Root/fs/namespace.c

1/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
11#include <linux/syscalls.h>
12#include <linux/slab.h>
13#include <linux/sched.h>
14#include <linux/smp_lock.h>
15#include <linux/init.h>
16#include <linux/kernel.h>
17#include <linux/acct.h>
18#include <linux/capability.h>
19#include <linux/cpumask.h>
20#include <linux/module.h>
21#include <linux/sysfs.h>
22#include <linux/seq_file.h>
23#include <linux/mnt_namespace.h>
24#include <linux/namei.h>
25#include <linux/nsproxy.h>
26#include <linux/security.h>
27#include <linux/mount.h>
28#include <linux/ramfs.h>
29#include <linux/log2.h>
30#include <linux/idr.h>
31#include <linux/fs_struct.h>
32#include <asm/uaccess.h>
33#include <asm/unistd.h>
34#include "pnode.h"
35#include "internal.h"
36
37#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
38#define HASH_SIZE (1UL << HASH_SHIFT)
39
40/* spinlock for vfsmount related operations, inplace of dcache_lock */
41__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
42
43static int event;
44static DEFINE_IDA(mnt_id_ida);
45static DEFINE_IDA(mnt_group_ida);
46static int mnt_id_start = 0;
47static int mnt_group_start = 1;
48
49static struct list_head *mount_hashtable __read_mostly;
50static struct kmem_cache *mnt_cache __read_mostly;
51static struct rw_semaphore namespace_sem;
52
53/* /sys/fs */
54struct kobject *fs_kobj;
55EXPORT_SYMBOL_GPL(fs_kobj);
56
57static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
58{
59    unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
60    tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
61    tmp = tmp + (tmp >> HASH_SHIFT);
62    return tmp & (HASH_SIZE - 1);
63}
64
65#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
66
67/* allocation is serialized by namespace_sem */
68static int mnt_alloc_id(struct vfsmount *mnt)
69{
70    int res;
71
72retry:
73    ida_pre_get(&mnt_id_ida, GFP_KERNEL);
74    spin_lock(&vfsmount_lock);
75    res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
76    if (!res)
77        mnt_id_start = mnt->mnt_id + 1;
78    spin_unlock(&vfsmount_lock);
79    if (res == -EAGAIN)
80        goto retry;
81
82    return res;
83}
84
85static void mnt_free_id(struct vfsmount *mnt)
86{
87    int id = mnt->mnt_id;
88    spin_lock(&vfsmount_lock);
89    ida_remove(&mnt_id_ida, id);
90    if (mnt_id_start > id)
91        mnt_id_start = id;
92    spin_unlock(&vfsmount_lock);
93}
94
95/*
96 * Allocate a new peer group ID
97 *
98 * mnt_group_ida is protected by namespace_sem
99 */
100static int mnt_alloc_group_id(struct vfsmount *mnt)
101{
102    int res;
103
104    if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
105        return -ENOMEM;
106
107    res = ida_get_new_above(&mnt_group_ida,
108                mnt_group_start,
109                &mnt->mnt_group_id);
110    if (!res)
111        mnt_group_start = mnt->mnt_group_id + 1;
112
113    return res;
114}
115
116/*
117 * Release a peer group ID
118 */
119void mnt_release_group_id(struct vfsmount *mnt)
120{
121    int id = mnt->mnt_group_id;
122    ida_remove(&mnt_group_ida, id);
123    if (mnt_group_start > id)
124        mnt_group_start = id;
125    mnt->mnt_group_id = 0;
126}
127
128struct vfsmount *alloc_vfsmnt(const char *name)
129{
130    struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
131    if (mnt) {
132        int err;
133
134        err = mnt_alloc_id(mnt);
135        if (err)
136            goto out_free_cache;
137
138        if (name) {
139            mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
140            if (!mnt->mnt_devname)
141                goto out_free_id;
142        }
143
144        atomic_set(&mnt->mnt_count, 1);
145        INIT_LIST_HEAD(&mnt->mnt_hash);
146        INIT_LIST_HEAD(&mnt->mnt_child);
147        INIT_LIST_HEAD(&mnt->mnt_mounts);
148        INIT_LIST_HEAD(&mnt->mnt_list);
149        INIT_LIST_HEAD(&mnt->mnt_expire);
150        INIT_LIST_HEAD(&mnt->mnt_share);
151        INIT_LIST_HEAD(&mnt->mnt_slave_list);
152        INIT_LIST_HEAD(&mnt->mnt_slave);
153#ifdef CONFIG_SMP
154        mnt->mnt_writers = alloc_percpu(int);
155        if (!mnt->mnt_writers)
156            goto out_free_devname;
157#else
158        mnt->mnt_writers = 0;
159#endif
160    }
161    return mnt;
162
163#ifdef CONFIG_SMP
164out_free_devname:
165    kfree(mnt->mnt_devname);
166#endif
167out_free_id:
168    mnt_free_id(mnt);
169out_free_cache:
170    kmem_cache_free(mnt_cache, mnt);
171    return NULL;
172}
173
174/*
175 * Most r/o checks on a fs are for operations that take
176 * discrete amounts of time, like a write() or unlink().
177 * We must keep track of when those operations start
178 * (for permission checks) and when they end, so that
179 * we can determine when writes are able to occur to
180 * a filesystem.
181 */
182/*
183 * __mnt_is_readonly: check whether a mount is read-only
184 * @mnt: the mount to check for its write status
185 *
186 * This shouldn't be used directly ouside of the VFS.
187 * It does not guarantee that the filesystem will stay
188 * r/w, just that it is right *now*. This can not and
189 * should not be used in place of IS_RDONLY(inode).
190 * mnt_want/drop_write() will _keep_ the filesystem
191 * r/w.
192 */
193int __mnt_is_readonly(struct vfsmount *mnt)
194{
195    if (mnt->mnt_flags & MNT_READONLY)
196        return 1;
197    if (mnt->mnt_sb->s_flags & MS_RDONLY)
198        return 1;
199    return 0;
200}
201EXPORT_SYMBOL_GPL(__mnt_is_readonly);
202
203static inline void inc_mnt_writers(struct vfsmount *mnt)
204{
205#ifdef CONFIG_SMP
206    (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++;
207#else
208    mnt->mnt_writers++;
209#endif
210}
211
212static inline void dec_mnt_writers(struct vfsmount *mnt)
213{
214#ifdef CONFIG_SMP
215    (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--;
216#else
217    mnt->mnt_writers--;
218#endif
219}
220
221static unsigned int count_mnt_writers(struct vfsmount *mnt)
222{
223#ifdef CONFIG_SMP
224    unsigned int count = 0;
225    int cpu;
226
227    for_each_possible_cpu(cpu) {
228        count += *per_cpu_ptr(mnt->mnt_writers, cpu);
229    }
230
231    return count;
232#else
233    return mnt->mnt_writers;
234#endif
235}
236
237/*
238 * Most r/o checks on a fs are for operations that take
239 * discrete amounts of time, like a write() or unlink().
240 * We must keep track of when those operations start
241 * (for permission checks) and when they end, so that
242 * we can determine when writes are able to occur to
243 * a filesystem.
244 */
245/**
246 * mnt_want_write - get write access to a mount
247 * @mnt: the mount on which to take a write
248 *
249 * This tells the low-level filesystem that a write is
250 * about to be performed to it, and makes sure that
251 * writes are allowed before returning success. When
252 * the write operation is finished, mnt_drop_write()
253 * must be called. This is effectively a refcount.
254 */
255int mnt_want_write(struct vfsmount *mnt)
256{
257    int ret = 0;
258
259    preempt_disable();
260    inc_mnt_writers(mnt);
261    /*
262     * The store to inc_mnt_writers must be visible before we pass
263     * MNT_WRITE_HOLD loop below, so that the slowpath can see our
264     * incremented count after it has set MNT_WRITE_HOLD.
265     */
266    smp_mb();
267    while (mnt->mnt_flags & MNT_WRITE_HOLD)
268        cpu_relax();
269    /*
270     * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
271     * be set to match its requirements. So we must not load that until
272     * MNT_WRITE_HOLD is cleared.
273     */
274    smp_rmb();
275    if (__mnt_is_readonly(mnt)) {
276        dec_mnt_writers(mnt);
277        ret = -EROFS;
278        goto out;
279    }
280out:
281    preempt_enable();
282    return ret;
283}
284EXPORT_SYMBOL_GPL(mnt_want_write);
285
286/**
287 * mnt_clone_write - get write access to a mount
288 * @mnt: the mount on which to take a write
289 *
290 * This is effectively like mnt_want_write, except
291 * it must only be used to take an extra write reference
292 * on a mountpoint that we already know has a write reference
293 * on it. This allows some optimisation.
294 *
295 * After finished, mnt_drop_write must be called as usual to
296 * drop the reference.
297 */
298int mnt_clone_write(struct vfsmount *mnt)
299{
300    /* superblock may be r/o */
301    if (__mnt_is_readonly(mnt))
302        return -EROFS;
303    preempt_disable();
304    inc_mnt_writers(mnt);
305    preempt_enable();
306    return 0;
307}
308EXPORT_SYMBOL_GPL(mnt_clone_write);
309
310/**
311 * mnt_want_write_file - get write access to a file's mount
312 * @file: the file who's mount on which to take a write
313 *
314 * This is like mnt_want_write, but it takes a file and can
315 * do some optimisations if the file is open for write already
316 */
317int mnt_want_write_file(struct file *file)
318{
319    struct inode *inode = file->f_dentry->d_inode;
320    if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
321        return mnt_want_write(file->f_path.mnt);
322    else
323        return mnt_clone_write(file->f_path.mnt);
324}
325EXPORT_SYMBOL_GPL(mnt_want_write_file);
326
327/**
328 * mnt_drop_write - give up write access to a mount
329 * @mnt: the mount on which to give up write access
330 *
331 * Tells the low-level filesystem that we are done
332 * performing writes to it. Must be matched with
333 * mnt_want_write() call above.
334 */
335void mnt_drop_write(struct vfsmount *mnt)
336{
337    preempt_disable();
338    dec_mnt_writers(mnt);
339    preempt_enable();
340}
341EXPORT_SYMBOL_GPL(mnt_drop_write);
342
343static int mnt_make_readonly(struct vfsmount *mnt)
344{
345    int ret = 0;
346
347    spin_lock(&vfsmount_lock);
348    mnt->mnt_flags |= MNT_WRITE_HOLD;
349    /*
350     * After storing MNT_WRITE_HOLD, we'll read the counters. This store
351     * should be visible before we do.
352     */
353    smp_mb();
354
355    /*
356     * With writers on hold, if this value is zero, then there are
357     * definitely no active writers (although held writers may subsequently
358     * increment the count, they'll have to wait, and decrement it after
359     * seeing MNT_READONLY).
360     *
361     * It is OK to have counter incremented on one CPU and decremented on
362     * another: the sum will add up correctly. The danger would be when we
363     * sum up each counter, if we read a counter before it is incremented,
364     * but then read another CPU's count which it has been subsequently
365     * decremented from -- we would see more decrements than we should.
366     * MNT_WRITE_HOLD protects against this scenario, because
367     * mnt_want_write first increments count, then smp_mb, then spins on
368     * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
369     * we're counting up here.
370     */
371    if (count_mnt_writers(mnt) > 0)
372        ret = -EBUSY;
373    else
374        mnt->mnt_flags |= MNT_READONLY;
375    /*
376     * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
377     * that become unheld will see MNT_READONLY.
378     */
379    smp_wmb();
380    mnt->mnt_flags &= ~MNT_WRITE_HOLD;
381    spin_unlock(&vfsmount_lock);
382    return ret;
383}
384
385static void __mnt_unmake_readonly(struct vfsmount *mnt)
386{
387    spin_lock(&vfsmount_lock);
388    mnt->mnt_flags &= ~MNT_READONLY;
389    spin_unlock(&vfsmount_lock);
390}
391
392void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
393{
394    mnt->mnt_sb = sb;
395    mnt->mnt_root = dget(sb->s_root);
396}
397
398EXPORT_SYMBOL(simple_set_mnt);
399
400void free_vfsmnt(struct vfsmount *mnt)
401{
402    kfree(mnt->mnt_devname);
403    mnt_free_id(mnt);
404#ifdef CONFIG_SMP
405    free_percpu(mnt->mnt_writers);
406#endif
407    kmem_cache_free(mnt_cache, mnt);
408}
409
410/*
411 * find the first or last mount at @dentry on vfsmount @mnt depending on
412 * @dir. If @dir is set return the first mount else return the last mount.
413 */
414struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
415                  int dir)
416{
417    struct list_head *head = mount_hashtable + hash(mnt, dentry);
418    struct list_head *tmp = head;
419    struct vfsmount *p, *found = NULL;
420
421    for (;;) {
422        tmp = dir ? tmp->next : tmp->prev;
423        p = NULL;
424        if (tmp == head)
425            break;
426        p = list_entry(tmp, struct vfsmount, mnt_hash);
427        if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
428            found = p;
429            break;
430        }
431    }
432    return found;
433}
434
435/*
436 * lookup_mnt increments the ref count before returning
437 * the vfsmount struct.
438 */
439struct vfsmount *lookup_mnt(struct path *path)
440{
441    struct vfsmount *child_mnt;
442    spin_lock(&vfsmount_lock);
443    if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
444        mntget(child_mnt);
445    spin_unlock(&vfsmount_lock);
446    return child_mnt;
447}
448
449static inline int check_mnt(struct vfsmount *mnt)
450{
451    return mnt->mnt_ns == current->nsproxy->mnt_ns;
452}
453
454static void touch_mnt_namespace(struct mnt_namespace *ns)
455{
456    if (ns) {
457        ns->event = ++event;
458        wake_up_interruptible(&ns->poll);
459    }
460}
461
462static void __touch_mnt_namespace(struct mnt_namespace *ns)
463{
464    if (ns && ns->event != event) {
465        ns->event = event;
466        wake_up_interruptible(&ns->poll);
467    }
468}
469
470static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
471{
472    old_path->dentry = mnt->mnt_mountpoint;
473    old_path->mnt = mnt->mnt_parent;
474    mnt->mnt_parent = mnt;
475    mnt->mnt_mountpoint = mnt->mnt_root;
476    list_del_init(&mnt->mnt_child);
477    list_del_init(&mnt->mnt_hash);
478    old_path->dentry->d_mounted--;
479}
480
481void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
482            struct vfsmount *child_mnt)
483{
484    child_mnt->mnt_parent = mntget(mnt);
485    child_mnt->mnt_mountpoint = dget(dentry);
486    dentry->d_mounted++;
487}
488
489static void attach_mnt(struct vfsmount *mnt, struct path *path)
490{
491    mnt_set_mountpoint(path->mnt, path->dentry, mnt);
492    list_add_tail(&mnt->mnt_hash, mount_hashtable +
493            hash(path->mnt, path->dentry));
494    list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
495}
496
497/*
498 * the caller must hold vfsmount_lock
499 */
500static void commit_tree(struct vfsmount *mnt)
501{
502    struct vfsmount *parent = mnt->mnt_parent;
503    struct vfsmount *m;
504    LIST_HEAD(head);
505    struct mnt_namespace *n = parent->mnt_ns;
506
507    BUG_ON(parent == mnt);
508
509    list_add_tail(&head, &mnt->mnt_list);
510    list_for_each_entry(m, &head, mnt_list)
511        m->mnt_ns = n;
512    list_splice(&head, n->list.prev);
513
514    list_add_tail(&mnt->mnt_hash, mount_hashtable +
515                hash(parent, mnt->mnt_mountpoint));
516    list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
517    touch_mnt_namespace(n);
518}
519
520static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
521{
522    struct list_head *next = p->mnt_mounts.next;
523    if (next == &p->mnt_mounts) {
524        while (1) {
525            if (p == root)
526                return NULL;
527            next = p->mnt_child.next;
528            if (next != &p->mnt_parent->mnt_mounts)
529                break;
530            p = p->mnt_parent;
531        }
532    }
533    return list_entry(next, struct vfsmount, mnt_child);
534}
535
536static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
537{
538    struct list_head *prev = p->mnt_mounts.prev;
539    while (prev != &p->mnt_mounts) {
540        p = list_entry(prev, struct vfsmount, mnt_child);
541        prev = p->mnt_mounts.prev;
542    }
543    return p;
544}
545
546static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
547                    int flag)
548{
549    struct super_block *sb = old->mnt_sb;
550    struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
551
552    if (mnt) {
553        if (flag & (CL_SLAVE | CL_PRIVATE))
554            mnt->mnt_group_id = 0; /* not a peer of original */
555        else
556            mnt->mnt_group_id = old->mnt_group_id;
557
558        if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
559            int err = mnt_alloc_group_id(mnt);
560            if (err)
561                goto out_free;
562        }
563
564        mnt->mnt_flags = old->mnt_flags;
565        atomic_inc(&sb->s_active);
566        mnt->mnt_sb = sb;
567        mnt->mnt_root = dget(root);
568        mnt->mnt_mountpoint = mnt->mnt_root;
569        mnt->mnt_parent = mnt;
570
571        if (flag & CL_SLAVE) {
572            list_add(&mnt->mnt_slave, &old->mnt_slave_list);
573            mnt->mnt_master = old;
574            CLEAR_MNT_SHARED(mnt);
575        } else if (!(flag & CL_PRIVATE)) {
576            if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
577                list_add(&mnt->mnt_share, &old->mnt_share);
578            if (IS_MNT_SLAVE(old))
579                list_add(&mnt->mnt_slave, &old->mnt_slave);
580            mnt->mnt_master = old->mnt_master;
581        }
582        if (flag & CL_MAKE_SHARED)
583            set_mnt_shared(mnt);
584
585        /* stick the duplicate mount on the same expiry list
586         * as the original if that was on one */
587        if (flag & CL_EXPIRE) {
588            if (!list_empty(&old->mnt_expire))
589                list_add(&mnt->mnt_expire, &old->mnt_expire);
590        }
591    }
592    return mnt;
593
594 out_free:
595    free_vfsmnt(mnt);
596    return NULL;
597}
598
599static inline void __mntput(struct vfsmount *mnt)
600{
601    struct super_block *sb = mnt->mnt_sb;
602    /*
603     * This probably indicates that somebody messed
604     * up a mnt_want/drop_write() pair. If this
605     * happens, the filesystem was probably unable
606     * to make r/w->r/o transitions.
607     */
608    /*
609     * atomic_dec_and_lock() used to deal with ->mnt_count decrements
610     * provides barriers, so count_mnt_writers() below is safe. AV
611     */
612    WARN_ON(count_mnt_writers(mnt));
613    dput(mnt->mnt_root);
614    free_vfsmnt(mnt);
615    deactivate_super(sb);
616}
617
618void mntput_no_expire(struct vfsmount *mnt)
619{
620repeat:
621    if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
622        if (likely(!mnt->mnt_pinned)) {
623            spin_unlock(&vfsmount_lock);
624            __mntput(mnt);
625            return;
626        }
627        atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
628        mnt->mnt_pinned = 0;
629        spin_unlock(&vfsmount_lock);
630        acct_auto_close_mnt(mnt);
631        security_sb_umount_close(mnt);
632        goto repeat;
633    }
634}
635
636EXPORT_SYMBOL(mntput_no_expire);
637
638void mnt_pin(struct vfsmount *mnt)
639{
640    spin_lock(&vfsmount_lock);
641    mnt->mnt_pinned++;
642    spin_unlock(&vfsmount_lock);
643}
644
645EXPORT_SYMBOL(mnt_pin);
646
647void mnt_unpin(struct vfsmount *mnt)
648{
649    spin_lock(&vfsmount_lock);
650    if (mnt->mnt_pinned) {
651        atomic_inc(&mnt->mnt_count);
652        mnt->mnt_pinned--;
653    }
654    spin_unlock(&vfsmount_lock);
655}
656
657EXPORT_SYMBOL(mnt_unpin);
658
659static inline void mangle(struct seq_file *m, const char *s)
660{
661    seq_escape(m, s, " \t\n\\");
662}
663
664/*
665 * Simple .show_options callback for filesystems which don't want to
666 * implement more complex mount option showing.
667 *
668 * See also save_mount_options().
669 */
670int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
671{
672    const char *options;
673
674    rcu_read_lock();
675    options = rcu_dereference(mnt->mnt_sb->s_options);
676
677    if (options != NULL && options[0]) {
678        seq_putc(m, ',');
679        mangle(m, options);
680    }
681    rcu_read_unlock();
682
683    return 0;
684}
685EXPORT_SYMBOL(generic_show_options);
686
687/*
688 * If filesystem uses generic_show_options(), this function should be
689 * called from the fill_super() callback.
690 *
691 * The .remount_fs callback usually needs to be handled in a special
692 * way, to make sure, that previous options are not overwritten if the
693 * remount fails.
694 *
695 * Also note, that if the filesystem's .remount_fs function doesn't
696 * reset all options to their default value, but changes only newly
697 * given options, then the displayed options will not reflect reality
698 * any more.
699 */
700void save_mount_options(struct super_block *sb, char *options)
701{
702    BUG_ON(sb->s_options);
703    rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
704}
705EXPORT_SYMBOL(save_mount_options);
706
707void replace_mount_options(struct super_block *sb, char *options)
708{
709    char *old = sb->s_options;
710    rcu_assign_pointer(sb->s_options, options);
711    if (old) {
712        synchronize_rcu();
713        kfree(old);
714    }
715}
716EXPORT_SYMBOL(replace_mount_options);
717
718#ifdef CONFIG_PROC_FS
719/* iterator */
720static void *m_start(struct seq_file *m, loff_t *pos)
721{
722    struct proc_mounts *p = m->private;
723
724    down_read(&namespace_sem);
725    return seq_list_start(&p->ns->list, *pos);
726}
727
728static void *m_next(struct seq_file *m, void *v, loff_t *pos)
729{
730    struct proc_mounts *p = m->private;
731
732    return seq_list_next(v, &p->ns->list, pos);
733}
734
735static void m_stop(struct seq_file *m, void *v)
736{
737    up_read(&namespace_sem);
738}
739
740struct proc_fs_info {
741    int flag;
742    const char *str;
743};
744
745static int show_sb_opts(struct seq_file *m, struct super_block *sb)
746{
747    static const struct proc_fs_info fs_info[] = {
748        { MS_SYNCHRONOUS, ",sync" },
749        { MS_DIRSYNC, ",dirsync" },
750        { MS_MANDLOCK, ",mand" },
751        { 0, NULL }
752    };
753    const struct proc_fs_info *fs_infop;
754
755    for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
756        if (sb->s_flags & fs_infop->flag)
757            seq_puts(m, fs_infop->str);
758    }
759
760    return security_sb_show_options(m, sb);
761}
762
763static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
764{
765    static const struct proc_fs_info mnt_info[] = {
766        { MNT_NOSUID, ",nosuid" },
767        { MNT_NODEV, ",nodev" },
768        { MNT_NOEXEC, ",noexec" },
769        { MNT_NOATIME, ",noatime" },
770        { MNT_NODIRATIME, ",nodiratime" },
771        { MNT_RELATIME, ",relatime" },
772        { MNT_STRICTATIME, ",strictatime" },
773        { 0, NULL }
774    };
775    const struct proc_fs_info *fs_infop;
776
777    for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
778        if (mnt->mnt_flags & fs_infop->flag)
779            seq_puts(m, fs_infop->str);
780    }
781}
782
783static void show_type(struct seq_file *m, struct super_block *sb)
784{
785    mangle(m, sb->s_type->name);
786    if (sb->s_subtype && sb->s_subtype[0]) {
787        seq_putc(m, '.');
788        mangle(m, sb->s_subtype);
789    }
790}
791
792static int show_vfsmnt(struct seq_file *m, void *v)
793{
794    struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
795    int err = 0;
796    struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
797
798    mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
799    seq_putc(m, ' ');
800    seq_path(m, &mnt_path, " \t\n\\");
801    seq_putc(m, ' ');
802    show_type(m, mnt->mnt_sb);
803    seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
804    err = show_sb_opts(m, mnt->mnt_sb);
805    if (err)
806        goto out;
807    show_mnt_opts(m, mnt);
808    if (mnt->mnt_sb->s_op->show_options)
809        err = mnt->mnt_sb->s_op->show_options(m, mnt);
810    seq_puts(m, " 0 0\n");
811out:
812    return err;
813}
814
815const struct seq_operations mounts_op = {
816    .start = m_start,
817    .next = m_next,
818    .stop = m_stop,
819    .show = show_vfsmnt
820};
821
822static int show_mountinfo(struct seq_file *m, void *v)
823{
824    struct proc_mounts *p = m->private;
825    struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
826    struct super_block *sb = mnt->mnt_sb;
827    struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
828    struct path root = p->root;
829    int err = 0;
830
831    seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
832           MAJOR(sb->s_dev), MINOR(sb->s_dev));
833    seq_dentry(m, mnt->mnt_root, " \t\n\\");
834    seq_putc(m, ' ');
835    seq_path_root(m, &mnt_path, &root, " \t\n\\");
836    if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
837        /*
838         * Mountpoint is outside root, discard that one. Ugly,
839         * but less so than trying to do that in iterator in a
840         * race-free way (due to renames).
841         */
842        return SEQ_SKIP;
843    }
844    seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
845    show_mnt_opts(m, mnt);
846
847    /* Tagged fields ("foo:X" or "bar") */
848    if (IS_MNT_SHARED(mnt))
849        seq_printf(m, " shared:%i", mnt->mnt_group_id);
850    if (IS_MNT_SLAVE(mnt)) {
851        int master = mnt->mnt_master->mnt_group_id;
852        int dom = get_dominating_id(mnt, &p->root);
853        seq_printf(m, " master:%i", master);
854        if (dom && dom != master)
855            seq_printf(m, " propagate_from:%i", dom);
856    }
857    if (IS_MNT_UNBINDABLE(mnt))
858        seq_puts(m, " unbindable");
859
860    /* Filesystem specific data */
861    seq_puts(m, " - ");
862    show_type(m, sb);
863    seq_putc(m, ' ');
864    mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
865    seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
866    err = show_sb_opts(m, sb);
867    if (err)
868        goto out;
869    if (sb->s_op->show_options)
870        err = sb->s_op->show_options(m, mnt);
871    seq_putc(m, '\n');
872out:
873    return err;
874}
875
876const struct seq_operations mountinfo_op = {
877    .start = m_start,
878    .next = m_next,
879    .stop = m_stop,
880    .show = show_mountinfo,
881};
882
883static int show_vfsstat(struct seq_file *m, void *v)
884{
885    struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
886    struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
887    int err = 0;
888
889    /* device */
890    if (mnt->mnt_devname) {
891        seq_puts(m, "device ");
892        mangle(m, mnt->mnt_devname);
893    } else
894        seq_puts(m, "no device");
895
896    /* mount point */
897    seq_puts(m, " mounted on ");
898    seq_path(m, &mnt_path, " \t\n\\");
899    seq_putc(m, ' ');
900
901    /* file system type */
902    seq_puts(m, "with fstype ");
903    show_type(m, mnt->mnt_sb);
904
905    /* optional statistics */
906    if (mnt->mnt_sb->s_op->show_stats) {
907        seq_putc(m, ' ');
908        err = mnt->mnt_sb->s_op->show_stats(m, mnt);
909    }
910
911    seq_putc(m, '\n');
912    return err;
913}
914
915const struct seq_operations mountstats_op = {
916    .start = m_start,
917    .next = m_next,
918    .stop = m_stop,
919    .show = show_vfsstat,
920};
921#endif /* CONFIG_PROC_FS */
922
923/**
924 * may_umount_tree - check if a mount tree is busy
925 * @mnt: root of mount tree
926 *
927 * This is called to check if a tree of mounts has any
928 * open files, pwds, chroots or sub mounts that are
929 * busy.
930 */
931int may_umount_tree(struct vfsmount *mnt)
932{
933    int actual_refs = 0;
934    int minimum_refs = 0;
935    struct vfsmount *p;
936
937    spin_lock(&vfsmount_lock);
938    for (p = mnt; p; p = next_mnt(p, mnt)) {
939        actual_refs += atomic_read(&p->mnt_count);
940        minimum_refs += 2;
941    }
942    spin_unlock(&vfsmount_lock);
943
944    if (actual_refs > minimum_refs)
945        return 0;
946
947    return 1;
948}
949
950EXPORT_SYMBOL(may_umount_tree);
951
952/**
953 * may_umount - check if a mount point is busy
954 * @mnt: root of mount
955 *
956 * This is called to check if a mount point has any
957 * open files, pwds, chroots or sub mounts. If the
958 * mount has sub mounts this will return busy
959 * regardless of whether the sub mounts are busy.
960 *
961 * Doesn't take quota and stuff into account. IOW, in some cases it will
962 * give false negatives. The main reason why it's here is that we need
963 * a non-destructive way to look for easily umountable filesystems.
964 */
965int may_umount(struct vfsmount *mnt)
966{
967    int ret = 1;
968    spin_lock(&vfsmount_lock);
969    if (propagate_mount_busy(mnt, 2))
970        ret = 0;
971    spin_unlock(&vfsmount_lock);
972    return ret;
973}
974
975EXPORT_SYMBOL(may_umount);
976
977void release_mounts(struct list_head *head)
978{
979    struct vfsmount *mnt;
980    while (!list_empty(head)) {
981        mnt = list_first_entry(head, struct vfsmount, mnt_hash);
982        list_del_init(&mnt->mnt_hash);
983        if (mnt->mnt_parent != mnt) {
984            struct dentry *dentry;
985            struct vfsmount *m;
986            spin_lock(&vfsmount_lock);
987            dentry = mnt->mnt_mountpoint;
988            m = mnt->mnt_parent;
989            mnt->mnt_mountpoint = mnt->mnt_root;
990            mnt->mnt_parent = mnt;
991            m->mnt_ghosts--;
992            spin_unlock(&vfsmount_lock);
993            dput(dentry);
994            mntput(m);
995        }
996        mntput(mnt);
997    }
998}
999
1000void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1001{
1002    struct vfsmount *p;
1003
1004    for (p = mnt; p; p = next_mnt(p, mnt))
1005        list_move(&p->mnt_hash, kill);
1006
1007    if (propagate)
1008        propagate_umount(kill);
1009
1010    list_for_each_entry(p, kill, mnt_hash) {
1011        list_del_init(&p->mnt_expire);
1012        list_del_init(&p->mnt_list);
1013        __touch_mnt_namespace(p->mnt_ns);
1014        p->mnt_ns = NULL;
1015        list_del_init(&p->mnt_child);
1016        if (p->mnt_parent != p) {
1017            p->mnt_parent->mnt_ghosts++;
1018            p->mnt_mountpoint->d_mounted--;
1019        }
1020        change_mnt_propagation(p, MS_PRIVATE);
1021    }
1022}
1023
1024static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1025
1026static int do_umount(struct vfsmount *mnt, int flags)
1027{
1028    struct super_block *sb = mnt->mnt_sb;
1029    int retval;
1030    LIST_HEAD(umount_list);
1031
1032    retval = security_sb_umount(mnt, flags);
1033    if (retval)
1034        return retval;
1035
1036    /*
1037     * Allow userspace to request a mountpoint be expired rather than
1038     * unmounting unconditionally. Unmount only happens if:
1039     * (1) the mark is already set (the mark is cleared by mntput())
1040     * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1041     */
1042    if (flags & MNT_EXPIRE) {
1043        if (mnt == current->fs->root.mnt ||
1044            flags & (MNT_FORCE | MNT_DETACH))
1045            return -EINVAL;
1046
1047        if (atomic_read(&mnt->mnt_count) != 2)
1048            return -EBUSY;
1049
1050        if (!xchg(&mnt->mnt_expiry_mark, 1))
1051            return -EAGAIN;
1052    }
1053
1054    /*
1055     * If we may have to abort operations to get out of this
1056     * mount, and they will themselves hold resources we must
1057     * allow the fs to do things. In the Unix tradition of
1058     * 'Gee thats tricky lets do it in userspace' the umount_begin
1059     * might fail to complete on the first run through as other tasks
1060     * must return, and the like. Thats for the mount program to worry
1061     * about for the moment.
1062     */
1063
1064    if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1065        sb->s_op->umount_begin(sb);
1066    }
1067
1068    /*
1069     * No sense to grab the lock for this test, but test itself looks
1070     * somewhat bogus. Suggestions for better replacement?
1071     * Ho-hum... In principle, we might treat that as umount + switch
1072     * to rootfs. GC would eventually take care of the old vfsmount.
1073     * Actually it makes sense, especially if rootfs would contain a
1074     * /reboot - static binary that would close all descriptors and
1075     * call reboot(9). Then init(8) could umount root and exec /reboot.
1076     */
1077    if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1078        /*
1079         * Special case for "unmounting" root ...
1080         * we just try to remount it readonly.
1081         */
1082        down_write(&sb->s_umount);
1083        if (!(sb->s_flags & MS_RDONLY))
1084            retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1085        up_write(&sb->s_umount);
1086        return retval;
1087    }
1088
1089    down_write(&namespace_sem);
1090    spin_lock(&vfsmount_lock);
1091    event++;
1092
1093    if (!(flags & MNT_DETACH))
1094        shrink_submounts(mnt, &umount_list);
1095
1096    retval = -EBUSY;
1097    if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1098        if (!list_empty(&mnt->mnt_list))
1099            umount_tree(mnt, 1, &umount_list);
1100        retval = 0;
1101    }
1102    spin_unlock(&vfsmount_lock);
1103    if (retval)
1104        security_sb_umount_busy(mnt);
1105    up_write(&namespace_sem);
1106    release_mounts(&umount_list);
1107    return retval;
1108}
1109
1110/*
1111 * Now umount can handle mount points as well as block devices.
1112 * This is important for filesystems which use unnamed block devices.
1113 *
1114 * We now support a flag for forced unmount like the other 'big iron'
1115 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1116 */
1117
1118SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1119{
1120    struct path path;
1121    int retval;
1122
1123    retval = user_path(name, &path);
1124    if (retval)
1125        goto out;
1126    retval = -EINVAL;
1127    if (path.dentry != path.mnt->mnt_root)
1128        goto dput_and_out;
1129    if (!check_mnt(path.mnt))
1130        goto dput_and_out;
1131
1132    retval = -EPERM;
1133    if (!capable(CAP_SYS_ADMIN))
1134        goto dput_and_out;
1135
1136    retval = do_umount(path.mnt, flags);
1137dput_and_out:
1138    /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1139    dput(path.dentry);
1140    mntput_no_expire(path.mnt);
1141out:
1142    return retval;
1143}
1144
1145#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1146
1147/*
1148 * The 2.0 compatible umount. No flags.
1149 */
1150SYSCALL_DEFINE1(oldumount, char __user *, name)
1151{
1152    return sys_umount(name, 0);
1153}
1154
1155#endif
1156
1157static int mount_is_safe(struct path *path)
1158{
1159    if (capable(CAP_SYS_ADMIN))
1160        return 0;
1161    return -EPERM;
1162#ifdef notyet
1163    if (S_ISLNK(path->dentry->d_inode->i_mode))
1164        return -EPERM;
1165    if (path->dentry->d_inode->i_mode & S_ISVTX) {
1166        if (current_uid() != path->dentry->d_inode->i_uid)
1167            return -EPERM;
1168    }
1169    if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1170        return -EPERM;
1171    return 0;
1172#endif
1173}
1174
1175struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1176                    int flag)
1177{
1178    struct vfsmount *res, *p, *q, *r, *s;
1179    struct path path;
1180
1181    if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1182        return NULL;
1183
1184    res = q = clone_mnt(mnt, dentry, flag);
1185    if (!q)
1186        goto Enomem;
1187    q->mnt_mountpoint = mnt->mnt_mountpoint;
1188
1189    p = mnt;
1190    list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1191        if (!is_subdir(r->mnt_mountpoint, dentry))
1192            continue;
1193
1194        for (s = r; s; s = next_mnt(s, r)) {
1195            if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1196                s = skip_mnt_tree(s);
1197                continue;
1198            }
1199            while (p != s->mnt_parent) {
1200                p = p->mnt_parent;
1201                q = q->mnt_parent;
1202            }
1203            p = s;
1204            path.mnt = q;
1205            path.dentry = p->mnt_mountpoint;
1206            q = clone_mnt(p, p->mnt_root, flag);
1207            if (!q)
1208                goto Enomem;
1209            spin_lock(&vfsmount_lock);
1210            list_add_tail(&q->mnt_list, &res->mnt_list);
1211            attach_mnt(q, &path);
1212            spin_unlock(&vfsmount_lock);
1213        }
1214    }
1215    return res;
1216Enomem:
1217    if (res) {
1218        LIST_HEAD(umount_list);
1219        spin_lock(&vfsmount_lock);
1220        umount_tree(res, 0, &umount_list);
1221        spin_unlock(&vfsmount_lock);
1222        release_mounts(&umount_list);
1223    }
1224    return NULL;
1225}
1226
1227struct vfsmount *collect_mounts(struct path *path)
1228{
1229    struct vfsmount *tree;
1230    down_write(&namespace_sem);
1231    tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1232    up_write(&namespace_sem);
1233    return tree;
1234}
1235
1236void drop_collected_mounts(struct vfsmount *mnt)
1237{
1238    LIST_HEAD(umount_list);
1239    down_write(&namespace_sem);
1240    spin_lock(&vfsmount_lock);
1241    umount_tree(mnt, 0, &umount_list);
1242    spin_unlock(&vfsmount_lock);
1243    up_write(&namespace_sem);
1244    release_mounts(&umount_list);
1245}
1246
1247static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1248{
1249    struct vfsmount *p;
1250
1251    for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1252        if (p->mnt_group_id && !IS_MNT_SHARED(p))
1253            mnt_release_group_id(p);
1254    }
1255}
1256
1257static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1258{
1259    struct vfsmount *p;
1260
1261    for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1262        if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1263            int err = mnt_alloc_group_id(p);
1264            if (err) {
1265                cleanup_group_ids(mnt, p);
1266                return err;
1267            }
1268        }
1269    }
1270
1271    return 0;
1272}
1273
1274/*
1275 * @source_mnt : mount tree to be attached
1276 * @nd : place the mount tree @source_mnt is attached
1277 * @parent_nd : if non-null, detach the source_mnt from its parent and
1278 * store the parent mount and mountpoint dentry.
1279 * (done when source_mnt is moved)
1280 *
1281 * NOTE: in the table below explains the semantics when a source mount
1282 * of a given type is attached to a destination mount of a given type.
1283 * ---------------------------------------------------------------------------
1284 * | BIND MOUNT OPERATION |
1285 * |**************************************************************************
1286 * | source-->| shared | private | slave | unbindable |
1287 * | dest | | | | |
1288 * | | | | | | |
1289 * | v | | | | |
1290 * |**************************************************************************
1291 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1292 * | | | | | |
1293 * |non-shared| shared (+) | private | slave (*) | invalid |
1294 * ***************************************************************************
1295 * A bind operation clones the source mount and mounts the clone on the
1296 * destination mount.
1297 *
1298 * (++) the cloned mount is propagated to all the mounts in the propagation
1299 * tree of the destination mount and the cloned mount is added to
1300 * the peer group of the source mount.
1301 * (+) the cloned mount is created under the destination mount and is marked
1302 * as shared. The cloned mount is added to the peer group of the source
1303 * mount.
1304 * (+++) the mount is propagated to all the mounts in the propagation tree
1305 * of the destination mount and the cloned mount is made slave
1306 * of the same master as that of the source mount. The cloned mount
1307 * is marked as 'shared and slave'.
1308 * (*) the cloned mount is made a slave of the same master as that of the
1309 * source mount.
1310 *
1311 * ---------------------------------------------------------------------------
1312 * | MOVE MOUNT OPERATION |
1313 * |**************************************************************************
1314 * | source-->| shared | private | slave | unbindable |
1315 * | dest | | | | |
1316 * | | | | | | |
1317 * | v | | | | |
1318 * |**************************************************************************
1319 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1320 * | | | | | |
1321 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1322 * ***************************************************************************
1323 *
1324 * (+) the mount is moved to the destination. And is then propagated to
1325 * all the mounts in the propagation tree of the destination mount.
1326 * (+*) the mount is moved to the destination.
1327 * (+++) the mount is moved to the destination and is then propagated to
1328 * all the mounts belonging to the destination mount's propagation tree.
1329 * the mount is marked as 'shared and slave'.
1330 * (*) the mount continues to be a slave at the new location.
1331 *
1332 * if the source mount is a tree, the operations explained above is
1333 * applied to each mount in the tree.
1334 * Must be called without spinlocks held, since this function can sleep
1335 * in allocations.
1336 */
1337static int attach_recursive_mnt(struct vfsmount *source_mnt,
1338            struct path *path, struct path *parent_path)
1339{
1340    LIST_HEAD(tree_list);
1341    struct vfsmount *dest_mnt = path->mnt;
1342    struct dentry *dest_dentry = path->dentry;
1343    struct vfsmount *child, *p;
1344    int err;
1345
1346    if (IS_MNT_SHARED(dest_mnt)) {
1347        err = invent_group_ids(source_mnt, true);
1348        if (err)
1349            goto out;
1350    }
1351    err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1352    if (err)
1353        goto out_cleanup_ids;
1354
1355    if (IS_MNT_SHARED(dest_mnt)) {
1356        for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1357            set_mnt_shared(p);
1358    }
1359
1360    spin_lock(&vfsmount_lock);
1361    if (parent_path) {
1362        detach_mnt(source_mnt, parent_path);
1363        attach_mnt(source_mnt, path);
1364        touch_mnt_namespace(parent_path->mnt->mnt_ns);
1365    } else {
1366        mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1367        commit_tree(source_mnt);
1368    }
1369
1370    list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1371        list_del_init(&child->mnt_hash);
1372        commit_tree(child);
1373    }
1374    spin_unlock(&vfsmount_lock);
1375    return 0;
1376
1377 out_cleanup_ids:
1378    if (IS_MNT_SHARED(dest_mnt))
1379        cleanup_group_ids(source_mnt, NULL);
1380 out:
1381    return err;
1382}
1383
1384static int graft_tree(struct vfsmount *mnt, struct path *path)
1385{
1386    int err;
1387    if (mnt->mnt_sb->s_flags & MS_NOUSER)
1388        return -EINVAL;
1389
1390    if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1391          S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1392        return -ENOTDIR;
1393
1394    err = -ENOENT;
1395    mutex_lock(&path->dentry->d_inode->i_mutex);
1396    if (IS_DEADDIR(path->dentry->d_inode))
1397        goto out_unlock;
1398
1399    err = security_sb_check_sb(mnt, path);
1400    if (err)
1401        goto out_unlock;
1402
1403    err = -ENOENT;
1404    if (!d_unlinked(path->dentry))
1405        err = attach_recursive_mnt(mnt, path, NULL);
1406out_unlock:
1407    mutex_unlock(&path->dentry->d_inode->i_mutex);
1408    if (!err)
1409        security_sb_post_addmount(mnt, path);
1410    return err;
1411}
1412
1413/*
1414 * recursively change the type of the mountpoint.
1415 */
1416static int do_change_type(struct path *path, int flag)
1417{
1418    struct vfsmount *m, *mnt = path->mnt;
1419    int recurse = flag & MS_REC;
1420    int type = flag & ~MS_REC;
1421    int err = 0;
1422
1423    if (!capable(CAP_SYS_ADMIN))
1424        return -EPERM;
1425
1426    if (path->dentry != path->mnt->mnt_root)
1427        return -EINVAL;
1428
1429    down_write(&namespace_sem);
1430    if (type == MS_SHARED) {
1431        err = invent_group_ids(mnt, recurse);
1432        if (err)
1433            goto out_unlock;
1434    }
1435
1436    spin_lock(&vfsmount_lock);
1437    for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1438        change_mnt_propagation(m, type);
1439    spin_unlock(&vfsmount_lock);
1440
1441 out_unlock:
1442    up_write(&namespace_sem);
1443    return err;
1444}
1445
1446/*
1447 * do loopback mount.
1448 */
1449static int do_loopback(struct path *path, char *old_name,
1450                int recurse)
1451{
1452    struct path old_path;
1453    struct vfsmount *mnt = NULL;
1454    int err = mount_is_safe(path);
1455    if (err)
1456        return err;
1457    if (!old_name || !*old_name)
1458        return -EINVAL;
1459    err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1460    if (err)
1461        return err;
1462
1463    down_write(&namespace_sem);
1464    err = -EINVAL;
1465    if (IS_MNT_UNBINDABLE(old_path.mnt))
1466        goto out;
1467
1468    if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1469        goto out;
1470
1471    err = -ENOMEM;
1472    if (recurse)
1473        mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1474    else
1475        mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1476
1477    if (!mnt)
1478        goto out;
1479
1480    err = graft_tree(mnt, path);
1481    if (err) {
1482        LIST_HEAD(umount_list);
1483        spin_lock(&vfsmount_lock);
1484        umount_tree(mnt, 0, &umount_list);
1485        spin_unlock(&vfsmount_lock);
1486        release_mounts(&umount_list);
1487    }
1488
1489out:
1490    up_write(&namespace_sem);
1491    path_put(&old_path);
1492    return err;
1493}
1494
1495static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1496{
1497    int error = 0;
1498    int readonly_request = 0;
1499
1500    if (ms_flags & MS_RDONLY)
1501        readonly_request = 1;
1502    if (readonly_request == __mnt_is_readonly(mnt))
1503        return 0;
1504
1505    if (readonly_request)
1506        error = mnt_make_readonly(mnt);
1507    else
1508        __mnt_unmake_readonly(mnt);
1509    return error;
1510}
1511
1512/*
1513 * change filesystem flags. dir should be a physical root of filesystem.
1514 * If you've mounted a non-root directory somewhere and want to do remount
1515 * on it - tough luck.
1516 */
1517static int do_remount(struct path *path, int flags, int mnt_flags,
1518              void *data)
1519{
1520    int err;
1521    struct super_block *sb = path->mnt->mnt_sb;
1522
1523    if (!capable(CAP_SYS_ADMIN))
1524        return -EPERM;
1525
1526    if (!check_mnt(path->mnt))
1527        return -EINVAL;
1528
1529    if (path->dentry != path->mnt->mnt_root)
1530        return -EINVAL;
1531
1532    down_write(&sb->s_umount);
1533    if (flags & MS_BIND)
1534        err = change_mount_flags(path->mnt, flags);
1535    else
1536        err = do_remount_sb(sb, flags, data, 0);
1537    if (!err)
1538        path->mnt->mnt_flags = mnt_flags;
1539    up_write(&sb->s_umount);
1540    if (!err) {
1541        security_sb_post_remount(path->mnt, flags, data);
1542
1543        spin_lock(&vfsmount_lock);
1544        touch_mnt_namespace(path->mnt->mnt_ns);
1545        spin_unlock(&vfsmount_lock);
1546    }
1547    return err;
1548}
1549
1550static inline int tree_contains_unbindable(struct vfsmount *mnt)
1551{
1552    struct vfsmount *p;
1553    for (p = mnt; p; p = next_mnt(p, mnt)) {
1554        if (IS_MNT_UNBINDABLE(p))
1555            return 1;
1556    }
1557    return 0;
1558}
1559
1560static int do_move_mount(struct path *path, char *old_name)
1561{
1562    struct path old_path, parent_path;
1563    struct vfsmount *p;
1564    int err = 0;
1565    if (!capable(CAP_SYS_ADMIN))
1566        return -EPERM;
1567    if (!old_name || !*old_name)
1568        return -EINVAL;
1569    err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1570    if (err)
1571        return err;
1572
1573    down_write(&namespace_sem);
1574    while (d_mountpoint(path->dentry) &&
1575           follow_down(path))
1576        ;
1577    err = -EINVAL;
1578    if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1579        goto out;
1580
1581    err = -ENOENT;
1582    mutex_lock(&path->dentry->d_inode->i_mutex);
1583    if (IS_DEADDIR(path->dentry->d_inode))
1584        goto out1;
1585
1586    if (d_unlinked(path->dentry))
1587        goto out1;
1588
1589    err = -EINVAL;
1590    if (old_path.dentry != old_path.mnt->mnt_root)
1591        goto out1;
1592
1593    if (old_path.mnt == old_path.mnt->mnt_parent)
1594        goto out1;
1595
1596    if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1597          S_ISDIR(old_path.dentry->d_inode->i_mode))
1598        goto out1;
1599    /*
1600     * Don't move a mount residing in a shared parent.
1601     */
1602    if (old_path.mnt->mnt_parent &&
1603        IS_MNT_SHARED(old_path.mnt->mnt_parent))
1604        goto out1;
1605    /*
1606     * Don't move a mount tree containing unbindable mounts to a destination
1607     * mount which is shared.
1608     */
1609    if (IS_MNT_SHARED(path->mnt) &&
1610        tree_contains_unbindable(old_path.mnt))
1611        goto out1;
1612    err = -ELOOP;
1613    for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1614        if (p == old_path.mnt)
1615            goto out1;
1616
1617    err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1618    if (err)
1619        goto out1;
1620
1621    /* if the mount is moved, it should no longer be expire
1622     * automatically */
1623    list_del_init(&old_path.mnt->mnt_expire);
1624out1:
1625    mutex_unlock(&path->dentry->d_inode->i_mutex);
1626out:
1627    up_write(&namespace_sem);
1628    if (!err)
1629        path_put(&parent_path);
1630    path_put(&old_path);
1631    return err;
1632}
1633
1634/*
1635 * create a new mount for userspace and request it to be added into the
1636 * namespace's tree
1637 */
1638static int do_new_mount(struct path *path, char *type, int flags,
1639            int mnt_flags, char *name, void *data)
1640{
1641    struct vfsmount *mnt;
1642
1643    if (!type || !memchr(type, 0, PAGE_SIZE))
1644        return -EINVAL;
1645
1646    /* we need capabilities... */
1647    if (!capable(CAP_SYS_ADMIN))
1648        return -EPERM;
1649
1650    lock_kernel();
1651    mnt = do_kern_mount(type, flags, name, data);
1652    unlock_kernel();
1653    if (IS_ERR(mnt))
1654        return PTR_ERR(mnt);
1655
1656    return do_add_mount(mnt, path, mnt_flags, NULL);
1657}
1658
1659/*
1660 * add a mount into a namespace's mount tree
1661 * - provide the option of adding the new mount to an expiration list
1662 */
1663int do_add_mount(struct vfsmount *newmnt, struct path *path,
1664         int mnt_flags, struct list_head *fslist)
1665{
1666    int err;
1667
1668    down_write(&namespace_sem);
1669    /* Something was mounted here while we slept */
1670    while (d_mountpoint(path->dentry) &&
1671           follow_down(path))
1672        ;
1673    err = -EINVAL;
1674    if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1675        goto unlock;
1676
1677    /* Refuse the same filesystem on the same mount point */
1678    err = -EBUSY;
1679    if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1680        path->mnt->mnt_root == path->dentry)
1681        goto unlock;
1682
1683    err = -EINVAL;
1684    if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1685        goto unlock;
1686
1687    newmnt->mnt_flags = mnt_flags;
1688    if ((err = graft_tree(newmnt, path)))
1689        goto unlock;
1690
1691    if (fslist) /* add to the specified expiration list */
1692        list_add_tail(&newmnt->mnt_expire, fslist);
1693
1694    up_write(&namespace_sem);
1695    return 0;
1696
1697unlock:
1698    up_write(&namespace_sem);
1699    mntput(newmnt);
1700    return err;
1701}
1702
1703EXPORT_SYMBOL_GPL(do_add_mount);
1704
1705/*
1706 * process a list of expirable mountpoints with the intent of discarding any
1707 * mountpoints that aren't in use and haven't been touched since last we came
1708 * here
1709 */
1710void mark_mounts_for_expiry(struct list_head *mounts)
1711{
1712    struct vfsmount *mnt, *next;
1713    LIST_HEAD(graveyard);
1714    LIST_HEAD(umounts);
1715
1716    if (list_empty(mounts))
1717        return;
1718
1719    down_write(&namespace_sem);
1720    spin_lock(&vfsmount_lock);
1721
1722    /* extract from the expiration list every vfsmount that matches the
1723     * following criteria:
1724     * - only referenced by its parent vfsmount
1725     * - still marked for expiry (marked on the last call here; marks are
1726     * cleared by mntput())
1727     */
1728    list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1729        if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1730            propagate_mount_busy(mnt, 1))
1731            continue;
1732        list_move(&mnt->mnt_expire, &graveyard);
1733    }
1734    while (!list_empty(&graveyard)) {
1735        mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1736        touch_mnt_namespace(mnt->mnt_ns);
1737        umount_tree(mnt, 1, &umounts);
1738    }
1739    spin_unlock(&vfsmount_lock);
1740    up_write(&namespace_sem);
1741
1742    release_mounts(&umounts);
1743}
1744
1745EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1746
1747/*
1748 * Ripoff of 'select_parent()'
1749 *
1750 * search the list of submounts for a given mountpoint, and move any
1751 * shrinkable submounts to the 'graveyard' list.
1752 */
1753static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1754{
1755    struct vfsmount *this_parent = parent;
1756    struct list_head *next;
1757    int found = 0;
1758
1759repeat:
1760    next = this_parent->mnt_mounts.next;
1761resume:
1762    while (next != &this_parent->mnt_mounts) {
1763        struct list_head *tmp = next;
1764        struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1765
1766        next = tmp->next;
1767        if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1768            continue;
1769        /*
1770         * Descend a level if the d_mounts list is non-empty.
1771         */
1772        if (!list_empty(&mnt->mnt_mounts)) {
1773            this_parent = mnt;
1774            goto repeat;
1775        }
1776
1777        if (!propagate_mount_busy(mnt, 1)) {
1778            list_move_tail(&mnt->mnt_expire, graveyard);
1779            found++;
1780        }
1781    }
1782    /*
1783     * All done at this level ... ascend and resume the search
1784     */
1785    if (this_parent != parent) {
1786        next = this_parent->mnt_child.next;
1787        this_parent = this_parent->mnt_parent;
1788        goto resume;
1789    }
1790    return found;
1791}
1792
1793/*
1794 * process a list of expirable mountpoints with the intent of discarding any
1795 * submounts of a specific parent mountpoint
1796 */
1797static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1798{
1799    LIST_HEAD(graveyard);
1800    struct vfsmount *m;
1801
1802    /* extract submounts of 'mountpoint' from the expiration list */
1803    while (select_submounts(mnt, &graveyard)) {
1804        while (!list_empty(&graveyard)) {
1805            m = list_first_entry(&graveyard, struct vfsmount,
1806                        mnt_expire);
1807            touch_mnt_namespace(m->mnt_ns);
1808            umount_tree(m, 1, umounts);
1809        }
1810    }
1811}
1812
1813/*
1814 * Some copy_from_user() implementations do not return the exact number of
1815 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1816 * Note that this function differs from copy_from_user() in that it will oops
1817 * on bad values of `to', rather than returning a short copy.
1818 */
1819static long exact_copy_from_user(void *to, const void __user * from,
1820                 unsigned long n)
1821{
1822    char *t = to;
1823    const char __user *f = from;
1824    char c;
1825
1826    if (!access_ok(VERIFY_READ, from, n))
1827        return n;
1828
1829    while (n) {
1830        if (__get_user(c, f)) {
1831            memset(t, 0, n);
1832            break;
1833        }
1834        *t++ = c;
1835        f++;
1836        n--;
1837    }
1838    return n;
1839}
1840
1841int copy_mount_options(const void __user * data, unsigned long *where)
1842{
1843    int i;
1844    unsigned long page;
1845    unsigned long size;
1846
1847    *where = 0;
1848    if (!data)
1849        return 0;
1850
1851    if (!(page = __get_free_page(GFP_KERNEL)))
1852        return -ENOMEM;
1853
1854    /* We only care that *some* data at the address the user
1855     * gave us is valid. Just in case, we'll zero
1856     * the remainder of the page.
1857     */
1858    /* copy_from_user cannot cross TASK_SIZE ! */
1859    size = TASK_SIZE - (unsigned long)data;
1860    if (size > PAGE_SIZE)
1861        size = PAGE_SIZE;
1862
1863    i = size - exact_copy_from_user((void *)page, data, size);
1864    if (!i) {
1865        free_page(page);
1866        return -EFAULT;
1867    }
1868    if (i != PAGE_SIZE)
1869        memset((char *)page + i, 0, PAGE_SIZE - i);
1870    *where = page;
1871    return 0;
1872}
1873
1874/*
1875 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1876 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1877 *
1878 * data is a (void *) that can point to any structure up to
1879 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1880 * information (or be NULL).
1881 *
1882 * Pre-0.97 versions of mount() didn't have a flags word.
1883 * When the flags word was introduced its top half was required
1884 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1885 * Therefore, if this magic number is present, it carries no information
1886 * and must be discarded.
1887 */
1888long do_mount(char *dev_name, char *dir_name, char *type_page,
1889          unsigned long flags, void *data_page)
1890{
1891    struct path path;
1892    int retval = 0;
1893    int mnt_flags = 0;
1894
1895    /* Discard magic */
1896    if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1897        flags &= ~MS_MGC_MSK;
1898
1899    /* Basic sanity checks */
1900
1901    if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1902        return -EINVAL;
1903    if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1904        return -EINVAL;
1905
1906    if (data_page)
1907        ((char *)data_page)[PAGE_SIZE - 1] = 0;
1908
1909    /* Default to relatime unless overriden */
1910    if (!(flags & MS_NOATIME))
1911        mnt_flags |= MNT_RELATIME;
1912
1913    /* Separate the per-mountpoint flags */
1914    if (flags & MS_NOSUID)
1915        mnt_flags |= MNT_NOSUID;
1916    if (flags & MS_NODEV)
1917        mnt_flags |= MNT_NODEV;
1918    if (flags & MS_NOEXEC)
1919        mnt_flags |= MNT_NOEXEC;
1920    if (flags & MS_NOATIME)
1921        mnt_flags |= MNT_NOATIME;
1922    if (flags & MS_NODIRATIME)
1923        mnt_flags |= MNT_NODIRATIME;
1924    if (flags & MS_STRICTATIME)
1925        mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
1926    if (flags & MS_RDONLY)
1927        mnt_flags |= MNT_READONLY;
1928
1929    flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1930           MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1931           MS_STRICTATIME);
1932
1933    /* ... and get the mountpoint */
1934    retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1935    if (retval)
1936        return retval;
1937
1938    retval = security_sb_mount(dev_name, &path,
1939                   type_page, flags, data_page);
1940    if (retval)
1941        goto dput_out;
1942
1943    if (flags & MS_REMOUNT)
1944        retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
1945                    data_page);
1946    else if (flags & MS_BIND)
1947        retval = do_loopback(&path, dev_name, flags & MS_REC);
1948    else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1949        retval = do_change_type(&path, flags);
1950    else if (flags & MS_MOVE)
1951        retval = do_move_mount(&path, dev_name);
1952    else
1953        retval = do_new_mount(&path, type_page, flags, mnt_flags,
1954                      dev_name, data_page);
1955dput_out:
1956    path_put(&path);
1957    return retval;
1958}
1959
1960static struct mnt_namespace *alloc_mnt_ns(void)
1961{
1962    struct mnt_namespace *new_ns;
1963
1964    new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1965    if (!new_ns)
1966        return ERR_PTR(-ENOMEM);
1967    atomic_set(&new_ns->count, 1);
1968    new_ns->root = NULL;
1969    INIT_LIST_HEAD(&new_ns->list);
1970    init_waitqueue_head(&new_ns->poll);
1971    new_ns->event = 0;
1972    return new_ns;
1973}
1974
1975/*
1976 * Allocate a new namespace structure and populate it with contents
1977 * copied from the namespace of the passed in task structure.
1978 */
1979static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1980        struct fs_struct *fs)
1981{
1982    struct mnt_namespace *new_ns;
1983    struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
1984    struct vfsmount *p, *q;
1985
1986    new_ns = alloc_mnt_ns();
1987    if (IS_ERR(new_ns))
1988        return new_ns;
1989
1990    down_write(&namespace_sem);
1991    /* First pass: copy the tree topology */
1992    new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1993                    CL_COPY_ALL | CL_EXPIRE);
1994    if (!new_ns->root) {
1995        up_write(&namespace_sem);
1996        kfree(new_ns);
1997        return ERR_PTR(-ENOMEM);
1998    }
1999    spin_lock(&vfsmount_lock);
2000    list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2001    spin_unlock(&vfsmount_lock);
2002
2003    /*
2004     * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2005     * as belonging to new namespace. We have already acquired a private
2006     * fs_struct, so tsk->fs->lock is not needed.
2007     */
2008    p = mnt_ns->root;
2009    q = new_ns->root;
2010    while (p) {
2011        q->mnt_ns = new_ns;
2012        if (fs) {
2013            if (p == fs->root.mnt) {
2014                rootmnt = p;
2015                fs->root.mnt = mntget(q);
2016            }
2017            if (p == fs->pwd.mnt) {
2018                pwdmnt = p;
2019                fs->pwd.mnt = mntget(q);
2020            }
2021        }
2022        p = next_mnt(p, mnt_ns->root);
2023        q = next_mnt(q, new_ns->root);
2024    }
2025    up_write(&namespace_sem);
2026
2027    if (rootmnt)
2028        mntput(rootmnt);
2029    if (pwdmnt)
2030        mntput(pwdmnt);
2031
2032    return new_ns;
2033}
2034
2035struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2036        struct fs_struct *new_fs)
2037{
2038    struct mnt_namespace *new_ns;
2039
2040    BUG_ON(!ns);
2041    get_mnt_ns(ns);
2042
2043    if (!(flags & CLONE_NEWNS))
2044        return ns;
2045
2046    new_ns = dup_mnt_ns(ns, new_fs);
2047
2048    put_mnt_ns(ns);
2049    return new_ns;
2050}
2051
2052/**
2053 * create_mnt_ns - creates a private namespace and adds a root filesystem
2054 * @mnt: pointer to the new root filesystem mountpoint
2055 */
2056struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2057{
2058    struct mnt_namespace *new_ns;
2059
2060    new_ns = alloc_mnt_ns();
2061    if (!IS_ERR(new_ns)) {
2062        mnt->mnt_ns = new_ns;
2063        new_ns->root = mnt;
2064        list_add(&new_ns->list, &new_ns->root->mnt_list);
2065    }
2066    return new_ns;
2067}
2068EXPORT_SYMBOL(create_mnt_ns);
2069
2070SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2071        char __user *, type, unsigned long, flags, void __user *, data)
2072{
2073    int retval;
2074    unsigned long data_page;
2075    unsigned long type_page;
2076    unsigned long dev_page;
2077    char *dir_page;
2078
2079    retval = copy_mount_options(type, &type_page);
2080    if (retval < 0)
2081        return retval;
2082
2083    dir_page = getname(dir_name);
2084    retval = PTR_ERR(dir_page);
2085    if (IS_ERR(dir_page))
2086        goto out1;
2087
2088    retval = copy_mount_options(dev_name, &dev_page);
2089    if (retval < 0)
2090        goto out2;
2091
2092    retval = copy_mount_options(data, &data_page);
2093    if (retval < 0)
2094        goto out3;
2095
2096    retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2097              flags, (void *)data_page);
2098    free_page(data_page);
2099
2100out3:
2101    free_page(dev_page);
2102out2:
2103    putname(dir_page);
2104out1:
2105    free_page(type_page);
2106    return retval;
2107}
2108
2109/*
2110 * pivot_root Semantics:
2111 * Moves the root file system of the current process to the directory put_old,
2112 * makes new_root as the new root file system of the current process, and sets
2113 * root/cwd of all processes which had them on the current root to new_root.
2114 *
2115 * Restrictions:
2116 * The new_root and put_old must be directories, and must not be on the
2117 * same file system as the current process root. The put_old must be
2118 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2119 * pointed to by put_old must yield the same directory as new_root. No other
2120 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2121 *
2122 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2123 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2124 * in this situation.
2125 *
2126 * Notes:
2127 * - we don't move root/cwd if they are not at the root (reason: if something
2128 * cared enough to change them, it's probably wrong to force them elsewhere)
2129 * - it's okay to pick a root that isn't the root of a file system, e.g.
2130 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2131 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2132 * first.
2133 */
2134SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2135        const char __user *, put_old)
2136{
2137    struct vfsmount *tmp;
2138    struct path new, old, parent_path, root_parent, root;
2139    int error;
2140
2141    if (!capable(CAP_SYS_ADMIN))
2142        return -EPERM;
2143
2144    error = user_path_dir(new_root, &new);
2145    if (error)
2146        goto out0;
2147    error = -EINVAL;
2148    if (!check_mnt(new.mnt))
2149        goto out1;
2150
2151    error = user_path_dir(put_old, &old);
2152    if (error)
2153        goto out1;
2154
2155    error = security_sb_pivotroot(&old, &new);
2156    if (error) {
2157        path_put(&old);
2158        goto out1;
2159    }
2160
2161    read_lock(&current->fs->lock);
2162    root = current->fs->root;
2163    path_get(&current->fs->root);
2164    read_unlock(&current->fs->lock);
2165    down_write(&namespace_sem);
2166    mutex_lock(&old.dentry->d_inode->i_mutex);
2167    error = -EINVAL;
2168    if (IS_MNT_SHARED(old.mnt) ||
2169        IS_MNT_SHARED(new.mnt->mnt_parent) ||
2170        IS_MNT_SHARED(root.mnt->mnt_parent))
2171        goto out2;
2172    if (!check_mnt(root.mnt))
2173        goto out2;
2174    error = -ENOENT;
2175    if (IS_DEADDIR(new.dentry->d_inode))
2176        goto out2;
2177    if (d_unlinked(new.dentry))
2178        goto out2;
2179    if (d_unlinked(old.dentry))
2180        goto out2;
2181    error = -EBUSY;
2182    if (new.mnt == root.mnt ||
2183        old.mnt == root.mnt)
2184        goto out2; /* loop, on the same file system */
2185    error = -EINVAL;
2186    if (root.mnt->mnt_root != root.dentry)
2187        goto out2; /* not a mountpoint */
2188    if (root.mnt->mnt_parent == root.mnt)
2189        goto out2; /* not attached */
2190    if (new.mnt->mnt_root != new.dentry)
2191        goto out2; /* not a mountpoint */
2192    if (new.mnt->mnt_parent == new.mnt)
2193        goto out2; /* not attached */
2194    /* make sure we can reach put_old from new_root */
2195    tmp = old.mnt;
2196    spin_lock(&vfsmount_lock);
2197    if (tmp != new.mnt) {
2198        for (;;) {
2199            if (tmp->mnt_parent == tmp)
2200                goto out3; /* already mounted on put_old */
2201            if (tmp->mnt_parent == new.mnt)
2202                break;
2203            tmp = tmp->mnt_parent;
2204        }
2205        if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2206            goto out3;
2207    } else if (!is_subdir(old.dentry, new.dentry))
2208        goto out3;
2209    detach_mnt(new.mnt, &parent_path);
2210    detach_mnt(root.mnt, &root_parent);
2211    /* mount old root on put_old */
2212    attach_mnt(root.mnt, &old);
2213    /* mount new_root on / */
2214    attach_mnt(new.mnt, &root_parent);
2215    touch_mnt_namespace(current->nsproxy->mnt_ns);
2216    spin_unlock(&vfsmount_lock);
2217    chroot_fs_refs(&root, &new);
2218    security_sb_post_pivotroot(&root, &new);
2219    error = 0;
2220    path_put(&root_parent);
2221    path_put(&parent_path);
2222out2:
2223    mutex_unlock(&old.dentry->d_inode->i_mutex);
2224    up_write(&namespace_sem);
2225    path_put(&root);
2226    path_put(&old);
2227out1:
2228    path_put(&new);
2229out0:
2230    return error;
2231out3:
2232    spin_unlock(&vfsmount_lock);
2233    goto out2;
2234}
2235
2236static void __init init_mount_tree(void)
2237{
2238    struct vfsmount *mnt;
2239    struct mnt_namespace *ns;
2240    struct path root;
2241
2242    mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2243    if (IS_ERR(mnt))
2244        panic("Can't create rootfs");
2245    ns = create_mnt_ns(mnt);
2246    if (IS_ERR(ns))
2247        panic("Can't allocate initial namespace");
2248
2249    init_task.nsproxy->mnt_ns = ns;
2250    get_mnt_ns(ns);
2251
2252    root.mnt = ns->root;
2253    root.dentry = ns->root->mnt_root;
2254
2255    set_fs_pwd(current->fs, &root);
2256    set_fs_root(current->fs, &root);
2257}
2258
2259void __init mnt_init(void)
2260{
2261    unsigned u;
2262    int err;
2263
2264    init_rwsem(&namespace_sem);
2265
2266    mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2267            0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2268
2269    mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2270
2271    if (!mount_hashtable)
2272        panic("Failed to allocate mount hash table\n");
2273
2274    printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2275
2276    for (u = 0; u < HASH_SIZE; u++)
2277        INIT_LIST_HEAD(&mount_hashtable[u]);
2278
2279    err = sysfs_init();
2280    if (err)
2281        printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2282            __func__, err);
2283    fs_kobj = kobject_create_and_add("fs", NULL);
2284    if (!fs_kobj)
2285        printk(KERN_WARNING "%s: kobj create error\n", __func__);
2286    init_rootfs();
2287    init_mount_tree();
2288}
2289
2290void put_mnt_ns(struct mnt_namespace *ns)
2291{
2292    struct vfsmount *root;
2293    LIST_HEAD(umount_list);
2294
2295    if (!atomic_dec_and_lock(&ns->count, &vfsmount_lock))
2296        return;
2297    root = ns->root;
2298    ns->root = NULL;
2299    spin_unlock(&vfsmount_lock);
2300    down_write(&namespace_sem);
2301    spin_lock(&vfsmount_lock);
2302    umount_tree(root, 0, &umount_list);
2303    spin_unlock(&vfsmount_lock);
2304    up_write(&namespace_sem);
2305    release_mounts(&umount_list);
2306    kfree(ns);
2307}
2308EXPORT_SYMBOL(put_mnt_ns);
2309

Archive Download this file



interactive