.
 830         */
 831        set_fs(USER_DS);
 832
 833        ptrace_event(PTRACE_EVENT_EXIT, code);
 834
 835        validate_creds_for_do_exit(tsk);
 836
 837        /*
 838         * We're taking recursive faults here in do_exit. Safest is to just
 839         * leave this task alone and wait for reboot.
 840         */
 841        if (unlikely(tsk->flags & PF_EXITING)) {
 842                printk(KERN_ALERT
 843                        "Fixing recursive fault but reboot is needed!\n");
 844                /*
 845                 * We can do this unlocked here. The futex code uses
 846                 * this flag just to verify whether the pi state
 847                 * cleanup has been done or not. In the worst case it
 848                 * loops once more. We pretend that the cleanup was
 849                 * done as there is no way to return. Either the
 850                 * OWNER_DIED bit is set by now or we push the blocked
 851                 * task into the wait for ever nirwana as well.
 852                 */
 853                tsk->flags |= PF_EXITPIDONE;
 854                set_current_state(TASK_UNINTERRUPTIBLE);
 855                schedule();
 856        }
 857
 858        exit_signals(tsk);  /* sets PF_EXITING */
 859        /*
 860         * tsk->flags are checked in the futex code to protect against
 861         * an exiting task cleaning up the robust pi futexes.
 862         */
 863        smp_mb();
 864        raw_spin_unlock_wait(&tsk->pi_lock);
 865
 866        if (unlikely(in_atomic()))
 867                printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 868                                current->comm, task_pid_nr(current),
 869                                preempt_count());
 870
 871        acct_update_integrals(tsk);
 872        /* sync mm's RSS info before statistics gathering */
 873        if (tsk->mm)
 874                sync_mm_rss(tsk->mm);
 875        group_dead = atomic_dec_and_test(&tsk->signal->live);
 876        if (group_dead) {
 877                hrtimer_cancel(&tsk->signal->real_timer);
 878                exit_itimers(tsk->signal);
 879                if (tsk->mm)
 880                        setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
 881        }
 882        acct_collect(code, group_dead);
 883        if (group_dead)
 884                tty_audit_exit();
 885        audit_free(tsk);
 886
 887        tsk->exit_code = code;
 888        taskstats_exit(tsk, group_dead);
 889
 890        exit_mm(tsk);
 891
 892        if (group_dead)
 893                acct_process();
 894        trace_sched_process_exit(tsk);
 895
 896        exit_sem(tsk);
 897        exit_shm(tsk);
 898        exit_files(tsk);
 899        exit_fs(tsk);
 900        exit_task_work(tsk);
 901        check_stack_usage();
 902        exit_thread();
 903
 904        /*
 905         * Flush inherited counters to the parent - before the parent
 906         * gets woken up by child-exit notifications.
 907         *
 908         * because of cgroup mode, must be called before cgroup_exit()
 909         */
 910        perf_event_exit_task(tsk);
 911
 912        cgroup_exit(tsk, 1);
 913
 914        if (group_dead)
 915                disassociate_ctty(1);
 916
 917        module_put(task_thread_info(tsk)->exec_domain->module);
 918
 919        proc_exit_connector(tsk);
 920
 921        /*
 922         * FIXME: do that only when needed, using sched_exit tracepoint
 923         */
 924        ptrace_put_breakpoints(tsk);
 925
 926        exit_notify(tsk, group_dead);
 927#ifdef CONFIG_NUMA
 928        task_lock(tsk);
 929        mpol_put(tsk->mempolicy);
 930        tsk->mempolicy = NULL;
 931        task_unlock(tsk);
 932#endif
 933#ifdef CONFIG_FUTEX
 934        if (unlikely(current->pi_state_cache))
 935                kfree(current->pi_state_cache);
 936#endif
 937        /*
 938         * Make sure we are holding no locks:
 939         */
 940        debug_check_no_locks_held(tsk);
 941        /*
 942         * We can do this unlocked here. The futex code uses this flag
 943         * just to verify whether the pi state cleanup has been done
 944         * or not. In the worst case it loops once more.
 945         */
 946        tsk->flags |= PF_EXITPIDONE;
 947
 948        if (tsk->io_context)
 949                exit_io_context(tsk);
 950
 951        if (tsk->splice_pipe)
 952                __free_pipe_info(tsk->splice_pipe);
 953
 954        if (tsk->task_frag.page)
 955                put_page(tsk->task_frag.page);
 956
 957        validate_creds_for_do_exit(tsk);
 958
 959        preempt_disable();
 960        if (tsk->nr_dirtied)
 961                __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
 962        exit_rcu();
 963
 964        /*
 965         * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
 966         * when the following two conditions become true.
 967         *   - There is race condition of mmap_sem (It is acquired by
 968         *     exit_mm()), and
 969         *   - SMI occurs before setting TASK_RUNINNG.
 970         *     (or hypervisor of virtual machine switches to other guest)
 971         *  As a result, we may become TASK_RUNNING after becoming TASK_DEAD
 972         *
 973         * To avoid it, we have to wait for releasing tsk->pi_lock which
 974         * is held by try_to_wake_up()
 975         */
 976        smp_mb();
 977        raw_spin_unlock_wait(&tsk->pi_lock);
 978
 979        /* causes final put_task_struct in finish_task_switch(). */
 980        tsk->state = TASK_DEAD;
 981        tsk->flags |= PF_NOFREEZE;      /* tell freezer to ignore us */
 982        schedule();
 983        BUG();
 984        /* Avoid "noreturn function does return".  */
 985        for (;;)
 986                cpu_relax();    /* For when BUG is null */
 987}
 988
 989EXPORT_SYMBOL_GPL(do_exit);
 990
 991void complete_and_exit(struct completion *comp, long code)
 992{
 993        if (comp)
 994                complete(comp);
 995
 996        do_exit(code);
 997}
 998
 999EXPORT_SYMBOL(complete_and_exit);
1000
1001SYSCALL_DEFINE1(exit, int, error_code)
1002{
1003        do_exit((error_code&0xff)<<8);
1004}
1005
1006/*
1007 * Take down every thread in the group.  This is called by fatal signals
1008 * as well as by sys_exit_group (below).
1009 */
1010void
1011do_group_exit(int exit_code)
1012{
1013        struct signal_struct *sig = current->signal;
1014
1015        BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1016
1017        if (signal_group_exit(sig))
1018                exit_code = sig->group_exit_code;
1019        else if (!thread_group_empty(current)) {
1020                struct sighand_struct *const sighand = current->sighand;
1021                spin_lock_irq(&sighand->siglock);
1022                if (signal_group_exit(sig))
1023                        /* Another thread got here before we took the lock.  */
1024                        exit_code = sig->group_exit_code;
1025                else {
1026                        sig->group_exit_code = exit_code;
1027                        sig->flags = SIGNAL_GROUP_EXIT;
1028                        zap_other_threads(current);
1029                }
1030                spin_unlock_irq(&sighand->siglock);
1031        }
1032
1033        do_exit(exit_code);
1034        /* NOTREACHED */
1035}
1036
1037/*
1038 * this kills every thread in the thread group. Note that any externally
1039 * wait4()-ing process will get the correct exit code - even if this
1040 * thread is not the thread group leader.
1041 */
1042SYSCALL_DEFINE1(exit_group, int, error_code)
1043{
1044        do_group_exit((error_code & 0xff) << 8);
1045        /* NOTREACHED */
1046        return 0;
1047}
1048
1049struct wait_opts {
1050        enum pid_type           wo_type;
1051        int                     wo_flags;
1052        struct pid              *wo_pid;
1053
1054        struct siginfo __user   *wo_info;
1055        int __user              *wo_stat;
1056        struct rusage __user    *wo_rusage;
1057
1058        wait_queue_t            child_wait;
1059        int                     notask_error;
1060};
1061
1062static inline
1063struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1064{
1065        if (type != PIDTYPE_PID)
1066                task = task->group_leader;
1067        return task->pids[type].pid;
1068}
1069
1070static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1071{
1072        return  wo->wo_type == PIDTYPE_MAX ||
1073                task_pid_type(p, wo->wo_type) == wo->wo_pid;
1074}
1075
1076static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1077{
1078        if (!eligible_pid(wo, p))
1079                return 0;
1080        /* Wait for all children (clone and not) if __WALL is set;
1081         * otherwise, wait for clone children *only* if __WCLONE is
1082         * set; otherwise, wait for non-clone children *only*.  (Note:
1083         * A "clone" child here is one that reports to its parent
1084         * using a signal other than SIGCHLD.) */
1085        if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1086            && !(wo->wo_flags & __WALL))
1087                return 0;
1088
1089        return 1;
1090}
1091
1092static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1093                                pid_t pid, uid_t uid, int why, int status)
1094{
1095        struct siginfo __user *infop;
1096        int retval = wo->wo_rusage
1097                ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1098
1099        put_task_struct(p);
1100        infop = wo->wo_info;
1101        if (infop) {
1102                if (!retval)
1103                        retval = put_user(SIGCHLD, &infop->si_signo);
1104                if (!retval)
1105                        retval = put_user(0, &infop->si_errno);
1106                if (!retval)
1107                        retval = put_user((short)why, &infop->si_code);
1108                if (!retval)
1109                        retval = put_user(pid, &infop->si_pid);
1110                if (!retval)
1111                        retval = put_user(uid, &infop->si_uid);
1112                if (!retval)
1113                        retval = put_user(status, &infop->si_status);
1114        }
1115        if (!retval)
1116                retval = pid;
1117        return retval;
1118}
1119
1120/*
1121 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1122 * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1123 * the lock and this task is uninteresting.  If we return nonzero, we have
1124 * released the lock and the system call should return.
1125 */
1126static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1127{
1128        unsigned long state;
1129        int retval, status, traced;
1130        pid_t pid = task_pid_vnr(p);
1131        uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1132        struct siginfo __user *infop;
1133
1134        if (!likely(wo->wo_flags & WEXITED))
1135                return 0;
1136
1137        if (unlikely(wo->wo_flags & WNOWAIT)) {
1138                int exit_code = p->exit_code;
1139                int why;
1140
1141                get_task_struct(p);
1142                read_unlock(&tasklist_lock);
1143                if ((exit_code & 0x7f) == 0) {
1144                        why = CLD_EXITED;
1145                        status = exit_code >> 8;
1146                } else {
1147                        why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1148                        status = exit_code & 0x7f;
1149                }
1150                return wait_noreap_copyout(wo, p, pid, uid, why, status);
1151        }
1152
1153        /*
1154         * Try to move the task's state to DEAD
1155         * only one thread is allowed to do this:
1156         */
1157        state = xchg(&p->exit_state, EXIT_DEAD);
1158        if (state != EXIT_ZOMBIE) {
1159                BUG_ON(state != EXIT_DEAD);
1160                return 0;
1161        }
1162
1163        traced = ptrace_reparented(p);
1164        /*
1165         * It can be ptraced but not reparented, check
1166         * thread_group_leader() to filter out sub-threads.
1167         */
1168        if (likely(!traced) && thread_group_leader(p)) {
1169                struct signal_struct *psig;
1170                struct signal_struct *sig;
1171                unsigned long maxrss;
1172                cputime_t tgutime, tgstime;
1173
1174                /*
1175                 * The resource counters for the group leader are in its
1176                 * own task_struct.  Those for dead threads in the group
1177                 * are in its signal_struct, as are those for the child
1178                 * processes it has previously reaped.  All these
1179                 * accumulate in the parent's signal_struct c* fields.
1180                 *
1181                 * We don't bother to take a lock here to protect these
1182                 * p->signal fields, because they are only touched by
1183                 * __exit_signal, which runs with tasklist_lock
1184                 * write-locked anyway, and so is excluded here.  We do
1185                 * need to protect the access to parent->signal fields,
1186                 * as other threads in the parent group can be right
1187                 * here reaping other children at the same time.
1188                 *
1189                 * We use thread_group_times() to get times for the thread
1190                 * group, which consolidates times for all threads in the
1191                 * group including the group leader.
1192                 */
1193                thread_group_times(p, &tgutime, &tgstime);
1194                spin_lock_irq(&p->real_parent->sighand->siglock);
1195                psig = p->real_parent->signal;
1196                sig = p->signal;
1197                psig->cutime += tgutime + sig->cutime;
1198                psig->cstime += tgstime + sig->cstime;
1199                psig->cgtime += p->gtime + sig->gtime + sig->cgtime;
1200                psig->cmin_flt +=
1201                        p->min_flt + sig->min_flt + sig->cmin_flt;
1202                psig->cmaj_flt +=
1203                        p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1204                psig->cnvcsw +=
1205                        p->nvcsw + sig->nvcsw + sig->cnvcsw;
1206                psig->cnivcsw +=
1207                        p->nivcsw + sig->nivcsw + sig->cnivcsw;
1208                psig->cinblock +=
1209                        task_io_get_inblock(p) +
1210                        sig->inblock + sig->cinblock;
1211                psig->coublock +=
1212                        task_io_get_oublock(p) +
1213                        sig->oublock + sig->coublock;
1214                maxrss = max(sig->maxrss, sig->cmaxrss);
1215                if (psig->cmaxrss < maxrss)
1216                        psig->cmaxrss = maxrss;
1217                task_io_accounting_add(&psig->ioac, &p->ioac);
1218                task_io_accounting_add(&psig->ioac, &sig->ioac);
1219                spin_unlock_irq(&p->real_parent->sighand->siglock);
1220        }
1221
1222        /*
1223         * Now we are sure this task is interesting, and no other
1224         * thread can reap it because we set its state to EXIT_DEAD.
1225         */
1226        read_unlock(&tasklist_lock);
1227
1228        retval = wo->wo_rusage
1229                ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1230        status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1231                ? p->signal->group_exit_code : p->exit_code;
1232        if (!retval && wo->wo_stat)
1233                retval = put_user(status, wo->wo_stat);
1234
1235        infop = wo->wo_info;
1236        if (!retval && infop)
1237                retval = put_user(SIGCHLD, &infop->si_signo);
1238        if (!retval && infop)
1239                retval = put_user(0, &infop->si_errno);
1240        if (!retval && infop) {
1241                int why;
1242
1243                if ((status & 0x7f) == 0) {
1244                        why = CLD_EXITED;
1245                        status >>= 8;
1246                } else {
1247                        why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1248                        status &= 0x7f;
1249                }
1250                retval = put_user((short)why, &infop->si_code);
1251                if (!retval)
1252                        retval = put_user(status, &infop->si_status);
1253        }
1254        if (!retval && infop)
1255                retval = put_user(pid, &infop->si_pid);
1256        if (!retval && infop)
1257                retval = put_user(uid, &infop->si_uid);
1258        if (!retval)
1259                retval = pid;
1260
1261        if (traced) {
1262                write_lock_irq(&tasklist_lock);
1263                /* We dropped tasklist, ptracer could die and untrace */
1264                ptrace_unlink(p);
1265                /*
1266                 * If this is not a sub-thread, notify the parent.
1267                 * If parent wants a zombie, don't release it now.
1268                 */
1269                if (thread_group_leader(p) &&
1270                    !do_notify_parent(p, p->exit_signal)) {
1271                        p->exit_state = EXIT_ZOMBIE;
1272                        p = NULL;
1273                }
1274                write_unlock_irq(&tasklist_lock);
1275        }
1276        if (p != NULL)
1277                release_task(p);
1278
1279        return retval;
1280}
1281
1282static int *task_stopped_code(struct task_struct *p, bool ptrace)
1283{
1284        if (ptrace) {
1285                if (task_is_stopped_or_traced(p) &&
1286                    !(p->jobctl & JOBCTL_LISTENING))
1287                        return &p->exit_code;
1288        } else {
1289                if (p->signal->flags & SIGNAL_STOP_STOPPED)
1290                        return &p->signal->group_exit_code;
1291        }
1292        return NULL;
1293}
1294
1295/**
1296 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1297 * @wo: wait options
1298 * @ptrace: is the wait for ptrace
1299 * @p: task to wait for
1300 *
1301 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1302 *
1303 * CONTEXT:
1304 * read_lock(&tasklist_lock), which is released if return value is
1305 * non-zero.  Also, grabs and releases @p->sighand->siglock.
1306 *
1307 * RETURNS:
1308 * 0 if wait condition didn't exist and search for other wait conditions
1309 * should continue.  Non-zero return, -errno on failure and @p's pid on
1310 * success, implies that tasklist_lock is released and wait condition
1311 * search should terminate.
1312 */
1313static int wait_task_stopped(struct wait_opts *wo,
1314                                int ptrace, struct task_struct *p)
1315{
1316        struct siginfo __user *infop;
1317        int retval, exit_code, *p_code, why;
1318        uid_t uid = 0; /* unneeded, required by compiler */
1319        pid_t pid;
1320
1321        /*
1322         * Traditionally we see ptrace'd stopped tasks regardless of options.
1323         */
1324        if (!ptrace && !(wo->wo_flags & WUNTRACED))
1325                return 0;
1326
1327        if (!task_stopped_code(p, ptrace))
1328                return 0;
1329
1330        exit_code = 0;
1331        spin_lock_irq(&p->sighand->siglock);
1332
1333        p_code = task_stopped_code(p, ptrace);
1334        if (unlikely(!p_code))
1335                goto unlock_sig;
1336
1337        exit_code = *p_code;
1338        if (!exit_code)
1339                goto unlock_sig;
1340
1341        if (!unlikely(wo->wo_flags & WNOWAIT))
1342                *p_code = 0;
1343
1344        uid = from_kuid_munged(current_user_ns(), task_uid(p));
1345unlock_sig:
1346        spin_unlock_irq(&p->sighand->siglock);
1347        if (!exit_code)
1348                return 0;
1349
1350        /*
1351         * Now we are pretty sure this task is interesting.
1352         * Make sure it doesn't get reaped out from under us while we
1353         * give up the lock and then examine it below.  We don't want to
1354         * keep holding onto the tasklist_lock while we call getrusage and
1355         * possibly take page faults for user memory.
1356         */
1357        get_task_struct(p);
1358        pid = task_pid_vnr(p);
1359        why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1360        read_unlock(&tasklist_lock);
1361
1362        if (unlikely(wo->wo_flags & WNOWAIT))
1363                return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1364
1365        retval = wo->wo_rusage
1366                ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1367        if (!retval && wo->wo_stat)
1368                retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1369
1370        infop = wo->wo_info;
1371        if (!retval && infop)
1372                retval = put_user(SIGCHLD, &infop->si_signo);
1373        if (!retval && infop)
1374                retval = put_user(0, &infop->si_errno);
1375        if (!retval && infop)
1376                retval = put_user((short)why, &infop->si_code);
1377        if (!retval && infop)
1378                retval = put_user(exit_code, &infop->si_status);
1379        if (!retval && infop)
1380                retval = put_user(pid, &infop->si_pid);
1381        if (!retval && infop)
1382                retval = put_user(uid, &infop->si_uid);
1383        if (!retval)
1384                retval = pid;
1385        put_task_struct(p);
1386
1387        BUG_ON(!retval);
1388        return retval;
1389}
1390
1391/*
1392 * Handle do_wait work for one task in a live, non-stopped state.
1393 * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1394 * the lock and this task is uninteresting.  If we return nonzero, we have
1395 * released the lock and the system call should return.
1396 */
1397static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1398{
1399        int retval;
1400        pid_t pid;
1401        uid_t uid;
1402
1403        if (!unlikely(wo->wo_flags & WCONTINUED))
1404                return 0;
1405
1406        if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1407                return 0;
1408
1409        spin_lock_irq(&p->sighand->siglock);
1410        /* Re-check with the lock held.  */
1411        if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1412                spin_unlock_irq(&p->sighand->siglock);
1413                return 0;
1414        }
1415        if (!unlikely(wo->wo_flags & WNOWAIT))
1416                p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1417        uid = from_kuid_munged(current_user_ns(), task_uid(p));
1418        spin_unlock_irq(&p->sighand->siglock);
1419
1420        pid = task_pid_vnr(p);
1421        get_task_struct(p);
1422        read_unlock(&tasklist_lock);
1423
1424        if (!wo->wo_info) {
1425                retval = wo->wo_rusage
1426                        ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1427                put_task_struct(p);
1428                if (!retval && wo->wo_stat)
1429                        retval = put_user(0xffff, wo->wo_stat);
1430                if (!retval)
1431                        retval = pid;
1432        } else {
1433                retval = wait_noreap_copyout(wo, p, pid, uid,
1434                                             CLD_CONTINUED, SIGCONT);
1435                BUG_ON(retval == 0);
1436        }
1437
1438        return retval;
1439}
1440
1441/*
1442 * Consider @p for a wait by @parent.
1443 *
1444 * -ECHILD should be in ->notask_error before the first call.
1445 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1446 * Returns zero if the search for a child should continue;
1447 * then ->notask_error is 0 if @p is an eligible child,
1448 * or another error from security_task_wait(), or still -ECHILD.
1449 */
1450static int wait_consider_task(struct wait_opts *wo, int ptrace,
1451                                struct task_struct *p)
1452{
1453        int ret = eligible_child(wo, p);
1454        if (!ret)
1455                return ret;
1456
1457        ret = security_task_wait(p);
1458        if (unlikely(ret < 0)) {
1459                /*
1460                 * If we have not yet seen any eligible child,
1461                 * then let this error code replace -ECHILD.
1462                 * A permission error will give the user a clue
1463                 * to look for security policy problems, rather
1464                 * than for mysterious wait bugs.
1465                 */
1466                if (wo->notask_error)
1467                        wo->notask_error = ret;
1468                return 0;
1469        }
1470
1471        /* dead body doesn't have much to contribute */
1472        if (unlikely(p->exit_state == EXIT_DEAD)) {
1473                /*
1474                 * But do not ignore this task until the tracer does
1475                 * wait_task_zombie()->do_notify_parent().
1476                 */
1477                if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
1478                        wo->notask_error = 0;
1479                return 0;
1480        }
1481
1482        /* slay zombie? */
1483        if (p->exit_state == EXIT_ZOMBIE) {
1484                /*
1485                 * A zombie ptracee is only visible to its ptracer.
1486                 * Notification and reaping will be cascaded to the real
1487                 * parent when the ptracer detaches.
1488                 */
1489                if (likely(!ptrace) && unlikely(p->ptrace)) {
1490                        /* it will become visible, clear notask_error */
1491                        wo->notask_error = 0;
1492                        return 0;
1493                }
1494
1495                /* we don't reap group leaders with subthreads */
1496                if (!delay_group_leader(p))
1497                        return wait_task_zombie(wo, p);
1498
1499                /*
1500                 * Allow access to stopped/continued state via zombie by
1501                 * falling through.  Clearing of notask_error is complex.
1502                 *
1503                 * When !@ptrace:
1504                 *
1505                 * If WEXITED is set, notask_error should naturally be
1506                 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1507                 * so, if there are live subthreads, there are events to
1508                 * wait for.  If all subthreads are dead, it's still safe
1509                 * to clear - this function will be called again in finite
1510                 * amount time once all the subthreads are released and
1511                 * will then return without clearing.
1512                 *
1513                 * When @ptrace:
1514                 *
1515                 * Stopped state is per-task and thus can't change once the
1516                 * target task dies.  Only continued and exited can happen.
1517                 * Clear notask_error if WCONTINUED | WEXITED.
1518                 */
1519                if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1520                        wo->notask_error = 0;
1521        } else {
1522                /*
1523                 * If @p is ptraced by a task in its real parent's group,
1524                 * hide group stop/continued state when looking at @p as
1525                 * the real parent; otherwise, a single stop can be
1526                 * reported twice as group and ptrace stops.
1527                 *
1528                 * If a ptracer wants to distinguish the two events for its
1529                 * own children, it should create a separate process which
1530                 * takes the role of real parent.
1531                 */
1532                if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1533                        return 0;
1534
1535                /*
1536                 * @p is alive and it's gonna stop, continue or exit, so
1537                 * there always is something to wait for.
1538                 */
1539                wo->notask_error = 0;
1540        }
1541
1542        /*
1543         * Wait for stopped.  Depending on @ptrace, different stopped state
1544         * is used and the two don't interact with each other.
1545         */
1546        ret = wait_task_stopped(wo, ptrace, p);
1547        if (ret)
1548                return ret;
1549
1550        /*
1551         * Wait for continued.  There's only one continued state and the
1552         * ptracer can consume it which can confuse the real parent.  Don't
1553         * use WCONTINUED from ptracer.  You don't need or want it.
1554         */
1555        return wait_task_continued(wo, p);
1556}
1557
1558/*
1559 * Do the work of do_wait() for one thread in the group, @tsk.
1560 *
1561 * -ECHILD should be in ->notask_error before the first call.
1562 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1563 * Returns zero if the search for a child should continue; then
1564 * ->notask_error is 0 if there were any eligible children,
1565 * or another error from security_task_wait(), or still -ECHILD.
1566 */
1567static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1568{
1569        struct task_struct *p;
1570
1571        list_for_each_entry(p, &tsk->children, sibling) {
1572                int ret = wait_consider_task(wo, 0, p);
1573                if (ret)
1574                        return ret;
1575        }
1576
1577        return 0;
1578}
1579
1580static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1581{
1582        struct task_struct *p;
1583
1584        list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1585                int ret = wait_consider_task(wo, 1, p);
1586                if (ret)
1587                        return ret;
1588        }
1589
1590        return 0;
1591}
1592
1593static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1594                                int sync, void *key)
1595{
1596        struct wait_opts *wo = container_of(wait, struct wait_opts,
1597                                                child_wait);
1598        struct task_struct *p = key;
1599
1600        if (!eligible_pid(wo, p))
1601                return 0;
1602
1603        if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1604                return 0;
1605
1606        return default_wake_function(wait, mode, sync, key);
1607}
1608
1609void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1610{
1611        __wake_up_sync_key(&parent->signal->wait_chldexit,
1612                                TASK_INTERRUPTIBLE, 1, p);
1613}
1614
1615static long do_wait(struct wait_opts *wo)
1616{
1617        struct task_struct *tsk;
1618        int retval;
1619
1620        trace_sched_process_wait(wo->wo_pid);
1621
1622        init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1623        wo->child_wait.private = current;
1624        add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1625repeat:
1626        /*
1627         * If there is nothing that can match our critiera just get out.
1628         * We will clear ->notask_error to zero if we see any child that
1629         * might later match our criteria, even if we are not able to reap
1630         * it yet.
1631         */
1632        wo->notask_error = -ECHILD;
1633        if ((wo->wo_type < PIDTYPE_MAX) &&
1634           (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1635                goto notask;
1636
1637        set_current_state(TASK_INTERRUPTIBLE);
1638        read_lock(&tasklist_lock);
1639        tsk = current;
1640        do {
1641                retval = do_wait_thread(wo, tsk);
1642                if (retval)
1643                        goto end;
1644
1645                retval = ptrace_do_wait(wo, tsk);
1646                if (retval)
1647                        goto end;
1648
1649                if (wo->wo_flags & __WNOTHREAD)
1650                        break;
1651        } while_each_thread(current, tsk);
1652        read_unlock(&tasklist_lock);
1653
1654notask:
1655        retval = wo->notask_error;
1656        if (!retval && !(wo->wo_flags & WNOHANG)) {
1657                retval = -ERESTARTSYS;
1658                if (!signal_pending(current)) {
1659                        schedule();
1660                        goto repeat;
1661                }
1662        }
1663end:
1664        __set_current_state(TASK_RUNNING);
1665        remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1666        return retval;
1667}
1668
1669SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1670                infop, int, options, struct rusage __user *, ru)
1671{
1672        struct wait_opts wo;
1673        struct pid *pid = NULL;
1674        enum pid_type type;
1675        long ret;
1676
1677        if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1678                return -EINVAL;
1679        if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1680                return -EINVAL;
1681
1682        switch (which) {
1683        case P_ALL:
1684                type = PIDTYPE_MAX;
1685                break;
1686        case P_PID:
1687                type = PIDTYPE_PID;
1688                if (upid <= 0)
1689                        return -EINVAL;
1690                break;
1691        case P_PGID:
1692                type = PIDTYPE_PGID;
1693                if (upid <= 0)
1694                        return -EINVAL;
1695                break;
1696        default:
1697                return -EINVAL;
1698        }
1699
1700        if (type < PIDTYPE_MAX)
1701                pid = find_get_pid(upid);
1702
1703        wo.wo_type      = type;
1704        wo.wo_pid       = pid;
1705        wo.wo_flags     = options;
1706        wo.wo_info      = infop;
1707        wo.wo_stat      = NULL;
1708        wo.wo_rusage    = ru;
1709        ret = do_wait(&wo);
1710
1711        if (ret > 0) {
1712                ret = 0;
1713        } else if (infop) {
1714                /*
1715                 * For a WNOHANG return, clear out all the fields
1716                 * we would set so the user can easily tell the
1717                 * difference.
1718                 */
1719                if (!ret)
1720                        ret = put_user(0, &infop->si_signo);
1721                if (!ret)
1722                        ret = put_user(0, &infop->si_errno);
1723                if (!ret)
1724                        ret = put_user(0, &infop->si_code);
1725                if (!ret)
1726                        ret = put_user(0, &infop->si_pid);
1727                if (!ret)
1728                        ret = put_user(0, &infop->si_uid);
1729                if (!ret)
1730                        ret = put_user(0, &infop->si_status);
1731        }
1732
1733        put_pid(pid);
1734
1735        /* avoid REGPARM breakage on x86: */
1736        asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1737        return ret;
1738}
1739
1740SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1741                int, options, struct rusage __user *, ru)
1742{
1743        struct wait_opts wo;
1744        struct pid *pid = NULL;
1745        enum pid_type type;
1746        long ret;
1747
1748        if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1749                        __WNOTHREAD|__WCLONE|__WALL))
1750                return -EINVAL;
1751
1752        if (upid == -1)
1753                type = PIDTYPE_MAX;
1754        else if (upid < 0) {
1755                type = PIDTYPE_PGID;
1756                pid = find_get_pid(-upid);
1757        } else if (upid == 0) {
1758                type = PIDTYPE_PGID;
1759                pid = get_task_pid(current, PIDTYPE_PGID);
1760        } else /* upid > 0 */ {
1761                type = PIDTYPE_PID;
1762                pid = find_get_pid(upid);
1763        }
1764
1765        wo.wo_type      = type;
1766        wo.wo_pid       = pid;
1767        wo.wo_flags     = options | WEXITED;
1768        wo.wo_info      = NULL;
1769        wo.wo_stat      = stat_addr;
1770        wo.wo_rusage    = ru;
1771        ret = do_wait(&wo);
1772        put_pid(pid);
1773
1774        /* avoid REGPARM breakage on x86: */
1775        asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1776        return ret;
1777}
1778
1779#ifdef __ARCH_WANT_SYS_WAITPID
1780
1781/*
1782 * sys_waitpid() remains for compatibility. waitpid() should be
1783 * implemented by calling sys_wait4() from libc.a.
1784 */
1785SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1786{
1787        return sys_wait4(pid, stat_addr, options, NULL);
1788}
1789
1790#endif
1791
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.