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/spinlock.h>
15#include <linux/percpu.h>
16#include <linux/init.h>
17#include <linux/kernel.h>
18#include <linux/acct.h>
19#include <linux/capability.h>
20#include <linux/cpumask.h>
21#include <linux/module.h>
22#include <linux/sysfs.h>
23#include <linux/seq_file.h>
24#include <linux/mnt_namespace.h>
25#include <linux/namei.h>
26#include <linux/nsproxy.h>
27#include <linux/security.h>
28#include <linux/mount.h>
29#include <linux/ramfs.h>
30#include <linux/log2.h>
31#include <linux/idr.h>
32#include <linux/fs_struct.h>
33#include <linux/fsnotify.h>
34#include <asm/uaccess.h>
35#include <asm/unistd.h>
36#include "pnode.h"
37#include "internal.h"
38
39#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
40#define HASH_SIZE (1UL << HASH_SHIFT)
41
42static int event;
43static DEFINE_IDA(mnt_id_ida);
44static DEFINE_IDA(mnt_group_ida);
45static DEFINE_SPINLOCK(mnt_id_lock);
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
57/*
58 * vfsmount lock may be taken for read to prevent changes to the
59 * vfsmount hash, ie. during mountpoint lookups or walking back
60 * up the tree.
61 *
62 * It should be taken for write in all cases where the vfsmount
63 * tree or hash is modified or when a vfsmount structure is modified.
64 */
65DEFINE_BRLOCK(vfsmount_lock);
66
67static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
68{
69    unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
70    tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
71    tmp = tmp + (tmp >> HASH_SHIFT);
72    return tmp & (HASH_SIZE - 1);
73}
74
75#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
76
77/*
78 * allocation is serialized by namespace_sem, but we need the spinlock to
79 * serialize with freeing.
80 */
81static int mnt_alloc_id(struct vfsmount *mnt)
82{
83    int res;
84
85retry:
86    ida_pre_get(&mnt_id_ida, GFP_KERNEL);
87    spin_lock(&mnt_id_lock);
88    res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
89    if (!res)
90        mnt_id_start = mnt->mnt_id + 1;
91    spin_unlock(&mnt_id_lock);
92    if (res == -EAGAIN)
93        goto retry;
94
95    return res;
96}
97
98static void mnt_free_id(struct vfsmount *mnt)
99{
100    int id = mnt->mnt_id;
101    spin_lock(&mnt_id_lock);
102    ida_remove(&mnt_id_ida, id);
103    if (mnt_id_start > id)
104        mnt_id_start = id;
105    spin_unlock(&mnt_id_lock);
106}
107
108/*
109 * Allocate a new peer group ID
110 *
111 * mnt_group_ida is protected by namespace_sem
112 */
113static int mnt_alloc_group_id(struct vfsmount *mnt)
114{
115    int res;
116
117    if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
118        return -ENOMEM;
119
120    res = ida_get_new_above(&mnt_group_ida,
121                mnt_group_start,
122                &mnt->mnt_group_id);
123    if (!res)
124        mnt_group_start = mnt->mnt_group_id + 1;
125
126    return res;
127}
128
129/*
130 * Release a peer group ID
131 */
132void mnt_release_group_id(struct vfsmount *mnt)
133{
134    int id = mnt->mnt_group_id;
135    ida_remove(&mnt_group_ida, id);
136    if (mnt_group_start > id)
137        mnt_group_start = id;
138    mnt->mnt_group_id = 0;
139}
140
141/*
142 * vfsmount lock must be held for read
143 */
144static inline void mnt_add_count(struct vfsmount *mnt, int n)
145{
146#ifdef CONFIG_SMP
147    this_cpu_add(mnt->mnt_pcp->mnt_count, n);
148#else
149    preempt_disable();
150    mnt->mnt_count += n;
151    preempt_enable();
152#endif
153}
154
155static inline void mnt_set_count(struct vfsmount *mnt, int n)
156{
157#ifdef CONFIG_SMP
158    this_cpu_write(mnt->mnt_pcp->mnt_count, n);
159#else
160    mnt->mnt_count = n;
161#endif
162}
163
164/*
165 * vfsmount lock must be held for read
166 */
167static inline void mnt_inc_count(struct vfsmount *mnt)
168{
169    mnt_add_count(mnt, 1);
170}
171
172/*
173 * vfsmount lock must be held for read
174 */
175static inline void mnt_dec_count(struct vfsmount *mnt)
176{
177    mnt_add_count(mnt, -1);
178}
179
180/*
181 * vfsmount lock must be held for write
182 */
183unsigned int mnt_get_count(struct vfsmount *mnt)
184{
185#ifdef CONFIG_SMP
186    unsigned int count = 0;
187    int cpu;
188
189    for_each_possible_cpu(cpu) {
190        count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
191    }
192
193    return count;
194#else
195    return mnt->mnt_count;
196#endif
197}
198
199static struct vfsmount *alloc_vfsmnt(const char *name)
200{
201    struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
202    if (mnt) {
203        int err;
204
205        err = mnt_alloc_id(mnt);
206        if (err)
207            goto out_free_cache;
208
209        if (name) {
210            mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
211            if (!mnt->mnt_devname)
212                goto out_free_id;
213        }
214
215#ifdef CONFIG_SMP
216        mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
217        if (!mnt->mnt_pcp)
218            goto out_free_devname;
219
220        this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
221#else
222        mnt->mnt_count = 1;
223        mnt->mnt_writers = 0;
224#endif
225
226        INIT_LIST_HEAD(&mnt->mnt_hash);
227        INIT_LIST_HEAD(&mnt->mnt_child);
228        INIT_LIST_HEAD(&mnt->mnt_mounts);
229        INIT_LIST_HEAD(&mnt->mnt_list);
230        INIT_LIST_HEAD(&mnt->mnt_expire);
231        INIT_LIST_HEAD(&mnt->mnt_share);
232        INIT_LIST_HEAD(&mnt->mnt_slave_list);
233        INIT_LIST_HEAD(&mnt->mnt_slave);
234#ifdef CONFIG_FSNOTIFY
235        INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
236#endif
237    }
238    return mnt;
239
240#ifdef CONFIG_SMP
241out_free_devname:
242    kfree(mnt->mnt_devname);
243#endif
244out_free_id:
245    mnt_free_id(mnt);
246out_free_cache:
247    kmem_cache_free(mnt_cache, mnt);
248    return NULL;
249}
250
251/*
252 * Most r/o checks on a fs are for operations that take
253 * discrete amounts of time, like a write() or unlink().
254 * We must keep track of when those operations start
255 * (for permission checks) and when they end, so that
256 * we can determine when writes are able to occur to
257 * a filesystem.
258 */
259/*
260 * __mnt_is_readonly: check whether a mount is read-only
261 * @mnt: the mount to check for its write status
262 *
263 * This shouldn't be used directly ouside of the VFS.
264 * It does not guarantee that the filesystem will stay
265 * r/w, just that it is right *now*. This can not and
266 * should not be used in place of IS_RDONLY(inode).
267 * mnt_want/drop_write() will _keep_ the filesystem
268 * r/w.
269 */
270int __mnt_is_readonly(struct vfsmount *mnt)
271{
272    if (mnt->mnt_flags & MNT_READONLY)
273        return 1;
274    if (mnt->mnt_sb->s_flags & MS_RDONLY)
275        return 1;
276    return 0;
277}
278EXPORT_SYMBOL_GPL(__mnt_is_readonly);
279
280static inline void mnt_inc_writers(struct vfsmount *mnt)
281{
282#ifdef CONFIG_SMP
283    this_cpu_inc(mnt->mnt_pcp->mnt_writers);
284#else
285    mnt->mnt_writers++;
286#endif
287}
288
289static inline void mnt_dec_writers(struct vfsmount *mnt)
290{
291#ifdef CONFIG_SMP
292    this_cpu_dec(mnt->mnt_pcp->mnt_writers);
293#else
294    mnt->mnt_writers--;
295#endif
296}
297
298static unsigned int mnt_get_writers(struct vfsmount *mnt)
299{
300#ifdef CONFIG_SMP
301    unsigned int count = 0;
302    int cpu;
303
304    for_each_possible_cpu(cpu) {
305        count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
306    }
307
308    return count;
309#else
310    return mnt->mnt_writers;
311#endif
312}
313
314/*
315 * Most r/o checks on a fs are for operations that take
316 * discrete amounts of time, like a write() or unlink().
317 * We must keep track of when those operations start
318 * (for permission checks) and when they end, so that
319 * we can determine when writes are able to occur to
320 * a filesystem.
321 */
322/**
323 * mnt_want_write - get write access to a mount
324 * @mnt: the mount on which to take a write
325 *
326 * This tells the low-level filesystem that a write is
327 * about to be performed to it, and makes sure that
328 * writes are allowed before returning success. When
329 * the write operation is finished, mnt_drop_write()
330 * must be called. This is effectively a refcount.
331 */
332int mnt_want_write(struct vfsmount *mnt)
333{
334    int ret = 0;
335
336    preempt_disable();
337    mnt_inc_writers(mnt);
338    /*
339     * The store to mnt_inc_writers must be visible before we pass
340     * MNT_WRITE_HOLD loop below, so that the slowpath can see our
341     * incremented count after it has set MNT_WRITE_HOLD.
342     */
343    smp_mb();
344    while (mnt->mnt_flags & MNT_WRITE_HOLD)
345        cpu_relax();
346    /*
347     * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
348     * be set to match its requirements. So we must not load that until
349     * MNT_WRITE_HOLD is cleared.
350     */
351    smp_rmb();
352    if (__mnt_is_readonly(mnt)) {
353        mnt_dec_writers(mnt);
354        ret = -EROFS;
355        goto out;
356    }
357out:
358    preempt_enable();
359    return ret;
360}
361EXPORT_SYMBOL_GPL(mnt_want_write);
362
363/**
364 * mnt_clone_write - get write access to a mount
365 * @mnt: the mount on which to take a write
366 *
367 * This is effectively like mnt_want_write, except
368 * it must only be used to take an extra write reference
369 * on a mountpoint that we already know has a write reference
370 * on it. This allows some optimisation.
371 *
372 * After finished, mnt_drop_write must be called as usual to
373 * drop the reference.
374 */
375int mnt_clone_write(struct vfsmount *mnt)
376{
377    /* superblock may be r/o */
378    if (__mnt_is_readonly(mnt))
379        return -EROFS;
380    preempt_disable();
381    mnt_inc_writers(mnt);
382    preempt_enable();
383    return 0;
384}
385EXPORT_SYMBOL_GPL(mnt_clone_write);
386
387/**
388 * mnt_want_write_file - get write access to a file's mount
389 * @file: the file who's mount on which to take a write
390 *
391 * This is like mnt_want_write, but it takes a file and can
392 * do some optimisations if the file is open for write already
393 */
394int mnt_want_write_file(struct file *file)
395{
396    struct inode *inode = file->f_dentry->d_inode;
397    if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
398        return mnt_want_write(file->f_path.mnt);
399    else
400        return mnt_clone_write(file->f_path.mnt);
401}
402EXPORT_SYMBOL_GPL(mnt_want_write_file);
403
404/**
405 * mnt_drop_write - give up write access to a mount
406 * @mnt: the mount on which to give up write access
407 *
408 * Tells the low-level filesystem that we are done
409 * performing writes to it. Must be matched with
410 * mnt_want_write() call above.
411 */
412void mnt_drop_write(struct vfsmount *mnt)
413{
414    preempt_disable();
415    mnt_dec_writers(mnt);
416    preempt_enable();
417}
418EXPORT_SYMBOL_GPL(mnt_drop_write);
419
420static int mnt_make_readonly(struct vfsmount *mnt)
421{
422    int ret = 0;
423
424    br_write_lock(vfsmount_lock);
425    mnt->mnt_flags |= MNT_WRITE_HOLD;
426    /*
427     * After storing MNT_WRITE_HOLD, we'll read the counters. This store
428     * should be visible before we do.
429     */
430    smp_mb();
431
432    /*
433     * With writers on hold, if this value is zero, then there are
434     * definitely no active writers (although held writers may subsequently
435     * increment the count, they'll have to wait, and decrement it after
436     * seeing MNT_READONLY).
437     *
438     * It is OK to have counter incremented on one CPU and decremented on
439     * another: the sum will add up correctly. The danger would be when we
440     * sum up each counter, if we read a counter before it is incremented,
441     * but then read another CPU's count which it has been subsequently
442     * decremented from -- we would see more decrements than we should.
443     * MNT_WRITE_HOLD protects against this scenario, because
444     * mnt_want_write first increments count, then smp_mb, then spins on
445     * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
446     * we're counting up here.
447     */
448    if (mnt_get_writers(mnt) > 0)
449        ret = -EBUSY;
450    else
451        mnt->mnt_flags |= MNT_READONLY;
452    /*
453     * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
454     * that become unheld will see MNT_READONLY.
455     */
456    smp_wmb();
457    mnt->mnt_flags &= ~MNT_WRITE_HOLD;
458    br_write_unlock(vfsmount_lock);
459    return ret;
460}
461
462static void __mnt_unmake_readonly(struct vfsmount *mnt)
463{
464    br_write_lock(vfsmount_lock);
465    mnt->mnt_flags &= ~MNT_READONLY;
466    br_write_unlock(vfsmount_lock);
467}
468
469static void free_vfsmnt(struct vfsmount *mnt)
470{
471    kfree(mnt->mnt_devname);
472    mnt_free_id(mnt);
473#ifdef CONFIG_SMP
474    free_percpu(mnt->mnt_pcp);
475#endif
476    kmem_cache_free(mnt_cache, mnt);
477}
478
479/*
480 * find the first or last mount at @dentry on vfsmount @mnt depending on
481 * @dir. If @dir is set return the first mount else return the last mount.
482 * vfsmount_lock must be held for read or write.
483 */
484struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
485                  int dir)
486{
487    struct list_head *head = mount_hashtable + hash(mnt, dentry);
488    struct list_head *tmp = head;
489    struct vfsmount *p, *found = NULL;
490
491    for (;;) {
492        tmp = dir ? tmp->next : tmp->prev;
493        p = NULL;
494        if (tmp == head)
495            break;
496        p = list_entry(tmp, struct vfsmount, mnt_hash);
497        if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
498            found = p;
499            break;
500        }
501    }
502    return found;
503}
504
505/*
506 * lookup_mnt increments the ref count before returning
507 * the vfsmount struct.
508 */
509struct vfsmount *lookup_mnt(struct path *path)
510{
511    struct vfsmount *child_mnt;
512
513    br_read_lock(vfsmount_lock);
514    if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
515        mntget(child_mnt);
516    br_read_unlock(vfsmount_lock);
517    return child_mnt;
518}
519
520static inline int check_mnt(struct vfsmount *mnt)
521{
522    return mnt->mnt_ns == current->nsproxy->mnt_ns;
523}
524
525/*
526 * vfsmount lock must be held for write
527 */
528static void touch_mnt_namespace(struct mnt_namespace *ns)
529{
530    if (ns) {
531        ns->event = ++event;
532        wake_up_interruptible(&ns->poll);
533    }
534}
535
536/*
537 * vfsmount lock must be held for write
538 */
539static void __touch_mnt_namespace(struct mnt_namespace *ns)
540{
541    if (ns && ns->event != event) {
542        ns->event = event;
543        wake_up_interruptible(&ns->poll);
544    }
545}
546
547/*
548 * Clear dentry's mounted state if it has no remaining mounts.
549 * vfsmount_lock must be held for write.
550 */
551static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
552{
553    unsigned u;
554
555    for (u = 0; u < HASH_SIZE; u++) {
556        struct vfsmount *p;
557
558        list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
559            if (p->mnt_mountpoint == dentry)
560                return;
561        }
562    }
563    spin_lock(&dentry->d_lock);
564    dentry->d_flags &= ~DCACHE_MOUNTED;
565    spin_unlock(&dentry->d_lock);
566}
567
568/*
569 * vfsmount lock must be held for write
570 */
571static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
572{
573    old_path->dentry = mnt->mnt_mountpoint;
574    old_path->mnt = mnt->mnt_parent;
575    mnt->mnt_parent = mnt;
576    mnt->mnt_mountpoint = mnt->mnt_root;
577    list_del_init(&mnt->mnt_child);
578    list_del_init(&mnt->mnt_hash);
579    dentry_reset_mounted(old_path->mnt, old_path->dentry);
580}
581
582/*
583 * vfsmount lock must be held for write
584 */
585void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
586            struct vfsmount *child_mnt)
587{
588    child_mnt->mnt_parent = mntget(mnt);
589    child_mnt->mnt_mountpoint = dget(dentry);
590    spin_lock(&dentry->d_lock);
591    dentry->d_flags |= DCACHE_MOUNTED;
592    spin_unlock(&dentry->d_lock);
593}
594
595/*
596 * vfsmount lock must be held for write
597 */
598static void attach_mnt(struct vfsmount *mnt, struct path *path)
599{
600    mnt_set_mountpoint(path->mnt, path->dentry, mnt);
601    list_add_tail(&mnt->mnt_hash, mount_hashtable +
602            hash(path->mnt, path->dentry));
603    list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
604}
605
606static inline void __mnt_make_longterm(struct vfsmount *mnt)
607{
608#ifdef CONFIG_SMP
609    atomic_inc(&mnt->mnt_longterm);
610#endif
611}
612
613/* needs vfsmount lock for write */
614static inline void __mnt_make_shortterm(struct vfsmount *mnt)
615{
616#ifdef CONFIG_SMP
617    atomic_dec(&mnt->mnt_longterm);
618#endif
619}
620
621/*
622 * vfsmount lock must be held for write
623 */
624static void commit_tree(struct vfsmount *mnt)
625{
626    struct vfsmount *parent = mnt->mnt_parent;
627    struct vfsmount *m;
628    LIST_HEAD(head);
629    struct mnt_namespace *n = parent->mnt_ns;
630
631    BUG_ON(parent == mnt);
632
633    list_add_tail(&head, &mnt->mnt_list);
634    list_for_each_entry(m, &head, mnt_list) {
635        m->mnt_ns = n;
636        __mnt_make_longterm(m);
637    }
638
639    list_splice(&head, n->list.prev);
640
641    list_add_tail(&mnt->mnt_hash, mount_hashtable +
642                hash(parent, mnt->mnt_mountpoint));
643    list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
644    touch_mnt_namespace(n);
645}
646
647static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
648{
649    struct list_head *next = p->mnt_mounts.next;
650    if (next == &p->mnt_mounts) {
651        while (1) {
652            if (p == root)
653                return NULL;
654            next = p->mnt_child.next;
655            if (next != &p->mnt_parent->mnt_mounts)
656                break;
657            p = p->mnt_parent;
658        }
659    }
660    return list_entry(next, struct vfsmount, mnt_child);
661}
662
663static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
664{
665    struct list_head *prev = p->mnt_mounts.prev;
666    while (prev != &p->mnt_mounts) {
667        p = list_entry(prev, struct vfsmount, mnt_child);
668        prev = p->mnt_mounts.prev;
669    }
670    return p;
671}
672
673struct vfsmount *
674vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
675{
676    struct vfsmount *mnt;
677    struct dentry *root;
678
679    if (!type)
680        return ERR_PTR(-ENODEV);
681
682    mnt = alloc_vfsmnt(name);
683    if (!mnt)
684        return ERR_PTR(-ENOMEM);
685
686    if (flags & MS_KERNMOUNT)
687        mnt->mnt_flags = MNT_INTERNAL;
688
689    root = mount_fs(type, flags, name, data);
690    if (IS_ERR(root)) {
691        free_vfsmnt(mnt);
692        return ERR_CAST(root);
693    }
694
695    mnt->mnt_root = root;
696    mnt->mnt_sb = root->d_sb;
697    mnt->mnt_mountpoint = mnt->mnt_root;
698    mnt->mnt_parent = mnt;
699    return mnt;
700}
701EXPORT_SYMBOL_GPL(vfs_kern_mount);
702
703static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
704                    int flag)
705{
706    struct super_block *sb = old->mnt_sb;
707    struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
708
709    if (mnt) {
710        if (flag & (CL_SLAVE | CL_PRIVATE))
711            mnt->mnt_group_id = 0; /* not a peer of original */
712        else
713            mnt->mnt_group_id = old->mnt_group_id;
714
715        if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
716            int err = mnt_alloc_group_id(mnt);
717            if (err)
718                goto out_free;
719        }
720
721        mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
722        atomic_inc(&sb->s_active);
723        mnt->mnt_sb = sb;
724        mnt->mnt_root = dget(root);
725        mnt->mnt_mountpoint = mnt->mnt_root;
726        mnt->mnt_parent = mnt;
727
728        if (flag & CL_SLAVE) {
729            list_add(&mnt->mnt_slave, &old->mnt_slave_list);
730            mnt->mnt_master = old;
731            CLEAR_MNT_SHARED(mnt);
732        } else if (!(flag & CL_PRIVATE)) {
733            if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
734                list_add(&mnt->mnt_share, &old->mnt_share);
735            if (IS_MNT_SLAVE(old))
736                list_add(&mnt->mnt_slave, &old->mnt_slave);
737            mnt->mnt_master = old->mnt_master;
738        }
739        if (flag & CL_MAKE_SHARED)
740            set_mnt_shared(mnt);
741
742        /* stick the duplicate mount on the same expiry list
743         * as the original if that was on one */
744        if (flag & CL_EXPIRE) {
745            if (!list_empty(&old->mnt_expire))
746                list_add(&mnt->mnt_expire, &old->mnt_expire);
747        }
748    }
749    return mnt;
750
751 out_free:
752    free_vfsmnt(mnt);
753    return NULL;
754}
755
756static inline void mntfree(struct vfsmount *mnt)
757{
758    struct super_block *sb = mnt->mnt_sb;
759
760    /*
761     * This probably indicates that somebody messed
762     * up a mnt_want/drop_write() pair. If this
763     * happens, the filesystem was probably unable
764     * to make r/w->r/o transitions.
765     */
766    /*
767     * The locking used to deal with mnt_count decrement provides barriers,
768     * so mnt_get_writers() below is safe.
769     */
770    WARN_ON(mnt_get_writers(mnt));
771    fsnotify_vfsmount_delete(mnt);
772    dput(mnt->mnt_root);
773    free_vfsmnt(mnt);
774    deactivate_super(sb);
775}
776
777static void mntput_no_expire(struct vfsmount *mnt)
778{
779put_again:
780#ifdef CONFIG_SMP
781    br_read_lock(vfsmount_lock);
782    if (likely(atomic_read(&mnt->mnt_longterm))) {
783        mnt_dec_count(mnt);
784        br_read_unlock(vfsmount_lock);
785        return;
786    }
787    br_read_unlock(vfsmount_lock);
788
789    br_write_lock(vfsmount_lock);
790    mnt_dec_count(mnt);
791    if (mnt_get_count(mnt)) {
792        br_write_unlock(vfsmount_lock);
793        return;
794    }
795#else
796    mnt_dec_count(mnt);
797    if (likely(mnt_get_count(mnt)))
798        return;
799    br_write_lock(vfsmount_lock);
800#endif
801    if (unlikely(mnt->mnt_pinned)) {
802        mnt_add_count(mnt, mnt->mnt_pinned + 1);
803        mnt->mnt_pinned = 0;
804        br_write_unlock(vfsmount_lock);
805        acct_auto_close_mnt(mnt);
806        goto put_again;
807    }
808    br_write_unlock(vfsmount_lock);
809    mntfree(mnt);
810}
811
812void mntput(struct vfsmount *mnt)
813{
814    if (mnt) {
815        /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
816        if (unlikely(mnt->mnt_expiry_mark))
817            mnt->mnt_expiry_mark = 0;
818        mntput_no_expire(mnt);
819    }
820}
821EXPORT_SYMBOL(mntput);
822
823struct vfsmount *mntget(struct vfsmount *mnt)
824{
825    if (mnt)
826        mnt_inc_count(mnt);
827    return mnt;
828}
829EXPORT_SYMBOL(mntget);
830
831void mnt_pin(struct vfsmount *mnt)
832{
833    br_write_lock(vfsmount_lock);
834    mnt->mnt_pinned++;
835    br_write_unlock(vfsmount_lock);
836}
837EXPORT_SYMBOL(mnt_pin);
838
839void mnt_unpin(struct vfsmount *mnt)
840{
841    br_write_lock(vfsmount_lock);
842    if (mnt->mnt_pinned) {
843        mnt_inc_count(mnt);
844        mnt->mnt_pinned--;
845    }
846    br_write_unlock(vfsmount_lock);
847}
848EXPORT_SYMBOL(mnt_unpin);
849
850static inline void mangle(struct seq_file *m, const char *s)
851{
852    seq_escape(m, s, " \t\n\\");
853}
854
855/*
856 * Simple .show_options callback for filesystems which don't want to
857 * implement more complex mount option showing.
858 *
859 * See also save_mount_options().
860 */
861int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
862{
863    const char *options;
864
865    rcu_read_lock();
866    options = rcu_dereference(mnt->mnt_sb->s_options);
867
868    if (options != NULL && options[0]) {
869        seq_putc(m, ',');
870        mangle(m, options);
871    }
872    rcu_read_unlock();
873
874    return 0;
875}
876EXPORT_SYMBOL(generic_show_options);
877
878/*
879 * If filesystem uses generic_show_options(), this function should be
880 * called from the fill_super() callback.
881 *
882 * The .remount_fs callback usually needs to be handled in a special
883 * way, to make sure, that previous options are not overwritten if the
884 * remount fails.
885 *
886 * Also note, that if the filesystem's .remount_fs function doesn't
887 * reset all options to their default value, but changes only newly
888 * given options, then the displayed options will not reflect reality
889 * any more.
890 */
891void save_mount_options(struct super_block *sb, char *options)
892{
893    BUG_ON(sb->s_options);
894    rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
895}
896EXPORT_SYMBOL(save_mount_options);
897
898void replace_mount_options(struct super_block *sb, char *options)
899{
900    char *old = sb->s_options;
901    rcu_assign_pointer(sb->s_options, options);
902    if (old) {
903        synchronize_rcu();
904        kfree(old);
905    }
906}
907EXPORT_SYMBOL(replace_mount_options);
908
909#ifdef CONFIG_PROC_FS
910/* iterator */
911static void *m_start(struct seq_file *m, loff_t *pos)
912{
913    struct proc_mounts *p = m->private;
914
915    down_read(&namespace_sem);
916    return seq_list_start(&p->ns->list, *pos);
917}
918
919static void *m_next(struct seq_file *m, void *v, loff_t *pos)
920{
921    struct proc_mounts *p = m->private;
922
923    return seq_list_next(v, &p->ns->list, pos);
924}
925
926static void m_stop(struct seq_file *m, void *v)
927{
928    up_read(&namespace_sem);
929}
930
931int mnt_had_events(struct proc_mounts *p)
932{
933    struct mnt_namespace *ns = p->ns;
934    int res = 0;
935
936    br_read_lock(vfsmount_lock);
937    if (p->event != ns->event) {
938        p->event = ns->event;
939        res = 1;
940    }
941    br_read_unlock(vfsmount_lock);
942
943    return res;
944}
945
946struct proc_fs_info {
947    int flag;
948    const char *str;
949};
950
951static int show_sb_opts(struct seq_file *m, struct super_block *sb)
952{
953    static const struct proc_fs_info fs_info[] = {
954        { MS_SYNCHRONOUS, ",sync" },
955        { MS_DIRSYNC, ",dirsync" },
956        { MS_MANDLOCK, ",mand" },
957        { 0, NULL }
958    };
959    const struct proc_fs_info *fs_infop;
960
961    for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
962        if (sb->s_flags & fs_infop->flag)
963            seq_puts(m, fs_infop->str);
964    }
965
966    return security_sb_show_options(m, sb);
967}
968
969static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
970{
971    static const struct proc_fs_info mnt_info[] = {
972        { MNT_NOSUID, ",nosuid" },
973        { MNT_NODEV, ",nodev" },
974        { MNT_NOEXEC, ",noexec" },
975        { MNT_NOATIME, ",noatime" },
976        { MNT_NODIRATIME, ",nodiratime" },
977        { MNT_RELATIME, ",relatime" },
978        { 0, NULL }
979    };
980    const struct proc_fs_info *fs_infop;
981
982    for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
983        if (mnt->mnt_flags & fs_infop->flag)
984            seq_puts(m, fs_infop->str);
985    }
986}
987
988static void show_type(struct seq_file *m, struct super_block *sb)
989{
990    mangle(m, sb->s_type->name);
991    if (sb->s_subtype && sb->s_subtype[0]) {
992        seq_putc(m, '.');
993        mangle(m, sb->s_subtype);
994    }
995}
996
997static int show_vfsmnt(struct seq_file *m, void *v)
998{
999    struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1000    int err = 0;
1001    struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1002
1003    if (mnt->mnt_sb->s_op->show_devname) {
1004        err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1005        if (err)
1006            goto out;
1007    } else {
1008        mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1009    }
1010    seq_putc(m, ' ');
1011    seq_path(m, &mnt_path, " \t\n\\");
1012    seq_putc(m, ' ');
1013    show_type(m, mnt->mnt_sb);
1014    seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
1015    err = show_sb_opts(m, mnt->mnt_sb);
1016    if (err)
1017        goto out;
1018    show_mnt_opts(m, mnt);
1019    if (mnt->mnt_sb->s_op->show_options)
1020        err = mnt->mnt_sb->s_op->show_options(m, mnt);
1021    seq_puts(m, " 0 0\n");
1022out:
1023    return err;
1024}
1025
1026const struct seq_operations mounts_op = {
1027    .start = m_start,
1028    .next = m_next,
1029    .stop = m_stop,
1030    .show = show_vfsmnt
1031};
1032
1033static int show_mountinfo(struct seq_file *m, void *v)
1034{
1035    struct proc_mounts *p = m->private;
1036    struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1037    struct super_block *sb = mnt->mnt_sb;
1038    struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1039    struct path root = p->root;
1040    int err = 0;
1041
1042    seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1043           MAJOR(sb->s_dev), MINOR(sb->s_dev));
1044    if (sb->s_op->show_path)
1045        err = sb->s_op->show_path(m, mnt);
1046    else
1047        seq_dentry(m, mnt->mnt_root, " \t\n\\");
1048    if (err)
1049        goto out;
1050    seq_putc(m, ' ');
1051    seq_path_root(m, &mnt_path, &root, " \t\n\\");
1052    if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
1053        /*
1054         * Mountpoint is outside root, discard that one. Ugly,
1055         * but less so than trying to do that in iterator in a
1056         * race-free way (due to renames).
1057         */
1058        return SEQ_SKIP;
1059    }
1060    seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1061    show_mnt_opts(m, mnt);
1062
1063    /* Tagged fields ("foo:X" or "bar") */
1064    if (IS_MNT_SHARED(mnt))
1065        seq_printf(m, " shared:%i", mnt->mnt_group_id);
1066    if (IS_MNT_SLAVE(mnt)) {
1067        int master = mnt->mnt_master->mnt_group_id;
1068        int dom = get_dominating_id(mnt, &p->root);
1069        seq_printf(m, " master:%i", master);
1070        if (dom && dom != master)
1071            seq_printf(m, " propagate_from:%i", dom);
1072    }
1073    if (IS_MNT_UNBINDABLE(mnt))
1074        seq_puts(m, " unbindable");
1075
1076    /* Filesystem specific data */
1077    seq_puts(m, " - ");
1078    show_type(m, sb);
1079    seq_putc(m, ' ');
1080    if (sb->s_op->show_devname)
1081        err = sb->s_op->show_devname(m, mnt);
1082    else
1083        mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1084    if (err)
1085        goto out;
1086    seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1087    err = show_sb_opts(m, sb);
1088    if (err)
1089        goto out;
1090    if (sb->s_op->show_options)
1091        err = sb->s_op->show_options(m, mnt);
1092    seq_putc(m, '\n');
1093out:
1094    return err;
1095}
1096
1097const struct seq_operations mountinfo_op = {
1098    .start = m_start,
1099    .next = m_next,
1100    .stop = m_stop,
1101    .show = show_mountinfo,
1102};
1103
1104static int show_vfsstat(struct seq_file *m, void *v)
1105{
1106    struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1107    struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1108    int err = 0;
1109
1110    /* device */
1111    if (mnt->mnt_sb->s_op->show_devname) {
1112        err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1113    } else {
1114        if (mnt->mnt_devname) {
1115            seq_puts(m, "device ");
1116            mangle(m, mnt->mnt_devname);
1117        } else
1118            seq_puts(m, "no device");
1119    }
1120
1121    /* mount point */
1122    seq_puts(m, " mounted on ");
1123    seq_path(m, &mnt_path, " \t\n\\");
1124    seq_putc(m, ' ');
1125
1126    /* file system type */
1127    seq_puts(m, "with fstype ");
1128    show_type(m, mnt->mnt_sb);
1129
1130    /* optional statistics */
1131    if (mnt->mnt_sb->s_op->show_stats) {
1132        seq_putc(m, ' ');
1133        if (!err)
1134            err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1135    }
1136
1137    seq_putc(m, '\n');
1138    return err;
1139}
1140
1141const struct seq_operations mountstats_op = {
1142    .start = m_start,
1143    .next = m_next,
1144    .stop = m_stop,
1145    .show = show_vfsstat,
1146};
1147#endif /* CONFIG_PROC_FS */
1148
1149/**
1150 * may_umount_tree - check if a mount tree is busy
1151 * @mnt: root of mount tree
1152 *
1153 * This is called to check if a tree of mounts has any
1154 * open files, pwds, chroots or sub mounts that are
1155 * busy.
1156 */
1157int may_umount_tree(struct vfsmount *mnt)
1158{
1159    int actual_refs = 0;
1160    int minimum_refs = 0;
1161    struct vfsmount *p;
1162
1163    /* write lock needed for mnt_get_count */
1164    br_write_lock(vfsmount_lock);
1165    for (p = mnt; p; p = next_mnt(p, mnt)) {
1166        actual_refs += mnt_get_count(p);
1167        minimum_refs += 2;
1168    }
1169    br_write_unlock(vfsmount_lock);
1170
1171    if (actual_refs > minimum_refs)
1172        return 0;
1173
1174    return 1;
1175}
1176
1177EXPORT_SYMBOL(may_umount_tree);
1178
1179/**
1180 * may_umount - check if a mount point is busy
1181 * @mnt: root of mount
1182 *
1183 * This is called to check if a mount point has any
1184 * open files, pwds, chroots or sub mounts. If the
1185 * mount has sub mounts this will return busy
1186 * regardless of whether the sub mounts are busy.
1187 *
1188 * Doesn't take quota and stuff into account. IOW, in some cases it will
1189 * give false negatives. The main reason why it's here is that we need
1190 * a non-destructive way to look for easily umountable filesystems.
1191 */
1192int may_umount(struct vfsmount *mnt)
1193{
1194    int ret = 1;
1195    down_read(&namespace_sem);
1196    br_write_lock(vfsmount_lock);
1197    if (propagate_mount_busy(mnt, 2))
1198        ret = 0;
1199    br_write_unlock(vfsmount_lock);
1200    up_read(&namespace_sem);
1201    return ret;
1202}
1203
1204EXPORT_SYMBOL(may_umount);
1205
1206void release_mounts(struct list_head *head)
1207{
1208    struct vfsmount *mnt;
1209    while (!list_empty(head)) {
1210        mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1211        list_del_init(&mnt->mnt_hash);
1212        if (mnt->mnt_parent != mnt) {
1213            struct dentry *dentry;
1214            struct vfsmount *m;
1215
1216            br_write_lock(vfsmount_lock);
1217            dentry = mnt->mnt_mountpoint;
1218            m = mnt->mnt_parent;
1219            mnt->mnt_mountpoint = mnt->mnt_root;
1220            mnt->mnt_parent = mnt;
1221            m->mnt_ghosts--;
1222            br_write_unlock(vfsmount_lock);
1223            dput(dentry);
1224            mntput(m);
1225        }
1226        mntput(mnt);
1227    }
1228}
1229
1230/*
1231 * vfsmount lock must be held for write
1232 * namespace_sem must be held for write
1233 */
1234void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1235{
1236    LIST_HEAD(tmp_list);
1237    struct vfsmount *p;
1238
1239    for (p = mnt; p; p = next_mnt(p, mnt))
1240        list_move(&p->mnt_hash, &tmp_list);
1241
1242    if (propagate)
1243        propagate_umount(&tmp_list);
1244
1245    list_for_each_entry(p, &tmp_list, mnt_hash) {
1246        list_del_init(&p->mnt_expire);
1247        list_del_init(&p->mnt_list);
1248        __touch_mnt_namespace(p->mnt_ns);
1249        p->mnt_ns = NULL;
1250        __mnt_make_shortterm(p);
1251        list_del_init(&p->mnt_child);
1252        if (p->mnt_parent != p) {
1253            p->mnt_parent->mnt_ghosts++;
1254            dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
1255        }
1256        change_mnt_propagation(p, MS_PRIVATE);
1257    }
1258    list_splice(&tmp_list, kill);
1259}
1260
1261static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1262
1263static int do_umount(struct vfsmount *mnt, int flags)
1264{
1265    struct super_block *sb = mnt->mnt_sb;
1266    int retval;
1267    LIST_HEAD(umount_list);
1268
1269    retval = security_sb_umount(mnt, flags);
1270    if (retval)
1271        return retval;
1272
1273    /*
1274     * Allow userspace to request a mountpoint be expired rather than
1275     * unmounting unconditionally. Unmount only happens if:
1276     * (1) the mark is already set (the mark is cleared by mntput())
1277     * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1278     */
1279    if (flags & MNT_EXPIRE) {
1280        if (mnt == current->fs->root.mnt ||
1281            flags & (MNT_FORCE | MNT_DETACH))
1282            return -EINVAL;
1283
1284        /*
1285         * probably don't strictly need the lock here if we examined
1286         * all race cases, but it's a slowpath.
1287         */
1288        br_write_lock(vfsmount_lock);
1289        if (mnt_get_count(mnt) != 2) {
1290            br_write_unlock(vfsmount_lock);
1291            return -EBUSY;
1292        }
1293        br_write_unlock(vfsmount_lock);
1294
1295        if (!xchg(&mnt->mnt_expiry_mark, 1))
1296            return -EAGAIN;
1297    }
1298
1299    /*
1300     * If we may have to abort operations to get out of this
1301     * mount, and they will themselves hold resources we must
1302     * allow the fs to do things. In the Unix tradition of
1303     * 'Gee thats tricky lets do it in userspace' the umount_begin
1304     * might fail to complete on the first run through as other tasks
1305     * must return, and the like. Thats for the mount program to worry
1306     * about for the moment.
1307     */
1308
1309    if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1310        sb->s_op->umount_begin(sb);
1311    }
1312
1313    /*
1314     * No sense to grab the lock for this test, but test itself looks
1315     * somewhat bogus. Suggestions for better replacement?
1316     * Ho-hum... In principle, we might treat that as umount + switch
1317     * to rootfs. GC would eventually take care of the old vfsmount.
1318     * Actually it makes sense, especially if rootfs would contain a
1319     * /reboot - static binary that would close all descriptors and
1320     * call reboot(9). Then init(8) could umount root and exec /reboot.
1321     */
1322    if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1323        /*
1324         * Special case for "unmounting" root ...
1325         * we just try to remount it readonly.
1326         */
1327        down_write(&sb->s_umount);
1328        if (!(sb->s_flags & MS_RDONLY))
1329            retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1330        up_write(&sb->s_umount);
1331        return retval;
1332    }
1333
1334    down_write(&namespace_sem);
1335    br_write_lock(vfsmount_lock);
1336    event++;
1337
1338    if (!(flags & MNT_DETACH))
1339        shrink_submounts(mnt, &umount_list);
1340
1341    retval = -EBUSY;
1342    if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1343        if (!list_empty(&mnt->mnt_list))
1344            umount_tree(mnt, 1, &umount_list);
1345        retval = 0;
1346    }
1347    br_write_unlock(vfsmount_lock);
1348    up_write(&namespace_sem);
1349    release_mounts(&umount_list);
1350    return retval;
1351}
1352
1353/*
1354 * Now umount can handle mount points as well as block devices.
1355 * This is important for filesystems which use unnamed block devices.
1356 *
1357 * We now support a flag for forced unmount like the other 'big iron'
1358 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1359 */
1360
1361SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1362{
1363    struct path path;
1364    int retval;
1365    int lookup_flags = 0;
1366
1367    if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1368        return -EINVAL;
1369
1370    if (!(flags & UMOUNT_NOFOLLOW))
1371        lookup_flags |= LOOKUP_FOLLOW;
1372
1373    retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1374    if (retval)
1375        goto out;
1376    retval = -EINVAL;
1377    if (path.dentry != path.mnt->mnt_root)
1378        goto dput_and_out;
1379    if (!check_mnt(path.mnt))
1380        goto dput_and_out;
1381
1382    retval = -EPERM;
1383    if (!capable(CAP_SYS_ADMIN))
1384        goto dput_and_out;
1385
1386    retval = do_umount(path.mnt, flags);
1387dput_and_out:
1388    /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1389    dput(path.dentry);
1390    mntput_no_expire(path.mnt);
1391out:
1392    return retval;
1393}
1394
1395#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1396
1397/*
1398 * The 2.0 compatible umount. No flags.
1399 */
1400SYSCALL_DEFINE1(oldumount, char __user *, name)
1401{
1402    return sys_umount(name, 0);
1403}
1404
1405#endif
1406
1407static int mount_is_safe(struct path *path)
1408{
1409    if (capable(CAP_SYS_ADMIN))
1410        return 0;
1411    return -EPERM;
1412#ifdef notyet
1413    if (S_ISLNK(path->dentry->d_inode->i_mode))
1414        return -EPERM;
1415    if (path->dentry->d_inode->i_mode & S_ISVTX) {
1416        if (current_uid() != path->dentry->d_inode->i_uid)
1417            return -EPERM;
1418    }
1419    if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1420        return -EPERM;
1421    return 0;
1422#endif
1423}
1424
1425struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1426                    int flag)
1427{
1428    struct vfsmount *res, *p, *q, *r, *s;
1429    struct path path;
1430
1431    if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1432        return NULL;
1433
1434    res = q = clone_mnt(mnt, dentry, flag);
1435    if (!q)
1436        goto Enomem;
1437    q->mnt_mountpoint = mnt->mnt_mountpoint;
1438
1439    p = mnt;
1440    list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1441        if (!is_subdir(r->mnt_mountpoint, dentry))
1442            continue;
1443
1444        for (s = r; s; s = next_mnt(s, r)) {
1445            if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1446                s = skip_mnt_tree(s);
1447                continue;
1448            }
1449            while (p != s->mnt_parent) {
1450                p = p->mnt_parent;
1451                q = q->mnt_parent;
1452            }
1453            p = s;
1454            path.mnt = q;
1455            path.dentry = p->mnt_mountpoint;
1456            q = clone_mnt(p, p->mnt_root, flag);
1457            if (!q)
1458                goto Enomem;
1459            br_write_lock(vfsmount_lock);
1460            list_add_tail(&q->mnt_list, &res->mnt_list);
1461            attach_mnt(q, &path);
1462            br_write_unlock(vfsmount_lock);
1463        }
1464    }
1465    return res;
1466Enomem:
1467    if (res) {
1468        LIST_HEAD(umount_list);
1469        br_write_lock(vfsmount_lock);
1470        umount_tree(res, 0, &umount_list);
1471        br_write_unlock(vfsmount_lock);
1472        release_mounts(&umount_list);
1473    }
1474    return NULL;
1475}
1476
1477struct vfsmount *collect_mounts(struct path *path)
1478{
1479    struct vfsmount *tree;
1480    down_write(&namespace_sem);
1481    tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1482    up_write(&namespace_sem);
1483    return tree;
1484}
1485
1486void drop_collected_mounts(struct vfsmount *mnt)
1487{
1488    LIST_HEAD(umount_list);
1489    down_write(&namespace_sem);
1490    br_write_lock(vfsmount_lock);
1491    umount_tree(mnt, 0, &umount_list);
1492    br_write_unlock(vfsmount_lock);
1493    up_write(&namespace_sem);
1494    release_mounts(&umount_list);
1495}
1496
1497int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1498           struct vfsmount *root)
1499{
1500    struct vfsmount *mnt;
1501    int res = f(root, arg);
1502    if (res)
1503        return res;
1504    list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1505        res = f(mnt, arg);
1506        if (res)
1507            return res;
1508    }
1509    return 0;
1510}
1511
1512static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1513{
1514    struct vfsmount *p;
1515
1516    for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1517        if (p->mnt_group_id && !IS_MNT_SHARED(p))
1518            mnt_release_group_id(p);
1519    }
1520}
1521
1522static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1523{
1524    struct vfsmount *p;
1525
1526    for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1527        if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1528            int err = mnt_alloc_group_id(p);
1529            if (err) {
1530                cleanup_group_ids(mnt, p);
1531                return err;
1532            }
1533        }
1534    }
1535
1536    return 0;
1537}
1538
1539/*
1540 * @source_mnt : mount tree to be attached
1541 * @nd : place the mount tree @source_mnt is attached
1542 * @parent_nd : if non-null, detach the source_mnt from its parent and
1543 * store the parent mount and mountpoint dentry.
1544 * (done when source_mnt is moved)
1545 *
1546 * NOTE: in the table below explains the semantics when a source mount
1547 * of a given type is attached to a destination mount of a given type.
1548 * ---------------------------------------------------------------------------
1549 * | BIND MOUNT OPERATION |
1550 * |**************************************************************************
1551 * | source-->| shared | private | slave | unbindable |
1552 * | dest | | | | |
1553 * | | | | | | |
1554 * | v | | | | |
1555 * |**************************************************************************
1556 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1557 * | | | | | |
1558 * |non-shared| shared (+) | private | slave (*) | invalid |
1559 * ***************************************************************************
1560 * A bind operation clones the source mount and mounts the clone on the
1561 * destination mount.
1562 *
1563 * (++) the cloned mount is propagated to all the mounts in the propagation
1564 * tree of the destination mount and the cloned mount is added to
1565 * the peer group of the source mount.
1566 * (+) the cloned mount is created under the destination mount and is marked
1567 * as shared. The cloned mount is added to the peer group of the source
1568 * mount.
1569 * (+++) the mount is propagated to all the mounts in the propagation tree
1570 * of the destination mount and the cloned mount is made slave
1571 * of the same master as that of the source mount. The cloned mount
1572 * is marked as 'shared and slave'.
1573 * (*) the cloned mount is made a slave of the same master as that of the
1574 * source mount.
1575 *
1576 * ---------------------------------------------------------------------------
1577 * | MOVE MOUNT OPERATION |
1578 * |**************************************************************************
1579 * | source-->| shared | private | slave | unbindable |
1580 * | dest | | | | |
1581 * | | | | | | |
1582 * | v | | | | |
1583 * |**************************************************************************
1584 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1585 * | | | | | |
1586 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1587 * ***************************************************************************
1588 *
1589 * (+) the mount is moved to the destination. And is then propagated to
1590 * all the mounts in the propagation tree of the destination mount.
1591 * (+*) the mount is moved to the destination.
1592 * (+++) the mount is moved to the destination and is then propagated to
1593 * all the mounts belonging to the destination mount's propagation tree.
1594 * the mount is marked as 'shared and slave'.
1595 * (*) the mount continues to be a slave at the new location.
1596 *
1597 * if the source mount is a tree, the operations explained above is
1598 * applied to each mount in the tree.
1599 * Must be called without spinlocks held, since this function can sleep
1600 * in allocations.
1601 */
1602static int attach_recursive_mnt(struct vfsmount *source_mnt,
1603            struct path *path, struct path *parent_path)
1604{
1605    LIST_HEAD(tree_list);
1606    struct vfsmount *dest_mnt = path->mnt;
1607    struct dentry *dest_dentry = path->dentry;
1608    struct vfsmount *child, *p;
1609    int err;
1610
1611    if (IS_MNT_SHARED(dest_mnt)) {
1612        err = invent_group_ids(source_mnt, true);
1613        if (err)
1614            goto out;
1615    }
1616    err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1617    if (err)
1618        goto out_cleanup_ids;
1619
1620    br_write_lock(vfsmount_lock);
1621
1622    if (IS_MNT_SHARED(dest_mnt)) {
1623        for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1624            set_mnt_shared(p);
1625    }
1626    if (parent_path) {
1627        detach_mnt(source_mnt, parent_path);
1628        attach_mnt(source_mnt, path);
1629        touch_mnt_namespace(parent_path->mnt->mnt_ns);
1630    } else {
1631        mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1632        commit_tree(source_mnt);
1633    }
1634
1635    list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1636        list_del_init(&child->mnt_hash);
1637        commit_tree(child);
1638    }
1639    br_write_unlock(vfsmount_lock);
1640
1641    return 0;
1642
1643 out_cleanup_ids:
1644    if (IS_MNT_SHARED(dest_mnt))
1645        cleanup_group_ids(source_mnt, NULL);
1646 out:
1647    return err;
1648}
1649
1650static int lock_mount(struct path *path)
1651{
1652    struct vfsmount *mnt;
1653retry:
1654    mutex_lock(&path->dentry->d_inode->i_mutex);
1655    if (unlikely(cant_mount(path->dentry))) {
1656        mutex_unlock(&path->dentry->d_inode->i_mutex);
1657        return -ENOENT;
1658    }
1659    down_write(&namespace_sem);
1660    mnt = lookup_mnt(path);
1661    if (likely(!mnt))
1662        return 0;
1663    up_write(&namespace_sem);
1664    mutex_unlock(&path->dentry->d_inode->i_mutex);
1665    path_put(path);
1666    path->mnt = mnt;
1667    path->dentry = dget(mnt->mnt_root);
1668    goto retry;
1669}
1670
1671static void unlock_mount(struct path *path)
1672{
1673    up_write(&namespace_sem);
1674    mutex_unlock(&path->dentry->d_inode->i_mutex);
1675}
1676
1677static int graft_tree(struct vfsmount *mnt, struct path *path)
1678{
1679    if (mnt->mnt_sb->s_flags & MS_NOUSER)
1680        return -EINVAL;
1681
1682    if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1683          S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1684        return -ENOTDIR;
1685
1686    if (d_unlinked(path->dentry))
1687        return -ENOENT;
1688
1689    return attach_recursive_mnt(mnt, path, NULL);
1690}
1691
1692/*
1693 * Sanity check the flags to change_mnt_propagation.
1694 */
1695
1696static int flags_to_propagation_type(int flags)
1697{
1698    int type = flags & ~(MS_REC | MS_SILENT);
1699
1700    /* Fail if any non-propagation flags are set */
1701    if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1702        return 0;
1703    /* Only one propagation flag should be set */
1704    if (!is_power_of_2(type))
1705        return 0;
1706    return type;
1707}
1708
1709/*
1710 * recursively change the type of the mountpoint.
1711 */
1712static int do_change_type(struct path *path, int flag)
1713{
1714    struct vfsmount *m, *mnt = path->mnt;
1715    int recurse = flag & MS_REC;
1716    int type;
1717    int err = 0;
1718
1719    if (!capable(CAP_SYS_ADMIN))
1720        return -EPERM;
1721
1722    if (path->dentry != path->mnt->mnt_root)
1723        return -EINVAL;
1724
1725    type = flags_to_propagation_type(flag);
1726    if (!type)
1727        return -EINVAL;
1728
1729    down_write(&namespace_sem);
1730    if (type == MS_SHARED) {
1731        err = invent_group_ids(mnt, recurse);
1732        if (err)
1733            goto out_unlock;
1734    }
1735
1736    br_write_lock(vfsmount_lock);
1737    for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1738        change_mnt_propagation(m, type);
1739    br_write_unlock(vfsmount_lock);
1740
1741 out_unlock:
1742    up_write(&namespace_sem);
1743    return err;
1744}
1745
1746/*
1747 * do loopback mount.
1748 */
1749static int do_loopback(struct path *path, char *old_name,
1750                int recurse)
1751{
1752    LIST_HEAD(umount_list);
1753    struct path old_path;
1754    struct vfsmount *mnt = NULL;
1755    int err = mount_is_safe(path);
1756    if (err)
1757        return err;
1758    if (!old_name || !*old_name)
1759        return -EINVAL;
1760    err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1761    if (err)
1762        return err;
1763
1764    err = lock_mount(path);
1765    if (err)
1766        goto out;
1767
1768    err = -EINVAL;
1769    if (IS_MNT_UNBINDABLE(old_path.mnt))
1770        goto out2;
1771
1772    if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1773        goto out2;
1774
1775    err = -ENOMEM;
1776    if (recurse)
1777        mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1778    else
1779        mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1780
1781    if (!mnt)
1782        goto out2;
1783
1784    err = graft_tree(mnt, path);
1785    if (err) {
1786        br_write_lock(vfsmount_lock);
1787        umount_tree(mnt, 0, &umount_list);
1788        br_write_unlock(vfsmount_lock);
1789    }
1790out2:
1791    unlock_mount(path);
1792    release_mounts(&umount_list);
1793out:
1794    path_put(&old_path);
1795    return err;
1796}
1797
1798static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1799{
1800    int error = 0;
1801    int readonly_request = 0;
1802
1803    if (ms_flags & MS_RDONLY)
1804        readonly_request = 1;
1805    if (readonly_request == __mnt_is_readonly(mnt))
1806        return 0;
1807
1808    if (readonly_request)
1809        error = mnt_make_readonly(mnt);
1810    else
1811        __mnt_unmake_readonly(mnt);
1812    return error;
1813}
1814
1815/*
1816 * change filesystem flags. dir should be a physical root of filesystem.
1817 * If you've mounted a non-root directory somewhere and want to do remount
1818 * on it - tough luck.
1819 */
1820static int do_remount(struct path *path, int flags, int mnt_flags,
1821              void *data)
1822{
1823    int err;
1824    struct super_block *sb = path->mnt->mnt_sb;
1825
1826    if (!capable(CAP_SYS_ADMIN))
1827        return -EPERM;
1828
1829    if (!check_mnt(path->mnt))
1830        return -EINVAL;
1831
1832    if (path->dentry != path->mnt->mnt_root)
1833        return -EINVAL;
1834
1835    err = security_sb_remount(sb, data);
1836    if (err)
1837        return err;
1838
1839    down_write(&sb->s_umount);
1840    if (flags & MS_BIND)
1841        err = change_mount_flags(path->mnt, flags);
1842    else
1843        err = do_remount_sb(sb, flags, data, 0);
1844    if (!err) {
1845        br_write_lock(vfsmount_lock);
1846        mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1847        path->mnt->mnt_flags = mnt_flags;
1848        br_write_unlock(vfsmount_lock);
1849    }
1850    up_write(&sb->s_umount);
1851    if (!err) {
1852        br_write_lock(vfsmount_lock);
1853        touch_mnt_namespace(path->mnt->mnt_ns);
1854        br_write_unlock(vfsmount_lock);
1855    }
1856    return err;
1857}
1858
1859static inline int tree_contains_unbindable(struct vfsmount *mnt)
1860{
1861    struct vfsmount *p;
1862    for (p = mnt; p; p = next_mnt(p, mnt)) {
1863        if (IS_MNT_UNBINDABLE(p))
1864            return 1;
1865    }
1866    return 0;
1867}
1868
1869static int do_move_mount(struct path *path, char *old_name)
1870{
1871    struct path old_path, parent_path;
1872    struct vfsmount *p;
1873    int err = 0;
1874    if (!capable(CAP_SYS_ADMIN))
1875        return -EPERM;
1876    if (!old_name || !*old_name)
1877        return -EINVAL;
1878    err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1879    if (err)
1880        return err;
1881
1882    err = lock_mount(path);
1883    if (err < 0)
1884        goto out;
1885
1886    err = -EINVAL;
1887    if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1888        goto out1;
1889
1890    if (d_unlinked(path->dentry))
1891        goto out1;
1892
1893    err = -EINVAL;
1894    if (old_path.dentry != old_path.mnt->mnt_root)
1895        goto out1;
1896
1897    if (old_path.mnt == old_path.mnt->mnt_parent)
1898        goto out1;
1899
1900    if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1901          S_ISDIR(old_path.dentry->d_inode->i_mode))
1902        goto out1;
1903    /*
1904     * Don't move a mount residing in a shared parent.
1905     */
1906    if (old_path.mnt->mnt_parent &&
1907        IS_MNT_SHARED(old_path.mnt->mnt_parent))
1908        goto out1;
1909    /*
1910     * Don't move a mount tree containing unbindable mounts to a destination
1911     * mount which is shared.
1912     */
1913    if (IS_MNT_SHARED(path->mnt) &&
1914        tree_contains_unbindable(old_path.mnt))
1915        goto out1;
1916    err = -ELOOP;
1917    for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1918        if (p == old_path.mnt)
1919            goto out1;
1920
1921    err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1922    if (err)
1923        goto out1;
1924
1925    /* if the mount is moved, it should no longer be expire
1926     * automatically */
1927    list_del_init(&old_path.mnt->mnt_expire);
1928out1:
1929    unlock_mount(path);
1930out:
1931    if (!err)
1932        path_put(&parent_path);
1933    path_put(&old_path);
1934    return err;
1935}
1936
1937static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1938{
1939    int err;
1940    const char *subtype = strchr(fstype, '.');
1941    if (subtype) {
1942        subtype++;
1943        err = -EINVAL;
1944        if (!subtype[0])
1945            goto err;
1946    } else
1947        subtype = "";
1948
1949    mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1950    err = -ENOMEM;
1951    if (!mnt->mnt_sb->s_subtype)
1952        goto err;
1953    return mnt;
1954
1955 err:
1956    mntput(mnt);
1957    return ERR_PTR(err);
1958}
1959
1960struct vfsmount *
1961do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1962{
1963    struct file_system_type *type = get_fs_type(fstype);
1964    struct vfsmount *mnt;
1965    if (!type)
1966        return ERR_PTR(-ENODEV);
1967    mnt = vfs_kern_mount(type, flags, name, data);
1968    if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1969        !mnt->mnt_sb->s_subtype)
1970        mnt = fs_set_subtype(mnt, fstype);
1971    put_filesystem(type);
1972    return mnt;
1973}
1974EXPORT_SYMBOL_GPL(do_kern_mount);
1975
1976/*
1977 * add a mount into a namespace's mount tree
1978 */
1979static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1980{
1981    int err;
1982
1983    mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1984
1985    err = lock_mount(path);
1986    if (err)
1987        return err;
1988
1989    err = -EINVAL;
1990    if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1991        goto unlock;
1992
1993    /* Refuse the same filesystem on the same mount point */
1994    err = -EBUSY;
1995    if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1996        path->mnt->mnt_root == path->dentry)
1997        goto unlock;
1998
1999    err = -EINVAL;
2000    if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
2001        goto unlock;
2002
2003    newmnt->mnt_flags = mnt_flags;
2004    err = graft_tree(newmnt, path);
2005
2006unlock:
2007    unlock_mount(path);
2008    return err;
2009}
2010
2011/*
2012 * create a new mount for userspace and request it to be added into the
2013 * namespace's tree
2014 */
2015static int do_new_mount(struct path *path, char *type, int flags,
2016            int mnt_flags, char *name, void *data)
2017{
2018    struct vfsmount *mnt;
2019    int err;
2020
2021    if (!type)
2022        return -EINVAL;
2023
2024    /* we need capabilities... */
2025    if (!capable(CAP_SYS_ADMIN))
2026        return -EPERM;
2027
2028    mnt = do_kern_mount(type, flags, name, data);
2029    if (IS_ERR(mnt))
2030        return PTR_ERR(mnt);
2031
2032    err = do_add_mount(mnt, path, mnt_flags);
2033    if (err)
2034        mntput(mnt);
2035    return err;
2036}
2037
2038int finish_automount(struct vfsmount *m, struct path *path)
2039{
2040    int err;
2041    /* The new mount record should have at least 2 refs to prevent it being
2042     * expired before we get a chance to add it
2043     */
2044    BUG_ON(mnt_get_count(m) < 2);
2045
2046    if (m->mnt_sb == path->mnt->mnt_sb &&
2047        m->mnt_root == path->dentry) {
2048        err = -ELOOP;
2049        goto fail;
2050    }
2051
2052    err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2053    if (!err)
2054        return 0;
2055fail:
2056    /* remove m from any expiration list it may be on */
2057    if (!list_empty(&m->mnt_expire)) {
2058        down_write(&namespace_sem);
2059        br_write_lock(vfsmount_lock);
2060        list_del_init(&m->mnt_expire);
2061        br_write_unlock(vfsmount_lock);
2062        up_write(&namespace_sem);
2063    }
2064    mntput(m);
2065    mntput(m);
2066    return err;
2067}
2068
2069/**
2070 * mnt_set_expiry - Put a mount on an expiration list
2071 * @mnt: The mount to list.
2072 * @expiry_list: The list to add the mount to.
2073 */
2074void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2075{
2076    down_write(&namespace_sem);
2077    br_write_lock(vfsmount_lock);
2078
2079    list_add_tail(&mnt->mnt_expire, expiry_list);
2080
2081    br_write_unlock(vfsmount_lock);
2082    up_write(&namespace_sem);
2083}
2084EXPORT_SYMBOL(mnt_set_expiry);
2085
2086/*
2087 * process a list of expirable mountpoints with the intent of discarding any
2088 * mountpoints that aren't in use and haven't been touched since last we came
2089 * here
2090 */
2091void mark_mounts_for_expiry(struct list_head *mounts)
2092{
2093    struct vfsmount *mnt, *next;
2094    LIST_HEAD(graveyard);
2095    LIST_HEAD(umounts);
2096
2097    if (list_empty(mounts))
2098        return;
2099
2100    down_write(&namespace_sem);
2101    br_write_lock(vfsmount_lock);
2102
2103    /* extract from the expiration list every vfsmount that matches the
2104     * following criteria:
2105     * - only referenced by its parent vfsmount
2106     * - still marked for expiry (marked on the last call here; marks are
2107     * cleared by mntput())
2108     */
2109    list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2110        if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2111            propagate_mount_busy(mnt, 1))
2112            continue;
2113        list_move(&mnt->mnt_expire, &graveyard);
2114    }
2115    while (!list_empty(&graveyard)) {
2116        mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2117        touch_mnt_namespace(mnt->mnt_ns);
2118        umount_tree(mnt, 1, &umounts);
2119    }
2120    br_write_unlock(vfsmount_lock);
2121    up_write(&namespace_sem);
2122
2123    release_mounts(&umounts);
2124}
2125
2126EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2127
2128/*
2129 * Ripoff of 'select_parent()'
2130 *
2131 * search the list of submounts for a given mountpoint, and move any
2132 * shrinkable submounts to the 'graveyard' list.
2133 */
2134static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2135{
2136    struct vfsmount *this_parent = parent;
2137    struct list_head *next;
2138    int found = 0;
2139
2140repeat:
2141    next = this_parent->mnt_mounts.next;
2142resume:
2143    while (next != &this_parent->mnt_mounts) {
2144        struct list_head *tmp = next;
2145        struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2146
2147        next = tmp->next;
2148        if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2149            continue;
2150        /*
2151         * Descend a level if the d_mounts list is non-empty.
2152         */
2153        if (!list_empty(&mnt->mnt_mounts)) {
2154            this_parent = mnt;
2155            goto repeat;
2156        }
2157
2158        if (!propagate_mount_busy(mnt, 1)) {
2159            list_move_tail(&mnt->mnt_expire, graveyard);
2160            found++;
2161        }
2162    }
2163    /*
2164     * All done at this level ... ascend and resume the search
2165     */
2166    if (this_parent != parent) {
2167        next = this_parent->mnt_child.next;
2168        this_parent = this_parent->mnt_parent;
2169        goto resume;
2170    }
2171    return found;
2172}
2173
2174/*
2175 * process a list of expirable mountpoints with the intent of discarding any
2176 * submounts of a specific parent mountpoint
2177 *
2178 * vfsmount_lock must be held for write
2179 */
2180static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
2181{
2182    LIST_HEAD(graveyard);
2183    struct vfsmount *m;
2184
2185    /* extract submounts of 'mountpoint' from the expiration list */
2186    while (select_submounts(mnt, &graveyard)) {
2187        while (!list_empty(&graveyard)) {
2188            m = list_first_entry(&graveyard, struct vfsmount,
2189                        mnt_expire);
2190            touch_mnt_namespace(m->mnt_ns);
2191            umount_tree(m, 1, umounts);
2192        }
2193    }
2194}
2195
2196/*
2197 * Some copy_from_user() implementations do not return the exact number of
2198 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2199 * Note that this function differs from copy_from_user() in that it will oops
2200 * on bad values of `to', rather than returning a short copy.
2201 */
2202static long exact_copy_from_user(void *to, const void __user * from,
2203                 unsigned long n)
2204{
2205    char *t = to;
2206    const char __user *f = from;
2207    char c;
2208
2209    if (!access_ok(VERIFY_READ, from, n))
2210        return n;
2211
2212    while (n) {
2213        if (__get_user(c, f)) {
2214            memset(t, 0, n);
2215            break;
2216        }
2217        *t++ = c;
2218        f++;
2219        n--;
2220    }
2221    return n;
2222}
2223
2224int copy_mount_options(const void __user * data, unsigned long *where)
2225{
2226    int i;
2227    unsigned long page;
2228    unsigned long size;
2229
2230    *where = 0;
2231    if (!data)
2232        return 0;
2233
2234    if (!(page = __get_free_page(GFP_KERNEL)))
2235        return -ENOMEM;
2236
2237    /* We only care that *some* data at the address the user
2238     * gave us is valid. Just in case, we'll zero
2239     * the remainder of the page.
2240     */
2241    /* copy_from_user cannot cross TASK_SIZE ! */
2242    size = TASK_SIZE - (unsigned long)data;
2243    if (size > PAGE_SIZE)
2244        size = PAGE_SIZE;
2245
2246    i = size - exact_copy_from_user((void *)page, data, size);
2247    if (!i) {
2248        free_page(page);
2249        return -EFAULT;
2250    }
2251    if (i != PAGE_SIZE)
2252        memset((char *)page + i, 0, PAGE_SIZE - i);
2253    *where = page;
2254    return 0;
2255}
2256
2257int copy_mount_string(const void __user *data, char **where)
2258{
2259    char *tmp;
2260
2261    if (!data) {
2262        *where = NULL;
2263        return 0;
2264    }
2265
2266    tmp = strndup_user(data, PAGE_SIZE);
2267    if (IS_ERR(tmp))
2268        return PTR_ERR(tmp);
2269
2270    *where = tmp;
2271    return 0;
2272}
2273
2274/*
2275 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2276 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2277 *
2278 * data is a (void *) that can point to any structure up to
2279 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2280 * information (or be NULL).
2281 *
2282 * Pre-0.97 versions of mount() didn't have a flags word.
2283 * When the flags word was introduced its top half was required
2284 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2285 * Therefore, if this magic number is present, it carries no information
2286 * and must be discarded.
2287 */
2288long do_mount(char *dev_name, char *dir_name, char *type_page,
2289          unsigned long flags, void *data_page)
2290{
2291    struct path path;
2292    int retval = 0;
2293    int mnt_flags = 0;
2294
2295    /* Discard magic */
2296    if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2297        flags &= ~MS_MGC_MSK;
2298
2299    /* Basic sanity checks */
2300
2301    if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2302        return -EINVAL;
2303
2304    if (data_page)
2305        ((char *)data_page)[PAGE_SIZE - 1] = 0;
2306
2307    /* ... and get the mountpoint */
2308    retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2309    if (retval)
2310        return retval;
2311
2312    retval = security_sb_mount(dev_name, &path,
2313                   type_page, flags, data_page);
2314    if (retval)
2315        goto dput_out;
2316
2317    /* Default to relatime unless overriden */
2318    if (!(flags & MS_NOATIME))
2319        mnt_flags |= MNT_RELATIME;
2320
2321    /* Separate the per-mountpoint flags */
2322    if (flags & MS_NOSUID)
2323        mnt_flags |= MNT_NOSUID;
2324    if (flags & MS_NODEV)
2325        mnt_flags |= MNT_NODEV;
2326    if (flags & MS_NOEXEC)
2327        mnt_flags |= MNT_NOEXEC;
2328    if (flags & MS_NOATIME)
2329        mnt_flags |= MNT_NOATIME;
2330    if (flags & MS_NODIRATIME)
2331        mnt_flags |= MNT_NODIRATIME;
2332    if (flags & MS_STRICTATIME)
2333        mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2334    if (flags & MS_RDONLY)
2335        mnt_flags |= MNT_READONLY;
2336
2337    flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2338           MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2339           MS_STRICTATIME);
2340
2341    if (flags & MS_REMOUNT)
2342        retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2343                    data_page);
2344    else if (flags & MS_BIND)
2345        retval = do_loopback(&path, dev_name, flags & MS_REC);
2346    else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2347        retval = do_change_type(&path, flags);
2348    else if (flags & MS_MOVE)
2349        retval = do_move_mount(&path, dev_name);
2350    else
2351        retval = do_new_mount(&path, type_page, flags, mnt_flags,
2352                      dev_name, data_page);
2353dput_out:
2354    path_put(&path);
2355    return retval;
2356}
2357
2358static struct mnt_namespace *alloc_mnt_ns(void)
2359{
2360    struct mnt_namespace *new_ns;
2361
2362    new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2363    if (!new_ns)
2364        return ERR_PTR(-ENOMEM);
2365    atomic_set(&new_ns->count, 1);
2366    new_ns->root = NULL;
2367    INIT_LIST_HEAD(&new_ns->list);
2368    init_waitqueue_head(&new_ns->poll);
2369    new_ns->event = 0;
2370    return new_ns;
2371}
2372
2373void mnt_make_longterm(struct vfsmount *mnt)
2374{
2375    __mnt_make_longterm(mnt);
2376}
2377
2378void mnt_make_shortterm(struct vfsmount *mnt)
2379{
2380#ifdef CONFIG_SMP
2381    if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2382        return;
2383    br_write_lock(vfsmount_lock);
2384    atomic_dec(&mnt->mnt_longterm);
2385    br_write_unlock(vfsmount_lock);
2386#endif
2387}
2388
2389/*
2390 * Allocate a new namespace structure and populate it with contents
2391 * copied from the namespace of the passed in task structure.
2392 */
2393static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2394        struct fs_struct *fs)
2395{
2396    struct mnt_namespace *new_ns;
2397    struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2398    struct vfsmount *p, *q;
2399
2400    new_ns = alloc_mnt_ns();
2401    if (IS_ERR(new_ns))
2402        return new_ns;
2403
2404    down_write(&namespace_sem);
2405    /* First pass: copy the tree topology */
2406    new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2407                    CL_COPY_ALL | CL_EXPIRE);
2408    if (!new_ns->root) {
2409        up_write(&namespace_sem);
2410        kfree(new_ns);
2411        return ERR_PTR(-ENOMEM);
2412    }
2413    br_write_lock(vfsmount_lock);
2414    list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2415    br_write_unlock(vfsmount_lock);
2416
2417    /*
2418     * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2419     * as belonging to new namespace. We have already acquired a private
2420     * fs_struct, so tsk->fs->lock is not needed.
2421     */
2422    p = mnt_ns->root;
2423    q = new_ns->root;
2424    while (p) {
2425        q->mnt_ns = new_ns;
2426        __mnt_make_longterm(q);
2427        if (fs) {
2428            if (p == fs->root.mnt) {
2429                fs->root.mnt = mntget(q);
2430                __mnt_make_longterm(q);
2431                mnt_make_shortterm(p);
2432                rootmnt = p;
2433            }
2434            if (p == fs->pwd.mnt) {
2435                fs->pwd.mnt = mntget(q);
2436                __mnt_make_longterm(q);
2437                mnt_make_shortterm(p);
2438                pwdmnt = p;
2439            }
2440        }
2441        p = next_mnt(p, mnt_ns->root);
2442        q = next_mnt(q, new_ns->root);
2443    }
2444    up_write(&namespace_sem);
2445
2446    if (rootmnt)
2447        mntput(rootmnt);
2448    if (pwdmnt)
2449        mntput(pwdmnt);
2450
2451    return new_ns;
2452}
2453
2454struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2455        struct fs_struct *new_fs)
2456{
2457    struct mnt_namespace *new_ns;
2458
2459    BUG_ON(!ns);
2460    get_mnt_ns(ns);
2461
2462    if (!(flags & CLONE_NEWNS))
2463        return ns;
2464
2465    new_ns = dup_mnt_ns(ns, new_fs);
2466
2467    put_mnt_ns(ns);
2468    return new_ns;
2469}
2470
2471/**
2472 * create_mnt_ns - creates a private namespace and adds a root filesystem
2473 * @mnt: pointer to the new root filesystem mountpoint
2474 */
2475struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2476{
2477    struct mnt_namespace *new_ns;
2478
2479    new_ns = alloc_mnt_ns();
2480    if (!IS_ERR(new_ns)) {
2481        mnt->mnt_ns = new_ns;
2482        __mnt_make_longterm(mnt);
2483        new_ns->root = mnt;
2484        list_add(&new_ns->list, &new_ns->root->mnt_list);
2485    }
2486    return new_ns;
2487}
2488EXPORT_SYMBOL(create_mnt_ns);
2489
2490SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2491        char __user *, type, unsigned long, flags, void __user *, data)
2492{
2493    int ret;
2494    char *kernel_type;
2495    char *kernel_dir;
2496    char *kernel_dev;
2497    unsigned long data_page;
2498
2499    ret = copy_mount_string(type, &kernel_type);
2500    if (ret < 0)
2501        goto out_type;
2502
2503    kernel_dir = getname(dir_name);
2504    if (IS_ERR(kernel_dir)) {
2505        ret = PTR_ERR(kernel_dir);
2506        goto out_dir;
2507    }
2508
2509    ret = copy_mount_string(dev_name, &kernel_dev);
2510    if (ret < 0)
2511        goto out_dev;
2512
2513    ret = copy_mount_options(data, &data_page);
2514    if (ret < 0)
2515        goto out_data;
2516
2517    ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2518        (void *) data_page);
2519
2520    free_page(data_page);
2521out_data:
2522    kfree(kernel_dev);
2523out_dev:
2524    putname(kernel_dir);
2525out_dir:
2526    kfree(kernel_type);
2527out_type:
2528    return ret;
2529}
2530
2531/*
2532 * pivot_root Semantics:
2533 * Moves the root file system of the current process to the directory put_old,
2534 * makes new_root as the new root file system of the current process, and sets
2535 * root/cwd of all processes which had them on the current root to new_root.
2536 *
2537 * Restrictions:
2538 * The new_root and put_old must be directories, and must not be on the
2539 * same file system as the current process root. The put_old must be
2540 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2541 * pointed to by put_old must yield the same directory as new_root. No other
2542 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2543 *
2544 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2545 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2546 * in this situation.
2547 *
2548 * Notes:
2549 * - we don't move root/cwd if they are not at the root (reason: if something
2550 * cared enough to change them, it's probably wrong to force them elsewhere)
2551 * - it's okay to pick a root that isn't the root of a file system, e.g.
2552 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2553 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2554 * first.
2555 */
2556SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2557        const char __user *, put_old)
2558{
2559    struct vfsmount *tmp;
2560    struct path new, old, parent_path, root_parent, root;
2561    int error;
2562
2563    if (!capable(CAP_SYS_ADMIN))
2564        return -EPERM;
2565
2566    error = user_path_dir(new_root, &new);
2567    if (error)
2568        goto out0;
2569
2570    error = user_path_dir(put_old, &old);
2571    if (error)
2572        goto out1;
2573
2574    error = security_sb_pivotroot(&old, &new);
2575    if (error)
2576        goto out2;
2577
2578    get_fs_root(current->fs, &root);
2579    error = lock_mount(&old);
2580    if (error)
2581        goto out3;
2582
2583    error = -EINVAL;
2584    if (IS_MNT_SHARED(old.mnt) ||
2585        IS_MNT_SHARED(new.mnt->mnt_parent) ||
2586        IS_MNT_SHARED(root.mnt->mnt_parent))
2587        goto out4;
2588    if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
2589        goto out4;
2590    error = -ENOENT;
2591    if (d_unlinked(new.dentry))
2592        goto out4;
2593    if (d_unlinked(old.dentry))
2594        goto out4;
2595    error = -EBUSY;
2596    if (new.mnt == root.mnt ||
2597        old.mnt == root.mnt)
2598        goto out4; /* loop, on the same file system */
2599    error = -EINVAL;
2600    if (root.mnt->mnt_root != root.dentry)
2601        goto out4; /* not a mountpoint */
2602    if (root.mnt->mnt_parent == root.mnt)
2603        goto out4; /* not attached */
2604    if (new.mnt->mnt_root != new.dentry)
2605        goto out4; /* not a mountpoint */
2606    if (new.mnt->mnt_parent == new.mnt)
2607        goto out4; /* not attached */
2608    /* make sure we can reach put_old from new_root */
2609    tmp = old.mnt;
2610    if (tmp != new.mnt) {
2611        for (;;) {
2612            if (tmp->mnt_parent == tmp)
2613                goto out4; /* already mounted on put_old */
2614            if (tmp->mnt_parent == new.mnt)
2615                break;
2616            tmp = tmp->mnt_parent;
2617        }
2618        if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2619            goto out4;
2620    } else if (!is_subdir(old.dentry, new.dentry))
2621        goto out4;
2622    br_write_lock(vfsmount_lock);
2623    detach_mnt(new.mnt, &parent_path);
2624    detach_mnt(root.mnt, &root_parent);
2625    /* mount old root on put_old */
2626    attach_mnt(root.mnt, &old);
2627    /* mount new_root on / */
2628    attach_mnt(new.mnt, &root_parent);
2629    touch_mnt_namespace(current->nsproxy->mnt_ns);
2630    br_write_unlock(vfsmount_lock);
2631    chroot_fs_refs(&root, &new);
2632    error = 0;
2633out4:
2634    unlock_mount(&old);
2635    if (!error) {
2636        path_put(&root_parent);
2637        path_put(&parent_path);
2638    }
2639out3:
2640    path_put(&root);
2641out2:
2642    path_put(&old);
2643out1:
2644    path_put(&new);
2645out0:
2646    return error;
2647}
2648
2649static void __init init_mount_tree(void)
2650{
2651    struct vfsmount *mnt;
2652    struct mnt_namespace *ns;
2653    struct path root;
2654
2655    mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2656    if (IS_ERR(mnt))
2657        panic("Can't create rootfs");
2658
2659    ns = create_mnt_ns(mnt);
2660    if (IS_ERR(ns))
2661        panic("Can't allocate initial namespace");
2662
2663    init_task.nsproxy->mnt_ns = ns;
2664    get_mnt_ns(ns);
2665
2666    root.mnt = ns->root;
2667    root.dentry = ns->root->mnt_root;
2668
2669    set_fs_pwd(current->fs, &root);
2670    set_fs_root(current->fs, &root);
2671}
2672
2673void __init mnt_init(void)
2674{
2675    unsigned u;
2676    int err;
2677
2678    init_rwsem(&namespace_sem);
2679
2680    mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2681            0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2682
2683    mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2684
2685    if (!mount_hashtable)
2686        panic("Failed to allocate mount hash table\n");
2687
2688    printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2689
2690    for (u = 0; u < HASH_SIZE; u++)
2691        INIT_LIST_HEAD(&mount_hashtable[u]);
2692
2693    br_lock_init(vfsmount_lock);
2694
2695    err = sysfs_init();
2696    if (err)
2697        printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2698            __func__, err);
2699    fs_kobj = kobject_create_and_add("fs", NULL);
2700    if (!fs_kobj)
2701        printk(KERN_WARNING "%s: kobj create error\n", __func__);
2702    init_rootfs();
2703    init_mount_tree();
2704}
2705
2706void put_mnt_ns(struct mnt_namespace *ns)
2707{
2708    LIST_HEAD(umount_list);
2709
2710    if (!atomic_dec_and_test(&ns->count))
2711        return;
2712    down_write(&namespace_sem);
2713    br_write_lock(vfsmount_lock);
2714    umount_tree(ns->root, 0, &umount_list);
2715    br_write_unlock(vfsmount_lock);
2716    up_write(&namespace_sem);
2717    release_mounts(&umount_list);
2718    kfree(ns);
2719}
2720EXPORT_SYMBOL(put_mnt_ns);
2721
2722struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2723{
2724    return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2725}
2726EXPORT_SYMBOL_GPL(kern_mount_data);
2727

Archive Download this file



interactive