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.
11 #include <linux/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched.h>
14 #include <linux/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/quotaops.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/security.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/log2.h>
30 #include <linux/idr.h>
31 #include <asm/uaccess.h>
32 #include <asm/unistd.h>
36 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
37 #define HASH_SIZE (1UL << HASH_SHIFT)
39 /* spinlock for vfsmount related operations, inplace of dcache_lock */
40 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(vfsmount_lock
);
43 static DEFINE_IDA(mnt_id_ida
);
44 static DEFINE_IDA(mnt_group_ida
);
46 static struct list_head
*mount_hashtable __read_mostly
;
47 static struct kmem_cache
*mnt_cache __read_mostly
;
48 static struct rw_semaphore namespace_sem
;
51 struct kobject
*fs_kobj
;
52 EXPORT_SYMBOL_GPL(fs_kobj
);
54 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
56 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
57 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
58 tmp
= tmp
+ (tmp
>> HASH_SHIFT
);
59 return tmp
& (HASH_SIZE
- 1);
62 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
64 /* allocation is serialized by namespace_sem */
65 static int mnt_alloc_id(struct vfsmount
*mnt
)
70 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
71 spin_lock(&vfsmount_lock
);
72 res
= ida_get_new(&mnt_id_ida
, &mnt
->mnt_id
);
73 spin_unlock(&vfsmount_lock
);
80 static void mnt_free_id(struct vfsmount
*mnt
)
82 spin_lock(&vfsmount_lock
);
83 ida_remove(&mnt_id_ida
, mnt
->mnt_id
);
84 spin_unlock(&vfsmount_lock
);
88 * Allocate a new peer group ID
90 * mnt_group_ida is protected by namespace_sem
92 static int mnt_alloc_group_id(struct vfsmount
*mnt
)
94 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
97 return ida_get_new_above(&mnt_group_ida
, 1, &mnt
->mnt_group_id
);
101 * Release a peer group ID
103 void mnt_release_group_id(struct vfsmount
*mnt
)
105 ida_remove(&mnt_group_ida
, mnt
->mnt_group_id
);
106 mnt
->mnt_group_id
= 0;
109 struct vfsmount
*alloc_vfsmnt(const char *name
)
111 struct vfsmount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
115 err
= mnt_alloc_id(mnt
);
117 kmem_cache_free(mnt_cache
, mnt
);
121 atomic_set(&mnt
->mnt_count
, 1);
122 INIT_LIST_HEAD(&mnt
->mnt_hash
);
123 INIT_LIST_HEAD(&mnt
->mnt_child
);
124 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
125 INIT_LIST_HEAD(&mnt
->mnt_list
);
126 INIT_LIST_HEAD(&mnt
->mnt_expire
);
127 INIT_LIST_HEAD(&mnt
->mnt_share
);
128 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
129 INIT_LIST_HEAD(&mnt
->mnt_slave
);
130 atomic_set(&mnt
->__mnt_writers
, 0);
132 int size
= strlen(name
) + 1;
133 char *newname
= kmalloc(size
, GFP_KERNEL
);
135 memcpy(newname
, name
, size
);
136 mnt
->mnt_devname
= newname
;
144 * Most r/o checks on a fs are for operations that take
145 * discrete amounts of time, like a write() or unlink().
146 * We must keep track of when those operations start
147 * (for permission checks) and when they end, so that
148 * we can determine when writes are able to occur to
152 * __mnt_is_readonly: check whether a mount is read-only
153 * @mnt: the mount to check for its write status
155 * This shouldn't be used directly ouside of the VFS.
156 * It does not guarantee that the filesystem will stay
157 * r/w, just that it is right *now*. This can not and
158 * should not be used in place of IS_RDONLY(inode).
159 * mnt_want/drop_write() will _keep_ the filesystem
162 int __mnt_is_readonly(struct vfsmount
*mnt
)
164 if (mnt
->mnt_flags
& MNT_READONLY
)
166 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
170 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
174 * If holding multiple instances of this lock, they
175 * must be ordered by cpu number.
178 struct lock_class_key lock_class
; /* compiles out with !lockdep */
180 struct vfsmount
*mnt
;
181 } ____cacheline_aligned_in_smp
;
182 static DEFINE_PER_CPU(struct mnt_writer
, mnt_writers
);
184 static int __init
init_mnt_writers(void)
187 for_each_possible_cpu(cpu
) {
188 struct mnt_writer
*writer
= &per_cpu(mnt_writers
, cpu
);
189 spin_lock_init(&writer
->lock
);
190 lockdep_set_class(&writer
->lock
, &writer
->lock_class
);
195 fs_initcall(init_mnt_writers
);
197 static void unlock_mnt_writers(void)
200 struct mnt_writer
*cpu_writer
;
202 for_each_possible_cpu(cpu
) {
203 cpu_writer
= &per_cpu(mnt_writers
, cpu
);
204 spin_unlock(&cpu_writer
->lock
);
208 static inline void __clear_mnt_count(struct mnt_writer
*cpu_writer
)
210 if (!cpu_writer
->mnt
)
213 * This is in case anyone ever leaves an invalid,
214 * old ->mnt and a count of 0.
216 if (!cpu_writer
->count
)
218 atomic_add(cpu_writer
->count
, &cpu_writer
->mnt
->__mnt_writers
);
219 cpu_writer
->count
= 0;
222 * must hold cpu_writer->lock
224 static inline void use_cpu_writer_for_mount(struct mnt_writer
*cpu_writer
,
225 struct vfsmount
*mnt
)
227 if (cpu_writer
->mnt
== mnt
)
229 __clear_mnt_count(cpu_writer
);
230 cpu_writer
->mnt
= mnt
;
234 * Most r/o checks on a fs are for operations that take
235 * discrete amounts of time, like a write() or unlink().
236 * We must keep track of when those operations start
237 * (for permission checks) and when they end, so that
238 * we can determine when writes are able to occur to
242 * mnt_want_write - get write access to a mount
243 * @mnt: the mount on which to take a write
245 * This tells the low-level filesystem that a write is
246 * about to be performed to it, and makes sure that
247 * writes are allowed before returning success. When
248 * the write operation is finished, mnt_drop_write()
249 * must be called. This is effectively a refcount.
251 int mnt_want_write(struct vfsmount
*mnt
)
254 struct mnt_writer
*cpu_writer
;
256 cpu_writer
= &get_cpu_var(mnt_writers
);
257 spin_lock(&cpu_writer
->lock
);
258 if (__mnt_is_readonly(mnt
)) {
262 use_cpu_writer_for_mount(cpu_writer
, mnt
);
265 spin_unlock(&cpu_writer
->lock
);
266 put_cpu_var(mnt_writers
);
269 EXPORT_SYMBOL_GPL(mnt_want_write
);
271 static void lock_mnt_writers(void)
274 struct mnt_writer
*cpu_writer
;
276 for_each_possible_cpu(cpu
) {
277 cpu_writer
= &per_cpu(mnt_writers
, cpu
);
278 spin_lock(&cpu_writer
->lock
);
279 __clear_mnt_count(cpu_writer
);
280 cpu_writer
->mnt
= NULL
;
285 * These per-cpu write counts are not guaranteed to have
286 * matched increments and decrements on any given cpu.
287 * A file open()ed for write on one cpu and close()d on
288 * another cpu will imbalance this count. Make sure it
289 * does not get too far out of whack.
291 static void handle_write_count_underflow(struct vfsmount
*mnt
)
293 if (atomic_read(&mnt
->__mnt_writers
) >=
294 MNT_WRITER_UNDERFLOW_LIMIT
)
297 * It isn't necessary to hold all of the locks
298 * at the same time, but doing it this way makes
299 * us share a lot more code.
303 * vfsmount_lock is for mnt_flags.
305 spin_lock(&vfsmount_lock
);
307 * If coalescing the per-cpu writer counts did not
308 * get us back to a positive writer count, we have
311 if ((atomic_read(&mnt
->__mnt_writers
) < 0) &&
312 !(mnt
->mnt_flags
& MNT_IMBALANCED_WRITE_COUNT
)) {
313 printk(KERN_DEBUG
"leak detected on mount(%p) writers "
315 mnt
, atomic_read(&mnt
->__mnt_writers
));
317 /* use the flag to keep the dmesg spam down */
318 mnt
->mnt_flags
|= MNT_IMBALANCED_WRITE_COUNT
;
320 spin_unlock(&vfsmount_lock
);
321 unlock_mnt_writers();
325 * mnt_drop_write - give up write access to a mount
326 * @mnt: the mount on which to give up write access
328 * Tells the low-level filesystem that we are done
329 * performing writes to it. Must be matched with
330 * mnt_want_write() call above.
332 void mnt_drop_write(struct vfsmount
*mnt
)
334 int must_check_underflow
= 0;
335 struct mnt_writer
*cpu_writer
;
337 cpu_writer
= &get_cpu_var(mnt_writers
);
338 spin_lock(&cpu_writer
->lock
);
340 use_cpu_writer_for_mount(cpu_writer
, mnt
);
341 if (cpu_writer
->count
> 0) {
344 must_check_underflow
= 1;
345 atomic_dec(&mnt
->__mnt_writers
);
348 spin_unlock(&cpu_writer
->lock
);
350 * Logically, we could call this each time,
351 * but the __mnt_writers cacheline tends to
352 * be cold, and makes this expensive.
354 if (must_check_underflow
)
355 handle_write_count_underflow(mnt
);
357 * This could be done right after the spinlock
358 * is taken because the spinlock keeps us on
359 * the cpu, and disables preemption. However,
360 * putting it here bounds the amount that
361 * __mnt_writers can underflow. Without it,
362 * we could theoretically wrap __mnt_writers.
364 put_cpu_var(mnt_writers
);
366 EXPORT_SYMBOL_GPL(mnt_drop_write
);
368 static int mnt_make_readonly(struct vfsmount
*mnt
)
374 * With all the locks held, this value is stable
376 if (atomic_read(&mnt
->__mnt_writers
) > 0) {
381 * nobody can do a successful mnt_want_write() with all
382 * of the counts in MNT_DENIED_WRITE and the locks held.
384 spin_lock(&vfsmount_lock
);
386 mnt
->mnt_flags
|= MNT_READONLY
;
387 spin_unlock(&vfsmount_lock
);
389 unlock_mnt_writers();
393 static void __mnt_unmake_readonly(struct vfsmount
*mnt
)
395 spin_lock(&vfsmount_lock
);
396 mnt
->mnt_flags
&= ~MNT_READONLY
;
397 spin_unlock(&vfsmount_lock
);
400 int simple_set_mnt(struct vfsmount
*mnt
, struct super_block
*sb
)
403 mnt
->mnt_root
= dget(sb
->s_root
);
407 EXPORT_SYMBOL(simple_set_mnt
);
409 void free_vfsmnt(struct vfsmount
*mnt
)
411 kfree(mnt
->mnt_devname
);
413 kmem_cache_free(mnt_cache
, mnt
);
417 * find the first or last mount at @dentry on vfsmount @mnt depending on
418 * @dir. If @dir is set return the first mount else return the last mount.
420 struct vfsmount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
423 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
424 struct list_head
*tmp
= head
;
425 struct vfsmount
*p
, *found
= NULL
;
428 tmp
= dir
? tmp
->next
: tmp
->prev
;
432 p
= list_entry(tmp
, struct vfsmount
, mnt_hash
);
433 if (p
->mnt_parent
== mnt
&& p
->mnt_mountpoint
== dentry
) {
442 * lookup_mnt increments the ref count before returning
443 * the vfsmount struct.
445 struct vfsmount
*lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
447 struct vfsmount
*child_mnt
;
448 spin_lock(&vfsmount_lock
);
449 if ((child_mnt
= __lookup_mnt(mnt
, dentry
, 1)))
451 spin_unlock(&vfsmount_lock
);
455 static inline int check_mnt(struct vfsmount
*mnt
)
457 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
460 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
464 wake_up_interruptible(&ns
->poll
);
468 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
470 if (ns
&& ns
->event
!= event
) {
472 wake_up_interruptible(&ns
->poll
);
476 static void detach_mnt(struct vfsmount
*mnt
, struct path
*old_path
)
478 old_path
->dentry
= mnt
->mnt_mountpoint
;
479 old_path
->mnt
= mnt
->mnt_parent
;
480 mnt
->mnt_parent
= mnt
;
481 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
482 list_del_init(&mnt
->mnt_child
);
483 list_del_init(&mnt
->mnt_hash
);
484 old_path
->dentry
->d_mounted
--;
487 void mnt_set_mountpoint(struct vfsmount
*mnt
, struct dentry
*dentry
,
488 struct vfsmount
*child_mnt
)
490 child_mnt
->mnt_parent
= mntget(mnt
);
491 child_mnt
->mnt_mountpoint
= dget(dentry
);
495 static void attach_mnt(struct vfsmount
*mnt
, struct path
*path
)
497 mnt_set_mountpoint(path
->mnt
, path
->dentry
, mnt
);
498 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
499 hash(path
->mnt
, path
->dentry
));
500 list_add_tail(&mnt
->mnt_child
, &path
->mnt
->mnt_mounts
);
504 * the caller must hold vfsmount_lock
506 static void commit_tree(struct vfsmount
*mnt
)
508 struct vfsmount
*parent
= mnt
->mnt_parent
;
511 struct mnt_namespace
*n
= parent
->mnt_ns
;
513 BUG_ON(parent
== mnt
);
515 list_add_tail(&head
, &mnt
->mnt_list
);
516 list_for_each_entry(m
, &head
, mnt_list
)
518 list_splice(&head
, n
->list
.prev
);
520 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
521 hash(parent
, mnt
->mnt_mountpoint
));
522 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
523 touch_mnt_namespace(n
);
526 static struct vfsmount
*next_mnt(struct vfsmount
*p
, struct vfsmount
*root
)
528 struct list_head
*next
= p
->mnt_mounts
.next
;
529 if (next
== &p
->mnt_mounts
) {
533 next
= p
->mnt_child
.next
;
534 if (next
!= &p
->mnt_parent
->mnt_mounts
)
539 return list_entry(next
, struct vfsmount
, mnt_child
);
542 static struct vfsmount
*skip_mnt_tree(struct vfsmount
*p
)
544 struct list_head
*prev
= p
->mnt_mounts
.prev
;
545 while (prev
!= &p
->mnt_mounts
) {
546 p
= list_entry(prev
, struct vfsmount
, mnt_child
);
547 prev
= p
->mnt_mounts
.prev
;
552 static struct vfsmount
*clone_mnt(struct vfsmount
*old
, struct dentry
*root
,
555 struct super_block
*sb
= old
->mnt_sb
;
556 struct vfsmount
*mnt
= alloc_vfsmnt(old
->mnt_devname
);
559 if (flag
& (CL_SLAVE
| CL_PRIVATE
))
560 mnt
->mnt_group_id
= 0; /* not a peer of original */
562 mnt
->mnt_group_id
= old
->mnt_group_id
;
564 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
565 int err
= mnt_alloc_group_id(mnt
);
570 mnt
->mnt_flags
= old
->mnt_flags
;
571 atomic_inc(&sb
->s_active
);
573 mnt
->mnt_root
= dget(root
);
574 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
575 mnt
->mnt_parent
= mnt
;
577 if (flag
& CL_SLAVE
) {
578 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
579 mnt
->mnt_master
= old
;
580 CLEAR_MNT_SHARED(mnt
);
581 } else if (!(flag
& CL_PRIVATE
)) {
582 if ((flag
& CL_PROPAGATION
) || IS_MNT_SHARED(old
))
583 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
584 if (IS_MNT_SLAVE(old
))
585 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
586 mnt
->mnt_master
= old
->mnt_master
;
588 if (flag
& CL_MAKE_SHARED
)
591 /* stick the duplicate mount on the same expiry list
592 * as the original if that was on one */
593 if (flag
& CL_EXPIRE
) {
594 if (!list_empty(&old
->mnt_expire
))
595 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
605 static inline void __mntput(struct vfsmount
*mnt
)
608 struct super_block
*sb
= mnt
->mnt_sb
;
610 * We don't have to hold all of the locks at the
611 * same time here because we know that we're the
612 * last reference to mnt and that no new writers
615 for_each_possible_cpu(cpu
) {
616 struct mnt_writer
*cpu_writer
= &per_cpu(mnt_writers
, cpu
);
617 if (cpu_writer
->mnt
!= mnt
)
619 spin_lock(&cpu_writer
->lock
);
620 atomic_add(cpu_writer
->count
, &mnt
->__mnt_writers
);
621 cpu_writer
->count
= 0;
623 * Might as well do this so that no one
624 * ever sees the pointer and expects
627 cpu_writer
->mnt
= NULL
;
628 spin_unlock(&cpu_writer
->lock
);
631 * This probably indicates that somebody messed
632 * up a mnt_want/drop_write() pair. If this
633 * happens, the filesystem was probably unable
634 * to make r/w->r/o transitions.
636 WARN_ON(atomic_read(&mnt
->__mnt_writers
));
639 deactivate_super(sb
);
642 void mntput_no_expire(struct vfsmount
*mnt
)
645 if (atomic_dec_and_lock(&mnt
->mnt_count
, &vfsmount_lock
)) {
646 if (likely(!mnt
->mnt_pinned
)) {
647 spin_unlock(&vfsmount_lock
);
651 atomic_add(mnt
->mnt_pinned
+ 1, &mnt
->mnt_count
);
653 spin_unlock(&vfsmount_lock
);
654 acct_auto_close_mnt(mnt
);
655 security_sb_umount_close(mnt
);
660 EXPORT_SYMBOL(mntput_no_expire
);
662 void mnt_pin(struct vfsmount
*mnt
)
664 spin_lock(&vfsmount_lock
);
666 spin_unlock(&vfsmount_lock
);
669 EXPORT_SYMBOL(mnt_pin
);
671 void mnt_unpin(struct vfsmount
*mnt
)
673 spin_lock(&vfsmount_lock
);
674 if (mnt
->mnt_pinned
) {
675 atomic_inc(&mnt
->mnt_count
);
678 spin_unlock(&vfsmount_lock
);
681 EXPORT_SYMBOL(mnt_unpin
);
683 static inline void mangle(struct seq_file
*m
, const char *s
)
685 seq_escape(m
, s
, " \t\n\\");
689 * Simple .show_options callback for filesystems which don't want to
690 * implement more complex mount option showing.
692 * See also save_mount_options().
694 int generic_show_options(struct seq_file
*m
, struct vfsmount
*mnt
)
696 const char *options
= mnt
->mnt_sb
->s_options
;
698 if (options
!= NULL
&& options
[0]) {
705 EXPORT_SYMBOL(generic_show_options
);
708 * If filesystem uses generic_show_options(), this function should be
709 * called from the fill_super() callback.
711 * The .remount_fs callback usually needs to be handled in a special
712 * way, to make sure, that previous options are not overwritten if the
715 * Also note, that if the filesystem's .remount_fs function doesn't
716 * reset all options to their default value, but changes only newly
717 * given options, then the displayed options will not reflect reality
720 void save_mount_options(struct super_block
*sb
, char *options
)
722 kfree(sb
->s_options
);
723 sb
->s_options
= kstrdup(options
, GFP_KERNEL
);
725 EXPORT_SYMBOL(save_mount_options
);
727 #ifdef CONFIG_PROC_FS
729 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
731 struct proc_mounts
*p
= m
->private;
733 down_read(&namespace_sem
);
734 return seq_list_start(&p
->ns
->list
, *pos
);
737 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
739 struct proc_mounts
*p
= m
->private;
741 return seq_list_next(v
, &p
->ns
->list
, pos
);
744 static void m_stop(struct seq_file
*m
, void *v
)
746 up_read(&namespace_sem
);
749 static int show_vfsmnt(struct seq_file
*m
, void *v
)
751 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
753 static struct proc_fs_info
{
757 { MS_SYNCHRONOUS
, ",sync" },
758 { MS_DIRSYNC
, ",dirsync" },
759 { MS_MANDLOCK
, ",mand" },
762 static struct proc_fs_info mnt_info
[] = {
763 { MNT_NOSUID
, ",nosuid" },
764 { MNT_NODEV
, ",nodev" },
765 { MNT_NOEXEC
, ",noexec" },
766 { MNT_NOATIME
, ",noatime" },
767 { MNT_NODIRATIME
, ",nodiratime" },
768 { MNT_RELATIME
, ",relatime" },
771 struct proc_fs_info
*fs_infop
;
772 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
774 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
776 seq_path(m
, &mnt_path
, " \t\n\\");
778 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
779 if (mnt
->mnt_sb
->s_subtype
&& mnt
->mnt_sb
->s_subtype
[0]) {
781 mangle(m
, mnt
->mnt_sb
->s_subtype
);
783 seq_puts(m
, __mnt_is_readonly(mnt
) ? " ro" : " rw");
784 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
785 if (mnt
->mnt_sb
->s_flags
& fs_infop
->flag
)
786 seq_puts(m
, fs_infop
->str
);
788 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
789 if (mnt
->mnt_flags
& fs_infop
->flag
)
790 seq_puts(m
, fs_infop
->str
);
792 if (mnt
->mnt_sb
->s_op
->show_options
)
793 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
794 seq_puts(m
, " 0 0\n");
798 const struct seq_operations mounts_op
= {
805 static int show_vfsstat(struct seq_file
*m
, void *v
)
807 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
808 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
812 if (mnt
->mnt_devname
) {
813 seq_puts(m
, "device ");
814 mangle(m
, mnt
->mnt_devname
);
816 seq_puts(m
, "no device");
819 seq_puts(m
, " mounted on ");
820 seq_path(m
, &mnt_path
, " \t\n\\");
823 /* file system type */
824 seq_puts(m
, "with fstype ");
825 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
827 /* optional statistics */
828 if (mnt
->mnt_sb
->s_op
->show_stats
) {
830 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
837 const struct seq_operations mountstats_op
= {
841 .show
= show_vfsstat
,
843 #endif /* CONFIG_PROC_FS */
846 * may_umount_tree - check if a mount tree is busy
847 * @mnt: root of mount tree
849 * This is called to check if a tree of mounts has any
850 * open files, pwds, chroots or sub mounts that are
853 int may_umount_tree(struct vfsmount
*mnt
)
856 int minimum_refs
= 0;
859 spin_lock(&vfsmount_lock
);
860 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
861 actual_refs
+= atomic_read(&p
->mnt_count
);
864 spin_unlock(&vfsmount_lock
);
866 if (actual_refs
> minimum_refs
)
872 EXPORT_SYMBOL(may_umount_tree
);
875 * may_umount - check if a mount point is busy
876 * @mnt: root of mount
878 * This is called to check if a mount point has any
879 * open files, pwds, chroots or sub mounts. If the
880 * mount has sub mounts this will return busy
881 * regardless of whether the sub mounts are busy.
883 * Doesn't take quota and stuff into account. IOW, in some cases it will
884 * give false negatives. The main reason why it's here is that we need
885 * a non-destructive way to look for easily umountable filesystems.
887 int may_umount(struct vfsmount
*mnt
)
890 spin_lock(&vfsmount_lock
);
891 if (propagate_mount_busy(mnt
, 2))
893 spin_unlock(&vfsmount_lock
);
897 EXPORT_SYMBOL(may_umount
);
899 void release_mounts(struct list_head
*head
)
901 struct vfsmount
*mnt
;
902 while (!list_empty(head
)) {
903 mnt
= list_first_entry(head
, struct vfsmount
, mnt_hash
);
904 list_del_init(&mnt
->mnt_hash
);
905 if (mnt
->mnt_parent
!= mnt
) {
906 struct dentry
*dentry
;
908 spin_lock(&vfsmount_lock
);
909 dentry
= mnt
->mnt_mountpoint
;
911 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
912 mnt
->mnt_parent
= mnt
;
914 spin_unlock(&vfsmount_lock
);
922 void umount_tree(struct vfsmount
*mnt
, int propagate
, struct list_head
*kill
)
926 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
927 list_move(&p
->mnt_hash
, kill
);
930 propagate_umount(kill
);
932 list_for_each_entry(p
, kill
, mnt_hash
) {
933 list_del_init(&p
->mnt_expire
);
934 list_del_init(&p
->mnt_list
);
935 __touch_mnt_namespace(p
->mnt_ns
);
937 list_del_init(&p
->mnt_child
);
938 if (p
->mnt_parent
!= p
) {
939 p
->mnt_parent
->mnt_ghosts
++;
940 p
->mnt_mountpoint
->d_mounted
--;
942 change_mnt_propagation(p
, MS_PRIVATE
);
946 static void shrink_submounts(struct vfsmount
*mnt
, struct list_head
*umounts
);
948 static int do_umount(struct vfsmount
*mnt
, int flags
)
950 struct super_block
*sb
= mnt
->mnt_sb
;
952 LIST_HEAD(umount_list
);
954 retval
= security_sb_umount(mnt
, flags
);
959 * Allow userspace to request a mountpoint be expired rather than
960 * unmounting unconditionally. Unmount only happens if:
961 * (1) the mark is already set (the mark is cleared by mntput())
962 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
964 if (flags
& MNT_EXPIRE
) {
965 if (mnt
== current
->fs
->root
.mnt
||
966 flags
& (MNT_FORCE
| MNT_DETACH
))
969 if (atomic_read(&mnt
->mnt_count
) != 2)
972 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
977 * If we may have to abort operations to get out of this
978 * mount, and they will themselves hold resources we must
979 * allow the fs to do things. In the Unix tradition of
980 * 'Gee thats tricky lets do it in userspace' the umount_begin
981 * might fail to complete on the first run through as other tasks
982 * must return, and the like. Thats for the mount program to worry
983 * about for the moment.
987 if (sb
->s_op
->umount_begin
)
988 sb
->s_op
->umount_begin(mnt
, flags
);
992 * No sense to grab the lock for this test, but test itself looks
993 * somewhat bogus. Suggestions for better replacement?
994 * Ho-hum... In principle, we might treat that as umount + switch
995 * to rootfs. GC would eventually take care of the old vfsmount.
996 * Actually it makes sense, especially if rootfs would contain a
997 * /reboot - static binary that would close all descriptors and
998 * call reboot(9). Then init(8) could umount root and exec /reboot.
1000 if (mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1002 * Special case for "unmounting" root ...
1003 * we just try to remount it readonly.
1005 down_write(&sb
->s_umount
);
1006 if (!(sb
->s_flags
& MS_RDONLY
)) {
1009 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1012 up_write(&sb
->s_umount
);
1016 down_write(&namespace_sem
);
1017 spin_lock(&vfsmount_lock
);
1020 if (!(flags
& MNT_DETACH
))
1021 shrink_submounts(mnt
, &umount_list
);
1024 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
1025 if (!list_empty(&mnt
->mnt_list
))
1026 umount_tree(mnt
, 1, &umount_list
);
1029 spin_unlock(&vfsmount_lock
);
1031 security_sb_umount_busy(mnt
);
1032 up_write(&namespace_sem
);
1033 release_mounts(&umount_list
);
1038 * Now umount can handle mount points as well as block devices.
1039 * This is important for filesystems which use unnamed block devices.
1041 * We now support a flag for forced unmount like the other 'big iron'
1042 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1045 asmlinkage
long sys_umount(char __user
* name
, int flags
)
1047 struct nameidata nd
;
1050 retval
= __user_walk(name
, LOOKUP_FOLLOW
, &nd
);
1054 if (nd
.path
.dentry
!= nd
.path
.mnt
->mnt_root
)
1056 if (!check_mnt(nd
.path
.mnt
))
1060 if (!capable(CAP_SYS_ADMIN
))
1063 retval
= do_umount(nd
.path
.mnt
, flags
);
1065 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1066 dput(nd
.path
.dentry
);
1067 mntput_no_expire(nd
.path
.mnt
);
1072 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1075 * The 2.0 compatible umount. No flags.
1077 asmlinkage
long sys_oldumount(char __user
* name
)
1079 return sys_umount(name
, 0);
1084 static int mount_is_safe(struct nameidata
*nd
)
1086 if (capable(CAP_SYS_ADMIN
))
1090 if (S_ISLNK(nd
->path
.dentry
->d_inode
->i_mode
))
1092 if (nd
->path
.dentry
->d_inode
->i_mode
& S_ISVTX
) {
1093 if (current
->uid
!= nd
->path
.dentry
->d_inode
->i_uid
)
1096 if (vfs_permission(nd
, MAY_WRITE
))
1102 static int lives_below_in_same_fs(struct dentry
*d
, struct dentry
*dentry
)
1107 if (d
== NULL
|| d
== d
->d_parent
)
1113 struct vfsmount
*copy_tree(struct vfsmount
*mnt
, struct dentry
*dentry
,
1116 struct vfsmount
*res
, *p
, *q
, *r
, *s
;
1119 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
1122 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1125 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1128 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1129 if (!lives_below_in_same_fs(r
->mnt_mountpoint
, dentry
))
1132 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1133 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
1134 s
= skip_mnt_tree(s
);
1137 while (p
!= s
->mnt_parent
) {
1143 path
.dentry
= p
->mnt_mountpoint
;
1144 q
= clone_mnt(p
, p
->mnt_root
, flag
);
1147 spin_lock(&vfsmount_lock
);
1148 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1149 attach_mnt(q
, &path
);
1150 spin_unlock(&vfsmount_lock
);
1156 LIST_HEAD(umount_list
);
1157 spin_lock(&vfsmount_lock
);
1158 umount_tree(res
, 0, &umount_list
);
1159 spin_unlock(&vfsmount_lock
);
1160 release_mounts(&umount_list
);
1165 struct vfsmount
*collect_mounts(struct vfsmount
*mnt
, struct dentry
*dentry
)
1167 struct vfsmount
*tree
;
1168 down_write(&namespace_sem
);
1169 tree
= copy_tree(mnt
, dentry
, CL_COPY_ALL
| CL_PRIVATE
);
1170 up_write(&namespace_sem
);
1174 void drop_collected_mounts(struct vfsmount
*mnt
)
1176 LIST_HEAD(umount_list
);
1177 down_write(&namespace_sem
);
1178 spin_lock(&vfsmount_lock
);
1179 umount_tree(mnt
, 0, &umount_list
);
1180 spin_unlock(&vfsmount_lock
);
1181 up_write(&namespace_sem
);
1182 release_mounts(&umount_list
);
1185 static void cleanup_group_ids(struct vfsmount
*mnt
, struct vfsmount
*end
)
1189 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1190 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1191 mnt_release_group_id(p
);
1195 static int invent_group_ids(struct vfsmount
*mnt
, bool recurse
)
1199 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1200 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1201 int err
= mnt_alloc_group_id(p
);
1203 cleanup_group_ids(mnt
, p
);
1213 * @source_mnt : mount tree to be attached
1214 * @nd : place the mount tree @source_mnt is attached
1215 * @parent_nd : if non-null, detach the source_mnt from its parent and
1216 * store the parent mount and mountpoint dentry.
1217 * (done when source_mnt is moved)
1219 * NOTE: in the table below explains the semantics when a source mount
1220 * of a given type is attached to a destination mount of a given type.
1221 * ---------------------------------------------------------------------------
1222 * | BIND MOUNT OPERATION |
1223 * |**************************************************************************
1224 * | source-->| shared | private | slave | unbindable |
1228 * |**************************************************************************
1229 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1231 * |non-shared| shared (+) | private | slave (*) | invalid |
1232 * ***************************************************************************
1233 * A bind operation clones the source mount and mounts the clone on the
1234 * destination mount.
1236 * (++) the cloned mount is propagated to all the mounts in the propagation
1237 * tree of the destination mount and the cloned mount is added to
1238 * the peer group of the source mount.
1239 * (+) the cloned mount is created under the destination mount and is marked
1240 * as shared. The cloned mount is added to the peer group of the source
1242 * (+++) the mount is propagated to all the mounts in the propagation tree
1243 * of the destination mount and the cloned mount is made slave
1244 * of the same master as that of the source mount. The cloned mount
1245 * is marked as 'shared and slave'.
1246 * (*) the cloned mount is made a slave of the same master as that of the
1249 * ---------------------------------------------------------------------------
1250 * | MOVE MOUNT OPERATION |
1251 * |**************************************************************************
1252 * | source-->| shared | private | slave | unbindable |
1256 * |**************************************************************************
1257 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1259 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1260 * ***************************************************************************
1262 * (+) the mount is moved to the destination. And is then propagated to
1263 * all the mounts in the propagation tree of the destination mount.
1264 * (+*) the mount is moved to the destination.
1265 * (+++) the mount is moved to the destination and is then propagated to
1266 * all the mounts belonging to the destination mount's propagation tree.
1267 * the mount is marked as 'shared and slave'.
1268 * (*) the mount continues to be a slave at the new location.
1270 * if the source mount is a tree, the operations explained above is
1271 * applied to each mount in the tree.
1272 * Must be called without spinlocks held, since this function can sleep
1275 static int attach_recursive_mnt(struct vfsmount
*source_mnt
,
1276 struct path
*path
, struct path
*parent_path
)
1278 LIST_HEAD(tree_list
);
1279 struct vfsmount
*dest_mnt
= path
->mnt
;
1280 struct dentry
*dest_dentry
= path
->dentry
;
1281 struct vfsmount
*child
, *p
;
1284 if (IS_MNT_SHARED(dest_mnt
)) {
1285 err
= invent_group_ids(source_mnt
, true);
1289 err
= propagate_mnt(dest_mnt
, dest_dentry
, source_mnt
, &tree_list
);
1291 goto out_cleanup_ids
;
1293 if (IS_MNT_SHARED(dest_mnt
)) {
1294 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
1298 spin_lock(&vfsmount_lock
);
1300 detach_mnt(source_mnt
, parent_path
);
1301 attach_mnt(source_mnt
, path
);
1302 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
1304 mnt_set_mountpoint(dest_mnt
, dest_dentry
, source_mnt
);
1305 commit_tree(source_mnt
);
1308 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
1309 list_del_init(&child
->mnt_hash
);
1312 spin_unlock(&vfsmount_lock
);
1316 if (IS_MNT_SHARED(dest_mnt
))
1317 cleanup_group_ids(source_mnt
, NULL
);
1322 static int graft_tree(struct vfsmount
*mnt
, struct path
*path
)
1325 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
1328 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1329 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
1333 mutex_lock(&path
->dentry
->d_inode
->i_mutex
);
1334 if (IS_DEADDIR(path
->dentry
->d_inode
))
1337 err
= security_sb_check_sb(mnt
, path
);
1342 if (IS_ROOT(path
->dentry
) || !d_unhashed(path
->dentry
))
1343 err
= attach_recursive_mnt(mnt
, path
, NULL
);
1345 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1347 security_sb_post_addmount(mnt
, path
);
1352 * recursively change the type of the mountpoint.
1353 * noinline this do_mount helper to save do_mount stack space.
1355 static noinline
int do_change_type(struct nameidata
*nd
, int flag
)
1357 struct vfsmount
*m
, *mnt
= nd
->path
.mnt
;
1358 int recurse
= flag
& MS_REC
;
1359 int type
= flag
& ~MS_REC
;
1362 if (!capable(CAP_SYS_ADMIN
))
1365 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
)
1368 down_write(&namespace_sem
);
1369 if (type
== MS_SHARED
) {
1370 err
= invent_group_ids(mnt
, recurse
);
1375 spin_lock(&vfsmount_lock
);
1376 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
1377 change_mnt_propagation(m
, type
);
1378 spin_unlock(&vfsmount_lock
);
1381 up_write(&namespace_sem
);
1386 * do loopback mount.
1387 * noinline this do_mount helper to save do_mount stack space.
1389 static noinline
int do_loopback(struct nameidata
*nd
, char *old_name
,
1392 struct nameidata old_nd
;
1393 struct vfsmount
*mnt
= NULL
;
1394 int err
= mount_is_safe(nd
);
1397 if (!old_name
|| !*old_name
)
1399 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
1403 down_write(&namespace_sem
);
1405 if (IS_MNT_UNBINDABLE(old_nd
.path
.mnt
))
1408 if (!check_mnt(nd
->path
.mnt
) || !check_mnt(old_nd
.path
.mnt
))
1413 mnt
= copy_tree(old_nd
.path
.mnt
, old_nd
.path
.dentry
, 0);
1415 mnt
= clone_mnt(old_nd
.path
.mnt
, old_nd
.path
.dentry
, 0);
1420 err
= graft_tree(mnt
, &nd
->path
);
1422 LIST_HEAD(umount_list
);
1423 spin_lock(&vfsmount_lock
);
1424 umount_tree(mnt
, 0, &umount_list
);
1425 spin_unlock(&vfsmount_lock
);
1426 release_mounts(&umount_list
);
1430 up_write(&namespace_sem
);
1431 path_put(&old_nd
.path
);
1435 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1438 int readonly_request
= 0;
1440 if (ms_flags
& MS_RDONLY
)
1441 readonly_request
= 1;
1442 if (readonly_request
== __mnt_is_readonly(mnt
))
1445 if (readonly_request
)
1446 error
= mnt_make_readonly(mnt
);
1448 __mnt_unmake_readonly(mnt
);
1453 * change filesystem flags. dir should be a physical root of filesystem.
1454 * If you've mounted a non-root directory somewhere and want to do remount
1455 * on it - tough luck.
1456 * noinline this do_mount helper to save do_mount stack space.
1458 static noinline
int do_remount(struct nameidata
*nd
, int flags
, int mnt_flags
,
1462 struct super_block
*sb
= nd
->path
.mnt
->mnt_sb
;
1464 if (!capable(CAP_SYS_ADMIN
))
1467 if (!check_mnt(nd
->path
.mnt
))
1470 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
)
1473 down_write(&sb
->s_umount
);
1474 if (flags
& MS_BIND
)
1475 err
= change_mount_flags(nd
->path
.mnt
, flags
);
1477 err
= do_remount_sb(sb
, flags
, data
, 0);
1479 nd
->path
.mnt
->mnt_flags
= mnt_flags
;
1480 up_write(&sb
->s_umount
);
1482 security_sb_post_remount(nd
->path
.mnt
, flags
, data
);
1486 static inline int tree_contains_unbindable(struct vfsmount
*mnt
)
1489 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1490 if (IS_MNT_UNBINDABLE(p
))
1497 * noinline this do_mount helper to save do_mount stack space.
1499 static noinline
int do_move_mount(struct nameidata
*nd
, char *old_name
)
1501 struct nameidata old_nd
;
1502 struct path parent_path
;
1505 if (!capable(CAP_SYS_ADMIN
))
1507 if (!old_name
|| !*old_name
)
1509 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
1513 down_write(&namespace_sem
);
1514 while (d_mountpoint(nd
->path
.dentry
) &&
1515 follow_down(&nd
->path
.mnt
, &nd
->path
.dentry
))
1518 if (!check_mnt(nd
->path
.mnt
) || !check_mnt(old_nd
.path
.mnt
))
1522 mutex_lock(&nd
->path
.dentry
->d_inode
->i_mutex
);
1523 if (IS_DEADDIR(nd
->path
.dentry
->d_inode
))
1526 if (!IS_ROOT(nd
->path
.dentry
) && d_unhashed(nd
->path
.dentry
))
1530 if (old_nd
.path
.dentry
!= old_nd
.path
.mnt
->mnt_root
)
1533 if (old_nd
.path
.mnt
== old_nd
.path
.mnt
->mnt_parent
)
1536 if (S_ISDIR(nd
->path
.dentry
->d_inode
->i_mode
) !=
1537 S_ISDIR(old_nd
.path
.dentry
->d_inode
->i_mode
))
1540 * Don't move a mount residing in a shared parent.
1542 if (old_nd
.path
.mnt
->mnt_parent
&&
1543 IS_MNT_SHARED(old_nd
.path
.mnt
->mnt_parent
))
1546 * Don't move a mount tree containing unbindable mounts to a destination
1547 * mount which is shared.
1549 if (IS_MNT_SHARED(nd
->path
.mnt
) &&
1550 tree_contains_unbindable(old_nd
.path
.mnt
))
1553 for (p
= nd
->path
.mnt
; p
->mnt_parent
!= p
; p
= p
->mnt_parent
)
1554 if (p
== old_nd
.path
.mnt
)
1557 err
= attach_recursive_mnt(old_nd
.path
.mnt
, &nd
->path
, &parent_path
);
1561 /* if the mount is moved, it should no longer be expire
1563 list_del_init(&old_nd
.path
.mnt
->mnt_expire
);
1565 mutex_unlock(&nd
->path
.dentry
->d_inode
->i_mutex
);
1567 up_write(&namespace_sem
);
1569 path_put(&parent_path
);
1570 path_put(&old_nd
.path
);
1575 * create a new mount for userspace and request it to be added into the
1577 * noinline this do_mount helper to save do_mount stack space.
1579 static noinline
int do_new_mount(struct nameidata
*nd
, char *type
, int flags
,
1580 int mnt_flags
, char *name
, void *data
)
1582 struct vfsmount
*mnt
;
1584 if (!type
|| !memchr(type
, 0, PAGE_SIZE
))
1587 /* we need capabilities... */
1588 if (!capable(CAP_SYS_ADMIN
))
1591 mnt
= do_kern_mount(type
, flags
, name
, data
);
1593 return PTR_ERR(mnt
);
1595 return do_add_mount(mnt
, nd
, mnt_flags
, NULL
);
1599 * add a mount into a namespace's mount tree
1600 * - provide the option of adding the new mount to an expiration list
1602 int do_add_mount(struct vfsmount
*newmnt
, struct nameidata
*nd
,
1603 int mnt_flags
, struct list_head
*fslist
)
1607 down_write(&namespace_sem
);
1608 /* Something was mounted here while we slept */
1609 while (d_mountpoint(nd
->path
.dentry
) &&
1610 follow_down(&nd
->path
.mnt
, &nd
->path
.dentry
))
1613 if (!check_mnt(nd
->path
.mnt
))
1616 /* Refuse the same filesystem on the same mount point */
1618 if (nd
->path
.mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1619 nd
->path
.mnt
->mnt_root
== nd
->path
.dentry
)
1623 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1626 newmnt
->mnt_flags
= mnt_flags
;
1627 if ((err
= graft_tree(newmnt
, &nd
->path
)))
1630 if (fslist
) /* add to the specified expiration list */
1631 list_add_tail(&newmnt
->mnt_expire
, fslist
);
1633 up_write(&namespace_sem
);
1637 up_write(&namespace_sem
);
1642 EXPORT_SYMBOL_GPL(do_add_mount
);
1645 * process a list of expirable mountpoints with the intent of discarding any
1646 * mountpoints that aren't in use and haven't been touched since last we came
1649 void mark_mounts_for_expiry(struct list_head
*mounts
)
1651 struct vfsmount
*mnt
, *next
;
1652 LIST_HEAD(graveyard
);
1655 if (list_empty(mounts
))
1658 down_write(&namespace_sem
);
1659 spin_lock(&vfsmount_lock
);
1661 /* extract from the expiration list every vfsmount that matches the
1662 * following criteria:
1663 * - only referenced by its parent vfsmount
1664 * - still marked for expiry (marked on the last call here; marks are
1665 * cleared by mntput())
1667 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
1668 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
1669 propagate_mount_busy(mnt
, 1))
1671 list_move(&mnt
->mnt_expire
, &graveyard
);
1673 while (!list_empty(&graveyard
)) {
1674 mnt
= list_first_entry(&graveyard
, struct vfsmount
, mnt_expire
);
1675 touch_mnt_namespace(mnt
->mnt_ns
);
1676 umount_tree(mnt
, 1, &umounts
);
1678 spin_unlock(&vfsmount_lock
);
1679 up_write(&namespace_sem
);
1681 release_mounts(&umounts
);
1684 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
1687 * Ripoff of 'select_parent()'
1689 * search the list of submounts for a given mountpoint, and move any
1690 * shrinkable submounts to the 'graveyard' list.
1692 static int select_submounts(struct vfsmount
*parent
, struct list_head
*graveyard
)
1694 struct vfsmount
*this_parent
= parent
;
1695 struct list_head
*next
;
1699 next
= this_parent
->mnt_mounts
.next
;
1701 while (next
!= &this_parent
->mnt_mounts
) {
1702 struct list_head
*tmp
= next
;
1703 struct vfsmount
*mnt
= list_entry(tmp
, struct vfsmount
, mnt_child
);
1706 if (!(mnt
->mnt_flags
& MNT_SHRINKABLE
))
1709 * Descend a level if the d_mounts list is non-empty.
1711 if (!list_empty(&mnt
->mnt_mounts
)) {
1716 if (!propagate_mount_busy(mnt
, 1)) {
1717 list_move_tail(&mnt
->mnt_expire
, graveyard
);
1722 * All done at this level ... ascend and resume the search
1724 if (this_parent
!= parent
) {
1725 next
= this_parent
->mnt_child
.next
;
1726 this_parent
= this_parent
->mnt_parent
;
1733 * process a list of expirable mountpoints with the intent of discarding any
1734 * submounts of a specific parent mountpoint
1736 static void shrink_submounts(struct vfsmount
*mnt
, struct list_head
*umounts
)
1738 LIST_HEAD(graveyard
);
1741 /* extract submounts of 'mountpoint' from the expiration list */
1742 while (select_submounts(mnt
, &graveyard
)) {
1743 while (!list_empty(&graveyard
)) {
1744 m
= list_first_entry(&graveyard
, struct vfsmount
,
1746 touch_mnt_namespace(mnt
->mnt_ns
);
1747 umount_tree(mnt
, 1, umounts
);
1753 * Some copy_from_user() implementations do not return the exact number of
1754 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1755 * Note that this function differs from copy_from_user() in that it will oops
1756 * on bad values of `to', rather than returning a short copy.
1758 static long exact_copy_from_user(void *to
, const void __user
* from
,
1762 const char __user
*f
= from
;
1765 if (!access_ok(VERIFY_READ
, from
, n
))
1769 if (__get_user(c
, f
)) {
1780 int copy_mount_options(const void __user
* data
, unsigned long *where
)
1790 if (!(page
= __get_free_page(GFP_KERNEL
)))
1793 /* We only care that *some* data at the address the user
1794 * gave us is valid. Just in case, we'll zero
1795 * the remainder of the page.
1797 /* copy_from_user cannot cross TASK_SIZE ! */
1798 size
= TASK_SIZE
- (unsigned long)data
;
1799 if (size
> PAGE_SIZE
)
1802 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
1808 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
1814 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1815 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1817 * data is a (void *) that can point to any structure up to
1818 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1819 * information (or be NULL).
1821 * Pre-0.97 versions of mount() didn't have a flags word.
1822 * When the flags word was introduced its top half was required
1823 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1824 * Therefore, if this magic number is present, it carries no information
1825 * and must be discarded.
1827 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
1828 unsigned long flags
, void *data_page
)
1830 struct nameidata nd
;
1835 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
1836 flags
&= ~MS_MGC_MSK
;
1838 /* Basic sanity checks */
1840 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
1842 if (dev_name
&& !memchr(dev_name
, 0, PAGE_SIZE
))
1846 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
1848 /* Separate the per-mountpoint flags */
1849 if (flags
& MS_NOSUID
)
1850 mnt_flags
|= MNT_NOSUID
;
1851 if (flags
& MS_NODEV
)
1852 mnt_flags
|= MNT_NODEV
;
1853 if (flags
& MS_NOEXEC
)
1854 mnt_flags
|= MNT_NOEXEC
;
1855 if (flags
& MS_NOATIME
)
1856 mnt_flags
|= MNT_NOATIME
;
1857 if (flags
& MS_NODIRATIME
)
1858 mnt_flags
|= MNT_NODIRATIME
;
1859 if (flags
& MS_RELATIME
)
1860 mnt_flags
|= MNT_RELATIME
;
1861 if (flags
& MS_RDONLY
)
1862 mnt_flags
|= MNT_READONLY
;
1864 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
|
1865 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
);
1867 /* ... and get the mountpoint */
1868 retval
= path_lookup(dir_name
, LOOKUP_FOLLOW
, &nd
);
1872 retval
= security_sb_mount(dev_name
, &nd
.path
,
1873 type_page
, flags
, data_page
);
1877 if (flags
& MS_REMOUNT
)
1878 retval
= do_remount(&nd
, flags
& ~MS_REMOUNT
, mnt_flags
,
1880 else if (flags
& MS_BIND
)
1881 retval
= do_loopback(&nd
, dev_name
, flags
& MS_REC
);
1882 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1883 retval
= do_change_type(&nd
, flags
);
1884 else if (flags
& MS_MOVE
)
1885 retval
= do_move_mount(&nd
, dev_name
);
1887 retval
= do_new_mount(&nd
, type_page
, flags
, mnt_flags
,
1888 dev_name
, data_page
);
1895 * Allocate a new namespace structure and populate it with contents
1896 * copied from the namespace of the passed in task structure.
1898 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
1899 struct fs_struct
*fs
)
1901 struct mnt_namespace
*new_ns
;
1902 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
, *altrootmnt
= NULL
;
1903 struct vfsmount
*p
, *q
;
1905 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
1907 return ERR_PTR(-ENOMEM
);
1909 atomic_set(&new_ns
->count
, 1);
1910 INIT_LIST_HEAD(&new_ns
->list
);
1911 init_waitqueue_head(&new_ns
->poll
);
1914 down_write(&namespace_sem
);
1915 /* First pass: copy the tree topology */
1916 new_ns
->root
= copy_tree(mnt_ns
->root
, mnt_ns
->root
->mnt_root
,
1917 CL_COPY_ALL
| CL_EXPIRE
);
1918 if (!new_ns
->root
) {
1919 up_write(&namespace_sem
);
1921 return ERR_PTR(-ENOMEM
);;
1923 spin_lock(&vfsmount_lock
);
1924 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
1925 spin_unlock(&vfsmount_lock
);
1928 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1929 * as belonging to new namespace. We have already acquired a private
1930 * fs_struct, so tsk->fs->lock is not needed.
1937 if (p
== fs
->root
.mnt
) {
1939 fs
->root
.mnt
= mntget(q
);
1941 if (p
== fs
->pwd
.mnt
) {
1943 fs
->pwd
.mnt
= mntget(q
);
1945 if (p
== fs
->altroot
.mnt
) {
1947 fs
->altroot
.mnt
= mntget(q
);
1950 p
= next_mnt(p
, mnt_ns
->root
);
1951 q
= next_mnt(q
, new_ns
->root
);
1953 up_write(&namespace_sem
);
1965 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
1966 struct fs_struct
*new_fs
)
1968 struct mnt_namespace
*new_ns
;
1973 if (!(flags
& CLONE_NEWNS
))
1976 new_ns
= dup_mnt_ns(ns
, new_fs
);
1982 asmlinkage
long sys_mount(char __user
* dev_name
, char __user
* dir_name
,
1983 char __user
* type
, unsigned long flags
,
1987 unsigned long data_page
;
1988 unsigned long type_page
;
1989 unsigned long dev_page
;
1992 retval
= copy_mount_options(type
, &type_page
);
1996 dir_page
= getname(dir_name
);
1997 retval
= PTR_ERR(dir_page
);
1998 if (IS_ERR(dir_page
))
2001 retval
= copy_mount_options(dev_name
, &dev_page
);
2005 retval
= copy_mount_options(data
, &data_page
);
2010 retval
= do_mount((char *)dev_page
, dir_page
, (char *)type_page
,
2011 flags
, (void *)data_page
);
2013 free_page(data_page
);
2016 free_page(dev_page
);
2020 free_page(type_page
);
2025 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
2026 * It can block. Requires the big lock held.
2028 void set_fs_root(struct fs_struct
*fs
, struct path
*path
)
2030 struct path old_root
;
2032 write_lock(&fs
->lock
);
2033 old_root
= fs
->root
;
2036 write_unlock(&fs
->lock
);
2037 if (old_root
.dentry
)
2038 path_put(&old_root
);
2042 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
2043 * It can block. Requires the big lock held.
2045 void set_fs_pwd(struct fs_struct
*fs
, struct path
*path
)
2047 struct path old_pwd
;
2049 write_lock(&fs
->lock
);
2053 write_unlock(&fs
->lock
);
2059 static void chroot_fs_refs(struct path
*old_root
, struct path
*new_root
)
2061 struct task_struct
*g
, *p
;
2062 struct fs_struct
*fs
;
2064 read_lock(&tasklist_lock
);
2065 do_each_thread(g
, p
) {
2069 atomic_inc(&fs
->count
);
2071 if (fs
->root
.dentry
== old_root
->dentry
2072 && fs
->root
.mnt
== old_root
->mnt
)
2073 set_fs_root(fs
, new_root
);
2074 if (fs
->pwd
.dentry
== old_root
->dentry
2075 && fs
->pwd
.mnt
== old_root
->mnt
)
2076 set_fs_pwd(fs
, new_root
);
2080 } while_each_thread(g
, p
);
2081 read_unlock(&tasklist_lock
);
2085 * pivot_root Semantics:
2086 * Moves the root file system of the current process to the directory put_old,
2087 * makes new_root as the new root file system of the current process, and sets
2088 * root/cwd of all processes which had them on the current root to new_root.
2091 * The new_root and put_old must be directories, and must not be on the
2092 * same file system as the current process root. The put_old must be
2093 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2094 * pointed to by put_old must yield the same directory as new_root. No other
2095 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2097 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2098 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2099 * in this situation.
2102 * - we don't move root/cwd if they are not at the root (reason: if something
2103 * cared enough to change them, it's probably wrong to force them elsewhere)
2104 * - it's okay to pick a root that isn't the root of a file system, e.g.
2105 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2106 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2109 asmlinkage
long sys_pivot_root(const char __user
* new_root
,
2110 const char __user
* put_old
)
2112 struct vfsmount
*tmp
;
2113 struct nameidata new_nd
, old_nd
;
2114 struct path parent_path
, root_parent
, root
;
2117 if (!capable(CAP_SYS_ADMIN
))
2120 error
= __user_walk(new_root
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
,
2125 if (!check_mnt(new_nd
.path
.mnt
))
2128 error
= __user_walk(put_old
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old_nd
);
2132 error
= security_sb_pivotroot(&old_nd
.path
, &new_nd
.path
);
2134 path_put(&old_nd
.path
);
2138 read_lock(¤t
->fs
->lock
);
2139 root
= current
->fs
->root
;
2140 path_get(¤t
->fs
->root
);
2141 read_unlock(¤t
->fs
->lock
);
2142 down_write(&namespace_sem
);
2143 mutex_lock(&old_nd
.path
.dentry
->d_inode
->i_mutex
);
2145 if (IS_MNT_SHARED(old_nd
.path
.mnt
) ||
2146 IS_MNT_SHARED(new_nd
.path
.mnt
->mnt_parent
) ||
2147 IS_MNT_SHARED(root
.mnt
->mnt_parent
))
2149 if (!check_mnt(root
.mnt
))
2152 if (IS_DEADDIR(new_nd
.path
.dentry
->d_inode
))
2154 if (d_unhashed(new_nd
.path
.dentry
) && !IS_ROOT(new_nd
.path
.dentry
))
2156 if (d_unhashed(old_nd
.path
.dentry
) && !IS_ROOT(old_nd
.path
.dentry
))
2159 if (new_nd
.path
.mnt
== root
.mnt
||
2160 old_nd
.path
.mnt
== root
.mnt
)
2161 goto out2
; /* loop, on the same file system */
2163 if (root
.mnt
->mnt_root
!= root
.dentry
)
2164 goto out2
; /* not a mountpoint */
2165 if (root
.mnt
->mnt_parent
== root
.mnt
)
2166 goto out2
; /* not attached */
2167 if (new_nd
.path
.mnt
->mnt_root
!= new_nd
.path
.dentry
)
2168 goto out2
; /* not a mountpoint */
2169 if (new_nd
.path
.mnt
->mnt_parent
== new_nd
.path
.mnt
)
2170 goto out2
; /* not attached */
2171 /* make sure we can reach put_old from new_root */
2172 tmp
= old_nd
.path
.mnt
;
2173 spin_lock(&vfsmount_lock
);
2174 if (tmp
!= new_nd
.path
.mnt
) {
2176 if (tmp
->mnt_parent
== tmp
)
2177 goto out3
; /* already mounted on put_old */
2178 if (tmp
->mnt_parent
== new_nd
.path
.mnt
)
2180 tmp
= tmp
->mnt_parent
;
2182 if (!is_subdir(tmp
->mnt_mountpoint
, new_nd
.path
.dentry
))
2184 } else if (!is_subdir(old_nd
.path
.dentry
, new_nd
.path
.dentry
))
2186 detach_mnt(new_nd
.path
.mnt
, &parent_path
);
2187 detach_mnt(root
.mnt
, &root_parent
);
2188 /* mount old root on put_old */
2189 attach_mnt(root
.mnt
, &old_nd
.path
);
2190 /* mount new_root on / */
2191 attach_mnt(new_nd
.path
.mnt
, &root_parent
);
2192 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2193 spin_unlock(&vfsmount_lock
);
2194 chroot_fs_refs(&root
, &new_nd
.path
);
2195 security_sb_post_pivotroot(&root
, &new_nd
.path
);
2197 path_put(&root_parent
);
2198 path_put(&parent_path
);
2200 mutex_unlock(&old_nd
.path
.dentry
->d_inode
->i_mutex
);
2201 up_write(&namespace_sem
);
2203 path_put(&old_nd
.path
);
2205 path_put(&new_nd
.path
);
2209 spin_unlock(&vfsmount_lock
);
2213 static void __init
init_mount_tree(void)
2215 struct vfsmount
*mnt
;
2216 struct mnt_namespace
*ns
;
2219 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
2221 panic("Can't create rootfs");
2222 ns
= kmalloc(sizeof(*ns
), GFP_KERNEL
);
2224 panic("Can't allocate initial namespace");
2225 atomic_set(&ns
->count
, 1);
2226 INIT_LIST_HEAD(&ns
->list
);
2227 init_waitqueue_head(&ns
->poll
);
2229 list_add(&mnt
->mnt_list
, &ns
->list
);
2233 init_task
.nsproxy
->mnt_ns
= ns
;
2236 root
.mnt
= ns
->root
;
2237 root
.dentry
= ns
->root
->mnt_root
;
2239 set_fs_pwd(current
->fs
, &root
);
2240 set_fs_root(current
->fs
, &root
);
2243 void __init
mnt_init(void)
2248 init_rwsem(&namespace_sem
);
2250 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct vfsmount
),
2251 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2253 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2255 if (!mount_hashtable
)
2256 panic("Failed to allocate mount hash table\n");
2258 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE
);
2260 for (u
= 0; u
< HASH_SIZE
; u
++)
2261 INIT_LIST_HEAD(&mount_hashtable
[u
]);
2265 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2267 fs_kobj
= kobject_create_and_add("fs", NULL
);
2269 printk(KERN_WARNING
"%s: kobj create error\n", __FUNCTION__
);
2274 void __put_mnt_ns(struct mnt_namespace
*ns
)
2276 struct vfsmount
*root
= ns
->root
;
2277 LIST_HEAD(umount_list
);
2279 spin_unlock(&vfsmount_lock
);
2280 down_write(&namespace_sem
);
2281 spin_lock(&vfsmount_lock
);
2282 umount_tree(root
, 0, &umount_list
);
2283 spin_unlock(&vfsmount_lock
);
2284 up_write(&namespace_sem
);
2285 release_mounts(&umount_list
);