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 * A task is exiting. If it owned this mm, find a new owner for the mm.
565 */
566void mm_update_next_owner(struct mm_struct *mm)
567{
568    struct task_struct *c, *g, *p = current;
569
570retry:
571    /*
572     * If the exiting or execing task is not the owner, it's
573     * someone else's problem.
574     */
575    if (mm->owner != p)
576        return;
577    /*
578     * The current owner is exiting/execing and there are no other
579     * candidates. Do not leave the mm pointing to a possibly
580     * freed task structure.
581     */
582    if (atomic_read(&mm->mm_users) <= 1) {
583        mm->owner = NULL;
584        return;
585    }
586
587    read_lock(&tasklist_lock);
588    /*
589     * Search in the children
590     */
591    list_for_each_entry(c, &p->children, sibling) {
592        if (c->mm == mm)
593            goto assign_new_owner;
594    }
595
596    /*
597     * Search in the siblings
598     */
599    list_for_each_entry(c, &p->real_parent->children, sibling) {
600        if (c->mm == mm)
601            goto assign_new_owner;
602    }
603
604    /*
605     * Search through everything else. We should not get
606     * here often
607     */
608    do_each_thread(g, c) {
609        if (c->mm == mm)
610            goto assign_new_owner;
611    } while_each_thread(g, c);
612
613    read_unlock(&tasklist_lock);
614    /*
615     * We found no owner yet mm_users > 1: this implies that we are
616     * most likely racing with swapoff (try_to_unuse()) or /proc or
617     * ptrace or page migration (get_task_mm()). Mark owner as NULL.
618     */
619    mm->owner = NULL;
620    return;
621
622assign_new_owner:
623    BUG_ON(c == p);
624    get_task_struct(c);
625    /*
626     * The task_lock protects c->mm from changing.
627     * We always want mm->owner->mm == mm
628     */
629    task_lock(c);
630    /*
631     * Delay read_unlock() till we have the task_lock()
632     * to ensure that c does not slip away underneath us
633     */
634    read_unlock(&tasklist_lock);
635    if (c->mm != mm) {
636        task_unlock(c);
637        put_task_struct(c);
638        goto retry;
639    }
640    mm->owner = c;
641    task_unlock(c);
642    put_task_struct(c);
643}
644#endif /* CONFIG_MM_OWNER */
645
646/*
647 * Turn us into a lazy TLB process if we
648 * aren't already..
649 */
650static void exit_mm(struct task_struct * tsk)
651{
652    struct mm_struct *mm = tsk->mm;
653    struct core_state *core_state;
654
655    mm_release(tsk, mm);
656    if (!mm)
657        return;
658    /*
659     * Serialize with any possible pending coredump.
660     * We must hold mmap_sem around checking core_state
661     * and clearing tsk->mm. The core-inducing thread
662     * will increment ->nr_threads for each thread in the
663     * group with ->mm != NULL.
664     */
665    down_read(&mm->mmap_sem);
666    core_state = mm->core_state;
667    if (core_state) {
668        struct core_thread self;
669        up_read(&mm->mmap_sem);
670
671        self.task = tsk;
672        self.next = xchg(&core_state->dumper.next, &self);
673        /*
674         * Implies mb(), the result of xchg() must be visible
675         * to core_state->dumper.
676         */
677        if (atomic_dec_and_test(&core_state->nr_threads))
678            complete(&core_state->startup);
679
680        for (;;) {
681            set_task_state(tsk, TASK_UNINTERRUPTIBLE);
682            if (!self.task) /* see coredump_finish() */
683                break;
684            schedule();
685        }
686        __set_task_state(tsk, TASK_RUNNING);
687        down_read(&mm->mmap_sem);
688    }
689    atomic_inc(&mm->mm_count);
690    BUG_ON(mm != tsk->active_mm);
691    /* more a memory barrier than a real lock */
692    task_lock(tsk);
693    tsk->mm = NULL;
694    up_read(&mm->mmap_sem);
695    enter_lazy_tlb(mm, current);
696    /* We don't want this task to be frozen prematurely */
697    clear_freeze_flag(tsk);
698    if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
699        atomic_dec(&mm->oom_disable_count);
700    task_unlock(tsk);
701    mm_update_next_owner(mm);
702    mmput(mm);
703}
704
705/*
706 * When we die, we re-parent all our children.
707 * Try to give them to another thread in our thread
708 * group, and if no such member exists, give it to
709 * the child reaper process (ie "init") in our pid
710 * space.
711 */
712static struct task_struct *find_new_reaper(struct task_struct *father)
713    __releases(&tasklist_lock)
714    __acquires(&tasklist_lock)
715{
716    struct pid_namespace *pid_ns = task_active_pid_ns(father);
717    struct task_struct *thread;
718
719    thread = father;
720    while_each_thread(father, thread) {
721        if (thread->flags & PF_EXITING)
722            continue;
723        if (unlikely(pid_ns->child_reaper == father))
724            pid_ns->child_reaper = thread;
725        return thread;
726    }
727
728    if (unlikely(pid_ns->child_reaper == father)) {
729        write_unlock_irq(&tasklist_lock);
730        if (unlikely(pid_ns == &init_pid_ns))
731            panic("Attempted to kill init!");
732
733        zap_pid_ns_processes(pid_ns);
734        write_lock_irq(&tasklist_lock);
735        /*
736         * We can not clear ->child_reaper or leave it alone.
737         * There may by stealth EXIT_DEAD tasks on ->children,
738         * forget_original_parent() must move them somewhere.
739         */
740        pid_ns->child_reaper = init_pid_ns.child_reaper;
741    }
742
743    return pid_ns->child_reaper;
744}
745
746/*
747* Any that need to be release_task'd are put on the @dead list.
748 */
749static void reparent_leader(struct task_struct *father, struct task_struct *p,
750                struct list_head *dead)
751{
752    list_move_tail(&p->sibling, &p->real_parent->children);
753
754    if (task_detached(p))
755        return;
756    /*
757     * If this is a threaded reparent there is no need to
758     * notify anyone anything has happened.
759     */
760    if (same_thread_group(p->real_parent, father))
761        return;
762
763    /* We don't want people slaying init. */
764    p->exit_signal = SIGCHLD;
765
766    /* If it has exited notify the new parent about this child's death. */
767    if (!task_ptrace(p) &&
768        p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
769        do_notify_parent(p, p->exit_signal);
770        if (task_detached(p)) {
771            p->exit_state = EXIT_DEAD;
772            list_move_tail(&p->sibling, dead);
773        }
774    }
775
776    kill_orphaned_pgrp(p, father);
777}
778
779static void forget_original_parent(struct task_struct *father)
780{
781    struct task_struct *p, *n, *reaper;
782    LIST_HEAD(dead_children);
783
784    write_lock_irq(&tasklist_lock);
785    /*
786     * Note that exit_ptrace() and find_new_reaper() might
787     * drop tasklist_lock and reacquire it.
788     */
789    exit_ptrace(father);
790    reaper = find_new_reaper(father);
791
792    list_for_each_entry_safe(p, n, &father->children, sibling) {
793        struct task_struct *t = p;
794        do {
795            t->real_parent = reaper;
796            if (t->parent == father) {
797                BUG_ON(task_ptrace(t));
798                t->parent = t->real_parent;
799            }
800            if (t->pdeath_signal)
801                group_send_sig_info(t->pdeath_signal,
802                            SEND_SIG_NOINFO, t);
803        } while_each_thread(p, t);
804        reparent_leader(father, p, &dead_children);
805    }
806    write_unlock_irq(&tasklist_lock);
807
808    BUG_ON(!list_empty(&father->children));
809
810    list_for_each_entry_safe(p, n, &dead_children, sibling) {
811        list_del_init(&p->sibling);
812        release_task(p);
813    }
814}
815
816/*
817 * Send signals to all our closest relatives so that they know
818 * to properly mourn us..
819 */
820static void exit_notify(struct task_struct *tsk, int group_dead)
821{
822    int signal;
823    void *cookie;
824
825    /*
826     * This does two things:
827     *
828       * A. Make init inherit all the child processes
829     * B. Check to see if any process groups have become orphaned
830     * as a result of our exiting, and if they have any stopped
831     * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
832     */
833    forget_original_parent(tsk);
834    exit_task_namespaces(tsk);
835
836    write_lock_irq(&tasklist_lock);
837    if (group_dead)
838        kill_orphaned_pgrp(tsk->group_leader, NULL);
839
840    /* Let father know we died
841     *
842     * Thread signals are configurable, but you aren't going to use
843     * that to send signals to arbitrary processes.
844     * That stops right now.
845     *
846     * If the parent exec id doesn't match the exec id we saved
847     * when we started then we know the parent has changed security
848     * domain.
849     *
850     * If our self_exec id doesn't match our parent_exec_id then
851     * we have changed execution domain as these two values started
852     * the same after a fork.
853     */
854    if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
855        (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
856         tsk->self_exec_id != tsk->parent_exec_id))
857        tsk->exit_signal = SIGCHLD;
858
859    signal = tracehook_notify_death(tsk, &cookie, group_dead);
860    if (signal >= 0)
861        signal = do_notify_parent(tsk, signal);
862
863    tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
864
865    /* mt-exec, de_thread() is waiting for group leader */
866    if (unlikely(tsk->signal->notify_count < 0))
867        wake_up_process(tsk->signal->group_exit_task);
868    write_unlock_irq(&tasklist_lock);
869
870    tracehook_report_death(tsk, signal, cookie, group_dead);
871
872    /* If the process is dead, release it - nobody will wait for it */
873    if (signal == DEATH_REAP)
874        release_task(tsk);
875}
876
877#ifdef CONFIG_DEBUG_STACK_USAGE
878static void check_stack_usage(void)
879{
880    static DEFINE_SPINLOCK(low_water_lock);
881    static int lowest_to_date = THREAD_SIZE;
882    unsigned long free;
883
884    free = stack_not_used(current);
885
886    if (free >= lowest_to_date)
887        return;
888
889    spin_lock(&low_water_lock);
890    if (free < lowest_to_date) {
891        printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
892                "left\n",
893                current->comm, free);
894        lowest_to_date = free;
895    }
896    spin_unlock(&low_water_lock);
897}
898#else
899static inline void check_stack_usage(void) {}
900#endif
901
902NORET_TYPE void do_exit(long code)
903{
904    struct task_struct *tsk = current;
905    int group_dead;
906
907    profile_task_exit(tsk);
908
909    WARN_ON(atomic_read(&tsk->fs_excl));
910    WARN_ON(blk_needs_flush_plug(tsk));
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    ptrace_put_breakpoints(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 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1381 * @wo: wait options
1382 * @ptrace: is the wait for ptrace
1383 * @p: task to wait for
1384 *
1385 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1386 *
1387 * CONTEXT:
1388 * read_lock(&tasklist_lock), which is released if return value is
1389 * non-zero. Also, grabs and releases @p->sighand->siglock.
1390 *
1391 * RETURNS:
1392 * 0 if wait condition didn't exist and search for other wait conditions
1393 * should continue. Non-zero return, -errno on failure and @p's pid on
1394 * success, implies that tasklist_lock is released and wait condition
1395 * search should terminate.
1396 */
1397static int wait_task_stopped(struct wait_opts *wo,
1398                int ptrace, struct task_struct *p)
1399{
1400    struct siginfo __user *infop;
1401    int retval, exit_code, *p_code, why;
1402    uid_t uid = 0; /* unneeded, required by compiler */
1403    pid_t pid;
1404
1405    /*
1406     * Traditionally we see ptrace'd stopped tasks regardless of options.
1407     */
1408    if (!ptrace && !(wo->wo_flags & WUNTRACED))
1409        return 0;
1410
1411    if (!task_stopped_code(p, ptrace))
1412        return 0;
1413
1414    exit_code = 0;
1415    spin_lock_irq(&p->sighand->siglock);
1416
1417    p_code = task_stopped_code(p, ptrace);
1418    if (unlikely(!p_code))
1419        goto unlock_sig;
1420
1421    exit_code = *p_code;
1422    if (!exit_code)
1423        goto unlock_sig;
1424
1425    if (!unlikely(wo->wo_flags & WNOWAIT))
1426        *p_code = 0;
1427
1428    uid = task_uid(p);
1429unlock_sig:
1430    spin_unlock_irq(&p->sighand->siglock);
1431    if (!exit_code)
1432        return 0;
1433
1434    /*
1435     * Now we are pretty sure this task is interesting.
1436     * Make sure it doesn't get reaped out from under us while we
1437     * give up the lock and then examine it below. We don't want to
1438     * keep holding onto the tasklist_lock while we call getrusage and
1439     * possibly take page faults for user memory.
1440     */
1441    get_task_struct(p);
1442    pid = task_pid_vnr(p);
1443    why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1444    read_unlock(&tasklist_lock);
1445
1446    if (unlikely(wo->wo_flags & WNOWAIT))
1447        return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1448
1449    retval = wo->wo_rusage
1450        ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1451    if (!retval && wo->wo_stat)
1452        retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1453
1454    infop = wo->wo_info;
1455    if (!retval && infop)
1456        retval = put_user(SIGCHLD, &infop->si_signo);
1457    if (!retval && infop)
1458        retval = put_user(0, &infop->si_errno);
1459    if (!retval && infop)
1460        retval = put_user((short)why, &infop->si_code);
1461    if (!retval && infop)
1462        retval = put_user(exit_code, &infop->si_status);
1463    if (!retval && infop)
1464        retval = put_user(pid, &infop->si_pid);
1465    if (!retval && infop)
1466        retval = put_user(uid, &infop->si_uid);
1467    if (!retval)
1468        retval = pid;
1469    put_task_struct(p);
1470
1471    BUG_ON(!retval);
1472    return retval;
1473}
1474
1475/*
1476 * Handle do_wait work for one task in a live, non-stopped state.
1477 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1478 * the lock and this task is uninteresting. If we return nonzero, we have
1479 * released the lock and the system call should return.
1480 */
1481static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1482{
1483    int retval;
1484    pid_t pid;
1485    uid_t uid;
1486
1487    if (!unlikely(wo->wo_flags & WCONTINUED))
1488        return 0;
1489
1490    if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1491        return 0;
1492
1493    spin_lock_irq(&p->sighand->siglock);
1494    /* Re-check with the lock held. */
1495    if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1496        spin_unlock_irq(&p->sighand->siglock);
1497        return 0;
1498    }
1499    if (!unlikely(wo->wo_flags & WNOWAIT))
1500        p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1501    uid = task_uid(p);
1502    spin_unlock_irq(&p->sighand->siglock);
1503
1504    pid = task_pid_vnr(p);
1505    get_task_struct(p);
1506    read_unlock(&tasklist_lock);
1507
1508    if (!wo->wo_info) {
1509        retval = wo->wo_rusage
1510            ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1511        put_task_struct(p);
1512        if (!retval && wo->wo_stat)
1513            retval = put_user(0xffff, wo->wo_stat);
1514        if (!retval)
1515            retval = pid;
1516    } else {
1517        retval = wait_noreap_copyout(wo, p, pid, uid,
1518                         CLD_CONTINUED, SIGCONT);
1519        BUG_ON(retval == 0);
1520    }
1521
1522    return retval;
1523}
1524
1525/*
1526 * Consider @p for a wait by @parent.
1527 *
1528 * -ECHILD should be in ->notask_error before the first call.
1529 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1530 * Returns zero if the search for a child should continue;
1531 * then ->notask_error is 0 if @p is an eligible child,
1532 * or another error from security_task_wait(), or still -ECHILD.
1533 */
1534static int wait_consider_task(struct wait_opts *wo, int ptrace,
1535                struct task_struct *p)
1536{
1537    int ret = eligible_child(wo, p);
1538    if (!ret)
1539        return ret;
1540
1541    ret = security_task_wait(p);
1542    if (unlikely(ret < 0)) {
1543        /*
1544         * If we have not yet seen any eligible child,
1545         * then let this error code replace -ECHILD.
1546         * A permission error will give the user a clue
1547         * to look for security policy problems, rather
1548         * than for mysterious wait bugs.
1549         */
1550        if (wo->notask_error)
1551            wo->notask_error = ret;
1552        return 0;
1553    }
1554
1555    /* dead body doesn't have much to contribute */
1556    if (p->exit_state == EXIT_DEAD)
1557        return 0;
1558
1559    /* slay zombie? */
1560    if (p->exit_state == EXIT_ZOMBIE) {
1561        /*
1562         * A zombie ptracee is only visible to its ptracer.
1563         * Notification and reaping will be cascaded to the real
1564         * parent when the ptracer detaches.
1565         */
1566        if (likely(!ptrace) && unlikely(task_ptrace(p))) {
1567            /* it will become visible, clear notask_error */
1568            wo->notask_error = 0;
1569            return 0;
1570        }
1571
1572        /* we don't reap group leaders with subthreads */
1573        if (!delay_group_leader(p))
1574            return wait_task_zombie(wo, p);
1575
1576        /*
1577         * Allow access to stopped/continued state via zombie by
1578         * falling through. Clearing of notask_error is complex.
1579         *
1580         * When !@ptrace:
1581         *
1582         * If WEXITED is set, notask_error should naturally be
1583         * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1584         * so, if there are live subthreads, there are events to
1585         * wait for. If all subthreads are dead, it's still safe
1586         * to clear - this function will be called again in finite
1587         * amount time once all the subthreads are released and
1588         * will then return without clearing.
1589         *
1590         * When @ptrace:
1591         *
1592         * Stopped state is per-task and thus can't change once the
1593         * target task dies. Only continued and exited can happen.
1594         * Clear notask_error if WCONTINUED | WEXITED.
1595         */
1596        if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1597            wo->notask_error = 0;
1598    } else {
1599        /*
1600         * If @p is ptraced by a task in its real parent's group,
1601         * hide group stop/continued state when looking at @p as
1602         * the real parent; otherwise, a single stop can be
1603         * reported twice as group and ptrace stops.
1604         *
1605         * If a ptracer wants to distinguish the two events for its
1606         * own children, it should create a separate process which
1607         * takes the role of real parent.
1608         */
1609        if (likely(!ptrace) && task_ptrace(p) &&
1610            same_thread_group(p->parent, p->real_parent))
1611            return 0;
1612
1613        /*
1614         * @p is alive and it's gonna stop, continue or exit, so
1615         * there always is something to wait for.
1616         */
1617        wo->notask_error = 0;
1618    }
1619
1620    /*
1621     * Wait for stopped. Depending on @ptrace, different stopped state
1622     * is used and the two don't interact with each other.
1623     */
1624    ret = wait_task_stopped(wo, ptrace, p);
1625    if (ret)
1626        return ret;
1627
1628    /*
1629     * Wait for continued. There's only one continued state and the
1630     * ptracer can consume it which can confuse the real parent. Don't
1631     * use WCONTINUED from ptracer. You don't need or want it.
1632     */
1633    return wait_task_continued(wo, p);
1634}
1635
1636/*
1637 * Do the work of do_wait() for one thread in the group, @tsk.
1638 *
1639 * -ECHILD should be in ->notask_error before the first call.
1640 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1641 * Returns zero if the search for a child should continue; then
1642 * ->notask_error is 0 if there were any eligible children,
1643 * or another error from security_task_wait(), or still -ECHILD.
1644 */
1645static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1646{
1647    struct task_struct *p;
1648
1649    list_for_each_entry(p, &tsk->children, sibling) {
1650        int ret = wait_consider_task(wo, 0, p);
1651        if (ret)
1652            return ret;
1653    }
1654
1655    return 0;
1656}
1657
1658static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1659{
1660    struct task_struct *p;
1661
1662    list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1663        int ret = wait_consider_task(wo, 1, p);
1664        if (ret)
1665            return ret;
1666    }
1667
1668    return 0;
1669}
1670
1671static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1672                int sync, void *key)
1673{
1674    struct wait_opts *wo = container_of(wait, struct wait_opts,
1675                        child_wait);
1676    struct task_struct *p = key;
1677
1678    if (!eligible_pid(wo, p))
1679        return 0;
1680
1681    if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1682        return 0;
1683
1684    return default_wake_function(wait, mode, sync, key);
1685}
1686
1687void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1688{
1689    __wake_up_sync_key(&parent->signal->wait_chldexit,
1690                TASK_INTERRUPTIBLE, 1, p);
1691}
1692
1693static long do_wait(struct wait_opts *wo)
1694{
1695    struct task_struct *tsk;
1696    int retval;
1697
1698    trace_sched_process_wait(wo->wo_pid);
1699
1700    init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1701    wo->child_wait.private = current;
1702    add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1703repeat:
1704    /*
1705     * If there is nothing that can match our critiera just get out.
1706     * We will clear ->notask_error to zero if we see any child that
1707     * might later match our criteria, even if we are not able to reap
1708     * it yet.
1709     */
1710    wo->notask_error = -ECHILD;
1711    if ((wo->wo_type < PIDTYPE_MAX) &&
1712       (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1713        goto notask;
1714
1715    set_current_state(TASK_INTERRUPTIBLE);
1716    read_lock(&tasklist_lock);
1717    tsk = current;
1718    do {
1719        retval = do_wait_thread(wo, tsk);
1720        if (retval)
1721            goto end;
1722
1723        retval = ptrace_do_wait(wo, tsk);
1724        if (retval)
1725            goto end;
1726
1727        if (wo->wo_flags & __WNOTHREAD)
1728            break;
1729    } while_each_thread(current, tsk);
1730    read_unlock(&tasklist_lock);
1731
1732notask:
1733    retval = wo->notask_error;
1734    if (!retval && !(wo->wo_flags & WNOHANG)) {
1735        retval = -ERESTARTSYS;
1736        if (!signal_pending(current)) {
1737            schedule();
1738            goto repeat;
1739        }
1740    }
1741end:
1742    __set_current_state(TASK_RUNNING);
1743    remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1744    return retval;
1745}
1746
1747SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1748        infop, int, options, struct rusage __user *, ru)
1749{
1750    struct wait_opts wo;
1751    struct pid *pid = NULL;
1752    enum pid_type type;
1753    long ret;
1754
1755    if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1756        return -EINVAL;
1757    if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1758        return -EINVAL;
1759
1760    switch (which) {
1761    case P_ALL:
1762        type = PIDTYPE_MAX;
1763        break;
1764    case P_PID:
1765        type = PIDTYPE_PID;
1766        if (upid <= 0)
1767            return -EINVAL;
1768        break;
1769    case P_PGID:
1770        type = PIDTYPE_PGID;
1771        if (upid <= 0)
1772            return -EINVAL;
1773        break;
1774    default:
1775        return -EINVAL;
1776    }
1777
1778    if (type < PIDTYPE_MAX)
1779        pid = find_get_pid(upid);
1780
1781    wo.wo_type = type;
1782    wo.wo_pid = pid;
1783    wo.wo_flags = options;
1784    wo.wo_info = infop;
1785    wo.wo_stat = NULL;
1786    wo.wo_rusage = ru;
1787    ret = do_wait(&wo);
1788
1789    if (ret > 0) {
1790        ret = 0;
1791    } else if (infop) {
1792        /*
1793         * For a WNOHANG return, clear out all the fields
1794         * we would set so the user can easily tell the
1795         * difference.
1796         */
1797        if (!ret)
1798            ret = put_user(0, &infop->si_signo);
1799        if (!ret)
1800            ret = put_user(0, &infop->si_errno);
1801        if (!ret)
1802            ret = put_user(0, &infop->si_code);
1803        if (!ret)
1804            ret = put_user(0, &infop->si_pid);
1805        if (!ret)
1806            ret = put_user(0, &infop->si_uid);
1807        if (!ret)
1808            ret = put_user(0, &infop->si_status);
1809    }
1810
1811    put_pid(pid);
1812
1813    /* avoid REGPARM breakage on x86: */
1814    asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1815    return ret;
1816}
1817
1818SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1819        int, options, struct rusage __user *, ru)
1820{
1821    struct wait_opts wo;
1822    struct pid *pid = NULL;
1823    enum pid_type type;
1824    long ret;
1825
1826    if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1827            __WNOTHREAD|__WCLONE|__WALL))
1828        return -EINVAL;
1829
1830    if (upid == -1)
1831        type = PIDTYPE_MAX;
1832    else if (upid < 0) {
1833        type = PIDTYPE_PGID;
1834        pid = find_get_pid(-upid);
1835    } else if (upid == 0) {
1836        type = PIDTYPE_PGID;
1837        pid = get_task_pid(current, PIDTYPE_PGID);
1838    } else /* upid > 0 */ {
1839        type = PIDTYPE_PID;
1840        pid = find_get_pid(upid);
1841    }
1842
1843    wo.wo_type = type;
1844    wo.wo_pid = pid;
1845    wo.wo_flags = options | WEXITED;
1846    wo.wo_info = NULL;
1847    wo.wo_stat = stat_addr;
1848    wo.wo_rusage = ru;
1849    ret = do_wait(&wo);
1850    put_pid(pid);
1851
1852    /* avoid REGPARM breakage on x86: */
1853    asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1854    return ret;
1855}
1856
1857#ifdef __ARCH_WANT_SYS_WAITPID
1858
1859/*
1860 * sys_waitpid() remains for compatibility. waitpid() should be
1861 * implemented by calling sys_wait4() from libc.a.
1862 */
1863SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1864{
1865    return sys_wait4(pid, stat_addr, options, NULL);
1866}
1867
1868#endif
1869

Archive Download this file



interactive