Merge branch 'driver-core-next' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / namespace.c
blob3ddfd9046c449199a8f75ffcaf54761dde37fa73
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/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"
39 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
40 #define HASH_SIZE (1UL << HASH_SHIFT)
42 static int event;
43 static DEFINE_IDA(mnt_id_ida);
44 static DEFINE_IDA(mnt_group_ida);
45 static DEFINE_SPINLOCK(mnt_id_lock);
46 static int mnt_id_start = 0;
47 static int mnt_group_start = 1;
49 static struct list_head *mount_hashtable __read_mostly;
50 static struct kmem_cache *mnt_cache __read_mostly;
51 static struct rw_semaphore namespace_sem;
53 /* /sys/fs */
54 struct kobject *fs_kobj;
55 EXPORT_SYMBOL_GPL(fs_kobj);
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.
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.
65 DEFINE_BRLOCK(vfsmount_lock);
67 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
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);
75 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
78 * allocation is serialized by namespace_sem, but we need the spinlock to
79 * serialize with freeing.
81 static int mnt_alloc_id(struct vfsmount *mnt)
83 int res;
85 retry:
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;
95 return res;
98 static void mnt_free_id(struct vfsmount *mnt)
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);
109 * Allocate a new peer group ID
111 * mnt_group_ida is protected by namespace_sem
113 static int mnt_alloc_group_id(struct vfsmount *mnt)
115 int res;
117 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
118 return -ENOMEM;
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;
126 return res;
130 * Release a peer group ID
132 void mnt_release_group_id(struct vfsmount *mnt)
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;
142 * vfsmount lock must be held for read
144 static inline void mnt_add_count(struct vfsmount *mnt, int n)
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
155 static inline void mnt_set_count(struct vfsmount *mnt, int n)
157 #ifdef CONFIG_SMP
158 this_cpu_write(mnt->mnt_pcp->mnt_count, n);
159 #else
160 mnt->mnt_count = n;
161 #endif
165 * vfsmount lock must be held for read
167 static inline void mnt_inc_count(struct vfsmount *mnt)
169 mnt_add_count(mnt, 1);
173 * vfsmount lock must be held for read
175 static inline void mnt_dec_count(struct vfsmount *mnt)
177 mnt_add_count(mnt, -1);
181 * vfsmount lock must be held for write
183 unsigned int mnt_get_count(struct vfsmount *mnt)
185 #ifdef CONFIG_SMP
186 unsigned int count = atomic_read(&mnt->mnt_longrefs);
187 int cpu;
189 for_each_possible_cpu(cpu) {
190 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
193 return count;
194 #else
195 return mnt->mnt_count;
196 #endif
199 struct vfsmount *alloc_vfsmnt(const char *name)
201 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
202 if (mnt) {
203 int err;
205 err = mnt_alloc_id(mnt);
206 if (err)
207 goto out_free_cache;
209 if (name) {
210 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
211 if (!mnt->mnt_devname)
212 goto out_free_id;
215 #ifdef CONFIG_SMP
216 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
217 if (!mnt->mnt_pcp)
218 goto out_free_devname;
220 atomic_set(&mnt->mnt_longrefs, 1);
221 #else
222 mnt->mnt_count = 1;
223 mnt->mnt_writers = 0;
224 #endif
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
238 return mnt;
240 #ifdef CONFIG_SMP
241 out_free_devname:
242 kfree(mnt->mnt_devname);
243 #endif
244 out_free_id:
245 mnt_free_id(mnt);
246 out_free_cache:
247 kmem_cache_free(mnt_cache, mnt);
248 return NULL;
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.
260 * __mnt_is_readonly: check whether a mount is read-only
261 * @mnt: the mount to check for its write status
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.
270 int __mnt_is_readonly(struct vfsmount *mnt)
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;
278 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
280 static inline void mnt_inc_writers(struct vfsmount *mnt)
282 #ifdef CONFIG_SMP
283 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
284 #else
285 mnt->mnt_writers++;
286 #endif
289 static inline void mnt_dec_writers(struct vfsmount *mnt)
291 #ifdef CONFIG_SMP
292 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
293 #else
294 mnt->mnt_writers--;
295 #endif
298 static unsigned int mnt_get_writers(struct vfsmount *mnt)
300 #ifdef CONFIG_SMP
301 unsigned int count = 0;
302 int cpu;
304 for_each_possible_cpu(cpu) {
305 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
308 return count;
309 #else
310 return mnt->mnt_writers;
311 #endif
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.
323 * mnt_want_write - get write access to a mount
324 * @mnt: the mount on which to take a write
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.
332 int mnt_want_write(struct vfsmount *mnt)
334 int ret = 0;
336 preempt_disable();
337 mnt_inc_writers(mnt);
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.
343 smp_mb();
344 while (mnt->mnt_flags & MNT_WRITE_HOLD)
345 cpu_relax();
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.
351 smp_rmb();
352 if (__mnt_is_readonly(mnt)) {
353 mnt_dec_writers(mnt);
354 ret = -EROFS;
355 goto out;
357 out:
358 preempt_enable();
359 return ret;
361 EXPORT_SYMBOL_GPL(mnt_want_write);
364 * mnt_clone_write - get write access to a mount
365 * @mnt: the mount on which to take a write
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.
372 * After finished, mnt_drop_write must be called as usual to
373 * drop the reference.
375 int mnt_clone_write(struct vfsmount *mnt)
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;
385 EXPORT_SYMBOL_GPL(mnt_clone_write);
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
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
394 int mnt_want_write_file(struct file *file)
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);
402 EXPORT_SYMBOL_GPL(mnt_want_write_file);
405 * mnt_drop_write - give up write access to a mount
406 * @mnt: the mount on which to give up write access
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.
412 void mnt_drop_write(struct vfsmount *mnt)
414 preempt_disable();
415 mnt_dec_writers(mnt);
416 preempt_enable();
418 EXPORT_SYMBOL_GPL(mnt_drop_write);
420 static int mnt_make_readonly(struct vfsmount *mnt)
422 int ret = 0;
424 br_write_lock(vfsmount_lock);
425 mnt->mnt_flags |= MNT_WRITE_HOLD;
427 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
428 * should be visible before we do.
430 smp_mb();
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).
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.
448 if (mnt_get_writers(mnt) > 0)
449 ret = -EBUSY;
450 else
451 mnt->mnt_flags |= MNT_READONLY;
453 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
454 * that become unheld will see MNT_READONLY.
456 smp_wmb();
457 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
458 br_write_unlock(vfsmount_lock);
459 return ret;
462 static void __mnt_unmake_readonly(struct vfsmount *mnt)
464 br_write_lock(vfsmount_lock);
465 mnt->mnt_flags &= ~MNT_READONLY;
466 br_write_unlock(vfsmount_lock);
469 void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
471 mnt->mnt_sb = sb;
472 mnt->mnt_root = dget(sb->s_root);
475 EXPORT_SYMBOL(simple_set_mnt);
477 void free_vfsmnt(struct vfsmount *mnt)
479 kfree(mnt->mnt_devname);
480 mnt_free_id(mnt);
481 #ifdef CONFIG_SMP
482 free_percpu(mnt->mnt_pcp);
483 #endif
484 kmem_cache_free(mnt_cache, mnt);
488 * find the first or last mount at @dentry on vfsmount @mnt depending on
489 * @dir. If @dir is set return the first mount else return the last mount.
490 * vfsmount_lock must be held for read or write.
492 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
493 int dir)
495 struct list_head *head = mount_hashtable + hash(mnt, dentry);
496 struct list_head *tmp = head;
497 struct vfsmount *p, *found = NULL;
499 for (;;) {
500 tmp = dir ? tmp->next : tmp->prev;
501 p = NULL;
502 if (tmp == head)
503 break;
504 p = list_entry(tmp, struct vfsmount, mnt_hash);
505 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
506 found = p;
507 break;
510 return found;
514 * lookup_mnt increments the ref count before returning
515 * the vfsmount struct.
517 struct vfsmount *lookup_mnt(struct path *path)
519 struct vfsmount *child_mnt;
521 br_read_lock(vfsmount_lock);
522 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
523 mntget(child_mnt);
524 br_read_unlock(vfsmount_lock);
525 return child_mnt;
528 static inline int check_mnt(struct vfsmount *mnt)
530 return mnt->mnt_ns == current->nsproxy->mnt_ns;
534 * vfsmount lock must be held for write
536 static void touch_mnt_namespace(struct mnt_namespace *ns)
538 if (ns) {
539 ns->event = ++event;
540 wake_up_interruptible(&ns->poll);
545 * vfsmount lock must be held for write
547 static void __touch_mnt_namespace(struct mnt_namespace *ns)
549 if (ns && ns->event != event) {
550 ns->event = event;
551 wake_up_interruptible(&ns->poll);
556 * Clear dentry's mounted state if it has no remaining mounts.
557 * vfsmount_lock must be held for write.
559 static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
561 unsigned u;
563 for (u = 0; u < HASH_SIZE; u++) {
564 struct vfsmount *p;
566 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
567 if (p->mnt_mountpoint == dentry)
568 return;
571 spin_lock(&dentry->d_lock);
572 dentry->d_flags &= ~DCACHE_MOUNTED;
573 spin_unlock(&dentry->d_lock);
577 * vfsmount lock must be held for write
579 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
581 old_path->dentry = mnt->mnt_mountpoint;
582 old_path->mnt = mnt->mnt_parent;
583 mnt->mnt_parent = mnt;
584 mnt->mnt_mountpoint = mnt->mnt_root;
585 list_del_init(&mnt->mnt_child);
586 list_del_init(&mnt->mnt_hash);
587 dentry_reset_mounted(old_path->mnt, old_path->dentry);
591 * vfsmount lock must be held for write
593 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
594 struct vfsmount *child_mnt)
596 child_mnt->mnt_parent = mntget(mnt);
597 child_mnt->mnt_mountpoint = dget(dentry);
598 spin_lock(&dentry->d_lock);
599 dentry->d_flags |= DCACHE_MOUNTED;
600 spin_unlock(&dentry->d_lock);
604 * vfsmount lock must be held for write
606 static void attach_mnt(struct vfsmount *mnt, struct path *path)
608 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
609 list_add_tail(&mnt->mnt_hash, mount_hashtable +
610 hash(path->mnt, path->dentry));
611 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
615 * vfsmount lock must be held for write
617 static void commit_tree(struct vfsmount *mnt)
619 struct vfsmount *parent = mnt->mnt_parent;
620 struct vfsmount *m;
621 LIST_HEAD(head);
622 struct mnt_namespace *n = parent->mnt_ns;
624 BUG_ON(parent == mnt);
626 list_add_tail(&head, &mnt->mnt_list);
627 list_for_each_entry(m, &head, mnt_list)
628 m->mnt_ns = n;
629 list_splice(&head, n->list.prev);
631 list_add_tail(&mnt->mnt_hash, mount_hashtable +
632 hash(parent, mnt->mnt_mountpoint));
633 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
634 touch_mnt_namespace(n);
637 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
639 struct list_head *next = p->mnt_mounts.next;
640 if (next == &p->mnt_mounts) {
641 while (1) {
642 if (p == root)
643 return NULL;
644 next = p->mnt_child.next;
645 if (next != &p->mnt_parent->mnt_mounts)
646 break;
647 p = p->mnt_parent;
650 return list_entry(next, struct vfsmount, mnt_child);
653 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
655 struct list_head *prev = p->mnt_mounts.prev;
656 while (prev != &p->mnt_mounts) {
657 p = list_entry(prev, struct vfsmount, mnt_child);
658 prev = p->mnt_mounts.prev;
660 return p;
663 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
664 int flag)
666 struct super_block *sb = old->mnt_sb;
667 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
669 if (mnt) {
670 if (flag & (CL_SLAVE | CL_PRIVATE))
671 mnt->mnt_group_id = 0; /* not a peer of original */
672 else
673 mnt->mnt_group_id = old->mnt_group_id;
675 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
676 int err = mnt_alloc_group_id(mnt);
677 if (err)
678 goto out_free;
681 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
682 atomic_inc(&sb->s_active);
683 mnt->mnt_sb = sb;
684 mnt->mnt_root = dget(root);
685 mnt->mnt_mountpoint = mnt->mnt_root;
686 mnt->mnt_parent = mnt;
688 if (flag & CL_SLAVE) {
689 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
690 mnt->mnt_master = old;
691 CLEAR_MNT_SHARED(mnt);
692 } else if (!(flag & CL_PRIVATE)) {
693 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
694 list_add(&mnt->mnt_share, &old->mnt_share);
695 if (IS_MNT_SLAVE(old))
696 list_add(&mnt->mnt_slave, &old->mnt_slave);
697 mnt->mnt_master = old->mnt_master;
699 if (flag & CL_MAKE_SHARED)
700 set_mnt_shared(mnt);
702 /* stick the duplicate mount on the same expiry list
703 * as the original if that was on one */
704 if (flag & CL_EXPIRE) {
705 if (!list_empty(&old->mnt_expire))
706 list_add(&mnt->mnt_expire, &old->mnt_expire);
709 return mnt;
711 out_free:
712 free_vfsmnt(mnt);
713 return NULL;
716 static inline void mntfree(struct vfsmount *mnt)
718 struct super_block *sb = mnt->mnt_sb;
721 * This probably indicates that somebody messed
722 * up a mnt_want/drop_write() pair. If this
723 * happens, the filesystem was probably unable
724 * to make r/w->r/o transitions.
727 * The locking used to deal with mnt_count decrement provides barriers,
728 * so mnt_get_writers() below is safe.
730 WARN_ON(mnt_get_writers(mnt));
731 fsnotify_vfsmount_delete(mnt);
732 dput(mnt->mnt_root);
733 free_vfsmnt(mnt);
734 deactivate_super(sb);
737 #ifdef CONFIG_SMP
738 static inline void __mntput(struct vfsmount *mnt, int longrefs)
740 if (!longrefs) {
741 put_again:
742 br_read_lock(vfsmount_lock);
743 if (likely(atomic_read(&mnt->mnt_longrefs))) {
744 mnt_dec_count(mnt);
745 br_read_unlock(vfsmount_lock);
746 return;
748 br_read_unlock(vfsmount_lock);
749 } else {
750 BUG_ON(!atomic_read(&mnt->mnt_longrefs));
751 if (atomic_add_unless(&mnt->mnt_longrefs, -1, 1))
752 return;
755 br_write_lock(vfsmount_lock);
756 if (!longrefs)
757 mnt_dec_count(mnt);
758 else
759 atomic_dec(&mnt->mnt_longrefs);
760 if (mnt_get_count(mnt)) {
761 br_write_unlock(vfsmount_lock);
762 return;
764 if (unlikely(mnt->mnt_pinned)) {
765 mnt_add_count(mnt, mnt->mnt_pinned + 1);
766 mnt->mnt_pinned = 0;
767 br_write_unlock(vfsmount_lock);
768 acct_auto_close_mnt(mnt);
769 goto put_again;
771 br_write_unlock(vfsmount_lock);
772 mntfree(mnt);
774 #else
775 static inline void __mntput(struct vfsmount *mnt, int longrefs)
777 put_again:
778 mnt_dec_count(mnt);
779 if (likely(mnt_get_count(mnt)))
780 return;
781 br_write_lock(vfsmount_lock);
782 if (unlikely(mnt->mnt_pinned)) {
783 mnt_add_count(mnt, mnt->mnt_pinned + 1);
784 mnt->mnt_pinned = 0;
785 br_write_unlock(vfsmount_lock);
786 acct_auto_close_mnt(mnt);
787 goto put_again;
789 br_write_unlock(vfsmount_lock);
790 mntfree(mnt);
792 #endif
794 static void mntput_no_expire(struct vfsmount *mnt)
796 __mntput(mnt, 0);
799 void mntput(struct vfsmount *mnt)
801 if (mnt) {
802 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
803 if (unlikely(mnt->mnt_expiry_mark))
804 mnt->mnt_expiry_mark = 0;
805 __mntput(mnt, 0);
808 EXPORT_SYMBOL(mntput);
810 struct vfsmount *mntget(struct vfsmount *mnt)
812 if (mnt)
813 mnt_inc_count(mnt);
814 return mnt;
816 EXPORT_SYMBOL(mntget);
818 void mntput_long(struct vfsmount *mnt)
820 #ifdef CONFIG_SMP
821 if (mnt) {
822 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
823 if (unlikely(mnt->mnt_expiry_mark))
824 mnt->mnt_expiry_mark = 0;
825 __mntput(mnt, 1);
827 #else
828 mntput(mnt);
829 #endif
831 EXPORT_SYMBOL(mntput_long);
833 struct vfsmount *mntget_long(struct vfsmount *mnt)
835 #ifdef CONFIG_SMP
836 if (mnt)
837 atomic_inc(&mnt->mnt_longrefs);
838 return mnt;
839 #else
840 return mntget(mnt);
841 #endif
843 EXPORT_SYMBOL(mntget_long);
845 void mnt_pin(struct vfsmount *mnt)
847 br_write_lock(vfsmount_lock);
848 mnt->mnt_pinned++;
849 br_write_unlock(vfsmount_lock);
851 EXPORT_SYMBOL(mnt_pin);
853 void mnt_unpin(struct vfsmount *mnt)
855 br_write_lock(vfsmount_lock);
856 if (mnt->mnt_pinned) {
857 mnt_inc_count(mnt);
858 mnt->mnt_pinned--;
860 br_write_unlock(vfsmount_lock);
862 EXPORT_SYMBOL(mnt_unpin);
864 static inline void mangle(struct seq_file *m, const char *s)
866 seq_escape(m, s, " \t\n\\");
870 * Simple .show_options callback for filesystems which don't want to
871 * implement more complex mount option showing.
873 * See also save_mount_options().
875 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
877 const char *options;
879 rcu_read_lock();
880 options = rcu_dereference(mnt->mnt_sb->s_options);
882 if (options != NULL && options[0]) {
883 seq_putc(m, ',');
884 mangle(m, options);
886 rcu_read_unlock();
888 return 0;
890 EXPORT_SYMBOL(generic_show_options);
893 * If filesystem uses generic_show_options(), this function should be
894 * called from the fill_super() callback.
896 * The .remount_fs callback usually needs to be handled in a special
897 * way, to make sure, that previous options are not overwritten if the
898 * remount fails.
900 * Also note, that if the filesystem's .remount_fs function doesn't
901 * reset all options to their default value, but changes only newly
902 * given options, then the displayed options will not reflect reality
903 * any more.
905 void save_mount_options(struct super_block *sb, char *options)
907 BUG_ON(sb->s_options);
908 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
910 EXPORT_SYMBOL(save_mount_options);
912 void replace_mount_options(struct super_block *sb, char *options)
914 char *old = sb->s_options;
915 rcu_assign_pointer(sb->s_options, options);
916 if (old) {
917 synchronize_rcu();
918 kfree(old);
921 EXPORT_SYMBOL(replace_mount_options);
923 #ifdef CONFIG_PROC_FS
924 /* iterator */
925 static void *m_start(struct seq_file *m, loff_t *pos)
927 struct proc_mounts *p = m->private;
929 down_read(&namespace_sem);
930 return seq_list_start(&p->ns->list, *pos);
933 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
935 struct proc_mounts *p = m->private;
937 return seq_list_next(v, &p->ns->list, pos);
940 static void m_stop(struct seq_file *m, void *v)
942 up_read(&namespace_sem);
945 int mnt_had_events(struct proc_mounts *p)
947 struct mnt_namespace *ns = p->ns;
948 int res = 0;
950 br_read_lock(vfsmount_lock);
951 if (p->event != ns->event) {
952 p->event = ns->event;
953 res = 1;
955 br_read_unlock(vfsmount_lock);
957 return res;
960 struct proc_fs_info {
961 int flag;
962 const char *str;
965 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
967 static const struct proc_fs_info fs_info[] = {
968 { MS_SYNCHRONOUS, ",sync" },
969 { MS_DIRSYNC, ",dirsync" },
970 { MS_MANDLOCK, ",mand" },
971 { 0, NULL }
973 const struct proc_fs_info *fs_infop;
975 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
976 if (sb->s_flags & fs_infop->flag)
977 seq_puts(m, fs_infop->str);
980 return security_sb_show_options(m, sb);
983 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
985 static const struct proc_fs_info mnt_info[] = {
986 { MNT_NOSUID, ",nosuid" },
987 { MNT_NODEV, ",nodev" },
988 { MNT_NOEXEC, ",noexec" },
989 { MNT_NOATIME, ",noatime" },
990 { MNT_NODIRATIME, ",nodiratime" },
991 { MNT_RELATIME, ",relatime" },
992 { 0, NULL }
994 const struct proc_fs_info *fs_infop;
996 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
997 if (mnt->mnt_flags & fs_infop->flag)
998 seq_puts(m, fs_infop->str);
1002 static void show_type(struct seq_file *m, struct super_block *sb)
1004 mangle(m, sb->s_type->name);
1005 if (sb->s_subtype && sb->s_subtype[0]) {
1006 seq_putc(m, '.');
1007 mangle(m, sb->s_subtype);
1011 static int show_vfsmnt(struct seq_file *m, void *v)
1013 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1014 int err = 0;
1015 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1017 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1018 seq_putc(m, ' ');
1019 seq_path(m, &mnt_path, " \t\n\\");
1020 seq_putc(m, ' ');
1021 show_type(m, mnt->mnt_sb);
1022 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
1023 err = show_sb_opts(m, mnt->mnt_sb);
1024 if (err)
1025 goto out;
1026 show_mnt_opts(m, mnt);
1027 if (mnt->mnt_sb->s_op->show_options)
1028 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1029 seq_puts(m, " 0 0\n");
1030 out:
1031 return err;
1034 const struct seq_operations mounts_op = {
1035 .start = m_start,
1036 .next = m_next,
1037 .stop = m_stop,
1038 .show = show_vfsmnt
1041 static int show_mountinfo(struct seq_file *m, void *v)
1043 struct proc_mounts *p = m->private;
1044 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1045 struct super_block *sb = mnt->mnt_sb;
1046 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1047 struct path root = p->root;
1048 int err = 0;
1050 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1051 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1052 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1053 seq_putc(m, ' ');
1054 seq_path_root(m, &mnt_path, &root, " \t\n\\");
1055 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
1057 * Mountpoint is outside root, discard that one. Ugly,
1058 * but less so than trying to do that in iterator in a
1059 * race-free way (due to renames).
1061 return SEQ_SKIP;
1063 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1064 show_mnt_opts(m, mnt);
1066 /* Tagged fields ("foo:X" or "bar") */
1067 if (IS_MNT_SHARED(mnt))
1068 seq_printf(m, " shared:%i", mnt->mnt_group_id);
1069 if (IS_MNT_SLAVE(mnt)) {
1070 int master = mnt->mnt_master->mnt_group_id;
1071 int dom = get_dominating_id(mnt, &p->root);
1072 seq_printf(m, " master:%i", master);
1073 if (dom && dom != master)
1074 seq_printf(m, " propagate_from:%i", dom);
1076 if (IS_MNT_UNBINDABLE(mnt))
1077 seq_puts(m, " unbindable");
1079 /* Filesystem specific data */
1080 seq_puts(m, " - ");
1081 show_type(m, sb);
1082 seq_putc(m, ' ');
1083 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1084 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1085 err = show_sb_opts(m, sb);
1086 if (err)
1087 goto out;
1088 if (sb->s_op->show_options)
1089 err = sb->s_op->show_options(m, mnt);
1090 seq_putc(m, '\n');
1091 out:
1092 return err;
1095 const struct seq_operations mountinfo_op = {
1096 .start = m_start,
1097 .next = m_next,
1098 .stop = m_stop,
1099 .show = show_mountinfo,
1102 static int show_vfsstat(struct seq_file *m, void *v)
1104 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1105 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1106 int err = 0;
1108 /* device */
1109 if (mnt->mnt_devname) {
1110 seq_puts(m, "device ");
1111 mangle(m, mnt->mnt_devname);
1112 } else
1113 seq_puts(m, "no device");
1115 /* mount point */
1116 seq_puts(m, " mounted on ");
1117 seq_path(m, &mnt_path, " \t\n\\");
1118 seq_putc(m, ' ');
1120 /* file system type */
1121 seq_puts(m, "with fstype ");
1122 show_type(m, mnt->mnt_sb);
1124 /* optional statistics */
1125 if (mnt->mnt_sb->s_op->show_stats) {
1126 seq_putc(m, ' ');
1127 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1130 seq_putc(m, '\n');
1131 return err;
1134 const struct seq_operations mountstats_op = {
1135 .start = m_start,
1136 .next = m_next,
1137 .stop = m_stop,
1138 .show = show_vfsstat,
1140 #endif /* CONFIG_PROC_FS */
1143 * may_umount_tree - check if a mount tree is busy
1144 * @mnt: root of mount tree
1146 * This is called to check if a tree of mounts has any
1147 * open files, pwds, chroots or sub mounts that are
1148 * busy.
1150 int may_umount_tree(struct vfsmount *mnt)
1152 int actual_refs = 0;
1153 int minimum_refs = 0;
1154 struct vfsmount *p;
1156 /* write lock needed for mnt_get_count */
1157 br_write_lock(vfsmount_lock);
1158 for (p = mnt; p; p = next_mnt(p, mnt)) {
1159 actual_refs += mnt_get_count(p);
1160 minimum_refs += 2;
1162 br_write_unlock(vfsmount_lock);
1164 if (actual_refs > minimum_refs)
1165 return 0;
1167 return 1;
1170 EXPORT_SYMBOL(may_umount_tree);
1173 * may_umount - check if a mount point is busy
1174 * @mnt: root of mount
1176 * This is called to check if a mount point has any
1177 * open files, pwds, chroots or sub mounts. If the
1178 * mount has sub mounts this will return busy
1179 * regardless of whether the sub mounts are busy.
1181 * Doesn't take quota and stuff into account. IOW, in some cases it will
1182 * give false negatives. The main reason why it's here is that we need
1183 * a non-destructive way to look for easily umountable filesystems.
1185 int may_umount(struct vfsmount *mnt)
1187 int ret = 1;
1188 down_read(&namespace_sem);
1189 br_write_lock(vfsmount_lock);
1190 if (propagate_mount_busy(mnt, 2))
1191 ret = 0;
1192 br_write_unlock(vfsmount_lock);
1193 up_read(&namespace_sem);
1194 return ret;
1197 EXPORT_SYMBOL(may_umount);
1199 void release_mounts(struct list_head *head)
1201 struct vfsmount *mnt;
1202 while (!list_empty(head)) {
1203 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1204 list_del_init(&mnt->mnt_hash);
1205 if (mnt->mnt_parent != mnt) {
1206 struct dentry *dentry;
1207 struct vfsmount *m;
1209 br_write_lock(vfsmount_lock);
1210 dentry = mnt->mnt_mountpoint;
1211 m = mnt->mnt_parent;
1212 mnt->mnt_mountpoint = mnt->mnt_root;
1213 mnt->mnt_parent = mnt;
1214 m->mnt_ghosts--;
1215 br_write_unlock(vfsmount_lock);
1216 dput(dentry);
1217 mntput(m);
1219 mntput_long(mnt);
1224 * vfsmount lock must be held for write
1225 * namespace_sem must be held for write
1227 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1229 struct vfsmount *p;
1231 for (p = mnt; p; p = next_mnt(p, mnt))
1232 list_move(&p->mnt_hash, kill);
1234 if (propagate)
1235 propagate_umount(kill);
1237 list_for_each_entry(p, kill, mnt_hash) {
1238 list_del_init(&p->mnt_expire);
1239 list_del_init(&p->mnt_list);
1240 __touch_mnt_namespace(p->mnt_ns);
1241 p->mnt_ns = NULL;
1242 list_del_init(&p->mnt_child);
1243 if (p->mnt_parent != p) {
1244 p->mnt_parent->mnt_ghosts++;
1245 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
1247 change_mnt_propagation(p, MS_PRIVATE);
1251 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1253 static int do_umount(struct vfsmount *mnt, int flags)
1255 struct super_block *sb = mnt->mnt_sb;
1256 int retval;
1257 LIST_HEAD(umount_list);
1259 retval = security_sb_umount(mnt, flags);
1260 if (retval)
1261 return retval;
1264 * Allow userspace to request a mountpoint be expired rather than
1265 * unmounting unconditionally. Unmount only happens if:
1266 * (1) the mark is already set (the mark is cleared by mntput())
1267 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1269 if (flags & MNT_EXPIRE) {
1270 if (mnt == current->fs->root.mnt ||
1271 flags & (MNT_FORCE | MNT_DETACH))
1272 return -EINVAL;
1275 * probably don't strictly need the lock here if we examined
1276 * all race cases, but it's a slowpath.
1278 br_write_lock(vfsmount_lock);
1279 if (mnt_get_count(mnt) != 2) {
1280 br_write_lock(vfsmount_lock);
1281 return -EBUSY;
1283 br_write_unlock(vfsmount_lock);
1285 if (!xchg(&mnt->mnt_expiry_mark, 1))
1286 return -EAGAIN;
1290 * If we may have to abort operations to get out of this
1291 * mount, and they will themselves hold resources we must
1292 * allow the fs to do things. In the Unix tradition of
1293 * 'Gee thats tricky lets do it in userspace' the umount_begin
1294 * might fail to complete on the first run through as other tasks
1295 * must return, and the like. Thats for the mount program to worry
1296 * about for the moment.
1299 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1300 sb->s_op->umount_begin(sb);
1304 * No sense to grab the lock for this test, but test itself looks
1305 * somewhat bogus. Suggestions for better replacement?
1306 * Ho-hum... In principle, we might treat that as umount + switch
1307 * to rootfs. GC would eventually take care of the old vfsmount.
1308 * Actually it makes sense, especially if rootfs would contain a
1309 * /reboot - static binary that would close all descriptors and
1310 * call reboot(9). Then init(8) could umount root and exec /reboot.
1312 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1314 * Special case for "unmounting" root ...
1315 * we just try to remount it readonly.
1317 down_write(&sb->s_umount);
1318 if (!(sb->s_flags & MS_RDONLY))
1319 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1320 up_write(&sb->s_umount);
1321 return retval;
1324 down_write(&namespace_sem);
1325 br_write_lock(vfsmount_lock);
1326 event++;
1328 if (!(flags & MNT_DETACH))
1329 shrink_submounts(mnt, &umount_list);
1331 retval = -EBUSY;
1332 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1333 if (!list_empty(&mnt->mnt_list))
1334 umount_tree(mnt, 1, &umount_list);
1335 retval = 0;
1337 br_write_unlock(vfsmount_lock);
1338 up_write(&namespace_sem);
1339 release_mounts(&umount_list);
1340 return retval;
1344 * Now umount can handle mount points as well as block devices.
1345 * This is important for filesystems which use unnamed block devices.
1347 * We now support a flag for forced unmount like the other 'big iron'
1348 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1351 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1353 struct path path;
1354 int retval;
1355 int lookup_flags = 0;
1357 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1358 return -EINVAL;
1360 if (!(flags & UMOUNT_NOFOLLOW))
1361 lookup_flags |= LOOKUP_FOLLOW;
1363 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1364 if (retval)
1365 goto out;
1366 retval = -EINVAL;
1367 if (path.dentry != path.mnt->mnt_root)
1368 goto dput_and_out;
1369 if (!check_mnt(path.mnt))
1370 goto dput_and_out;
1372 retval = -EPERM;
1373 if (!capable(CAP_SYS_ADMIN))
1374 goto dput_and_out;
1376 retval = do_umount(path.mnt, flags);
1377 dput_and_out:
1378 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1379 dput(path.dentry);
1380 mntput_no_expire(path.mnt);
1381 out:
1382 return retval;
1385 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1388 * The 2.0 compatible umount. No flags.
1390 SYSCALL_DEFINE1(oldumount, char __user *, name)
1392 return sys_umount(name, 0);
1395 #endif
1397 static int mount_is_safe(struct path *path)
1399 if (capable(CAP_SYS_ADMIN))
1400 return 0;
1401 return -EPERM;
1402 #ifdef notyet
1403 if (S_ISLNK(path->dentry->d_inode->i_mode))
1404 return -EPERM;
1405 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1406 if (current_uid() != path->dentry->d_inode->i_uid)
1407 return -EPERM;
1409 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1410 return -EPERM;
1411 return 0;
1412 #endif
1415 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1416 int flag)
1418 struct vfsmount *res, *p, *q, *r, *s;
1419 struct path path;
1421 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1422 return NULL;
1424 res = q = clone_mnt(mnt, dentry, flag);
1425 if (!q)
1426 goto Enomem;
1427 q->mnt_mountpoint = mnt->mnt_mountpoint;
1429 p = mnt;
1430 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1431 if (!is_subdir(r->mnt_mountpoint, dentry))
1432 continue;
1434 for (s = r; s; s = next_mnt(s, r)) {
1435 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1436 s = skip_mnt_tree(s);
1437 continue;
1439 while (p != s->mnt_parent) {
1440 p = p->mnt_parent;
1441 q = q->mnt_parent;
1443 p = s;
1444 path.mnt = q;
1445 path.dentry = p->mnt_mountpoint;
1446 q = clone_mnt(p, p->mnt_root, flag);
1447 if (!q)
1448 goto Enomem;
1449 br_write_lock(vfsmount_lock);
1450 list_add_tail(&q->mnt_list, &res->mnt_list);
1451 attach_mnt(q, &path);
1452 br_write_unlock(vfsmount_lock);
1455 return res;
1456 Enomem:
1457 if (res) {
1458 LIST_HEAD(umount_list);
1459 br_write_lock(vfsmount_lock);
1460 umount_tree(res, 0, &umount_list);
1461 br_write_unlock(vfsmount_lock);
1462 release_mounts(&umount_list);
1464 return NULL;
1467 struct vfsmount *collect_mounts(struct path *path)
1469 struct vfsmount *tree;
1470 down_write(&namespace_sem);
1471 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1472 up_write(&namespace_sem);
1473 return tree;
1476 void drop_collected_mounts(struct vfsmount *mnt)
1478 LIST_HEAD(umount_list);
1479 down_write(&namespace_sem);
1480 br_write_lock(vfsmount_lock);
1481 umount_tree(mnt, 0, &umount_list);
1482 br_write_unlock(vfsmount_lock);
1483 up_write(&namespace_sem);
1484 release_mounts(&umount_list);
1487 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1488 struct vfsmount *root)
1490 struct vfsmount *mnt;
1491 int res = f(root, arg);
1492 if (res)
1493 return res;
1494 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1495 res = f(mnt, arg);
1496 if (res)
1497 return res;
1499 return 0;
1502 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1504 struct vfsmount *p;
1506 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1507 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1508 mnt_release_group_id(p);
1512 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1514 struct vfsmount *p;
1516 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1517 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1518 int err = mnt_alloc_group_id(p);
1519 if (err) {
1520 cleanup_group_ids(mnt, p);
1521 return err;
1526 return 0;
1530 * @source_mnt : mount tree to be attached
1531 * @nd : place the mount tree @source_mnt is attached
1532 * @parent_nd : if non-null, detach the source_mnt from its parent and
1533 * store the parent mount and mountpoint dentry.
1534 * (done when source_mnt is moved)
1536 * NOTE: in the table below explains the semantics when a source mount
1537 * of a given type is attached to a destination mount of a given type.
1538 * ---------------------------------------------------------------------------
1539 * | BIND MOUNT OPERATION |
1540 * |**************************************************************************
1541 * | source-->| shared | private | slave | unbindable |
1542 * | dest | | | | |
1543 * | | | | | | |
1544 * | v | | | | |
1545 * |**************************************************************************
1546 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1547 * | | | | | |
1548 * |non-shared| shared (+) | private | slave (*) | invalid |
1549 * ***************************************************************************
1550 * A bind operation clones the source mount and mounts the clone on the
1551 * destination mount.
1553 * (++) the cloned mount is propagated to all the mounts in the propagation
1554 * tree of the destination mount and the cloned mount is added to
1555 * the peer group of the source mount.
1556 * (+) the cloned mount is created under the destination mount and is marked
1557 * as shared. The cloned mount is added to the peer group of the source
1558 * mount.
1559 * (+++) the mount is propagated to all the mounts in the propagation tree
1560 * of the destination mount and the cloned mount is made slave
1561 * of the same master as that of the source mount. The cloned mount
1562 * is marked as 'shared and slave'.
1563 * (*) the cloned mount is made a slave of the same master as that of the
1564 * source mount.
1566 * ---------------------------------------------------------------------------
1567 * | MOVE MOUNT OPERATION |
1568 * |**************************************************************************
1569 * | source-->| shared | private | slave | unbindable |
1570 * | dest | | | | |
1571 * | | | | | | |
1572 * | v | | | | |
1573 * |**************************************************************************
1574 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1575 * | | | | | |
1576 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1577 * ***************************************************************************
1579 * (+) the mount is moved to the destination. And is then propagated to
1580 * all the mounts in the propagation tree of the destination mount.
1581 * (+*) the mount is moved to the destination.
1582 * (+++) the mount is moved to the destination and is then propagated to
1583 * all the mounts belonging to the destination mount's propagation tree.
1584 * the mount is marked as 'shared and slave'.
1585 * (*) the mount continues to be a slave at the new location.
1587 * if the source mount is a tree, the operations explained above is
1588 * applied to each mount in the tree.
1589 * Must be called without spinlocks held, since this function can sleep
1590 * in allocations.
1592 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1593 struct path *path, struct path *parent_path)
1595 LIST_HEAD(tree_list);
1596 struct vfsmount *dest_mnt = path->mnt;
1597 struct dentry *dest_dentry = path->dentry;
1598 struct vfsmount *child, *p;
1599 int err;
1601 if (IS_MNT_SHARED(dest_mnt)) {
1602 err = invent_group_ids(source_mnt, true);
1603 if (err)
1604 goto out;
1606 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1607 if (err)
1608 goto out_cleanup_ids;
1610 br_write_lock(vfsmount_lock);
1612 if (IS_MNT_SHARED(dest_mnt)) {
1613 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1614 set_mnt_shared(p);
1616 if (parent_path) {
1617 detach_mnt(source_mnt, parent_path);
1618 attach_mnt(source_mnt, path);
1619 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1620 } else {
1621 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1622 commit_tree(source_mnt);
1625 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1626 list_del_init(&child->mnt_hash);
1627 commit_tree(child);
1629 br_write_unlock(vfsmount_lock);
1631 return 0;
1633 out_cleanup_ids:
1634 if (IS_MNT_SHARED(dest_mnt))
1635 cleanup_group_ids(source_mnt, NULL);
1636 out:
1637 return err;
1640 static int graft_tree(struct vfsmount *mnt, struct path *path)
1642 int err;
1643 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1644 return -EINVAL;
1646 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1647 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1648 return -ENOTDIR;
1650 err = -ENOENT;
1651 mutex_lock(&path->dentry->d_inode->i_mutex);
1652 if (cant_mount(path->dentry))
1653 goto out_unlock;
1655 if (!d_unlinked(path->dentry))
1656 err = attach_recursive_mnt(mnt, path, NULL);
1657 out_unlock:
1658 mutex_unlock(&path->dentry->d_inode->i_mutex);
1659 return err;
1663 * Sanity check the flags to change_mnt_propagation.
1666 static int flags_to_propagation_type(int flags)
1668 int type = flags & ~MS_REC;
1670 /* Fail if any non-propagation flags are set */
1671 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1672 return 0;
1673 /* Only one propagation flag should be set */
1674 if (!is_power_of_2(type))
1675 return 0;
1676 return type;
1680 * recursively change the type of the mountpoint.
1682 static int do_change_type(struct path *path, int flag)
1684 struct vfsmount *m, *mnt = path->mnt;
1685 int recurse = flag & MS_REC;
1686 int type;
1687 int err = 0;
1689 if (!capable(CAP_SYS_ADMIN))
1690 return -EPERM;
1692 if (path->dentry != path->mnt->mnt_root)
1693 return -EINVAL;
1695 type = flags_to_propagation_type(flag);
1696 if (!type)
1697 return -EINVAL;
1699 down_write(&namespace_sem);
1700 if (type == MS_SHARED) {
1701 err = invent_group_ids(mnt, recurse);
1702 if (err)
1703 goto out_unlock;
1706 br_write_lock(vfsmount_lock);
1707 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1708 change_mnt_propagation(m, type);
1709 br_write_unlock(vfsmount_lock);
1711 out_unlock:
1712 up_write(&namespace_sem);
1713 return err;
1717 * do loopback mount.
1719 static int do_loopback(struct path *path, char *old_name,
1720 int recurse)
1722 struct path old_path;
1723 struct vfsmount *mnt = NULL;
1724 int err = mount_is_safe(path);
1725 if (err)
1726 return err;
1727 if (!old_name || !*old_name)
1728 return -EINVAL;
1729 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1730 if (err)
1731 return err;
1733 down_write(&namespace_sem);
1734 err = -EINVAL;
1735 if (IS_MNT_UNBINDABLE(old_path.mnt))
1736 goto out;
1738 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1739 goto out;
1741 err = -ENOMEM;
1742 if (recurse)
1743 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1744 else
1745 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1747 if (!mnt)
1748 goto out;
1750 err = graft_tree(mnt, path);
1751 if (err) {
1752 LIST_HEAD(umount_list);
1754 br_write_lock(vfsmount_lock);
1755 umount_tree(mnt, 0, &umount_list);
1756 br_write_unlock(vfsmount_lock);
1757 release_mounts(&umount_list);
1760 out:
1761 up_write(&namespace_sem);
1762 path_put(&old_path);
1763 return err;
1766 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1768 int error = 0;
1769 int readonly_request = 0;
1771 if (ms_flags & MS_RDONLY)
1772 readonly_request = 1;
1773 if (readonly_request == __mnt_is_readonly(mnt))
1774 return 0;
1776 if (readonly_request)
1777 error = mnt_make_readonly(mnt);
1778 else
1779 __mnt_unmake_readonly(mnt);
1780 return error;
1784 * change filesystem flags. dir should be a physical root of filesystem.
1785 * If you've mounted a non-root directory somewhere and want to do remount
1786 * on it - tough luck.
1788 static int do_remount(struct path *path, int flags, int mnt_flags,
1789 void *data)
1791 int err;
1792 struct super_block *sb = path->mnt->mnt_sb;
1794 if (!capable(CAP_SYS_ADMIN))
1795 return -EPERM;
1797 if (!check_mnt(path->mnt))
1798 return -EINVAL;
1800 if (path->dentry != path->mnt->mnt_root)
1801 return -EINVAL;
1803 down_write(&sb->s_umount);
1804 if (flags & MS_BIND)
1805 err = change_mount_flags(path->mnt, flags);
1806 else
1807 err = do_remount_sb(sb, flags, data, 0);
1808 if (!err) {
1809 br_write_lock(vfsmount_lock);
1810 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1811 path->mnt->mnt_flags = mnt_flags;
1812 br_write_unlock(vfsmount_lock);
1814 up_write(&sb->s_umount);
1815 if (!err) {
1816 br_write_lock(vfsmount_lock);
1817 touch_mnt_namespace(path->mnt->mnt_ns);
1818 br_write_unlock(vfsmount_lock);
1820 return err;
1823 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1825 struct vfsmount *p;
1826 for (p = mnt; p; p = next_mnt(p, mnt)) {
1827 if (IS_MNT_UNBINDABLE(p))
1828 return 1;
1830 return 0;
1833 static int do_move_mount(struct path *path, char *old_name)
1835 struct path old_path, parent_path;
1836 struct vfsmount *p;
1837 int err = 0;
1838 if (!capable(CAP_SYS_ADMIN))
1839 return -EPERM;
1840 if (!old_name || !*old_name)
1841 return -EINVAL;
1842 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1843 if (err)
1844 return err;
1846 down_write(&namespace_sem);
1847 while (d_mountpoint(path->dentry) &&
1848 follow_down(path))
1850 err = -EINVAL;
1851 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1852 goto out;
1854 err = -ENOENT;
1855 mutex_lock(&path->dentry->d_inode->i_mutex);
1856 if (cant_mount(path->dentry))
1857 goto out1;
1859 if (d_unlinked(path->dentry))
1860 goto out1;
1862 err = -EINVAL;
1863 if (old_path.dentry != old_path.mnt->mnt_root)
1864 goto out1;
1866 if (old_path.mnt == old_path.mnt->mnt_parent)
1867 goto out1;
1869 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1870 S_ISDIR(old_path.dentry->d_inode->i_mode))
1871 goto out1;
1873 * Don't move a mount residing in a shared parent.
1875 if (old_path.mnt->mnt_parent &&
1876 IS_MNT_SHARED(old_path.mnt->mnt_parent))
1877 goto out1;
1879 * Don't move a mount tree containing unbindable mounts to a destination
1880 * mount which is shared.
1882 if (IS_MNT_SHARED(path->mnt) &&
1883 tree_contains_unbindable(old_path.mnt))
1884 goto out1;
1885 err = -ELOOP;
1886 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1887 if (p == old_path.mnt)
1888 goto out1;
1890 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1891 if (err)
1892 goto out1;
1894 /* if the mount is moved, it should no longer be expire
1895 * automatically */
1896 list_del_init(&old_path.mnt->mnt_expire);
1897 out1:
1898 mutex_unlock(&path->dentry->d_inode->i_mutex);
1899 out:
1900 up_write(&namespace_sem);
1901 if (!err)
1902 path_put(&parent_path);
1903 path_put(&old_path);
1904 return err;
1908 * create a new mount for userspace and request it to be added into the
1909 * namespace's tree
1911 static int do_new_mount(struct path *path, char *type, int flags,
1912 int mnt_flags, char *name, void *data)
1914 struct vfsmount *mnt;
1916 if (!type)
1917 return -EINVAL;
1919 /* we need capabilities... */
1920 if (!capable(CAP_SYS_ADMIN))
1921 return -EPERM;
1923 mnt = do_kern_mount(type, flags, name, data);
1924 if (IS_ERR(mnt))
1925 return PTR_ERR(mnt);
1927 return do_add_mount(mnt, path, mnt_flags, NULL);
1931 * add a mount into a namespace's mount tree
1932 * - provide the option of adding the new mount to an expiration list
1934 int do_add_mount(struct vfsmount *newmnt, struct path *path,
1935 int mnt_flags, struct list_head *fslist)
1937 int err;
1939 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1941 down_write(&namespace_sem);
1942 /* Something was mounted here while we slept */
1943 while (d_mountpoint(path->dentry) &&
1944 follow_down(path))
1946 err = -EINVAL;
1947 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1948 goto unlock;
1950 /* Refuse the same filesystem on the same mount point */
1951 err = -EBUSY;
1952 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1953 path->mnt->mnt_root == path->dentry)
1954 goto unlock;
1956 err = -EINVAL;
1957 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1958 goto unlock;
1960 newmnt->mnt_flags = mnt_flags;
1961 if ((err = graft_tree(newmnt, path)))
1962 goto unlock;
1964 if (fslist) /* add to the specified expiration list */
1965 list_add_tail(&newmnt->mnt_expire, fslist);
1967 up_write(&namespace_sem);
1968 return 0;
1970 unlock:
1971 up_write(&namespace_sem);
1972 mntput_long(newmnt);
1973 return err;
1976 EXPORT_SYMBOL_GPL(do_add_mount);
1979 * process a list of expirable mountpoints with the intent of discarding any
1980 * mountpoints that aren't in use and haven't been touched since last we came
1981 * here
1983 void mark_mounts_for_expiry(struct list_head *mounts)
1985 struct vfsmount *mnt, *next;
1986 LIST_HEAD(graveyard);
1987 LIST_HEAD(umounts);
1989 if (list_empty(mounts))
1990 return;
1992 down_write(&namespace_sem);
1993 br_write_lock(vfsmount_lock);
1995 /* extract from the expiration list every vfsmount that matches the
1996 * following criteria:
1997 * - only referenced by its parent vfsmount
1998 * - still marked for expiry (marked on the last call here; marks are
1999 * cleared by mntput())
2001 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2002 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2003 propagate_mount_busy(mnt, 1))
2004 continue;
2005 list_move(&mnt->mnt_expire, &graveyard);
2007 while (!list_empty(&graveyard)) {
2008 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2009 touch_mnt_namespace(mnt->mnt_ns);
2010 umount_tree(mnt, 1, &umounts);
2012 br_write_unlock(vfsmount_lock);
2013 up_write(&namespace_sem);
2015 release_mounts(&umounts);
2018 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2021 * Ripoff of 'select_parent()'
2023 * search the list of submounts for a given mountpoint, and move any
2024 * shrinkable submounts to the 'graveyard' list.
2026 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2028 struct vfsmount *this_parent = parent;
2029 struct list_head *next;
2030 int found = 0;
2032 repeat:
2033 next = this_parent->mnt_mounts.next;
2034 resume:
2035 while (next != &this_parent->mnt_mounts) {
2036 struct list_head *tmp = next;
2037 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2039 next = tmp->next;
2040 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2041 continue;
2043 * Descend a level if the d_mounts list is non-empty.
2045 if (!list_empty(&mnt->mnt_mounts)) {
2046 this_parent = mnt;
2047 goto repeat;
2050 if (!propagate_mount_busy(mnt, 1)) {
2051 list_move_tail(&mnt->mnt_expire, graveyard);
2052 found++;
2056 * All done at this level ... ascend and resume the search
2058 if (this_parent != parent) {
2059 next = this_parent->mnt_child.next;
2060 this_parent = this_parent->mnt_parent;
2061 goto resume;
2063 return found;
2067 * process a list of expirable mountpoints with the intent of discarding any
2068 * submounts of a specific parent mountpoint
2070 * vfsmount_lock must be held for write
2072 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
2074 LIST_HEAD(graveyard);
2075 struct vfsmount *m;
2077 /* extract submounts of 'mountpoint' from the expiration list */
2078 while (select_submounts(mnt, &graveyard)) {
2079 while (!list_empty(&graveyard)) {
2080 m = list_first_entry(&graveyard, struct vfsmount,
2081 mnt_expire);
2082 touch_mnt_namespace(m->mnt_ns);
2083 umount_tree(m, 1, umounts);
2089 * Some copy_from_user() implementations do not return the exact number of
2090 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2091 * Note that this function differs from copy_from_user() in that it will oops
2092 * on bad values of `to', rather than returning a short copy.
2094 static long exact_copy_from_user(void *to, const void __user * from,
2095 unsigned long n)
2097 char *t = to;
2098 const char __user *f = from;
2099 char c;
2101 if (!access_ok(VERIFY_READ, from, n))
2102 return n;
2104 while (n) {
2105 if (__get_user(c, f)) {
2106 memset(t, 0, n);
2107 break;
2109 *t++ = c;
2110 f++;
2111 n--;
2113 return n;
2116 int copy_mount_options(const void __user * data, unsigned long *where)
2118 int i;
2119 unsigned long page;
2120 unsigned long size;
2122 *where = 0;
2123 if (!data)
2124 return 0;
2126 if (!(page = __get_free_page(GFP_KERNEL)))
2127 return -ENOMEM;
2129 /* We only care that *some* data at the address the user
2130 * gave us is valid. Just in case, we'll zero
2131 * the remainder of the page.
2133 /* copy_from_user cannot cross TASK_SIZE ! */
2134 size = TASK_SIZE - (unsigned long)data;
2135 if (size > PAGE_SIZE)
2136 size = PAGE_SIZE;
2138 i = size - exact_copy_from_user((void *)page, data, size);
2139 if (!i) {
2140 free_page(page);
2141 return -EFAULT;
2143 if (i != PAGE_SIZE)
2144 memset((char *)page + i, 0, PAGE_SIZE - i);
2145 *where = page;
2146 return 0;
2149 int copy_mount_string(const void __user *data, char **where)
2151 char *tmp;
2153 if (!data) {
2154 *where = NULL;
2155 return 0;
2158 tmp = strndup_user(data, PAGE_SIZE);
2159 if (IS_ERR(tmp))
2160 return PTR_ERR(tmp);
2162 *where = tmp;
2163 return 0;
2167 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2168 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2170 * data is a (void *) that can point to any structure up to
2171 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2172 * information (or be NULL).
2174 * Pre-0.97 versions of mount() didn't have a flags word.
2175 * When the flags word was introduced its top half was required
2176 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2177 * Therefore, if this magic number is present, it carries no information
2178 * and must be discarded.
2180 long do_mount(char *dev_name, char *dir_name, char *type_page,
2181 unsigned long flags, void *data_page)
2183 struct path path;
2184 int retval = 0;
2185 int mnt_flags = 0;
2187 /* Discard magic */
2188 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2189 flags &= ~MS_MGC_MSK;
2191 /* Basic sanity checks */
2193 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2194 return -EINVAL;
2196 if (data_page)
2197 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2199 /* ... and get the mountpoint */
2200 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2201 if (retval)
2202 return retval;
2204 retval = security_sb_mount(dev_name, &path,
2205 type_page, flags, data_page);
2206 if (retval)
2207 goto dput_out;
2209 /* Default to relatime unless overriden */
2210 if (!(flags & MS_NOATIME))
2211 mnt_flags |= MNT_RELATIME;
2213 /* Separate the per-mountpoint flags */
2214 if (flags & MS_NOSUID)
2215 mnt_flags |= MNT_NOSUID;
2216 if (flags & MS_NODEV)
2217 mnt_flags |= MNT_NODEV;
2218 if (flags & MS_NOEXEC)
2219 mnt_flags |= MNT_NOEXEC;
2220 if (flags & MS_NOATIME)
2221 mnt_flags |= MNT_NOATIME;
2222 if (flags & MS_NODIRATIME)
2223 mnt_flags |= MNT_NODIRATIME;
2224 if (flags & MS_STRICTATIME)
2225 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2226 if (flags & MS_RDONLY)
2227 mnt_flags |= MNT_READONLY;
2229 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2230 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2231 MS_STRICTATIME);
2233 if (flags & MS_REMOUNT)
2234 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2235 data_page);
2236 else if (flags & MS_BIND)
2237 retval = do_loopback(&path, dev_name, flags & MS_REC);
2238 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2239 retval = do_change_type(&path, flags);
2240 else if (flags & MS_MOVE)
2241 retval = do_move_mount(&path, dev_name);
2242 else
2243 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2244 dev_name, data_page);
2245 dput_out:
2246 path_put(&path);
2247 return retval;
2250 static struct mnt_namespace *alloc_mnt_ns(void)
2252 struct mnt_namespace *new_ns;
2254 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2255 if (!new_ns)
2256 return ERR_PTR(-ENOMEM);
2257 atomic_set(&new_ns->count, 1);
2258 new_ns->root = NULL;
2259 INIT_LIST_HEAD(&new_ns->list);
2260 init_waitqueue_head(&new_ns->poll);
2261 new_ns->event = 0;
2262 return new_ns;
2266 * Allocate a new namespace structure and populate it with contents
2267 * copied from the namespace of the passed in task structure.
2269 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2270 struct fs_struct *fs)
2272 struct mnt_namespace *new_ns;
2273 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2274 struct vfsmount *p, *q;
2276 new_ns = alloc_mnt_ns();
2277 if (IS_ERR(new_ns))
2278 return new_ns;
2280 down_write(&namespace_sem);
2281 /* First pass: copy the tree topology */
2282 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2283 CL_COPY_ALL | CL_EXPIRE);
2284 if (!new_ns->root) {
2285 up_write(&namespace_sem);
2286 kfree(new_ns);
2287 return ERR_PTR(-ENOMEM);
2289 br_write_lock(vfsmount_lock);
2290 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2291 br_write_unlock(vfsmount_lock);
2294 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2295 * as belonging to new namespace. We have already acquired a private
2296 * fs_struct, so tsk->fs->lock is not needed.
2298 p = mnt_ns->root;
2299 q = new_ns->root;
2300 while (p) {
2301 q->mnt_ns = new_ns;
2302 if (fs) {
2303 if (p == fs->root.mnt) {
2304 rootmnt = p;
2305 fs->root.mnt = mntget_long(q);
2307 if (p == fs->pwd.mnt) {
2308 pwdmnt = p;
2309 fs->pwd.mnt = mntget_long(q);
2312 p = next_mnt(p, mnt_ns->root);
2313 q = next_mnt(q, new_ns->root);
2315 up_write(&namespace_sem);
2317 if (rootmnt)
2318 mntput_long(rootmnt);
2319 if (pwdmnt)
2320 mntput_long(pwdmnt);
2322 return new_ns;
2325 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2326 struct fs_struct *new_fs)
2328 struct mnt_namespace *new_ns;
2330 BUG_ON(!ns);
2331 get_mnt_ns(ns);
2333 if (!(flags & CLONE_NEWNS))
2334 return ns;
2336 new_ns = dup_mnt_ns(ns, new_fs);
2338 put_mnt_ns(ns);
2339 return new_ns;
2343 * create_mnt_ns - creates a private namespace and adds a root filesystem
2344 * @mnt: pointer to the new root filesystem mountpoint
2346 struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2348 struct mnt_namespace *new_ns;
2350 new_ns = alloc_mnt_ns();
2351 if (!IS_ERR(new_ns)) {
2352 mnt->mnt_ns = new_ns;
2353 new_ns->root = mnt;
2354 list_add(&new_ns->list, &new_ns->root->mnt_list);
2356 return new_ns;
2358 EXPORT_SYMBOL(create_mnt_ns);
2360 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2361 char __user *, type, unsigned long, flags, void __user *, data)
2363 int ret;
2364 char *kernel_type;
2365 char *kernel_dir;
2366 char *kernel_dev;
2367 unsigned long data_page;
2369 ret = copy_mount_string(type, &kernel_type);
2370 if (ret < 0)
2371 goto out_type;
2373 kernel_dir = getname(dir_name);
2374 if (IS_ERR(kernel_dir)) {
2375 ret = PTR_ERR(kernel_dir);
2376 goto out_dir;
2379 ret = copy_mount_string(dev_name, &kernel_dev);
2380 if (ret < 0)
2381 goto out_dev;
2383 ret = copy_mount_options(data, &data_page);
2384 if (ret < 0)
2385 goto out_data;
2387 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2388 (void *) data_page);
2390 free_page(data_page);
2391 out_data:
2392 kfree(kernel_dev);
2393 out_dev:
2394 putname(kernel_dir);
2395 out_dir:
2396 kfree(kernel_type);
2397 out_type:
2398 return ret;
2402 * pivot_root Semantics:
2403 * Moves the root file system of the current process to the directory put_old,
2404 * makes new_root as the new root file system of the current process, and sets
2405 * root/cwd of all processes which had them on the current root to new_root.
2407 * Restrictions:
2408 * The new_root and put_old must be directories, and must not be on the
2409 * same file system as the current process root. The put_old must be
2410 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2411 * pointed to by put_old must yield the same directory as new_root. No other
2412 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2414 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2415 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2416 * in this situation.
2418 * Notes:
2419 * - we don't move root/cwd if they are not at the root (reason: if something
2420 * cared enough to change them, it's probably wrong to force them elsewhere)
2421 * - it's okay to pick a root that isn't the root of a file system, e.g.
2422 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2423 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2424 * first.
2426 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2427 const char __user *, put_old)
2429 struct vfsmount *tmp;
2430 struct path new, old, parent_path, root_parent, root;
2431 int error;
2433 if (!capable(CAP_SYS_ADMIN))
2434 return -EPERM;
2436 error = user_path_dir(new_root, &new);
2437 if (error)
2438 goto out0;
2439 error = -EINVAL;
2440 if (!check_mnt(new.mnt))
2441 goto out1;
2443 error = user_path_dir(put_old, &old);
2444 if (error)
2445 goto out1;
2447 error = security_sb_pivotroot(&old, &new);
2448 if (error) {
2449 path_put(&old);
2450 goto out1;
2453 get_fs_root(current->fs, &root);
2454 down_write(&namespace_sem);
2455 mutex_lock(&old.dentry->d_inode->i_mutex);
2456 error = -EINVAL;
2457 if (IS_MNT_SHARED(old.mnt) ||
2458 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2459 IS_MNT_SHARED(root.mnt->mnt_parent))
2460 goto out2;
2461 if (!check_mnt(root.mnt))
2462 goto out2;
2463 error = -ENOENT;
2464 if (cant_mount(old.dentry))
2465 goto out2;
2466 if (d_unlinked(new.dentry))
2467 goto out2;
2468 if (d_unlinked(old.dentry))
2469 goto out2;
2470 error = -EBUSY;
2471 if (new.mnt == root.mnt ||
2472 old.mnt == root.mnt)
2473 goto out2; /* loop, on the same file system */
2474 error = -EINVAL;
2475 if (root.mnt->mnt_root != root.dentry)
2476 goto out2; /* not a mountpoint */
2477 if (root.mnt->mnt_parent == root.mnt)
2478 goto out2; /* not attached */
2479 if (new.mnt->mnt_root != new.dentry)
2480 goto out2; /* not a mountpoint */
2481 if (new.mnt->mnt_parent == new.mnt)
2482 goto out2; /* not attached */
2483 /* make sure we can reach put_old from new_root */
2484 tmp = old.mnt;
2485 br_write_lock(vfsmount_lock);
2486 if (tmp != new.mnt) {
2487 for (;;) {
2488 if (tmp->mnt_parent == tmp)
2489 goto out3; /* already mounted on put_old */
2490 if (tmp->mnt_parent == new.mnt)
2491 break;
2492 tmp = tmp->mnt_parent;
2494 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2495 goto out3;
2496 } else if (!is_subdir(old.dentry, new.dentry))
2497 goto out3;
2498 detach_mnt(new.mnt, &parent_path);
2499 detach_mnt(root.mnt, &root_parent);
2500 /* mount old root on put_old */
2501 attach_mnt(root.mnt, &old);
2502 /* mount new_root on / */
2503 attach_mnt(new.mnt, &root_parent);
2504 touch_mnt_namespace(current->nsproxy->mnt_ns);
2505 br_write_unlock(vfsmount_lock);
2506 chroot_fs_refs(&root, &new);
2508 error = 0;
2509 path_put(&root_parent);
2510 path_put(&parent_path);
2511 out2:
2512 mutex_unlock(&old.dentry->d_inode->i_mutex);
2513 up_write(&namespace_sem);
2514 path_put(&root);
2515 path_put(&old);
2516 out1:
2517 path_put(&new);
2518 out0:
2519 return error;
2520 out3:
2521 br_write_unlock(vfsmount_lock);
2522 goto out2;
2525 static void __init init_mount_tree(void)
2527 struct vfsmount *mnt;
2528 struct mnt_namespace *ns;
2529 struct path root;
2531 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2532 if (IS_ERR(mnt))
2533 panic("Can't create rootfs");
2535 ns = create_mnt_ns(mnt);
2536 if (IS_ERR(ns))
2537 panic("Can't allocate initial namespace");
2539 init_task.nsproxy->mnt_ns = ns;
2540 get_mnt_ns(ns);
2542 root.mnt = ns->root;
2543 root.dentry = ns->root->mnt_root;
2545 set_fs_pwd(current->fs, &root);
2546 set_fs_root(current->fs, &root);
2549 void __init mnt_init(void)
2551 unsigned u;
2552 int err;
2554 init_rwsem(&namespace_sem);
2556 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2557 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2559 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2561 if (!mount_hashtable)
2562 panic("Failed to allocate mount hash table\n");
2564 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2566 for (u = 0; u < HASH_SIZE; u++)
2567 INIT_LIST_HEAD(&mount_hashtable[u]);
2569 br_lock_init(vfsmount_lock);
2571 err = sysfs_init();
2572 if (err)
2573 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2574 __func__, err);
2575 fs_kobj = kobject_create_and_add("fs", NULL);
2576 if (!fs_kobj)
2577 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2578 init_rootfs();
2579 init_mount_tree();
2582 void put_mnt_ns(struct mnt_namespace *ns)
2584 LIST_HEAD(umount_list);
2586 if (!atomic_dec_and_test(&ns->count))
2587 return;
2588 down_write(&namespace_sem);
2589 br_write_lock(vfsmount_lock);
2590 umount_tree(ns->root, 0, &umount_list);
2591 br_write_unlock(vfsmount_lock);
2592 up_write(&namespace_sem);
2593 release_mounts(&umount_list);
2594 kfree(ns);
2596 EXPORT_SYMBOL(put_mnt_ns);