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/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/acct.h> /* acct_auto_close_mnt */
20 #include <linux/ramfs.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 //#define UMOUNT_LOG //enable kernel layer unmount log when unmount fail
26 #include <linux/types.h>
28 #include <linux/proc_ns.h>
29 #include <linux/magic.h>
33 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
34 #define HASH_SIZE (1UL << HASH_SHIFT)
37 static DEFINE_IDA(mnt_id_ida
);
38 static DEFINE_IDA(mnt_group_ida
);
39 static DEFINE_SPINLOCK(mnt_id_lock
);
40 static int mnt_id_start
= 0;
41 static int mnt_group_start
= 1;
43 static struct list_head
*mount_hashtable __read_mostly
;
44 static struct list_head
*mountpoint_hashtable __read_mostly
;
45 static struct kmem_cache
*mnt_cache __read_mostly
;
46 static struct rw_semaphore namespace_sem
;
49 struct kobject
*fs_kobj
;
50 EXPORT_SYMBOL_GPL(fs_kobj
);
53 * vfsmount lock may be taken for read to prevent changes to the
54 * vfsmount hash, ie. during mountpoint lookups or walking back
57 * It should be taken for write in all cases where the vfsmount
58 * tree or hash is modified or when a vfsmount structure is modified.
60 DEFINE_BRLOCK(vfsmount_lock
);
62 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
64 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
65 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
66 tmp
= tmp
+ (tmp
>> HASH_SHIFT
);
67 return tmp
& (HASH_SIZE
- 1);
70 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
73 * allocation is serialized by namespace_sem, but we need the spinlock to
74 * serialize with freeing.
76 static int mnt_alloc_id(struct mount
*mnt
)
81 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
82 spin_lock(&mnt_id_lock
);
83 res
= ida_get_new_above(&mnt_id_ida
, mnt_id_start
, &mnt
->mnt_id
);
85 mnt_id_start
= mnt
->mnt_id
+ 1;
86 spin_unlock(&mnt_id_lock
);
93 static void mnt_free_id(struct mount
*mnt
)
96 spin_lock(&mnt_id_lock
);
97 ida_remove(&mnt_id_ida
, id
);
98 if (mnt_id_start
> id
)
100 spin_unlock(&mnt_id_lock
);
104 * Allocate a new peer group ID
106 * mnt_group_ida is protected by namespace_sem
108 static int mnt_alloc_group_id(struct mount
*mnt
)
112 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
115 res
= ida_get_new_above(&mnt_group_ida
,
119 mnt_group_start
= mnt
->mnt_group_id
+ 1;
125 * Release a peer group ID
127 void mnt_release_group_id(struct mount
*mnt
)
129 int id
= mnt
->mnt_group_id
;
130 ida_remove(&mnt_group_ida
, id
);
131 if (mnt_group_start
> id
)
132 mnt_group_start
= id
;
133 mnt
->mnt_group_id
= 0;
137 * vfsmount lock must be held for read
140 #define UMOUNT_Partition "/emmc@usrdata"
141 struct record_ref_count
{
143 char name
[TASK_COMM_LEN
];
145 struct record_ref_count
*next
;
148 struct record_ref_count
*ref_head
= NULL
;
149 struct record_ref_count
*ref_current
= NULL
;
150 struct record_ref_count
*ref_prev
= NULL
;
151 int s_total_count
= 0;
153 static inline void mnt_add_count(struct mount
*mnt
, int n
)
156 int print_link_list
=0;
161 if (strcmp(UMOUNT_Partition
,mnt
->mnt_devname
)==0)
163 //if (strcmp("mobile_log_d",current->comm)!=0)
165 //if (current->pid < 100) //((current->pid < 70) && (current->pid > 60))
167 //printk("Ahsin n=%d current->pid=%d name=%s \n",n,current->pid,current->comm);
168 spin_lock(&mnt_id_lock
);
169 if (ref_head
== NULL
) //linked list head (start)
171 printk("Ahsin link list init mnt_get_count=%d \n",mnt_get_count(mnt
));
173 ref_current
= kmalloc(sizeof(struct record_ref_count
), GFP_KERNEL
);
174 if (ref_current
== NULL
)
175 printk("Ahsin can't allocate memory for ref_current /n");
177 ref_current
->next
= NULL
;
178 ref_current
->pid
= current
->pid
;
179 strncpy(ref_current
->name
, current
->comm
, TASK_COMM_LEN
-1);
180 ref_current
->name
[TASK_COMM_LEN
-1] = '\0';
181 ref_current
->count
= n
;
182 s_total_count
= s_total_count
+ n
;
183 ref_head
= ref_current
;
185 printk("Ahsin ref_head == NULL pid=%d name=%s counter=%d n=%d \n",ref_current
->pid
,ref_current
->name
,ref_current
->count
,n
);
187 else //check exist first and then add linked list or modify counter
190 while(ref_prev
!= NULL
)
192 //printk("Ahsin PID= %d, Name= %s, Count= %d n=%d current->pid=%d \n", ref_prev->pid, ref_prev->name, ref_prev->count,n,current->pid);
193 if (strcmp(ref_prev
->name
,current
->comm
)==0) //(ref_prev->pid==current->pid)//exist and find, modify counter
195 ref_prev
->count
= ref_prev
->count
+ n
;
196 s_total_count
= s_total_count
+ n
;
197 //printk("Ahsin (ref_prev->name,current->comm) pid=%d name=%s counter=%d n=%d \n",ref_prev->pid,ref_prev->name,ref_prev->count,n);
202 if (ref_prev
->next
!= NULL
)
203 ref_prev
= ref_prev
->next
;
205 { // end of link list
206 ref_current
= kmalloc(sizeof(struct record_ref_count
), GFP_KERNEL
);
207 if (ref_current
== NULL
)
208 printk("Ahsin can't allocate memory for ref_prev /n");
210 ref_current
->next
= NULL
;
211 ref_current
->pid
= current
->pid
;
212 strncpy(ref_current
->name
, current
->comm
, TASK_COMM_LEN
-1);
213 ref_current
->name
[TASK_COMM_LEN
-1] = '\0';
214 ref_current
->count
= n
;
215 s_total_count
= s_total_count
+ n
;
216 ref_prev
->next
= ref_current
;
217 //printk("Ahsin new node(end of link list) pid=%d name=%s counter=%d n=%d \n",ref_current->pid,ref_current->name,ref_current->count,n);
224 spin_unlock(&mnt_id_lock
);
234 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
237 if (strcmp(UMOUNT_Partition
,mnt
->mnt_devname
)==0)
239 if (strcmp("mobile_log_d",current
->comm
)!=0)
241 printk("Ahsin s_total_count=%d mnt_get_count=%d n=%d current->pid=%d \n",s_total_count
,mnt_get_count(mnt
),n
,current
->pid
);
242 //if (current->pid < 100)
245 spin_lock(&mnt_id_lock
);
246 ref_current
= ref_head
;
247 while(ref_current
!= NULL
)
249 if (ref_current
->count
)
251 print_link_list
= print_link_list
+ ref_current
->count
;
252 //printk("Ahsin PID= %d, Name = %s, Count= %d \n", ref_current->pid, ref_current->name, ref_current->count);
254 ref_current
= ref_current
->next
;
256 spin_unlock(&mnt_id_lock
);
257 printk("Ahsin print_link_list=%d \n",print_link_list
);
271 * vfsmount lock must be held for write
273 unsigned int mnt_get_count(struct mount
*mnt
)
276 unsigned int count
= 0;
279 for_each_possible_cpu(cpu
) {
280 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
285 return mnt
->mnt_count
;
289 static struct mount
*alloc_vfsmnt(const char *name
)
291 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
295 err
= mnt_alloc_id(mnt
);
300 mnt
->mnt_devname
= kstrdup(name
, GFP_KERNEL
);
301 if (!mnt
->mnt_devname
)
306 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
308 goto out_free_devname
;
311 if (strcmp(UMOUNT_Partition
,mnt
->mnt_devname
)==0)
313 printk("Ahsin alloc_vfsmnt current->pid=%d name=%s \n",current
->pid
,current
->comm
);
316 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
319 mnt
->mnt_writers
= 0;
322 INIT_LIST_HEAD(&mnt
->mnt_hash
);
323 INIT_LIST_HEAD(&mnt
->mnt_child
);
324 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
325 INIT_LIST_HEAD(&mnt
->mnt_list
);
326 INIT_LIST_HEAD(&mnt
->mnt_expire
);
327 INIT_LIST_HEAD(&mnt
->mnt_share
);
328 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
329 INIT_LIST_HEAD(&mnt
->mnt_slave
);
330 #ifdef CONFIG_FSNOTIFY
331 INIT_HLIST_HEAD(&mnt
->mnt_fsnotify_marks
);
338 kfree(mnt
->mnt_devname
);
343 kmem_cache_free(mnt_cache
, mnt
);
348 * Most r/o checks on a fs are for operations that take
349 * discrete amounts of time, like a write() or unlink().
350 * We must keep track of when those operations start
351 * (for permission checks) and when they end, so that
352 * we can determine when writes are able to occur to
356 * __mnt_is_readonly: check whether a mount is read-only
357 * @mnt: the mount to check for its write status
359 * This shouldn't be used directly ouside of the VFS.
360 * It does not guarantee that the filesystem will stay
361 * r/w, just that it is right *now*. This can not and
362 * should not be used in place of IS_RDONLY(inode).
363 * mnt_want/drop_write() will _keep_ the filesystem
366 int __mnt_is_readonly(struct vfsmount
*mnt
)
368 if (mnt
->mnt_flags
& MNT_READONLY
)
370 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
374 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
376 static inline void mnt_inc_writers(struct mount
*mnt
)
379 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
385 static inline void mnt_dec_writers(struct mount
*mnt
)
388 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
394 static unsigned int mnt_get_writers(struct mount
*mnt
)
397 unsigned int count
= 0;
400 for_each_possible_cpu(cpu
) {
401 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
406 return mnt
->mnt_writers
;
410 static int mnt_is_readonly(struct vfsmount
*mnt
)
412 if (mnt
->mnt_sb
->s_readonly_remount
)
414 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
416 return __mnt_is_readonly(mnt
);
420 * Most r/o & frozen checks on a fs are for operations that take discrete
421 * amounts of time, like a write() or unlink(). We must keep track of when
422 * those operations start (for permission checks) and when they end, so that we
423 * can determine when writes are able to occur to a filesystem.
426 * __mnt_want_write - get write access to a mount without freeze protection
427 * @m: the mount on which to take a write
429 * This tells the low-level filesystem that a write is about to be performed to
430 * it, and makes sure that writes are allowed (mnt it read-write) before
431 * returning success. This operation does not protect against filesystem being
432 * frozen. When the write operation is finished, __mnt_drop_write() must be
433 * called. This is effectively a refcount.
435 int __mnt_want_write(struct vfsmount
*m
)
437 struct mount
*mnt
= real_mount(m
);
441 mnt_inc_writers(mnt
);
443 * The store to mnt_inc_writers must be visible before we pass
444 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
445 * incremented count after it has set MNT_WRITE_HOLD.
448 while (ACCESS_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
451 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
452 * be set to match its requirements. So we must not load that until
453 * MNT_WRITE_HOLD is cleared.
456 if (mnt_is_readonly(m
)) {
457 mnt_dec_writers(mnt
);
466 * mnt_want_write - get write access to a mount
467 * @m: the mount on which to take a write
469 * This tells the low-level filesystem that a write is about to be performed to
470 * it, and makes sure that writes are allowed (mount is read-write, filesystem
471 * is not frozen) before returning success. When the write operation is
472 * finished, mnt_drop_write() must be called. This is effectively a refcount.
474 int mnt_want_write(struct vfsmount
*m
)
478 sb_start_write(m
->mnt_sb
);
479 ret
= __mnt_want_write(m
);
481 sb_end_write(m
->mnt_sb
);
484 EXPORT_SYMBOL_GPL(mnt_want_write
);
487 * mnt_clone_write - get write access to a mount
488 * @mnt: the mount on which to take a write
490 * This is effectively like mnt_want_write, except
491 * it must only be used to take an extra write reference
492 * on a mountpoint that we already know has a write reference
493 * on it. This allows some optimisation.
495 * After finished, mnt_drop_write must be called as usual to
496 * drop the reference.
498 int mnt_clone_write(struct vfsmount
*mnt
)
500 /* superblock may be r/o */
501 if (__mnt_is_readonly(mnt
))
504 mnt_inc_writers(real_mount(mnt
));
508 EXPORT_SYMBOL_GPL(mnt_clone_write
);
511 * __mnt_want_write_file - get write access to a file's mount
512 * @file: the file who's mount on which to take a write
514 * This is like __mnt_want_write, but it takes a file and can
515 * do some optimisations if the file is open for write already
517 int __mnt_want_write_file(struct file
*file
)
519 struct inode
*inode
= file_inode(file
);
521 if (!(file
->f_mode
& FMODE_WRITE
) || special_file(inode
->i_mode
))
522 return __mnt_want_write(file
->f_path
.mnt
);
524 return mnt_clone_write(file
->f_path
.mnt
);
528 * mnt_want_write_file - get write access to a file's mount
529 * @file: the file who's mount on which to take a write
531 * This is like mnt_want_write, but it takes a file and can
532 * do some optimisations if the file is open for write already
534 int mnt_want_write_file(struct file
*file
)
538 sb_start_write(file
->f_path
.mnt
->mnt_sb
);
539 ret
= __mnt_want_write_file(file
);
541 sb_end_write(file
->f_path
.mnt
->mnt_sb
);
544 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
547 * __mnt_drop_write - give up write access to a mount
548 * @mnt: the mount on which to give up write access
550 * Tells the low-level filesystem that we are done
551 * performing writes to it. Must be matched with
552 * __mnt_want_write() call above.
554 void __mnt_drop_write(struct vfsmount
*mnt
)
557 mnt_dec_writers(real_mount(mnt
));
562 * mnt_drop_write - give up write access to a mount
563 * @mnt: the mount on which to give up write access
565 * Tells the low-level filesystem that we are done performing writes to it and
566 * also allows filesystem to be frozen again. Must be matched with
567 * mnt_want_write() call above.
569 void mnt_drop_write(struct vfsmount
*mnt
)
571 __mnt_drop_write(mnt
);
572 sb_end_write(mnt
->mnt_sb
);
574 EXPORT_SYMBOL_GPL(mnt_drop_write
);
576 void __mnt_drop_write_file(struct file
*file
)
578 __mnt_drop_write(file
->f_path
.mnt
);
581 void mnt_drop_write_file(struct file
*file
)
583 mnt_drop_write(file
->f_path
.mnt
);
585 EXPORT_SYMBOL(mnt_drop_write_file
);
587 static int mnt_make_readonly(struct mount
*mnt
)
591 br_write_lock(&vfsmount_lock
);
592 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
594 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
595 * should be visible before we do.
600 * With writers on hold, if this value is zero, then there are
601 * definitely no active writers (although held writers may subsequently
602 * increment the count, they'll have to wait, and decrement it after
603 * seeing MNT_READONLY).
605 * It is OK to have counter incremented on one CPU and decremented on
606 * another: the sum will add up correctly. The danger would be when we
607 * sum up each counter, if we read a counter before it is incremented,
608 * but then read another CPU's count which it has been subsequently
609 * decremented from -- we would see more decrements than we should.
610 * MNT_WRITE_HOLD protects against this scenario, because
611 * mnt_want_write first increments count, then smp_mb, then spins on
612 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
613 * we're counting up here.
615 if (mnt_get_writers(mnt
) > 0)
618 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
620 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
621 * that become unheld will see MNT_READONLY.
624 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
625 br_write_unlock(&vfsmount_lock
);
629 static void __mnt_unmake_readonly(struct mount
*mnt
)
631 br_write_lock(&vfsmount_lock
);
632 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
633 br_write_unlock(&vfsmount_lock
);
636 int sb_prepare_remount_readonly(struct super_block
*sb
)
641 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
642 if (atomic_long_read(&sb
->s_remove_count
))
645 br_write_lock(&vfsmount_lock
);
646 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
647 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
648 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
650 if (mnt_get_writers(mnt
) > 0) {
656 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
660 sb
->s_readonly_remount
= 1;
663 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
664 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
665 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
667 br_write_unlock(&vfsmount_lock
);
672 static void free_vfsmnt(struct mount
*mnt
)
674 kfree(mnt
->mnt_devname
);
677 free_percpu(mnt
->mnt_pcp
);
679 kmem_cache_free(mnt_cache
, mnt
);
683 * find the first or last mount at @dentry on vfsmount @mnt depending on
684 * @dir. If @dir is set return the first mount else return the last mount.
685 * vfsmount_lock must be held for read or write.
687 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
690 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
691 struct list_head
*tmp
= head
;
692 struct mount
*p
, *found
= NULL
;
695 tmp
= dir
? tmp
->next
: tmp
->prev
;
699 p
= list_entry(tmp
, struct mount
, mnt_hash
);
700 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
) {
709 * lookup_mnt - Return the first child mount mounted at path
711 * "First" means first mounted chronologically. If you create the
714 * mount /dev/sda1 /mnt
715 * mount /dev/sda2 /mnt
716 * mount /dev/sda3 /mnt
718 * Then lookup_mnt() on the base /mnt dentry in the root mount will
719 * return successively the root dentry and vfsmount of /dev/sda1, then
720 * /dev/sda2, then /dev/sda3, then NULL.
722 * lookup_mnt takes a reference to the found vfsmount.
724 struct vfsmount
*lookup_mnt(struct path
*path
)
726 struct mount
*child_mnt
;
728 br_read_lock(&vfsmount_lock
);
729 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
731 mnt_add_count(child_mnt
, 1);
732 br_read_unlock(&vfsmount_lock
);
733 return &child_mnt
->mnt
;
735 br_read_unlock(&vfsmount_lock
);
740 static struct mountpoint
*new_mountpoint(struct dentry
*dentry
)
742 struct list_head
*chain
= mountpoint_hashtable
+ hash(NULL
, dentry
);
743 struct mountpoint
*mp
;
745 list_for_each_entry(mp
, chain
, m_hash
) {
746 if (mp
->m_dentry
== dentry
) {
747 /* might be worth a WARN_ON() */
748 if (d_unlinked(dentry
))
749 return ERR_PTR(-ENOENT
);
755 mp
= kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
757 return ERR_PTR(-ENOMEM
);
759 spin_lock(&dentry
->d_lock
);
760 if (d_unlinked(dentry
)) {
761 spin_unlock(&dentry
->d_lock
);
763 return ERR_PTR(-ENOENT
);
765 dentry
->d_flags
|= DCACHE_MOUNTED
;
766 spin_unlock(&dentry
->d_lock
);
767 mp
->m_dentry
= dentry
;
769 list_add(&mp
->m_hash
, chain
);
773 static void put_mountpoint(struct mountpoint
*mp
)
775 if (!--mp
->m_count
) {
776 struct dentry
*dentry
= mp
->m_dentry
;
777 spin_lock(&dentry
->d_lock
);
778 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
779 spin_unlock(&dentry
->d_lock
);
780 list_del(&mp
->m_hash
);
785 static inline int check_mnt(struct mount
*mnt
)
787 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
791 * vfsmount lock must be held for write
793 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
797 wake_up_interruptible(&ns
->poll
);
802 * vfsmount lock must be held for write
804 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
806 if (ns
&& ns
->event
!= event
) {
808 wake_up_interruptible(&ns
->poll
);
813 * vfsmount lock must be held for write
815 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
817 old_path
->dentry
= mnt
->mnt_mountpoint
;
818 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
819 mnt
->mnt_parent
= mnt
;
820 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
821 list_del_init(&mnt
->mnt_child
);
822 list_del_init(&mnt
->mnt_hash
);
823 put_mountpoint(mnt
->mnt_mp
);
828 * vfsmount lock must be held for write
830 void mnt_set_mountpoint(struct mount
*mnt
,
831 struct mountpoint
*mp
,
832 struct mount
*child_mnt
)
835 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
836 child_mnt
->mnt_mountpoint
= dget(mp
->m_dentry
);
837 child_mnt
->mnt_parent
= mnt
;
838 child_mnt
->mnt_mp
= mp
;
842 * vfsmount lock must be held for write
844 static void attach_mnt(struct mount
*mnt
,
845 struct mount
*parent
,
846 struct mountpoint
*mp
)
848 mnt_set_mountpoint(parent
, mp
, mnt
);
849 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
850 hash(&parent
->mnt
, mp
->m_dentry
));
851 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
855 * vfsmount lock must be held for write
857 static void commit_tree(struct mount
*mnt
)
859 struct mount
*parent
= mnt
->mnt_parent
;
862 struct mnt_namespace
*n
= parent
->mnt_ns
;
864 BUG_ON(parent
== mnt
);
866 list_add_tail(&head
, &mnt
->mnt_list
);
867 list_for_each_entry(m
, &head
, mnt_list
)
870 list_splice(&head
, n
->list
.prev
);
872 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
873 hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
874 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
875 touch_mnt_namespace(n
);
878 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
880 struct list_head
*next
= p
->mnt_mounts
.next
;
881 if (next
== &p
->mnt_mounts
) {
885 next
= p
->mnt_child
.next
;
886 if (next
!= &p
->mnt_parent
->mnt_mounts
)
891 return list_entry(next
, struct mount
, mnt_child
);
894 static struct mount
*skip_mnt_tree(struct mount
*p
)
896 struct list_head
*prev
= p
->mnt_mounts
.prev
;
897 while (prev
!= &p
->mnt_mounts
) {
898 p
= list_entry(prev
, struct mount
, mnt_child
);
899 prev
= p
->mnt_mounts
.prev
;
905 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
911 return ERR_PTR(-ENODEV
);
913 mnt
= alloc_vfsmnt(name
);
915 return ERR_PTR(-ENOMEM
);
917 if (flags
& MS_KERNMOUNT
)
918 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
920 root
= mount_fs(type
, flags
, name
, data
);
923 return ERR_CAST(root
);
926 mnt
->mnt
.mnt_root
= root
;
927 mnt
->mnt
.mnt_sb
= root
->d_sb
;
928 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
929 mnt
->mnt_parent
= mnt
;
930 br_write_lock(&vfsmount_lock
);
931 list_add_tail(&mnt
->mnt_instance
, &root
->d_sb
->s_mounts
);
932 br_write_unlock(&vfsmount_lock
);
935 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
937 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
940 struct super_block
*sb
= old
->mnt
.mnt_sb
;
944 mnt
= alloc_vfsmnt(old
->mnt_devname
);
946 return ERR_PTR(-ENOMEM
);
948 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
949 mnt
->mnt_group_id
= 0; /* not a peer of original */
951 mnt
->mnt_group_id
= old
->mnt_group_id
;
953 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
954 err
= mnt_alloc_group_id(mnt
);
959 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~MNT_WRITE_HOLD
;
960 /* Don't allow unprivileged users to change mount flags */
961 if ((flag
& CL_UNPRIVILEGED
) && (mnt
->mnt
.mnt_flags
& MNT_READONLY
))
962 mnt
->mnt
.mnt_flags
|= MNT_LOCK_READONLY
;
964 atomic_inc(&sb
->s_active
);
965 mnt
->mnt
.mnt_sb
= sb
;
966 mnt
->mnt
.mnt_root
= dget(root
);
967 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
968 mnt
->mnt_parent
= mnt
;
969 br_write_lock(&vfsmount_lock
);
970 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
971 br_write_unlock(&vfsmount_lock
);
973 if ((flag
& CL_SLAVE
) ||
974 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
975 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
976 mnt
->mnt_master
= old
;
977 CLEAR_MNT_SHARED(mnt
);
978 } else if (!(flag
& CL_PRIVATE
)) {
979 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
980 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
981 if (IS_MNT_SLAVE(old
))
982 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
983 mnt
->mnt_master
= old
->mnt_master
;
985 if (flag
& CL_MAKE_SHARED
)
988 /* stick the duplicate mount on the same expiry list
989 * as the original if that was on one */
990 if (flag
& CL_EXPIRE
) {
991 if (!list_empty(&old
->mnt_expire
))
992 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
1002 static inline void mntfree(struct mount
*mnt
)
1004 struct vfsmount
*m
= &mnt
->mnt
;
1005 struct super_block
*sb
= m
->mnt_sb
;
1008 * This probably indicates that somebody messed
1009 * up a mnt_want/drop_write() pair. If this
1010 * happens, the filesystem was probably unable
1011 * to make r/w->r/o transitions.
1014 * The locking used to deal with mnt_count decrement provides barriers,
1015 * so mnt_get_writers() below is safe.
1017 WARN_ON(mnt_get_writers(mnt
));
1018 fsnotify_vfsmount_delete(m
);
1021 deactivate_super(sb
);
1024 static void mntput_no_expire(struct mount
*mnt
)
1028 br_read_lock(&vfsmount_lock
);
1029 if (likely(mnt
->mnt_ns
)) {
1030 /* shouldn't be the last one */
1031 mnt_add_count(mnt
, -1);
1032 br_read_unlock(&vfsmount_lock
);
1035 br_read_unlock(&vfsmount_lock
);
1037 br_write_lock(&vfsmount_lock
);
1038 mnt_add_count(mnt
, -1);
1039 if (mnt_get_count(mnt
)) {
1040 br_write_unlock(&vfsmount_lock
);
1044 mnt_add_count(mnt
, -1);
1045 if (likely(mnt_get_count(mnt
)))
1047 br_write_lock(&vfsmount_lock
);
1049 if (unlikely(mnt
->mnt_pinned
)) {
1050 mnt_add_count(mnt
, mnt
->mnt_pinned
+ 1);
1051 mnt
->mnt_pinned
= 0;
1052 br_write_unlock(&vfsmount_lock
);
1053 acct_auto_close_mnt(&mnt
->mnt
);
1057 list_del(&mnt
->mnt_instance
);
1058 br_write_unlock(&vfsmount_lock
);
1062 void mntput(struct vfsmount
*mnt
)
1065 struct mount
*m
= real_mount(mnt
);
1066 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1067 if (unlikely(m
->mnt_expiry_mark
))
1068 m
->mnt_expiry_mark
= 0;
1069 mntput_no_expire(m
);
1072 EXPORT_SYMBOL(mntput
);
1074 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1077 mnt_add_count(real_mount(mnt
), 1);
1080 EXPORT_SYMBOL(mntget
);
1082 void mnt_pin(struct vfsmount
*mnt
)
1084 br_write_lock(&vfsmount_lock
);
1085 real_mount(mnt
)->mnt_pinned
++;
1086 br_write_unlock(&vfsmount_lock
);
1088 EXPORT_SYMBOL(mnt_pin
);
1090 void mnt_unpin(struct vfsmount
*m
)
1092 struct mount
*mnt
= real_mount(m
);
1093 br_write_lock(&vfsmount_lock
);
1094 if (mnt
->mnt_pinned
) {
1095 mnt_add_count(mnt
, 1);
1098 br_write_unlock(&vfsmount_lock
);
1100 EXPORT_SYMBOL(mnt_unpin
);
1102 static inline void mangle(struct seq_file
*m
, const char *s
)
1104 seq_escape(m
, s
, " \t\n\\");
1108 * Simple .show_options callback for filesystems which don't want to
1109 * implement more complex mount option showing.
1111 * See also save_mount_options().
1113 int generic_show_options(struct seq_file
*m
, struct dentry
*root
)
1115 const char *options
;
1118 options
= rcu_dereference(root
->d_sb
->s_options
);
1120 if (options
!= NULL
&& options
[0]) {
1128 EXPORT_SYMBOL(generic_show_options
);
1131 * If filesystem uses generic_show_options(), this function should be
1132 * called from the fill_super() callback.
1134 * The .remount_fs callback usually needs to be handled in a special
1135 * way, to make sure, that previous options are not overwritten if the
1138 * Also note, that if the filesystem's .remount_fs function doesn't
1139 * reset all options to their default value, but changes only newly
1140 * given options, then the displayed options will not reflect reality
1143 void save_mount_options(struct super_block
*sb
, char *options
)
1145 BUG_ON(sb
->s_options
);
1146 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
1148 EXPORT_SYMBOL(save_mount_options
);
1150 void replace_mount_options(struct super_block
*sb
, char *options
)
1152 char *old
= sb
->s_options
;
1153 rcu_assign_pointer(sb
->s_options
, options
);
1159 EXPORT_SYMBOL(replace_mount_options
);
1161 #ifdef CONFIG_PROC_FS
1162 /* iterator; we want it to have access to namespace_sem, thus here... */
1163 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1165 struct proc_mounts
*p
= proc_mounts(m
);
1167 down_read(&namespace_sem
);
1168 return seq_list_start(&p
->ns
->list
, *pos
);
1171 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1173 struct proc_mounts
*p
= proc_mounts(m
);
1175 return seq_list_next(v
, &p
->ns
->list
, pos
);
1178 static void m_stop(struct seq_file
*m
, void *v
)
1180 up_read(&namespace_sem
);
1183 static int m_show(struct seq_file
*m
, void *v
)
1185 struct proc_mounts
*p
= proc_mounts(m
);
1186 struct mount
*r
= list_entry(v
, struct mount
, mnt_list
);
1187 return p
->show(m
, &r
->mnt
);
1190 const struct seq_operations mounts_op
= {
1196 #endif /* CONFIG_PROC_FS */
1199 * may_umount_tree - check if a mount tree is busy
1200 * @mnt: root of mount tree
1202 * This is called to check if a tree of mounts has any
1203 * open files, pwds, chroots or sub mounts that are
1206 int may_umount_tree(struct vfsmount
*m
)
1208 struct mount
*mnt
= real_mount(m
);
1209 int actual_refs
= 0;
1210 int minimum_refs
= 0;
1214 /* write lock needed for mnt_get_count */
1215 br_write_lock(&vfsmount_lock
);
1216 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1217 actual_refs
+= mnt_get_count(p
);
1220 br_write_unlock(&vfsmount_lock
);
1222 if (actual_refs
> minimum_refs
)
1228 EXPORT_SYMBOL(may_umount_tree
);
1231 * may_umount - check if a mount point is busy
1232 * @mnt: root of mount
1234 * This is called to check if a mount point has any
1235 * open files, pwds, chroots or sub mounts. If the
1236 * mount has sub mounts this will return busy
1237 * regardless of whether the sub mounts are busy.
1239 * Doesn't take quota and stuff into account. IOW, in some cases it will
1240 * give false negatives. The main reason why it's here is that we need
1241 * a non-destructive way to look for easily umountable filesystems.
1243 int may_umount(struct vfsmount
*mnt
)
1246 down_read(&namespace_sem
);
1247 br_write_lock(&vfsmount_lock
);
1248 if (propagate_mount_busy(real_mount(mnt
), 2))
1250 br_write_unlock(&vfsmount_lock
);
1251 up_read(&namespace_sem
);
1255 EXPORT_SYMBOL(may_umount
);
1257 static LIST_HEAD(unmounted
); /* protected by namespace_sem */
1259 static void namespace_unlock(void)
1264 if (likely(list_empty(&unmounted
))) {
1265 up_write(&namespace_sem
);
1269 list_splice_init(&unmounted
, &head
);
1270 up_write(&namespace_sem
);
1272 while (!list_empty(&head
)) {
1273 mnt
= list_first_entry(&head
, struct mount
, mnt_hash
);
1274 list_del_init(&mnt
->mnt_hash
);
1275 if (mnt_has_parent(mnt
)) {
1276 struct dentry
*dentry
;
1279 br_write_lock(&vfsmount_lock
);
1280 dentry
= mnt
->mnt_mountpoint
;
1281 m
= mnt
->mnt_parent
;
1282 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1283 mnt
->mnt_parent
= mnt
;
1285 br_write_unlock(&vfsmount_lock
);
1293 static inline void namespace_lock(void)
1295 down_write(&namespace_sem
);
1299 * vfsmount lock must be held for write
1300 * namespace_sem must be held for write
1302 void umount_tree(struct mount
*mnt
, int propagate
)
1304 LIST_HEAD(tmp_list
);
1307 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
1308 list_move(&p
->mnt_hash
, &tmp_list
);
1311 propagate_umount(&tmp_list
);
1313 list_for_each_entry(p
, &tmp_list
, mnt_hash
) {
1314 list_del_init(&p
->mnt_expire
);
1315 list_del_init(&p
->mnt_list
);
1316 __touch_mnt_namespace(p
->mnt_ns
);
1318 list_del_init(&p
->mnt_child
);
1319 if (mnt_has_parent(p
)) {
1320 p
->mnt_parent
->mnt_ghosts
++;
1321 put_mountpoint(p
->mnt_mp
);
1324 change_mnt_propagation(p
, MS_PRIVATE
);
1326 list_splice(&tmp_list
, &unmounted
);
1329 static void shrink_submounts(struct mount
*mnt
);
1331 static int do_umount(struct mount
*mnt
, int flags
)
1333 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1336 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1341 * Allow userspace to request a mountpoint be expired rather than
1342 * unmounting unconditionally. Unmount only happens if:
1343 * (1) the mark is already set (the mark is cleared by mntput())
1344 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1346 if (flags
& MNT_EXPIRE
) {
1347 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1348 flags
& (MNT_FORCE
| MNT_DETACH
))
1352 * probably don't strictly need the lock here if we examined
1353 * all race cases, but it's a slowpath.
1355 br_write_lock(&vfsmount_lock
);
1356 if (mnt_get_count(mnt
) != 2) {
1357 br_write_unlock(&vfsmount_lock
);
1360 br_write_unlock(&vfsmount_lock
);
1362 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1367 * If we may have to abort operations to get out of this
1368 * mount, and they will themselves hold resources we must
1369 * allow the fs to do things. In the Unix tradition of
1370 * 'Gee thats tricky lets do it in userspace' the umount_begin
1371 * might fail to complete on the first run through as other tasks
1372 * must return, and the like. Thats for the mount program to worry
1373 * about for the moment.
1376 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1377 sb
->s_op
->umount_begin(sb
);
1381 * No sense to grab the lock for this test, but test itself looks
1382 * somewhat bogus. Suggestions for better replacement?
1383 * Ho-hum... In principle, we might treat that as umount + switch
1384 * to rootfs. GC would eventually take care of the old vfsmount.
1385 * Actually it makes sense, especially if rootfs would contain a
1386 * /reboot - static binary that would close all descriptors and
1387 * call reboot(9). Then init(8) could umount root and exec /reboot.
1389 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1391 * Special case for "unmounting" root ...
1392 * we just try to remount it readonly.
1394 down_write(&sb
->s_umount
);
1395 if (!(sb
->s_flags
& MS_RDONLY
))
1396 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1397 up_write(&sb
->s_umount
);
1402 br_write_lock(&vfsmount_lock
);
1405 if (!(flags
& MNT_DETACH
))
1406 shrink_submounts(mnt
);
1409 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
1410 if (!list_empty(&mnt
->mnt_list
))
1411 umount_tree(mnt
, 1);
1414 br_write_unlock(&vfsmount_lock
);
1420 * Is the caller allowed to modify his namespace?
1422 static inline bool may_mount(void)
1424 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1428 * Now umount can handle mount points as well as block devices.
1429 * This is important for filesystems which use unnamed block devices.
1431 * We now support a flag for forced unmount like the other 'big iron'
1432 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1435 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1440 int lookup_flags
= 0;
1444 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1450 if (!(flags
& UMOUNT_NOFOLLOW
))
1451 lookup_flags
|= LOOKUP_FOLLOW
;
1453 retval
= user_path_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1456 mnt
= real_mount(path
.mnt
);
1458 if (path
.dentry
!= path
.mnt
->mnt_root
)
1460 if (!check_mnt(mnt
))
1463 retval
= do_umount(mnt
, flags
);
1466 printk("Ahsin do_umount retval=%d \n",retval
);
1467 //do_umount success: 0, do_umount busy: -16
1468 //if do_umount fail, need to dump the link list here
1471 printk("Ahsin do_umount fail; mnt_get_count=%d mnt->mnt_devname=%s\n",mnt_get_count(mnt
),mnt
->mnt_devname
);
1473 printk("Ahsin do_umount success; mnt_get_count=%d mnt->mnt_devname=%s\n",mnt_get_count(mnt
),mnt
->mnt_devname
);
1475 // print linked list
1476 spin_lock(&mnt_id_lock
);
1477 ref_current
= ref_head
;
1478 while(ref_current
!= NULL
)
1480 total_value
= total_value
+ ref_current
->count
;
1482 if (ref_current
->count
)
1483 printk("Ahsin PID= %d, Name = %s, Count= %d \n", ref_current
->pid
, ref_current
->name
, ref_current
->count
);
1484 ref_current
= ref_current
->next
;
1486 spin_unlock(&mnt_id_lock
);
1487 printk("Ahsin total_value=%d \n",total_value
);
1491 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1493 mntput_no_expire(mnt
);
1498 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1501 * The 2.0 compatible umount. No flags.
1503 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1505 return sys_umount(name
, 0);
1510 static bool mnt_ns_loop(struct path
*path
)
1512 /* Could bind mounting the mount namespace inode cause a
1513 * mount namespace loop?
1515 struct inode
*inode
= path
->dentry
->d_inode
;
1517 struct mnt_namespace
*mnt_ns
;
1519 if (!proc_ns_inode(inode
))
1522 ei
= get_proc_ns(inode
);
1523 if (ei
->ns_ops
!= &mntns_operations
)
1527 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1530 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1533 struct mount
*res
, *p
, *q
, *r
, *parent
;
1535 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
1536 return ERR_PTR(-EINVAL
);
1538 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1542 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1545 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1547 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1550 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1551 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
1552 s
= skip_mnt_tree(s
);
1555 while (p
!= s
->mnt_parent
) {
1561 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1564 br_write_lock(&vfsmount_lock
);
1565 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1566 attach_mnt(q
, parent
, p
->mnt_mp
);
1567 br_write_unlock(&vfsmount_lock
);
1573 br_write_lock(&vfsmount_lock
);
1574 umount_tree(res
, 0);
1575 br_write_unlock(&vfsmount_lock
);
1580 /* Caller should check returned pointer for errors */
1582 struct vfsmount
*collect_mounts(struct path
*path
)
1586 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1587 CL_COPY_ALL
| CL_PRIVATE
);
1590 return ERR_CAST(tree
);
1594 void drop_collected_mounts(struct vfsmount
*mnt
)
1597 br_write_lock(&vfsmount_lock
);
1598 umount_tree(real_mount(mnt
), 0);
1599 br_write_unlock(&vfsmount_lock
);
1603 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1604 struct vfsmount
*root
)
1607 int res
= f(root
, arg
);
1610 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1611 res
= f(&mnt
->mnt
, arg
);
1618 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1622 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1623 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1624 mnt_release_group_id(p
);
1628 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1632 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1633 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1634 int err
= mnt_alloc_group_id(p
);
1636 cleanup_group_ids(mnt
, p
);
1646 * @source_mnt : mount tree to be attached
1647 * @nd : place the mount tree @source_mnt is attached
1648 * @parent_nd : if non-null, detach the source_mnt from its parent and
1649 * store the parent mount and mountpoint dentry.
1650 * (done when source_mnt is moved)
1652 * NOTE: in the table below explains the semantics when a source mount
1653 * of a given type is attached to a destination mount of a given type.
1654 * ---------------------------------------------------------------------------
1655 * | BIND MOUNT OPERATION |
1656 * |**************************************************************************
1657 * | source-->| shared | private | slave | unbindable |
1661 * |**************************************************************************
1662 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1664 * |non-shared| shared (+) | private | slave (*) | invalid |
1665 * ***************************************************************************
1666 * A bind operation clones the source mount and mounts the clone on the
1667 * destination mount.
1669 * (++) the cloned mount is propagated to all the mounts in the propagation
1670 * tree of the destination mount and the cloned mount is added to
1671 * the peer group of the source mount.
1672 * (+) the cloned mount is created under the destination mount and is marked
1673 * as shared. The cloned mount is added to the peer group of the source
1675 * (+++) the mount is propagated to all the mounts in the propagation tree
1676 * of the destination mount and the cloned mount is made slave
1677 * of the same master as that of the source mount. The cloned mount
1678 * is marked as 'shared and slave'.
1679 * (*) the cloned mount is made a slave of the same master as that of the
1682 * ---------------------------------------------------------------------------
1683 * | MOVE MOUNT OPERATION |
1684 * |**************************************************************************
1685 * | source-->| shared | private | slave | unbindable |
1689 * |**************************************************************************
1690 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1692 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1693 * ***************************************************************************
1695 * (+) the mount is moved to the destination. And is then propagated to
1696 * all the mounts in the propagation tree of the destination mount.
1697 * (+*) the mount is moved to the destination.
1698 * (+++) the mount is moved to the destination and is then propagated to
1699 * all the mounts belonging to the destination mount's propagation tree.
1700 * the mount is marked as 'shared and slave'.
1701 * (*) the mount continues to be a slave at the new location.
1703 * if the source mount is a tree, the operations explained above is
1704 * applied to each mount in the tree.
1705 * Must be called without spinlocks held, since this function can sleep
1708 static int attach_recursive_mnt(struct mount
*source_mnt
,
1709 struct mount
*dest_mnt
,
1710 struct mountpoint
*dest_mp
,
1711 struct path
*parent_path
)
1713 LIST_HEAD(tree_list
);
1714 struct mount
*child
, *p
;
1717 if (IS_MNT_SHARED(dest_mnt
)) {
1718 err
= invent_group_ids(source_mnt
, true);
1722 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
1724 goto out_cleanup_ids
;
1726 br_write_lock(&vfsmount_lock
);
1728 if (IS_MNT_SHARED(dest_mnt
)) {
1729 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
1733 detach_mnt(source_mnt
, parent_path
);
1734 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
1735 touch_mnt_namespace(source_mnt
->mnt_ns
);
1737 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
1738 commit_tree(source_mnt
);
1741 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
1742 list_del_init(&child
->mnt_hash
);
1745 br_write_unlock(&vfsmount_lock
);
1750 if (IS_MNT_SHARED(dest_mnt
))
1751 cleanup_group_ids(source_mnt
, NULL
);
1756 static struct mountpoint
*lock_mount(struct path
*path
)
1758 struct vfsmount
*mnt
;
1759 struct dentry
*dentry
= path
->dentry
;
1761 mutex_lock(&dentry
->d_inode
->i_mutex
);
1762 if (unlikely(cant_mount(dentry
))) {
1763 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1764 return ERR_PTR(-ENOENT
);
1767 mnt
= lookup_mnt(path
);
1769 struct mountpoint
*mp
= new_mountpoint(dentry
);
1772 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1778 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1781 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
1785 static void unlock_mount(struct mountpoint
*where
)
1787 struct dentry
*dentry
= where
->m_dentry
;
1788 put_mountpoint(where
);
1790 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1793 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
1795 if (mnt
->mnt
.mnt_sb
->s_flags
& MS_NOUSER
)
1798 if (S_ISDIR(mp
->m_dentry
->d_inode
->i_mode
) !=
1799 S_ISDIR(mnt
->mnt
.mnt_root
->d_inode
->i_mode
))
1802 return attach_recursive_mnt(mnt
, p
, mp
, NULL
);
1806 * Sanity check the flags to change_mnt_propagation.
1809 static int flags_to_propagation_type(int flags
)
1811 int type
= flags
& ~(MS_REC
| MS_SILENT
);
1813 /* Fail if any non-propagation flags are set */
1814 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1816 /* Only one propagation flag should be set */
1817 if (!is_power_of_2(type
))
1823 * recursively change the type of the mountpoint.
1825 static int do_change_type(struct path
*path
, int flag
)
1828 struct mount
*mnt
= real_mount(path
->mnt
);
1829 int recurse
= flag
& MS_REC
;
1833 if (path
->dentry
!= path
->mnt
->mnt_root
)
1836 type
= flags_to_propagation_type(flag
);
1841 if (type
== MS_SHARED
) {
1842 err
= invent_group_ids(mnt
, recurse
);
1847 br_write_lock(&vfsmount_lock
);
1848 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
1849 change_mnt_propagation(m
, type
);
1850 br_write_unlock(&vfsmount_lock
);
1858 * do loopback mount.
1860 static int do_loopback(struct path
*path
, const char *old_name
,
1863 struct path old_path
;
1864 struct mount
*mnt
= NULL
, *old
, *parent
;
1865 struct mountpoint
*mp
;
1867 if (!old_name
|| !*old_name
)
1869 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
1874 if (mnt_ns_loop(&old_path
))
1877 mp
= lock_mount(path
);
1882 old
= real_mount(old_path
.mnt
);
1883 parent
= real_mount(path
->mnt
);
1886 if (IS_MNT_UNBINDABLE(old
))
1889 if (!check_mnt(parent
) || !check_mnt(old
))
1893 mnt
= copy_tree(old
, old_path
.dentry
, 0);
1895 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
1902 err
= graft_tree(mnt
, parent
, mp
);
1904 br_write_lock(&vfsmount_lock
);
1905 umount_tree(mnt
, 0);
1906 br_write_unlock(&vfsmount_lock
);
1911 path_put(&old_path
);
1915 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1918 int readonly_request
= 0;
1920 if (ms_flags
& MS_RDONLY
)
1921 readonly_request
= 1;
1922 if (readonly_request
== __mnt_is_readonly(mnt
))
1925 if (mnt
->mnt_flags
& MNT_LOCK_READONLY
)
1928 if (readonly_request
)
1929 error
= mnt_make_readonly(real_mount(mnt
));
1931 __mnt_unmake_readonly(real_mount(mnt
));
1936 * change filesystem flags. dir should be a physical root of filesystem.
1937 * If you've mounted a non-root directory somewhere and want to do remount
1938 * on it - tough luck.
1940 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
1944 struct super_block
*sb
= path
->mnt
->mnt_sb
;
1945 struct mount
*mnt
= real_mount(path
->mnt
);
1947 if (!check_mnt(mnt
))
1950 if (path
->dentry
!= path
->mnt
->mnt_root
)
1953 err
= security_sb_remount(sb
, data
);
1957 down_write(&sb
->s_umount
);
1958 if (flags
& MS_BIND
)
1959 err
= change_mount_flags(path
->mnt
, flags
);
1960 else if (!capable(CAP_SYS_ADMIN
))
1963 err
= do_remount_sb(sb
, flags
, data
, 0);
1965 br_write_lock(&vfsmount_lock
);
1966 mnt_flags
|= mnt
->mnt
.mnt_flags
& MNT_PROPAGATION_MASK
;
1967 mnt
->mnt
.mnt_flags
= mnt_flags
;
1968 br_write_unlock(&vfsmount_lock
);
1970 up_write(&sb
->s_umount
);
1972 br_write_lock(&vfsmount_lock
);
1973 touch_mnt_namespace(mnt
->mnt_ns
);
1974 br_write_unlock(&vfsmount_lock
);
1979 static inline int tree_contains_unbindable(struct mount
*mnt
)
1982 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1983 if (IS_MNT_UNBINDABLE(p
))
1989 static int do_move_mount(struct path
*path
, const char *old_name
)
1991 struct path old_path
, parent_path
;
1994 struct mountpoint
*mp
;
1996 if (!old_name
|| !*old_name
)
1998 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
2002 mp
= lock_mount(path
);
2007 old
= real_mount(old_path
.mnt
);
2008 p
= real_mount(path
->mnt
);
2011 if (!check_mnt(p
) || !check_mnt(old
))
2015 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
2018 if (!mnt_has_parent(old
))
2021 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
2022 S_ISDIR(old_path
.dentry
->d_inode
->i_mode
))
2025 * Don't move a mount residing in a shared parent.
2027 if (IS_MNT_SHARED(old
->mnt_parent
))
2030 * Don't move a mount tree containing unbindable mounts to a destination
2031 * mount which is shared.
2033 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
2036 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
2040 err
= attach_recursive_mnt(old
, real_mount(path
->mnt
), mp
, &parent_path
);
2044 /* if the mount is moved, it should no longer be expire
2046 list_del_init(&old
->mnt_expire
);
2051 path_put(&parent_path
);
2052 path_put(&old_path
);
2056 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
2059 const char *subtype
= strchr(fstype
, '.');
2068 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
2070 if (!mnt
->mnt_sb
->s_subtype
)
2076 return ERR_PTR(err
);
2080 * add a mount into a namespace's mount tree
2082 static int do_add_mount(struct mount
*newmnt
, struct path
*path
, int mnt_flags
)
2084 struct mountpoint
*mp
;
2085 struct mount
*parent
;
2088 mnt_flags
&= ~(MNT_SHARED
| MNT_WRITE_HOLD
| MNT_INTERNAL
);
2090 mp
= lock_mount(path
);
2094 parent
= real_mount(path
->mnt
);
2096 if (unlikely(!check_mnt(parent
))) {
2097 /* that's acceptable only for automounts done in private ns */
2098 if (!(mnt_flags
& MNT_SHRINKABLE
))
2100 /* ... and for those we'd better have mountpoint still alive */
2101 if (!parent
->mnt_ns
)
2105 /* Refuse the same filesystem on the same mount point */
2107 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
2108 path
->mnt
->mnt_root
== path
->dentry
)
2112 if (S_ISLNK(newmnt
->mnt
.mnt_root
->d_inode
->i_mode
))
2115 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2116 err
= graft_tree(newmnt
, parent
, mp
);
2124 * create a new mount for userspace and request it to be added into the
2127 static int do_new_mount(struct path
*path
, const char *fstype
, int flags
,
2128 int mnt_flags
, const char *name
, void *data
)
2130 struct file_system_type
*type
;
2131 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2132 struct vfsmount
*mnt
;
2138 type
= get_fs_type(fstype
);
2142 if (user_ns
!= &init_user_ns
) {
2143 if (!(type
->fs_flags
& FS_USERNS_MOUNT
)) {
2144 put_filesystem(type
);
2147 /* Only in special cases allow devices from mounts
2148 * created outside the initial user namespace.
2150 if (!(type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
2152 mnt_flags
|= MNT_NODEV
;
2156 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
2157 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
2158 !mnt
->mnt_sb
->s_subtype
)
2159 mnt
= fs_set_subtype(mnt
, fstype
);
2161 put_filesystem(type
);
2163 return PTR_ERR(mnt
);
2165 err
= do_add_mount(real_mount(mnt
), path
, mnt_flags
);
2171 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2173 struct mount
*mnt
= real_mount(m
);
2175 /* The new mount record should have at least 2 refs to prevent it being
2176 * expired before we get a chance to add it
2178 BUG_ON(mnt_get_count(mnt
) < 2);
2180 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2181 m
->mnt_root
== path
->dentry
) {
2186 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2190 /* remove m from any expiration list it may be on */
2191 if (!list_empty(&mnt
->mnt_expire
)) {
2193 br_write_lock(&vfsmount_lock
);
2194 list_del_init(&mnt
->mnt_expire
);
2195 br_write_unlock(&vfsmount_lock
);
2204 * mnt_set_expiry - Put a mount on an expiration list
2205 * @mnt: The mount to list.
2206 * @expiry_list: The list to add the mount to.
2208 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2211 br_write_lock(&vfsmount_lock
);
2213 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2215 br_write_unlock(&vfsmount_lock
);
2218 EXPORT_SYMBOL(mnt_set_expiry
);
2221 * process a list of expirable mountpoints with the intent of discarding any
2222 * mountpoints that aren't in use and haven't been touched since last we came
2225 void mark_mounts_for_expiry(struct list_head
*mounts
)
2227 struct mount
*mnt
, *next
;
2228 LIST_HEAD(graveyard
);
2230 if (list_empty(mounts
))
2234 br_write_lock(&vfsmount_lock
);
2236 /* extract from the expiration list every vfsmount that matches the
2237 * following criteria:
2238 * - only referenced by its parent vfsmount
2239 * - still marked for expiry (marked on the last call here; marks are
2240 * cleared by mntput())
2242 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2243 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2244 propagate_mount_busy(mnt
, 1))
2246 list_move(&mnt
->mnt_expire
, &graveyard
);
2248 while (!list_empty(&graveyard
)) {
2249 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
2250 touch_mnt_namespace(mnt
->mnt_ns
);
2251 umount_tree(mnt
, 1);
2253 br_write_unlock(&vfsmount_lock
);
2257 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2260 * Ripoff of 'select_parent()'
2262 * search the list of submounts for a given mountpoint, and move any
2263 * shrinkable submounts to the 'graveyard' list.
2265 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2267 struct mount
*this_parent
= parent
;
2268 struct list_head
*next
;
2272 next
= this_parent
->mnt_mounts
.next
;
2274 while (next
!= &this_parent
->mnt_mounts
) {
2275 struct list_head
*tmp
= next
;
2276 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2279 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2282 * Descend a level if the d_mounts list is non-empty.
2284 if (!list_empty(&mnt
->mnt_mounts
)) {
2289 if (!propagate_mount_busy(mnt
, 1)) {
2290 list_move_tail(&mnt
->mnt_expire
, graveyard
);
2295 * All done at this level ... ascend and resume the search
2297 if (this_parent
!= parent
) {
2298 next
= this_parent
->mnt_child
.next
;
2299 this_parent
= this_parent
->mnt_parent
;
2306 * process a list of expirable mountpoints with the intent of discarding any
2307 * submounts of a specific parent mountpoint
2309 * vfsmount_lock must be held for write
2311 static void shrink_submounts(struct mount
*mnt
)
2313 LIST_HEAD(graveyard
);
2316 /* extract submounts of 'mountpoint' from the expiration list */
2317 while (select_submounts(mnt
, &graveyard
)) {
2318 while (!list_empty(&graveyard
)) {
2319 m
= list_first_entry(&graveyard
, struct mount
,
2321 touch_mnt_namespace(m
->mnt_ns
);
2328 * Some copy_from_user() implementations do not return the exact number of
2329 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2330 * Note that this function differs from copy_from_user() in that it will oops
2331 * on bad values of `to', rather than returning a short copy.
2333 static long exact_copy_from_user(void *to
, const void __user
* from
,
2337 const char __user
*f
= from
;
2340 if (!access_ok(VERIFY_READ
, from
, n
))
2344 if (__get_user(c
, f
)) {
2355 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2365 if (!(page
= __get_free_page(GFP_KERNEL
)))
2368 /* We only care that *some* data at the address the user
2369 * gave us is valid. Just in case, we'll zero
2370 * the remainder of the page.
2372 /* copy_from_user cannot cross TASK_SIZE ! */
2373 size
= TASK_SIZE
- (unsigned long)data
;
2374 if (size
> PAGE_SIZE
)
2377 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2383 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2388 int copy_mount_string(const void __user
*data
, char **where
)
2397 tmp
= strndup_user(data
, PAGE_SIZE
);
2399 return PTR_ERR(tmp
);
2406 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2407 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2409 * data is a (void *) that can point to any structure up to
2410 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2411 * information (or be NULL).
2413 * Pre-0.97 versions of mount() didn't have a flags word.
2414 * When the flags word was introduced its top half was required
2415 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2416 * Therefore, if this magic number is present, it carries no information
2417 * and must be discarded.
2419 long do_mount(const char *dev_name
, const char *dir_name
,
2420 const char *type_page
, unsigned long flags
, void *data_page
)
2427 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2428 flags
&= ~MS_MGC_MSK
;
2430 /* Basic sanity checks */
2432 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
2436 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2438 /* ... and get the mountpoint */
2439 retval
= kern_path(dir_name
, LOOKUP_FOLLOW
, &path
);
2443 retval
= security_sb_mount(dev_name
, &path
,
2444 type_page
, flags
, data_page
);
2445 if (!retval
&& !may_mount())
2450 /* Default to relatime unless overriden */
2451 if (!(flags
& MS_NOATIME
))
2452 mnt_flags
|= MNT_RELATIME
;
2454 /* Separate the per-mountpoint flags */
2455 if (flags
& MS_NOSUID
)
2456 mnt_flags
|= MNT_NOSUID
;
2457 if (flags
& MS_NODEV
)
2458 mnt_flags
|= MNT_NODEV
;
2459 if (flags
& MS_NOEXEC
)
2460 mnt_flags
|= MNT_NOEXEC
;
2461 if (flags
& MS_NOATIME
)
2462 mnt_flags
|= MNT_NOATIME
;
2463 if (flags
& MS_NODIRATIME
)
2464 mnt_flags
|= MNT_NODIRATIME
;
2465 if (flags
& MS_STRICTATIME
)
2466 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2467 if (flags
& MS_RDONLY
)
2468 mnt_flags
|= MNT_READONLY
;
2470 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2471 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2474 if (flags
& MS_REMOUNT
)
2475 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2477 else if (flags
& MS_BIND
)
2478 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2479 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2480 retval
= do_change_type(&path
, flags
);
2481 else if (flags
& MS_MOVE
)
2482 retval
= do_move_mount(&path
, dev_name
);
2484 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2485 dev_name
, data_page
);
2491 static void free_mnt_ns(struct mnt_namespace
*ns
)
2493 proc_free_inum(ns
->proc_inum
);
2494 put_user_ns(ns
->user_ns
);
2499 * Assign a sequence number so we can detect when we attempt to bind
2500 * mount a reference to an older mount namespace into the current
2501 * mount namespace, preventing reference counting loops. A 64bit
2502 * number incrementing at 10Ghz will take 12,427 years to wrap which
2503 * is effectively never, so we can ignore the possibility.
2505 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
2507 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
)
2509 struct mnt_namespace
*new_ns
;
2512 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2514 return ERR_PTR(-ENOMEM
);
2515 ret
= proc_alloc_inum(&new_ns
->proc_inum
);
2518 return ERR_PTR(ret
);
2520 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
2521 atomic_set(&new_ns
->count
, 1);
2522 new_ns
->root
= NULL
;
2523 INIT_LIST_HEAD(&new_ns
->list
);
2524 init_waitqueue_head(&new_ns
->poll
);
2526 new_ns
->user_ns
= get_user_ns(user_ns
);
2531 * Allocate a new namespace structure and populate it with contents
2532 * copied from the namespace of the passed in task structure.
2534 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
2535 struct user_namespace
*user_ns
, struct fs_struct
*fs
)
2537 struct mnt_namespace
*new_ns
;
2538 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2539 struct mount
*p
, *q
;
2540 struct mount
*old
= mnt_ns
->root
;
2544 new_ns
= alloc_mnt_ns(user_ns
);
2549 /* First pass: copy the tree topology */
2550 copy_flags
= CL_COPY_ALL
| CL_EXPIRE
;
2551 if (user_ns
!= mnt_ns
->user_ns
)
2552 copy_flags
|= CL_SHARED_TO_SLAVE
| CL_UNPRIVILEGED
;
2553 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
2556 free_mnt_ns(new_ns
);
2557 return ERR_CAST(new);
2560 br_write_lock(&vfsmount_lock
);
2561 list_add_tail(&new_ns
->list
, &new->mnt_list
);
2562 br_write_unlock(&vfsmount_lock
);
2565 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2566 * as belonging to new namespace. We have already acquired a private
2567 * fs_struct, so tsk->fs->lock is not needed.
2574 if (&p
->mnt
== fs
->root
.mnt
) {
2575 fs
->root
.mnt
= mntget(&q
->mnt
);
2578 if (&p
->mnt
== fs
->pwd
.mnt
) {
2579 fs
->pwd
.mnt
= mntget(&q
->mnt
);
2583 p
= next_mnt(p
, old
);
2584 q
= next_mnt(q
, new);
2596 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2597 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
2599 struct mnt_namespace
*new_ns
;
2604 if (!(flags
& CLONE_NEWNS
))
2607 new_ns
= dup_mnt_ns(ns
, user_ns
, new_fs
);
2614 * create_mnt_ns - creates a private namespace and adds a root filesystem
2615 * @mnt: pointer to the new root filesystem mountpoint
2617 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*m
)
2619 struct mnt_namespace
*new_ns
= alloc_mnt_ns(&init_user_ns
);
2620 if (!IS_ERR(new_ns
)) {
2621 struct mount
*mnt
= real_mount(m
);
2622 mnt
->mnt_ns
= new_ns
;
2624 list_add(&mnt
->mnt_list
, &new_ns
->list
);
2631 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2633 struct mnt_namespace
*ns
;
2634 struct super_block
*s
;
2638 ns
= create_mnt_ns(mnt
);
2640 return ERR_CAST(ns
);
2642 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2643 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2648 return ERR_PTR(err
);
2650 /* trade a vfsmount reference for active sb one */
2651 s
= path
.mnt
->mnt_sb
;
2652 atomic_inc(&s
->s_active
);
2654 /* lock the sucker */
2655 down_write(&s
->s_umount
);
2656 /* ... and return the root of (sub)tree on it */
2659 EXPORT_SYMBOL(mount_subtree
);
2661 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2662 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2666 struct filename
*kernel_dir
;
2668 unsigned long data_page
;
2670 ret
= copy_mount_string(type
, &kernel_type
);
2674 kernel_dir
= getname(dir_name
);
2675 if (IS_ERR(kernel_dir
)) {
2676 ret
= PTR_ERR(kernel_dir
);
2680 ret
= copy_mount_string(dev_name
, &kernel_dev
);
2684 ret
= copy_mount_options(data
, &data_page
);
2688 ret
= do_mount(kernel_dev
, kernel_dir
->name
, kernel_type
, flags
,
2689 (void *) data_page
);
2691 free_page(data_page
);
2695 putname(kernel_dir
);
2703 * Return true if path is reachable from root
2705 * namespace_sem or vfsmount_lock is held
2707 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
2708 const struct path
*root
)
2710 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
2711 dentry
= mnt
->mnt_mountpoint
;
2712 mnt
= mnt
->mnt_parent
;
2714 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
2717 int path_is_under(struct path
*path1
, struct path
*path2
)
2720 br_read_lock(&vfsmount_lock
);
2721 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
2722 br_read_unlock(&vfsmount_lock
);
2725 EXPORT_SYMBOL(path_is_under
);
2728 * pivot_root Semantics:
2729 * Moves the root file system of the current process to the directory put_old,
2730 * makes new_root as the new root file system of the current process, and sets
2731 * root/cwd of all processes which had them on the current root to new_root.
2734 * The new_root and put_old must be directories, and must not be on the
2735 * same file system as the current process root. The put_old must be
2736 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2737 * pointed to by put_old must yield the same directory as new_root. No other
2738 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2740 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2741 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2742 * in this situation.
2745 * - we don't move root/cwd if they are not at the root (reason: if something
2746 * cared enough to change them, it's probably wrong to force them elsewhere)
2747 * - it's okay to pick a root that isn't the root of a file system, e.g.
2748 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2749 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2752 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
2753 const char __user
*, put_old
)
2755 struct path
new, old
, parent_path
, root_parent
, root
;
2756 struct mount
*new_mnt
, *root_mnt
, *old_mnt
;
2757 struct mountpoint
*old_mp
, *root_mp
;
2763 error
= user_path_dir(new_root
, &new);
2767 error
= user_path_dir(put_old
, &old
);
2771 error
= security_sb_pivotroot(&old
, &new);
2775 get_fs_root(current
->fs
, &root
);
2776 old_mp
= lock_mount(&old
);
2777 error
= PTR_ERR(old_mp
);
2782 new_mnt
= real_mount(new.mnt
);
2783 root_mnt
= real_mount(root
.mnt
);
2784 old_mnt
= real_mount(old
.mnt
);
2785 if (IS_MNT_SHARED(old_mnt
) ||
2786 IS_MNT_SHARED(new_mnt
->mnt_parent
) ||
2787 IS_MNT_SHARED(root_mnt
->mnt_parent
))
2789 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
2792 if (d_unlinked(new.dentry
))
2795 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
2796 goto out4
; /* loop, on the same file system */
2798 if (root
.mnt
->mnt_root
!= root
.dentry
)
2799 goto out4
; /* not a mountpoint */
2800 if (!mnt_has_parent(root_mnt
))
2801 goto out4
; /* not attached */
2802 root_mp
= root_mnt
->mnt_mp
;
2803 if (new.mnt
->mnt_root
!= new.dentry
)
2804 goto out4
; /* not a mountpoint */
2805 if (!mnt_has_parent(new_mnt
))
2806 goto out4
; /* not attached */
2807 /* make sure we can reach put_old from new_root */
2808 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
2810 root_mp
->m_count
++; /* pin it so it won't go away */
2811 br_write_lock(&vfsmount_lock
);
2812 detach_mnt(new_mnt
, &parent_path
);
2813 detach_mnt(root_mnt
, &root_parent
);
2814 /* mount old root on put_old */
2815 attach_mnt(root_mnt
, old_mnt
, old_mp
);
2816 /* mount new_root on / */
2817 attach_mnt(new_mnt
, real_mount(root_parent
.mnt
), root_mp
);
2818 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2819 br_write_unlock(&vfsmount_lock
);
2820 chroot_fs_refs(&root
, &new);
2821 put_mountpoint(root_mp
);
2824 unlock_mount(old_mp
);
2826 path_put(&root_parent
);
2827 path_put(&parent_path
);
2839 static void __init
init_mount_tree(void)
2841 struct vfsmount
*mnt
;
2842 struct mnt_namespace
*ns
;
2844 struct file_system_type
*type
;
2846 type
= get_fs_type("rootfs");
2848 panic("Can't find rootfs type");
2849 mnt
= vfs_kern_mount(type
, 0, "rootfs", NULL
);
2850 put_filesystem(type
);
2852 panic("Can't create rootfs");
2854 ns
= create_mnt_ns(mnt
);
2856 panic("Can't allocate initial namespace");
2858 init_task
.nsproxy
->mnt_ns
= ns
;
2862 root
.dentry
= mnt
->mnt_root
;
2864 set_fs_pwd(current
->fs
, &root
);
2865 set_fs_root(current
->fs
, &root
);
2868 void __init
mnt_init(void)
2873 init_rwsem(&namespace_sem
);
2875 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
2876 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2878 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2879 mountpoint_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2881 if (!mount_hashtable
|| !mountpoint_hashtable
)
2882 panic("Failed to allocate mount hash table\n");
2884 printk(KERN_INFO
"Mount-cache hash table entries: %lu\n", HASH_SIZE
);
2886 for (u
= 0; u
< HASH_SIZE
; u
++)
2887 INIT_LIST_HEAD(&mount_hashtable
[u
]);
2888 for (u
= 0; u
< HASH_SIZE
; u
++)
2889 INIT_LIST_HEAD(&mountpoint_hashtable
[u
]);
2891 br_lock_init(&vfsmount_lock
);
2895 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2897 fs_kobj
= kobject_create_and_add("fs", NULL
);
2899 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
2904 void put_mnt_ns(struct mnt_namespace
*ns
)
2906 if (!atomic_dec_and_test(&ns
->count
))
2909 br_write_lock(&vfsmount_lock
);
2910 umount_tree(ns
->root
, 0);
2911 br_write_unlock(&vfsmount_lock
);
2916 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
2918 struct vfsmount
*mnt
;
2919 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
2922 * it is a longterm mount, don't release mnt until
2923 * we unmount before file sys is unregistered
2925 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
2929 EXPORT_SYMBOL_GPL(kern_mount_data
);
2931 void kern_unmount(struct vfsmount
*mnt
)
2933 /* release long term mount so mount point can be released */
2934 if (!IS_ERR_OR_NULL(mnt
)) {
2935 br_write_lock(&vfsmount_lock
);
2936 real_mount(mnt
)->mnt_ns
= NULL
;
2937 br_write_unlock(&vfsmount_lock
);
2941 EXPORT_SYMBOL(kern_unmount
);
2943 bool our_mnt(struct vfsmount
*mnt
)
2945 return check_mnt(real_mount(mnt
));
2948 bool current_chrooted(void)
2950 /* Does the current process have a non-standard root */
2951 struct path ns_root
;
2952 struct path fs_root
;
2955 /* Find the namespace root */
2956 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
2957 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
2959 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
2962 get_fs_root(current
->fs
, &fs_root
);
2964 chrooted
= !path_equal(&fs_root
, &ns_root
);
2972 void update_mnt_policy(struct user_namespace
*userns
)
2974 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
2977 down_read(&namespace_sem
);
2978 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
2979 switch (mnt
->mnt
.mnt_sb
->s_magic
) {
2981 userns
->may_mount_sysfs
= true;
2983 case PROC_SUPER_MAGIC
:
2984 userns
->may_mount_proc
= true;
2987 if (userns
->may_mount_sysfs
&& userns
->may_mount_proc
)
2990 up_read(&namespace_sem
);
2993 static void *mntns_get(struct task_struct
*task
)
2995 struct mnt_namespace
*ns
= NULL
;
2996 struct nsproxy
*nsproxy
;
2999 nsproxy
= task_nsproxy(task
);
3001 ns
= nsproxy
->mnt_ns
;
3009 static void mntns_put(void *ns
)
3014 static int mntns_install(struct nsproxy
*nsproxy
, void *ns
)
3016 struct fs_struct
*fs
= current
->fs
;
3017 struct mnt_namespace
*mnt_ns
= ns
;
3020 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
3021 !nsown_capable(CAP_SYS_CHROOT
) ||
3022 !nsown_capable(CAP_SYS_ADMIN
))
3029 put_mnt_ns(nsproxy
->mnt_ns
);
3030 nsproxy
->mnt_ns
= mnt_ns
;
3033 root
.mnt
= &mnt_ns
->root
->mnt
;
3034 root
.dentry
= mnt_ns
->root
->mnt
.mnt_root
;
3036 while(d_mountpoint(root
.dentry
) && follow_down_one(&root
))
3039 /* Update the pwd and root */
3040 set_fs_pwd(fs
, &root
);
3041 set_fs_root(fs
, &root
);
3047 static unsigned int mntns_inum(void *ns
)
3049 struct mnt_namespace
*mnt_ns
= ns
;
3050 return mnt_ns
->proc_inum
;
3053 const struct proc_ns_operations mntns_operations
= {
3055 .type
= CLONE_NEWNS
,
3058 .install
= mntns_install
,