Root/kernel/exit.c

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

Archive Download this file



interactive