BKL: remove extraneous #include <smp_lock.h>
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / namespace.c
blob3dbfc072ec70aa0099b79b3ff1040455543c46ca
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;
141 struct vfsmount *alloc_vfsmnt(const char *name)
143 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
144 if (mnt) {
145 int err;
147 err = mnt_alloc_id(mnt);
148 if (err)
149 goto out_free_cache;
151 if (name) {
152 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
153 if (!mnt->mnt_devname)
154 goto out_free_id;
157 atomic_set(&mnt->mnt_count, 1);
158 INIT_LIST_HEAD(&mnt->mnt_hash);
159 INIT_LIST_HEAD(&mnt->mnt_child);
160 INIT_LIST_HEAD(&mnt->mnt_mounts);
161 INIT_LIST_HEAD(&mnt->mnt_list);
162 INIT_LIST_HEAD(&mnt->mnt_expire);
163 INIT_LIST_HEAD(&mnt->mnt_share);
164 INIT_LIST_HEAD(&mnt->mnt_slave_list);
165 INIT_LIST_HEAD(&mnt->mnt_slave);
166 #ifdef CONFIG_FSNOTIFY
167 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
168 #endif
169 #ifdef CONFIG_SMP
170 mnt->mnt_writers = alloc_percpu(int);
171 if (!mnt->mnt_writers)
172 goto out_free_devname;
173 #else
174 mnt->mnt_writers = 0;
175 #endif
177 return mnt;
179 #ifdef CONFIG_SMP
180 out_free_devname:
181 kfree(mnt->mnt_devname);
182 #endif
183 out_free_id:
184 mnt_free_id(mnt);
185 out_free_cache:
186 kmem_cache_free(mnt_cache, mnt);
187 return NULL;
191 * Most r/o checks on a fs are for operations that take
192 * discrete amounts of time, like a write() or unlink().
193 * We must keep track of when those operations start
194 * (for permission checks) and when they end, so that
195 * we can determine when writes are able to occur to
196 * a filesystem.
199 * __mnt_is_readonly: check whether a mount is read-only
200 * @mnt: the mount to check for its write status
202 * This shouldn't be used directly ouside of the VFS.
203 * It does not guarantee that the filesystem will stay
204 * r/w, just that it is right *now*. This can not and
205 * should not be used in place of IS_RDONLY(inode).
206 * mnt_want/drop_write() will _keep_ the filesystem
207 * r/w.
209 int __mnt_is_readonly(struct vfsmount *mnt)
211 if (mnt->mnt_flags & MNT_READONLY)
212 return 1;
213 if (mnt->mnt_sb->s_flags & MS_RDONLY)
214 return 1;
215 return 0;
217 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
219 static inline void inc_mnt_writers(struct vfsmount *mnt)
221 #ifdef CONFIG_SMP
222 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++;
223 #else
224 mnt->mnt_writers++;
225 #endif
228 static inline void dec_mnt_writers(struct vfsmount *mnt)
230 #ifdef CONFIG_SMP
231 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--;
232 #else
233 mnt->mnt_writers--;
234 #endif
237 static unsigned int count_mnt_writers(struct vfsmount *mnt)
239 #ifdef CONFIG_SMP
240 unsigned int count = 0;
241 int cpu;
243 for_each_possible_cpu(cpu) {
244 count += *per_cpu_ptr(mnt->mnt_writers, cpu);
247 return count;
248 #else
249 return mnt->mnt_writers;
250 #endif
254 * Most r/o checks on a fs are for operations that take
255 * discrete amounts of time, like a write() or unlink().
256 * We must keep track of when those operations start
257 * (for permission checks) and when they end, so that
258 * we can determine when writes are able to occur to
259 * a filesystem.
262 * mnt_want_write - get write access to a mount
263 * @mnt: the mount on which to take a write
265 * This tells the low-level filesystem that a write is
266 * about to be performed to it, and makes sure that
267 * writes are allowed before returning success. When
268 * the write operation is finished, mnt_drop_write()
269 * must be called. This is effectively a refcount.
271 int mnt_want_write(struct vfsmount *mnt)
273 int ret = 0;
275 preempt_disable();
276 inc_mnt_writers(mnt);
278 * The store to inc_mnt_writers must be visible before we pass
279 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
280 * incremented count after it has set MNT_WRITE_HOLD.
282 smp_mb();
283 while (mnt->mnt_flags & MNT_WRITE_HOLD)
284 cpu_relax();
286 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
287 * be set to match its requirements. So we must not load that until
288 * MNT_WRITE_HOLD is cleared.
290 smp_rmb();
291 if (__mnt_is_readonly(mnt)) {
292 dec_mnt_writers(mnt);
293 ret = -EROFS;
294 goto out;
296 out:
297 preempt_enable();
298 return ret;
300 EXPORT_SYMBOL_GPL(mnt_want_write);
303 * mnt_clone_write - get write access to a mount
304 * @mnt: the mount on which to take a write
306 * This is effectively like mnt_want_write, except
307 * it must only be used to take an extra write reference
308 * on a mountpoint that we already know has a write reference
309 * on it. This allows some optimisation.
311 * After finished, mnt_drop_write must be called as usual to
312 * drop the reference.
314 int mnt_clone_write(struct vfsmount *mnt)
316 /* superblock may be r/o */
317 if (__mnt_is_readonly(mnt))
318 return -EROFS;
319 preempt_disable();
320 inc_mnt_writers(mnt);
321 preempt_enable();
322 return 0;
324 EXPORT_SYMBOL_GPL(mnt_clone_write);
327 * mnt_want_write_file - get write access to a file's mount
328 * @file: the file who's mount on which to take a write
330 * This is like mnt_want_write, but it takes a file and can
331 * do some optimisations if the file is open for write already
333 int mnt_want_write_file(struct file *file)
335 struct inode *inode = file->f_dentry->d_inode;
336 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
337 return mnt_want_write(file->f_path.mnt);
338 else
339 return mnt_clone_write(file->f_path.mnt);
341 EXPORT_SYMBOL_GPL(mnt_want_write_file);
344 * mnt_drop_write - give up write access to a mount
345 * @mnt: the mount on which to give up write access
347 * Tells the low-level filesystem that we are done
348 * performing writes to it. Must be matched with
349 * mnt_want_write() call above.
351 void mnt_drop_write(struct vfsmount *mnt)
353 preempt_disable();
354 dec_mnt_writers(mnt);
355 preempt_enable();
357 EXPORT_SYMBOL_GPL(mnt_drop_write);
359 static int mnt_make_readonly(struct vfsmount *mnt)
361 int ret = 0;
363 br_write_lock(vfsmount_lock);
364 mnt->mnt_flags |= MNT_WRITE_HOLD;
366 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
367 * should be visible before we do.
369 smp_mb();
372 * With writers on hold, if this value is zero, then there are
373 * definitely no active writers (although held writers may subsequently
374 * increment the count, they'll have to wait, and decrement it after
375 * seeing MNT_READONLY).
377 * It is OK to have counter incremented on one CPU and decremented on
378 * another: the sum will add up correctly. The danger would be when we
379 * sum up each counter, if we read a counter before it is incremented,
380 * but then read another CPU's count which it has been subsequently
381 * decremented from -- we would see more decrements than we should.
382 * MNT_WRITE_HOLD protects against this scenario, because
383 * mnt_want_write first increments count, then smp_mb, then spins on
384 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
385 * we're counting up here.
387 if (count_mnt_writers(mnt) > 0)
388 ret = -EBUSY;
389 else
390 mnt->mnt_flags |= MNT_READONLY;
392 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
393 * that become unheld will see MNT_READONLY.
395 smp_wmb();
396 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
397 br_write_unlock(vfsmount_lock);
398 return ret;
401 static void __mnt_unmake_readonly(struct vfsmount *mnt)
403 br_write_lock(vfsmount_lock);
404 mnt->mnt_flags &= ~MNT_READONLY;
405 br_write_unlock(vfsmount_lock);
408 void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
410 mnt->mnt_sb = sb;
411 mnt->mnt_root = dget(sb->s_root);
414 EXPORT_SYMBOL(simple_set_mnt);
416 void free_vfsmnt(struct vfsmount *mnt)
418 kfree(mnt->mnt_devname);
419 mnt_free_id(mnt);
420 #ifdef CONFIG_SMP
421 free_percpu(mnt->mnt_writers);
422 #endif
423 kmem_cache_free(mnt_cache, mnt);
427 * find the first or last mount at @dentry on vfsmount @mnt depending on
428 * @dir. If @dir is set return the first mount else return the last mount.
429 * vfsmount_lock must be held for read or write.
431 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
432 int dir)
434 struct list_head *head = mount_hashtable + hash(mnt, dentry);
435 struct list_head *tmp = head;
436 struct vfsmount *p, *found = NULL;
438 for (;;) {
439 tmp = dir ? tmp->next : tmp->prev;
440 p = NULL;
441 if (tmp == head)
442 break;
443 p = list_entry(tmp, struct vfsmount, mnt_hash);
444 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
445 found = p;
446 break;
449 return found;
453 * lookup_mnt increments the ref count before returning
454 * the vfsmount struct.
456 struct vfsmount *lookup_mnt(struct path *path)
458 struct vfsmount *child_mnt;
460 br_read_lock(vfsmount_lock);
461 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
462 mntget(child_mnt);
463 br_read_unlock(vfsmount_lock);
464 return child_mnt;
467 static inline int check_mnt(struct vfsmount *mnt)
469 return mnt->mnt_ns == current->nsproxy->mnt_ns;
473 * vfsmount lock must be held for write
475 static void touch_mnt_namespace(struct mnt_namespace *ns)
477 if (ns) {
478 ns->event = ++event;
479 wake_up_interruptible(&ns->poll);
484 * vfsmount lock must be held for write
486 static void __touch_mnt_namespace(struct mnt_namespace *ns)
488 if (ns && ns->event != event) {
489 ns->event = event;
490 wake_up_interruptible(&ns->poll);
495 * vfsmount lock must be held for write
497 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
499 old_path->dentry = mnt->mnt_mountpoint;
500 old_path->mnt = mnt->mnt_parent;
501 mnt->mnt_parent = mnt;
502 mnt->mnt_mountpoint = mnt->mnt_root;
503 list_del_init(&mnt->mnt_child);
504 list_del_init(&mnt->mnt_hash);
505 old_path->dentry->d_mounted--;
509 * vfsmount lock must be held for write
511 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
512 struct vfsmount *child_mnt)
514 child_mnt->mnt_parent = mntget(mnt);
515 child_mnt->mnt_mountpoint = dget(dentry);
516 dentry->d_mounted++;
520 * vfsmount lock must be held for write
522 static void attach_mnt(struct vfsmount *mnt, struct path *path)
524 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
525 list_add_tail(&mnt->mnt_hash, mount_hashtable +
526 hash(path->mnt, path->dentry));
527 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
531 * vfsmount lock must be held for write
533 static void commit_tree(struct vfsmount *mnt)
535 struct vfsmount *parent = mnt->mnt_parent;
536 struct vfsmount *m;
537 LIST_HEAD(head);
538 struct mnt_namespace *n = parent->mnt_ns;
540 BUG_ON(parent == mnt);
542 list_add_tail(&head, &mnt->mnt_list);
543 list_for_each_entry(m, &head, mnt_list)
544 m->mnt_ns = n;
545 list_splice(&head, n->list.prev);
547 list_add_tail(&mnt->mnt_hash, mount_hashtable +
548 hash(parent, mnt->mnt_mountpoint));
549 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
550 touch_mnt_namespace(n);
553 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
555 struct list_head *next = p->mnt_mounts.next;
556 if (next == &p->mnt_mounts) {
557 while (1) {
558 if (p == root)
559 return NULL;
560 next = p->mnt_child.next;
561 if (next != &p->mnt_parent->mnt_mounts)
562 break;
563 p = p->mnt_parent;
566 return list_entry(next, struct vfsmount, mnt_child);
569 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
571 struct list_head *prev = p->mnt_mounts.prev;
572 while (prev != &p->mnt_mounts) {
573 p = list_entry(prev, struct vfsmount, mnt_child);
574 prev = p->mnt_mounts.prev;
576 return p;
579 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
580 int flag)
582 struct super_block *sb = old->mnt_sb;
583 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
585 if (mnt) {
586 if (flag & (CL_SLAVE | CL_PRIVATE))
587 mnt->mnt_group_id = 0; /* not a peer of original */
588 else
589 mnt->mnt_group_id = old->mnt_group_id;
591 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
592 int err = mnt_alloc_group_id(mnt);
593 if (err)
594 goto out_free;
597 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
598 atomic_inc(&sb->s_active);
599 mnt->mnt_sb = sb;
600 mnt->mnt_root = dget(root);
601 mnt->mnt_mountpoint = mnt->mnt_root;
602 mnt->mnt_parent = mnt;
604 if (flag & CL_SLAVE) {
605 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
606 mnt->mnt_master = old;
607 CLEAR_MNT_SHARED(mnt);
608 } else if (!(flag & CL_PRIVATE)) {
609 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
610 list_add(&mnt->mnt_share, &old->mnt_share);
611 if (IS_MNT_SLAVE(old))
612 list_add(&mnt->mnt_slave, &old->mnt_slave);
613 mnt->mnt_master = old->mnt_master;
615 if (flag & CL_MAKE_SHARED)
616 set_mnt_shared(mnt);
618 /* stick the duplicate mount on the same expiry list
619 * as the original if that was on one */
620 if (flag & CL_EXPIRE) {
621 if (!list_empty(&old->mnt_expire))
622 list_add(&mnt->mnt_expire, &old->mnt_expire);
625 return mnt;
627 out_free:
628 free_vfsmnt(mnt);
629 return NULL;
632 static inline void __mntput(struct vfsmount *mnt)
634 struct super_block *sb = mnt->mnt_sb;
636 * This probably indicates that somebody messed
637 * up a mnt_want/drop_write() pair. If this
638 * happens, the filesystem was probably unable
639 * to make r/w->r/o transitions.
642 * atomic_dec_and_lock() used to deal with ->mnt_count decrements
643 * provides barriers, so count_mnt_writers() below is safe. AV
645 WARN_ON(count_mnt_writers(mnt));
646 fsnotify_vfsmount_delete(mnt);
647 dput(mnt->mnt_root);
648 free_vfsmnt(mnt);
649 deactivate_super(sb);
652 void mntput_no_expire(struct vfsmount *mnt)
654 repeat:
655 if (atomic_add_unless(&mnt->mnt_count, -1, 1))
656 return;
657 br_write_lock(vfsmount_lock);
658 if (!atomic_dec_and_test(&mnt->mnt_count)) {
659 br_write_unlock(vfsmount_lock);
660 return;
662 if (likely(!mnt->mnt_pinned)) {
663 br_write_unlock(vfsmount_lock);
664 __mntput(mnt);
665 return;
667 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
668 mnt->mnt_pinned = 0;
669 br_write_unlock(vfsmount_lock);
670 acct_auto_close_mnt(mnt);
671 goto repeat;
673 EXPORT_SYMBOL(mntput_no_expire);
675 void mnt_pin(struct vfsmount *mnt)
677 br_write_lock(vfsmount_lock);
678 mnt->mnt_pinned++;
679 br_write_unlock(vfsmount_lock);
682 EXPORT_SYMBOL(mnt_pin);
684 void mnt_unpin(struct vfsmount *mnt)
686 br_write_lock(vfsmount_lock);
687 if (mnt->mnt_pinned) {
688 atomic_inc(&mnt->mnt_count);
689 mnt->mnt_pinned--;
691 br_write_unlock(vfsmount_lock);
694 EXPORT_SYMBOL(mnt_unpin);
696 static inline void mangle(struct seq_file *m, const char *s)
698 seq_escape(m, s, " \t\n\\");
702 * Simple .show_options callback for filesystems which don't want to
703 * implement more complex mount option showing.
705 * See also save_mount_options().
707 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
709 const char *options;
711 rcu_read_lock();
712 options = rcu_dereference(mnt->mnt_sb->s_options);
714 if (options != NULL && options[0]) {
715 seq_putc(m, ',');
716 mangle(m, options);
718 rcu_read_unlock();
720 return 0;
722 EXPORT_SYMBOL(generic_show_options);
725 * If filesystem uses generic_show_options(), this function should be
726 * called from the fill_super() callback.
728 * The .remount_fs callback usually needs to be handled in a special
729 * way, to make sure, that previous options are not overwritten if the
730 * remount fails.
732 * Also note, that if the filesystem's .remount_fs function doesn't
733 * reset all options to their default value, but changes only newly
734 * given options, then the displayed options will not reflect reality
735 * any more.
737 void save_mount_options(struct super_block *sb, char *options)
739 BUG_ON(sb->s_options);
740 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
742 EXPORT_SYMBOL(save_mount_options);
744 void replace_mount_options(struct super_block *sb, char *options)
746 char *old = sb->s_options;
747 rcu_assign_pointer(sb->s_options, options);
748 if (old) {
749 synchronize_rcu();
750 kfree(old);
753 EXPORT_SYMBOL(replace_mount_options);
755 #ifdef CONFIG_PROC_FS
756 /* iterator */
757 static void *m_start(struct seq_file *m, loff_t *pos)
759 struct proc_mounts *p = m->private;
761 down_read(&namespace_sem);
762 return seq_list_start(&p->ns->list, *pos);
765 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
767 struct proc_mounts *p = m->private;
769 return seq_list_next(v, &p->ns->list, pos);
772 static void m_stop(struct seq_file *m, void *v)
774 up_read(&namespace_sem);
777 int mnt_had_events(struct proc_mounts *p)
779 struct mnt_namespace *ns = p->ns;
780 int res = 0;
782 br_read_lock(vfsmount_lock);
783 if (p->event != ns->event) {
784 p->event = ns->event;
785 res = 1;
787 br_read_unlock(vfsmount_lock);
789 return res;
792 struct proc_fs_info {
793 int flag;
794 const char *str;
797 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
799 static const struct proc_fs_info fs_info[] = {
800 { MS_SYNCHRONOUS, ",sync" },
801 { MS_DIRSYNC, ",dirsync" },
802 { MS_MANDLOCK, ",mand" },
803 { 0, NULL }
805 const struct proc_fs_info *fs_infop;
807 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
808 if (sb->s_flags & fs_infop->flag)
809 seq_puts(m, fs_infop->str);
812 return security_sb_show_options(m, sb);
815 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
817 static const struct proc_fs_info mnt_info[] = {
818 { MNT_NOSUID, ",nosuid" },
819 { MNT_NODEV, ",nodev" },
820 { MNT_NOEXEC, ",noexec" },
821 { MNT_NOATIME, ",noatime" },
822 { MNT_NODIRATIME, ",nodiratime" },
823 { MNT_RELATIME, ",relatime" },
824 { 0, NULL }
826 const struct proc_fs_info *fs_infop;
828 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
829 if (mnt->mnt_flags & fs_infop->flag)
830 seq_puts(m, fs_infop->str);
834 static void show_type(struct seq_file *m, struct super_block *sb)
836 mangle(m, sb->s_type->name);
837 if (sb->s_subtype && sb->s_subtype[0]) {
838 seq_putc(m, '.');
839 mangle(m, sb->s_subtype);
843 static int show_vfsmnt(struct seq_file *m, void *v)
845 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
846 int err = 0;
847 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
849 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
850 seq_putc(m, ' ');
851 seq_path(m, &mnt_path, " \t\n\\");
852 seq_putc(m, ' ');
853 show_type(m, mnt->mnt_sb);
854 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
855 err = show_sb_opts(m, mnt->mnt_sb);
856 if (err)
857 goto out;
858 show_mnt_opts(m, mnt);
859 if (mnt->mnt_sb->s_op->show_options)
860 err = mnt->mnt_sb->s_op->show_options(m, mnt);
861 seq_puts(m, " 0 0\n");
862 out:
863 return err;
866 const struct seq_operations mounts_op = {
867 .start = m_start,
868 .next = m_next,
869 .stop = m_stop,
870 .show = show_vfsmnt
873 static int show_mountinfo(struct seq_file *m, void *v)
875 struct proc_mounts *p = m->private;
876 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
877 struct super_block *sb = mnt->mnt_sb;
878 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
879 struct path root = p->root;
880 int err = 0;
882 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
883 MAJOR(sb->s_dev), MINOR(sb->s_dev));
884 seq_dentry(m, mnt->mnt_root, " \t\n\\");
885 seq_putc(m, ' ');
886 seq_path_root(m, &mnt_path, &root, " \t\n\\");
887 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
889 * Mountpoint is outside root, discard that one. Ugly,
890 * but less so than trying to do that in iterator in a
891 * race-free way (due to renames).
893 return SEQ_SKIP;
895 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
896 show_mnt_opts(m, mnt);
898 /* Tagged fields ("foo:X" or "bar") */
899 if (IS_MNT_SHARED(mnt))
900 seq_printf(m, " shared:%i", mnt->mnt_group_id);
901 if (IS_MNT_SLAVE(mnt)) {
902 int master = mnt->mnt_master->mnt_group_id;
903 int dom = get_dominating_id(mnt, &p->root);
904 seq_printf(m, " master:%i", master);
905 if (dom && dom != master)
906 seq_printf(m, " propagate_from:%i", dom);
908 if (IS_MNT_UNBINDABLE(mnt))
909 seq_puts(m, " unbindable");
911 /* Filesystem specific data */
912 seq_puts(m, " - ");
913 show_type(m, sb);
914 seq_putc(m, ' ');
915 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
916 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
917 err = show_sb_opts(m, sb);
918 if (err)
919 goto out;
920 if (sb->s_op->show_options)
921 err = sb->s_op->show_options(m, mnt);
922 seq_putc(m, '\n');
923 out:
924 return err;
927 const struct seq_operations mountinfo_op = {
928 .start = m_start,
929 .next = m_next,
930 .stop = m_stop,
931 .show = show_mountinfo,
934 static int show_vfsstat(struct seq_file *m, void *v)
936 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
937 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
938 int err = 0;
940 /* device */
941 if (mnt->mnt_devname) {
942 seq_puts(m, "device ");
943 mangle(m, mnt->mnt_devname);
944 } else
945 seq_puts(m, "no device");
947 /* mount point */
948 seq_puts(m, " mounted on ");
949 seq_path(m, &mnt_path, " \t\n\\");
950 seq_putc(m, ' ');
952 /* file system type */
953 seq_puts(m, "with fstype ");
954 show_type(m, mnt->mnt_sb);
956 /* optional statistics */
957 if (mnt->mnt_sb->s_op->show_stats) {
958 seq_putc(m, ' ');
959 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
962 seq_putc(m, '\n');
963 return err;
966 const struct seq_operations mountstats_op = {
967 .start = m_start,
968 .next = m_next,
969 .stop = m_stop,
970 .show = show_vfsstat,
972 #endif /* CONFIG_PROC_FS */
975 * may_umount_tree - check if a mount tree is busy
976 * @mnt: root of mount tree
978 * This is called to check if a tree of mounts has any
979 * open files, pwds, chroots or sub mounts that are
980 * busy.
982 int may_umount_tree(struct vfsmount *mnt)
984 int actual_refs = 0;
985 int minimum_refs = 0;
986 struct vfsmount *p;
988 br_read_lock(vfsmount_lock);
989 for (p = mnt; p; p = next_mnt(p, mnt)) {
990 actual_refs += atomic_read(&p->mnt_count);
991 minimum_refs += 2;
993 br_read_unlock(vfsmount_lock);
995 if (actual_refs > minimum_refs)
996 return 0;
998 return 1;
1001 EXPORT_SYMBOL(may_umount_tree);
1004 * may_umount - check if a mount point is busy
1005 * @mnt: root of mount
1007 * This is called to check if a mount point has any
1008 * open files, pwds, chroots or sub mounts. If the
1009 * mount has sub mounts this will return busy
1010 * regardless of whether the sub mounts are busy.
1012 * Doesn't take quota and stuff into account. IOW, in some cases it will
1013 * give false negatives. The main reason why it's here is that we need
1014 * a non-destructive way to look for easily umountable filesystems.
1016 int may_umount(struct vfsmount *mnt)
1018 int ret = 1;
1019 down_read(&namespace_sem);
1020 br_read_lock(vfsmount_lock);
1021 if (propagate_mount_busy(mnt, 2))
1022 ret = 0;
1023 br_read_unlock(vfsmount_lock);
1024 up_read(&namespace_sem);
1025 return ret;
1028 EXPORT_SYMBOL(may_umount);
1030 void release_mounts(struct list_head *head)
1032 struct vfsmount *mnt;
1033 while (!list_empty(head)) {
1034 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1035 list_del_init(&mnt->mnt_hash);
1036 if (mnt->mnt_parent != mnt) {
1037 struct dentry *dentry;
1038 struct vfsmount *m;
1040 br_write_lock(vfsmount_lock);
1041 dentry = mnt->mnt_mountpoint;
1042 m = mnt->mnt_parent;
1043 mnt->mnt_mountpoint = mnt->mnt_root;
1044 mnt->mnt_parent = mnt;
1045 m->mnt_ghosts--;
1046 br_write_unlock(vfsmount_lock);
1047 dput(dentry);
1048 mntput(m);
1050 mntput(mnt);
1055 * vfsmount lock must be held for write
1056 * namespace_sem must be held for write
1058 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1060 struct vfsmount *p;
1062 for (p = mnt; p; p = next_mnt(p, mnt))
1063 list_move(&p->mnt_hash, kill);
1065 if (propagate)
1066 propagate_umount(kill);
1068 list_for_each_entry(p, kill, mnt_hash) {
1069 list_del_init(&p->mnt_expire);
1070 list_del_init(&p->mnt_list);
1071 __touch_mnt_namespace(p->mnt_ns);
1072 p->mnt_ns = NULL;
1073 list_del_init(&p->mnt_child);
1074 if (p->mnt_parent != p) {
1075 p->mnt_parent->mnt_ghosts++;
1076 p->mnt_mountpoint->d_mounted--;
1078 change_mnt_propagation(p, MS_PRIVATE);
1082 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1084 static int do_umount(struct vfsmount *mnt, int flags)
1086 struct super_block *sb = mnt->mnt_sb;
1087 int retval;
1088 LIST_HEAD(umount_list);
1090 retval = security_sb_umount(mnt, flags);
1091 if (retval)
1092 return retval;
1095 * Allow userspace to request a mountpoint be expired rather than
1096 * unmounting unconditionally. Unmount only happens if:
1097 * (1) the mark is already set (the mark is cleared by mntput())
1098 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1100 if (flags & MNT_EXPIRE) {
1101 if (mnt == current->fs->root.mnt ||
1102 flags & (MNT_FORCE | MNT_DETACH))
1103 return -EINVAL;
1105 if (atomic_read(&mnt->mnt_count) != 2)
1106 return -EBUSY;
1108 if (!xchg(&mnt->mnt_expiry_mark, 1))
1109 return -EAGAIN;
1113 * If we may have to abort operations to get out of this
1114 * mount, and they will themselves hold resources we must
1115 * allow the fs to do things. In the Unix tradition of
1116 * 'Gee thats tricky lets do it in userspace' the umount_begin
1117 * might fail to complete on the first run through as other tasks
1118 * must return, and the like. Thats for the mount program to worry
1119 * about for the moment.
1122 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1123 sb->s_op->umount_begin(sb);
1127 * No sense to grab the lock for this test, but test itself looks
1128 * somewhat bogus. Suggestions for better replacement?
1129 * Ho-hum... In principle, we might treat that as umount + switch
1130 * to rootfs. GC would eventually take care of the old vfsmount.
1131 * Actually it makes sense, especially if rootfs would contain a
1132 * /reboot - static binary that would close all descriptors and
1133 * call reboot(9). Then init(8) could umount root and exec /reboot.
1135 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1137 * Special case for "unmounting" root ...
1138 * we just try to remount it readonly.
1140 down_write(&sb->s_umount);
1141 if (!(sb->s_flags & MS_RDONLY))
1142 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1143 up_write(&sb->s_umount);
1144 return retval;
1147 down_write(&namespace_sem);
1148 br_write_lock(vfsmount_lock);
1149 event++;
1151 if (!(flags & MNT_DETACH))
1152 shrink_submounts(mnt, &umount_list);
1154 retval = -EBUSY;
1155 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1156 if (!list_empty(&mnt->mnt_list))
1157 umount_tree(mnt, 1, &umount_list);
1158 retval = 0;
1160 br_write_unlock(vfsmount_lock);
1161 up_write(&namespace_sem);
1162 release_mounts(&umount_list);
1163 return retval;
1167 * Now umount can handle mount points as well as block devices.
1168 * This is important for filesystems which use unnamed block devices.
1170 * We now support a flag for forced unmount like the other 'big iron'
1171 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1174 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1176 struct path path;
1177 int retval;
1178 int lookup_flags = 0;
1180 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1181 return -EINVAL;
1183 if (!(flags & UMOUNT_NOFOLLOW))
1184 lookup_flags |= LOOKUP_FOLLOW;
1186 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1187 if (retval)
1188 goto out;
1189 retval = -EINVAL;
1190 if (path.dentry != path.mnt->mnt_root)
1191 goto dput_and_out;
1192 if (!check_mnt(path.mnt))
1193 goto dput_and_out;
1195 retval = -EPERM;
1196 if (!capable(CAP_SYS_ADMIN))
1197 goto dput_and_out;
1199 retval = do_umount(path.mnt, flags);
1200 dput_and_out:
1201 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1202 dput(path.dentry);
1203 mntput_no_expire(path.mnt);
1204 out:
1205 return retval;
1208 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1211 * The 2.0 compatible umount. No flags.
1213 SYSCALL_DEFINE1(oldumount, char __user *, name)
1215 return sys_umount(name, 0);
1218 #endif
1220 static int mount_is_safe(struct path *path)
1222 if (capable(CAP_SYS_ADMIN))
1223 return 0;
1224 return -EPERM;
1225 #ifdef notyet
1226 if (S_ISLNK(path->dentry->d_inode->i_mode))
1227 return -EPERM;
1228 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1229 if (current_uid() != path->dentry->d_inode->i_uid)
1230 return -EPERM;
1232 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1233 return -EPERM;
1234 return 0;
1235 #endif
1238 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1239 int flag)
1241 struct vfsmount *res, *p, *q, *r, *s;
1242 struct path path;
1244 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1245 return NULL;
1247 res = q = clone_mnt(mnt, dentry, flag);
1248 if (!q)
1249 goto Enomem;
1250 q->mnt_mountpoint = mnt->mnt_mountpoint;
1252 p = mnt;
1253 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1254 if (!is_subdir(r->mnt_mountpoint, dentry))
1255 continue;
1257 for (s = r; s; s = next_mnt(s, r)) {
1258 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1259 s = skip_mnt_tree(s);
1260 continue;
1262 while (p != s->mnt_parent) {
1263 p = p->mnt_parent;
1264 q = q->mnt_parent;
1266 p = s;
1267 path.mnt = q;
1268 path.dentry = p->mnt_mountpoint;
1269 q = clone_mnt(p, p->mnt_root, flag);
1270 if (!q)
1271 goto Enomem;
1272 br_write_lock(vfsmount_lock);
1273 list_add_tail(&q->mnt_list, &res->mnt_list);
1274 attach_mnt(q, &path);
1275 br_write_unlock(vfsmount_lock);
1278 return res;
1279 Enomem:
1280 if (res) {
1281 LIST_HEAD(umount_list);
1282 br_write_lock(vfsmount_lock);
1283 umount_tree(res, 0, &umount_list);
1284 br_write_unlock(vfsmount_lock);
1285 release_mounts(&umount_list);
1287 return NULL;
1290 struct vfsmount *collect_mounts(struct path *path)
1292 struct vfsmount *tree;
1293 down_write(&namespace_sem);
1294 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1295 up_write(&namespace_sem);
1296 return tree;
1299 void drop_collected_mounts(struct vfsmount *mnt)
1301 LIST_HEAD(umount_list);
1302 down_write(&namespace_sem);
1303 br_write_lock(vfsmount_lock);
1304 umount_tree(mnt, 0, &umount_list);
1305 br_write_unlock(vfsmount_lock);
1306 up_write(&namespace_sem);
1307 release_mounts(&umount_list);
1310 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1311 struct vfsmount *root)
1313 struct vfsmount *mnt;
1314 int res = f(root, arg);
1315 if (res)
1316 return res;
1317 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1318 res = f(mnt, arg);
1319 if (res)
1320 return res;
1322 return 0;
1325 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1327 struct vfsmount *p;
1329 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1330 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1331 mnt_release_group_id(p);
1335 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1337 struct vfsmount *p;
1339 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1340 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1341 int err = mnt_alloc_group_id(p);
1342 if (err) {
1343 cleanup_group_ids(mnt, p);
1344 return err;
1349 return 0;
1353 * @source_mnt : mount tree to be attached
1354 * @nd : place the mount tree @source_mnt is attached
1355 * @parent_nd : if non-null, detach the source_mnt from its parent and
1356 * store the parent mount and mountpoint dentry.
1357 * (done when source_mnt is moved)
1359 * NOTE: in the table below explains the semantics when a source mount
1360 * of a given type is attached to a destination mount of a given type.
1361 * ---------------------------------------------------------------------------
1362 * | BIND MOUNT OPERATION |
1363 * |**************************************************************************
1364 * | source-->| shared | private | slave | unbindable |
1365 * | dest | | | | |
1366 * | | | | | | |
1367 * | v | | | | |
1368 * |**************************************************************************
1369 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1370 * | | | | | |
1371 * |non-shared| shared (+) | private | slave (*) | invalid |
1372 * ***************************************************************************
1373 * A bind operation clones the source mount and mounts the clone on the
1374 * destination mount.
1376 * (++) the cloned mount is propagated to all the mounts in the propagation
1377 * tree of the destination mount and the cloned mount is added to
1378 * the peer group of the source mount.
1379 * (+) the cloned mount is created under the destination mount and is marked
1380 * as shared. The cloned mount is added to the peer group of the source
1381 * mount.
1382 * (+++) the mount is propagated to all the mounts in the propagation tree
1383 * of the destination mount and the cloned mount is made slave
1384 * of the same master as that of the source mount. The cloned mount
1385 * is marked as 'shared and slave'.
1386 * (*) the cloned mount is made a slave of the same master as that of the
1387 * source mount.
1389 * ---------------------------------------------------------------------------
1390 * | MOVE MOUNT OPERATION |
1391 * |**************************************************************************
1392 * | source-->| shared | private | slave | unbindable |
1393 * | dest | | | | |
1394 * | | | | | | |
1395 * | v | | | | |
1396 * |**************************************************************************
1397 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1398 * | | | | | |
1399 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1400 * ***************************************************************************
1402 * (+) the mount is moved to the destination. And is then propagated to
1403 * all the mounts in the propagation tree of the destination mount.
1404 * (+*) the mount is moved to the destination.
1405 * (+++) the mount is moved to the destination and is then propagated to
1406 * all the mounts belonging to the destination mount's propagation tree.
1407 * the mount is marked as 'shared and slave'.
1408 * (*) the mount continues to be a slave at the new location.
1410 * if the source mount is a tree, the operations explained above is
1411 * applied to each mount in the tree.
1412 * Must be called without spinlocks held, since this function can sleep
1413 * in allocations.
1415 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1416 struct path *path, struct path *parent_path)
1418 LIST_HEAD(tree_list);
1419 struct vfsmount *dest_mnt = path->mnt;
1420 struct dentry *dest_dentry = path->dentry;
1421 struct vfsmount *child, *p;
1422 int err;
1424 if (IS_MNT_SHARED(dest_mnt)) {
1425 err = invent_group_ids(source_mnt, true);
1426 if (err)
1427 goto out;
1429 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1430 if (err)
1431 goto out_cleanup_ids;
1433 br_write_lock(vfsmount_lock);
1435 if (IS_MNT_SHARED(dest_mnt)) {
1436 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1437 set_mnt_shared(p);
1439 if (parent_path) {
1440 detach_mnt(source_mnt, parent_path);
1441 attach_mnt(source_mnt, path);
1442 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1443 } else {
1444 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1445 commit_tree(source_mnt);
1448 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1449 list_del_init(&child->mnt_hash);
1450 commit_tree(child);
1452 br_write_unlock(vfsmount_lock);
1454 return 0;
1456 out_cleanup_ids:
1457 if (IS_MNT_SHARED(dest_mnt))
1458 cleanup_group_ids(source_mnt, NULL);
1459 out:
1460 return err;
1463 static int graft_tree(struct vfsmount *mnt, struct path *path)
1465 int err;
1466 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1467 return -EINVAL;
1469 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1470 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1471 return -ENOTDIR;
1473 err = -ENOENT;
1474 mutex_lock(&path->dentry->d_inode->i_mutex);
1475 if (cant_mount(path->dentry))
1476 goto out_unlock;
1478 if (!d_unlinked(path->dentry))
1479 err = attach_recursive_mnt(mnt, path, NULL);
1480 out_unlock:
1481 mutex_unlock(&path->dentry->d_inode->i_mutex);
1482 return err;
1486 * Sanity check the flags to change_mnt_propagation.
1489 static int flags_to_propagation_type(int flags)
1491 int type = flags & ~MS_REC;
1493 /* Fail if any non-propagation flags are set */
1494 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1495 return 0;
1496 /* Only one propagation flag should be set */
1497 if (!is_power_of_2(type))
1498 return 0;
1499 return type;
1503 * recursively change the type of the mountpoint.
1505 static int do_change_type(struct path *path, int flag)
1507 struct vfsmount *m, *mnt = path->mnt;
1508 int recurse = flag & MS_REC;
1509 int type;
1510 int err = 0;
1512 if (!capable(CAP_SYS_ADMIN))
1513 return -EPERM;
1515 if (path->dentry != path->mnt->mnt_root)
1516 return -EINVAL;
1518 type = flags_to_propagation_type(flag);
1519 if (!type)
1520 return -EINVAL;
1522 down_write(&namespace_sem);
1523 if (type == MS_SHARED) {
1524 err = invent_group_ids(mnt, recurse);
1525 if (err)
1526 goto out_unlock;
1529 br_write_lock(vfsmount_lock);
1530 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1531 change_mnt_propagation(m, type);
1532 br_write_unlock(vfsmount_lock);
1534 out_unlock:
1535 up_write(&namespace_sem);
1536 return err;
1540 * do loopback mount.
1542 static int do_loopback(struct path *path, char *old_name,
1543 int recurse)
1545 struct path old_path;
1546 struct vfsmount *mnt = NULL;
1547 int err = mount_is_safe(path);
1548 if (err)
1549 return err;
1550 if (!old_name || !*old_name)
1551 return -EINVAL;
1552 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1553 if (err)
1554 return err;
1556 down_write(&namespace_sem);
1557 err = -EINVAL;
1558 if (IS_MNT_UNBINDABLE(old_path.mnt))
1559 goto out;
1561 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1562 goto out;
1564 err = -ENOMEM;
1565 if (recurse)
1566 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1567 else
1568 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1570 if (!mnt)
1571 goto out;
1573 err = graft_tree(mnt, path);
1574 if (err) {
1575 LIST_HEAD(umount_list);
1577 br_write_lock(vfsmount_lock);
1578 umount_tree(mnt, 0, &umount_list);
1579 br_write_unlock(vfsmount_lock);
1580 release_mounts(&umount_list);
1583 out:
1584 up_write(&namespace_sem);
1585 path_put(&old_path);
1586 return err;
1589 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1591 int error = 0;
1592 int readonly_request = 0;
1594 if (ms_flags & MS_RDONLY)
1595 readonly_request = 1;
1596 if (readonly_request == __mnt_is_readonly(mnt))
1597 return 0;
1599 if (readonly_request)
1600 error = mnt_make_readonly(mnt);
1601 else
1602 __mnt_unmake_readonly(mnt);
1603 return error;
1607 * change filesystem flags. dir should be a physical root of filesystem.
1608 * If you've mounted a non-root directory somewhere and want to do remount
1609 * on it - tough luck.
1611 static int do_remount(struct path *path, int flags, int mnt_flags,
1612 void *data)
1614 int err;
1615 struct super_block *sb = path->mnt->mnt_sb;
1617 if (!capable(CAP_SYS_ADMIN))
1618 return -EPERM;
1620 if (!check_mnt(path->mnt))
1621 return -EINVAL;
1623 if (path->dentry != path->mnt->mnt_root)
1624 return -EINVAL;
1626 down_write(&sb->s_umount);
1627 if (flags & MS_BIND)
1628 err = change_mount_flags(path->mnt, flags);
1629 else
1630 err = do_remount_sb(sb, flags, data, 0);
1631 if (!err) {
1632 br_write_lock(vfsmount_lock);
1633 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1634 path->mnt->mnt_flags = mnt_flags;
1635 br_write_unlock(vfsmount_lock);
1637 up_write(&sb->s_umount);
1638 if (!err) {
1639 br_write_lock(vfsmount_lock);
1640 touch_mnt_namespace(path->mnt->mnt_ns);
1641 br_write_unlock(vfsmount_lock);
1643 return err;
1646 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1648 struct vfsmount *p;
1649 for (p = mnt; p; p = next_mnt(p, mnt)) {
1650 if (IS_MNT_UNBINDABLE(p))
1651 return 1;
1653 return 0;
1656 static int do_move_mount(struct path *path, char *old_name)
1658 struct path old_path, parent_path;
1659 struct vfsmount *p;
1660 int err = 0;
1661 if (!capable(CAP_SYS_ADMIN))
1662 return -EPERM;
1663 if (!old_name || !*old_name)
1664 return -EINVAL;
1665 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1666 if (err)
1667 return err;
1669 down_write(&namespace_sem);
1670 while (d_mountpoint(path->dentry) &&
1671 follow_down(path))
1673 err = -EINVAL;
1674 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1675 goto out;
1677 err = -ENOENT;
1678 mutex_lock(&path->dentry->d_inode->i_mutex);
1679 if (cant_mount(path->dentry))
1680 goto out1;
1682 if (d_unlinked(path->dentry))
1683 goto out1;
1685 err = -EINVAL;
1686 if (old_path.dentry != old_path.mnt->mnt_root)
1687 goto out1;
1689 if (old_path.mnt == old_path.mnt->mnt_parent)
1690 goto out1;
1692 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1693 S_ISDIR(old_path.dentry->d_inode->i_mode))
1694 goto out1;
1696 * Don't move a mount residing in a shared parent.
1698 if (old_path.mnt->mnt_parent &&
1699 IS_MNT_SHARED(old_path.mnt->mnt_parent))
1700 goto out1;
1702 * Don't move a mount tree containing unbindable mounts to a destination
1703 * mount which is shared.
1705 if (IS_MNT_SHARED(path->mnt) &&
1706 tree_contains_unbindable(old_path.mnt))
1707 goto out1;
1708 err = -ELOOP;
1709 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1710 if (p == old_path.mnt)
1711 goto out1;
1713 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1714 if (err)
1715 goto out1;
1717 /* if the mount is moved, it should no longer be expire
1718 * automatically */
1719 list_del_init(&old_path.mnt->mnt_expire);
1720 out1:
1721 mutex_unlock(&path->dentry->d_inode->i_mutex);
1722 out:
1723 up_write(&namespace_sem);
1724 if (!err)
1725 path_put(&parent_path);
1726 path_put(&old_path);
1727 return err;
1731 * create a new mount for userspace and request it to be added into the
1732 * namespace's tree
1734 static int do_new_mount(struct path *path, char *type, int flags,
1735 int mnt_flags, char *name, void *data)
1737 struct vfsmount *mnt;
1739 if (!type)
1740 return -EINVAL;
1742 /* we need capabilities... */
1743 if (!capable(CAP_SYS_ADMIN))
1744 return -EPERM;
1746 mnt = do_kern_mount(type, flags, name, data);
1747 if (IS_ERR(mnt))
1748 return PTR_ERR(mnt);
1750 return do_add_mount(mnt, path, mnt_flags, NULL);
1754 * add a mount into a namespace's mount tree
1755 * - provide the option of adding the new mount to an expiration list
1757 int do_add_mount(struct vfsmount *newmnt, struct path *path,
1758 int mnt_flags, struct list_head *fslist)
1760 int err;
1762 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1764 down_write(&namespace_sem);
1765 /* Something was mounted here while we slept */
1766 while (d_mountpoint(path->dentry) &&
1767 follow_down(path))
1769 err = -EINVAL;
1770 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1771 goto unlock;
1773 /* Refuse the same filesystem on the same mount point */
1774 err = -EBUSY;
1775 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1776 path->mnt->mnt_root == path->dentry)
1777 goto unlock;
1779 err = -EINVAL;
1780 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1781 goto unlock;
1783 newmnt->mnt_flags = mnt_flags;
1784 if ((err = graft_tree(newmnt, path)))
1785 goto unlock;
1787 if (fslist) /* add to the specified expiration list */
1788 list_add_tail(&newmnt->mnt_expire, fslist);
1790 up_write(&namespace_sem);
1791 return 0;
1793 unlock:
1794 up_write(&namespace_sem);
1795 mntput(newmnt);
1796 return err;
1799 EXPORT_SYMBOL_GPL(do_add_mount);
1802 * process a list of expirable mountpoints with the intent of discarding any
1803 * mountpoints that aren't in use and haven't been touched since last we came
1804 * here
1806 void mark_mounts_for_expiry(struct list_head *mounts)
1808 struct vfsmount *mnt, *next;
1809 LIST_HEAD(graveyard);
1810 LIST_HEAD(umounts);
1812 if (list_empty(mounts))
1813 return;
1815 down_write(&namespace_sem);
1816 br_write_lock(vfsmount_lock);
1818 /* extract from the expiration list every vfsmount that matches the
1819 * following criteria:
1820 * - only referenced by its parent vfsmount
1821 * - still marked for expiry (marked on the last call here; marks are
1822 * cleared by mntput())
1824 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1825 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1826 propagate_mount_busy(mnt, 1))
1827 continue;
1828 list_move(&mnt->mnt_expire, &graveyard);
1830 while (!list_empty(&graveyard)) {
1831 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1832 touch_mnt_namespace(mnt->mnt_ns);
1833 umount_tree(mnt, 1, &umounts);
1835 br_write_unlock(vfsmount_lock);
1836 up_write(&namespace_sem);
1838 release_mounts(&umounts);
1841 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1844 * Ripoff of 'select_parent()'
1846 * search the list of submounts for a given mountpoint, and move any
1847 * shrinkable submounts to the 'graveyard' list.
1849 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1851 struct vfsmount *this_parent = parent;
1852 struct list_head *next;
1853 int found = 0;
1855 repeat:
1856 next = this_parent->mnt_mounts.next;
1857 resume:
1858 while (next != &this_parent->mnt_mounts) {
1859 struct list_head *tmp = next;
1860 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1862 next = tmp->next;
1863 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1864 continue;
1866 * Descend a level if the d_mounts list is non-empty.
1868 if (!list_empty(&mnt->mnt_mounts)) {
1869 this_parent = mnt;
1870 goto repeat;
1873 if (!propagate_mount_busy(mnt, 1)) {
1874 list_move_tail(&mnt->mnt_expire, graveyard);
1875 found++;
1879 * All done at this level ... ascend and resume the search
1881 if (this_parent != parent) {
1882 next = this_parent->mnt_child.next;
1883 this_parent = this_parent->mnt_parent;
1884 goto resume;
1886 return found;
1890 * process a list of expirable mountpoints with the intent of discarding any
1891 * submounts of a specific parent mountpoint
1893 * vfsmount_lock must be held for write
1895 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1897 LIST_HEAD(graveyard);
1898 struct vfsmount *m;
1900 /* extract submounts of 'mountpoint' from the expiration list */
1901 while (select_submounts(mnt, &graveyard)) {
1902 while (!list_empty(&graveyard)) {
1903 m = list_first_entry(&graveyard, struct vfsmount,
1904 mnt_expire);
1905 touch_mnt_namespace(m->mnt_ns);
1906 umount_tree(m, 1, umounts);
1912 * Some copy_from_user() implementations do not return the exact number of
1913 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1914 * Note that this function differs from copy_from_user() in that it will oops
1915 * on bad values of `to', rather than returning a short copy.
1917 static long exact_copy_from_user(void *to, const void __user * from,
1918 unsigned long n)
1920 char *t = to;
1921 const char __user *f = from;
1922 char c;
1924 if (!access_ok(VERIFY_READ, from, n))
1925 return n;
1927 while (n) {
1928 if (__get_user(c, f)) {
1929 memset(t, 0, n);
1930 break;
1932 *t++ = c;
1933 f++;
1934 n--;
1936 return n;
1939 int copy_mount_options(const void __user * data, unsigned long *where)
1941 int i;
1942 unsigned long page;
1943 unsigned long size;
1945 *where = 0;
1946 if (!data)
1947 return 0;
1949 if (!(page = __get_free_page(GFP_KERNEL)))
1950 return -ENOMEM;
1952 /* We only care that *some* data at the address the user
1953 * gave us is valid. Just in case, we'll zero
1954 * the remainder of the page.
1956 /* copy_from_user cannot cross TASK_SIZE ! */
1957 size = TASK_SIZE - (unsigned long)data;
1958 if (size > PAGE_SIZE)
1959 size = PAGE_SIZE;
1961 i = size - exact_copy_from_user((void *)page, data, size);
1962 if (!i) {
1963 free_page(page);
1964 return -EFAULT;
1966 if (i != PAGE_SIZE)
1967 memset((char *)page + i, 0, PAGE_SIZE - i);
1968 *where = page;
1969 return 0;
1972 int copy_mount_string(const void __user *data, char **where)
1974 char *tmp;
1976 if (!data) {
1977 *where = NULL;
1978 return 0;
1981 tmp = strndup_user(data, PAGE_SIZE);
1982 if (IS_ERR(tmp))
1983 return PTR_ERR(tmp);
1985 *where = tmp;
1986 return 0;
1990 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1991 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1993 * data is a (void *) that can point to any structure up to
1994 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1995 * information (or be NULL).
1997 * Pre-0.97 versions of mount() didn't have a flags word.
1998 * When the flags word was introduced its top half was required
1999 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2000 * Therefore, if this magic number is present, it carries no information
2001 * and must be discarded.
2003 long do_mount(char *dev_name, char *dir_name, char *type_page,
2004 unsigned long flags, void *data_page)
2006 struct path path;
2007 int retval = 0;
2008 int mnt_flags = 0;
2010 /* Discard magic */
2011 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2012 flags &= ~MS_MGC_MSK;
2014 /* Basic sanity checks */
2016 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2017 return -EINVAL;
2019 if (data_page)
2020 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2022 /* ... and get the mountpoint */
2023 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2024 if (retval)
2025 return retval;
2027 retval = security_sb_mount(dev_name, &path,
2028 type_page, flags, data_page);
2029 if (retval)
2030 goto dput_out;
2032 /* Default to relatime unless overriden */
2033 if (!(flags & MS_NOATIME))
2034 mnt_flags |= MNT_RELATIME;
2036 /* Separate the per-mountpoint flags */
2037 if (flags & MS_NOSUID)
2038 mnt_flags |= MNT_NOSUID;
2039 if (flags & MS_NODEV)
2040 mnt_flags |= MNT_NODEV;
2041 if (flags & MS_NOEXEC)
2042 mnt_flags |= MNT_NOEXEC;
2043 if (flags & MS_NOATIME)
2044 mnt_flags |= MNT_NOATIME;
2045 if (flags & MS_NODIRATIME)
2046 mnt_flags |= MNT_NODIRATIME;
2047 if (flags & MS_STRICTATIME)
2048 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2049 if (flags & MS_RDONLY)
2050 mnt_flags |= MNT_READONLY;
2052 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2053 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2054 MS_STRICTATIME);
2056 if (flags & MS_REMOUNT)
2057 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2058 data_page);
2059 else if (flags & MS_BIND)
2060 retval = do_loopback(&path, dev_name, flags & MS_REC);
2061 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2062 retval = do_change_type(&path, flags);
2063 else if (flags & MS_MOVE)
2064 retval = do_move_mount(&path, dev_name);
2065 else
2066 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2067 dev_name, data_page);
2068 dput_out:
2069 path_put(&path);
2070 return retval;
2073 static struct mnt_namespace *alloc_mnt_ns(void)
2075 struct mnt_namespace *new_ns;
2077 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2078 if (!new_ns)
2079 return ERR_PTR(-ENOMEM);
2080 atomic_set(&new_ns->count, 1);
2081 new_ns->root = NULL;
2082 INIT_LIST_HEAD(&new_ns->list);
2083 init_waitqueue_head(&new_ns->poll);
2084 new_ns->event = 0;
2085 return new_ns;
2089 * Allocate a new namespace structure and populate it with contents
2090 * copied from the namespace of the passed in task structure.
2092 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2093 struct fs_struct *fs)
2095 struct mnt_namespace *new_ns;
2096 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2097 struct vfsmount *p, *q;
2099 new_ns = alloc_mnt_ns();
2100 if (IS_ERR(new_ns))
2101 return new_ns;
2103 down_write(&namespace_sem);
2104 /* First pass: copy the tree topology */
2105 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2106 CL_COPY_ALL | CL_EXPIRE);
2107 if (!new_ns->root) {
2108 up_write(&namespace_sem);
2109 kfree(new_ns);
2110 return ERR_PTR(-ENOMEM);
2112 br_write_lock(vfsmount_lock);
2113 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2114 br_write_unlock(vfsmount_lock);
2117 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2118 * as belonging to new namespace. We have already acquired a private
2119 * fs_struct, so tsk->fs->lock is not needed.
2121 p = mnt_ns->root;
2122 q = new_ns->root;
2123 while (p) {
2124 q->mnt_ns = new_ns;
2125 if (fs) {
2126 if (p == fs->root.mnt) {
2127 rootmnt = p;
2128 fs->root.mnt = mntget(q);
2130 if (p == fs->pwd.mnt) {
2131 pwdmnt = p;
2132 fs->pwd.mnt = mntget(q);
2135 p = next_mnt(p, mnt_ns->root);
2136 q = next_mnt(q, new_ns->root);
2138 up_write(&namespace_sem);
2140 if (rootmnt)
2141 mntput(rootmnt);
2142 if (pwdmnt)
2143 mntput(pwdmnt);
2145 return new_ns;
2148 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2149 struct fs_struct *new_fs)
2151 struct mnt_namespace *new_ns;
2153 BUG_ON(!ns);
2154 get_mnt_ns(ns);
2156 if (!(flags & CLONE_NEWNS))
2157 return ns;
2159 new_ns = dup_mnt_ns(ns, new_fs);
2161 put_mnt_ns(ns);
2162 return new_ns;
2166 * create_mnt_ns - creates a private namespace and adds a root filesystem
2167 * @mnt: pointer to the new root filesystem mountpoint
2169 struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2171 struct mnt_namespace *new_ns;
2173 new_ns = alloc_mnt_ns();
2174 if (!IS_ERR(new_ns)) {
2175 mnt->mnt_ns = new_ns;
2176 new_ns->root = mnt;
2177 list_add(&new_ns->list, &new_ns->root->mnt_list);
2179 return new_ns;
2181 EXPORT_SYMBOL(create_mnt_ns);
2183 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2184 char __user *, type, unsigned long, flags, void __user *, data)
2186 int ret;
2187 char *kernel_type;
2188 char *kernel_dir;
2189 char *kernel_dev;
2190 unsigned long data_page;
2192 ret = copy_mount_string(type, &kernel_type);
2193 if (ret < 0)
2194 goto out_type;
2196 kernel_dir = getname(dir_name);
2197 if (IS_ERR(kernel_dir)) {
2198 ret = PTR_ERR(kernel_dir);
2199 goto out_dir;
2202 ret = copy_mount_string(dev_name, &kernel_dev);
2203 if (ret < 0)
2204 goto out_dev;
2206 ret = copy_mount_options(data, &data_page);
2207 if (ret < 0)
2208 goto out_data;
2210 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2211 (void *) data_page);
2213 free_page(data_page);
2214 out_data:
2215 kfree(kernel_dev);
2216 out_dev:
2217 putname(kernel_dir);
2218 out_dir:
2219 kfree(kernel_type);
2220 out_type:
2221 return ret;
2225 * pivot_root Semantics:
2226 * Moves the root file system of the current process to the directory put_old,
2227 * makes new_root as the new root file system of the current process, and sets
2228 * root/cwd of all processes which had them on the current root to new_root.
2230 * Restrictions:
2231 * The new_root and put_old must be directories, and must not be on the
2232 * same file system as the current process root. The put_old must be
2233 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2234 * pointed to by put_old must yield the same directory as new_root. No other
2235 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2237 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2238 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2239 * in this situation.
2241 * Notes:
2242 * - we don't move root/cwd if they are not at the root (reason: if something
2243 * cared enough to change them, it's probably wrong to force them elsewhere)
2244 * - it's okay to pick a root that isn't the root of a file system, e.g.
2245 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2246 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2247 * first.
2249 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2250 const char __user *, put_old)
2252 struct vfsmount *tmp;
2253 struct path new, old, parent_path, root_parent, root;
2254 int error;
2256 if (!capable(CAP_SYS_ADMIN))
2257 return -EPERM;
2259 error = user_path_dir(new_root, &new);
2260 if (error)
2261 goto out0;
2262 error = -EINVAL;
2263 if (!check_mnt(new.mnt))
2264 goto out1;
2266 error = user_path_dir(put_old, &old);
2267 if (error)
2268 goto out1;
2270 error = security_sb_pivotroot(&old, &new);
2271 if (error) {
2272 path_put(&old);
2273 goto out1;
2276 get_fs_root(current->fs, &root);
2277 down_write(&namespace_sem);
2278 mutex_lock(&old.dentry->d_inode->i_mutex);
2279 error = -EINVAL;
2280 if (IS_MNT_SHARED(old.mnt) ||
2281 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2282 IS_MNT_SHARED(root.mnt->mnt_parent))
2283 goto out2;
2284 if (!check_mnt(root.mnt))
2285 goto out2;
2286 error = -ENOENT;
2287 if (cant_mount(old.dentry))
2288 goto out2;
2289 if (d_unlinked(new.dentry))
2290 goto out2;
2291 if (d_unlinked(old.dentry))
2292 goto out2;
2293 error = -EBUSY;
2294 if (new.mnt == root.mnt ||
2295 old.mnt == root.mnt)
2296 goto out2; /* loop, on the same file system */
2297 error = -EINVAL;
2298 if (root.mnt->mnt_root != root.dentry)
2299 goto out2; /* not a mountpoint */
2300 if (root.mnt->mnt_parent == root.mnt)
2301 goto out2; /* not attached */
2302 if (new.mnt->mnt_root != new.dentry)
2303 goto out2; /* not a mountpoint */
2304 if (new.mnt->mnt_parent == new.mnt)
2305 goto out2; /* not attached */
2306 /* make sure we can reach put_old from new_root */
2307 tmp = old.mnt;
2308 br_write_lock(vfsmount_lock);
2309 if (tmp != new.mnt) {
2310 for (;;) {
2311 if (tmp->mnt_parent == tmp)
2312 goto out3; /* already mounted on put_old */
2313 if (tmp->mnt_parent == new.mnt)
2314 break;
2315 tmp = tmp->mnt_parent;
2317 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2318 goto out3;
2319 } else if (!is_subdir(old.dentry, new.dentry))
2320 goto out3;
2321 detach_mnt(new.mnt, &parent_path);
2322 detach_mnt(root.mnt, &root_parent);
2323 /* mount old root on put_old */
2324 attach_mnt(root.mnt, &old);
2325 /* mount new_root on / */
2326 attach_mnt(new.mnt, &root_parent);
2327 touch_mnt_namespace(current->nsproxy->mnt_ns);
2328 br_write_unlock(vfsmount_lock);
2329 chroot_fs_refs(&root, &new);
2330 error = 0;
2331 path_put(&root_parent);
2332 path_put(&parent_path);
2333 out2:
2334 mutex_unlock(&old.dentry->d_inode->i_mutex);
2335 up_write(&namespace_sem);
2336 path_put(&root);
2337 path_put(&old);
2338 out1:
2339 path_put(&new);
2340 out0:
2341 return error;
2342 out3:
2343 br_write_unlock(vfsmount_lock);
2344 goto out2;
2347 static void __init init_mount_tree(void)
2349 struct vfsmount *mnt;
2350 struct mnt_namespace *ns;
2351 struct path root;
2353 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2354 if (IS_ERR(mnt))
2355 panic("Can't create rootfs");
2356 ns = create_mnt_ns(mnt);
2357 if (IS_ERR(ns))
2358 panic("Can't allocate initial namespace");
2360 init_task.nsproxy->mnt_ns = ns;
2361 get_mnt_ns(ns);
2363 root.mnt = ns->root;
2364 root.dentry = ns->root->mnt_root;
2366 set_fs_pwd(current->fs, &root);
2367 set_fs_root(current->fs, &root);
2370 void __init mnt_init(void)
2372 unsigned u;
2373 int err;
2375 init_rwsem(&namespace_sem);
2377 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2378 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2380 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2382 if (!mount_hashtable)
2383 panic("Failed to allocate mount hash table\n");
2385 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2387 for (u = 0; u < HASH_SIZE; u++)
2388 INIT_LIST_HEAD(&mount_hashtable[u]);
2390 br_lock_init(vfsmount_lock);
2392 err = sysfs_init();
2393 if (err)
2394 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2395 __func__, err);
2396 fs_kobj = kobject_create_and_add("fs", NULL);
2397 if (!fs_kobj)
2398 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2399 init_rootfs();
2400 init_mount_tree();
2403 void put_mnt_ns(struct mnt_namespace *ns)
2405 LIST_HEAD(umount_list);
2407 if (!atomic_dec_and_test(&ns->count))
2408 return;
2409 down_write(&namespace_sem);
2410 br_write_lock(vfsmount_lock);
2411 umount_tree(ns->root, 0, &umount_list);
2412 br_write_unlock(vfsmount_lock);
2413 up_write(&namespace_sem);
2414 release_mounts(&umount_list);
2415 kfree(ns);
2417 EXPORT_SYMBOL(put_mnt_ns);