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_MAKE_SHARED) || 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
740int mnt_had_events(struct proc_mounts *p)
741{
742    struct mnt_namespace *ns = p->ns;
743    int res = 0;
744
745    spin_lock(&vfsmount_lock);
746    if (p->event != ns->event) {
747        p->event = ns->event;
748        res = 1;
749    }
750    spin_unlock(&vfsmount_lock);
751
752    return res;
753}
754
755struct proc_fs_info {
756    int flag;
757    const char *str;
758};
759
760static int show_sb_opts(struct seq_file *m, struct super_block *sb)
761{
762    static const struct proc_fs_info fs_info[] = {
763        { MS_SYNCHRONOUS, ",sync" },
764        { MS_DIRSYNC, ",dirsync" },
765        { MS_MANDLOCK, ",mand" },
766        { 0, NULL }
767    };
768    const struct proc_fs_info *fs_infop;
769
770    for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
771        if (sb->s_flags & fs_infop->flag)
772            seq_puts(m, fs_infop->str);
773    }
774
775    return security_sb_show_options(m, sb);
776}
777
778static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
779{
780    static const struct proc_fs_info mnt_info[] = {
781        { MNT_NOSUID, ",nosuid" },
782        { MNT_NODEV, ",nodev" },
783        { MNT_NOEXEC, ",noexec" },
784        { MNT_NOATIME, ",noatime" },
785        { MNT_NODIRATIME, ",nodiratime" },
786        { MNT_RELATIME, ",relatime" },
787        { MNT_STRICTATIME, ",strictatime" },
788        { 0, NULL }
789    };
790    const struct proc_fs_info *fs_infop;
791
792    for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
793        if (mnt->mnt_flags & fs_infop->flag)
794            seq_puts(m, fs_infop->str);
795    }
796}
797
798static void show_type(struct seq_file *m, struct super_block *sb)
799{
800    mangle(m, sb->s_type->name);
801    if (sb->s_subtype && sb->s_subtype[0]) {
802        seq_putc(m, '.');
803        mangle(m, sb->s_subtype);
804    }
805}
806
807static int show_vfsmnt(struct seq_file *m, void *v)
808{
809    struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
810    int err = 0;
811    struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
812
813    mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
814    seq_putc(m, ' ');
815    seq_path(m, &mnt_path, " \t\n\\");
816    seq_putc(m, ' ');
817    show_type(m, mnt->mnt_sb);
818    seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
819    err = show_sb_opts(m, mnt->mnt_sb);
820    if (err)
821        goto out;
822    show_mnt_opts(m, mnt);
823    if (mnt->mnt_sb->s_op->show_options)
824        err = mnt->mnt_sb->s_op->show_options(m, mnt);
825    seq_puts(m, " 0 0\n");
826out:
827    return err;
828}
829
830const struct seq_operations mounts_op = {
831    .start = m_start,
832    .next = m_next,
833    .stop = m_stop,
834    .show = show_vfsmnt
835};
836
837static int show_mountinfo(struct seq_file *m, void *v)
838{
839    struct proc_mounts *p = m->private;
840    struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
841    struct super_block *sb = mnt->mnt_sb;
842    struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
843    struct path root = p->root;
844    int err = 0;
845
846    seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
847           MAJOR(sb->s_dev), MINOR(sb->s_dev));
848    seq_dentry(m, mnt->mnt_root, " \t\n\\");
849    seq_putc(m, ' ');
850    seq_path_root(m, &mnt_path, &root, " \t\n\\");
851    if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
852        /*
853         * Mountpoint is outside root, discard that one. Ugly,
854         * but less so than trying to do that in iterator in a
855         * race-free way (due to renames).
856         */
857        return SEQ_SKIP;
858    }
859    seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
860    show_mnt_opts(m, mnt);
861
862    /* Tagged fields ("foo:X" or "bar") */
863    if (IS_MNT_SHARED(mnt))
864        seq_printf(m, " shared:%i", mnt->mnt_group_id);
865    if (IS_MNT_SLAVE(mnt)) {
866        int master = mnt->mnt_master->mnt_group_id;
867        int dom = get_dominating_id(mnt, &p->root);
868        seq_printf(m, " master:%i", master);
869        if (dom && dom != master)
870            seq_printf(m, " propagate_from:%i", dom);
871    }
872    if (IS_MNT_UNBINDABLE(mnt))
873        seq_puts(m, " unbindable");
874
875    /* Filesystem specific data */
876    seq_puts(m, " - ");
877    show_type(m, sb);
878    seq_putc(m, ' ');
879    mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
880    seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
881    err = show_sb_opts(m, sb);
882    if (err)
883        goto out;
884    if (sb->s_op->show_options)
885        err = sb->s_op->show_options(m, mnt);
886    seq_putc(m, '\n');
887out:
888    return err;
889}
890
891const struct seq_operations mountinfo_op = {
892    .start = m_start,
893    .next = m_next,
894    .stop = m_stop,
895    .show = show_mountinfo,
896};
897
898static int show_vfsstat(struct seq_file *m, void *v)
899{
900    struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
901    struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
902    int err = 0;
903
904    /* device */
905    if (mnt->mnt_devname) {
906        seq_puts(m, "device ");
907        mangle(m, mnt->mnt_devname);
908    } else
909        seq_puts(m, "no device");
910
911    /* mount point */
912    seq_puts(m, " mounted on ");
913    seq_path(m, &mnt_path, " \t\n\\");
914    seq_putc(m, ' ');
915
916    /* file system type */
917    seq_puts(m, "with fstype ");
918    show_type(m, mnt->mnt_sb);
919
920    /* optional statistics */
921    if (mnt->mnt_sb->s_op->show_stats) {
922        seq_putc(m, ' ');
923        err = mnt->mnt_sb->s_op->show_stats(m, mnt);
924    }
925
926    seq_putc(m, '\n');
927    return err;
928}
929
930const struct seq_operations mountstats_op = {
931    .start = m_start,
932    .next = m_next,
933    .stop = m_stop,
934    .show = show_vfsstat,
935};
936#endif /* CONFIG_PROC_FS */
937
938/**
939 * may_umount_tree - check if a mount tree is busy
940 * @mnt: root of mount tree
941 *
942 * This is called to check if a tree of mounts has any
943 * open files, pwds, chroots or sub mounts that are
944 * busy.
945 */
946int may_umount_tree(struct vfsmount *mnt)
947{
948    int actual_refs = 0;
949    int minimum_refs = 0;
950    struct vfsmount *p;
951
952    spin_lock(&vfsmount_lock);
953    for (p = mnt; p; p = next_mnt(p, mnt)) {
954        actual_refs += atomic_read(&p->mnt_count);
955        minimum_refs += 2;
956    }
957    spin_unlock(&vfsmount_lock);
958
959    if (actual_refs > minimum_refs)
960        return 0;
961
962    return 1;
963}
964
965EXPORT_SYMBOL(may_umount_tree);
966
967/**
968 * may_umount - check if a mount point is busy
969 * @mnt: root of mount
970 *
971 * This is called to check if a mount point has any
972 * open files, pwds, chroots or sub mounts. If the
973 * mount has sub mounts this will return busy
974 * regardless of whether the sub mounts are busy.
975 *
976 * Doesn't take quota and stuff into account. IOW, in some cases it will
977 * give false negatives. The main reason why it's here is that we need
978 * a non-destructive way to look for easily umountable filesystems.
979 */
980int may_umount(struct vfsmount *mnt)
981{
982    int ret = 1;
983    down_read(&namespace_sem);
984    spin_lock(&vfsmount_lock);
985    if (propagate_mount_busy(mnt, 2))
986        ret = 0;
987    spin_unlock(&vfsmount_lock);
988    up_read(&namespace_sem);
989    return ret;
990}
991
992EXPORT_SYMBOL(may_umount);
993
994void release_mounts(struct list_head *head)
995{
996    struct vfsmount *mnt;
997    while (!list_empty(head)) {
998        mnt = list_first_entry(head, struct vfsmount, mnt_hash);
999        list_del_init(&mnt->mnt_hash);
1000        if (mnt->mnt_parent != mnt) {
1001            struct dentry *dentry;
1002            struct vfsmount *m;
1003            spin_lock(&vfsmount_lock);
1004            dentry = mnt->mnt_mountpoint;
1005            m = mnt->mnt_parent;
1006            mnt->mnt_mountpoint = mnt->mnt_root;
1007            mnt->mnt_parent = mnt;
1008            m->mnt_ghosts--;
1009            spin_unlock(&vfsmount_lock);
1010            dput(dentry);
1011            mntput(m);
1012        }
1013        mntput(mnt);
1014    }
1015}
1016
1017void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1018{
1019    struct vfsmount *p;
1020
1021    for (p = mnt; p; p = next_mnt(p, mnt))
1022        list_move(&p->mnt_hash, kill);
1023
1024    if (propagate)
1025        propagate_umount(kill);
1026
1027    list_for_each_entry(p, kill, mnt_hash) {
1028        list_del_init(&p->mnt_expire);
1029        list_del_init(&p->mnt_list);
1030        __touch_mnt_namespace(p->mnt_ns);
1031        p->mnt_ns = NULL;
1032        list_del_init(&p->mnt_child);
1033        if (p->mnt_parent != p) {
1034            p->mnt_parent->mnt_ghosts++;
1035            p->mnt_mountpoint->d_mounted--;
1036        }
1037        change_mnt_propagation(p, MS_PRIVATE);
1038    }
1039}
1040
1041static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1042
1043static int do_umount(struct vfsmount *mnt, int flags)
1044{
1045    struct super_block *sb = mnt->mnt_sb;
1046    int retval;
1047    LIST_HEAD(umount_list);
1048
1049    retval = security_sb_umount(mnt, flags);
1050    if (retval)
1051        return retval;
1052
1053    /*
1054     * Allow userspace to request a mountpoint be expired rather than
1055     * unmounting unconditionally. Unmount only happens if:
1056     * (1) the mark is already set (the mark is cleared by mntput())
1057     * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1058     */
1059    if (flags & MNT_EXPIRE) {
1060        if (mnt == current->fs->root.mnt ||
1061            flags & (MNT_FORCE | MNT_DETACH))
1062            return -EINVAL;
1063
1064        if (atomic_read(&mnt->mnt_count) != 2)
1065            return -EBUSY;
1066
1067        if (!xchg(&mnt->mnt_expiry_mark, 1))
1068            return -EAGAIN;
1069    }
1070
1071    /*
1072     * If we may have to abort operations to get out of this
1073     * mount, and they will themselves hold resources we must
1074     * allow the fs to do things. In the Unix tradition of
1075     * 'Gee thats tricky lets do it in userspace' the umount_begin
1076     * might fail to complete on the first run through as other tasks
1077     * must return, and the like. Thats for the mount program to worry
1078     * about for the moment.
1079     */
1080
1081    if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1082        sb->s_op->umount_begin(sb);
1083    }
1084
1085    /*
1086     * No sense to grab the lock for this test, but test itself looks
1087     * somewhat bogus. Suggestions for better replacement?
1088     * Ho-hum... In principle, we might treat that as umount + switch
1089     * to rootfs. GC would eventually take care of the old vfsmount.
1090     * Actually it makes sense, especially if rootfs would contain a
1091     * /reboot - static binary that would close all descriptors and
1092     * call reboot(9). Then init(8) could umount root and exec /reboot.
1093     */
1094    if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1095        /*
1096         * Special case for "unmounting" root ...
1097         * we just try to remount it readonly.
1098         */
1099        down_write(&sb->s_umount);
1100        if (!(sb->s_flags & MS_RDONLY))
1101            retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1102        up_write(&sb->s_umount);
1103        return retval;
1104    }
1105
1106    down_write(&namespace_sem);
1107    spin_lock(&vfsmount_lock);
1108    event++;
1109
1110    if (!(flags & MNT_DETACH))
1111        shrink_submounts(mnt, &umount_list);
1112
1113    retval = -EBUSY;
1114    if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1115        if (!list_empty(&mnt->mnt_list))
1116            umount_tree(mnt, 1, &umount_list);
1117        retval = 0;
1118    }
1119    spin_unlock(&vfsmount_lock);
1120    if (retval)
1121        security_sb_umount_busy(mnt);
1122    up_write(&namespace_sem);
1123    release_mounts(&umount_list);
1124    return retval;
1125}
1126
1127/*
1128 * Now umount can handle mount points as well as block devices.
1129 * This is important for filesystems which use unnamed block devices.
1130 *
1131 * We now support a flag for forced unmount like the other 'big iron'
1132 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1133 */
1134
1135SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1136{
1137    struct path path;
1138    int retval;
1139    int lookup_flags = 0;
1140
1141    if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1142        return -EINVAL;
1143
1144    if (!(flags & UMOUNT_NOFOLLOW))
1145        lookup_flags |= LOOKUP_FOLLOW;
1146
1147    retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1148    if (retval)
1149        goto out;
1150    retval = -EINVAL;
1151    if (path.dentry != path.mnt->mnt_root)
1152        goto dput_and_out;
1153    if (!check_mnt(path.mnt))
1154        goto dput_and_out;
1155
1156    retval = -EPERM;
1157    if (!capable(CAP_SYS_ADMIN))
1158        goto dput_and_out;
1159
1160    retval = do_umount(path.mnt, flags);
1161dput_and_out:
1162    /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1163    dput(path.dentry);
1164    mntput_no_expire(path.mnt);
1165out:
1166    return retval;
1167}
1168
1169#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1170
1171/*
1172 * The 2.0 compatible umount. No flags.
1173 */
1174SYSCALL_DEFINE1(oldumount, char __user *, name)
1175{
1176    return sys_umount(name, 0);
1177}
1178
1179#endif
1180
1181static int mount_is_safe(struct path *path)
1182{
1183    if (capable(CAP_SYS_ADMIN))
1184        return 0;
1185    return -EPERM;
1186#ifdef notyet
1187    if (S_ISLNK(path->dentry->d_inode->i_mode))
1188        return -EPERM;
1189    if (path->dentry->d_inode->i_mode & S_ISVTX) {
1190        if (current_uid() != path->dentry->d_inode->i_uid)
1191            return -EPERM;
1192    }
1193    if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1194        return -EPERM;
1195    return 0;
1196#endif
1197}
1198
1199struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1200                    int flag)
1201{
1202    struct vfsmount *res, *p, *q, *r, *s;
1203    struct path path;
1204
1205    if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1206        return NULL;
1207
1208    res = q = clone_mnt(mnt, dentry, flag);
1209    if (!q)
1210        goto Enomem;
1211    q->mnt_mountpoint = mnt->mnt_mountpoint;
1212
1213    p = mnt;
1214    list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1215        if (!is_subdir(r->mnt_mountpoint, dentry))
1216            continue;
1217
1218        for (s = r; s; s = next_mnt(s, r)) {
1219            if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1220                s = skip_mnt_tree(s);
1221                continue;
1222            }
1223            while (p != s->mnt_parent) {
1224                p = p->mnt_parent;
1225                q = q->mnt_parent;
1226            }
1227            p = s;
1228            path.mnt = q;
1229            path.dentry = p->mnt_mountpoint;
1230            q = clone_mnt(p, p->mnt_root, flag);
1231            if (!q)
1232                goto Enomem;
1233            spin_lock(&vfsmount_lock);
1234            list_add_tail(&q->mnt_list, &res->mnt_list);
1235            attach_mnt(q, &path);
1236            spin_unlock(&vfsmount_lock);
1237        }
1238    }
1239    return res;
1240Enomem:
1241    if (res) {
1242        LIST_HEAD(umount_list);
1243        spin_lock(&vfsmount_lock);
1244        umount_tree(res, 0, &umount_list);
1245        spin_unlock(&vfsmount_lock);
1246        release_mounts(&umount_list);
1247    }
1248    return NULL;
1249}
1250
1251struct vfsmount *collect_mounts(struct path *path)
1252{
1253    struct vfsmount *tree;
1254    down_write(&namespace_sem);
1255    tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1256    up_write(&namespace_sem);
1257    return tree;
1258}
1259
1260void drop_collected_mounts(struct vfsmount *mnt)
1261{
1262    LIST_HEAD(umount_list);
1263    down_write(&namespace_sem);
1264    spin_lock(&vfsmount_lock);
1265    umount_tree(mnt, 0, &umount_list);
1266    spin_unlock(&vfsmount_lock);
1267    up_write(&namespace_sem);
1268    release_mounts(&umount_list);
1269}
1270
1271int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1272           struct vfsmount *root)
1273{
1274    struct vfsmount *mnt;
1275    int res = f(root, arg);
1276    if (res)
1277        return res;
1278    list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1279        res = f(mnt, arg);
1280        if (res)
1281            return res;
1282    }
1283    return 0;
1284}
1285
1286static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1287{
1288    struct vfsmount *p;
1289
1290    for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1291        if (p->mnt_group_id && !IS_MNT_SHARED(p))
1292            mnt_release_group_id(p);
1293    }
1294}
1295
1296static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1297{
1298    struct vfsmount *p;
1299
1300    for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1301        if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1302            int err = mnt_alloc_group_id(p);
1303            if (err) {
1304                cleanup_group_ids(mnt, p);
1305                return err;
1306            }
1307        }
1308    }
1309
1310    return 0;
1311}
1312
1313/*
1314 * @source_mnt : mount tree to be attached
1315 * @nd : place the mount tree @source_mnt is attached
1316 * @parent_nd : if non-null, detach the source_mnt from its parent and
1317 * store the parent mount and mountpoint dentry.
1318 * (done when source_mnt is moved)
1319 *
1320 * NOTE: in the table below explains the semantics when a source mount
1321 * of a given type is attached to a destination mount of a given type.
1322 * ---------------------------------------------------------------------------
1323 * | BIND MOUNT OPERATION |
1324 * |**************************************************************************
1325 * | source-->| shared | private | slave | unbindable |
1326 * | dest | | | | |
1327 * | | | | | | |
1328 * | v | | | | |
1329 * |**************************************************************************
1330 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1331 * | | | | | |
1332 * |non-shared| shared (+) | private | slave (*) | invalid |
1333 * ***************************************************************************
1334 * A bind operation clones the source mount and mounts the clone on the
1335 * destination mount.
1336 *
1337 * (++) the cloned mount is propagated to all the mounts in the propagation
1338 * tree of the destination mount and the cloned mount is added to
1339 * the peer group of the source mount.
1340 * (+) the cloned mount is created under the destination mount and is marked
1341 * as shared. The cloned mount is added to the peer group of the source
1342 * mount.
1343 * (+++) the mount is propagated to all the mounts in the propagation tree
1344 * of the destination mount and the cloned mount is made slave
1345 * of the same master as that of the source mount. The cloned mount
1346 * is marked as 'shared and slave'.
1347 * (*) the cloned mount is made a slave of the same master as that of the
1348 * source mount.
1349 *
1350 * ---------------------------------------------------------------------------
1351 * | MOVE MOUNT OPERATION |
1352 * |**************************************************************************
1353 * | source-->| shared | private | slave | unbindable |
1354 * | dest | | | | |
1355 * | | | | | | |
1356 * | v | | | | |
1357 * |**************************************************************************
1358 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1359 * | | | | | |
1360 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1361 * ***************************************************************************
1362 *
1363 * (+) the mount is moved to the destination. And is then propagated to
1364 * all the mounts in the propagation tree of the destination mount.
1365 * (+*) the mount is moved to the destination.
1366 * (+++) the mount is moved to the destination and is then propagated to
1367 * all the mounts belonging to the destination mount's propagation tree.
1368 * the mount is marked as 'shared and slave'.
1369 * (*) the mount continues to be a slave at the new location.
1370 *
1371 * if the source mount is a tree, the operations explained above is
1372 * applied to each mount in the tree.
1373 * Must be called without spinlocks held, since this function can sleep
1374 * in allocations.
1375 */
1376static int attach_recursive_mnt(struct vfsmount *source_mnt,
1377            struct path *path, struct path *parent_path)
1378{
1379    LIST_HEAD(tree_list);
1380    struct vfsmount *dest_mnt = path->mnt;
1381    struct dentry *dest_dentry = path->dentry;
1382    struct vfsmount *child, *p;
1383    int err;
1384
1385    if (IS_MNT_SHARED(dest_mnt)) {
1386        err = invent_group_ids(source_mnt, true);
1387        if (err)
1388            goto out;
1389    }
1390    err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1391    if (err)
1392        goto out_cleanup_ids;
1393
1394    spin_lock(&vfsmount_lock);
1395
1396    if (IS_MNT_SHARED(dest_mnt)) {
1397        for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1398            set_mnt_shared(p);
1399    }
1400    if (parent_path) {
1401        detach_mnt(source_mnt, parent_path);
1402        attach_mnt(source_mnt, path);
1403        touch_mnt_namespace(parent_path->mnt->mnt_ns);
1404    } else {
1405        mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1406        commit_tree(source_mnt);
1407    }
1408
1409    list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1410        list_del_init(&child->mnt_hash);
1411        commit_tree(child);
1412    }
1413    spin_unlock(&vfsmount_lock);
1414    return 0;
1415
1416 out_cleanup_ids:
1417    if (IS_MNT_SHARED(dest_mnt))
1418        cleanup_group_ids(source_mnt, NULL);
1419 out:
1420    return err;
1421}
1422
1423static int graft_tree(struct vfsmount *mnt, struct path *path)
1424{
1425    int err;
1426    if (mnt->mnt_sb->s_flags & MS_NOUSER)
1427        return -EINVAL;
1428
1429    if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1430          S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1431        return -ENOTDIR;
1432
1433    err = -ENOENT;
1434    mutex_lock(&path->dentry->d_inode->i_mutex);
1435    if (IS_DEADDIR(path->dentry->d_inode))
1436        goto out_unlock;
1437
1438    err = security_sb_check_sb(mnt, path);
1439    if (err)
1440        goto out_unlock;
1441
1442    err = -ENOENT;
1443    if (!d_unlinked(path->dentry))
1444        err = attach_recursive_mnt(mnt, path, NULL);
1445out_unlock:
1446    mutex_unlock(&path->dentry->d_inode->i_mutex);
1447    if (!err)
1448        security_sb_post_addmount(mnt, path);
1449    return err;
1450}
1451
1452/*
1453 * recursively change the type of the mountpoint.
1454 */
1455static int do_change_type(struct path *path, int flag)
1456{
1457    struct vfsmount *m, *mnt = path->mnt;
1458    int recurse = flag & MS_REC;
1459    int type = flag & ~MS_REC;
1460    int err = 0;
1461
1462    if (!capable(CAP_SYS_ADMIN))
1463        return -EPERM;
1464
1465    if (path->dentry != path->mnt->mnt_root)
1466        return -EINVAL;
1467
1468    down_write(&namespace_sem);
1469    if (type == MS_SHARED) {
1470        err = invent_group_ids(mnt, recurse);
1471        if (err)
1472            goto out_unlock;
1473    }
1474
1475    spin_lock(&vfsmount_lock);
1476    for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1477        change_mnt_propagation(m, type);
1478    spin_unlock(&vfsmount_lock);
1479
1480 out_unlock:
1481    up_write(&namespace_sem);
1482    return err;
1483}
1484
1485/*
1486 * do loopback mount.
1487 */
1488static int do_loopback(struct path *path, char *old_name,
1489                int recurse)
1490{
1491    struct path old_path;
1492    struct vfsmount *mnt = NULL;
1493    int err = mount_is_safe(path);
1494    if (err)
1495        return err;
1496    if (!old_name || !*old_name)
1497        return -EINVAL;
1498    err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1499    if (err)
1500        return err;
1501
1502    down_write(&namespace_sem);
1503    err = -EINVAL;
1504    if (IS_MNT_UNBINDABLE(old_path.mnt))
1505        goto out;
1506
1507    if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1508        goto out;
1509
1510    err = -ENOMEM;
1511    if (recurse)
1512        mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1513    else
1514        mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1515
1516    if (!mnt)
1517        goto out;
1518
1519    err = graft_tree(mnt, path);
1520    if (err) {
1521        LIST_HEAD(umount_list);
1522        spin_lock(&vfsmount_lock);
1523        umount_tree(mnt, 0, &umount_list);
1524        spin_unlock(&vfsmount_lock);
1525        release_mounts(&umount_list);
1526    }
1527
1528out:
1529    up_write(&namespace_sem);
1530    path_put(&old_path);
1531    return err;
1532}
1533
1534static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1535{
1536    int error = 0;
1537    int readonly_request = 0;
1538
1539    if (ms_flags & MS_RDONLY)
1540        readonly_request = 1;
1541    if (readonly_request == __mnt_is_readonly(mnt))
1542        return 0;
1543
1544    if (readonly_request)
1545        error = mnt_make_readonly(mnt);
1546    else
1547        __mnt_unmake_readonly(mnt);
1548    return error;
1549}
1550
1551/*
1552 * change filesystem flags. dir should be a physical root of filesystem.
1553 * If you've mounted a non-root directory somewhere and want to do remount
1554 * on it - tough luck.
1555 */
1556static int do_remount(struct path *path, int flags, int mnt_flags,
1557              void *data)
1558{
1559    int err;
1560    struct super_block *sb = path->mnt->mnt_sb;
1561
1562    if (!capable(CAP_SYS_ADMIN))
1563        return -EPERM;
1564
1565    if (!check_mnt(path->mnt))
1566        return -EINVAL;
1567
1568    if (path->dentry != path->mnt->mnt_root)
1569        return -EINVAL;
1570
1571    down_write(&sb->s_umount);
1572    if (flags & MS_BIND)
1573        err = change_mount_flags(path->mnt, flags);
1574    else
1575        err = do_remount_sb(sb, flags, data, 0);
1576    if (!err) {
1577        spin_lock(&vfsmount_lock);
1578        mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1579        path->mnt->mnt_flags = mnt_flags;
1580        spin_unlock(&vfsmount_lock);
1581    }
1582    up_write(&sb->s_umount);
1583    if (!err) {
1584        security_sb_post_remount(path->mnt, flags, data);
1585
1586        spin_lock(&vfsmount_lock);
1587        touch_mnt_namespace(path->mnt->mnt_ns);
1588        spin_unlock(&vfsmount_lock);
1589    }
1590    return err;
1591}
1592
1593static inline int tree_contains_unbindable(struct vfsmount *mnt)
1594{
1595    struct vfsmount *p;
1596    for (p = mnt; p; p = next_mnt(p, mnt)) {
1597        if (IS_MNT_UNBINDABLE(p))
1598            return 1;
1599    }
1600    return 0;
1601}
1602
1603static int do_move_mount(struct path *path, char *old_name)
1604{
1605    struct path old_path, parent_path;
1606    struct vfsmount *p;
1607    int err = 0;
1608    if (!capable(CAP_SYS_ADMIN))
1609        return -EPERM;
1610    if (!old_name || !*old_name)
1611        return -EINVAL;
1612    err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1613    if (err)
1614        return err;
1615
1616    down_write(&namespace_sem);
1617    while (d_mountpoint(path->dentry) &&
1618           follow_down(path))
1619        ;
1620    err = -EINVAL;
1621    if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1622        goto out;
1623
1624    err = -ENOENT;
1625    mutex_lock(&path->dentry->d_inode->i_mutex);
1626    if (IS_DEADDIR(path->dentry->d_inode))
1627        goto out1;
1628
1629    if (d_unlinked(path->dentry))
1630        goto out1;
1631
1632    err = -EINVAL;
1633    if (old_path.dentry != old_path.mnt->mnt_root)
1634        goto out1;
1635
1636    if (old_path.mnt == old_path.mnt->mnt_parent)
1637        goto out1;
1638
1639    if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1640          S_ISDIR(old_path.dentry->d_inode->i_mode))
1641        goto out1;
1642    /*
1643     * Don't move a mount residing in a shared parent.
1644     */
1645    if (old_path.mnt->mnt_parent &&
1646        IS_MNT_SHARED(old_path.mnt->mnt_parent))
1647        goto out1;
1648    /*
1649     * Don't move a mount tree containing unbindable mounts to a destination
1650     * mount which is shared.
1651     */
1652    if (IS_MNT_SHARED(path->mnt) &&
1653        tree_contains_unbindable(old_path.mnt))
1654        goto out1;
1655    err = -ELOOP;
1656    for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1657        if (p == old_path.mnt)
1658            goto out1;
1659
1660    err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1661    if (err)
1662        goto out1;
1663
1664    /* if the mount is moved, it should no longer be expire
1665     * automatically */
1666    list_del_init(&old_path.mnt->mnt_expire);
1667out1:
1668    mutex_unlock(&path->dentry->d_inode->i_mutex);
1669out:
1670    up_write(&namespace_sem);
1671    if (!err)
1672        path_put(&parent_path);
1673    path_put(&old_path);
1674    return err;
1675}
1676
1677/*
1678 * create a new mount for userspace and request it to be added into the
1679 * namespace's tree
1680 */
1681static int do_new_mount(struct path *path, char *type, int flags,
1682            int mnt_flags, char *name, void *data)
1683{
1684    struct vfsmount *mnt;
1685
1686    if (!type)
1687        return -EINVAL;
1688
1689    /* we need capabilities... */
1690    if (!capable(CAP_SYS_ADMIN))
1691        return -EPERM;
1692
1693    lock_kernel();
1694    mnt = do_kern_mount(type, flags, name, data);
1695    unlock_kernel();
1696    if (IS_ERR(mnt))
1697        return PTR_ERR(mnt);
1698
1699    return do_add_mount(mnt, path, mnt_flags, NULL);
1700}
1701
1702/*
1703 * add a mount into a namespace's mount tree
1704 * - provide the option of adding the new mount to an expiration list
1705 */
1706int do_add_mount(struct vfsmount *newmnt, struct path *path,
1707         int mnt_flags, struct list_head *fslist)
1708{
1709    int err;
1710
1711    mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1712
1713    down_write(&namespace_sem);
1714    /* Something was mounted here while we slept */
1715    while (d_mountpoint(path->dentry) &&
1716           follow_down(path))
1717        ;
1718    err = -EINVAL;
1719    if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1720        goto unlock;
1721
1722    /* Refuse the same filesystem on the same mount point */
1723    err = -EBUSY;
1724    if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1725        path->mnt->mnt_root == path->dentry)
1726        goto unlock;
1727
1728    err = -EINVAL;
1729    if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1730        goto unlock;
1731
1732    newmnt->mnt_flags = mnt_flags;
1733    if ((err = graft_tree(newmnt, path)))
1734        goto unlock;
1735
1736    if (fslist) /* add to the specified expiration list */
1737        list_add_tail(&newmnt->mnt_expire, fslist);
1738
1739    up_write(&namespace_sem);
1740    return 0;
1741
1742unlock:
1743    up_write(&namespace_sem);
1744    mntput(newmnt);
1745    return err;
1746}
1747
1748EXPORT_SYMBOL_GPL(do_add_mount);
1749
1750/*
1751 * process a list of expirable mountpoints with the intent of discarding any
1752 * mountpoints that aren't in use and haven't been touched since last we came
1753 * here
1754 */
1755void mark_mounts_for_expiry(struct list_head *mounts)
1756{
1757    struct vfsmount *mnt, *next;
1758    LIST_HEAD(graveyard);
1759    LIST_HEAD(umounts);
1760
1761    if (list_empty(mounts))
1762        return;
1763
1764    down_write(&namespace_sem);
1765    spin_lock(&vfsmount_lock);
1766
1767    /* extract from the expiration list every vfsmount that matches the
1768     * following criteria:
1769     * - only referenced by its parent vfsmount
1770     * - still marked for expiry (marked on the last call here; marks are
1771     * cleared by mntput())
1772     */
1773    list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1774        if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1775            propagate_mount_busy(mnt, 1))
1776            continue;
1777        list_move(&mnt->mnt_expire, &graveyard);
1778    }
1779    while (!list_empty(&graveyard)) {
1780        mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1781        touch_mnt_namespace(mnt->mnt_ns);
1782        umount_tree(mnt, 1, &umounts);
1783    }
1784    spin_unlock(&vfsmount_lock);
1785    up_write(&namespace_sem);
1786
1787    release_mounts(&umounts);
1788}
1789
1790EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1791
1792/*
1793 * Ripoff of 'select_parent()'
1794 *
1795 * search the list of submounts for a given mountpoint, and move any
1796 * shrinkable submounts to the 'graveyard' list.
1797 */
1798static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1799{
1800    struct vfsmount *this_parent = parent;
1801    struct list_head *next;
1802    int found = 0;
1803
1804repeat:
1805    next = this_parent->mnt_mounts.next;
1806resume:
1807    while (next != &this_parent->mnt_mounts) {
1808        struct list_head *tmp = next;
1809        struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1810
1811        next = tmp->next;
1812        if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1813            continue;
1814        /*
1815         * Descend a level if the d_mounts list is non-empty.
1816         */
1817        if (!list_empty(&mnt->mnt_mounts)) {
1818            this_parent = mnt;
1819            goto repeat;
1820        }
1821
1822        if (!propagate_mount_busy(mnt, 1)) {
1823            list_move_tail(&mnt->mnt_expire, graveyard);
1824            found++;
1825        }
1826    }
1827    /*
1828     * All done at this level ... ascend and resume the search
1829     */
1830    if (this_parent != parent) {
1831        next = this_parent->mnt_child.next;
1832        this_parent = this_parent->mnt_parent;
1833        goto resume;
1834    }
1835    return found;
1836}
1837
1838/*
1839 * process a list of expirable mountpoints with the intent of discarding any
1840 * submounts of a specific parent mountpoint
1841 */
1842static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1843{
1844    LIST_HEAD(graveyard);
1845    struct vfsmount *m;
1846
1847    /* extract submounts of 'mountpoint' from the expiration list */
1848    while (select_submounts(mnt, &graveyard)) {
1849        while (!list_empty(&graveyard)) {
1850            m = list_first_entry(&graveyard, struct vfsmount,
1851                        mnt_expire);
1852            touch_mnt_namespace(m->mnt_ns);
1853            umount_tree(m, 1, umounts);
1854        }
1855    }
1856}
1857
1858/*
1859 * Some copy_from_user() implementations do not return the exact number of
1860 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1861 * Note that this function differs from copy_from_user() in that it will oops
1862 * on bad values of `to', rather than returning a short copy.
1863 */
1864static long exact_copy_from_user(void *to, const void __user * from,
1865                 unsigned long n)
1866{
1867    char *t = to;
1868    const char __user *f = from;
1869    char c;
1870
1871    if (!access_ok(VERIFY_READ, from, n))
1872        return n;
1873
1874    while (n) {
1875        if (__get_user(c, f)) {
1876            memset(t, 0, n);
1877            break;
1878        }
1879        *t++ = c;
1880        f++;
1881        n--;
1882    }
1883    return n;
1884}
1885
1886int copy_mount_options(const void __user * data, unsigned long *where)
1887{
1888    int i;
1889    unsigned long page;
1890    unsigned long size;
1891
1892    *where = 0;
1893    if (!data)
1894        return 0;
1895
1896    if (!(page = __get_free_page(GFP_KERNEL)))
1897        return -ENOMEM;
1898
1899    /* We only care that *some* data at the address the user
1900     * gave us is valid. Just in case, we'll zero
1901     * the remainder of the page.
1902     */
1903    /* copy_from_user cannot cross TASK_SIZE ! */
1904    size = TASK_SIZE - (unsigned long)data;
1905    if (size > PAGE_SIZE)
1906        size = PAGE_SIZE;
1907
1908    i = size - exact_copy_from_user((void *)page, data, size);
1909    if (!i) {
1910        free_page(page);
1911        return -EFAULT;
1912    }
1913    if (i != PAGE_SIZE)
1914        memset((char *)page + i, 0, PAGE_SIZE - i);
1915    *where = page;
1916    return 0;
1917}
1918
1919int copy_mount_string(const void __user *data, char **where)
1920{
1921    char *tmp;
1922
1923    if (!data) {
1924        *where = NULL;
1925        return 0;
1926    }
1927
1928    tmp = strndup_user(data, PAGE_SIZE);
1929    if (IS_ERR(tmp))
1930        return PTR_ERR(tmp);
1931
1932    *where = tmp;
1933    return 0;
1934}
1935
1936/*
1937 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1938 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1939 *
1940 * data is a (void *) that can point to any structure up to
1941 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1942 * information (or be NULL).
1943 *
1944 * Pre-0.97 versions of mount() didn't have a flags word.
1945 * When the flags word was introduced its top half was required
1946 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1947 * Therefore, if this magic number is present, it carries no information
1948 * and must be discarded.
1949 */
1950long do_mount(char *dev_name, char *dir_name, char *type_page,
1951          unsigned long flags, void *data_page)
1952{
1953    struct path path;
1954    int retval = 0;
1955    int mnt_flags = 0;
1956
1957    /* Discard magic */
1958    if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1959        flags &= ~MS_MGC_MSK;
1960
1961    /* Basic sanity checks */
1962
1963    if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1964        return -EINVAL;
1965
1966    if (data_page)
1967        ((char *)data_page)[PAGE_SIZE - 1] = 0;
1968
1969    /* ... and get the mountpoint */
1970    retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1971    if (retval)
1972        return retval;
1973
1974    retval = security_sb_mount(dev_name, &path,
1975                   type_page, flags, data_page);
1976    if (retval)
1977        goto dput_out;
1978
1979    /* Default to relatime unless overriden */
1980    if (!(flags & MS_NOATIME))
1981        mnt_flags |= MNT_RELATIME;
1982
1983    /* Separate the per-mountpoint flags */
1984    if (flags & MS_NOSUID)
1985        mnt_flags |= MNT_NOSUID;
1986    if (flags & MS_NODEV)
1987        mnt_flags |= MNT_NODEV;
1988    if (flags & MS_NOEXEC)
1989        mnt_flags |= MNT_NOEXEC;
1990    if (flags & MS_NOATIME)
1991        mnt_flags |= MNT_NOATIME;
1992    if (flags & MS_NODIRATIME)
1993        mnt_flags |= MNT_NODIRATIME;
1994    if (flags & MS_STRICTATIME)
1995        mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
1996    if (flags & MS_RDONLY)
1997        mnt_flags |= MNT_READONLY;
1998
1999    flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
2000           MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2001           MS_STRICTATIME);
2002
2003    if (flags & MS_REMOUNT)
2004        retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2005                    data_page);
2006    else if (flags & MS_BIND)
2007        retval = do_loopback(&path, dev_name, flags & MS_REC);
2008    else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2009        retval = do_change_type(&path, flags);
2010    else if (flags & MS_MOVE)
2011        retval = do_move_mount(&path, dev_name);
2012    else
2013        retval = do_new_mount(&path, type_page, flags, mnt_flags,
2014                      dev_name, data_page);
2015dput_out:
2016    path_put(&path);
2017    return retval;
2018}
2019
2020static struct mnt_namespace *alloc_mnt_ns(void)
2021{
2022    struct mnt_namespace *new_ns;
2023
2024    new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2025    if (!new_ns)
2026        return ERR_PTR(-ENOMEM);
2027    atomic_set(&new_ns->count, 1);
2028    new_ns->root = NULL;
2029    INIT_LIST_HEAD(&new_ns->list);
2030    init_waitqueue_head(&new_ns->poll);
2031    new_ns->event = 0;
2032    return new_ns;
2033}
2034
2035/*
2036 * Allocate a new namespace structure and populate it with contents
2037 * copied from the namespace of the passed in task structure.
2038 */
2039static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2040        struct fs_struct *fs)
2041{
2042    struct mnt_namespace *new_ns;
2043    struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2044    struct vfsmount *p, *q;
2045
2046    new_ns = alloc_mnt_ns();
2047    if (IS_ERR(new_ns))
2048        return new_ns;
2049
2050    down_write(&namespace_sem);
2051    /* First pass: copy the tree topology */
2052    new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2053                    CL_COPY_ALL | CL_EXPIRE);
2054    if (!new_ns->root) {
2055        up_write(&namespace_sem);
2056        kfree(new_ns);
2057        return ERR_PTR(-ENOMEM);
2058    }
2059    spin_lock(&vfsmount_lock);
2060    list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2061    spin_unlock(&vfsmount_lock);
2062
2063    /*
2064     * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2065     * as belonging to new namespace. We have already acquired a private
2066     * fs_struct, so tsk->fs->lock is not needed.
2067     */
2068    p = mnt_ns->root;
2069    q = new_ns->root;
2070    while (p) {
2071        q->mnt_ns = new_ns;
2072        if (fs) {
2073            if (p == fs->root.mnt) {
2074                rootmnt = p;
2075                fs->root.mnt = mntget(q);
2076            }
2077            if (p == fs->pwd.mnt) {
2078                pwdmnt = p;
2079                fs->pwd.mnt = mntget(q);
2080            }
2081        }
2082        p = next_mnt(p, mnt_ns->root);
2083        q = next_mnt(q, new_ns->root);
2084    }
2085    up_write(&namespace_sem);
2086
2087    if (rootmnt)
2088        mntput(rootmnt);
2089    if (pwdmnt)
2090        mntput(pwdmnt);
2091
2092    return new_ns;
2093}
2094
2095struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2096        struct fs_struct *new_fs)
2097{
2098    struct mnt_namespace *new_ns;
2099
2100    BUG_ON(!ns);
2101    get_mnt_ns(ns);
2102
2103    if (!(flags & CLONE_NEWNS))
2104        return ns;
2105
2106    new_ns = dup_mnt_ns(ns, new_fs);
2107
2108    put_mnt_ns(ns);
2109    return new_ns;
2110}
2111
2112/**
2113 * create_mnt_ns - creates a private namespace and adds a root filesystem
2114 * @mnt: pointer to the new root filesystem mountpoint
2115 */
2116struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2117{
2118    struct mnt_namespace *new_ns;
2119
2120    new_ns = alloc_mnt_ns();
2121    if (!IS_ERR(new_ns)) {
2122        mnt->mnt_ns = new_ns;
2123        new_ns->root = mnt;
2124        list_add(&new_ns->list, &new_ns->root->mnt_list);
2125    }
2126    return new_ns;
2127}
2128EXPORT_SYMBOL(create_mnt_ns);
2129
2130SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2131        char __user *, type, unsigned long, flags, void __user *, data)
2132{
2133    int ret;
2134    char *kernel_type;
2135    char *kernel_dir;
2136    char *kernel_dev;
2137    unsigned long data_page;
2138
2139    ret = copy_mount_string(type, &kernel_type);
2140    if (ret < 0)
2141        goto out_type;
2142
2143    kernel_dir = getname(dir_name);
2144    if (IS_ERR(kernel_dir)) {
2145        ret = PTR_ERR(kernel_dir);
2146        goto out_dir;
2147    }
2148
2149    ret = copy_mount_string(dev_name, &kernel_dev);
2150    if (ret < 0)
2151        goto out_dev;
2152
2153    ret = copy_mount_options(data, &data_page);
2154    if (ret < 0)
2155        goto out_data;
2156
2157    ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2158        (void *) data_page);
2159
2160    free_page(data_page);
2161out_data:
2162    kfree(kernel_dev);
2163out_dev:
2164    putname(kernel_dir);
2165out_dir:
2166    kfree(kernel_type);
2167out_type:
2168    return ret;
2169}
2170
2171/*
2172 * pivot_root Semantics:
2173 * Moves the root file system of the current process to the directory put_old,
2174 * makes new_root as the new root file system of the current process, and sets
2175 * root/cwd of all processes which had them on the current root to new_root.
2176 *
2177 * Restrictions:
2178 * The new_root and put_old must be directories, and must not be on the
2179 * same file system as the current process root. The put_old must be
2180 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2181 * pointed to by put_old must yield the same directory as new_root. No other
2182 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2183 *
2184 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2185 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2186 * in this situation.
2187 *
2188 * Notes:
2189 * - we don't move root/cwd if they are not at the root (reason: if something
2190 * cared enough to change them, it's probably wrong to force them elsewhere)
2191 * - it's okay to pick a root that isn't the root of a file system, e.g.
2192 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2193 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2194 * first.
2195 */
2196SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2197        const char __user *, put_old)
2198{
2199    struct vfsmount *tmp;
2200    struct path new, old, parent_path, root_parent, root;
2201    int error;
2202
2203    if (!capable(CAP_SYS_ADMIN))
2204        return -EPERM;
2205
2206    error = user_path_dir(new_root, &new);
2207    if (error)
2208        goto out0;
2209    error = -EINVAL;
2210    if (!check_mnt(new.mnt))
2211        goto out1;
2212
2213    error = user_path_dir(put_old, &old);
2214    if (error)
2215        goto out1;
2216
2217    error = security_sb_pivotroot(&old, &new);
2218    if (error) {
2219        path_put(&old);
2220        goto out1;
2221    }
2222
2223    read_lock(&current->fs->lock);
2224    root = current->fs->root;
2225    path_get(&current->fs->root);
2226    read_unlock(&current->fs->lock);
2227    down_write(&namespace_sem);
2228    mutex_lock(&old.dentry->d_inode->i_mutex);
2229    error = -EINVAL;
2230    if (IS_MNT_SHARED(old.mnt) ||
2231        IS_MNT_SHARED(new.mnt->mnt_parent) ||
2232        IS_MNT_SHARED(root.mnt->mnt_parent))
2233        goto out2;
2234    if (!check_mnt(root.mnt))
2235        goto out2;
2236    error = -ENOENT;
2237    if (IS_DEADDIR(new.dentry->d_inode))
2238        goto out2;
2239    if (d_unlinked(new.dentry))
2240        goto out2;
2241    if (d_unlinked(old.dentry))
2242        goto out2;
2243    error = -EBUSY;
2244    if (new.mnt == root.mnt ||
2245        old.mnt == root.mnt)
2246        goto out2; /* loop, on the same file system */
2247    error = -EINVAL;
2248    if (root.mnt->mnt_root != root.dentry)
2249        goto out2; /* not a mountpoint */
2250    if (root.mnt->mnt_parent == root.mnt)
2251        goto out2; /* not attached */
2252    if (new.mnt->mnt_root != new.dentry)
2253        goto out2; /* not a mountpoint */
2254    if (new.mnt->mnt_parent == new.mnt)
2255        goto out2; /* not attached */
2256    /* make sure we can reach put_old from new_root */
2257    tmp = old.mnt;
2258    spin_lock(&vfsmount_lock);
2259    if (tmp != new.mnt) {
2260        for (;;) {
2261            if (tmp->mnt_parent == tmp)
2262                goto out3; /* already mounted on put_old */
2263            if (tmp->mnt_parent == new.mnt)
2264                break;
2265            tmp = tmp->mnt_parent;
2266        }
2267        if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2268            goto out3;
2269    } else if (!is_subdir(old.dentry, new.dentry))
2270        goto out3;
2271    detach_mnt(new.mnt, &parent_path);
2272    detach_mnt(root.mnt, &root_parent);
2273    /* mount old root on put_old */
2274    attach_mnt(root.mnt, &old);
2275    /* mount new_root on / */
2276    attach_mnt(new.mnt, &root_parent);
2277    touch_mnt_namespace(current->nsproxy->mnt_ns);
2278    spin_unlock(&vfsmount_lock);
2279    chroot_fs_refs(&root, &new);
2280    security_sb_post_pivotroot(&root, &new);
2281    error = 0;
2282    path_put(&root_parent);
2283    path_put(&parent_path);
2284out2:
2285    mutex_unlock(&old.dentry->d_inode->i_mutex);
2286    up_write(&namespace_sem);
2287    path_put(&root);
2288    path_put(&old);
2289out1:
2290    path_put(&new);
2291out0:
2292    return error;
2293out3:
2294    spin_unlock(&vfsmount_lock);
2295    goto out2;
2296}
2297
2298static void __init init_mount_tree(void)
2299{
2300    struct vfsmount *mnt;
2301    struct mnt_namespace *ns;
2302    struct path root;
2303
2304    mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2305    if (IS_ERR(mnt))
2306        panic("Can't create rootfs");
2307    ns = create_mnt_ns(mnt);
2308    if (IS_ERR(ns))
2309        panic("Can't allocate initial namespace");
2310
2311    init_task.nsproxy->mnt_ns = ns;
2312    get_mnt_ns(ns);
2313
2314    root.mnt = ns->root;
2315    root.dentry = ns->root->mnt_root;
2316
2317    set_fs_pwd(current->fs, &root);
2318    set_fs_root(current->fs, &root);
2319}
2320
2321void __init mnt_init(void)
2322{
2323    unsigned u;
2324    int err;
2325
2326    init_rwsem(&namespace_sem);
2327
2328    mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2329            0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2330
2331    mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2332
2333    if (!mount_hashtable)
2334        panic("Failed to allocate mount hash table\n");
2335
2336    printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2337
2338    for (u = 0; u < HASH_SIZE; u++)
2339        INIT_LIST_HEAD(&mount_hashtable[u]);
2340
2341    err = sysfs_init();
2342    if (err)
2343        printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2344            __func__, err);
2345    fs_kobj = kobject_create_and_add("fs", NULL);
2346    if (!fs_kobj)
2347        printk(KERN_WARNING "%s: kobj create error\n", __func__);
2348    init_rootfs();
2349    init_mount_tree();
2350}
2351
2352void put_mnt_ns(struct mnt_namespace *ns)
2353{
2354    LIST_HEAD(umount_list);
2355
2356    if (!atomic_dec_and_test(&ns->count))
2357        return;
2358    down_write(&namespace_sem);
2359    spin_lock(&vfsmount_lock);
2360    umount_tree(ns->root, 0, &umount_list);
2361    spin_unlock(&vfsmount_lock);
2362    up_write(&namespace_sem);
2363    release_mounts(&umount_list);
2364    kfree(ns);
2365}
2366EXPORT_SYMBOL(put_mnt_ns);
2367

Archive Download this file



interactive