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

Archive Download this file



interactive