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 <asm/uaccess.h>
31 #include <asm/unistd.h>
35 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
36 #define HASH_SIZE (1UL << HASH_SHIFT)
38 /* spinlock for vfsmount related operations, inplace of dcache_lock */
39 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(vfsmount_lock
);
43 static struct list_head
*mount_hashtable __read_mostly
;
44 static struct kmem_cache
*mnt_cache __read_mostly
;
45 static struct rw_semaphore namespace_sem
;
48 struct kobject
*fs_kobj
;
49 EXPORT_SYMBOL_GPL(fs_kobj
);
51 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
53 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
54 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
55 tmp
= tmp
+ (tmp
>> HASH_SHIFT
);
56 return tmp
& (HASH_SIZE
- 1);
59 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
61 struct vfsmount
*alloc_vfsmnt(const char *name
)
63 struct vfsmount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
65 atomic_set(&mnt
->mnt_count
, 1);
66 INIT_LIST_HEAD(&mnt
->mnt_hash
);
67 INIT_LIST_HEAD(&mnt
->mnt_child
);
68 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
69 INIT_LIST_HEAD(&mnt
->mnt_list
);
70 INIT_LIST_HEAD(&mnt
->mnt_expire
);
71 INIT_LIST_HEAD(&mnt
->mnt_share
);
72 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
73 INIT_LIST_HEAD(&mnt
->mnt_slave
);
74 atomic_set(&mnt
->__mnt_writers
, 0);
76 int size
= strlen(name
) + 1;
77 char *newname
= kmalloc(size
, GFP_KERNEL
);
79 memcpy(newname
, name
, size
);
80 mnt
->mnt_devname
= newname
;
88 * Most r/o checks on a fs are for operations that take
89 * discrete amounts of time, like a write() or unlink().
90 * We must keep track of when those operations start
91 * (for permission checks) and when they end, so that
92 * we can determine when writes are able to occur to
96 * __mnt_is_readonly: check whether a mount is read-only
97 * @mnt: the mount to check for its write status
99 * This shouldn't be used directly ouside of the VFS.
100 * It does not guarantee that the filesystem will stay
101 * r/w, just that it is right *now*. This can not and
102 * should not be used in place of IS_RDONLY(inode).
103 * mnt_want/drop_write() will _keep_ the filesystem
106 int __mnt_is_readonly(struct vfsmount
*mnt
)
108 if (mnt
->mnt_flags
& MNT_READONLY
)
110 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
114 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
118 * If holding multiple instances of this lock, they
119 * must be ordered by cpu number.
122 struct lock_class_key lock_class
; /* compiles out with !lockdep */
124 struct vfsmount
*mnt
;
125 } ____cacheline_aligned_in_smp
;
126 static DEFINE_PER_CPU(struct mnt_writer
, mnt_writers
);
128 static int __init
init_mnt_writers(void)
131 for_each_possible_cpu(cpu
) {
132 struct mnt_writer
*writer
= &per_cpu(mnt_writers
, cpu
);
133 spin_lock_init(&writer
->lock
);
134 lockdep_set_class(&writer
->lock
, &writer
->lock_class
);
139 fs_initcall(init_mnt_writers
);
141 static void unlock_mnt_writers(void)
144 struct mnt_writer
*cpu_writer
;
146 for_each_possible_cpu(cpu
) {
147 cpu_writer
= &per_cpu(mnt_writers
, cpu
);
148 spin_unlock(&cpu_writer
->lock
);
152 static inline void __clear_mnt_count(struct mnt_writer
*cpu_writer
)
154 if (!cpu_writer
->mnt
)
157 * This is in case anyone ever leaves an invalid,
158 * old ->mnt and a count of 0.
160 if (!cpu_writer
->count
)
162 atomic_add(cpu_writer
->count
, &cpu_writer
->mnt
->__mnt_writers
);
163 cpu_writer
->count
= 0;
166 * must hold cpu_writer->lock
168 static inline void use_cpu_writer_for_mount(struct mnt_writer
*cpu_writer
,
169 struct vfsmount
*mnt
)
171 if (cpu_writer
->mnt
== mnt
)
173 __clear_mnt_count(cpu_writer
);
174 cpu_writer
->mnt
= mnt
;
178 * Most r/o checks on a fs are for operations that take
179 * discrete amounts of time, like a write() or unlink().
180 * We must keep track of when those operations start
181 * (for permission checks) and when they end, so that
182 * we can determine when writes are able to occur to
186 * mnt_want_write - get write access to a mount
187 * @mnt: the mount on which to take a write
189 * This tells the low-level filesystem that a write is
190 * about to be performed to it, and makes sure that
191 * writes are allowed before returning success. When
192 * the write operation is finished, mnt_drop_write()
193 * must be called. This is effectively a refcount.
195 int mnt_want_write(struct vfsmount
*mnt
)
198 struct mnt_writer
*cpu_writer
;
200 cpu_writer
= &get_cpu_var(mnt_writers
);
201 spin_lock(&cpu_writer
->lock
);
202 if (__mnt_is_readonly(mnt
)) {
206 use_cpu_writer_for_mount(cpu_writer
, mnt
);
209 spin_unlock(&cpu_writer
->lock
);
210 put_cpu_var(mnt_writers
);
213 EXPORT_SYMBOL_GPL(mnt_want_write
);
215 static void lock_mnt_writers(void)
218 struct mnt_writer
*cpu_writer
;
220 for_each_possible_cpu(cpu
) {
221 cpu_writer
= &per_cpu(mnt_writers
, cpu
);
222 spin_lock(&cpu_writer
->lock
);
223 __clear_mnt_count(cpu_writer
);
224 cpu_writer
->mnt
= NULL
;
229 * These per-cpu write counts are not guaranteed to have
230 * matched increments and decrements on any given cpu.
231 * A file open()ed for write on one cpu and close()d on
232 * another cpu will imbalance this count. Make sure it
233 * does not get too far out of whack.
235 static void handle_write_count_underflow(struct vfsmount
*mnt
)
237 if (atomic_read(&mnt
->__mnt_writers
) >=
238 MNT_WRITER_UNDERFLOW_LIMIT
)
241 * It isn't necessary to hold all of the locks
242 * at the same time, but doing it this way makes
243 * us share a lot more code.
247 * vfsmount_lock is for mnt_flags.
249 spin_lock(&vfsmount_lock
);
251 * If coalescing the per-cpu writer counts did not
252 * get us back to a positive writer count, we have
255 if ((atomic_read(&mnt
->__mnt_writers
) < 0) &&
256 !(mnt
->mnt_flags
& MNT_IMBALANCED_WRITE_COUNT
)) {
257 printk(KERN_DEBUG
"leak detected on mount(%p) writers "
259 mnt
, atomic_read(&mnt
->__mnt_writers
));
261 /* use the flag to keep the dmesg spam down */
262 mnt
->mnt_flags
|= MNT_IMBALANCED_WRITE_COUNT
;
264 spin_unlock(&vfsmount_lock
);
265 unlock_mnt_writers();
269 * mnt_drop_write - give up write access to a mount
270 * @mnt: the mount on which to give up write access
272 * Tells the low-level filesystem that we are done
273 * performing writes to it. Must be matched with
274 * mnt_want_write() call above.
276 void mnt_drop_write(struct vfsmount
*mnt
)
278 int must_check_underflow
= 0;
279 struct mnt_writer
*cpu_writer
;
281 cpu_writer
= &get_cpu_var(mnt_writers
);
282 spin_lock(&cpu_writer
->lock
);
284 use_cpu_writer_for_mount(cpu_writer
, mnt
);
285 if (cpu_writer
->count
> 0) {
288 must_check_underflow
= 1;
289 atomic_dec(&mnt
->__mnt_writers
);
292 spin_unlock(&cpu_writer
->lock
);
294 * Logically, we could call this each time,
295 * but the __mnt_writers cacheline tends to
296 * be cold, and makes this expensive.
298 if (must_check_underflow
)
299 handle_write_count_underflow(mnt
);
301 * This could be done right after the spinlock
302 * is taken because the spinlock keeps us on
303 * the cpu, and disables preemption. However,
304 * putting it here bounds the amount that
305 * __mnt_writers can underflow. Without it,
306 * we could theoretically wrap __mnt_writers.
308 put_cpu_var(mnt_writers
);
310 EXPORT_SYMBOL_GPL(mnt_drop_write
);
312 static int mnt_make_readonly(struct vfsmount
*mnt
)
318 * With all the locks held, this value is stable
320 if (atomic_read(&mnt
->__mnt_writers
) > 0) {
325 * nobody can do a successful mnt_want_write() with all
326 * of the counts in MNT_DENIED_WRITE and the locks held.
328 spin_lock(&vfsmount_lock
);
330 mnt
->mnt_flags
|= MNT_READONLY
;
331 spin_unlock(&vfsmount_lock
);
333 unlock_mnt_writers();
337 static void __mnt_unmake_readonly(struct vfsmount
*mnt
)
339 spin_lock(&vfsmount_lock
);
340 mnt
->mnt_flags
&= ~MNT_READONLY
;
341 spin_unlock(&vfsmount_lock
);
344 int simple_set_mnt(struct vfsmount
*mnt
, struct super_block
*sb
)
347 mnt
->mnt_root
= dget(sb
->s_root
);
351 EXPORT_SYMBOL(simple_set_mnt
);
353 void free_vfsmnt(struct vfsmount
*mnt
)
355 kfree(mnt
->mnt_devname
);
356 kmem_cache_free(mnt_cache
, mnt
);
360 * find the first or last mount at @dentry on vfsmount @mnt depending on
361 * @dir. If @dir is set return the first mount else return the last mount.
363 struct vfsmount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
366 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
367 struct list_head
*tmp
= head
;
368 struct vfsmount
*p
, *found
= NULL
;
371 tmp
= dir
? tmp
->next
: tmp
->prev
;
375 p
= list_entry(tmp
, struct vfsmount
, mnt_hash
);
376 if (p
->mnt_parent
== mnt
&& p
->mnt_mountpoint
== dentry
) {
385 * lookup_mnt increments the ref count before returning
386 * the vfsmount struct.
388 struct vfsmount
*lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
390 struct vfsmount
*child_mnt
;
391 spin_lock(&vfsmount_lock
);
392 if ((child_mnt
= __lookup_mnt(mnt
, dentry
, 1)))
394 spin_unlock(&vfsmount_lock
);
398 static inline int check_mnt(struct vfsmount
*mnt
)
400 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
403 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
407 wake_up_interruptible(&ns
->poll
);
411 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
413 if (ns
&& ns
->event
!= event
) {
415 wake_up_interruptible(&ns
->poll
);
419 static void detach_mnt(struct vfsmount
*mnt
, struct path
*old_path
)
421 old_path
->dentry
= mnt
->mnt_mountpoint
;
422 old_path
->mnt
= mnt
->mnt_parent
;
423 mnt
->mnt_parent
= mnt
;
424 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
425 list_del_init(&mnt
->mnt_child
);
426 list_del_init(&mnt
->mnt_hash
);
427 old_path
->dentry
->d_mounted
--;
430 void mnt_set_mountpoint(struct vfsmount
*mnt
, struct dentry
*dentry
,
431 struct vfsmount
*child_mnt
)
433 child_mnt
->mnt_parent
= mntget(mnt
);
434 child_mnt
->mnt_mountpoint
= dget(dentry
);
438 static void attach_mnt(struct vfsmount
*mnt
, struct path
*path
)
440 mnt_set_mountpoint(path
->mnt
, path
->dentry
, mnt
);
441 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
442 hash(path
->mnt
, path
->dentry
));
443 list_add_tail(&mnt
->mnt_child
, &path
->mnt
->mnt_mounts
);
447 * the caller must hold vfsmount_lock
449 static void commit_tree(struct vfsmount
*mnt
)
451 struct vfsmount
*parent
= mnt
->mnt_parent
;
454 struct mnt_namespace
*n
= parent
->mnt_ns
;
456 BUG_ON(parent
== mnt
);
458 list_add_tail(&head
, &mnt
->mnt_list
);
459 list_for_each_entry(m
, &head
, mnt_list
)
461 list_splice(&head
, n
->list
.prev
);
463 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
464 hash(parent
, mnt
->mnt_mountpoint
));
465 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
466 touch_mnt_namespace(n
);
469 static struct vfsmount
*next_mnt(struct vfsmount
*p
, struct vfsmount
*root
)
471 struct list_head
*next
= p
->mnt_mounts
.next
;
472 if (next
== &p
->mnt_mounts
) {
476 next
= p
->mnt_child
.next
;
477 if (next
!= &p
->mnt_parent
->mnt_mounts
)
482 return list_entry(next
, struct vfsmount
, mnt_child
);
485 static struct vfsmount
*skip_mnt_tree(struct vfsmount
*p
)
487 struct list_head
*prev
= p
->mnt_mounts
.prev
;
488 while (prev
!= &p
->mnt_mounts
) {
489 p
= list_entry(prev
, struct vfsmount
, mnt_child
);
490 prev
= p
->mnt_mounts
.prev
;
495 static struct vfsmount
*clone_mnt(struct vfsmount
*old
, struct dentry
*root
,
498 struct super_block
*sb
= old
->mnt_sb
;
499 struct vfsmount
*mnt
= alloc_vfsmnt(old
->mnt_devname
);
502 mnt
->mnt_flags
= old
->mnt_flags
;
503 atomic_inc(&sb
->s_active
);
505 mnt
->mnt_root
= dget(root
);
506 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
507 mnt
->mnt_parent
= mnt
;
509 if (flag
& CL_SLAVE
) {
510 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
511 mnt
->mnt_master
= old
;
512 CLEAR_MNT_SHARED(mnt
);
513 } else if (!(flag
& CL_PRIVATE
)) {
514 if ((flag
& CL_PROPAGATION
) || IS_MNT_SHARED(old
))
515 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
516 if (IS_MNT_SLAVE(old
))
517 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
518 mnt
->mnt_master
= old
->mnt_master
;
520 if (flag
& CL_MAKE_SHARED
)
523 /* stick the duplicate mount on the same expiry list
524 * as the original if that was on one */
525 if (flag
& CL_EXPIRE
) {
526 if (!list_empty(&old
->mnt_expire
))
527 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
533 static inline void __mntput(struct vfsmount
*mnt
)
536 struct super_block
*sb
= mnt
->mnt_sb
;
538 * We don't have to hold all of the locks at the
539 * same time here because we know that we're the
540 * last reference to mnt and that no new writers
543 for_each_possible_cpu(cpu
) {
544 struct mnt_writer
*cpu_writer
= &per_cpu(mnt_writers
, cpu
);
545 if (cpu_writer
->mnt
!= mnt
)
547 spin_lock(&cpu_writer
->lock
);
548 atomic_add(cpu_writer
->count
, &mnt
->__mnt_writers
);
549 cpu_writer
->count
= 0;
551 * Might as well do this so that no one
552 * ever sees the pointer and expects
555 cpu_writer
->mnt
= NULL
;
556 spin_unlock(&cpu_writer
->lock
);
559 * This probably indicates that somebody messed
560 * up a mnt_want/drop_write() pair. If this
561 * happens, the filesystem was probably unable
562 * to make r/w->r/o transitions.
564 WARN_ON(atomic_read(&mnt
->__mnt_writers
));
567 deactivate_super(sb
);
570 void mntput_no_expire(struct vfsmount
*mnt
)
573 if (atomic_dec_and_lock(&mnt
->mnt_count
, &vfsmount_lock
)) {
574 if (likely(!mnt
->mnt_pinned
)) {
575 spin_unlock(&vfsmount_lock
);
579 atomic_add(mnt
->mnt_pinned
+ 1, &mnt
->mnt_count
);
581 spin_unlock(&vfsmount_lock
);
582 acct_auto_close_mnt(mnt
);
583 security_sb_umount_close(mnt
);
588 EXPORT_SYMBOL(mntput_no_expire
);
590 void mnt_pin(struct vfsmount
*mnt
)
592 spin_lock(&vfsmount_lock
);
594 spin_unlock(&vfsmount_lock
);
597 EXPORT_SYMBOL(mnt_pin
);
599 void mnt_unpin(struct vfsmount
*mnt
)
601 spin_lock(&vfsmount_lock
);
602 if (mnt
->mnt_pinned
) {
603 atomic_inc(&mnt
->mnt_count
);
606 spin_unlock(&vfsmount_lock
);
609 EXPORT_SYMBOL(mnt_unpin
);
611 static inline void mangle(struct seq_file
*m
, const char *s
)
613 seq_escape(m
, s
, " \t\n\\");
617 * Simple .show_options callback for filesystems which don't want to
618 * implement more complex mount option showing.
620 * See also save_mount_options().
622 int generic_show_options(struct seq_file
*m
, struct vfsmount
*mnt
)
624 const char *options
= mnt
->mnt_sb
->s_options
;
626 if (options
!= NULL
&& options
[0]) {
633 EXPORT_SYMBOL(generic_show_options
);
636 * If filesystem uses generic_show_options(), this function should be
637 * called from the fill_super() callback.
639 * The .remount_fs callback usually needs to be handled in a special
640 * way, to make sure, that previous options are not overwritten if the
643 * Also note, that if the filesystem's .remount_fs function doesn't
644 * reset all options to their default value, but changes only newly
645 * given options, then the displayed options will not reflect reality
648 void save_mount_options(struct super_block
*sb
, char *options
)
650 kfree(sb
->s_options
);
651 sb
->s_options
= kstrdup(options
, GFP_KERNEL
);
653 EXPORT_SYMBOL(save_mount_options
);
656 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
658 struct mnt_namespace
*n
= m
->private;
660 down_read(&namespace_sem
);
661 return seq_list_start(&n
->list
, *pos
);
664 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
666 struct mnt_namespace
*n
= m
->private;
668 return seq_list_next(v
, &n
->list
, pos
);
671 static void m_stop(struct seq_file
*m
, void *v
)
673 up_read(&namespace_sem
);
676 static int show_vfsmnt(struct seq_file
*m
, void *v
)
678 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
680 static struct proc_fs_info
{
684 { MS_SYNCHRONOUS
, ",sync" },
685 { MS_DIRSYNC
, ",dirsync" },
686 { MS_MANDLOCK
, ",mand" },
689 static struct proc_fs_info mnt_info
[] = {
690 { MNT_NOSUID
, ",nosuid" },
691 { MNT_NODEV
, ",nodev" },
692 { MNT_NOEXEC
, ",noexec" },
693 { MNT_NOATIME
, ",noatime" },
694 { MNT_NODIRATIME
, ",nodiratime" },
695 { MNT_RELATIME
, ",relatime" },
698 struct proc_fs_info
*fs_infop
;
699 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
701 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
703 seq_path(m
, &mnt_path
, " \t\n\\");
705 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
706 if (mnt
->mnt_sb
->s_subtype
&& mnt
->mnt_sb
->s_subtype
[0]) {
708 mangle(m
, mnt
->mnt_sb
->s_subtype
);
710 seq_puts(m
, __mnt_is_readonly(mnt
) ? " ro" : " rw");
711 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
712 if (mnt
->mnt_sb
->s_flags
& fs_infop
->flag
)
713 seq_puts(m
, fs_infop
->str
);
715 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
716 if (mnt
->mnt_flags
& fs_infop
->flag
)
717 seq_puts(m
, fs_infop
->str
);
719 if (mnt
->mnt_sb
->s_op
->show_options
)
720 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
721 seq_puts(m
, " 0 0\n");
725 struct seq_operations mounts_op
= {
732 static int show_vfsstat(struct seq_file
*m
, void *v
)
734 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
735 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
739 if (mnt
->mnt_devname
) {
740 seq_puts(m
, "device ");
741 mangle(m
, mnt
->mnt_devname
);
743 seq_puts(m
, "no device");
746 seq_puts(m
, " mounted on ");
747 seq_path(m
, &mnt_path
, " \t\n\\");
750 /* file system type */
751 seq_puts(m
, "with fstype ");
752 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
754 /* optional statistics */
755 if (mnt
->mnt_sb
->s_op
->show_stats
) {
757 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
764 struct seq_operations mountstats_op
= {
768 .show
= show_vfsstat
,
772 * may_umount_tree - check if a mount tree is busy
773 * @mnt: root of mount tree
775 * This is called to check if a tree of mounts has any
776 * open files, pwds, chroots or sub mounts that are
779 int may_umount_tree(struct vfsmount
*mnt
)
782 int minimum_refs
= 0;
785 spin_lock(&vfsmount_lock
);
786 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
787 actual_refs
+= atomic_read(&p
->mnt_count
);
790 spin_unlock(&vfsmount_lock
);
792 if (actual_refs
> minimum_refs
)
798 EXPORT_SYMBOL(may_umount_tree
);
801 * may_umount - check if a mount point is busy
802 * @mnt: root of mount
804 * This is called to check if a mount point has any
805 * open files, pwds, chroots or sub mounts. If the
806 * mount has sub mounts this will return busy
807 * regardless of whether the sub mounts are busy.
809 * Doesn't take quota and stuff into account. IOW, in some cases it will
810 * give false negatives. The main reason why it's here is that we need
811 * a non-destructive way to look for easily umountable filesystems.
813 int may_umount(struct vfsmount
*mnt
)
816 spin_lock(&vfsmount_lock
);
817 if (propagate_mount_busy(mnt
, 2))
819 spin_unlock(&vfsmount_lock
);
823 EXPORT_SYMBOL(may_umount
);
825 void release_mounts(struct list_head
*head
)
827 struct vfsmount
*mnt
;
828 while (!list_empty(head
)) {
829 mnt
= list_first_entry(head
, struct vfsmount
, mnt_hash
);
830 list_del_init(&mnt
->mnt_hash
);
831 if (mnt
->mnt_parent
!= mnt
) {
832 struct dentry
*dentry
;
834 spin_lock(&vfsmount_lock
);
835 dentry
= mnt
->mnt_mountpoint
;
837 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
838 mnt
->mnt_parent
= mnt
;
840 spin_unlock(&vfsmount_lock
);
848 void umount_tree(struct vfsmount
*mnt
, int propagate
, struct list_head
*kill
)
852 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
853 list_move(&p
->mnt_hash
, kill
);
856 propagate_umount(kill
);
858 list_for_each_entry(p
, kill
, mnt_hash
) {
859 list_del_init(&p
->mnt_expire
);
860 list_del_init(&p
->mnt_list
);
861 __touch_mnt_namespace(p
->mnt_ns
);
863 list_del_init(&p
->mnt_child
);
864 if (p
->mnt_parent
!= p
) {
865 p
->mnt_parent
->mnt_ghosts
++;
866 p
->mnt_mountpoint
->d_mounted
--;
868 change_mnt_propagation(p
, MS_PRIVATE
);
872 static void shrink_submounts(struct vfsmount
*mnt
, struct list_head
*umounts
);
874 static int do_umount(struct vfsmount
*mnt
, int flags
)
876 struct super_block
*sb
= mnt
->mnt_sb
;
878 LIST_HEAD(umount_list
);
880 retval
= security_sb_umount(mnt
, flags
);
885 * Allow userspace to request a mountpoint be expired rather than
886 * unmounting unconditionally. Unmount only happens if:
887 * (1) the mark is already set (the mark is cleared by mntput())
888 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
890 if (flags
& MNT_EXPIRE
) {
891 if (mnt
== current
->fs
->root
.mnt
||
892 flags
& (MNT_FORCE
| MNT_DETACH
))
895 if (atomic_read(&mnt
->mnt_count
) != 2)
898 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
903 * If we may have to abort operations to get out of this
904 * mount, and they will themselves hold resources we must
905 * allow the fs to do things. In the Unix tradition of
906 * 'Gee thats tricky lets do it in userspace' the umount_begin
907 * might fail to complete on the first run through as other tasks
908 * must return, and the like. Thats for the mount program to worry
909 * about for the moment.
913 if (sb
->s_op
->umount_begin
)
914 sb
->s_op
->umount_begin(mnt
, flags
);
918 * No sense to grab the lock for this test, but test itself looks
919 * somewhat bogus. Suggestions for better replacement?
920 * Ho-hum... In principle, we might treat that as umount + switch
921 * to rootfs. GC would eventually take care of the old vfsmount.
922 * Actually it makes sense, especially if rootfs would contain a
923 * /reboot - static binary that would close all descriptors and
924 * call reboot(9). Then init(8) could umount root and exec /reboot.
926 if (mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
928 * Special case for "unmounting" root ...
929 * we just try to remount it readonly.
931 down_write(&sb
->s_umount
);
932 if (!(sb
->s_flags
& MS_RDONLY
)) {
935 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
938 up_write(&sb
->s_umount
);
942 down_write(&namespace_sem
);
943 spin_lock(&vfsmount_lock
);
946 if (!(flags
& MNT_DETACH
))
947 shrink_submounts(mnt
, &umount_list
);
950 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
951 if (!list_empty(&mnt
->mnt_list
))
952 umount_tree(mnt
, 1, &umount_list
);
955 spin_unlock(&vfsmount_lock
);
957 security_sb_umount_busy(mnt
);
958 up_write(&namespace_sem
);
959 release_mounts(&umount_list
);
964 * Now umount can handle mount points as well as block devices.
965 * This is important for filesystems which use unnamed block devices.
967 * We now support a flag for forced unmount like the other 'big iron'
968 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
971 asmlinkage
long sys_umount(char __user
* name
, int flags
)
976 retval
= __user_walk(name
, LOOKUP_FOLLOW
, &nd
);
980 if (nd
.path
.dentry
!= nd
.path
.mnt
->mnt_root
)
982 if (!check_mnt(nd
.path
.mnt
))
986 if (!capable(CAP_SYS_ADMIN
))
989 retval
= do_umount(nd
.path
.mnt
, flags
);
991 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
992 dput(nd
.path
.dentry
);
993 mntput_no_expire(nd
.path
.mnt
);
998 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1001 * The 2.0 compatible umount. No flags.
1003 asmlinkage
long sys_oldumount(char __user
* name
)
1005 return sys_umount(name
, 0);
1010 static int mount_is_safe(struct nameidata
*nd
)
1012 if (capable(CAP_SYS_ADMIN
))
1016 if (S_ISLNK(nd
->path
.dentry
->d_inode
->i_mode
))
1018 if (nd
->path
.dentry
->d_inode
->i_mode
& S_ISVTX
) {
1019 if (current
->uid
!= nd
->path
.dentry
->d_inode
->i_uid
)
1022 if (vfs_permission(nd
, MAY_WRITE
))
1028 static int lives_below_in_same_fs(struct dentry
*d
, struct dentry
*dentry
)
1033 if (d
== NULL
|| d
== d
->d_parent
)
1039 struct vfsmount
*copy_tree(struct vfsmount
*mnt
, struct dentry
*dentry
,
1042 struct vfsmount
*res
, *p
, *q
, *r
, *s
;
1045 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
1048 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1051 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1054 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1055 if (!lives_below_in_same_fs(r
->mnt_mountpoint
, dentry
))
1058 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1059 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
1060 s
= skip_mnt_tree(s
);
1063 while (p
!= s
->mnt_parent
) {
1069 path
.dentry
= p
->mnt_mountpoint
;
1070 q
= clone_mnt(p
, p
->mnt_root
, flag
);
1073 spin_lock(&vfsmount_lock
);
1074 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1075 attach_mnt(q
, &path
);
1076 spin_unlock(&vfsmount_lock
);
1082 LIST_HEAD(umount_list
);
1083 spin_lock(&vfsmount_lock
);
1084 umount_tree(res
, 0, &umount_list
);
1085 spin_unlock(&vfsmount_lock
);
1086 release_mounts(&umount_list
);
1091 struct vfsmount
*collect_mounts(struct vfsmount
*mnt
, struct dentry
*dentry
)
1093 struct vfsmount
*tree
;
1094 down_write(&namespace_sem
);
1095 tree
= copy_tree(mnt
, dentry
, CL_COPY_ALL
| CL_PRIVATE
);
1096 up_write(&namespace_sem
);
1100 void drop_collected_mounts(struct vfsmount
*mnt
)
1102 LIST_HEAD(umount_list
);
1103 down_write(&namespace_sem
);
1104 spin_lock(&vfsmount_lock
);
1105 umount_tree(mnt
, 0, &umount_list
);
1106 spin_unlock(&vfsmount_lock
);
1107 up_write(&namespace_sem
);
1108 release_mounts(&umount_list
);
1112 * @source_mnt : mount tree to be attached
1113 * @nd : place the mount tree @source_mnt is attached
1114 * @parent_nd : if non-null, detach the source_mnt from its parent and
1115 * store the parent mount and mountpoint dentry.
1116 * (done when source_mnt is moved)
1118 * NOTE: in the table below explains the semantics when a source mount
1119 * of a given type is attached to a destination mount of a given type.
1120 * ---------------------------------------------------------------------------
1121 * | BIND MOUNT OPERATION |
1122 * |**************************************************************************
1123 * | source-->| shared | private | slave | unbindable |
1127 * |**************************************************************************
1128 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1130 * |non-shared| shared (+) | private | slave (*) | invalid |
1131 * ***************************************************************************
1132 * A bind operation clones the source mount and mounts the clone on the
1133 * destination mount.
1135 * (++) the cloned mount is propagated to all the mounts in the propagation
1136 * tree of the destination mount and the cloned mount is added to
1137 * the peer group of the source mount.
1138 * (+) the cloned mount is created under the destination mount and is marked
1139 * as shared. The cloned mount is added to the peer group of the source
1141 * (+++) the mount is propagated to all the mounts in the propagation tree
1142 * of the destination mount and the cloned mount is made slave
1143 * of the same master as that of the source mount. The cloned mount
1144 * is marked as 'shared and slave'.
1145 * (*) the cloned mount is made a slave of the same master as that of the
1148 * ---------------------------------------------------------------------------
1149 * | MOVE MOUNT OPERATION |
1150 * |**************************************************************************
1151 * | source-->| shared | private | slave | unbindable |
1155 * |**************************************************************************
1156 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1158 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1159 * ***************************************************************************
1161 * (+) the mount is moved to the destination. And is then propagated to
1162 * all the mounts in the propagation tree of the destination mount.
1163 * (+*) the mount is moved to the destination.
1164 * (+++) the mount is moved to the destination and is then propagated to
1165 * all the mounts belonging to the destination mount's propagation tree.
1166 * the mount is marked as 'shared and slave'.
1167 * (*) the mount continues to be a slave at the new location.
1169 * if the source mount is a tree, the operations explained above is
1170 * applied to each mount in the tree.
1171 * Must be called without spinlocks held, since this function can sleep
1174 static int attach_recursive_mnt(struct vfsmount
*source_mnt
,
1175 struct path
*path
, struct path
*parent_path
)
1177 LIST_HEAD(tree_list
);
1178 struct vfsmount
*dest_mnt
= path
->mnt
;
1179 struct dentry
*dest_dentry
= path
->dentry
;
1180 struct vfsmount
*child
, *p
;
1182 if (propagate_mnt(dest_mnt
, dest_dentry
, source_mnt
, &tree_list
))
1185 if (IS_MNT_SHARED(dest_mnt
)) {
1186 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
1190 spin_lock(&vfsmount_lock
);
1192 detach_mnt(source_mnt
, parent_path
);
1193 attach_mnt(source_mnt
, path
);
1194 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
1196 mnt_set_mountpoint(dest_mnt
, dest_dentry
, source_mnt
);
1197 commit_tree(source_mnt
);
1200 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
1201 list_del_init(&child
->mnt_hash
);
1204 spin_unlock(&vfsmount_lock
);
1208 static int graft_tree(struct vfsmount
*mnt
, struct path
*path
)
1211 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
1214 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1215 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
1219 mutex_lock(&path
->dentry
->d_inode
->i_mutex
);
1220 if (IS_DEADDIR(path
->dentry
->d_inode
))
1223 err
= security_sb_check_sb(mnt
, path
);
1228 if (IS_ROOT(path
->dentry
) || !d_unhashed(path
->dentry
))
1229 err
= attach_recursive_mnt(mnt
, path
, NULL
);
1231 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1233 security_sb_post_addmount(mnt
, path
);
1238 * recursively change the type of the mountpoint.
1239 * noinline this do_mount helper to save do_mount stack space.
1241 static noinline
int do_change_type(struct nameidata
*nd
, int flag
)
1243 struct vfsmount
*m
, *mnt
= nd
->path
.mnt
;
1244 int recurse
= flag
& MS_REC
;
1245 int type
= flag
& ~MS_REC
;
1247 if (!capable(CAP_SYS_ADMIN
))
1250 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
)
1253 down_write(&namespace_sem
);
1254 spin_lock(&vfsmount_lock
);
1255 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
1256 change_mnt_propagation(m
, type
);
1257 spin_unlock(&vfsmount_lock
);
1258 up_write(&namespace_sem
);
1263 * do loopback mount.
1264 * noinline this do_mount helper to save do_mount stack space.
1266 static noinline
int do_loopback(struct nameidata
*nd
, char *old_name
,
1269 struct nameidata old_nd
;
1270 struct vfsmount
*mnt
= NULL
;
1271 int err
= mount_is_safe(nd
);
1274 if (!old_name
|| !*old_name
)
1276 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
1280 down_write(&namespace_sem
);
1282 if (IS_MNT_UNBINDABLE(old_nd
.path
.mnt
))
1285 if (!check_mnt(nd
->path
.mnt
) || !check_mnt(old_nd
.path
.mnt
))
1290 mnt
= copy_tree(old_nd
.path
.mnt
, old_nd
.path
.dentry
, 0);
1292 mnt
= clone_mnt(old_nd
.path
.mnt
, old_nd
.path
.dentry
, 0);
1297 err
= graft_tree(mnt
, &nd
->path
);
1299 LIST_HEAD(umount_list
);
1300 spin_lock(&vfsmount_lock
);
1301 umount_tree(mnt
, 0, &umount_list
);
1302 spin_unlock(&vfsmount_lock
);
1303 release_mounts(&umount_list
);
1307 up_write(&namespace_sem
);
1308 path_put(&old_nd
.path
);
1312 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1315 int readonly_request
= 0;
1317 if (ms_flags
& MS_RDONLY
)
1318 readonly_request
= 1;
1319 if (readonly_request
== __mnt_is_readonly(mnt
))
1322 if (readonly_request
)
1323 error
= mnt_make_readonly(mnt
);
1325 __mnt_unmake_readonly(mnt
);
1330 * change filesystem flags. dir should be a physical root of filesystem.
1331 * If you've mounted a non-root directory somewhere and want to do remount
1332 * on it - tough luck.
1333 * noinline this do_mount helper to save do_mount stack space.
1335 static noinline
int do_remount(struct nameidata
*nd
, int flags
, int mnt_flags
,
1339 struct super_block
*sb
= nd
->path
.mnt
->mnt_sb
;
1341 if (!capable(CAP_SYS_ADMIN
))
1344 if (!check_mnt(nd
->path
.mnt
))
1347 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
)
1350 down_write(&sb
->s_umount
);
1351 if (flags
& MS_BIND
)
1352 err
= change_mount_flags(nd
->path
.mnt
, flags
);
1354 err
= do_remount_sb(sb
, flags
, data
, 0);
1356 nd
->path
.mnt
->mnt_flags
= mnt_flags
;
1357 up_write(&sb
->s_umount
);
1359 security_sb_post_remount(nd
->path
.mnt
, flags
, data
);
1363 static inline int tree_contains_unbindable(struct vfsmount
*mnt
)
1366 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1367 if (IS_MNT_UNBINDABLE(p
))
1374 * noinline this do_mount helper to save do_mount stack space.
1376 static noinline
int do_move_mount(struct nameidata
*nd
, char *old_name
)
1378 struct nameidata old_nd
;
1379 struct path parent_path
;
1382 if (!capable(CAP_SYS_ADMIN
))
1384 if (!old_name
|| !*old_name
)
1386 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
1390 down_write(&namespace_sem
);
1391 while (d_mountpoint(nd
->path
.dentry
) &&
1392 follow_down(&nd
->path
.mnt
, &nd
->path
.dentry
))
1395 if (!check_mnt(nd
->path
.mnt
) || !check_mnt(old_nd
.path
.mnt
))
1399 mutex_lock(&nd
->path
.dentry
->d_inode
->i_mutex
);
1400 if (IS_DEADDIR(nd
->path
.dentry
->d_inode
))
1403 if (!IS_ROOT(nd
->path
.dentry
) && d_unhashed(nd
->path
.dentry
))
1407 if (old_nd
.path
.dentry
!= old_nd
.path
.mnt
->mnt_root
)
1410 if (old_nd
.path
.mnt
== old_nd
.path
.mnt
->mnt_parent
)
1413 if (S_ISDIR(nd
->path
.dentry
->d_inode
->i_mode
) !=
1414 S_ISDIR(old_nd
.path
.dentry
->d_inode
->i_mode
))
1417 * Don't move a mount residing in a shared parent.
1419 if (old_nd
.path
.mnt
->mnt_parent
&&
1420 IS_MNT_SHARED(old_nd
.path
.mnt
->mnt_parent
))
1423 * Don't move a mount tree containing unbindable mounts to a destination
1424 * mount which is shared.
1426 if (IS_MNT_SHARED(nd
->path
.mnt
) &&
1427 tree_contains_unbindable(old_nd
.path
.mnt
))
1430 for (p
= nd
->path
.mnt
; p
->mnt_parent
!= p
; p
= p
->mnt_parent
)
1431 if (p
== old_nd
.path
.mnt
)
1434 err
= attach_recursive_mnt(old_nd
.path
.mnt
, &nd
->path
, &parent_path
);
1438 /* if the mount is moved, it should no longer be expire
1440 list_del_init(&old_nd
.path
.mnt
->mnt_expire
);
1442 mutex_unlock(&nd
->path
.dentry
->d_inode
->i_mutex
);
1444 up_write(&namespace_sem
);
1446 path_put(&parent_path
);
1447 path_put(&old_nd
.path
);
1452 * create a new mount for userspace and request it to be added into the
1454 * noinline this do_mount helper to save do_mount stack space.
1456 static noinline
int do_new_mount(struct nameidata
*nd
, char *type
, int flags
,
1457 int mnt_flags
, char *name
, void *data
)
1459 struct vfsmount
*mnt
;
1461 if (!type
|| !memchr(type
, 0, PAGE_SIZE
))
1464 /* we need capabilities... */
1465 if (!capable(CAP_SYS_ADMIN
))
1468 mnt
= do_kern_mount(type
, flags
, name
, data
);
1470 return PTR_ERR(mnt
);
1472 return do_add_mount(mnt
, nd
, mnt_flags
, NULL
);
1476 * add a mount into a namespace's mount tree
1477 * - provide the option of adding the new mount to an expiration list
1479 int do_add_mount(struct vfsmount
*newmnt
, struct nameidata
*nd
,
1480 int mnt_flags
, struct list_head
*fslist
)
1484 down_write(&namespace_sem
);
1485 /* Something was mounted here while we slept */
1486 while (d_mountpoint(nd
->path
.dentry
) &&
1487 follow_down(&nd
->path
.mnt
, &nd
->path
.dentry
))
1490 if (!check_mnt(nd
->path
.mnt
))
1493 /* Refuse the same filesystem on the same mount point */
1495 if (nd
->path
.mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1496 nd
->path
.mnt
->mnt_root
== nd
->path
.dentry
)
1500 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1503 newmnt
->mnt_flags
= mnt_flags
;
1504 if ((err
= graft_tree(newmnt
, &nd
->path
)))
1507 if (fslist
) /* add to the specified expiration list */
1508 list_add_tail(&newmnt
->mnt_expire
, fslist
);
1510 up_write(&namespace_sem
);
1514 up_write(&namespace_sem
);
1519 EXPORT_SYMBOL_GPL(do_add_mount
);
1522 * process a list of expirable mountpoints with the intent of discarding any
1523 * mountpoints that aren't in use and haven't been touched since last we came
1526 void mark_mounts_for_expiry(struct list_head
*mounts
)
1528 struct vfsmount
*mnt
, *next
;
1529 LIST_HEAD(graveyard
);
1532 if (list_empty(mounts
))
1535 down_write(&namespace_sem
);
1536 spin_lock(&vfsmount_lock
);
1538 /* extract from the expiration list every vfsmount that matches the
1539 * following criteria:
1540 * - only referenced by its parent vfsmount
1541 * - still marked for expiry (marked on the last call here; marks are
1542 * cleared by mntput())
1544 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
1545 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
1546 propagate_mount_busy(mnt
, 1))
1548 list_move(&mnt
->mnt_expire
, &graveyard
);
1550 while (!list_empty(&graveyard
)) {
1551 mnt
= list_first_entry(&graveyard
, struct vfsmount
, mnt_expire
);
1552 touch_mnt_namespace(mnt
->mnt_ns
);
1553 umount_tree(mnt
, 1, &umounts
);
1555 spin_unlock(&vfsmount_lock
);
1556 up_write(&namespace_sem
);
1558 release_mounts(&umounts
);
1561 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
1564 * Ripoff of 'select_parent()'
1566 * search the list of submounts for a given mountpoint, and move any
1567 * shrinkable submounts to the 'graveyard' list.
1569 static int select_submounts(struct vfsmount
*parent
, struct list_head
*graveyard
)
1571 struct vfsmount
*this_parent
= parent
;
1572 struct list_head
*next
;
1576 next
= this_parent
->mnt_mounts
.next
;
1578 while (next
!= &this_parent
->mnt_mounts
) {
1579 struct list_head
*tmp
= next
;
1580 struct vfsmount
*mnt
= list_entry(tmp
, struct vfsmount
, mnt_child
);
1583 if (!(mnt
->mnt_flags
& MNT_SHRINKABLE
))
1586 * Descend a level if the d_mounts list is non-empty.
1588 if (!list_empty(&mnt
->mnt_mounts
)) {
1593 if (!propagate_mount_busy(mnt
, 1)) {
1594 list_move_tail(&mnt
->mnt_expire
, graveyard
);
1599 * All done at this level ... ascend and resume the search
1601 if (this_parent
!= parent
) {
1602 next
= this_parent
->mnt_child
.next
;
1603 this_parent
= this_parent
->mnt_parent
;
1610 * process a list of expirable mountpoints with the intent of discarding any
1611 * submounts of a specific parent mountpoint
1613 static void shrink_submounts(struct vfsmount
*mnt
, struct list_head
*umounts
)
1615 LIST_HEAD(graveyard
);
1618 /* extract submounts of 'mountpoint' from the expiration list */
1619 while (select_submounts(mnt
, &graveyard
)) {
1620 while (!list_empty(&graveyard
)) {
1621 m
= list_first_entry(&graveyard
, struct vfsmount
,
1623 touch_mnt_namespace(mnt
->mnt_ns
);
1624 umount_tree(mnt
, 1, umounts
);
1630 * Some copy_from_user() implementations do not return the exact number of
1631 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1632 * Note that this function differs from copy_from_user() in that it will oops
1633 * on bad values of `to', rather than returning a short copy.
1635 static long exact_copy_from_user(void *to
, const void __user
* from
,
1639 const char __user
*f
= from
;
1642 if (!access_ok(VERIFY_READ
, from
, n
))
1646 if (__get_user(c
, f
)) {
1657 int copy_mount_options(const void __user
* data
, unsigned long *where
)
1667 if (!(page
= __get_free_page(GFP_KERNEL
)))
1670 /* We only care that *some* data at the address the user
1671 * gave us is valid. Just in case, we'll zero
1672 * the remainder of the page.
1674 /* copy_from_user cannot cross TASK_SIZE ! */
1675 size
= TASK_SIZE
- (unsigned long)data
;
1676 if (size
> PAGE_SIZE
)
1679 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
1685 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
1691 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1692 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1694 * data is a (void *) that can point to any structure up to
1695 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1696 * information (or be NULL).
1698 * Pre-0.97 versions of mount() didn't have a flags word.
1699 * When the flags word was introduced its top half was required
1700 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1701 * Therefore, if this magic number is present, it carries no information
1702 * and must be discarded.
1704 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
1705 unsigned long flags
, void *data_page
)
1707 struct nameidata nd
;
1712 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
1713 flags
&= ~MS_MGC_MSK
;
1715 /* Basic sanity checks */
1717 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
1719 if (dev_name
&& !memchr(dev_name
, 0, PAGE_SIZE
))
1723 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
1725 /* Separate the per-mountpoint flags */
1726 if (flags
& MS_NOSUID
)
1727 mnt_flags
|= MNT_NOSUID
;
1728 if (flags
& MS_NODEV
)
1729 mnt_flags
|= MNT_NODEV
;
1730 if (flags
& MS_NOEXEC
)
1731 mnt_flags
|= MNT_NOEXEC
;
1732 if (flags
& MS_NOATIME
)
1733 mnt_flags
|= MNT_NOATIME
;
1734 if (flags
& MS_NODIRATIME
)
1735 mnt_flags
|= MNT_NODIRATIME
;
1736 if (flags
& MS_RELATIME
)
1737 mnt_flags
|= MNT_RELATIME
;
1738 if (flags
& MS_RDONLY
)
1739 mnt_flags
|= MNT_READONLY
;
1741 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
|
1742 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
);
1744 /* ... and get the mountpoint */
1745 retval
= path_lookup(dir_name
, LOOKUP_FOLLOW
, &nd
);
1749 retval
= security_sb_mount(dev_name
, &nd
.path
,
1750 type_page
, flags
, data_page
);
1754 if (flags
& MS_REMOUNT
)
1755 retval
= do_remount(&nd
, flags
& ~MS_REMOUNT
, mnt_flags
,
1757 else if (flags
& MS_BIND
)
1758 retval
= do_loopback(&nd
, dev_name
, flags
& MS_REC
);
1759 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1760 retval
= do_change_type(&nd
, flags
);
1761 else if (flags
& MS_MOVE
)
1762 retval
= do_move_mount(&nd
, dev_name
);
1764 retval
= do_new_mount(&nd
, type_page
, flags
, mnt_flags
,
1765 dev_name
, data_page
);
1772 * Allocate a new namespace structure and populate it with contents
1773 * copied from the namespace of the passed in task structure.
1775 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
1776 struct fs_struct
*fs
)
1778 struct mnt_namespace
*new_ns
;
1779 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
, *altrootmnt
= NULL
;
1780 struct vfsmount
*p
, *q
;
1782 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
1784 return ERR_PTR(-ENOMEM
);
1786 atomic_set(&new_ns
->count
, 1);
1787 INIT_LIST_HEAD(&new_ns
->list
);
1788 init_waitqueue_head(&new_ns
->poll
);
1791 down_write(&namespace_sem
);
1792 /* First pass: copy the tree topology */
1793 new_ns
->root
= copy_tree(mnt_ns
->root
, mnt_ns
->root
->mnt_root
,
1794 CL_COPY_ALL
| CL_EXPIRE
);
1795 if (!new_ns
->root
) {
1796 up_write(&namespace_sem
);
1798 return ERR_PTR(-ENOMEM
);;
1800 spin_lock(&vfsmount_lock
);
1801 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
1802 spin_unlock(&vfsmount_lock
);
1805 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1806 * as belonging to new namespace. We have already acquired a private
1807 * fs_struct, so tsk->fs->lock is not needed.
1814 if (p
== fs
->root
.mnt
) {
1816 fs
->root
.mnt
= mntget(q
);
1818 if (p
== fs
->pwd
.mnt
) {
1820 fs
->pwd
.mnt
= mntget(q
);
1822 if (p
== fs
->altroot
.mnt
) {
1824 fs
->altroot
.mnt
= mntget(q
);
1827 p
= next_mnt(p
, mnt_ns
->root
);
1828 q
= next_mnt(q
, new_ns
->root
);
1830 up_write(&namespace_sem
);
1842 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
1843 struct fs_struct
*new_fs
)
1845 struct mnt_namespace
*new_ns
;
1850 if (!(flags
& CLONE_NEWNS
))
1853 new_ns
= dup_mnt_ns(ns
, new_fs
);
1859 asmlinkage
long sys_mount(char __user
* dev_name
, char __user
* dir_name
,
1860 char __user
* type
, unsigned long flags
,
1864 unsigned long data_page
;
1865 unsigned long type_page
;
1866 unsigned long dev_page
;
1869 retval
= copy_mount_options(type
, &type_page
);
1873 dir_page
= getname(dir_name
);
1874 retval
= PTR_ERR(dir_page
);
1875 if (IS_ERR(dir_page
))
1878 retval
= copy_mount_options(dev_name
, &dev_page
);
1882 retval
= copy_mount_options(data
, &data_page
);
1887 retval
= do_mount((char *)dev_page
, dir_page
, (char *)type_page
,
1888 flags
, (void *)data_page
);
1890 free_page(data_page
);
1893 free_page(dev_page
);
1897 free_page(type_page
);
1902 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1903 * It can block. Requires the big lock held.
1905 void set_fs_root(struct fs_struct
*fs
, struct path
*path
)
1907 struct path old_root
;
1909 write_lock(&fs
->lock
);
1910 old_root
= fs
->root
;
1913 write_unlock(&fs
->lock
);
1914 if (old_root
.dentry
)
1915 path_put(&old_root
);
1919 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1920 * It can block. Requires the big lock held.
1922 void set_fs_pwd(struct fs_struct
*fs
, struct path
*path
)
1924 struct path old_pwd
;
1926 write_lock(&fs
->lock
);
1930 write_unlock(&fs
->lock
);
1936 static void chroot_fs_refs(struct path
*old_root
, struct path
*new_root
)
1938 struct task_struct
*g
, *p
;
1939 struct fs_struct
*fs
;
1941 read_lock(&tasklist_lock
);
1942 do_each_thread(g
, p
) {
1946 atomic_inc(&fs
->count
);
1948 if (fs
->root
.dentry
== old_root
->dentry
1949 && fs
->root
.mnt
== old_root
->mnt
)
1950 set_fs_root(fs
, new_root
);
1951 if (fs
->pwd
.dentry
== old_root
->dentry
1952 && fs
->pwd
.mnt
== old_root
->mnt
)
1953 set_fs_pwd(fs
, new_root
);
1957 } while_each_thread(g
, p
);
1958 read_unlock(&tasklist_lock
);
1962 * pivot_root Semantics:
1963 * Moves the root file system of the current process to the directory put_old,
1964 * makes new_root as the new root file system of the current process, and sets
1965 * root/cwd of all processes which had them on the current root to new_root.
1968 * The new_root and put_old must be directories, and must not be on the
1969 * same file system as the current process root. The put_old must be
1970 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1971 * pointed to by put_old must yield the same directory as new_root. No other
1972 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1974 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1975 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1976 * in this situation.
1979 * - we don't move root/cwd if they are not at the root (reason: if something
1980 * cared enough to change them, it's probably wrong to force them elsewhere)
1981 * - it's okay to pick a root that isn't the root of a file system, e.g.
1982 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1983 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1986 asmlinkage
long sys_pivot_root(const char __user
* new_root
,
1987 const char __user
* put_old
)
1989 struct vfsmount
*tmp
;
1990 struct nameidata new_nd
, old_nd
;
1991 struct path parent_path
, root_parent
, root
;
1994 if (!capable(CAP_SYS_ADMIN
))
1997 error
= __user_walk(new_root
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
,
2002 if (!check_mnt(new_nd
.path
.mnt
))
2005 error
= __user_walk(put_old
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old_nd
);
2009 error
= security_sb_pivotroot(&old_nd
.path
, &new_nd
.path
);
2011 path_put(&old_nd
.path
);
2015 read_lock(¤t
->fs
->lock
);
2016 root
= current
->fs
->root
;
2017 path_get(¤t
->fs
->root
);
2018 read_unlock(¤t
->fs
->lock
);
2019 down_write(&namespace_sem
);
2020 mutex_lock(&old_nd
.path
.dentry
->d_inode
->i_mutex
);
2022 if (IS_MNT_SHARED(old_nd
.path
.mnt
) ||
2023 IS_MNT_SHARED(new_nd
.path
.mnt
->mnt_parent
) ||
2024 IS_MNT_SHARED(root
.mnt
->mnt_parent
))
2026 if (!check_mnt(root
.mnt
))
2029 if (IS_DEADDIR(new_nd
.path
.dentry
->d_inode
))
2031 if (d_unhashed(new_nd
.path
.dentry
) && !IS_ROOT(new_nd
.path
.dentry
))
2033 if (d_unhashed(old_nd
.path
.dentry
) && !IS_ROOT(old_nd
.path
.dentry
))
2036 if (new_nd
.path
.mnt
== root
.mnt
||
2037 old_nd
.path
.mnt
== root
.mnt
)
2038 goto out2
; /* loop, on the same file system */
2040 if (root
.mnt
->mnt_root
!= root
.dentry
)
2041 goto out2
; /* not a mountpoint */
2042 if (root
.mnt
->mnt_parent
== root
.mnt
)
2043 goto out2
; /* not attached */
2044 if (new_nd
.path
.mnt
->mnt_root
!= new_nd
.path
.dentry
)
2045 goto out2
; /* not a mountpoint */
2046 if (new_nd
.path
.mnt
->mnt_parent
== new_nd
.path
.mnt
)
2047 goto out2
; /* not attached */
2048 /* make sure we can reach put_old from new_root */
2049 tmp
= old_nd
.path
.mnt
;
2050 spin_lock(&vfsmount_lock
);
2051 if (tmp
!= new_nd
.path
.mnt
) {
2053 if (tmp
->mnt_parent
== tmp
)
2054 goto out3
; /* already mounted on put_old */
2055 if (tmp
->mnt_parent
== new_nd
.path
.mnt
)
2057 tmp
= tmp
->mnt_parent
;
2059 if (!is_subdir(tmp
->mnt_mountpoint
, new_nd
.path
.dentry
))
2061 } else if (!is_subdir(old_nd
.path
.dentry
, new_nd
.path
.dentry
))
2063 detach_mnt(new_nd
.path
.mnt
, &parent_path
);
2064 detach_mnt(root
.mnt
, &root_parent
);
2065 /* mount old root on put_old */
2066 attach_mnt(root
.mnt
, &old_nd
.path
);
2067 /* mount new_root on / */
2068 attach_mnt(new_nd
.path
.mnt
, &root_parent
);
2069 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2070 spin_unlock(&vfsmount_lock
);
2071 chroot_fs_refs(&root
, &new_nd
.path
);
2072 security_sb_post_pivotroot(&root
, &new_nd
.path
);
2074 path_put(&root_parent
);
2075 path_put(&parent_path
);
2077 mutex_unlock(&old_nd
.path
.dentry
->d_inode
->i_mutex
);
2078 up_write(&namespace_sem
);
2080 path_put(&old_nd
.path
);
2082 path_put(&new_nd
.path
);
2086 spin_unlock(&vfsmount_lock
);
2090 static void __init
init_mount_tree(void)
2092 struct vfsmount
*mnt
;
2093 struct mnt_namespace
*ns
;
2096 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
2098 panic("Can't create rootfs");
2099 ns
= kmalloc(sizeof(*ns
), GFP_KERNEL
);
2101 panic("Can't allocate initial namespace");
2102 atomic_set(&ns
->count
, 1);
2103 INIT_LIST_HEAD(&ns
->list
);
2104 init_waitqueue_head(&ns
->poll
);
2106 list_add(&mnt
->mnt_list
, &ns
->list
);
2110 init_task
.nsproxy
->mnt_ns
= ns
;
2113 root
.mnt
= ns
->root
;
2114 root
.dentry
= ns
->root
->mnt_root
;
2116 set_fs_pwd(current
->fs
, &root
);
2117 set_fs_root(current
->fs
, &root
);
2120 void __init
mnt_init(void)
2125 init_rwsem(&namespace_sem
);
2127 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct vfsmount
),
2128 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2130 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2132 if (!mount_hashtable
)
2133 panic("Failed to allocate mount hash table\n");
2135 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE
);
2137 for (u
= 0; u
< HASH_SIZE
; u
++)
2138 INIT_LIST_HEAD(&mount_hashtable
[u
]);
2142 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2144 fs_kobj
= kobject_create_and_add("fs", NULL
);
2146 printk(KERN_WARNING
"%s: kobj create error\n", __FUNCTION__
);
2151 void __put_mnt_ns(struct mnt_namespace
*ns
)
2153 struct vfsmount
*root
= ns
->root
;
2154 LIST_HEAD(umount_list
);
2156 spin_unlock(&vfsmount_lock
);
2157 down_write(&namespace_sem
);
2158 spin_lock(&vfsmount_lock
);
2159 umount_tree(root
, 0, &umount_list
);
2160 spin_unlock(&vfsmount_lock
);
2161 up_write(&namespace_sem
);
2162 release_mounts(&umount_list
);