Merge branch 'linux-next' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad...
[zen-stable.git] / fs / namespace.c
blobcfc6d4448aa54bdc538131b1e53285a92bb20073
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 = 0;
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 static 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 this_cpu_add(mnt->mnt_pcp->mnt_count, 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 static void free_vfsmnt(struct vfsmount *mnt)
471 kfree(mnt->mnt_devname);
472 mnt_free_id(mnt);
473 #ifdef CONFIG_SMP
474 free_percpu(mnt->mnt_pcp);
475 #endif
476 kmem_cache_free(mnt_cache, mnt);
480 * find the first or last mount at @dentry on vfsmount @mnt depending on
481 * @dir. If @dir is set return the first mount else return the last mount.
482 * vfsmount_lock must be held for read or write.
484 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
485 int dir)
487 struct list_head *head = mount_hashtable + hash(mnt, dentry);
488 struct list_head *tmp = head;
489 struct vfsmount *p, *found = NULL;
491 for (;;) {
492 tmp = dir ? tmp->next : tmp->prev;
493 p = NULL;
494 if (tmp == head)
495 break;
496 p = list_entry(tmp, struct vfsmount, mnt_hash);
497 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
498 found = p;
499 break;
502 return found;
506 * lookup_mnt increments the ref count before returning
507 * the vfsmount struct.
509 struct vfsmount *lookup_mnt(struct path *path)
511 struct vfsmount *child_mnt;
513 br_read_lock(vfsmount_lock);
514 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
515 mntget(child_mnt);
516 br_read_unlock(vfsmount_lock);
517 return child_mnt;
520 static inline int check_mnt(struct vfsmount *mnt)
522 return mnt->mnt_ns == current->nsproxy->mnt_ns;
526 * vfsmount lock must be held for write
528 static void touch_mnt_namespace(struct mnt_namespace *ns)
530 if (ns) {
531 ns->event = ++event;
532 wake_up_interruptible(&ns->poll);
537 * vfsmount lock must be held for write
539 static void __touch_mnt_namespace(struct mnt_namespace *ns)
541 if (ns && ns->event != event) {
542 ns->event = event;
543 wake_up_interruptible(&ns->poll);
548 * Clear dentry's mounted state if it has no remaining mounts.
549 * vfsmount_lock must be held for write.
551 static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
553 unsigned u;
555 for (u = 0; u < HASH_SIZE; u++) {
556 struct vfsmount *p;
558 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
559 if (p->mnt_mountpoint == dentry)
560 return;
563 spin_lock(&dentry->d_lock);
564 dentry->d_flags &= ~DCACHE_MOUNTED;
565 spin_unlock(&dentry->d_lock);
569 * vfsmount lock must be held for write
571 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
573 old_path->dentry = mnt->mnt_mountpoint;
574 old_path->mnt = mnt->mnt_parent;
575 mnt->mnt_parent = mnt;
576 mnt->mnt_mountpoint = mnt->mnt_root;
577 list_del_init(&mnt->mnt_child);
578 list_del_init(&mnt->mnt_hash);
579 dentry_reset_mounted(old_path->mnt, old_path->dentry);
583 * vfsmount lock must be held for write
585 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
586 struct vfsmount *child_mnt)
588 child_mnt->mnt_parent = mntget(mnt);
589 child_mnt->mnt_mountpoint = dget(dentry);
590 spin_lock(&dentry->d_lock);
591 dentry->d_flags |= DCACHE_MOUNTED;
592 spin_unlock(&dentry->d_lock);
596 * vfsmount lock must be held for write
598 static void attach_mnt(struct vfsmount *mnt, struct path *path)
600 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
601 list_add_tail(&mnt->mnt_hash, mount_hashtable +
602 hash(path->mnt, path->dentry));
603 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
606 static inline void __mnt_make_longterm(struct vfsmount *mnt)
608 #ifdef CONFIG_SMP
609 atomic_inc(&mnt->mnt_longterm);
610 #endif
613 /* needs vfsmount lock for write */
614 static inline void __mnt_make_shortterm(struct vfsmount *mnt)
616 #ifdef CONFIG_SMP
617 atomic_dec(&mnt->mnt_longterm);
618 #endif
622 * vfsmount lock must be held for write
624 static void commit_tree(struct vfsmount *mnt)
626 struct vfsmount *parent = mnt->mnt_parent;
627 struct vfsmount *m;
628 LIST_HEAD(head);
629 struct mnt_namespace *n = parent->mnt_ns;
631 BUG_ON(parent == mnt);
633 list_add_tail(&head, &mnt->mnt_list);
634 list_for_each_entry(m, &head, mnt_list) {
635 m->mnt_ns = n;
636 __mnt_make_longterm(m);
639 list_splice(&head, n->list.prev);
641 list_add_tail(&mnt->mnt_hash, mount_hashtable +
642 hash(parent, mnt->mnt_mountpoint));
643 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
644 touch_mnt_namespace(n);
647 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
649 struct list_head *next = p->mnt_mounts.next;
650 if (next == &p->mnt_mounts) {
651 while (1) {
652 if (p == root)
653 return NULL;
654 next = p->mnt_child.next;
655 if (next != &p->mnt_parent->mnt_mounts)
656 break;
657 p = p->mnt_parent;
660 return list_entry(next, struct vfsmount, mnt_child);
663 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
665 struct list_head *prev = p->mnt_mounts.prev;
666 while (prev != &p->mnt_mounts) {
667 p = list_entry(prev, struct vfsmount, mnt_child);
668 prev = p->mnt_mounts.prev;
670 return p;
673 struct vfsmount *
674 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
676 struct vfsmount *mnt;
677 struct dentry *root;
679 if (!type)
680 return ERR_PTR(-ENODEV);
682 mnt = alloc_vfsmnt(name);
683 if (!mnt)
684 return ERR_PTR(-ENOMEM);
686 if (flags & MS_KERNMOUNT)
687 mnt->mnt_flags = MNT_INTERNAL;
689 root = mount_fs(type, flags, name, data);
690 if (IS_ERR(root)) {
691 free_vfsmnt(mnt);
692 return ERR_CAST(root);
695 mnt->mnt_root = root;
696 mnt->mnt_sb = root->d_sb;
697 mnt->mnt_mountpoint = mnt->mnt_root;
698 mnt->mnt_parent = mnt;
699 return mnt;
701 EXPORT_SYMBOL_GPL(vfs_kern_mount);
703 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
704 int flag)
706 struct super_block *sb = old->mnt_sb;
707 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
709 if (mnt) {
710 if (flag & (CL_SLAVE | CL_PRIVATE))
711 mnt->mnt_group_id = 0; /* not a peer of original */
712 else
713 mnt->mnt_group_id = old->mnt_group_id;
715 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
716 int err = mnt_alloc_group_id(mnt);
717 if (err)
718 goto out_free;
721 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
722 atomic_inc(&sb->s_active);
723 mnt->mnt_sb = sb;
724 mnt->mnt_root = dget(root);
725 mnt->mnt_mountpoint = mnt->mnt_root;
726 mnt->mnt_parent = mnt;
728 if (flag & CL_SLAVE) {
729 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
730 mnt->mnt_master = old;
731 CLEAR_MNT_SHARED(mnt);
732 } else if (!(flag & CL_PRIVATE)) {
733 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
734 list_add(&mnt->mnt_share, &old->mnt_share);
735 if (IS_MNT_SLAVE(old))
736 list_add(&mnt->mnt_slave, &old->mnt_slave);
737 mnt->mnt_master = old->mnt_master;
739 if (flag & CL_MAKE_SHARED)
740 set_mnt_shared(mnt);
742 /* stick the duplicate mount on the same expiry list
743 * as the original if that was on one */
744 if (flag & CL_EXPIRE) {
745 if (!list_empty(&old->mnt_expire))
746 list_add(&mnt->mnt_expire, &old->mnt_expire);
749 return mnt;
751 out_free:
752 free_vfsmnt(mnt);
753 return NULL;
756 static inline void mntfree(struct vfsmount *mnt)
758 struct super_block *sb = mnt->mnt_sb;
761 * This probably indicates that somebody messed
762 * up a mnt_want/drop_write() pair. If this
763 * happens, the filesystem was probably unable
764 * to make r/w->r/o transitions.
767 * The locking used to deal with mnt_count decrement provides barriers,
768 * so mnt_get_writers() below is safe.
770 WARN_ON(mnt_get_writers(mnt));
771 fsnotify_vfsmount_delete(mnt);
772 dput(mnt->mnt_root);
773 free_vfsmnt(mnt);
774 deactivate_super(sb);
777 static void mntput_no_expire(struct vfsmount *mnt)
779 put_again:
780 #ifdef CONFIG_SMP
781 br_read_lock(vfsmount_lock);
782 if (likely(atomic_read(&mnt->mnt_longterm))) {
783 mnt_dec_count(mnt);
784 br_read_unlock(vfsmount_lock);
785 return;
787 br_read_unlock(vfsmount_lock);
789 br_write_lock(vfsmount_lock);
790 mnt_dec_count(mnt);
791 if (mnt_get_count(mnt)) {
792 br_write_unlock(vfsmount_lock);
793 return;
795 #else
796 mnt_dec_count(mnt);
797 if (likely(mnt_get_count(mnt)))
798 return;
799 br_write_lock(vfsmount_lock);
800 #endif
801 if (unlikely(mnt->mnt_pinned)) {
802 mnt_add_count(mnt, mnt->mnt_pinned + 1);
803 mnt->mnt_pinned = 0;
804 br_write_unlock(vfsmount_lock);
805 acct_auto_close_mnt(mnt);
806 goto put_again;
808 br_write_unlock(vfsmount_lock);
809 mntfree(mnt);
812 void mntput(struct vfsmount *mnt)
814 if (mnt) {
815 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
816 if (unlikely(mnt->mnt_expiry_mark))
817 mnt->mnt_expiry_mark = 0;
818 mntput_no_expire(mnt);
821 EXPORT_SYMBOL(mntput);
823 struct vfsmount *mntget(struct vfsmount *mnt)
825 if (mnt)
826 mnt_inc_count(mnt);
827 return mnt;
829 EXPORT_SYMBOL(mntget);
831 void mnt_pin(struct vfsmount *mnt)
833 br_write_lock(vfsmount_lock);
834 mnt->mnt_pinned++;
835 br_write_unlock(vfsmount_lock);
837 EXPORT_SYMBOL(mnt_pin);
839 void mnt_unpin(struct vfsmount *mnt)
841 br_write_lock(vfsmount_lock);
842 if (mnt->mnt_pinned) {
843 mnt_inc_count(mnt);
844 mnt->mnt_pinned--;
846 br_write_unlock(vfsmount_lock);
848 EXPORT_SYMBOL(mnt_unpin);
850 static inline void mangle(struct seq_file *m, const char *s)
852 seq_escape(m, s, " \t\n\\");
856 * Simple .show_options callback for filesystems which don't want to
857 * implement more complex mount option showing.
859 * See also save_mount_options().
861 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
863 const char *options;
865 rcu_read_lock();
866 options = rcu_dereference(mnt->mnt_sb->s_options);
868 if (options != NULL && options[0]) {
869 seq_putc(m, ',');
870 mangle(m, options);
872 rcu_read_unlock();
874 return 0;
876 EXPORT_SYMBOL(generic_show_options);
879 * If filesystem uses generic_show_options(), this function should be
880 * called from the fill_super() callback.
882 * The .remount_fs callback usually needs to be handled in a special
883 * way, to make sure, that previous options are not overwritten if the
884 * remount fails.
886 * Also note, that if the filesystem's .remount_fs function doesn't
887 * reset all options to their default value, but changes only newly
888 * given options, then the displayed options will not reflect reality
889 * any more.
891 void save_mount_options(struct super_block *sb, char *options)
893 BUG_ON(sb->s_options);
894 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
896 EXPORT_SYMBOL(save_mount_options);
898 void replace_mount_options(struct super_block *sb, char *options)
900 char *old = sb->s_options;
901 rcu_assign_pointer(sb->s_options, options);
902 if (old) {
903 synchronize_rcu();
904 kfree(old);
907 EXPORT_SYMBOL(replace_mount_options);
909 #ifdef CONFIG_PROC_FS
910 /* iterator */
911 static void *m_start(struct seq_file *m, loff_t *pos)
913 struct proc_mounts *p = m->private;
915 down_read(&namespace_sem);
916 return seq_list_start(&p->ns->list, *pos);
919 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
921 struct proc_mounts *p = m->private;
923 return seq_list_next(v, &p->ns->list, pos);
926 static void m_stop(struct seq_file *m, void *v)
928 up_read(&namespace_sem);
931 int mnt_had_events(struct proc_mounts *p)
933 struct mnt_namespace *ns = p->ns;
934 int res = 0;
936 br_read_lock(vfsmount_lock);
937 if (p->m.poll_event != ns->event) {
938 p->m.poll_event = ns->event;
939 res = 1;
941 br_read_unlock(vfsmount_lock);
943 return res;
946 struct proc_fs_info {
947 int flag;
948 const char *str;
951 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
953 static const struct proc_fs_info fs_info[] = {
954 { MS_SYNCHRONOUS, ",sync" },
955 { MS_DIRSYNC, ",dirsync" },
956 { MS_MANDLOCK, ",mand" },
957 { 0, NULL }
959 const struct proc_fs_info *fs_infop;
961 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
962 if (sb->s_flags & fs_infop->flag)
963 seq_puts(m, fs_infop->str);
966 return security_sb_show_options(m, sb);
969 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
971 static const struct proc_fs_info mnt_info[] = {
972 { MNT_NOSUID, ",nosuid" },
973 { MNT_NODEV, ",nodev" },
974 { MNT_NOEXEC, ",noexec" },
975 { MNT_NOATIME, ",noatime" },
976 { MNT_NODIRATIME, ",nodiratime" },
977 { MNT_RELATIME, ",relatime" },
978 { 0, NULL }
980 const struct proc_fs_info *fs_infop;
982 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
983 if (mnt->mnt_flags & fs_infop->flag)
984 seq_puts(m, fs_infop->str);
988 static void show_type(struct seq_file *m, struct super_block *sb)
990 mangle(m, sb->s_type->name);
991 if (sb->s_subtype && sb->s_subtype[0]) {
992 seq_putc(m, '.');
993 mangle(m, sb->s_subtype);
997 static int show_vfsmnt(struct seq_file *m, void *v)
999 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1000 int err = 0;
1001 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1003 if (mnt->mnt_sb->s_op->show_devname) {
1004 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1005 if (err)
1006 goto out;
1007 } else {
1008 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1010 seq_putc(m, ' ');
1011 seq_path(m, &mnt_path, " \t\n\\");
1012 seq_putc(m, ' ');
1013 show_type(m, mnt->mnt_sb);
1014 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
1015 err = show_sb_opts(m, mnt->mnt_sb);
1016 if (err)
1017 goto out;
1018 show_mnt_opts(m, mnt);
1019 if (mnt->mnt_sb->s_op->show_options)
1020 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1021 seq_puts(m, " 0 0\n");
1022 out:
1023 return err;
1026 const struct seq_operations mounts_op = {
1027 .start = m_start,
1028 .next = m_next,
1029 .stop = m_stop,
1030 .show = show_vfsmnt
1033 static int show_mountinfo(struct seq_file *m, void *v)
1035 struct proc_mounts *p = m->private;
1036 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1037 struct super_block *sb = mnt->mnt_sb;
1038 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1039 struct path root = p->root;
1040 int err = 0;
1042 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1043 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1044 if (sb->s_op->show_path)
1045 err = sb->s_op->show_path(m, mnt);
1046 else
1047 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1048 if (err)
1049 goto out;
1050 seq_putc(m, ' ');
1052 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1053 err = seq_path_root(m, &mnt_path, &root, " \t\n\\");
1054 if (err)
1055 goto out;
1057 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1058 show_mnt_opts(m, mnt);
1060 /* Tagged fields ("foo:X" or "bar") */
1061 if (IS_MNT_SHARED(mnt))
1062 seq_printf(m, " shared:%i", mnt->mnt_group_id);
1063 if (IS_MNT_SLAVE(mnt)) {
1064 int master = mnt->mnt_master->mnt_group_id;
1065 int dom = get_dominating_id(mnt, &p->root);
1066 seq_printf(m, " master:%i", master);
1067 if (dom && dom != master)
1068 seq_printf(m, " propagate_from:%i", dom);
1070 if (IS_MNT_UNBINDABLE(mnt))
1071 seq_puts(m, " unbindable");
1073 /* Filesystem specific data */
1074 seq_puts(m, " - ");
1075 show_type(m, sb);
1076 seq_putc(m, ' ');
1077 if (sb->s_op->show_devname)
1078 err = sb->s_op->show_devname(m, mnt);
1079 else
1080 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1081 if (err)
1082 goto out;
1083 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1084 err = show_sb_opts(m, sb);
1085 if (err)
1086 goto out;
1087 if (sb->s_op->show_options)
1088 err = sb->s_op->show_options(m, mnt);
1089 seq_putc(m, '\n');
1090 out:
1091 return err;
1094 const struct seq_operations mountinfo_op = {
1095 .start = m_start,
1096 .next = m_next,
1097 .stop = m_stop,
1098 .show = show_mountinfo,
1101 static int show_vfsstat(struct seq_file *m, void *v)
1103 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1104 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1105 int err = 0;
1107 /* device */
1108 if (mnt->mnt_sb->s_op->show_devname) {
1109 seq_puts(m, "device ");
1110 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1111 } else {
1112 if (mnt->mnt_devname) {
1113 seq_puts(m, "device ");
1114 mangle(m, mnt->mnt_devname);
1115 } else
1116 seq_puts(m, "no device");
1119 /* mount point */
1120 seq_puts(m, " mounted on ");
1121 seq_path(m, &mnt_path, " \t\n\\");
1122 seq_putc(m, ' ');
1124 /* file system type */
1125 seq_puts(m, "with fstype ");
1126 show_type(m, mnt->mnt_sb);
1128 /* optional statistics */
1129 if (mnt->mnt_sb->s_op->show_stats) {
1130 seq_putc(m, ' ');
1131 if (!err)
1132 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1135 seq_putc(m, '\n');
1136 return err;
1139 const struct seq_operations mountstats_op = {
1140 .start = m_start,
1141 .next = m_next,
1142 .stop = m_stop,
1143 .show = show_vfsstat,
1145 #endif /* CONFIG_PROC_FS */
1148 * may_umount_tree - check if a mount tree is busy
1149 * @mnt: root of mount tree
1151 * This is called to check if a tree of mounts has any
1152 * open files, pwds, chroots or sub mounts that are
1153 * busy.
1155 int may_umount_tree(struct vfsmount *mnt)
1157 int actual_refs = 0;
1158 int minimum_refs = 0;
1159 struct vfsmount *p;
1161 /* write lock needed for mnt_get_count */
1162 br_write_lock(vfsmount_lock);
1163 for (p = mnt; p; p = next_mnt(p, mnt)) {
1164 actual_refs += mnt_get_count(p);
1165 minimum_refs += 2;
1167 br_write_unlock(vfsmount_lock);
1169 if (actual_refs > minimum_refs)
1170 return 0;
1172 return 1;
1175 EXPORT_SYMBOL(may_umount_tree);
1178 * may_umount - check if a mount point is busy
1179 * @mnt: root of mount
1181 * This is called to check if a mount point has any
1182 * open files, pwds, chroots or sub mounts. If the
1183 * mount has sub mounts this will return busy
1184 * regardless of whether the sub mounts are busy.
1186 * Doesn't take quota and stuff into account. IOW, in some cases it will
1187 * give false negatives. The main reason why it's here is that we need
1188 * a non-destructive way to look for easily umountable filesystems.
1190 int may_umount(struct vfsmount *mnt)
1192 int ret = 1;
1193 down_read(&namespace_sem);
1194 br_write_lock(vfsmount_lock);
1195 if (propagate_mount_busy(mnt, 2))
1196 ret = 0;
1197 br_write_unlock(vfsmount_lock);
1198 up_read(&namespace_sem);
1199 return ret;
1202 EXPORT_SYMBOL(may_umount);
1204 void release_mounts(struct list_head *head)
1206 struct vfsmount *mnt;
1207 while (!list_empty(head)) {
1208 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1209 list_del_init(&mnt->mnt_hash);
1210 if (mnt->mnt_parent != mnt) {
1211 struct dentry *dentry;
1212 struct vfsmount *m;
1214 br_write_lock(vfsmount_lock);
1215 dentry = mnt->mnt_mountpoint;
1216 m = mnt->mnt_parent;
1217 mnt->mnt_mountpoint = mnt->mnt_root;
1218 mnt->mnt_parent = mnt;
1219 m->mnt_ghosts--;
1220 br_write_unlock(vfsmount_lock);
1221 dput(dentry);
1222 mntput(m);
1224 mntput(mnt);
1229 * vfsmount lock must be held for write
1230 * namespace_sem must be held for write
1232 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1234 LIST_HEAD(tmp_list);
1235 struct vfsmount *p;
1237 for (p = mnt; p; p = next_mnt(p, mnt))
1238 list_move(&p->mnt_hash, &tmp_list);
1240 if (propagate)
1241 propagate_umount(&tmp_list);
1243 list_for_each_entry(p, &tmp_list, mnt_hash) {
1244 list_del_init(&p->mnt_expire);
1245 list_del_init(&p->mnt_list);
1246 __touch_mnt_namespace(p->mnt_ns);
1247 p->mnt_ns = NULL;
1248 __mnt_make_shortterm(p);
1249 list_del_init(&p->mnt_child);
1250 if (p->mnt_parent != p) {
1251 p->mnt_parent->mnt_ghosts++;
1252 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
1254 change_mnt_propagation(p, MS_PRIVATE);
1256 list_splice(&tmp_list, kill);
1259 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1261 static int do_umount(struct vfsmount *mnt, int flags)
1263 struct super_block *sb = mnt->mnt_sb;
1264 int retval;
1265 LIST_HEAD(umount_list);
1267 retval = security_sb_umount(mnt, flags);
1268 if (retval)
1269 return retval;
1272 * Allow userspace to request a mountpoint be expired rather than
1273 * unmounting unconditionally. Unmount only happens if:
1274 * (1) the mark is already set (the mark is cleared by mntput())
1275 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1277 if (flags & MNT_EXPIRE) {
1278 if (mnt == current->fs->root.mnt ||
1279 flags & (MNT_FORCE | MNT_DETACH))
1280 return -EINVAL;
1283 * probably don't strictly need the lock here if we examined
1284 * all race cases, but it's a slowpath.
1286 br_write_lock(vfsmount_lock);
1287 if (mnt_get_count(mnt) != 2) {
1288 br_write_unlock(vfsmount_lock);
1289 return -EBUSY;
1291 br_write_unlock(vfsmount_lock);
1293 if (!xchg(&mnt->mnt_expiry_mark, 1))
1294 return -EAGAIN;
1298 * If we may have to abort operations to get out of this
1299 * mount, and they will themselves hold resources we must
1300 * allow the fs to do things. In the Unix tradition of
1301 * 'Gee thats tricky lets do it in userspace' the umount_begin
1302 * might fail to complete on the first run through as other tasks
1303 * must return, and the like. Thats for the mount program to worry
1304 * about for the moment.
1307 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1308 sb->s_op->umount_begin(sb);
1312 * No sense to grab the lock for this test, but test itself looks
1313 * somewhat bogus. Suggestions for better replacement?
1314 * Ho-hum... In principle, we might treat that as umount + switch
1315 * to rootfs. GC would eventually take care of the old vfsmount.
1316 * Actually it makes sense, especially if rootfs would contain a
1317 * /reboot - static binary that would close all descriptors and
1318 * call reboot(9). Then init(8) could umount root and exec /reboot.
1320 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1322 * Special case for "unmounting" root ...
1323 * we just try to remount it readonly.
1325 down_write(&sb->s_umount);
1326 if (!(sb->s_flags & MS_RDONLY))
1327 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1328 up_write(&sb->s_umount);
1329 return retval;
1332 down_write(&namespace_sem);
1333 br_write_lock(vfsmount_lock);
1334 event++;
1336 if (!(flags & MNT_DETACH))
1337 shrink_submounts(mnt, &umount_list);
1339 retval = -EBUSY;
1340 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1341 if (!list_empty(&mnt->mnt_list))
1342 umount_tree(mnt, 1, &umount_list);
1343 retval = 0;
1345 br_write_unlock(vfsmount_lock);
1346 up_write(&namespace_sem);
1347 release_mounts(&umount_list);
1348 return retval;
1352 * Now umount can handle mount points as well as block devices.
1353 * This is important for filesystems which use unnamed block devices.
1355 * We now support a flag for forced unmount like the other 'big iron'
1356 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1359 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1361 struct path path;
1362 int retval;
1363 int lookup_flags = 0;
1365 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1366 return -EINVAL;
1368 if (!(flags & UMOUNT_NOFOLLOW))
1369 lookup_flags |= LOOKUP_FOLLOW;
1371 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1372 if (retval)
1373 goto out;
1374 retval = -EINVAL;
1375 if (path.dentry != path.mnt->mnt_root)
1376 goto dput_and_out;
1377 if (!check_mnt(path.mnt))
1378 goto dput_and_out;
1380 retval = -EPERM;
1381 if (!capable(CAP_SYS_ADMIN))
1382 goto dput_and_out;
1384 retval = do_umount(path.mnt, flags);
1385 dput_and_out:
1386 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1387 dput(path.dentry);
1388 mntput_no_expire(path.mnt);
1389 out:
1390 return retval;
1393 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1396 * The 2.0 compatible umount. No flags.
1398 SYSCALL_DEFINE1(oldumount, char __user *, name)
1400 return sys_umount(name, 0);
1403 #endif
1405 static int mount_is_safe(struct path *path)
1407 if (capable(CAP_SYS_ADMIN))
1408 return 0;
1409 return -EPERM;
1410 #ifdef notyet
1411 if (S_ISLNK(path->dentry->d_inode->i_mode))
1412 return -EPERM;
1413 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1414 if (current_uid() != path->dentry->d_inode->i_uid)
1415 return -EPERM;
1417 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1418 return -EPERM;
1419 return 0;
1420 #endif
1423 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1424 int flag)
1426 struct vfsmount *res, *p, *q, *r, *s;
1427 struct path path;
1429 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1430 return NULL;
1432 res = q = clone_mnt(mnt, dentry, flag);
1433 if (!q)
1434 goto Enomem;
1435 q->mnt_mountpoint = mnt->mnt_mountpoint;
1437 p = mnt;
1438 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1439 if (!is_subdir(r->mnt_mountpoint, dentry))
1440 continue;
1442 for (s = r; s; s = next_mnt(s, r)) {
1443 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1444 s = skip_mnt_tree(s);
1445 continue;
1447 while (p != s->mnt_parent) {
1448 p = p->mnt_parent;
1449 q = q->mnt_parent;
1451 p = s;
1452 path.mnt = q;
1453 path.dentry = p->mnt_mountpoint;
1454 q = clone_mnt(p, p->mnt_root, flag);
1455 if (!q)
1456 goto Enomem;
1457 br_write_lock(vfsmount_lock);
1458 list_add_tail(&q->mnt_list, &res->mnt_list);
1459 attach_mnt(q, &path);
1460 br_write_unlock(vfsmount_lock);
1463 return res;
1464 Enomem:
1465 if (res) {
1466 LIST_HEAD(umount_list);
1467 br_write_lock(vfsmount_lock);
1468 umount_tree(res, 0, &umount_list);
1469 br_write_unlock(vfsmount_lock);
1470 release_mounts(&umount_list);
1472 return NULL;
1475 struct vfsmount *collect_mounts(struct path *path)
1477 struct vfsmount *tree;
1478 down_write(&namespace_sem);
1479 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1480 up_write(&namespace_sem);
1481 return tree;
1484 void drop_collected_mounts(struct vfsmount *mnt)
1486 LIST_HEAD(umount_list);
1487 down_write(&namespace_sem);
1488 br_write_lock(vfsmount_lock);
1489 umount_tree(mnt, 0, &umount_list);
1490 br_write_unlock(vfsmount_lock);
1491 up_write(&namespace_sem);
1492 release_mounts(&umount_list);
1495 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1496 struct vfsmount *root)
1498 struct vfsmount *mnt;
1499 int res = f(root, arg);
1500 if (res)
1501 return res;
1502 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1503 res = f(mnt, arg);
1504 if (res)
1505 return res;
1507 return 0;
1510 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1512 struct vfsmount *p;
1514 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1515 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1516 mnt_release_group_id(p);
1520 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1522 struct vfsmount *p;
1524 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1525 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1526 int err = mnt_alloc_group_id(p);
1527 if (err) {
1528 cleanup_group_ids(mnt, p);
1529 return err;
1534 return 0;
1538 * @source_mnt : mount tree to be attached
1539 * @nd : place the mount tree @source_mnt is attached
1540 * @parent_nd : if non-null, detach the source_mnt from its parent and
1541 * store the parent mount and mountpoint dentry.
1542 * (done when source_mnt is moved)
1544 * NOTE: in the table below explains the semantics when a source mount
1545 * of a given type is attached to a destination mount of a given type.
1546 * ---------------------------------------------------------------------------
1547 * | BIND MOUNT OPERATION |
1548 * |**************************************************************************
1549 * | source-->| shared | private | slave | unbindable |
1550 * | dest | | | | |
1551 * | | | | | | |
1552 * | v | | | | |
1553 * |**************************************************************************
1554 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1555 * | | | | | |
1556 * |non-shared| shared (+) | private | slave (*) | invalid |
1557 * ***************************************************************************
1558 * A bind operation clones the source mount and mounts the clone on the
1559 * destination mount.
1561 * (++) the cloned mount is propagated to all the mounts in the propagation
1562 * tree of the destination mount and the cloned mount is added to
1563 * the peer group of the source mount.
1564 * (+) the cloned mount is created under the destination mount and is marked
1565 * as shared. The cloned mount is added to the peer group of the source
1566 * mount.
1567 * (+++) the mount is propagated to all the mounts in the propagation tree
1568 * of the destination mount and the cloned mount is made slave
1569 * of the same master as that of the source mount. The cloned mount
1570 * is marked as 'shared and slave'.
1571 * (*) the cloned mount is made a slave of the same master as that of the
1572 * source mount.
1574 * ---------------------------------------------------------------------------
1575 * | MOVE MOUNT OPERATION |
1576 * |**************************************************************************
1577 * | source-->| shared | private | slave | unbindable |
1578 * | dest | | | | |
1579 * | | | | | | |
1580 * | v | | | | |
1581 * |**************************************************************************
1582 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1583 * | | | | | |
1584 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1585 * ***************************************************************************
1587 * (+) the mount is moved to the destination. And is then propagated to
1588 * all the mounts in the propagation tree of the destination mount.
1589 * (+*) the mount is moved to the destination.
1590 * (+++) the mount is moved to the destination and is then propagated to
1591 * all the mounts belonging to the destination mount's propagation tree.
1592 * the mount is marked as 'shared and slave'.
1593 * (*) the mount continues to be a slave at the new location.
1595 * if the source mount is a tree, the operations explained above is
1596 * applied to each mount in the tree.
1597 * Must be called without spinlocks held, since this function can sleep
1598 * in allocations.
1600 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1601 struct path *path, struct path *parent_path)
1603 LIST_HEAD(tree_list);
1604 struct vfsmount *dest_mnt = path->mnt;
1605 struct dentry *dest_dentry = path->dentry;
1606 struct vfsmount *child, *p;
1607 int err;
1609 if (IS_MNT_SHARED(dest_mnt)) {
1610 err = invent_group_ids(source_mnt, true);
1611 if (err)
1612 goto out;
1614 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1615 if (err)
1616 goto out_cleanup_ids;
1618 br_write_lock(vfsmount_lock);
1620 if (IS_MNT_SHARED(dest_mnt)) {
1621 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1622 set_mnt_shared(p);
1624 if (parent_path) {
1625 detach_mnt(source_mnt, parent_path);
1626 attach_mnt(source_mnt, path);
1627 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1628 } else {
1629 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1630 commit_tree(source_mnt);
1633 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1634 list_del_init(&child->mnt_hash);
1635 commit_tree(child);
1637 br_write_unlock(vfsmount_lock);
1639 return 0;
1641 out_cleanup_ids:
1642 if (IS_MNT_SHARED(dest_mnt))
1643 cleanup_group_ids(source_mnt, NULL);
1644 out:
1645 return err;
1648 static int lock_mount(struct path *path)
1650 struct vfsmount *mnt;
1651 retry:
1652 mutex_lock(&path->dentry->d_inode->i_mutex);
1653 if (unlikely(cant_mount(path->dentry))) {
1654 mutex_unlock(&path->dentry->d_inode->i_mutex);
1655 return -ENOENT;
1657 down_write(&namespace_sem);
1658 mnt = lookup_mnt(path);
1659 if (likely(!mnt))
1660 return 0;
1661 up_write(&namespace_sem);
1662 mutex_unlock(&path->dentry->d_inode->i_mutex);
1663 path_put(path);
1664 path->mnt = mnt;
1665 path->dentry = dget(mnt->mnt_root);
1666 goto retry;
1669 static void unlock_mount(struct path *path)
1671 up_write(&namespace_sem);
1672 mutex_unlock(&path->dentry->d_inode->i_mutex);
1675 static int graft_tree(struct vfsmount *mnt, struct path *path)
1677 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1678 return -EINVAL;
1680 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1681 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1682 return -ENOTDIR;
1684 if (d_unlinked(path->dentry))
1685 return -ENOENT;
1687 return attach_recursive_mnt(mnt, path, NULL);
1691 * Sanity check the flags to change_mnt_propagation.
1694 static int flags_to_propagation_type(int flags)
1696 int type = flags & ~(MS_REC | MS_SILENT);
1698 /* Fail if any non-propagation flags are set */
1699 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1700 return 0;
1701 /* Only one propagation flag should be set */
1702 if (!is_power_of_2(type))
1703 return 0;
1704 return type;
1708 * recursively change the type of the mountpoint.
1710 static int do_change_type(struct path *path, int flag)
1712 struct vfsmount *m, *mnt = path->mnt;
1713 int recurse = flag & MS_REC;
1714 int type;
1715 int err = 0;
1717 if (!capable(CAP_SYS_ADMIN))
1718 return -EPERM;
1720 if (path->dentry != path->mnt->mnt_root)
1721 return -EINVAL;
1723 type = flags_to_propagation_type(flag);
1724 if (!type)
1725 return -EINVAL;
1727 down_write(&namespace_sem);
1728 if (type == MS_SHARED) {
1729 err = invent_group_ids(mnt, recurse);
1730 if (err)
1731 goto out_unlock;
1734 br_write_lock(vfsmount_lock);
1735 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1736 change_mnt_propagation(m, type);
1737 br_write_unlock(vfsmount_lock);
1739 out_unlock:
1740 up_write(&namespace_sem);
1741 return err;
1745 * do loopback mount.
1747 static int do_loopback(struct path *path, char *old_name,
1748 int recurse)
1750 LIST_HEAD(umount_list);
1751 struct path old_path;
1752 struct vfsmount *mnt = NULL;
1753 int err = mount_is_safe(path);
1754 if (err)
1755 return err;
1756 if (!old_name || !*old_name)
1757 return -EINVAL;
1758 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1759 if (err)
1760 return err;
1762 err = lock_mount(path);
1763 if (err)
1764 goto out;
1766 err = -EINVAL;
1767 if (IS_MNT_UNBINDABLE(old_path.mnt))
1768 goto out2;
1770 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1771 goto out2;
1773 err = -ENOMEM;
1774 if (recurse)
1775 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1776 else
1777 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1779 if (!mnt)
1780 goto out2;
1782 err = graft_tree(mnt, path);
1783 if (err) {
1784 br_write_lock(vfsmount_lock);
1785 umount_tree(mnt, 0, &umount_list);
1786 br_write_unlock(vfsmount_lock);
1788 out2:
1789 unlock_mount(path);
1790 release_mounts(&umount_list);
1791 out:
1792 path_put(&old_path);
1793 return err;
1796 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1798 int error = 0;
1799 int readonly_request = 0;
1801 if (ms_flags & MS_RDONLY)
1802 readonly_request = 1;
1803 if (readonly_request == __mnt_is_readonly(mnt))
1804 return 0;
1806 if (readonly_request)
1807 error = mnt_make_readonly(mnt);
1808 else
1809 __mnt_unmake_readonly(mnt);
1810 return error;
1814 * change filesystem flags. dir should be a physical root of filesystem.
1815 * If you've mounted a non-root directory somewhere and want to do remount
1816 * on it - tough luck.
1818 static int do_remount(struct path *path, int flags, int mnt_flags,
1819 void *data)
1821 int err;
1822 struct super_block *sb = path->mnt->mnt_sb;
1824 if (!capable(CAP_SYS_ADMIN))
1825 return -EPERM;
1827 if (!check_mnt(path->mnt))
1828 return -EINVAL;
1830 if (path->dentry != path->mnt->mnt_root)
1831 return -EINVAL;
1833 err = security_sb_remount(sb, data);
1834 if (err)
1835 return err;
1837 down_write(&sb->s_umount);
1838 if (flags & MS_BIND)
1839 err = change_mount_flags(path->mnt, flags);
1840 else
1841 err = do_remount_sb(sb, flags, data, 0);
1842 if (!err) {
1843 br_write_lock(vfsmount_lock);
1844 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1845 path->mnt->mnt_flags = mnt_flags;
1846 br_write_unlock(vfsmount_lock);
1848 up_write(&sb->s_umount);
1849 if (!err) {
1850 br_write_lock(vfsmount_lock);
1851 touch_mnt_namespace(path->mnt->mnt_ns);
1852 br_write_unlock(vfsmount_lock);
1854 return err;
1857 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1859 struct vfsmount *p;
1860 for (p = mnt; p; p = next_mnt(p, mnt)) {
1861 if (IS_MNT_UNBINDABLE(p))
1862 return 1;
1864 return 0;
1867 static int do_move_mount(struct path *path, char *old_name)
1869 struct path old_path, parent_path;
1870 struct vfsmount *p;
1871 int err = 0;
1872 if (!capable(CAP_SYS_ADMIN))
1873 return -EPERM;
1874 if (!old_name || !*old_name)
1875 return -EINVAL;
1876 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1877 if (err)
1878 return err;
1880 err = lock_mount(path);
1881 if (err < 0)
1882 goto out;
1884 err = -EINVAL;
1885 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1886 goto out1;
1888 if (d_unlinked(path->dentry))
1889 goto out1;
1891 err = -EINVAL;
1892 if (old_path.dentry != old_path.mnt->mnt_root)
1893 goto out1;
1895 if (old_path.mnt == old_path.mnt->mnt_parent)
1896 goto out1;
1898 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1899 S_ISDIR(old_path.dentry->d_inode->i_mode))
1900 goto out1;
1902 * Don't move a mount residing in a shared parent.
1904 if (old_path.mnt->mnt_parent &&
1905 IS_MNT_SHARED(old_path.mnt->mnt_parent))
1906 goto out1;
1908 * Don't move a mount tree containing unbindable mounts to a destination
1909 * mount which is shared.
1911 if (IS_MNT_SHARED(path->mnt) &&
1912 tree_contains_unbindable(old_path.mnt))
1913 goto out1;
1914 err = -ELOOP;
1915 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1916 if (p == old_path.mnt)
1917 goto out1;
1919 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1920 if (err)
1921 goto out1;
1923 /* if the mount is moved, it should no longer be expire
1924 * automatically */
1925 list_del_init(&old_path.mnt->mnt_expire);
1926 out1:
1927 unlock_mount(path);
1928 out:
1929 if (!err)
1930 path_put(&parent_path);
1931 path_put(&old_path);
1932 return err;
1935 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1937 int err;
1938 const char *subtype = strchr(fstype, '.');
1939 if (subtype) {
1940 subtype++;
1941 err = -EINVAL;
1942 if (!subtype[0])
1943 goto err;
1944 } else
1945 subtype = "";
1947 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1948 err = -ENOMEM;
1949 if (!mnt->mnt_sb->s_subtype)
1950 goto err;
1951 return mnt;
1953 err:
1954 mntput(mnt);
1955 return ERR_PTR(err);
1958 struct vfsmount *
1959 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1961 struct file_system_type *type = get_fs_type(fstype);
1962 struct vfsmount *mnt;
1963 if (!type)
1964 return ERR_PTR(-ENODEV);
1965 mnt = vfs_kern_mount(type, flags, name, data);
1966 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1967 !mnt->mnt_sb->s_subtype)
1968 mnt = fs_set_subtype(mnt, fstype);
1969 put_filesystem(type);
1970 return mnt;
1972 EXPORT_SYMBOL_GPL(do_kern_mount);
1975 * add a mount into a namespace's mount tree
1977 static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1979 int err;
1981 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1983 err = lock_mount(path);
1984 if (err)
1985 return err;
1987 err = -EINVAL;
1988 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1989 goto unlock;
1991 /* Refuse the same filesystem on the same mount point */
1992 err = -EBUSY;
1993 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1994 path->mnt->mnt_root == path->dentry)
1995 goto unlock;
1997 err = -EINVAL;
1998 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1999 goto unlock;
2001 newmnt->mnt_flags = mnt_flags;
2002 err = graft_tree(newmnt, path);
2004 unlock:
2005 unlock_mount(path);
2006 return err;
2010 * create a new mount for userspace and request it to be added into the
2011 * namespace's tree
2013 static int do_new_mount(struct path *path, char *type, int flags,
2014 int mnt_flags, char *name, void *data)
2016 struct vfsmount *mnt;
2017 int err;
2019 if (!type)
2020 return -EINVAL;
2022 /* we need capabilities... */
2023 if (!capable(CAP_SYS_ADMIN))
2024 return -EPERM;
2026 mnt = do_kern_mount(type, flags, name, data);
2027 if (IS_ERR(mnt))
2028 return PTR_ERR(mnt);
2030 err = do_add_mount(mnt, path, mnt_flags);
2031 if (err)
2032 mntput(mnt);
2033 return err;
2036 int finish_automount(struct vfsmount *m, struct path *path)
2038 int err;
2039 /* The new mount record should have at least 2 refs to prevent it being
2040 * expired before we get a chance to add it
2042 BUG_ON(mnt_get_count(m) < 2);
2044 if (m->mnt_sb == path->mnt->mnt_sb &&
2045 m->mnt_root == path->dentry) {
2046 err = -ELOOP;
2047 goto fail;
2050 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2051 if (!err)
2052 return 0;
2053 fail:
2054 /* remove m from any expiration list it may be on */
2055 if (!list_empty(&m->mnt_expire)) {
2056 down_write(&namespace_sem);
2057 br_write_lock(vfsmount_lock);
2058 list_del_init(&m->mnt_expire);
2059 br_write_unlock(vfsmount_lock);
2060 up_write(&namespace_sem);
2062 mntput(m);
2063 mntput(m);
2064 return err;
2068 * mnt_set_expiry - Put a mount on an expiration list
2069 * @mnt: The mount to list.
2070 * @expiry_list: The list to add the mount to.
2072 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2074 down_write(&namespace_sem);
2075 br_write_lock(vfsmount_lock);
2077 list_add_tail(&mnt->mnt_expire, expiry_list);
2079 br_write_unlock(vfsmount_lock);
2080 up_write(&namespace_sem);
2082 EXPORT_SYMBOL(mnt_set_expiry);
2085 * process a list of expirable mountpoints with the intent of discarding any
2086 * mountpoints that aren't in use and haven't been touched since last we came
2087 * here
2089 void mark_mounts_for_expiry(struct list_head *mounts)
2091 struct vfsmount *mnt, *next;
2092 LIST_HEAD(graveyard);
2093 LIST_HEAD(umounts);
2095 if (list_empty(mounts))
2096 return;
2098 down_write(&namespace_sem);
2099 br_write_lock(vfsmount_lock);
2101 /* extract from the expiration list every vfsmount that matches the
2102 * following criteria:
2103 * - only referenced by its parent vfsmount
2104 * - still marked for expiry (marked on the last call here; marks are
2105 * cleared by mntput())
2107 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2108 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2109 propagate_mount_busy(mnt, 1))
2110 continue;
2111 list_move(&mnt->mnt_expire, &graveyard);
2113 while (!list_empty(&graveyard)) {
2114 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2115 touch_mnt_namespace(mnt->mnt_ns);
2116 umount_tree(mnt, 1, &umounts);
2118 br_write_unlock(vfsmount_lock);
2119 up_write(&namespace_sem);
2121 release_mounts(&umounts);
2124 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2127 * Ripoff of 'select_parent()'
2129 * search the list of submounts for a given mountpoint, and move any
2130 * shrinkable submounts to the 'graveyard' list.
2132 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2134 struct vfsmount *this_parent = parent;
2135 struct list_head *next;
2136 int found = 0;
2138 repeat:
2139 next = this_parent->mnt_mounts.next;
2140 resume:
2141 while (next != &this_parent->mnt_mounts) {
2142 struct list_head *tmp = next;
2143 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2145 next = tmp->next;
2146 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2147 continue;
2149 * Descend a level if the d_mounts list is non-empty.
2151 if (!list_empty(&mnt->mnt_mounts)) {
2152 this_parent = mnt;
2153 goto repeat;
2156 if (!propagate_mount_busy(mnt, 1)) {
2157 list_move_tail(&mnt->mnt_expire, graveyard);
2158 found++;
2162 * All done at this level ... ascend and resume the search
2164 if (this_parent != parent) {
2165 next = this_parent->mnt_child.next;
2166 this_parent = this_parent->mnt_parent;
2167 goto resume;
2169 return found;
2173 * process a list of expirable mountpoints with the intent of discarding any
2174 * submounts of a specific parent mountpoint
2176 * vfsmount_lock must be held for write
2178 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
2180 LIST_HEAD(graveyard);
2181 struct vfsmount *m;
2183 /* extract submounts of 'mountpoint' from the expiration list */
2184 while (select_submounts(mnt, &graveyard)) {
2185 while (!list_empty(&graveyard)) {
2186 m = list_first_entry(&graveyard, struct vfsmount,
2187 mnt_expire);
2188 touch_mnt_namespace(m->mnt_ns);
2189 umount_tree(m, 1, umounts);
2195 * Some copy_from_user() implementations do not return the exact number of
2196 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2197 * Note that this function differs from copy_from_user() in that it will oops
2198 * on bad values of `to', rather than returning a short copy.
2200 static long exact_copy_from_user(void *to, const void __user * from,
2201 unsigned long n)
2203 char *t = to;
2204 const char __user *f = from;
2205 char c;
2207 if (!access_ok(VERIFY_READ, from, n))
2208 return n;
2210 while (n) {
2211 if (__get_user(c, f)) {
2212 memset(t, 0, n);
2213 break;
2215 *t++ = c;
2216 f++;
2217 n--;
2219 return n;
2222 int copy_mount_options(const void __user * data, unsigned long *where)
2224 int i;
2225 unsigned long page;
2226 unsigned long size;
2228 *where = 0;
2229 if (!data)
2230 return 0;
2232 if (!(page = __get_free_page(GFP_KERNEL)))
2233 return -ENOMEM;
2235 /* We only care that *some* data at the address the user
2236 * gave us is valid. Just in case, we'll zero
2237 * the remainder of the page.
2239 /* copy_from_user cannot cross TASK_SIZE ! */
2240 size = TASK_SIZE - (unsigned long)data;
2241 if (size > PAGE_SIZE)
2242 size = PAGE_SIZE;
2244 i = size - exact_copy_from_user((void *)page, data, size);
2245 if (!i) {
2246 free_page(page);
2247 return -EFAULT;
2249 if (i != PAGE_SIZE)
2250 memset((char *)page + i, 0, PAGE_SIZE - i);
2251 *where = page;
2252 return 0;
2255 int copy_mount_string(const void __user *data, char **where)
2257 char *tmp;
2259 if (!data) {
2260 *where = NULL;
2261 return 0;
2264 tmp = strndup_user(data, PAGE_SIZE);
2265 if (IS_ERR(tmp))
2266 return PTR_ERR(tmp);
2268 *where = tmp;
2269 return 0;
2273 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2274 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2276 * data is a (void *) that can point to any structure up to
2277 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2278 * information (or be NULL).
2280 * Pre-0.97 versions of mount() didn't have a flags word.
2281 * When the flags word was introduced its top half was required
2282 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2283 * Therefore, if this magic number is present, it carries no information
2284 * and must be discarded.
2286 long do_mount(char *dev_name, char *dir_name, char *type_page,
2287 unsigned long flags, void *data_page)
2289 struct path path;
2290 int retval = 0;
2291 int mnt_flags = 0;
2293 /* Discard magic */
2294 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2295 flags &= ~MS_MGC_MSK;
2297 /* Basic sanity checks */
2299 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2300 return -EINVAL;
2302 if (data_page)
2303 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2305 /* ... and get the mountpoint */
2306 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2307 if (retval)
2308 return retval;
2310 retval = security_sb_mount(dev_name, &path,
2311 type_page, flags, data_page);
2312 if (retval)
2313 goto dput_out;
2315 /* Default to relatime unless overriden */
2316 if (!(flags & MS_NOATIME))
2317 mnt_flags |= MNT_RELATIME;
2319 /* Separate the per-mountpoint flags */
2320 if (flags & MS_NOSUID)
2321 mnt_flags |= MNT_NOSUID;
2322 if (flags & MS_NODEV)
2323 mnt_flags |= MNT_NODEV;
2324 if (flags & MS_NOEXEC)
2325 mnt_flags |= MNT_NOEXEC;
2326 if (flags & MS_NOATIME)
2327 mnt_flags |= MNT_NOATIME;
2328 if (flags & MS_NODIRATIME)
2329 mnt_flags |= MNT_NODIRATIME;
2330 if (flags & MS_STRICTATIME)
2331 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2332 if (flags & MS_RDONLY)
2333 mnt_flags |= MNT_READONLY;
2335 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2336 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2337 MS_STRICTATIME);
2339 if (flags & MS_REMOUNT)
2340 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2341 data_page);
2342 else if (flags & MS_BIND)
2343 retval = do_loopback(&path, dev_name, flags & MS_REC);
2344 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2345 retval = do_change_type(&path, flags);
2346 else if (flags & MS_MOVE)
2347 retval = do_move_mount(&path, dev_name);
2348 else
2349 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2350 dev_name, data_page);
2351 dput_out:
2352 path_put(&path);
2353 return retval;
2356 static struct mnt_namespace *alloc_mnt_ns(void)
2358 struct mnt_namespace *new_ns;
2360 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2361 if (!new_ns)
2362 return ERR_PTR(-ENOMEM);
2363 atomic_set(&new_ns->count, 1);
2364 new_ns->root = NULL;
2365 INIT_LIST_HEAD(&new_ns->list);
2366 init_waitqueue_head(&new_ns->poll);
2367 new_ns->event = 0;
2368 return new_ns;
2371 void mnt_make_longterm(struct vfsmount *mnt)
2373 __mnt_make_longterm(mnt);
2376 void mnt_make_shortterm(struct vfsmount *mnt)
2378 #ifdef CONFIG_SMP
2379 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2380 return;
2381 br_write_lock(vfsmount_lock);
2382 atomic_dec(&mnt->mnt_longterm);
2383 br_write_unlock(vfsmount_lock);
2384 #endif
2388 * Allocate a new namespace structure and populate it with contents
2389 * copied from the namespace of the passed in task structure.
2391 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2392 struct fs_struct *fs)
2394 struct mnt_namespace *new_ns;
2395 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2396 struct vfsmount *p, *q;
2398 new_ns = alloc_mnt_ns();
2399 if (IS_ERR(new_ns))
2400 return new_ns;
2402 down_write(&namespace_sem);
2403 /* First pass: copy the tree topology */
2404 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2405 CL_COPY_ALL | CL_EXPIRE);
2406 if (!new_ns->root) {
2407 up_write(&namespace_sem);
2408 kfree(new_ns);
2409 return ERR_PTR(-ENOMEM);
2411 br_write_lock(vfsmount_lock);
2412 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2413 br_write_unlock(vfsmount_lock);
2416 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2417 * as belonging to new namespace. We have already acquired a private
2418 * fs_struct, so tsk->fs->lock is not needed.
2420 p = mnt_ns->root;
2421 q = new_ns->root;
2422 while (p) {
2423 q->mnt_ns = new_ns;
2424 __mnt_make_longterm(q);
2425 if (fs) {
2426 if (p == fs->root.mnt) {
2427 fs->root.mnt = mntget(q);
2428 __mnt_make_longterm(q);
2429 mnt_make_shortterm(p);
2430 rootmnt = p;
2432 if (p == fs->pwd.mnt) {
2433 fs->pwd.mnt = mntget(q);
2434 __mnt_make_longterm(q);
2435 mnt_make_shortterm(p);
2436 pwdmnt = p;
2439 p = next_mnt(p, mnt_ns->root);
2440 q = next_mnt(q, new_ns->root);
2442 up_write(&namespace_sem);
2444 if (rootmnt)
2445 mntput(rootmnt);
2446 if (pwdmnt)
2447 mntput(pwdmnt);
2449 return new_ns;
2452 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2453 struct fs_struct *new_fs)
2455 struct mnt_namespace *new_ns;
2457 BUG_ON(!ns);
2458 get_mnt_ns(ns);
2460 if (!(flags & CLONE_NEWNS))
2461 return ns;
2463 new_ns = dup_mnt_ns(ns, new_fs);
2465 put_mnt_ns(ns);
2466 return new_ns;
2470 * create_mnt_ns - creates a private namespace and adds a root filesystem
2471 * @mnt: pointer to the new root filesystem mountpoint
2473 struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2475 struct mnt_namespace *new_ns;
2477 new_ns = alloc_mnt_ns();
2478 if (!IS_ERR(new_ns)) {
2479 mnt->mnt_ns = new_ns;
2480 __mnt_make_longterm(mnt);
2481 new_ns->root = mnt;
2482 list_add(&new_ns->list, &new_ns->root->mnt_list);
2483 } else {
2484 mntput(mnt);
2486 return new_ns;
2488 EXPORT_SYMBOL(create_mnt_ns);
2490 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2492 struct mnt_namespace *ns;
2493 struct super_block *s;
2494 struct path path;
2495 int err;
2497 ns = create_mnt_ns(mnt);
2498 if (IS_ERR(ns))
2499 return ERR_CAST(ns);
2501 err = vfs_path_lookup(mnt->mnt_root, mnt,
2502 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2504 put_mnt_ns(ns);
2506 if (err)
2507 return ERR_PTR(err);
2509 /* trade a vfsmount reference for active sb one */
2510 s = path.mnt->mnt_sb;
2511 atomic_inc(&s->s_active);
2512 mntput(path.mnt);
2513 /* lock the sucker */
2514 down_write(&s->s_umount);
2515 /* ... and return the root of (sub)tree on it */
2516 return path.dentry;
2518 EXPORT_SYMBOL(mount_subtree);
2520 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2521 char __user *, type, unsigned long, flags, void __user *, data)
2523 int ret;
2524 char *kernel_type;
2525 char *kernel_dir;
2526 char *kernel_dev;
2527 unsigned long data_page;
2529 ret = copy_mount_string(type, &kernel_type);
2530 if (ret < 0)
2531 goto out_type;
2533 kernel_dir = getname(dir_name);
2534 if (IS_ERR(kernel_dir)) {
2535 ret = PTR_ERR(kernel_dir);
2536 goto out_dir;
2539 ret = copy_mount_string(dev_name, &kernel_dev);
2540 if (ret < 0)
2541 goto out_dev;
2543 ret = copy_mount_options(data, &data_page);
2544 if (ret < 0)
2545 goto out_data;
2547 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2548 (void *) data_page);
2550 free_page(data_page);
2551 out_data:
2552 kfree(kernel_dev);
2553 out_dev:
2554 putname(kernel_dir);
2555 out_dir:
2556 kfree(kernel_type);
2557 out_type:
2558 return ret;
2562 * pivot_root Semantics:
2563 * Moves the root file system of the current process to the directory put_old,
2564 * makes new_root as the new root file system of the current process, and sets
2565 * root/cwd of all processes which had them on the current root to new_root.
2567 * Restrictions:
2568 * The new_root and put_old must be directories, and must not be on the
2569 * same file system as the current process root. The put_old must be
2570 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2571 * pointed to by put_old must yield the same directory as new_root. No other
2572 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2574 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2575 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2576 * in this situation.
2578 * Notes:
2579 * - we don't move root/cwd if they are not at the root (reason: if something
2580 * cared enough to change them, it's probably wrong to force them elsewhere)
2581 * - it's okay to pick a root that isn't the root of a file system, e.g.
2582 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2583 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2584 * first.
2586 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2587 const char __user *, put_old)
2589 struct vfsmount *tmp;
2590 struct path new, old, parent_path, root_parent, root;
2591 int error;
2593 if (!capable(CAP_SYS_ADMIN))
2594 return -EPERM;
2596 error = user_path_dir(new_root, &new);
2597 if (error)
2598 goto out0;
2600 error = user_path_dir(put_old, &old);
2601 if (error)
2602 goto out1;
2604 error = security_sb_pivotroot(&old, &new);
2605 if (error)
2606 goto out2;
2608 get_fs_root(current->fs, &root);
2609 error = lock_mount(&old);
2610 if (error)
2611 goto out3;
2613 error = -EINVAL;
2614 if (IS_MNT_SHARED(old.mnt) ||
2615 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2616 IS_MNT_SHARED(root.mnt->mnt_parent))
2617 goto out4;
2618 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
2619 goto out4;
2620 error = -ENOENT;
2621 if (d_unlinked(new.dentry))
2622 goto out4;
2623 if (d_unlinked(old.dentry))
2624 goto out4;
2625 error = -EBUSY;
2626 if (new.mnt == root.mnt ||
2627 old.mnt == root.mnt)
2628 goto out4; /* loop, on the same file system */
2629 error = -EINVAL;
2630 if (root.mnt->mnt_root != root.dentry)
2631 goto out4; /* not a mountpoint */
2632 if (root.mnt->mnt_parent == root.mnt)
2633 goto out4; /* not attached */
2634 if (new.mnt->mnt_root != new.dentry)
2635 goto out4; /* not a mountpoint */
2636 if (new.mnt->mnt_parent == new.mnt)
2637 goto out4; /* not attached */
2638 /* make sure we can reach put_old from new_root */
2639 tmp = old.mnt;
2640 if (tmp != new.mnt) {
2641 for (;;) {
2642 if (tmp->mnt_parent == tmp)
2643 goto out4; /* already mounted on put_old */
2644 if (tmp->mnt_parent == new.mnt)
2645 break;
2646 tmp = tmp->mnt_parent;
2648 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2649 goto out4;
2650 } else if (!is_subdir(old.dentry, new.dentry))
2651 goto out4;
2652 br_write_lock(vfsmount_lock);
2653 detach_mnt(new.mnt, &parent_path);
2654 detach_mnt(root.mnt, &root_parent);
2655 /* mount old root on put_old */
2656 attach_mnt(root.mnt, &old);
2657 /* mount new_root on / */
2658 attach_mnt(new.mnt, &root_parent);
2659 touch_mnt_namespace(current->nsproxy->mnt_ns);
2660 br_write_unlock(vfsmount_lock);
2661 chroot_fs_refs(&root, &new);
2662 error = 0;
2663 out4:
2664 unlock_mount(&old);
2665 if (!error) {
2666 path_put(&root_parent);
2667 path_put(&parent_path);
2669 out3:
2670 path_put(&root);
2671 out2:
2672 path_put(&old);
2673 out1:
2674 path_put(&new);
2675 out0:
2676 return error;
2679 static void __init init_mount_tree(void)
2681 struct vfsmount *mnt;
2682 struct mnt_namespace *ns;
2683 struct path root;
2685 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2686 if (IS_ERR(mnt))
2687 panic("Can't create rootfs");
2689 ns = create_mnt_ns(mnt);
2690 if (IS_ERR(ns))
2691 panic("Can't allocate initial namespace");
2693 init_task.nsproxy->mnt_ns = ns;
2694 get_mnt_ns(ns);
2696 root.mnt = ns->root;
2697 root.dentry = ns->root->mnt_root;
2699 set_fs_pwd(current->fs, &root);
2700 set_fs_root(current->fs, &root);
2703 void __init mnt_init(void)
2705 unsigned u;
2706 int err;
2708 init_rwsem(&namespace_sem);
2710 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2711 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2713 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2715 if (!mount_hashtable)
2716 panic("Failed to allocate mount hash table\n");
2718 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2720 for (u = 0; u < HASH_SIZE; u++)
2721 INIT_LIST_HEAD(&mount_hashtable[u]);
2723 br_lock_init(vfsmount_lock);
2725 err = sysfs_init();
2726 if (err)
2727 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2728 __func__, err);
2729 fs_kobj = kobject_create_and_add("fs", NULL);
2730 if (!fs_kobj)
2731 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2732 init_rootfs();
2733 init_mount_tree();
2736 void put_mnt_ns(struct mnt_namespace *ns)
2738 LIST_HEAD(umount_list);
2740 if (!atomic_dec_and_test(&ns->count))
2741 return;
2742 down_write(&namespace_sem);
2743 br_write_lock(vfsmount_lock);
2744 umount_tree(ns->root, 0, &umount_list);
2745 br_write_unlock(vfsmount_lock);
2746 up_write(&namespace_sem);
2747 release_mounts(&umount_list);
2748 kfree(ns);
2750 EXPORT_SYMBOL(put_mnt_ns);
2752 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2754 struct vfsmount *mnt;
2755 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2756 if (!IS_ERR(mnt)) {
2758 * it is a longterm mount, don't release mnt until
2759 * we unmount before file sys is unregistered
2761 mnt_make_longterm(mnt);
2763 return mnt;
2765 EXPORT_SYMBOL_GPL(kern_mount_data);
2767 void kern_unmount(struct vfsmount *mnt)
2769 /* release long term mount so mount point can be released */
2770 if (!IS_ERR_OR_NULL(mnt)) {
2771 mnt_make_shortterm(mnt);
2772 mntput(mnt);
2775 EXPORT_SYMBOL(kern_unmount);
2777 bool our_mnt(struct vfsmount *mnt)
2779 return check_mnt(mnt);