Merge tag 'gpio-for-v3.11-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
[linux-2.6.git] / fs / namespace.c
blob7b1ca9ba0b0a70213915f6d7bce688184ab7a56f
1 /*
2 * linux/fs/namespace.c
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
11 #include <linux/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/acct.h> /* acct_auto_close_mnt */
20 #include <linux/ramfs.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
26 #include "pnode.h"
27 #include "internal.h"
29 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
30 #define HASH_SIZE (1UL << HASH_SHIFT)
32 static int event;
33 static DEFINE_IDA(mnt_id_ida);
34 static DEFINE_IDA(mnt_group_ida);
35 static DEFINE_SPINLOCK(mnt_id_lock);
36 static int mnt_id_start = 0;
37 static int mnt_group_start = 1;
39 static struct list_head *mount_hashtable __read_mostly;
40 static struct list_head *mountpoint_hashtable __read_mostly;
41 static struct kmem_cache *mnt_cache __read_mostly;
42 static struct rw_semaphore namespace_sem;
44 /* /sys/fs */
45 struct kobject *fs_kobj;
46 EXPORT_SYMBOL_GPL(fs_kobj);
49 * vfsmount lock may be taken for read to prevent changes to the
50 * vfsmount hash, ie. during mountpoint lookups or walking back
51 * up the tree.
53 * It should be taken for write in all cases where the vfsmount
54 * tree or hash is modified or when a vfsmount structure is modified.
56 DEFINE_BRLOCK(vfsmount_lock);
58 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
60 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
61 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
62 tmp = tmp + (tmp >> HASH_SHIFT);
63 return tmp & (HASH_SIZE - 1);
66 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
69 * allocation is serialized by namespace_sem, but we need the spinlock to
70 * serialize with freeing.
72 static int mnt_alloc_id(struct mount *mnt)
74 int res;
76 retry:
77 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
78 spin_lock(&mnt_id_lock);
79 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
80 if (!res)
81 mnt_id_start = mnt->mnt_id + 1;
82 spin_unlock(&mnt_id_lock);
83 if (res == -EAGAIN)
84 goto retry;
86 return res;
89 static void mnt_free_id(struct mount *mnt)
91 int id = mnt->mnt_id;
92 spin_lock(&mnt_id_lock);
93 ida_remove(&mnt_id_ida, id);
94 if (mnt_id_start > id)
95 mnt_id_start = id;
96 spin_unlock(&mnt_id_lock);
100 * Allocate a new peer group ID
102 * mnt_group_ida is protected by namespace_sem
104 static int mnt_alloc_group_id(struct mount *mnt)
106 int res;
108 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
109 return -ENOMEM;
111 res = ida_get_new_above(&mnt_group_ida,
112 mnt_group_start,
113 &mnt->mnt_group_id);
114 if (!res)
115 mnt_group_start = mnt->mnt_group_id + 1;
117 return res;
121 * Release a peer group ID
123 void mnt_release_group_id(struct mount *mnt)
125 int id = mnt->mnt_group_id;
126 ida_remove(&mnt_group_ida, id);
127 if (mnt_group_start > id)
128 mnt_group_start = id;
129 mnt->mnt_group_id = 0;
133 * vfsmount lock must be held for read
135 static inline void mnt_add_count(struct mount *mnt, int n)
137 #ifdef CONFIG_SMP
138 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
139 #else
140 preempt_disable();
141 mnt->mnt_count += n;
142 preempt_enable();
143 #endif
147 * vfsmount lock must be held for write
149 unsigned int mnt_get_count(struct mount *mnt)
151 #ifdef CONFIG_SMP
152 unsigned int count = 0;
153 int cpu;
155 for_each_possible_cpu(cpu) {
156 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
159 return count;
160 #else
161 return mnt->mnt_count;
162 #endif
165 static struct mount *alloc_vfsmnt(const char *name)
167 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
168 if (mnt) {
169 int err;
171 err = mnt_alloc_id(mnt);
172 if (err)
173 goto out_free_cache;
175 if (name) {
176 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
177 if (!mnt->mnt_devname)
178 goto out_free_id;
181 #ifdef CONFIG_SMP
182 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
183 if (!mnt->mnt_pcp)
184 goto out_free_devname;
186 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
187 #else
188 mnt->mnt_count = 1;
189 mnt->mnt_writers = 0;
190 #endif
192 INIT_LIST_HEAD(&mnt->mnt_hash);
193 INIT_LIST_HEAD(&mnt->mnt_child);
194 INIT_LIST_HEAD(&mnt->mnt_mounts);
195 INIT_LIST_HEAD(&mnt->mnt_list);
196 INIT_LIST_HEAD(&mnt->mnt_expire);
197 INIT_LIST_HEAD(&mnt->mnt_share);
198 INIT_LIST_HEAD(&mnt->mnt_slave_list);
199 INIT_LIST_HEAD(&mnt->mnt_slave);
200 #ifdef CONFIG_FSNOTIFY
201 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
202 #endif
204 return mnt;
206 #ifdef CONFIG_SMP
207 out_free_devname:
208 kfree(mnt->mnt_devname);
209 #endif
210 out_free_id:
211 mnt_free_id(mnt);
212 out_free_cache:
213 kmem_cache_free(mnt_cache, mnt);
214 return NULL;
218 * Most r/o checks on a fs are for operations that take
219 * discrete amounts of time, like a write() or unlink().
220 * We must keep track of when those operations start
221 * (for permission checks) and when they end, so that
222 * we can determine when writes are able to occur to
223 * a filesystem.
226 * __mnt_is_readonly: check whether a mount is read-only
227 * @mnt: the mount to check for its write status
229 * This shouldn't be used directly ouside of the VFS.
230 * It does not guarantee that the filesystem will stay
231 * r/w, just that it is right *now*. This can not and
232 * should not be used in place of IS_RDONLY(inode).
233 * mnt_want/drop_write() will _keep_ the filesystem
234 * r/w.
236 int __mnt_is_readonly(struct vfsmount *mnt)
238 if (mnt->mnt_flags & MNT_READONLY)
239 return 1;
240 if (mnt->mnt_sb->s_flags & MS_RDONLY)
241 return 1;
242 return 0;
244 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
246 static inline void mnt_inc_writers(struct mount *mnt)
248 #ifdef CONFIG_SMP
249 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
250 #else
251 mnt->mnt_writers++;
252 #endif
255 static inline void mnt_dec_writers(struct mount *mnt)
257 #ifdef CONFIG_SMP
258 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
259 #else
260 mnt->mnt_writers--;
261 #endif
264 static unsigned int mnt_get_writers(struct mount *mnt)
266 #ifdef CONFIG_SMP
267 unsigned int count = 0;
268 int cpu;
270 for_each_possible_cpu(cpu) {
271 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
274 return count;
275 #else
276 return mnt->mnt_writers;
277 #endif
280 static int mnt_is_readonly(struct vfsmount *mnt)
282 if (mnt->mnt_sb->s_readonly_remount)
283 return 1;
284 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
285 smp_rmb();
286 return __mnt_is_readonly(mnt);
290 * Most r/o & frozen checks on a fs are for operations that take discrete
291 * amounts of time, like a write() or unlink(). We must keep track of when
292 * those operations start (for permission checks) and when they end, so that we
293 * can determine when writes are able to occur to a filesystem.
296 * __mnt_want_write - get write access to a mount without freeze protection
297 * @m: the mount on which to take a write
299 * This tells the low-level filesystem that a write is about to be performed to
300 * it, and makes sure that writes are allowed (mnt it read-write) before
301 * returning success. This operation does not protect against filesystem being
302 * frozen. When the write operation is finished, __mnt_drop_write() must be
303 * called. This is effectively a refcount.
305 int __mnt_want_write(struct vfsmount *m)
307 struct mount *mnt = real_mount(m);
308 int ret = 0;
310 preempt_disable();
311 mnt_inc_writers(mnt);
313 * The store to mnt_inc_writers must be visible before we pass
314 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
315 * incremented count after it has set MNT_WRITE_HOLD.
317 smp_mb();
318 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
319 cpu_relax();
321 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
322 * be set to match its requirements. So we must not load that until
323 * MNT_WRITE_HOLD is cleared.
325 smp_rmb();
326 if (mnt_is_readonly(m)) {
327 mnt_dec_writers(mnt);
328 ret = -EROFS;
330 preempt_enable();
332 return ret;
336 * mnt_want_write - get write access to a mount
337 * @m: the mount on which to take a write
339 * This tells the low-level filesystem that a write is about to be performed to
340 * it, and makes sure that writes are allowed (mount is read-write, filesystem
341 * is not frozen) before returning success. When the write operation is
342 * finished, mnt_drop_write() must be called. This is effectively a refcount.
344 int mnt_want_write(struct vfsmount *m)
346 int ret;
348 sb_start_write(m->mnt_sb);
349 ret = __mnt_want_write(m);
350 if (ret)
351 sb_end_write(m->mnt_sb);
352 return ret;
354 EXPORT_SYMBOL_GPL(mnt_want_write);
357 * mnt_clone_write - get write access to a mount
358 * @mnt: the mount on which to take a write
360 * This is effectively like mnt_want_write, except
361 * it must only be used to take an extra write reference
362 * on a mountpoint that we already know has a write reference
363 * on it. This allows some optimisation.
365 * After finished, mnt_drop_write must be called as usual to
366 * drop the reference.
368 int mnt_clone_write(struct vfsmount *mnt)
370 /* superblock may be r/o */
371 if (__mnt_is_readonly(mnt))
372 return -EROFS;
373 preempt_disable();
374 mnt_inc_writers(real_mount(mnt));
375 preempt_enable();
376 return 0;
378 EXPORT_SYMBOL_GPL(mnt_clone_write);
381 * __mnt_want_write_file - get write access to a file's mount
382 * @file: the file who's mount on which to take a write
384 * This is like __mnt_want_write, but it takes a file and can
385 * do some optimisations if the file is open for write already
387 int __mnt_want_write_file(struct file *file)
389 struct inode *inode = file_inode(file);
391 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
392 return __mnt_want_write(file->f_path.mnt);
393 else
394 return mnt_clone_write(file->f_path.mnt);
398 * mnt_want_write_file - get write access to a file's mount
399 * @file: the file who's mount on which to take a write
401 * This is like mnt_want_write, but it takes a file and can
402 * do some optimisations if the file is open for write already
404 int mnt_want_write_file(struct file *file)
406 int ret;
408 sb_start_write(file->f_path.mnt->mnt_sb);
409 ret = __mnt_want_write_file(file);
410 if (ret)
411 sb_end_write(file->f_path.mnt->mnt_sb);
412 return ret;
414 EXPORT_SYMBOL_GPL(mnt_want_write_file);
417 * __mnt_drop_write - give up write access to a mount
418 * @mnt: the mount on which to give up write access
420 * Tells the low-level filesystem that we are done
421 * performing writes to it. Must be matched with
422 * __mnt_want_write() call above.
424 void __mnt_drop_write(struct vfsmount *mnt)
426 preempt_disable();
427 mnt_dec_writers(real_mount(mnt));
428 preempt_enable();
432 * mnt_drop_write - give up write access to a mount
433 * @mnt: the mount on which to give up write access
435 * Tells the low-level filesystem that we are done performing writes to it and
436 * also allows filesystem to be frozen again. Must be matched with
437 * mnt_want_write() call above.
439 void mnt_drop_write(struct vfsmount *mnt)
441 __mnt_drop_write(mnt);
442 sb_end_write(mnt->mnt_sb);
444 EXPORT_SYMBOL_GPL(mnt_drop_write);
446 void __mnt_drop_write_file(struct file *file)
448 __mnt_drop_write(file->f_path.mnt);
451 void mnt_drop_write_file(struct file *file)
453 mnt_drop_write(file->f_path.mnt);
455 EXPORT_SYMBOL(mnt_drop_write_file);
457 static int mnt_make_readonly(struct mount *mnt)
459 int ret = 0;
461 br_write_lock(&vfsmount_lock);
462 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
464 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
465 * should be visible before we do.
467 smp_mb();
470 * With writers on hold, if this value is zero, then there are
471 * definitely no active writers (although held writers may subsequently
472 * increment the count, they'll have to wait, and decrement it after
473 * seeing MNT_READONLY).
475 * It is OK to have counter incremented on one CPU and decremented on
476 * another: the sum will add up correctly. The danger would be when we
477 * sum up each counter, if we read a counter before it is incremented,
478 * but then read another CPU's count which it has been subsequently
479 * decremented from -- we would see more decrements than we should.
480 * MNT_WRITE_HOLD protects against this scenario, because
481 * mnt_want_write first increments count, then smp_mb, then spins on
482 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
483 * we're counting up here.
485 if (mnt_get_writers(mnt) > 0)
486 ret = -EBUSY;
487 else
488 mnt->mnt.mnt_flags |= MNT_READONLY;
490 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
491 * that become unheld will see MNT_READONLY.
493 smp_wmb();
494 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
495 br_write_unlock(&vfsmount_lock);
496 return ret;
499 static void __mnt_unmake_readonly(struct mount *mnt)
501 br_write_lock(&vfsmount_lock);
502 mnt->mnt.mnt_flags &= ~MNT_READONLY;
503 br_write_unlock(&vfsmount_lock);
506 int sb_prepare_remount_readonly(struct super_block *sb)
508 struct mount *mnt;
509 int err = 0;
511 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
512 if (atomic_long_read(&sb->s_remove_count))
513 return -EBUSY;
515 br_write_lock(&vfsmount_lock);
516 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
517 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
518 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
519 smp_mb();
520 if (mnt_get_writers(mnt) > 0) {
521 err = -EBUSY;
522 break;
526 if (!err && atomic_long_read(&sb->s_remove_count))
527 err = -EBUSY;
529 if (!err) {
530 sb->s_readonly_remount = 1;
531 smp_wmb();
533 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
534 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
535 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
537 br_write_unlock(&vfsmount_lock);
539 return err;
542 static void free_vfsmnt(struct mount *mnt)
544 kfree(mnt->mnt_devname);
545 mnt_free_id(mnt);
546 #ifdef CONFIG_SMP
547 free_percpu(mnt->mnt_pcp);
548 #endif
549 kmem_cache_free(mnt_cache, mnt);
553 * find the first or last mount at @dentry on vfsmount @mnt depending on
554 * @dir. If @dir is set return the first mount else return the last mount.
555 * vfsmount_lock must be held for read or write.
557 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
558 int dir)
560 struct list_head *head = mount_hashtable + hash(mnt, dentry);
561 struct list_head *tmp = head;
562 struct mount *p, *found = NULL;
564 for (;;) {
565 tmp = dir ? tmp->next : tmp->prev;
566 p = NULL;
567 if (tmp == head)
568 break;
569 p = list_entry(tmp, struct mount, mnt_hash);
570 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
571 found = p;
572 break;
575 return found;
579 * lookup_mnt - Return the first child mount mounted at path
581 * "First" means first mounted chronologically. If you create the
582 * following mounts:
584 * mount /dev/sda1 /mnt
585 * mount /dev/sda2 /mnt
586 * mount /dev/sda3 /mnt
588 * Then lookup_mnt() on the base /mnt dentry in the root mount will
589 * return successively the root dentry and vfsmount of /dev/sda1, then
590 * /dev/sda2, then /dev/sda3, then NULL.
592 * lookup_mnt takes a reference to the found vfsmount.
594 struct vfsmount *lookup_mnt(struct path *path)
596 struct mount *child_mnt;
598 br_read_lock(&vfsmount_lock);
599 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
600 if (child_mnt) {
601 mnt_add_count(child_mnt, 1);
602 br_read_unlock(&vfsmount_lock);
603 return &child_mnt->mnt;
604 } else {
605 br_read_unlock(&vfsmount_lock);
606 return NULL;
610 static struct mountpoint *new_mountpoint(struct dentry *dentry)
612 struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry);
613 struct mountpoint *mp;
615 list_for_each_entry(mp, chain, m_hash) {
616 if (mp->m_dentry == dentry) {
617 /* might be worth a WARN_ON() */
618 if (d_unlinked(dentry))
619 return ERR_PTR(-ENOENT);
620 mp->m_count++;
621 return mp;
625 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
626 if (!mp)
627 return ERR_PTR(-ENOMEM);
629 spin_lock(&dentry->d_lock);
630 if (d_unlinked(dentry)) {
631 spin_unlock(&dentry->d_lock);
632 kfree(mp);
633 return ERR_PTR(-ENOENT);
635 dentry->d_flags |= DCACHE_MOUNTED;
636 spin_unlock(&dentry->d_lock);
637 mp->m_dentry = dentry;
638 mp->m_count = 1;
639 list_add(&mp->m_hash, chain);
640 return mp;
643 static void put_mountpoint(struct mountpoint *mp)
645 if (!--mp->m_count) {
646 struct dentry *dentry = mp->m_dentry;
647 spin_lock(&dentry->d_lock);
648 dentry->d_flags &= ~DCACHE_MOUNTED;
649 spin_unlock(&dentry->d_lock);
650 list_del(&mp->m_hash);
651 kfree(mp);
655 static inline int check_mnt(struct mount *mnt)
657 return mnt->mnt_ns == current->nsproxy->mnt_ns;
661 * vfsmount lock must be held for write
663 static void touch_mnt_namespace(struct mnt_namespace *ns)
665 if (ns) {
666 ns->event = ++event;
667 wake_up_interruptible(&ns->poll);
672 * vfsmount lock must be held for write
674 static void __touch_mnt_namespace(struct mnt_namespace *ns)
676 if (ns && ns->event != event) {
677 ns->event = event;
678 wake_up_interruptible(&ns->poll);
683 * vfsmount lock must be held for write
685 static void detach_mnt(struct mount *mnt, struct path *old_path)
687 old_path->dentry = mnt->mnt_mountpoint;
688 old_path->mnt = &mnt->mnt_parent->mnt;
689 mnt->mnt_parent = mnt;
690 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
691 list_del_init(&mnt->mnt_child);
692 list_del_init(&mnt->mnt_hash);
693 put_mountpoint(mnt->mnt_mp);
694 mnt->mnt_mp = NULL;
698 * vfsmount lock must be held for write
700 void mnt_set_mountpoint(struct mount *mnt,
701 struct mountpoint *mp,
702 struct mount *child_mnt)
704 mp->m_count++;
705 mnt_add_count(mnt, 1); /* essentially, that's mntget */
706 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
707 child_mnt->mnt_parent = mnt;
708 child_mnt->mnt_mp = mp;
712 * vfsmount lock must be held for write
714 static void attach_mnt(struct mount *mnt,
715 struct mount *parent,
716 struct mountpoint *mp)
718 mnt_set_mountpoint(parent, mp, mnt);
719 list_add_tail(&mnt->mnt_hash, mount_hashtable +
720 hash(&parent->mnt, mp->m_dentry));
721 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
725 * vfsmount lock must be held for write
727 static void commit_tree(struct mount *mnt)
729 struct mount *parent = mnt->mnt_parent;
730 struct mount *m;
731 LIST_HEAD(head);
732 struct mnt_namespace *n = parent->mnt_ns;
734 BUG_ON(parent == mnt);
736 list_add_tail(&head, &mnt->mnt_list);
737 list_for_each_entry(m, &head, mnt_list)
738 m->mnt_ns = n;
740 list_splice(&head, n->list.prev);
742 list_add_tail(&mnt->mnt_hash, mount_hashtable +
743 hash(&parent->mnt, mnt->mnt_mountpoint));
744 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
745 touch_mnt_namespace(n);
748 static struct mount *next_mnt(struct mount *p, struct mount *root)
750 struct list_head *next = p->mnt_mounts.next;
751 if (next == &p->mnt_mounts) {
752 while (1) {
753 if (p == root)
754 return NULL;
755 next = p->mnt_child.next;
756 if (next != &p->mnt_parent->mnt_mounts)
757 break;
758 p = p->mnt_parent;
761 return list_entry(next, struct mount, mnt_child);
764 static struct mount *skip_mnt_tree(struct mount *p)
766 struct list_head *prev = p->mnt_mounts.prev;
767 while (prev != &p->mnt_mounts) {
768 p = list_entry(prev, struct mount, mnt_child);
769 prev = p->mnt_mounts.prev;
771 return p;
774 struct vfsmount *
775 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
777 struct mount *mnt;
778 struct dentry *root;
780 if (!type)
781 return ERR_PTR(-ENODEV);
783 mnt = alloc_vfsmnt(name);
784 if (!mnt)
785 return ERR_PTR(-ENOMEM);
787 if (flags & MS_KERNMOUNT)
788 mnt->mnt.mnt_flags = MNT_INTERNAL;
790 root = mount_fs(type, flags, name, data);
791 if (IS_ERR(root)) {
792 free_vfsmnt(mnt);
793 return ERR_CAST(root);
796 mnt->mnt.mnt_root = root;
797 mnt->mnt.mnt_sb = root->d_sb;
798 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
799 mnt->mnt_parent = mnt;
800 br_write_lock(&vfsmount_lock);
801 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
802 br_write_unlock(&vfsmount_lock);
803 return &mnt->mnt;
805 EXPORT_SYMBOL_GPL(vfs_kern_mount);
807 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
808 int flag)
810 struct super_block *sb = old->mnt.mnt_sb;
811 struct mount *mnt;
812 int err;
814 mnt = alloc_vfsmnt(old->mnt_devname);
815 if (!mnt)
816 return ERR_PTR(-ENOMEM);
818 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
819 mnt->mnt_group_id = 0; /* not a peer of original */
820 else
821 mnt->mnt_group_id = old->mnt_group_id;
823 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
824 err = mnt_alloc_group_id(mnt);
825 if (err)
826 goto out_free;
829 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
830 /* Don't allow unprivileged users to change mount flags */
831 if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
832 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
834 atomic_inc(&sb->s_active);
835 mnt->mnt.mnt_sb = sb;
836 mnt->mnt.mnt_root = dget(root);
837 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
838 mnt->mnt_parent = mnt;
839 br_write_lock(&vfsmount_lock);
840 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
841 br_write_unlock(&vfsmount_lock);
843 if ((flag & CL_SLAVE) ||
844 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
845 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
846 mnt->mnt_master = old;
847 CLEAR_MNT_SHARED(mnt);
848 } else if (!(flag & CL_PRIVATE)) {
849 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
850 list_add(&mnt->mnt_share, &old->mnt_share);
851 if (IS_MNT_SLAVE(old))
852 list_add(&mnt->mnt_slave, &old->mnt_slave);
853 mnt->mnt_master = old->mnt_master;
855 if (flag & CL_MAKE_SHARED)
856 set_mnt_shared(mnt);
858 /* stick the duplicate mount on the same expiry list
859 * as the original if that was on one */
860 if (flag & CL_EXPIRE) {
861 if (!list_empty(&old->mnt_expire))
862 list_add(&mnt->mnt_expire, &old->mnt_expire);
865 return mnt;
867 out_free:
868 free_vfsmnt(mnt);
869 return ERR_PTR(err);
872 static inline void mntfree(struct mount *mnt)
874 struct vfsmount *m = &mnt->mnt;
875 struct super_block *sb = m->mnt_sb;
878 * This probably indicates that somebody messed
879 * up a mnt_want/drop_write() pair. If this
880 * happens, the filesystem was probably unable
881 * to make r/w->r/o transitions.
884 * The locking used to deal with mnt_count decrement provides barriers,
885 * so mnt_get_writers() below is safe.
887 WARN_ON(mnt_get_writers(mnt));
888 fsnotify_vfsmount_delete(m);
889 dput(m->mnt_root);
890 free_vfsmnt(mnt);
891 deactivate_super(sb);
894 static void mntput_no_expire(struct mount *mnt)
896 put_again:
897 #ifdef CONFIG_SMP
898 br_read_lock(&vfsmount_lock);
899 if (likely(mnt->mnt_ns)) {
900 /* shouldn't be the last one */
901 mnt_add_count(mnt, -1);
902 br_read_unlock(&vfsmount_lock);
903 return;
905 br_read_unlock(&vfsmount_lock);
907 br_write_lock(&vfsmount_lock);
908 mnt_add_count(mnt, -1);
909 if (mnt_get_count(mnt)) {
910 br_write_unlock(&vfsmount_lock);
911 return;
913 #else
914 mnt_add_count(mnt, -1);
915 if (likely(mnt_get_count(mnt)))
916 return;
917 br_write_lock(&vfsmount_lock);
918 #endif
919 if (unlikely(mnt->mnt_pinned)) {
920 mnt_add_count(mnt, mnt->mnt_pinned + 1);
921 mnt->mnt_pinned = 0;
922 br_write_unlock(&vfsmount_lock);
923 acct_auto_close_mnt(&mnt->mnt);
924 goto put_again;
927 list_del(&mnt->mnt_instance);
928 br_write_unlock(&vfsmount_lock);
929 mntfree(mnt);
932 void mntput(struct vfsmount *mnt)
934 if (mnt) {
935 struct mount *m = real_mount(mnt);
936 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
937 if (unlikely(m->mnt_expiry_mark))
938 m->mnt_expiry_mark = 0;
939 mntput_no_expire(m);
942 EXPORT_SYMBOL(mntput);
944 struct vfsmount *mntget(struct vfsmount *mnt)
946 if (mnt)
947 mnt_add_count(real_mount(mnt), 1);
948 return mnt;
950 EXPORT_SYMBOL(mntget);
952 void mnt_pin(struct vfsmount *mnt)
954 br_write_lock(&vfsmount_lock);
955 real_mount(mnt)->mnt_pinned++;
956 br_write_unlock(&vfsmount_lock);
958 EXPORT_SYMBOL(mnt_pin);
960 void mnt_unpin(struct vfsmount *m)
962 struct mount *mnt = real_mount(m);
963 br_write_lock(&vfsmount_lock);
964 if (mnt->mnt_pinned) {
965 mnt_add_count(mnt, 1);
966 mnt->mnt_pinned--;
968 br_write_unlock(&vfsmount_lock);
970 EXPORT_SYMBOL(mnt_unpin);
972 static inline void mangle(struct seq_file *m, const char *s)
974 seq_escape(m, s, " \t\n\\");
978 * Simple .show_options callback for filesystems which don't want to
979 * implement more complex mount option showing.
981 * See also save_mount_options().
983 int generic_show_options(struct seq_file *m, struct dentry *root)
985 const char *options;
987 rcu_read_lock();
988 options = rcu_dereference(root->d_sb->s_options);
990 if (options != NULL && options[0]) {
991 seq_putc(m, ',');
992 mangle(m, options);
994 rcu_read_unlock();
996 return 0;
998 EXPORT_SYMBOL(generic_show_options);
1001 * If filesystem uses generic_show_options(), this function should be
1002 * called from the fill_super() callback.
1004 * The .remount_fs callback usually needs to be handled in a special
1005 * way, to make sure, that previous options are not overwritten if the
1006 * remount fails.
1008 * Also note, that if the filesystem's .remount_fs function doesn't
1009 * reset all options to their default value, but changes only newly
1010 * given options, then the displayed options will not reflect reality
1011 * any more.
1013 void save_mount_options(struct super_block *sb, char *options)
1015 BUG_ON(sb->s_options);
1016 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1018 EXPORT_SYMBOL(save_mount_options);
1020 void replace_mount_options(struct super_block *sb, char *options)
1022 char *old = sb->s_options;
1023 rcu_assign_pointer(sb->s_options, options);
1024 if (old) {
1025 synchronize_rcu();
1026 kfree(old);
1029 EXPORT_SYMBOL(replace_mount_options);
1031 #ifdef CONFIG_PROC_FS
1032 /* iterator; we want it to have access to namespace_sem, thus here... */
1033 static void *m_start(struct seq_file *m, loff_t *pos)
1035 struct proc_mounts *p = proc_mounts(m);
1037 down_read(&namespace_sem);
1038 return seq_list_start(&p->ns->list, *pos);
1041 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1043 struct proc_mounts *p = proc_mounts(m);
1045 return seq_list_next(v, &p->ns->list, pos);
1048 static void m_stop(struct seq_file *m, void *v)
1050 up_read(&namespace_sem);
1053 static int m_show(struct seq_file *m, void *v)
1055 struct proc_mounts *p = proc_mounts(m);
1056 struct mount *r = list_entry(v, struct mount, mnt_list);
1057 return p->show(m, &r->mnt);
1060 const struct seq_operations mounts_op = {
1061 .start = m_start,
1062 .next = m_next,
1063 .stop = m_stop,
1064 .show = m_show,
1066 #endif /* CONFIG_PROC_FS */
1069 * may_umount_tree - check if a mount tree is busy
1070 * @mnt: root of mount tree
1072 * This is called to check if a tree of mounts has any
1073 * open files, pwds, chroots or sub mounts that are
1074 * busy.
1076 int may_umount_tree(struct vfsmount *m)
1078 struct mount *mnt = real_mount(m);
1079 int actual_refs = 0;
1080 int minimum_refs = 0;
1081 struct mount *p;
1082 BUG_ON(!m);
1084 /* write lock needed for mnt_get_count */
1085 br_write_lock(&vfsmount_lock);
1086 for (p = mnt; p; p = next_mnt(p, mnt)) {
1087 actual_refs += mnt_get_count(p);
1088 minimum_refs += 2;
1090 br_write_unlock(&vfsmount_lock);
1092 if (actual_refs > minimum_refs)
1093 return 0;
1095 return 1;
1098 EXPORT_SYMBOL(may_umount_tree);
1101 * may_umount - check if a mount point is busy
1102 * @mnt: root of mount
1104 * This is called to check if a mount point has any
1105 * open files, pwds, chroots or sub mounts. If the
1106 * mount has sub mounts this will return busy
1107 * regardless of whether the sub mounts are busy.
1109 * Doesn't take quota and stuff into account. IOW, in some cases it will
1110 * give false negatives. The main reason why it's here is that we need
1111 * a non-destructive way to look for easily umountable filesystems.
1113 int may_umount(struct vfsmount *mnt)
1115 int ret = 1;
1116 down_read(&namespace_sem);
1117 br_write_lock(&vfsmount_lock);
1118 if (propagate_mount_busy(real_mount(mnt), 2))
1119 ret = 0;
1120 br_write_unlock(&vfsmount_lock);
1121 up_read(&namespace_sem);
1122 return ret;
1125 EXPORT_SYMBOL(may_umount);
1127 static LIST_HEAD(unmounted); /* protected by namespace_sem */
1129 static void namespace_unlock(void)
1131 struct mount *mnt;
1132 LIST_HEAD(head);
1134 if (likely(list_empty(&unmounted))) {
1135 up_write(&namespace_sem);
1136 return;
1139 list_splice_init(&unmounted, &head);
1140 up_write(&namespace_sem);
1142 while (!list_empty(&head)) {
1143 mnt = list_first_entry(&head, struct mount, mnt_hash);
1144 list_del_init(&mnt->mnt_hash);
1145 if (mnt_has_parent(mnt)) {
1146 struct dentry *dentry;
1147 struct mount *m;
1149 br_write_lock(&vfsmount_lock);
1150 dentry = mnt->mnt_mountpoint;
1151 m = mnt->mnt_parent;
1152 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1153 mnt->mnt_parent = mnt;
1154 m->mnt_ghosts--;
1155 br_write_unlock(&vfsmount_lock);
1156 dput(dentry);
1157 mntput(&m->mnt);
1159 mntput(&mnt->mnt);
1163 static inline void namespace_lock(void)
1165 down_write(&namespace_sem);
1169 * vfsmount lock must be held for write
1170 * namespace_sem must be held for write
1172 void umount_tree(struct mount *mnt, int propagate)
1174 LIST_HEAD(tmp_list);
1175 struct mount *p;
1177 for (p = mnt; p; p = next_mnt(p, mnt))
1178 list_move(&p->mnt_hash, &tmp_list);
1180 if (propagate)
1181 propagate_umount(&tmp_list);
1183 list_for_each_entry(p, &tmp_list, mnt_hash) {
1184 list_del_init(&p->mnt_expire);
1185 list_del_init(&p->mnt_list);
1186 __touch_mnt_namespace(p->mnt_ns);
1187 p->mnt_ns = NULL;
1188 list_del_init(&p->mnt_child);
1189 if (mnt_has_parent(p)) {
1190 p->mnt_parent->mnt_ghosts++;
1191 put_mountpoint(p->mnt_mp);
1192 p->mnt_mp = NULL;
1194 change_mnt_propagation(p, MS_PRIVATE);
1196 list_splice(&tmp_list, &unmounted);
1199 static void shrink_submounts(struct mount *mnt);
1201 static int do_umount(struct mount *mnt, int flags)
1203 struct super_block *sb = mnt->mnt.mnt_sb;
1204 int retval;
1206 retval = security_sb_umount(&mnt->mnt, flags);
1207 if (retval)
1208 return retval;
1211 * Allow userspace to request a mountpoint be expired rather than
1212 * unmounting unconditionally. Unmount only happens if:
1213 * (1) the mark is already set (the mark is cleared by mntput())
1214 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1216 if (flags & MNT_EXPIRE) {
1217 if (&mnt->mnt == current->fs->root.mnt ||
1218 flags & (MNT_FORCE | MNT_DETACH))
1219 return -EINVAL;
1222 * probably don't strictly need the lock here if we examined
1223 * all race cases, but it's a slowpath.
1225 br_write_lock(&vfsmount_lock);
1226 if (mnt_get_count(mnt) != 2) {
1227 br_write_unlock(&vfsmount_lock);
1228 return -EBUSY;
1230 br_write_unlock(&vfsmount_lock);
1232 if (!xchg(&mnt->mnt_expiry_mark, 1))
1233 return -EAGAIN;
1237 * If we may have to abort operations to get out of this
1238 * mount, and they will themselves hold resources we must
1239 * allow the fs to do things. In the Unix tradition of
1240 * 'Gee thats tricky lets do it in userspace' the umount_begin
1241 * might fail to complete on the first run through as other tasks
1242 * must return, and the like. Thats for the mount program to worry
1243 * about for the moment.
1246 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1247 sb->s_op->umount_begin(sb);
1251 * No sense to grab the lock for this test, but test itself looks
1252 * somewhat bogus. Suggestions for better replacement?
1253 * Ho-hum... In principle, we might treat that as umount + switch
1254 * to rootfs. GC would eventually take care of the old vfsmount.
1255 * Actually it makes sense, especially if rootfs would contain a
1256 * /reboot - static binary that would close all descriptors and
1257 * call reboot(9). Then init(8) could umount root and exec /reboot.
1259 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1261 * Special case for "unmounting" root ...
1262 * we just try to remount it readonly.
1264 down_write(&sb->s_umount);
1265 if (!(sb->s_flags & MS_RDONLY))
1266 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1267 up_write(&sb->s_umount);
1268 return retval;
1271 namespace_lock();
1272 br_write_lock(&vfsmount_lock);
1273 event++;
1275 if (!(flags & MNT_DETACH))
1276 shrink_submounts(mnt);
1278 retval = -EBUSY;
1279 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1280 if (!list_empty(&mnt->mnt_list))
1281 umount_tree(mnt, 1);
1282 retval = 0;
1284 br_write_unlock(&vfsmount_lock);
1285 namespace_unlock();
1286 return retval;
1290 * Is the caller allowed to modify his namespace?
1292 static inline bool may_mount(void)
1294 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1298 * Now umount can handle mount points as well as block devices.
1299 * This is important for filesystems which use unnamed block devices.
1301 * We now support a flag for forced unmount like the other 'big iron'
1302 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1305 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1307 struct path path;
1308 struct mount *mnt;
1309 int retval;
1310 int lookup_flags = 0;
1312 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1313 return -EINVAL;
1315 if (!may_mount())
1316 return -EPERM;
1318 if (!(flags & UMOUNT_NOFOLLOW))
1319 lookup_flags |= LOOKUP_FOLLOW;
1321 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1322 if (retval)
1323 goto out;
1324 mnt = real_mount(path.mnt);
1325 retval = -EINVAL;
1326 if (path.dentry != path.mnt->mnt_root)
1327 goto dput_and_out;
1328 if (!check_mnt(mnt))
1329 goto dput_and_out;
1331 retval = do_umount(mnt, flags);
1332 dput_and_out:
1333 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1334 dput(path.dentry);
1335 mntput_no_expire(mnt);
1336 out:
1337 return retval;
1340 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1343 * The 2.0 compatible umount. No flags.
1345 SYSCALL_DEFINE1(oldumount, char __user *, name)
1347 return sys_umount(name, 0);
1350 #endif
1352 static bool mnt_ns_loop(struct path *path)
1354 /* Could bind mounting the mount namespace inode cause a
1355 * mount namespace loop?
1357 struct inode *inode = path->dentry->d_inode;
1358 struct proc_ns *ei;
1359 struct mnt_namespace *mnt_ns;
1361 if (!proc_ns_inode(inode))
1362 return false;
1364 ei = get_proc_ns(inode);
1365 if (ei->ns_ops != &mntns_operations)
1366 return false;
1368 mnt_ns = ei->ns;
1369 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1372 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1373 int flag)
1375 struct mount *res, *p, *q, *r, *parent;
1377 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1378 return ERR_PTR(-EINVAL);
1380 res = q = clone_mnt(mnt, dentry, flag);
1381 if (IS_ERR(q))
1382 return q;
1384 q->mnt_mountpoint = mnt->mnt_mountpoint;
1386 p = mnt;
1387 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1388 struct mount *s;
1389 if (!is_subdir(r->mnt_mountpoint, dentry))
1390 continue;
1392 for (s = r; s; s = next_mnt(s, r)) {
1393 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1394 s = skip_mnt_tree(s);
1395 continue;
1397 while (p != s->mnt_parent) {
1398 p = p->mnt_parent;
1399 q = q->mnt_parent;
1401 p = s;
1402 parent = q;
1403 q = clone_mnt(p, p->mnt.mnt_root, flag);
1404 if (IS_ERR(q))
1405 goto out;
1406 br_write_lock(&vfsmount_lock);
1407 list_add_tail(&q->mnt_list, &res->mnt_list);
1408 attach_mnt(q, parent, p->mnt_mp);
1409 br_write_unlock(&vfsmount_lock);
1412 return res;
1413 out:
1414 if (res) {
1415 br_write_lock(&vfsmount_lock);
1416 umount_tree(res, 0);
1417 br_write_unlock(&vfsmount_lock);
1419 return q;
1422 /* Caller should check returned pointer for errors */
1424 struct vfsmount *collect_mounts(struct path *path)
1426 struct mount *tree;
1427 namespace_lock();
1428 tree = copy_tree(real_mount(path->mnt), path->dentry,
1429 CL_COPY_ALL | CL_PRIVATE);
1430 namespace_unlock();
1431 if (IS_ERR(tree))
1432 return NULL;
1433 return &tree->mnt;
1436 void drop_collected_mounts(struct vfsmount *mnt)
1438 namespace_lock();
1439 br_write_lock(&vfsmount_lock);
1440 umount_tree(real_mount(mnt), 0);
1441 br_write_unlock(&vfsmount_lock);
1442 namespace_unlock();
1445 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1446 struct vfsmount *root)
1448 struct mount *mnt;
1449 int res = f(root, arg);
1450 if (res)
1451 return res;
1452 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1453 res = f(&mnt->mnt, arg);
1454 if (res)
1455 return res;
1457 return 0;
1460 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1462 struct mount *p;
1464 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1465 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1466 mnt_release_group_id(p);
1470 static int invent_group_ids(struct mount *mnt, bool recurse)
1472 struct mount *p;
1474 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1475 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1476 int err = mnt_alloc_group_id(p);
1477 if (err) {
1478 cleanup_group_ids(mnt, p);
1479 return err;
1484 return 0;
1488 * @source_mnt : mount tree to be attached
1489 * @nd : place the mount tree @source_mnt is attached
1490 * @parent_nd : if non-null, detach the source_mnt from its parent and
1491 * store the parent mount and mountpoint dentry.
1492 * (done when source_mnt is moved)
1494 * NOTE: in the table below explains the semantics when a source mount
1495 * of a given type is attached to a destination mount of a given type.
1496 * ---------------------------------------------------------------------------
1497 * | BIND MOUNT OPERATION |
1498 * |**************************************************************************
1499 * | source-->| shared | private | slave | unbindable |
1500 * | dest | | | | |
1501 * | | | | | | |
1502 * | v | | | | |
1503 * |**************************************************************************
1504 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1505 * | | | | | |
1506 * |non-shared| shared (+) | private | slave (*) | invalid |
1507 * ***************************************************************************
1508 * A bind operation clones the source mount and mounts the clone on the
1509 * destination mount.
1511 * (++) the cloned mount is propagated to all the mounts in the propagation
1512 * tree of the destination mount and the cloned mount is added to
1513 * the peer group of the source mount.
1514 * (+) the cloned mount is created under the destination mount and is marked
1515 * as shared. The cloned mount is added to the peer group of the source
1516 * mount.
1517 * (+++) the mount is propagated to all the mounts in the propagation tree
1518 * of the destination mount and the cloned mount is made slave
1519 * of the same master as that of the source mount. The cloned mount
1520 * is marked as 'shared and slave'.
1521 * (*) the cloned mount is made a slave of the same master as that of the
1522 * source mount.
1524 * ---------------------------------------------------------------------------
1525 * | MOVE MOUNT OPERATION |
1526 * |**************************************************************************
1527 * | source-->| shared | private | slave | unbindable |
1528 * | dest | | | | |
1529 * | | | | | | |
1530 * | v | | | | |
1531 * |**************************************************************************
1532 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1533 * | | | | | |
1534 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1535 * ***************************************************************************
1537 * (+) the mount is moved to the destination. And is then propagated to
1538 * all the mounts in the propagation tree of the destination mount.
1539 * (+*) the mount is moved to the destination.
1540 * (+++) the mount is moved to the destination and is then propagated to
1541 * all the mounts belonging to the destination mount's propagation tree.
1542 * the mount is marked as 'shared and slave'.
1543 * (*) the mount continues to be a slave at the new location.
1545 * if the source mount is a tree, the operations explained above is
1546 * applied to each mount in the tree.
1547 * Must be called without spinlocks held, since this function can sleep
1548 * in allocations.
1550 static int attach_recursive_mnt(struct mount *source_mnt,
1551 struct mount *dest_mnt,
1552 struct mountpoint *dest_mp,
1553 struct path *parent_path)
1555 LIST_HEAD(tree_list);
1556 struct mount *child, *p;
1557 int err;
1559 if (IS_MNT_SHARED(dest_mnt)) {
1560 err = invent_group_ids(source_mnt, true);
1561 if (err)
1562 goto out;
1564 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1565 if (err)
1566 goto out_cleanup_ids;
1568 br_write_lock(&vfsmount_lock);
1570 if (IS_MNT_SHARED(dest_mnt)) {
1571 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1572 set_mnt_shared(p);
1574 if (parent_path) {
1575 detach_mnt(source_mnt, parent_path);
1576 attach_mnt(source_mnt, dest_mnt, dest_mp);
1577 touch_mnt_namespace(source_mnt->mnt_ns);
1578 } else {
1579 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1580 commit_tree(source_mnt);
1583 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1584 list_del_init(&child->mnt_hash);
1585 commit_tree(child);
1587 br_write_unlock(&vfsmount_lock);
1589 return 0;
1591 out_cleanup_ids:
1592 if (IS_MNT_SHARED(dest_mnt))
1593 cleanup_group_ids(source_mnt, NULL);
1594 out:
1595 return err;
1598 static struct mountpoint *lock_mount(struct path *path)
1600 struct vfsmount *mnt;
1601 struct dentry *dentry = path->dentry;
1602 retry:
1603 mutex_lock(&dentry->d_inode->i_mutex);
1604 if (unlikely(cant_mount(dentry))) {
1605 mutex_unlock(&dentry->d_inode->i_mutex);
1606 return ERR_PTR(-ENOENT);
1608 namespace_lock();
1609 mnt = lookup_mnt(path);
1610 if (likely(!mnt)) {
1611 struct mountpoint *mp = new_mountpoint(dentry);
1612 if (IS_ERR(mp)) {
1613 namespace_unlock();
1614 mutex_unlock(&dentry->d_inode->i_mutex);
1615 return mp;
1617 return mp;
1619 namespace_unlock();
1620 mutex_unlock(&path->dentry->d_inode->i_mutex);
1621 path_put(path);
1622 path->mnt = mnt;
1623 dentry = path->dentry = dget(mnt->mnt_root);
1624 goto retry;
1627 static void unlock_mount(struct mountpoint *where)
1629 struct dentry *dentry = where->m_dentry;
1630 put_mountpoint(where);
1631 namespace_unlock();
1632 mutex_unlock(&dentry->d_inode->i_mutex);
1635 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1637 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1638 return -EINVAL;
1640 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
1641 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1642 return -ENOTDIR;
1644 return attach_recursive_mnt(mnt, p, mp, NULL);
1648 * Sanity check the flags to change_mnt_propagation.
1651 static int flags_to_propagation_type(int flags)
1653 int type = flags & ~(MS_REC | MS_SILENT);
1655 /* Fail if any non-propagation flags are set */
1656 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1657 return 0;
1658 /* Only one propagation flag should be set */
1659 if (!is_power_of_2(type))
1660 return 0;
1661 return type;
1665 * recursively change the type of the mountpoint.
1667 static int do_change_type(struct path *path, int flag)
1669 struct mount *m;
1670 struct mount *mnt = real_mount(path->mnt);
1671 int recurse = flag & MS_REC;
1672 int type;
1673 int err = 0;
1675 if (path->dentry != path->mnt->mnt_root)
1676 return -EINVAL;
1678 type = flags_to_propagation_type(flag);
1679 if (!type)
1680 return -EINVAL;
1682 namespace_lock();
1683 if (type == MS_SHARED) {
1684 err = invent_group_ids(mnt, recurse);
1685 if (err)
1686 goto out_unlock;
1689 br_write_lock(&vfsmount_lock);
1690 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1691 change_mnt_propagation(m, type);
1692 br_write_unlock(&vfsmount_lock);
1694 out_unlock:
1695 namespace_unlock();
1696 return err;
1700 * do loopback mount.
1702 static int do_loopback(struct path *path, const char *old_name,
1703 int recurse)
1705 struct path old_path;
1706 struct mount *mnt = NULL, *old, *parent;
1707 struct mountpoint *mp;
1708 int err;
1709 if (!old_name || !*old_name)
1710 return -EINVAL;
1711 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1712 if (err)
1713 return err;
1715 err = -EINVAL;
1716 if (mnt_ns_loop(&old_path))
1717 goto out;
1719 mp = lock_mount(path);
1720 err = PTR_ERR(mp);
1721 if (IS_ERR(mp))
1722 goto out;
1724 old = real_mount(old_path.mnt);
1725 parent = real_mount(path->mnt);
1727 err = -EINVAL;
1728 if (IS_MNT_UNBINDABLE(old))
1729 goto out2;
1731 if (!check_mnt(parent) || !check_mnt(old))
1732 goto out2;
1734 if (recurse)
1735 mnt = copy_tree(old, old_path.dentry, 0);
1736 else
1737 mnt = clone_mnt(old, old_path.dentry, 0);
1739 if (IS_ERR(mnt)) {
1740 err = PTR_ERR(mnt);
1741 goto out2;
1744 err = graft_tree(mnt, parent, mp);
1745 if (err) {
1746 br_write_lock(&vfsmount_lock);
1747 umount_tree(mnt, 0);
1748 br_write_unlock(&vfsmount_lock);
1750 out2:
1751 unlock_mount(mp);
1752 out:
1753 path_put(&old_path);
1754 return err;
1757 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1759 int error = 0;
1760 int readonly_request = 0;
1762 if (ms_flags & MS_RDONLY)
1763 readonly_request = 1;
1764 if (readonly_request == __mnt_is_readonly(mnt))
1765 return 0;
1767 if (mnt->mnt_flags & MNT_LOCK_READONLY)
1768 return -EPERM;
1770 if (readonly_request)
1771 error = mnt_make_readonly(real_mount(mnt));
1772 else
1773 __mnt_unmake_readonly(real_mount(mnt));
1774 return error;
1778 * change filesystem flags. dir should be a physical root of filesystem.
1779 * If you've mounted a non-root directory somewhere and want to do remount
1780 * on it - tough luck.
1782 static int do_remount(struct path *path, int flags, int mnt_flags,
1783 void *data)
1785 int err;
1786 struct super_block *sb = path->mnt->mnt_sb;
1787 struct mount *mnt = real_mount(path->mnt);
1789 if (!check_mnt(mnt))
1790 return -EINVAL;
1792 if (path->dentry != path->mnt->mnt_root)
1793 return -EINVAL;
1795 err = security_sb_remount(sb, data);
1796 if (err)
1797 return err;
1799 down_write(&sb->s_umount);
1800 if (flags & MS_BIND)
1801 err = change_mount_flags(path->mnt, flags);
1802 else if (!capable(CAP_SYS_ADMIN))
1803 err = -EPERM;
1804 else
1805 err = do_remount_sb(sb, flags, data, 0);
1806 if (!err) {
1807 br_write_lock(&vfsmount_lock);
1808 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1809 mnt->mnt.mnt_flags = mnt_flags;
1810 br_write_unlock(&vfsmount_lock);
1812 up_write(&sb->s_umount);
1813 if (!err) {
1814 br_write_lock(&vfsmount_lock);
1815 touch_mnt_namespace(mnt->mnt_ns);
1816 br_write_unlock(&vfsmount_lock);
1818 return err;
1821 static inline int tree_contains_unbindable(struct mount *mnt)
1823 struct mount *p;
1824 for (p = mnt; p; p = next_mnt(p, mnt)) {
1825 if (IS_MNT_UNBINDABLE(p))
1826 return 1;
1828 return 0;
1831 static int do_move_mount(struct path *path, const char *old_name)
1833 struct path old_path, parent_path;
1834 struct mount *p;
1835 struct mount *old;
1836 struct mountpoint *mp;
1837 int err;
1838 if (!old_name || !*old_name)
1839 return -EINVAL;
1840 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1841 if (err)
1842 return err;
1844 mp = lock_mount(path);
1845 err = PTR_ERR(mp);
1846 if (IS_ERR(mp))
1847 goto out;
1849 old = real_mount(old_path.mnt);
1850 p = real_mount(path->mnt);
1852 err = -EINVAL;
1853 if (!check_mnt(p) || !check_mnt(old))
1854 goto out1;
1856 err = -EINVAL;
1857 if (old_path.dentry != old_path.mnt->mnt_root)
1858 goto out1;
1860 if (!mnt_has_parent(old))
1861 goto out1;
1863 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1864 S_ISDIR(old_path.dentry->d_inode->i_mode))
1865 goto out1;
1867 * Don't move a mount residing in a shared parent.
1869 if (IS_MNT_SHARED(old->mnt_parent))
1870 goto out1;
1872 * Don't move a mount tree containing unbindable mounts to a destination
1873 * mount which is shared.
1875 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1876 goto out1;
1877 err = -ELOOP;
1878 for (; mnt_has_parent(p); p = p->mnt_parent)
1879 if (p == old)
1880 goto out1;
1882 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
1883 if (err)
1884 goto out1;
1886 /* if the mount is moved, it should no longer be expire
1887 * automatically */
1888 list_del_init(&old->mnt_expire);
1889 out1:
1890 unlock_mount(mp);
1891 out:
1892 if (!err)
1893 path_put(&parent_path);
1894 path_put(&old_path);
1895 return err;
1898 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1900 int err;
1901 const char *subtype = strchr(fstype, '.');
1902 if (subtype) {
1903 subtype++;
1904 err = -EINVAL;
1905 if (!subtype[0])
1906 goto err;
1907 } else
1908 subtype = "";
1910 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1911 err = -ENOMEM;
1912 if (!mnt->mnt_sb->s_subtype)
1913 goto err;
1914 return mnt;
1916 err:
1917 mntput(mnt);
1918 return ERR_PTR(err);
1922 * add a mount into a namespace's mount tree
1924 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1926 struct mountpoint *mp;
1927 struct mount *parent;
1928 int err;
1930 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1932 mp = lock_mount(path);
1933 if (IS_ERR(mp))
1934 return PTR_ERR(mp);
1936 parent = real_mount(path->mnt);
1937 err = -EINVAL;
1938 if (unlikely(!check_mnt(parent))) {
1939 /* that's acceptable only for automounts done in private ns */
1940 if (!(mnt_flags & MNT_SHRINKABLE))
1941 goto unlock;
1942 /* ... and for those we'd better have mountpoint still alive */
1943 if (!parent->mnt_ns)
1944 goto unlock;
1947 /* Refuse the same filesystem on the same mount point */
1948 err = -EBUSY;
1949 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1950 path->mnt->mnt_root == path->dentry)
1951 goto unlock;
1953 err = -EINVAL;
1954 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1955 goto unlock;
1957 newmnt->mnt.mnt_flags = mnt_flags;
1958 err = graft_tree(newmnt, parent, mp);
1960 unlock:
1961 unlock_mount(mp);
1962 return err;
1966 * create a new mount for userspace and request it to be added into the
1967 * namespace's tree
1969 static int do_new_mount(struct path *path, const char *fstype, int flags,
1970 int mnt_flags, const char *name, void *data)
1972 struct file_system_type *type;
1973 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
1974 struct vfsmount *mnt;
1975 int err;
1977 if (!fstype)
1978 return -EINVAL;
1980 type = get_fs_type(fstype);
1981 if (!type)
1982 return -ENODEV;
1984 if (user_ns != &init_user_ns) {
1985 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
1986 put_filesystem(type);
1987 return -EPERM;
1989 /* Only in special cases allow devices from mounts
1990 * created outside the initial user namespace.
1992 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
1993 flags |= MS_NODEV;
1994 mnt_flags |= MNT_NODEV;
1998 mnt = vfs_kern_mount(type, flags, name, data);
1999 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2000 !mnt->mnt_sb->s_subtype)
2001 mnt = fs_set_subtype(mnt, fstype);
2003 put_filesystem(type);
2004 if (IS_ERR(mnt))
2005 return PTR_ERR(mnt);
2007 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2008 if (err)
2009 mntput(mnt);
2010 return err;
2013 int finish_automount(struct vfsmount *m, struct path *path)
2015 struct mount *mnt = real_mount(m);
2016 int err;
2017 /* The new mount record should have at least 2 refs to prevent it being
2018 * expired before we get a chance to add it
2020 BUG_ON(mnt_get_count(mnt) < 2);
2022 if (m->mnt_sb == path->mnt->mnt_sb &&
2023 m->mnt_root == path->dentry) {
2024 err = -ELOOP;
2025 goto fail;
2028 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2029 if (!err)
2030 return 0;
2031 fail:
2032 /* remove m from any expiration list it may be on */
2033 if (!list_empty(&mnt->mnt_expire)) {
2034 namespace_lock();
2035 br_write_lock(&vfsmount_lock);
2036 list_del_init(&mnt->mnt_expire);
2037 br_write_unlock(&vfsmount_lock);
2038 namespace_unlock();
2040 mntput(m);
2041 mntput(m);
2042 return err;
2046 * mnt_set_expiry - Put a mount on an expiration list
2047 * @mnt: The mount to list.
2048 * @expiry_list: The list to add the mount to.
2050 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2052 namespace_lock();
2053 br_write_lock(&vfsmount_lock);
2055 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2057 br_write_unlock(&vfsmount_lock);
2058 namespace_unlock();
2060 EXPORT_SYMBOL(mnt_set_expiry);
2063 * process a list of expirable mountpoints with the intent of discarding any
2064 * mountpoints that aren't in use and haven't been touched since last we came
2065 * here
2067 void mark_mounts_for_expiry(struct list_head *mounts)
2069 struct mount *mnt, *next;
2070 LIST_HEAD(graveyard);
2072 if (list_empty(mounts))
2073 return;
2075 namespace_lock();
2076 br_write_lock(&vfsmount_lock);
2078 /* extract from the expiration list every vfsmount that matches the
2079 * following criteria:
2080 * - only referenced by its parent vfsmount
2081 * - still marked for expiry (marked on the last call here; marks are
2082 * cleared by mntput())
2084 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2085 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2086 propagate_mount_busy(mnt, 1))
2087 continue;
2088 list_move(&mnt->mnt_expire, &graveyard);
2090 while (!list_empty(&graveyard)) {
2091 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2092 touch_mnt_namespace(mnt->mnt_ns);
2093 umount_tree(mnt, 1);
2095 br_write_unlock(&vfsmount_lock);
2096 namespace_unlock();
2099 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2102 * Ripoff of 'select_parent()'
2104 * search the list of submounts for a given mountpoint, and move any
2105 * shrinkable submounts to the 'graveyard' list.
2107 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2109 struct mount *this_parent = parent;
2110 struct list_head *next;
2111 int found = 0;
2113 repeat:
2114 next = this_parent->mnt_mounts.next;
2115 resume:
2116 while (next != &this_parent->mnt_mounts) {
2117 struct list_head *tmp = next;
2118 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2120 next = tmp->next;
2121 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2122 continue;
2124 * Descend a level if the d_mounts list is non-empty.
2126 if (!list_empty(&mnt->mnt_mounts)) {
2127 this_parent = mnt;
2128 goto repeat;
2131 if (!propagate_mount_busy(mnt, 1)) {
2132 list_move_tail(&mnt->mnt_expire, graveyard);
2133 found++;
2137 * All done at this level ... ascend and resume the search
2139 if (this_parent != parent) {
2140 next = this_parent->mnt_child.next;
2141 this_parent = this_parent->mnt_parent;
2142 goto resume;
2144 return found;
2148 * process a list of expirable mountpoints with the intent of discarding any
2149 * submounts of a specific parent mountpoint
2151 * vfsmount_lock must be held for write
2153 static void shrink_submounts(struct mount *mnt)
2155 LIST_HEAD(graveyard);
2156 struct mount *m;
2158 /* extract submounts of 'mountpoint' from the expiration list */
2159 while (select_submounts(mnt, &graveyard)) {
2160 while (!list_empty(&graveyard)) {
2161 m = list_first_entry(&graveyard, struct mount,
2162 mnt_expire);
2163 touch_mnt_namespace(m->mnt_ns);
2164 umount_tree(m, 1);
2170 * Some copy_from_user() implementations do not return the exact number of
2171 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2172 * Note that this function differs from copy_from_user() in that it will oops
2173 * on bad values of `to', rather than returning a short copy.
2175 static long exact_copy_from_user(void *to, const void __user * from,
2176 unsigned long n)
2178 char *t = to;
2179 const char __user *f = from;
2180 char c;
2182 if (!access_ok(VERIFY_READ, from, n))
2183 return n;
2185 while (n) {
2186 if (__get_user(c, f)) {
2187 memset(t, 0, n);
2188 break;
2190 *t++ = c;
2191 f++;
2192 n--;
2194 return n;
2197 int copy_mount_options(const void __user * data, unsigned long *where)
2199 int i;
2200 unsigned long page;
2201 unsigned long size;
2203 *where = 0;
2204 if (!data)
2205 return 0;
2207 if (!(page = __get_free_page(GFP_KERNEL)))
2208 return -ENOMEM;
2210 /* We only care that *some* data at the address the user
2211 * gave us is valid. Just in case, we'll zero
2212 * the remainder of the page.
2214 /* copy_from_user cannot cross TASK_SIZE ! */
2215 size = TASK_SIZE - (unsigned long)data;
2216 if (size > PAGE_SIZE)
2217 size = PAGE_SIZE;
2219 i = size - exact_copy_from_user((void *)page, data, size);
2220 if (!i) {
2221 free_page(page);
2222 return -EFAULT;
2224 if (i != PAGE_SIZE)
2225 memset((char *)page + i, 0, PAGE_SIZE - i);
2226 *where = page;
2227 return 0;
2230 int copy_mount_string(const void __user *data, char **where)
2232 char *tmp;
2234 if (!data) {
2235 *where = NULL;
2236 return 0;
2239 tmp = strndup_user(data, PAGE_SIZE);
2240 if (IS_ERR(tmp))
2241 return PTR_ERR(tmp);
2243 *where = tmp;
2244 return 0;
2248 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2249 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2251 * data is a (void *) that can point to any structure up to
2252 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2253 * information (or be NULL).
2255 * Pre-0.97 versions of mount() didn't have a flags word.
2256 * When the flags word was introduced its top half was required
2257 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2258 * Therefore, if this magic number is present, it carries no information
2259 * and must be discarded.
2261 long do_mount(const char *dev_name, const char *dir_name,
2262 const char *type_page, unsigned long flags, void *data_page)
2264 struct path path;
2265 int retval = 0;
2266 int mnt_flags = 0;
2268 /* Discard magic */
2269 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2270 flags &= ~MS_MGC_MSK;
2272 /* Basic sanity checks */
2274 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2275 return -EINVAL;
2277 if (data_page)
2278 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2280 /* ... and get the mountpoint */
2281 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2282 if (retval)
2283 return retval;
2285 retval = security_sb_mount(dev_name, &path,
2286 type_page, flags, data_page);
2287 if (!retval && !may_mount())
2288 retval = -EPERM;
2289 if (retval)
2290 goto dput_out;
2292 /* Default to relatime unless overriden */
2293 if (!(flags & MS_NOATIME))
2294 mnt_flags |= MNT_RELATIME;
2296 /* Separate the per-mountpoint flags */
2297 if (flags & MS_NOSUID)
2298 mnt_flags |= MNT_NOSUID;
2299 if (flags & MS_NODEV)
2300 mnt_flags |= MNT_NODEV;
2301 if (flags & MS_NOEXEC)
2302 mnt_flags |= MNT_NOEXEC;
2303 if (flags & MS_NOATIME)
2304 mnt_flags |= MNT_NOATIME;
2305 if (flags & MS_NODIRATIME)
2306 mnt_flags |= MNT_NODIRATIME;
2307 if (flags & MS_STRICTATIME)
2308 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2309 if (flags & MS_RDONLY)
2310 mnt_flags |= MNT_READONLY;
2312 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2313 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2314 MS_STRICTATIME);
2316 if (flags & MS_REMOUNT)
2317 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2318 data_page);
2319 else if (flags & MS_BIND)
2320 retval = do_loopback(&path, dev_name, flags & MS_REC);
2321 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2322 retval = do_change_type(&path, flags);
2323 else if (flags & MS_MOVE)
2324 retval = do_move_mount(&path, dev_name);
2325 else
2326 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2327 dev_name, data_page);
2328 dput_out:
2329 path_put(&path);
2330 return retval;
2333 static void free_mnt_ns(struct mnt_namespace *ns)
2335 proc_free_inum(ns->proc_inum);
2336 put_user_ns(ns->user_ns);
2337 kfree(ns);
2341 * Assign a sequence number so we can detect when we attempt to bind
2342 * mount a reference to an older mount namespace into the current
2343 * mount namespace, preventing reference counting loops. A 64bit
2344 * number incrementing at 10Ghz will take 12,427 years to wrap which
2345 * is effectively never, so we can ignore the possibility.
2347 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2349 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2351 struct mnt_namespace *new_ns;
2352 int ret;
2354 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2355 if (!new_ns)
2356 return ERR_PTR(-ENOMEM);
2357 ret = proc_alloc_inum(&new_ns->proc_inum);
2358 if (ret) {
2359 kfree(new_ns);
2360 return ERR_PTR(ret);
2362 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2363 atomic_set(&new_ns->count, 1);
2364 new_ns->root = NULL;
2365 INIT_LIST_HEAD(&new_ns->list);
2366 init_waitqueue_head(&new_ns->poll);
2367 new_ns->event = 0;
2368 new_ns->user_ns = get_user_ns(user_ns);
2369 return new_ns;
2373 * Allocate a new namespace structure and populate it with contents
2374 * copied from the namespace of the passed in task structure.
2376 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2377 struct user_namespace *user_ns, struct fs_struct *fs)
2379 struct mnt_namespace *new_ns;
2380 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2381 struct mount *p, *q;
2382 struct mount *old = mnt_ns->root;
2383 struct mount *new;
2384 int copy_flags;
2386 new_ns = alloc_mnt_ns(user_ns);
2387 if (IS_ERR(new_ns))
2388 return new_ns;
2390 namespace_lock();
2391 /* First pass: copy the tree topology */
2392 copy_flags = CL_COPY_ALL | CL_EXPIRE;
2393 if (user_ns != mnt_ns->user_ns)
2394 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2395 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2396 if (IS_ERR(new)) {
2397 namespace_unlock();
2398 free_mnt_ns(new_ns);
2399 return ERR_CAST(new);
2401 new_ns->root = new;
2402 br_write_lock(&vfsmount_lock);
2403 list_add_tail(&new_ns->list, &new->mnt_list);
2404 br_write_unlock(&vfsmount_lock);
2407 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2408 * as belonging to new namespace. We have already acquired a private
2409 * fs_struct, so tsk->fs->lock is not needed.
2411 p = old;
2412 q = new;
2413 while (p) {
2414 q->mnt_ns = new_ns;
2415 if (fs) {
2416 if (&p->mnt == fs->root.mnt) {
2417 fs->root.mnt = mntget(&q->mnt);
2418 rootmnt = &p->mnt;
2420 if (&p->mnt == fs->pwd.mnt) {
2421 fs->pwd.mnt = mntget(&q->mnt);
2422 pwdmnt = &p->mnt;
2425 p = next_mnt(p, old);
2426 q = next_mnt(q, new);
2428 namespace_unlock();
2430 if (rootmnt)
2431 mntput(rootmnt);
2432 if (pwdmnt)
2433 mntput(pwdmnt);
2435 return new_ns;
2438 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2439 struct user_namespace *user_ns, struct fs_struct *new_fs)
2441 struct mnt_namespace *new_ns;
2443 BUG_ON(!ns);
2444 get_mnt_ns(ns);
2446 if (!(flags & CLONE_NEWNS))
2447 return ns;
2449 new_ns = dup_mnt_ns(ns, user_ns, new_fs);
2451 put_mnt_ns(ns);
2452 return new_ns;
2456 * create_mnt_ns - creates a private namespace and adds a root filesystem
2457 * @mnt: pointer to the new root filesystem mountpoint
2459 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2461 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2462 if (!IS_ERR(new_ns)) {
2463 struct mount *mnt = real_mount(m);
2464 mnt->mnt_ns = new_ns;
2465 new_ns->root = mnt;
2466 list_add(&mnt->mnt_list, &new_ns->list);
2467 } else {
2468 mntput(m);
2470 return new_ns;
2473 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2475 struct mnt_namespace *ns;
2476 struct super_block *s;
2477 struct path path;
2478 int err;
2480 ns = create_mnt_ns(mnt);
2481 if (IS_ERR(ns))
2482 return ERR_CAST(ns);
2484 err = vfs_path_lookup(mnt->mnt_root, mnt,
2485 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2487 put_mnt_ns(ns);
2489 if (err)
2490 return ERR_PTR(err);
2492 /* trade a vfsmount reference for active sb one */
2493 s = path.mnt->mnt_sb;
2494 atomic_inc(&s->s_active);
2495 mntput(path.mnt);
2496 /* lock the sucker */
2497 down_write(&s->s_umount);
2498 /* ... and return the root of (sub)tree on it */
2499 return path.dentry;
2501 EXPORT_SYMBOL(mount_subtree);
2503 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2504 char __user *, type, unsigned long, flags, void __user *, data)
2506 int ret;
2507 char *kernel_type;
2508 struct filename *kernel_dir;
2509 char *kernel_dev;
2510 unsigned long data_page;
2512 ret = copy_mount_string(type, &kernel_type);
2513 if (ret < 0)
2514 goto out_type;
2516 kernel_dir = getname(dir_name);
2517 if (IS_ERR(kernel_dir)) {
2518 ret = PTR_ERR(kernel_dir);
2519 goto out_dir;
2522 ret = copy_mount_string(dev_name, &kernel_dev);
2523 if (ret < 0)
2524 goto out_dev;
2526 ret = copy_mount_options(data, &data_page);
2527 if (ret < 0)
2528 goto out_data;
2530 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2531 (void *) data_page);
2533 free_page(data_page);
2534 out_data:
2535 kfree(kernel_dev);
2536 out_dev:
2537 putname(kernel_dir);
2538 out_dir:
2539 kfree(kernel_type);
2540 out_type:
2541 return ret;
2545 * Return true if path is reachable from root
2547 * namespace_sem or vfsmount_lock is held
2549 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2550 const struct path *root)
2552 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2553 dentry = mnt->mnt_mountpoint;
2554 mnt = mnt->mnt_parent;
2556 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2559 int path_is_under(struct path *path1, struct path *path2)
2561 int res;
2562 br_read_lock(&vfsmount_lock);
2563 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2564 br_read_unlock(&vfsmount_lock);
2565 return res;
2567 EXPORT_SYMBOL(path_is_under);
2570 * pivot_root Semantics:
2571 * Moves the root file system of the current process to the directory put_old,
2572 * makes new_root as the new root file system of the current process, and sets
2573 * root/cwd of all processes which had them on the current root to new_root.
2575 * Restrictions:
2576 * The new_root and put_old must be directories, and must not be on the
2577 * same file system as the current process root. The put_old must be
2578 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2579 * pointed to by put_old must yield the same directory as new_root. No other
2580 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2582 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2583 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2584 * in this situation.
2586 * Notes:
2587 * - we don't move root/cwd if they are not at the root (reason: if something
2588 * cared enough to change them, it's probably wrong to force them elsewhere)
2589 * - it's okay to pick a root that isn't the root of a file system, e.g.
2590 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2591 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2592 * first.
2594 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2595 const char __user *, put_old)
2597 struct path new, old, parent_path, root_parent, root;
2598 struct mount *new_mnt, *root_mnt, *old_mnt;
2599 struct mountpoint *old_mp, *root_mp;
2600 int error;
2602 if (!may_mount())
2603 return -EPERM;
2605 error = user_path_dir(new_root, &new);
2606 if (error)
2607 goto out0;
2609 error = user_path_dir(put_old, &old);
2610 if (error)
2611 goto out1;
2613 error = security_sb_pivotroot(&old, &new);
2614 if (error)
2615 goto out2;
2617 get_fs_root(current->fs, &root);
2618 old_mp = lock_mount(&old);
2619 error = PTR_ERR(old_mp);
2620 if (IS_ERR(old_mp))
2621 goto out3;
2623 error = -EINVAL;
2624 new_mnt = real_mount(new.mnt);
2625 root_mnt = real_mount(root.mnt);
2626 old_mnt = real_mount(old.mnt);
2627 if (IS_MNT_SHARED(old_mnt) ||
2628 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2629 IS_MNT_SHARED(root_mnt->mnt_parent))
2630 goto out4;
2631 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2632 goto out4;
2633 error = -ENOENT;
2634 if (d_unlinked(new.dentry))
2635 goto out4;
2636 error = -EBUSY;
2637 if (new_mnt == root_mnt || old_mnt == root_mnt)
2638 goto out4; /* loop, on the same file system */
2639 error = -EINVAL;
2640 if (root.mnt->mnt_root != root.dentry)
2641 goto out4; /* not a mountpoint */
2642 if (!mnt_has_parent(root_mnt))
2643 goto out4; /* not attached */
2644 root_mp = root_mnt->mnt_mp;
2645 if (new.mnt->mnt_root != new.dentry)
2646 goto out4; /* not a mountpoint */
2647 if (!mnt_has_parent(new_mnt))
2648 goto out4; /* not attached */
2649 /* make sure we can reach put_old from new_root */
2650 if (!is_path_reachable(old_mnt, old.dentry, &new))
2651 goto out4;
2652 root_mp->m_count++; /* pin it so it won't go away */
2653 br_write_lock(&vfsmount_lock);
2654 detach_mnt(new_mnt, &parent_path);
2655 detach_mnt(root_mnt, &root_parent);
2656 /* mount old root on put_old */
2657 attach_mnt(root_mnt, old_mnt, old_mp);
2658 /* mount new_root on / */
2659 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2660 touch_mnt_namespace(current->nsproxy->mnt_ns);
2661 br_write_unlock(&vfsmount_lock);
2662 chroot_fs_refs(&root, &new);
2663 put_mountpoint(root_mp);
2664 error = 0;
2665 out4:
2666 unlock_mount(old_mp);
2667 if (!error) {
2668 path_put(&root_parent);
2669 path_put(&parent_path);
2671 out3:
2672 path_put(&root);
2673 out2:
2674 path_put(&old);
2675 out1:
2676 path_put(&new);
2677 out0:
2678 return error;
2681 static void __init init_mount_tree(void)
2683 struct vfsmount *mnt;
2684 struct mnt_namespace *ns;
2685 struct path root;
2686 struct file_system_type *type;
2688 type = get_fs_type("rootfs");
2689 if (!type)
2690 panic("Can't find rootfs type");
2691 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2692 put_filesystem(type);
2693 if (IS_ERR(mnt))
2694 panic("Can't create rootfs");
2696 ns = create_mnt_ns(mnt);
2697 if (IS_ERR(ns))
2698 panic("Can't allocate initial namespace");
2700 init_task.nsproxy->mnt_ns = ns;
2701 get_mnt_ns(ns);
2703 root.mnt = mnt;
2704 root.dentry = mnt->mnt_root;
2706 set_fs_pwd(current->fs, &root);
2707 set_fs_root(current->fs, &root);
2710 void __init mnt_init(void)
2712 unsigned u;
2713 int err;
2715 init_rwsem(&namespace_sem);
2717 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2718 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2720 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2721 mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2723 if (!mount_hashtable || !mountpoint_hashtable)
2724 panic("Failed to allocate mount hash table\n");
2726 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2728 for (u = 0; u < HASH_SIZE; u++)
2729 INIT_LIST_HEAD(&mount_hashtable[u]);
2730 for (u = 0; u < HASH_SIZE; u++)
2731 INIT_LIST_HEAD(&mountpoint_hashtable[u]);
2733 br_lock_init(&vfsmount_lock);
2735 err = sysfs_init();
2736 if (err)
2737 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2738 __func__, err);
2739 fs_kobj = kobject_create_and_add("fs", NULL);
2740 if (!fs_kobj)
2741 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2742 init_rootfs();
2743 init_mount_tree();
2746 void put_mnt_ns(struct mnt_namespace *ns)
2748 if (!atomic_dec_and_test(&ns->count))
2749 return;
2750 namespace_lock();
2751 br_write_lock(&vfsmount_lock);
2752 umount_tree(ns->root, 0);
2753 br_write_unlock(&vfsmount_lock);
2754 namespace_unlock();
2755 free_mnt_ns(ns);
2758 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2760 struct vfsmount *mnt;
2761 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2762 if (!IS_ERR(mnt)) {
2764 * it is a longterm mount, don't release mnt until
2765 * we unmount before file sys is unregistered
2767 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2769 return mnt;
2771 EXPORT_SYMBOL_GPL(kern_mount_data);
2773 void kern_unmount(struct vfsmount *mnt)
2775 /* release long term mount so mount point can be released */
2776 if (!IS_ERR_OR_NULL(mnt)) {
2777 br_write_lock(&vfsmount_lock);
2778 real_mount(mnt)->mnt_ns = NULL;
2779 br_write_unlock(&vfsmount_lock);
2780 mntput(mnt);
2783 EXPORT_SYMBOL(kern_unmount);
2785 bool our_mnt(struct vfsmount *mnt)
2787 return check_mnt(real_mount(mnt));
2790 bool current_chrooted(void)
2792 /* Does the current process have a non-standard root */
2793 struct path ns_root;
2794 struct path fs_root;
2795 bool chrooted;
2797 /* Find the namespace root */
2798 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
2799 ns_root.dentry = ns_root.mnt->mnt_root;
2800 path_get(&ns_root);
2801 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2804 get_fs_root(current->fs, &fs_root);
2806 chrooted = !path_equal(&fs_root, &ns_root);
2808 path_put(&fs_root);
2809 path_put(&ns_root);
2811 return chrooted;
2814 void update_mnt_policy(struct user_namespace *userns)
2816 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2817 struct mount *mnt;
2819 down_read(&namespace_sem);
2820 list_for_each_entry(mnt, &ns->list, mnt_list) {
2821 switch (mnt->mnt.mnt_sb->s_magic) {
2822 case SYSFS_MAGIC:
2823 userns->may_mount_sysfs = true;
2824 break;
2825 case PROC_SUPER_MAGIC:
2826 userns->may_mount_proc = true;
2827 break;
2829 if (userns->may_mount_sysfs && userns->may_mount_proc)
2830 break;
2832 up_read(&namespace_sem);
2835 static void *mntns_get(struct task_struct *task)
2837 struct mnt_namespace *ns = NULL;
2838 struct nsproxy *nsproxy;
2840 rcu_read_lock();
2841 nsproxy = task_nsproxy(task);
2842 if (nsproxy) {
2843 ns = nsproxy->mnt_ns;
2844 get_mnt_ns(ns);
2846 rcu_read_unlock();
2848 return ns;
2851 static void mntns_put(void *ns)
2853 put_mnt_ns(ns);
2856 static int mntns_install(struct nsproxy *nsproxy, void *ns)
2858 struct fs_struct *fs = current->fs;
2859 struct mnt_namespace *mnt_ns = ns;
2860 struct path root;
2862 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
2863 !nsown_capable(CAP_SYS_CHROOT) ||
2864 !nsown_capable(CAP_SYS_ADMIN))
2865 return -EPERM;
2867 if (fs->users != 1)
2868 return -EINVAL;
2870 get_mnt_ns(mnt_ns);
2871 put_mnt_ns(nsproxy->mnt_ns);
2872 nsproxy->mnt_ns = mnt_ns;
2874 /* Find the root */
2875 root.mnt = &mnt_ns->root->mnt;
2876 root.dentry = mnt_ns->root->mnt.mnt_root;
2877 path_get(&root);
2878 while(d_mountpoint(root.dentry) && follow_down_one(&root))
2881 /* Update the pwd and root */
2882 set_fs_pwd(fs, &root);
2883 set_fs_root(fs, &root);
2885 path_put(&root);
2886 return 0;
2889 static unsigned int mntns_inum(void *ns)
2891 struct mnt_namespace *mnt_ns = ns;
2892 return mnt_ns->proc_inum;
2895 const struct proc_ns_operations mntns_operations = {
2896 .name = "mnt",
2897 .type = CLONE_NEWNS,
2898 .get = mntns_get,
2899 .put = mntns_put,
2900 .install = mntns_install,
2901 .inum = mntns_inum,