blob: 24960626bb6bfc7b9eff78631876d4eb7ca7a40e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
Al Virod10577a2011-12-07 13:06:11 -050012#include <linux/export.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080013#include <linux/capability.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080014#include <linux/mnt_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/namei.h>
16#include <linux/security.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010017#include <linux/idr.h>
Al Virod10577a2011-12-07 13:06:11 -050018#include <linux/acct.h> /* acct_auto_close_mnt */
19#include <linux/ramfs.h> /* init_rootfs */
20#include <linux/fs_struct.h> /* get_fs_root et.al. */
21#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22#include <linux/uaccess.h>
Ram Pai07b20882005-11-07 17:19:07 -050023#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070024#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
Eric Dumazet13f14b42008-02-06 01:37:57 -080026#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
27#define HASH_SIZE (1UL << HASH_SHIFT)
28
Al Viro5addc5d2005-11-07 17:15:49 -050029static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010030static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010031static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100032static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040033static int mnt_id_start = 0;
34static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Eric Dumazetfa3536c2006-03-26 01:37:24 -080036static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080037static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050038static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080040/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060041struct kobject *fs_kobj;
42EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080043
Nick Piggin99b7db72010-08-18 04:37:39 +100044/*
45 * vfsmount lock may be taken for read to prevent changes to the
46 * vfsmount hash, ie. during mountpoint lookups or walking back
47 * up the tree.
48 *
49 * It should be taken for write in all cases where the vfsmount
50 * tree or hash is modified or when a vfsmount structure is modified.
51 */
52DEFINE_BRLOCK(vfsmount_lock);
53
Linus Torvalds1da177e2005-04-16 15:20:36 -070054static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
55{
Ram Paib58fed82005-11-07 17:16:09 -050056 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
57 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080058 tmp = tmp + (tmp >> HASH_SHIFT);
59 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060}
61
Dave Hansen3d733632008-02-15 14:37:59 -080062#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
63
Nick Piggin99b7db72010-08-18 04:37:39 +100064/*
65 * allocation is serialized by namespace_sem, but we need the spinlock to
66 * serialize with freeing.
67 */
Al Virob105e272011-11-24 20:38:33 -050068static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010069{
70 int res;
71
72retry:
73 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +100074 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -050075 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -040076 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -050077 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +100078 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010079 if (res == -EAGAIN)
80 goto retry;
81
82 return res;
83}
84
Al Virob105e272011-11-24 20:38:33 -050085static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010086{
Al Viro15169fe2011-11-25 00:50:41 -050087 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +100088 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040089 ida_remove(&mnt_id_ida, id);
90 if (mnt_id_start > id)
91 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +100092 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010093}
94
Miklos Szeredi719f5d72008-03-27 13:06:23 +010095/*
96 * Allocate a new peer group ID
97 *
98 * mnt_group_ida is protected by namespace_sem
99 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500100static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100101{
Al Virof21f6222009-06-24 03:12:00 -0400102 int res;
103
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100104 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
105 return -ENOMEM;
106
Al Virof21f6222009-06-24 03:12:00 -0400107 res = ida_get_new_above(&mnt_group_ida,
108 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500109 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400110 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500111 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400112
113 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100114}
115
116/*
117 * Release a peer group ID
118 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500119void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100120{
Al Viro15169fe2011-11-25 00:50:41 -0500121 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400122 ida_remove(&mnt_group_ida, id);
123 if (mnt_group_start > id)
124 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500125 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100126}
127
Nick Pigginb3e19d92011-01-07 17:50:11 +1100128/*
129 * vfsmount lock must be held for read
130 */
Al Viro83adc752011-11-24 22:37:54 -0500131static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100132{
133#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500134 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100135#else
136 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500137 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100138 preempt_enable();
139#endif
140}
141
Nick Pigginb3e19d92011-01-07 17:50:11 +1100142/*
143 * vfsmount lock must be held for write
144 */
Al Viro83adc752011-11-24 22:37:54 -0500145unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100146{
147#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500148 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100149 int cpu;
150
151 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500152 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100153 }
154
155 return count;
156#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500157 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100158#endif
159}
160
Al Virob105e272011-11-24 20:38:33 -0500161static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162{
Al Viroc63181e2011-11-25 02:35:16 -0500163 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
164 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100165 int err;
166
Al Viroc63181e2011-11-25 02:35:16 -0500167 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800168 if (err)
169 goto out_free_cache;
170
171 if (name) {
Al Viroc63181e2011-11-25 02:35:16 -0500172 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
173 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800174 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100175 }
176
Nick Pigginb3e19d92011-01-07 17:50:11 +1100177#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500178 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
179 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100180 goto out_free_devname;
181
Al Viroc63181e2011-11-25 02:35:16 -0500182 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100183#else
Al Viroc63181e2011-11-25 02:35:16 -0500184 mnt->mnt_count = 1;
185 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100186#endif
187
Al Viroc63181e2011-11-25 02:35:16 -0500188 INIT_LIST_HEAD(&mnt->mnt_hash);
189 INIT_LIST_HEAD(&mnt->mnt_child);
190 INIT_LIST_HEAD(&mnt->mnt_mounts);
191 INIT_LIST_HEAD(&mnt->mnt_list);
192 INIT_LIST_HEAD(&mnt->mnt_expire);
193 INIT_LIST_HEAD(&mnt->mnt_share);
194 INIT_LIST_HEAD(&mnt->mnt_slave_list);
195 INIT_LIST_HEAD(&mnt->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500196#ifdef CONFIG_FSNOTIFY
197 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
198#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 }
Al Viroc63181e2011-11-25 02:35:16 -0500200 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800201
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000202#ifdef CONFIG_SMP
203out_free_devname:
Al Viroc63181e2011-11-25 02:35:16 -0500204 kfree(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000205#endif
Li Zefan88b38782008-07-21 18:06:36 +0800206out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500207 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800208out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500209 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800210 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
Dave Hansen83660252008-02-15 14:37:30 -0800213/*
214 * Most r/o checks on a fs are for operations that take
215 * discrete amounts of time, like a write() or unlink().
216 * We must keep track of when those operations start
217 * (for permission checks) and when they end, so that
218 * we can determine when writes are able to occur to
219 * a filesystem.
220 */
Dave Hansen3d733632008-02-15 14:37:59 -0800221/*
222 * __mnt_is_readonly: check whether a mount is read-only
223 * @mnt: the mount to check for its write status
224 *
225 * This shouldn't be used directly ouside of the VFS.
226 * It does not guarantee that the filesystem will stay
227 * r/w, just that it is right *now*. This can not and
228 * should not be used in place of IS_RDONLY(inode).
229 * mnt_want/drop_write() will _keep_ the filesystem
230 * r/w.
231 */
232int __mnt_is_readonly(struct vfsmount *mnt)
233{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800234 if (mnt->mnt_flags & MNT_READONLY)
235 return 1;
236 if (mnt->mnt_sb->s_flags & MS_RDONLY)
237 return 1;
238 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800239}
240EXPORT_SYMBOL_GPL(__mnt_is_readonly);
241
Al Viro83adc752011-11-24 22:37:54 -0500242static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800243{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000244#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500245 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000246#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500247 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000248#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800249}
Dave Hansen3d733632008-02-15 14:37:59 -0800250
Al Viro83adc752011-11-24 22:37:54 -0500251static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800252{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000253#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500254 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000255#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500256 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000257#endif
258}
259
Al Viro83adc752011-11-24 22:37:54 -0500260static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000261{
262#ifdef CONFIG_SMP
263 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800264 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800265
266 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500267 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800268 }
Dave Hansen3d733632008-02-15 14:37:59 -0800269
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000270 return count;
271#else
272 return mnt->mnt_writers;
273#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800274}
275
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100276static int mnt_is_readonly(struct vfsmount *mnt)
277{
278 if (mnt->mnt_sb->s_readonly_remount)
279 return 1;
280 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
281 smp_rmb();
282 return __mnt_is_readonly(mnt);
283}
284
Dave Hansen3d733632008-02-15 14:37:59 -0800285/*
Jan Karaeb04c282012-06-12 16:20:35 +0200286 * Most r/o & frozen checks on a fs are for operations that take discrete
287 * amounts of time, like a write() or unlink(). We must keep track of when
288 * those operations start (for permission checks) and when they end, so that we
289 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800290 */
Dave Hansen83660252008-02-15 14:37:30 -0800291/**
Jan Karaeb04c282012-06-12 16:20:35 +0200292 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500293 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800294 *
Jan Karaeb04c282012-06-12 16:20:35 +0200295 * This tells the low-level filesystem that a write is about to be performed to
296 * it, and makes sure that writes are allowed (mnt it read-write) before
297 * returning success. This operation does not protect against filesystem being
298 * frozen. When the write operation is finished, __mnt_drop_write() must be
299 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800300 */
Jan Karaeb04c282012-06-12 16:20:35 +0200301int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800302{
Al Viro83adc752011-11-24 22:37:54 -0500303 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800304 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800305
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000306 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100307 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000308 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100309 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000310 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
311 * incremented count after it has set MNT_WRITE_HOLD.
312 */
313 smp_mb();
Al Viro83adc752011-11-24 22:37:54 -0500314 while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000315 cpu_relax();
316 /*
317 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
318 * be set to match its requirements. So we must not load that until
319 * MNT_WRITE_HOLD is cleared.
320 */
321 smp_rmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100322 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100323 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800324 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800325 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000326 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200327
328 return ret;
329}
330
331/**
332 * mnt_want_write - get write access to a mount
333 * @m: the mount on which to take a write
334 *
335 * This tells the low-level filesystem that a write is about to be performed to
336 * it, and makes sure that writes are allowed (mount is read-write, filesystem
337 * is not frozen) before returning success. When the write operation is
338 * finished, mnt_drop_write() must be called. This is effectively a refcount.
339 */
340int mnt_want_write(struct vfsmount *m)
341{
342 int ret;
343
344 sb_start_write(m->mnt_sb);
345 ret = __mnt_want_write(m);
346 if (ret)
347 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800348 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800349}
350EXPORT_SYMBOL_GPL(mnt_want_write);
351
352/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000353 * mnt_clone_write - get write access to a mount
354 * @mnt: the mount on which to take a write
355 *
356 * This is effectively like mnt_want_write, except
357 * it must only be used to take an extra write reference
358 * on a mountpoint that we already know has a write reference
359 * on it. This allows some optimisation.
360 *
361 * After finished, mnt_drop_write must be called as usual to
362 * drop the reference.
363 */
364int mnt_clone_write(struct vfsmount *mnt)
365{
366 /* superblock may be r/o */
367 if (__mnt_is_readonly(mnt))
368 return -EROFS;
369 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500370 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000371 preempt_enable();
372 return 0;
373}
374EXPORT_SYMBOL_GPL(mnt_clone_write);
375
376/**
Jan Karaeb04c282012-06-12 16:20:35 +0200377 * __mnt_want_write_file - get write access to a file's mount
378 * @file: the file who's mount on which to take a write
379 *
380 * This is like __mnt_want_write, but it takes a file and can
381 * do some optimisations if the file is open for write already
382 */
383int __mnt_want_write_file(struct file *file)
384{
385 struct inode *inode = file->f_dentry->d_inode;
386
387 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
388 return __mnt_want_write(file->f_path.mnt);
389 else
390 return mnt_clone_write(file->f_path.mnt);
391}
392
393/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000394 * mnt_want_write_file - get write access to a file's mount
395 * @file: the file who's mount on which to take a write
396 *
397 * This is like mnt_want_write, but it takes a file and can
398 * do some optimisations if the file is open for write already
399 */
400int mnt_want_write_file(struct file *file)
401{
Jan Karaeb04c282012-06-12 16:20:35 +0200402 int ret;
403
404 sb_start_write(file->f_path.mnt->mnt_sb);
405 ret = __mnt_want_write_file(file);
406 if (ret)
407 sb_end_write(file->f_path.mnt->mnt_sb);
408 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000409}
410EXPORT_SYMBOL_GPL(mnt_want_write_file);
411
412/**
Jan Karaeb04c282012-06-12 16:20:35 +0200413 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800414 * @mnt: the mount on which to give up write access
415 *
416 * Tells the low-level filesystem that we are done
417 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200418 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800419 */
Jan Karaeb04c282012-06-12 16:20:35 +0200420void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800421{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000422 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500423 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000424 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800425}
Jan Karaeb04c282012-06-12 16:20:35 +0200426
427/**
428 * mnt_drop_write - give up write access to a mount
429 * @mnt: the mount on which to give up write access
430 *
431 * Tells the low-level filesystem that we are done performing writes to it and
432 * also allows filesystem to be frozen again. Must be matched with
433 * mnt_want_write() call above.
434 */
435void mnt_drop_write(struct vfsmount *mnt)
436{
437 __mnt_drop_write(mnt);
438 sb_end_write(mnt->mnt_sb);
439}
Dave Hansen83660252008-02-15 14:37:30 -0800440EXPORT_SYMBOL_GPL(mnt_drop_write);
441
Jan Karaeb04c282012-06-12 16:20:35 +0200442void __mnt_drop_write_file(struct file *file)
443{
444 __mnt_drop_write(file->f_path.mnt);
445}
446
Al Viro2a79f172011-12-09 08:06:57 -0500447void mnt_drop_write_file(struct file *file)
448{
449 mnt_drop_write(file->f_path.mnt);
450}
451EXPORT_SYMBOL(mnt_drop_write_file);
452
Al Viro83adc752011-11-24 22:37:54 -0500453static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800454{
Dave Hansen3d733632008-02-15 14:37:59 -0800455 int ret = 0;
456
Andi Kleen962830d2012-05-08 13:32:02 +0930457 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500458 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000459 /*
460 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
461 * should be visible before we do.
462 */
463 smp_mb();
464
465 /*
466 * With writers on hold, if this value is zero, then there are
467 * definitely no active writers (although held writers may subsequently
468 * increment the count, they'll have to wait, and decrement it after
469 * seeing MNT_READONLY).
470 *
471 * It is OK to have counter incremented on one CPU and decremented on
472 * another: the sum will add up correctly. The danger would be when we
473 * sum up each counter, if we read a counter before it is incremented,
474 * but then read another CPU's count which it has been subsequently
475 * decremented from -- we would see more decrements than we should.
476 * MNT_WRITE_HOLD protects against this scenario, because
477 * mnt_want_write first increments count, then smp_mb, then spins on
478 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
479 * we're counting up here.
480 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100481 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000482 ret = -EBUSY;
483 else
Al Viro83adc752011-11-24 22:37:54 -0500484 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000485 /*
486 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
487 * that become unheld will see MNT_READONLY.
488 */
489 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500490 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Andi Kleen962830d2012-05-08 13:32:02 +0930491 br_write_unlock(&vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800492 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800493}
Dave Hansen83660252008-02-15 14:37:30 -0800494
Al Viro83adc752011-11-24 22:37:54 -0500495static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800496{
Andi Kleen962830d2012-05-08 13:32:02 +0930497 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500498 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Andi Kleen962830d2012-05-08 13:32:02 +0930499 br_write_unlock(&vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800500}
501
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100502int sb_prepare_remount_readonly(struct super_block *sb)
503{
504 struct mount *mnt;
505 int err = 0;
506
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100507 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
508 if (atomic_long_read(&sb->s_remove_count))
509 return -EBUSY;
510
Andi Kleen962830d2012-05-08 13:32:02 +0930511 br_write_lock(&vfsmount_lock);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100512 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
513 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
514 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
515 smp_mb();
516 if (mnt_get_writers(mnt) > 0) {
517 err = -EBUSY;
518 break;
519 }
520 }
521 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100522 if (!err && atomic_long_read(&sb->s_remove_count))
523 err = -EBUSY;
524
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100525 if (!err) {
526 sb->s_readonly_remount = 1;
527 smp_wmb();
528 }
529 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
530 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
531 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
532 }
Andi Kleen962830d2012-05-08 13:32:02 +0930533 br_write_unlock(&vfsmount_lock);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100534
535 return err;
536}
537
Al Virob105e272011-11-24 20:38:33 -0500538static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
Al Viro52ba1622011-11-25 02:25:17 -0500540 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100541 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000542#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500543 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000544#endif
Al Virob105e272011-11-24 20:38:33 -0500545 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546}
547
548/*
Ram Paia05964f2005-11-07 17:20:17 -0500549 * find the first or last mount at @dentry on vfsmount @mnt depending on
550 * @dir. If @dir is set return the first mount else return the last mount.
Nick Piggin99b7db72010-08-18 04:37:39 +1000551 * vfsmount_lock must be held for read or write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 */
Al Viroc7105362011-11-24 18:22:03 -0500553struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
Ram Paia05964f2005-11-07 17:20:17 -0500554 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555{
Ram Paib58fed82005-11-07 17:16:09 -0500556 struct list_head *head = mount_hashtable + hash(mnt, dentry);
557 struct list_head *tmp = head;
Al Viroc7105362011-11-24 18:22:03 -0500558 struct mount *p, *found = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500561 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 p = NULL;
563 if (tmp == head)
564 break;
Al Viro1b8e5562011-11-24 21:01:32 -0500565 p = list_entry(tmp, struct mount, mnt_hash);
Al Viroa73324d2011-11-24 22:25:07 -0500566 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500567 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 break;
569 }
570 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 return found;
572}
573
Ram Paia05964f2005-11-07 17:20:17 -0500574/*
David Howellsf015f1262012-06-25 12:55:28 +0100575 * lookup_mnt - Return the first child mount mounted at path
576 *
577 * "First" means first mounted chronologically. If you create the
578 * following mounts:
579 *
580 * mount /dev/sda1 /mnt
581 * mount /dev/sda2 /mnt
582 * mount /dev/sda3 /mnt
583 *
584 * Then lookup_mnt() on the base /mnt dentry in the root mount will
585 * return successively the root dentry and vfsmount of /dev/sda1, then
586 * /dev/sda2, then /dev/sda3, then NULL.
587 *
588 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500589 */
Al Viro1c755af2009-04-18 14:06:57 -0400590struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500591{
Al Viroc7105362011-11-24 18:22:03 -0500592 struct mount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000593
Andi Kleen962830d2012-05-08 13:32:02 +0930594 br_read_lock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500595 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
596 if (child_mnt) {
597 mnt_add_count(child_mnt, 1);
Andi Kleen962830d2012-05-08 13:32:02 +0930598 br_read_unlock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500599 return &child_mnt->mnt;
600 } else {
Andi Kleen962830d2012-05-08 13:32:02 +0930601 br_read_unlock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500602 return NULL;
603 }
Ram Paia05964f2005-11-07 17:20:17 -0500604}
605
Al Viro143c8c92011-11-25 00:46:35 -0500606static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800608 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
610
Nick Piggin99b7db72010-08-18 04:37:39 +1000611/*
612 * vfsmount lock must be held for write
613 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800614static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500615{
616 if (ns) {
617 ns->event = ++event;
618 wake_up_interruptible(&ns->poll);
619 }
620}
621
Nick Piggin99b7db72010-08-18 04:37:39 +1000622/*
623 * vfsmount lock must be held for write
624 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800625static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500626{
627 if (ns && ns->event != event) {
628 ns->event = event;
629 wake_up_interruptible(&ns->poll);
630 }
631}
632
Nick Piggin99b7db72010-08-18 04:37:39 +1000633/*
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100634 * Clear dentry's mounted state if it has no remaining mounts.
635 * vfsmount_lock must be held for write.
636 */
Al Viroaa0a4cf2011-11-24 19:31:36 -0500637static void dentry_reset_mounted(struct dentry *dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100638{
639 unsigned u;
640
641 for (u = 0; u < HASH_SIZE; u++) {
Al Virod5e50f72011-11-24 20:58:57 -0500642 struct mount *p;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100643
Al Viro1b8e5562011-11-24 21:01:32 -0500644 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
Al Viroa73324d2011-11-24 22:25:07 -0500645 if (p->mnt_mountpoint == dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100646 return;
647 }
648 }
649 spin_lock(&dentry->d_lock);
650 dentry->d_flags &= ~DCACHE_MOUNTED;
651 spin_unlock(&dentry->d_lock);
652}
653
654/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000655 * vfsmount lock must be held for write
656 */
Al Viro419148d2011-11-24 19:41:16 -0500657static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658{
Al Viroa73324d2011-11-24 22:25:07 -0500659 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500660 old_path->mnt = &mnt->mnt_parent->mnt;
661 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500662 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500663 list_del_init(&mnt->mnt_child);
Al Viro1b8e5562011-11-24 21:01:32 -0500664 list_del_init(&mnt->mnt_hash);
Al Viroaa0a4cf2011-11-24 19:31:36 -0500665 dentry_reset_mounted(old_path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666}
667
Nick Piggin99b7db72010-08-18 04:37:39 +1000668/*
669 * vfsmount lock must be held for write
670 */
Al Viro14cf1fa2011-11-25 00:01:17 -0500671void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
Al Viro44d964d62011-11-24 21:28:22 -0500672 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500673{
Al Viro3a2393d2011-11-25 03:19:09 -0500674 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viroa73324d2011-11-24 22:25:07 -0500675 child_mnt->mnt_mountpoint = dget(dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500676 child_mnt->mnt_parent = mnt;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100677 spin_lock(&dentry->d_lock);
678 dentry->d_flags |= DCACHE_MOUNTED;
679 spin_unlock(&dentry->d_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500680}
681
Nick Piggin99b7db72010-08-18 04:37:39 +1000682/*
683 * vfsmount lock must be held for write
684 */
Al Viro419148d2011-11-24 19:41:16 -0500685static void attach_mnt(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
Al Viro14cf1fa2011-11-25 00:01:17 -0500687 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
Al Viro1b8e5562011-11-24 21:01:32 -0500688 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400689 hash(path->mnt, path->dentry));
Al Viro6b41d532011-11-24 23:24:33 -0500690 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500691}
692
693/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000694 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500695 */
Al Viro4b2619a2011-11-24 19:47:15 -0500696static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500697{
Al Viro0714a532011-11-24 22:19:58 -0500698 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500699 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500700 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500701 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500702
Al Viro0714a532011-11-24 22:19:58 -0500703 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500704
Al Viro1a4eeaf2011-11-25 02:19:55 -0500705 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400706 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500707 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500708
Ram Paib90fa9a2005-11-07 17:19:50 -0500709 list_splice(&head, n->list.prev);
710
Al Viro1b8e5562011-11-24 21:01:32 -0500711 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viroa73324d2011-11-24 22:25:07 -0500712 hash(&parent->mnt, mnt->mnt_mountpoint));
Al Viro6b41d532011-11-24 23:24:33 -0500713 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800714 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715}
716
Al Viro909b0a82011-11-25 03:06:56 -0500717static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
Al Viro6b41d532011-11-24 23:24:33 -0500719 struct list_head *next = p->mnt_mounts.next;
720 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500722 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500724 next = p->mnt_child.next;
725 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 break;
Al Viro0714a532011-11-24 22:19:58 -0500727 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
729 }
Al Viro6b41d532011-11-24 23:24:33 -0500730 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731}
732
Al Viro315fc832011-11-24 18:57:30 -0500733static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500734{
Al Viro6b41d532011-11-24 23:24:33 -0500735 struct list_head *prev = p->mnt_mounts.prev;
736 while (prev != &p->mnt_mounts) {
737 p = list_entry(prev, struct mount, mnt_child);
738 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500739 }
740 return p;
741}
742
Al Viro9d412a42011-03-17 22:08:28 -0400743struct vfsmount *
744vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
745{
Al Virob105e272011-11-24 20:38:33 -0500746 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400747 struct dentry *root;
748
749 if (!type)
750 return ERR_PTR(-ENODEV);
751
752 mnt = alloc_vfsmnt(name);
753 if (!mnt)
754 return ERR_PTR(-ENOMEM);
755
756 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500757 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400758
759 root = mount_fs(type, flags, name, data);
760 if (IS_ERR(root)) {
761 free_vfsmnt(mnt);
762 return ERR_CAST(root);
763 }
764
Al Virob105e272011-11-24 20:38:33 -0500765 mnt->mnt.mnt_root = root;
766 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500767 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500768 mnt->mnt_parent = mnt;
Andi Kleen962830d2012-05-08 13:32:02 +0930769 br_write_lock(&vfsmount_lock);
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100770 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Andi Kleen962830d2012-05-08 13:32:02 +0930771 br_write_unlock(&vfsmount_lock);
Al Virob105e272011-11-24 20:38:33 -0500772 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400773}
774EXPORT_SYMBOL_GPL(vfs_kern_mount);
775
Al Viro87129cc2011-11-24 21:24:27 -0500776static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500777 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778{
Al Viro87129cc2011-11-24 21:24:27 -0500779 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100780 struct mount *mnt;
781 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
David Howellsbe34d1a2012-06-25 12:55:18 +0100783 mnt = alloc_vfsmnt(old->mnt_devname);
784 if (!mnt)
785 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100786
David Howellsbe34d1a2012-06-25 12:55:18 +0100787 if (flag & (CL_SLAVE | CL_PRIVATE))
788 mnt->mnt_group_id = 0; /* not a peer of original */
789 else
790 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100791
David Howellsbe34d1a2012-06-25 12:55:18 +0100792 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
793 err = mnt_alloc_group_id(mnt);
794 if (err)
795 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100797
798 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
799 atomic_inc(&sb->s_active);
800 mnt->mnt.mnt_sb = sb;
801 mnt->mnt.mnt_root = dget(root);
802 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
803 mnt->mnt_parent = mnt;
804 br_write_lock(&vfsmount_lock);
805 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
806 br_write_unlock(&vfsmount_lock);
807
808 if (flag & CL_SLAVE) {
809 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
810 mnt->mnt_master = old;
811 CLEAR_MNT_SHARED(mnt);
812 } else if (!(flag & CL_PRIVATE)) {
813 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
814 list_add(&mnt->mnt_share, &old->mnt_share);
815 if (IS_MNT_SLAVE(old))
816 list_add(&mnt->mnt_slave, &old->mnt_slave);
817 mnt->mnt_master = old->mnt_master;
818 }
819 if (flag & CL_MAKE_SHARED)
820 set_mnt_shared(mnt);
821
822 /* stick the duplicate mount on the same expiry list
823 * as the original if that was on one */
824 if (flag & CL_EXPIRE) {
825 if (!list_empty(&old->mnt_expire))
826 list_add(&mnt->mnt_expire, &old->mnt_expire);
827 }
828
Al Virocb338d02011-11-24 20:55:08 -0500829 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100830
831 out_free:
832 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +0100833 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834}
835
Al Viro83adc752011-11-24 22:37:54 -0500836static inline void mntfree(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837{
Al Viro83adc752011-11-24 22:37:54 -0500838 struct vfsmount *m = &mnt->mnt;
839 struct super_block *sb = m->mnt_sb;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100840
Dave Hansen3d733632008-02-15 14:37:59 -0800841 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800842 * This probably indicates that somebody messed
843 * up a mnt_want/drop_write() pair. If this
844 * happens, the filesystem was probably unable
845 * to make r/w->r/o transitions.
846 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000847 /*
Nick Pigginb3e19d92011-01-07 17:50:11 +1100848 * The locking used to deal with mnt_count decrement provides barriers,
849 * so mnt_get_writers() below is safe.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000850 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100851 WARN_ON(mnt_get_writers(mnt));
Al Viro83adc752011-11-24 22:37:54 -0500852 fsnotify_vfsmount_delete(m);
853 dput(m->mnt_root);
854 free_vfsmnt(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 deactivate_super(sb);
856}
857
Al Viro900148d2011-11-25 00:33:11 -0500858static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500859{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100860put_again:
Al Virof03c6592011-01-14 22:30:21 -0500861#ifdef CONFIG_SMP
Andi Kleen962830d2012-05-08 13:32:02 +0930862 br_read_lock(&vfsmount_lock);
Al Virof7a99c52012-06-09 00:59:08 -0400863 if (likely(mnt->mnt_ns)) {
864 /* shouldn't be the last one */
Al Viroaa9c0e02011-11-23 19:04:52 -0500865 mnt_add_count(mnt, -1);
Andi Kleen962830d2012-05-08 13:32:02 +0930866 br_read_unlock(&vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500867 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100868 }
Andi Kleen962830d2012-05-08 13:32:02 +0930869 br_read_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100870
Andi Kleen962830d2012-05-08 13:32:02 +0930871 br_write_lock(&vfsmount_lock);
Al Viroaa9c0e02011-11-23 19:04:52 -0500872 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100873 if (mnt_get_count(mnt)) {
Andi Kleen962830d2012-05-08 13:32:02 +0930874 br_write_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100875 return;
876 }
Nick Pigginb3e19d92011-01-07 17:50:11 +1100877#else
Al Viroaa9c0e02011-11-23 19:04:52 -0500878 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100879 if (likely(mnt_get_count(mnt)))
Nick Piggin99b7db72010-08-18 04:37:39 +1000880 return;
Andi Kleen962830d2012-05-08 13:32:02 +0930881 br_write_lock(&vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500882#endif
Al Viro863d6842011-11-25 00:57:42 -0500883 if (unlikely(mnt->mnt_pinned)) {
884 mnt_add_count(mnt, mnt->mnt_pinned + 1);
885 mnt->mnt_pinned = 0;
Andi Kleen962830d2012-05-08 13:32:02 +0930886 br_write_unlock(&vfsmount_lock);
Al Viro900148d2011-11-25 00:33:11 -0500887 acct_auto_close_mnt(&mnt->mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100888 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500889 }
Andi Kleen962830d2012-05-08 13:32:02 +0930890
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100891 list_del(&mnt->mnt_instance);
Andi Kleen962830d2012-05-08 13:32:02 +0930892 br_write_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100893 mntfree(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500894}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100895
896void mntput(struct vfsmount *mnt)
897{
898 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -0500899 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100900 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -0500901 if (unlikely(m->mnt_expiry_mark))
902 m->mnt_expiry_mark = 0;
903 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100904 }
905}
906EXPORT_SYMBOL(mntput);
907
908struct vfsmount *mntget(struct vfsmount *mnt)
909{
910 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -0500911 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100912 return mnt;
913}
914EXPORT_SYMBOL(mntget);
915
Al Viro7b7b1ac2005-11-07 17:13:39 -0500916void mnt_pin(struct vfsmount *mnt)
917{
Andi Kleen962830d2012-05-08 13:32:02 +0930918 br_write_lock(&vfsmount_lock);
Al Viro863d6842011-11-25 00:57:42 -0500919 real_mount(mnt)->mnt_pinned++;
Andi Kleen962830d2012-05-08 13:32:02 +0930920 br_write_unlock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500921}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500922EXPORT_SYMBOL(mnt_pin);
923
Al Viro863d6842011-11-25 00:57:42 -0500924void mnt_unpin(struct vfsmount *m)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500925{
Al Viro863d6842011-11-25 00:57:42 -0500926 struct mount *mnt = real_mount(m);
Andi Kleen962830d2012-05-08 13:32:02 +0930927 br_write_lock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500928 if (mnt->mnt_pinned) {
Al Viro863d6842011-11-25 00:57:42 -0500929 mnt_add_count(mnt, 1);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500930 mnt->mnt_pinned--;
931 }
Andi Kleen962830d2012-05-08 13:32:02 +0930932 br_write_unlock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500933}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500934EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800936static inline void mangle(struct seq_file *m, const char *s)
937{
938 seq_escape(m, s, " \t\n\\");
939}
940
941/*
942 * Simple .show_options callback for filesystems which don't want to
943 * implement more complex mount option showing.
944 *
945 * See also save_mount_options().
946 */
Al Viro34c80b12011-12-08 21:32:45 -0500947int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800948{
Al Viro2a32ceb2009-05-08 16:05:57 -0400949 const char *options;
950
951 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -0500952 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800953
954 if (options != NULL && options[0]) {
955 seq_putc(m, ',');
956 mangle(m, options);
957 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400958 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800959
960 return 0;
961}
962EXPORT_SYMBOL(generic_show_options);
963
964/*
965 * If filesystem uses generic_show_options(), this function should be
966 * called from the fill_super() callback.
967 *
968 * The .remount_fs callback usually needs to be handled in a special
969 * way, to make sure, that previous options are not overwritten if the
970 * remount fails.
971 *
972 * Also note, that if the filesystem's .remount_fs function doesn't
973 * reset all options to their default value, but changes only newly
974 * given options, then the displayed options will not reflect reality
975 * any more.
976 */
977void save_mount_options(struct super_block *sb, char *options)
978{
Al Viro2a32ceb2009-05-08 16:05:57 -0400979 BUG_ON(sb->s_options);
980 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800981}
982EXPORT_SYMBOL(save_mount_options);
983
Al Viro2a32ceb2009-05-08 16:05:57 -0400984void replace_mount_options(struct super_block *sb, char *options)
985{
986 char *old = sb->s_options;
987 rcu_assign_pointer(sb->s_options, options);
988 if (old) {
989 synchronize_rcu();
990 kfree(old);
991 }
992}
993EXPORT_SYMBOL(replace_mount_options);
994
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100995#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -0500996/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997static void *m_start(struct seq_file *m, loff_t *pos)
998{
Al Viro6ce6e242012-06-09 01:16:59 -0400999 struct proc_mounts *p = proc_mounts(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Ram Pai390c6842005-11-07 17:17:51 -05001001 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001002 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003}
1004
1005static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1006{
Al Viro6ce6e242012-06-09 01:16:59 -04001007 struct proc_mounts *p = proc_mounts(m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001008
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001009 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010}
1011
1012static void m_stop(struct seq_file *m, void *v)
1013{
Ram Pai390c6842005-11-07 17:17:51 -05001014 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015}
1016
Al Viro0226f492011-12-06 12:21:54 -05001017static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001018{
Al Viro6ce6e242012-06-09 01:16:59 -04001019 struct proc_mounts *p = proc_mounts(m);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001020 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001021 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022}
1023
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001024const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 .start = m_start,
1026 .next = m_next,
1027 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001028 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001029};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001030#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032/**
1033 * may_umount_tree - check if a mount tree is busy
1034 * @mnt: root of mount tree
1035 *
1036 * This is called to check if a tree of mounts has any
1037 * open files, pwds, chroots or sub mounts that are
1038 * busy.
1039 */
Al Viro909b0a82011-11-25 03:06:56 -05001040int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041{
Al Viro909b0a82011-11-25 03:06:56 -05001042 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001043 int actual_refs = 0;
1044 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001045 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001046 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Nick Pigginb3e19d92011-01-07 17:50:11 +11001048 /* write lock needed for mnt_get_count */
Andi Kleen962830d2012-05-08 13:32:02 +09301049 br_write_lock(&vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -05001050 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001051 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 }
Andi Kleen962830d2012-05-08 13:32:02 +09301054 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
1056 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001057 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
Ian Kente3474a82006-03-27 01:14:51 -08001059 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060}
1061
1062EXPORT_SYMBOL(may_umount_tree);
1063
1064/**
1065 * may_umount - check if a mount point is busy
1066 * @mnt: root of mount
1067 *
1068 * This is called to check if a mount point has any
1069 * open files, pwds, chroots or sub mounts. If the
1070 * mount has sub mounts this will return busy
1071 * regardless of whether the sub mounts are busy.
1072 *
1073 * Doesn't take quota and stuff into account. IOW, in some cases it will
1074 * give false negatives. The main reason why it's here is that we need
1075 * a non-destructive way to look for easily umountable filesystems.
1076 */
1077int may_umount(struct vfsmount *mnt)
1078{
Ian Kente3474a82006-03-27 01:14:51 -08001079 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001080 down_read(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301081 br_write_lock(&vfsmount_lock);
Al Viro1ab59732011-11-24 21:35:16 -05001082 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001083 ret = 0;
Andi Kleen962830d2012-05-08 13:32:02 +09301084 br_write_unlock(&vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001085 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001086 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087}
1088
1089EXPORT_SYMBOL(may_umount);
1090
Ram Paib90fa9a2005-11-07 17:19:50 -05001091void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092{
Al Virod5e50f72011-11-24 20:58:57 -05001093 struct mount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001094 while (!list_empty(head)) {
Al Viro1b8e5562011-11-24 21:01:32 -05001095 mnt = list_first_entry(head, struct mount, mnt_hash);
1096 list_del_init(&mnt->mnt_hash);
Al Viro676da582011-11-24 21:47:05 -05001097 if (mnt_has_parent(mnt)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001098 struct dentry *dentry;
Al Viro863d6842011-11-25 00:57:42 -05001099 struct mount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001100
Andi Kleen962830d2012-05-08 13:32:02 +09301101 br_write_lock(&vfsmount_lock);
Al Viroa73324d2011-11-24 22:25:07 -05001102 dentry = mnt->mnt_mountpoint;
Al Viro863d6842011-11-25 00:57:42 -05001103 m = mnt->mnt_parent;
Al Viroa73324d2011-11-24 22:25:07 -05001104 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -05001105 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001106 m->mnt_ghosts--;
Andi Kleen962830d2012-05-08 13:32:02 +09301107 br_write_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001108 dput(dentry);
Al Viro863d6842011-11-25 00:57:42 -05001109 mntput(&m->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
Al Virod5e50f72011-11-24 20:58:57 -05001111 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001112 }
1113}
1114
Nick Piggin99b7db72010-08-18 04:37:39 +10001115/*
1116 * vfsmount lock must be held for write
1117 * namespace_sem must be held for write
1118 */
Al Viro761d5c32011-11-24 21:07:43 -05001119void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001120{
Al Viro7b8a53f2011-01-15 20:08:44 -05001121 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001122 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001123
Al Viro909b0a82011-11-25 03:06:56 -05001124 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro1b8e5562011-11-24 21:01:32 -05001125 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001126
Ram Paia05964f2005-11-07 17:20:17 -05001127 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001128 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001129
Al Viro1b8e5562011-11-24 21:01:32 -05001130 list_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001131 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001132 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001133 __touch_mnt_namespace(p->mnt_ns);
1134 p->mnt_ns = NULL;
Al Viro6b41d532011-11-24 23:24:33 -05001135 list_del_init(&p->mnt_child);
Al Viro676da582011-11-24 21:47:05 -05001136 if (mnt_has_parent(p)) {
Al Viro863d6842011-11-25 00:57:42 -05001137 p->mnt_parent->mnt_ghosts++;
Al Viroa73324d2011-11-24 22:25:07 -05001138 dentry_reset_mounted(p->mnt_mountpoint);
Al Viro7c4b93d2008-03-21 23:59:49 -04001139 }
Al Viro0f0afb12011-11-24 20:43:10 -05001140 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 }
Al Viro7b8a53f2011-01-15 20:08:44 -05001142 list_splice(&tmp_list, kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143}
1144
Al Viro692afc32011-11-24 21:15:14 -05001145static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
Al Viroc35038b2008-03-22 00:46:23 -04001146
Al Viro1ab59732011-11-24 21:35:16 -05001147static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148{
Al Viro1ab59732011-11-24 21:35:16 -05001149 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001151 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
Al Viro1ab59732011-11-24 21:35:16 -05001153 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if (retval)
1155 return retval;
1156
1157 /*
1158 * Allow userspace to request a mountpoint be expired rather than
1159 * unmounting unconditionally. Unmount only happens if:
1160 * (1) the mark is already set (the mark is cleared by mntput())
1161 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1162 */
1163 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001164 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 flags & (MNT_FORCE | MNT_DETACH))
1166 return -EINVAL;
1167
Nick Pigginb3e19d92011-01-07 17:50:11 +11001168 /*
1169 * probably don't strictly need the lock here if we examined
1170 * all race cases, but it's a slowpath.
1171 */
Andi Kleen962830d2012-05-08 13:32:02 +09301172 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -05001173 if (mnt_get_count(mnt) != 2) {
Andi Kleen962830d2012-05-08 13:32:02 +09301174 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001176 }
Andi Kleen962830d2012-05-08 13:32:02 +09301177 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
Al Viro863d6842011-11-25 00:57:42 -05001179 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 return -EAGAIN;
1181 }
1182
1183 /*
1184 * If we may have to abort operations to get out of this
1185 * mount, and they will themselves hold resources we must
1186 * allow the fs to do things. In the Unix tradition of
1187 * 'Gee thats tricky lets do it in userspace' the umount_begin
1188 * might fail to complete on the first run through as other tasks
1189 * must return, and the like. Thats for the mount program to worry
1190 * about for the moment.
1191 */
1192
Al Viro42faad92008-04-24 07:21:56 -04001193 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001194 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
1197 /*
1198 * No sense to grab the lock for this test, but test itself looks
1199 * somewhat bogus. Suggestions for better replacement?
1200 * Ho-hum... In principle, we might treat that as umount + switch
1201 * to rootfs. GC would eventually take care of the old vfsmount.
1202 * Actually it makes sense, especially if rootfs would contain a
1203 * /reboot - static binary that would close all descriptors and
1204 * call reboot(9). Then init(8) could umount root and exec /reboot.
1205 */
Al Viro1ab59732011-11-24 21:35:16 -05001206 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 /*
1208 * Special case for "unmounting" root ...
1209 * we just try to remount it readonly.
1210 */
1211 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001212 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 up_write(&sb->s_umount);
1215 return retval;
1216 }
1217
Ram Pai390c6842005-11-07 17:17:51 -05001218 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301219 br_write_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001220 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Al Viroc35038b2008-03-22 00:46:23 -04001222 if (!(flags & MNT_DETACH))
Al Viro1ab59732011-11-24 21:35:16 -05001223 shrink_submounts(mnt, &umount_list);
Al Viroc35038b2008-03-22 00:46:23 -04001224
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001226 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001227 if (!list_empty(&mnt->mnt_list))
Al Viro1ab59732011-11-24 21:35:16 -05001228 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 retval = 0;
1230 }
Andi Kleen962830d2012-05-08 13:32:02 +09301231 br_write_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001232 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001233 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 return retval;
1235}
1236
1237/*
1238 * Now umount can handle mount points as well as block devices.
1239 * This is important for filesystems which use unnamed block devices.
1240 *
1241 * We now support a flag for forced unmount like the other 'big iron'
1242 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1243 */
1244
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001245SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246{
Al Viro2d8f3032008-07-22 09:59:21 -04001247 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001248 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001250 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001252 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1253 return -EINVAL;
1254
1255 if (!(flags & UMOUNT_NOFOLLOW))
1256 lookup_flags |= LOOKUP_FOLLOW;
1257
1258 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 if (retval)
1260 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001261 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001263 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001265 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 goto dput_and_out;
1267
1268 retval = -EPERM;
1269 if (!capable(CAP_SYS_ADMIN))
1270 goto dput_and_out;
1271
Al Viro900148d2011-11-25 00:33:11 -05001272 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001274 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001275 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001276 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277out:
1278 return retval;
1279}
1280
1281#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1282
1283/*
Ram Paib58fed82005-11-07 17:16:09 -05001284 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001286SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287{
Ram Paib58fed82005-11-07 17:16:09 -05001288 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289}
1290
1291#endif
1292
Al Viro2d92ab32008-08-02 00:51:11 -04001293static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294{
1295 if (capable(CAP_SYS_ADMIN))
1296 return 0;
1297 return -EPERM;
1298#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001299 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001301 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001302 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 return -EPERM;
1304 }
Al Viro2d92ab32008-08-02 00:51:11 -04001305 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 return -EPERM;
1307 return 0;
1308#endif
1309}
1310
Al Viro87129cc2011-11-24 21:24:27 -05001311struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001312 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313{
Al Viroa73324d2011-11-24 22:25:07 -05001314 struct mount *res, *p, *q, *r;
Al Viro1a390682008-03-21 20:48:19 -04001315 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Al Virofc7be132011-11-25 01:05:37 -05001317 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
David Howellsbe34d1a2012-06-25 12:55:18 +01001318 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001319
Ram Pai36341f62005-11-07 17:17:22 -05001320 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001321 if (IS_ERR(q))
1322 return q;
1323
Al Viroa73324d2011-11-24 22:25:07 -05001324 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
1326 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001327 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001328 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001329 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 continue;
1331
Al Viro909b0a82011-11-25 03:06:56 -05001332 for (s = r; s; s = next_mnt(s, r)) {
Al Virofc7be132011-11-25 01:05:37 -05001333 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001334 s = skip_mnt_tree(s);
1335 continue;
1336 }
Al Viro0714a532011-11-24 22:19:58 -05001337 while (p != s->mnt_parent) {
1338 p = p->mnt_parent;
1339 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 }
Al Viro87129cc2011-11-24 21:24:27 -05001341 p = s;
Al Virocb338d02011-11-24 20:55:08 -05001342 path.mnt = &q->mnt;
Al Viroa73324d2011-11-24 22:25:07 -05001343 path.dentry = p->mnt_mountpoint;
Al Viro87129cc2011-11-24 21:24:27 -05001344 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001345 if (IS_ERR(q))
1346 goto out;
Andi Kleen962830d2012-05-08 13:32:02 +09301347 br_write_lock(&vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001348 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Virocb338d02011-11-24 20:55:08 -05001349 attach_mnt(q, &path);
Andi Kleen962830d2012-05-08 13:32:02 +09301350 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 }
1352 }
1353 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001354out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001356 LIST_HEAD(umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301357 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001358 umount_tree(res, 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301359 br_write_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001360 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001362 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363}
1364
David Howellsbe34d1a2012-06-25 12:55:18 +01001365/* Caller should check returned pointer for errors */
1366
Al Viro589ff872009-04-18 03:28:19 -04001367struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001368{
Al Virocb338d02011-11-24 20:55:08 -05001369 struct mount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001370 down_write(&namespace_sem);
Al Viro87129cc2011-11-24 21:24:27 -05001371 tree = copy_tree(real_mount(path->mnt), path->dentry,
1372 CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001373 up_write(&namespace_sem);
David Howellsbe34d1a2012-06-25 12:55:18 +01001374 if (IS_ERR(tree))
1375 return NULL;
1376 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001377}
1378
1379void drop_collected_mounts(struct vfsmount *mnt)
1380{
1381 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001382 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301383 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001384 umount_tree(real_mount(mnt), 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301385 br_write_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001386 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001387 release_mounts(&umount_list);
1388}
1389
Al Viro1f707132010-01-30 22:51:25 -05001390int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1391 struct vfsmount *root)
1392{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001393 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001394 int res = f(root, arg);
1395 if (res)
1396 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001397 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1398 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001399 if (res)
1400 return res;
1401 }
1402 return 0;
1403}
1404
Al Viro4b8b21f2011-11-24 19:54:23 -05001405static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001406{
Al Viro315fc832011-11-24 18:57:30 -05001407 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001408
Al Viro909b0a82011-11-25 03:06:56 -05001409 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001410 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001411 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001412 }
1413}
1414
Al Viro4b8b21f2011-11-24 19:54:23 -05001415static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001416{
Al Viro315fc832011-11-24 18:57:30 -05001417 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001418
Al Viro909b0a82011-11-25 03:06:56 -05001419 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001420 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001421 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001422 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001423 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001424 return err;
1425 }
1426 }
1427 }
1428
1429 return 0;
1430}
1431
Ram Paib90fa9a2005-11-07 17:19:50 -05001432/*
1433 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001434 * @nd : place the mount tree @source_mnt is attached
1435 * @parent_nd : if non-null, detach the source_mnt from its parent and
1436 * store the parent mount and mountpoint dentry.
1437 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001438 *
1439 * NOTE: in the table below explains the semantics when a source mount
1440 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001441 * ---------------------------------------------------------------------------
1442 * | BIND MOUNT OPERATION |
1443 * |**************************************************************************
1444 * | source-->| shared | private | slave | unbindable |
1445 * | dest | | | | |
1446 * | | | | | | |
1447 * | v | | | | |
1448 * |**************************************************************************
1449 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1450 * | | | | | |
1451 * |non-shared| shared (+) | private | slave (*) | invalid |
1452 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001453 * A bind operation clones the source mount and mounts the clone on the
1454 * destination mount.
1455 *
1456 * (++) the cloned mount is propagated to all the mounts in the propagation
1457 * tree of the destination mount and the cloned mount is added to
1458 * the peer group of the source mount.
1459 * (+) the cloned mount is created under the destination mount and is marked
1460 * as shared. The cloned mount is added to the peer group of the source
1461 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001462 * (+++) the mount is propagated to all the mounts in the propagation tree
1463 * of the destination mount and the cloned mount is made slave
1464 * of the same master as that of the source mount. The cloned mount
1465 * is marked as 'shared and slave'.
1466 * (*) the cloned mount is made a slave of the same master as that of the
1467 * source mount.
1468 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001469 * ---------------------------------------------------------------------------
1470 * | MOVE MOUNT OPERATION |
1471 * |**************************************************************************
1472 * | source-->| shared | private | slave | unbindable |
1473 * | dest | | | | |
1474 * | | | | | | |
1475 * | v | | | | |
1476 * |**************************************************************************
1477 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1478 * | | | | | |
1479 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1480 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001481 *
1482 * (+) the mount is moved to the destination. And is then propagated to
1483 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001484 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001485 * (+++) the mount is moved to the destination and is then propagated to
1486 * all the mounts belonging to the destination mount's propagation tree.
1487 * the mount is marked as 'shared and slave'.
1488 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001489 *
1490 * if the source mount is a tree, the operations explained above is
1491 * applied to each mount in the tree.
1492 * Must be called without spinlocks held, since this function can sleep
1493 * in allocations.
1494 */
Al Viro0fb54e52011-11-24 19:59:16 -05001495static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001496 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001497{
1498 LIST_HEAD(tree_list);
Al Viroa8d56d82011-11-24 23:59:29 -05001499 struct mount *dest_mnt = real_mount(path->mnt);
Al Viro1a390682008-03-21 20:48:19 -04001500 struct dentry *dest_dentry = path->dentry;
Al Viro315fc832011-11-24 18:57:30 -05001501 struct mount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001502 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001503
Al Virofc7be132011-11-25 01:05:37 -05001504 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001505 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001506 if (err)
1507 goto out;
1508 }
Al Viroa8d56d82011-11-24 23:59:29 -05001509 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001510 if (err)
1511 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001512
Andi Kleen962830d2012-05-08 13:32:02 +09301513 br_write_lock(&vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001514
Al Virofc7be132011-11-25 01:05:37 -05001515 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro909b0a82011-11-25 03:06:56 -05001516 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001517 set_mnt_shared(p);
Ram Paib90fa9a2005-11-07 17:19:50 -05001518 }
Al Viro1a390682008-03-21 20:48:19 -04001519 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001520 detach_mnt(source_mnt, parent_path);
1521 attach_mnt(source_mnt, path);
Al Viro143c8c92011-11-25 00:46:35 -05001522 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001523 } else {
Al Viro14cf1fa2011-11-25 00:01:17 -05001524 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
Al Viro0fb54e52011-11-24 19:59:16 -05001525 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05001526 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001527
Al Viro1b8e5562011-11-24 21:01:32 -05001528 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1529 list_del_init(&child->mnt_hash);
Al Viro4b2619a2011-11-24 19:47:15 -05001530 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05001531 }
Andi Kleen962830d2012-05-08 13:32:02 +09301532 br_write_unlock(&vfsmount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +10001533
Ram Paib90fa9a2005-11-07 17:19:50 -05001534 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001535
1536 out_cleanup_ids:
Al Virofc7be132011-11-25 01:05:37 -05001537 if (IS_MNT_SHARED(dest_mnt))
Al Viro0fb54e52011-11-24 19:59:16 -05001538 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001539 out:
1540 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001541}
1542
Al Virob12cea92011-03-18 08:55:38 -04001543static int lock_mount(struct path *path)
1544{
1545 struct vfsmount *mnt;
1546retry:
1547 mutex_lock(&path->dentry->d_inode->i_mutex);
1548 if (unlikely(cant_mount(path->dentry))) {
1549 mutex_unlock(&path->dentry->d_inode->i_mutex);
1550 return -ENOENT;
1551 }
1552 down_write(&namespace_sem);
1553 mnt = lookup_mnt(path);
1554 if (likely(!mnt))
1555 return 0;
1556 up_write(&namespace_sem);
1557 mutex_unlock(&path->dentry->d_inode->i_mutex);
1558 path_put(path);
1559 path->mnt = mnt;
1560 path->dentry = dget(mnt->mnt_root);
1561 goto retry;
1562}
1563
1564static void unlock_mount(struct path *path)
1565{
1566 up_write(&namespace_sem);
1567 mutex_unlock(&path->dentry->d_inode->i_mutex);
1568}
1569
Al Viro95bc5f22011-11-25 00:30:56 -05001570static int graft_tree(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571{
Al Viro95bc5f22011-11-25 00:30:56 -05001572 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 return -EINVAL;
1574
Al Viro8c3ee422008-03-22 18:00:39 -04001575 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001576 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 return -ENOTDIR;
1578
Al Virob12cea92011-03-18 08:55:38 -04001579 if (d_unlinked(path->dentry))
1580 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
Al Viro95bc5f22011-11-25 00:30:56 -05001582 return attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583}
1584
1585/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001586 * Sanity check the flags to change_mnt_propagation.
1587 */
1588
1589static int flags_to_propagation_type(int flags)
1590{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001591 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001592
1593 /* Fail if any non-propagation flags are set */
1594 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1595 return 0;
1596 /* Only one propagation flag should be set */
1597 if (!is_power_of_2(type))
1598 return 0;
1599 return type;
1600}
1601
1602/*
Ram Pai07b20882005-11-07 17:19:07 -05001603 * recursively change the type of the mountpoint.
1604 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001605static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001606{
Al Viro315fc832011-11-24 18:57:30 -05001607 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001608 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001609 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001610 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001611 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001612
Miklos Szerediee6f9582007-05-08 00:30:40 -07001613 if (!capable(CAP_SYS_ADMIN))
1614 return -EPERM;
1615
Al Viro2d92ab32008-08-02 00:51:11 -04001616 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001617 return -EINVAL;
1618
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001619 type = flags_to_propagation_type(flag);
1620 if (!type)
1621 return -EINVAL;
1622
Ram Pai07b20882005-11-07 17:19:07 -05001623 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001624 if (type == MS_SHARED) {
1625 err = invent_group_ids(mnt, recurse);
1626 if (err)
1627 goto out_unlock;
1628 }
1629
Andi Kleen962830d2012-05-08 13:32:02 +09301630 br_write_lock(&vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -05001631 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001632 change_mnt_propagation(m, type);
Andi Kleen962830d2012-05-08 13:32:02 +09301633 br_write_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001634
1635 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001636 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001637 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001638}
1639
1640/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 * do loopback mount.
1642 */
Al Viro808d4e32012-10-11 11:42:01 -04001643static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001644 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645{
Al Virob12cea92011-03-18 08:55:38 -04001646 LIST_HEAD(umount_list);
Al Viro2d92ab32008-08-02 00:51:11 -04001647 struct path old_path;
Al Viro87129cc2011-11-24 21:24:27 -05001648 struct mount *mnt = NULL, *old;
Al Viro2d92ab32008-08-02 00:51:11 -04001649 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 if (err)
1651 return err;
1652 if (!old_name || !*old_name)
1653 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001654 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 if (err)
1656 return err;
1657
Al Virob12cea92011-03-18 08:55:38 -04001658 err = lock_mount(path);
1659 if (err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001660 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001661
Al Viro87129cc2011-11-24 21:24:27 -05001662 old = real_mount(old_path.mnt);
1663
Al Virob12cea92011-03-18 08:55:38 -04001664 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001665 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04001666 goto out2;
1667
Al Viro143c8c92011-11-25 00:46:35 -05001668 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
Al Virob12cea92011-03-18 08:55:38 -04001669 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
Al Viroccd48bc2005-11-07 17:15:04 -05001671 if (recurse)
Al Viro87129cc2011-11-24 21:24:27 -05001672 mnt = copy_tree(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001673 else
Al Viro87129cc2011-11-24 21:24:27 -05001674 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001675
David Howellsbe34d1a2012-06-25 12:55:18 +01001676 if (IS_ERR(mnt)) {
1677 err = PTR_ERR(mnt);
1678 goto out;
1679 }
Al Viroccd48bc2005-11-07 17:15:04 -05001680
Al Viro95bc5f22011-11-25 00:30:56 -05001681 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001682 if (err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301683 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001684 umount_tree(mnt, 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301685 br_write_unlock(&vfsmount_lock);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001686 }
Al Virob12cea92011-03-18 08:55:38 -04001687out2:
1688 unlock_mount(path);
1689 release_mounts(&umount_list);
Al Viroccd48bc2005-11-07 17:15:04 -05001690out:
Al Viro2d92ab32008-08-02 00:51:11 -04001691 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 return err;
1693}
1694
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001695static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1696{
1697 int error = 0;
1698 int readonly_request = 0;
1699
1700 if (ms_flags & MS_RDONLY)
1701 readonly_request = 1;
1702 if (readonly_request == __mnt_is_readonly(mnt))
1703 return 0;
1704
1705 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05001706 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001707 else
Al Viro83adc752011-11-24 22:37:54 -05001708 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001709 return error;
1710}
1711
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712/*
1713 * change filesystem flags. dir should be a physical root of filesystem.
1714 * If you've mounted a non-root directory somewhere and want to do remount
1715 * on it - tough luck.
1716 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001717static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 void *data)
1719{
1720 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001721 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05001722 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
1724 if (!capable(CAP_SYS_ADMIN))
1725 return -EPERM;
1726
Al Viro143c8c92011-11-25 00:46:35 -05001727 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 return -EINVAL;
1729
Al Viro2d92ab32008-08-02 00:51:11 -04001730 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 return -EINVAL;
1732
Eric Parisff36fe22011-03-03 16:09:14 -05001733 err = security_sb_remount(sb, data);
1734 if (err)
1735 return err;
1736
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001738 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001739 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001740 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001741 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001742 if (!err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301743 br_write_lock(&vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001744 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1745 mnt->mnt.mnt_flags = mnt_flags;
Andi Kleen962830d2012-05-08 13:32:02 +09301746 br_write_unlock(&vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001747 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001749 if (!err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301750 br_write_lock(&vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001751 touch_mnt_namespace(mnt->mnt_ns);
Andi Kleen962830d2012-05-08 13:32:02 +09301752 br_write_unlock(&vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001753 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 return err;
1755}
1756
Al Virocbbe3622011-11-24 20:01:19 -05001757static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05001758{
Al Viro315fc832011-11-24 18:57:30 -05001759 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001760 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001761 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05001762 return 1;
1763 }
1764 return 0;
1765}
1766
Al Viro808d4e32012-10-11 11:42:01 -04001767static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768{
Al Viro2d92ab32008-08-02 00:51:11 -04001769 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05001770 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05001771 struct mount *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 int err = 0;
1773 if (!capable(CAP_SYS_ADMIN))
1774 return -EPERM;
1775 if (!old_name || !*old_name)
1776 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001777 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 if (err)
1779 return err;
1780
Al Virob12cea92011-03-18 08:55:38 -04001781 err = lock_mount(path);
David Howellscc53ce52011-01-14 18:45:26 +00001782 if (err < 0)
1783 goto out;
1784
Al Viro143c8c92011-11-25 00:46:35 -05001785 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05001786 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05001787
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001789 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 goto out1;
1791
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001792 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001793 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
1795 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001796 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001797 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
Al Viro676da582011-11-24 21:47:05 -05001799 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05001800 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
Al Viro2d92ab32008-08-02 00:51:11 -04001802 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1803 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001804 goto out1;
1805 /*
1806 * Don't move a mount residing in a shared parent.
1807 */
Al Virofc7be132011-11-25 01:05:37 -05001808 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001809 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001810 /*
1811 * Don't move a mount tree containing unbindable mounts to a destination
1812 * mount which is shared.
1813 */
Al Virofc7be132011-11-25 01:05:37 -05001814 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05001815 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05001817 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05001818 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05001819 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
Al Viro0fb54e52011-11-24 19:59:16 -05001821 err = attach_recursive_mnt(old, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001822 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001823 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824
1825 /* if the mount is moved, it should no longer be expire
1826 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05001827 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828out1:
Al Virob12cea92011-03-18 08:55:38 -04001829 unlock_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001832 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001833 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 return err;
1835}
1836
Al Viro9d412a42011-03-17 22:08:28 -04001837static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1838{
1839 int err;
1840 const char *subtype = strchr(fstype, '.');
1841 if (subtype) {
1842 subtype++;
1843 err = -EINVAL;
1844 if (!subtype[0])
1845 goto err;
1846 } else
1847 subtype = "";
1848
1849 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1850 err = -ENOMEM;
1851 if (!mnt->mnt_sb->s_subtype)
1852 goto err;
1853 return mnt;
1854
1855 err:
1856 mntput(mnt);
1857 return ERR_PTR(err);
1858}
1859
Al Viro79e801a2011-12-12 23:07:05 -05001860static struct vfsmount *
Al Viro9d412a42011-03-17 22:08:28 -04001861do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1862{
1863 struct file_system_type *type = get_fs_type(fstype);
1864 struct vfsmount *mnt;
1865 if (!type)
1866 return ERR_PTR(-ENODEV);
1867 mnt = vfs_kern_mount(type, flags, name, data);
1868 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1869 !mnt->mnt_sb->s_subtype)
1870 mnt = fs_set_subtype(mnt, fstype);
1871 put_filesystem(type);
1872 return mnt;
1873}
Al Viro9d412a42011-03-17 22:08:28 -04001874
1875/*
1876 * add a mount into a namespace's mount tree
1877 */
Al Viro95bc5f22011-11-25 00:30:56 -05001878static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04001879{
1880 int err;
1881
1882 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1883
Al Virob12cea92011-03-18 08:55:38 -04001884 err = lock_mount(path);
1885 if (err)
1886 return err;
Al Viro9d412a42011-03-17 22:08:28 -04001887
1888 err = -EINVAL;
Al Viro156cacb2012-09-21 08:19:02 -04001889 if (unlikely(!check_mnt(real_mount(path->mnt)))) {
1890 /* that's acceptable only for automounts done in private ns */
1891 if (!(mnt_flags & MNT_SHRINKABLE))
1892 goto unlock;
1893 /* ... and for those we'd better have mountpoint still alive */
1894 if (!real_mount(path->mnt)->mnt_ns)
1895 goto unlock;
1896 }
Al Viro9d412a42011-03-17 22:08:28 -04001897
1898 /* Refuse the same filesystem on the same mount point */
1899 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05001900 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04001901 path->mnt->mnt_root == path->dentry)
1902 goto unlock;
1903
1904 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05001905 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04001906 goto unlock;
1907
Al Viro95bc5f22011-11-25 00:30:56 -05001908 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro9d412a42011-03-17 22:08:28 -04001909 err = graft_tree(newmnt, path);
1910
1911unlock:
Al Virob12cea92011-03-18 08:55:38 -04001912 unlock_mount(path);
Al Viro9d412a42011-03-17 22:08:28 -04001913 return err;
1914}
Al Virob1e75df2011-01-17 01:47:59 -05001915
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916/*
1917 * create a new mount for userspace and request it to be added into the
1918 * namespace's tree
1919 */
Al Viro808d4e32012-10-11 11:42:01 -04001920static int do_new_mount(struct path *path, const char *type, int flags,
1921 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922{
1923 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05001924 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
Vegard Nossumeca6f532009-09-18 13:05:45 -07001926 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 return -EINVAL;
1928
1929 /* we need capabilities... */
1930 if (!capable(CAP_SYS_ADMIN))
1931 return -EPERM;
1932
1933 mnt = do_kern_mount(type, flags, name, data);
1934 if (IS_ERR(mnt))
1935 return PTR_ERR(mnt);
1936
Al Viro95bc5f22011-11-25 00:30:56 -05001937 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05001938 if (err)
1939 mntput(mnt);
1940 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941}
1942
Al Viro19a167a2011-01-17 01:35:23 -05001943int finish_automount(struct vfsmount *m, struct path *path)
1944{
Al Viro6776db3d2011-11-25 00:22:05 -05001945 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05001946 int err;
1947 /* The new mount record should have at least 2 refs to prevent it being
1948 * expired before we get a chance to add it
1949 */
Al Viro6776db3d2011-11-25 00:22:05 -05001950 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05001951
1952 if (m->mnt_sb == path->mnt->mnt_sb &&
1953 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05001954 err = -ELOOP;
1955 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05001956 }
1957
Al Viro95bc5f22011-11-25 00:30:56 -05001958 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05001959 if (!err)
1960 return 0;
1961fail:
1962 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05001963 if (!list_empty(&mnt->mnt_expire)) {
Al Virob1e75df2011-01-17 01:47:59 -05001964 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301965 br_write_lock(&vfsmount_lock);
Al Viro6776db3d2011-11-25 00:22:05 -05001966 list_del_init(&mnt->mnt_expire);
Andi Kleen962830d2012-05-08 13:32:02 +09301967 br_write_unlock(&vfsmount_lock);
Al Virob1e75df2011-01-17 01:47:59 -05001968 up_write(&namespace_sem);
Al Viro19a167a2011-01-17 01:35:23 -05001969 }
Al Virob1e75df2011-01-17 01:47:59 -05001970 mntput(m);
1971 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05001972 return err;
1973}
1974
David Howellsea5b7782011-01-14 19:10:03 +00001975/**
1976 * mnt_set_expiry - Put a mount on an expiration list
1977 * @mnt: The mount to list.
1978 * @expiry_list: The list to add the mount to.
1979 */
1980void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1981{
1982 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301983 br_write_lock(&vfsmount_lock);
David Howellsea5b7782011-01-14 19:10:03 +00001984
Al Viro6776db3d2011-11-25 00:22:05 -05001985 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00001986
Andi Kleen962830d2012-05-08 13:32:02 +09301987 br_write_unlock(&vfsmount_lock);
David Howellsea5b7782011-01-14 19:10:03 +00001988 up_write(&namespace_sem);
1989}
1990EXPORT_SYMBOL(mnt_set_expiry);
1991
1992/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 * process a list of expirable mountpoints with the intent of discarding any
1994 * mountpoints that aren't in use and haven't been touched since last we came
1995 * here
1996 */
1997void mark_mounts_for_expiry(struct list_head *mounts)
1998{
Al Viro761d5c32011-11-24 21:07:43 -05001999 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04002001 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
2003 if (list_empty(mounts))
2004 return;
2005
Al Virobcc5c7d2008-03-22 00:21:53 -04002006 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09302007 br_write_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
2009 /* extract from the expiration list every vfsmount that matches the
2010 * following criteria:
2011 * - only referenced by its parent vfsmount
2012 * - still marked for expiry (marked on the last call here; marks are
2013 * cleared by mntput())
2014 */
Al Viro6776db3d2011-11-25 00:22:05 -05002015 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002016 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002017 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002019 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002021 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002022 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002023 touch_mnt_namespace(mnt->mnt_ns);
Al Virobcc5c7d2008-03-22 00:21:53 -04002024 umount_tree(mnt, 1, &umounts);
2025 }
Andi Kleen962830d2012-05-08 13:32:02 +09302026 br_write_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04002027 up_write(&namespace_sem);
2028
2029 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030}
2031
2032EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2033
2034/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002035 * Ripoff of 'select_parent()'
2036 *
2037 * search the list of submounts for a given mountpoint, and move any
2038 * shrinkable submounts to the 'graveyard' list.
2039 */
Al Viro692afc32011-11-24 21:15:14 -05002040static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002041{
Al Viro692afc32011-11-24 21:15:14 -05002042 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002043 struct list_head *next;
2044 int found = 0;
2045
2046repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002047 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002048resume:
Al Viro6b41d532011-11-24 23:24:33 -05002049 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002050 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002051 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002052
2053 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002054 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002055 continue;
2056 /*
2057 * Descend a level if the d_mounts list is non-empty.
2058 */
Al Viro6b41d532011-11-24 23:24:33 -05002059 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002060 this_parent = mnt;
2061 goto repeat;
2062 }
2063
Al Viro1ab59732011-11-24 21:35:16 -05002064 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002065 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002066 found++;
2067 }
2068 }
2069 /*
2070 * All done at this level ... ascend and resume the search
2071 */
2072 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002073 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002074 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002075 goto resume;
2076 }
2077 return found;
2078}
2079
2080/*
2081 * process a list of expirable mountpoints with the intent of discarding any
2082 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002083 *
2084 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002085 */
Al Viro692afc32011-11-24 21:15:14 -05002086static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002087{
2088 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002089 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002090
Trond Myklebust5528f9112006-06-09 09:34:17 -04002091 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002092 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002093 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002094 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002095 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002096 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08002097 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04002098 }
2099 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002100}
2101
Trond Myklebust5528f9112006-06-09 09:34:17 -04002102/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 * Some copy_from_user() implementations do not return the exact number of
2104 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2105 * Note that this function differs from copy_from_user() in that it will oops
2106 * on bad values of `to', rather than returning a short copy.
2107 */
Ram Paib58fed82005-11-07 17:16:09 -05002108static long exact_copy_from_user(void *to, const void __user * from,
2109 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110{
2111 char *t = to;
2112 const char __user *f = from;
2113 char c;
2114
2115 if (!access_ok(VERIFY_READ, from, n))
2116 return n;
2117
2118 while (n) {
2119 if (__get_user(c, f)) {
2120 memset(t, 0, n);
2121 break;
2122 }
2123 *t++ = c;
2124 f++;
2125 n--;
2126 }
2127 return n;
2128}
2129
Ram Paib58fed82005-11-07 17:16:09 -05002130int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131{
2132 int i;
2133 unsigned long page;
2134 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002135
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 *where = 0;
2137 if (!data)
2138 return 0;
2139
2140 if (!(page = __get_free_page(GFP_KERNEL)))
2141 return -ENOMEM;
2142
2143 /* We only care that *some* data at the address the user
2144 * gave us is valid. Just in case, we'll zero
2145 * the remainder of the page.
2146 */
2147 /* copy_from_user cannot cross TASK_SIZE ! */
2148 size = TASK_SIZE - (unsigned long)data;
2149 if (size > PAGE_SIZE)
2150 size = PAGE_SIZE;
2151
2152 i = size - exact_copy_from_user((void *)page, data, size);
2153 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002154 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 return -EFAULT;
2156 }
2157 if (i != PAGE_SIZE)
2158 memset((char *)page + i, 0, PAGE_SIZE - i);
2159 *where = page;
2160 return 0;
2161}
2162
Vegard Nossumeca6f532009-09-18 13:05:45 -07002163int copy_mount_string(const void __user *data, char **where)
2164{
2165 char *tmp;
2166
2167 if (!data) {
2168 *where = NULL;
2169 return 0;
2170 }
2171
2172 tmp = strndup_user(data, PAGE_SIZE);
2173 if (IS_ERR(tmp))
2174 return PTR_ERR(tmp);
2175
2176 *where = tmp;
2177 return 0;
2178}
2179
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180/*
2181 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2182 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2183 *
2184 * data is a (void *) that can point to any structure up to
2185 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2186 * information (or be NULL).
2187 *
2188 * Pre-0.97 versions of mount() didn't have a flags word.
2189 * When the flags word was introduced its top half was required
2190 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2191 * Therefore, if this magic number is present, it carries no information
2192 * and must be discarded.
2193 */
Al Viro808d4e32012-10-11 11:42:01 -04002194long do_mount(const char *dev_name, const char *dir_name,
2195 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196{
Al Viro2d92ab32008-08-02 00:51:11 -04002197 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 int retval = 0;
2199 int mnt_flags = 0;
2200
2201 /* Discard magic */
2202 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2203 flags &= ~MS_MGC_MSK;
2204
2205 /* Basic sanity checks */
2206
2207 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2208 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
2210 if (data_page)
2211 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2212
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002213 /* ... and get the mountpoint */
2214 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2215 if (retval)
2216 return retval;
2217
2218 retval = security_sb_mount(dev_name, &path,
2219 type_page, flags, data_page);
2220 if (retval)
2221 goto dput_out;
2222
Andi Kleen613cbe32009-04-19 18:40:43 +02002223 /* Default to relatime unless overriden */
2224 if (!(flags & MS_NOATIME))
2225 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002226
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 /* Separate the per-mountpoint flags */
2228 if (flags & MS_NOSUID)
2229 mnt_flags |= MNT_NOSUID;
2230 if (flags & MS_NODEV)
2231 mnt_flags |= MNT_NODEV;
2232 if (flags & MS_NOEXEC)
2233 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002234 if (flags & MS_NOATIME)
2235 mnt_flags |= MNT_NOATIME;
2236 if (flags & MS_NODIRATIME)
2237 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002238 if (flags & MS_STRICTATIME)
2239 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002240 if (flags & MS_RDONLY)
2241 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002242
Al Viro7a4dec52010-08-09 12:05:43 -04002243 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002244 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2245 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002248 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 data_page);
2250 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002251 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002252 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002253 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002255 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 else
Al Viro2d92ab32008-08-02 00:51:11 -04002257 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 dev_name, data_page);
2259dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002260 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 return retval;
2262}
2263
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002264static struct mnt_namespace *alloc_mnt_ns(void)
2265{
2266 struct mnt_namespace *new_ns;
2267
2268 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2269 if (!new_ns)
2270 return ERR_PTR(-ENOMEM);
2271 atomic_set(&new_ns->count, 1);
2272 new_ns->root = NULL;
2273 INIT_LIST_HEAD(&new_ns->list);
2274 init_waitqueue_head(&new_ns->poll);
2275 new_ns->event = 0;
2276 return new_ns;
2277}
2278
JANAK DESAI741a2952006-02-07 12:59:00 -08002279/*
2280 * Allocate a new namespace structure and populate it with contents
2281 * copied from the namespace of the passed in task structure.
2282 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002283static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002284 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002286 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002287 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002288 struct mount *p, *q;
Al Virobe08d6d2011-12-06 13:32:36 -05002289 struct mount *old = mnt_ns->root;
Al Virocb338d02011-11-24 20:55:08 -05002290 struct mount *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002292 new_ns = alloc_mnt_ns();
2293 if (IS_ERR(new_ns))
2294 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295
Ram Pai390c6842005-11-07 17:17:51 -05002296 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 /* First pass: copy the tree topology */
Al Viro909b0a82011-11-25 03:06:56 -05002298 new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE);
David Howellsbe34d1a2012-06-25 12:55:18 +01002299 if (IS_ERR(new)) {
Ram Pai390c6842005-11-07 17:17:51 -05002300 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 kfree(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002302 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 }
Al Virobe08d6d2011-12-06 13:32:36 -05002304 new_ns->root = new;
Andi Kleen962830d2012-05-08 13:32:02 +09302305 br_write_lock(&vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05002306 list_add_tail(&new_ns->list, &new->mnt_list);
Andi Kleen962830d2012-05-08 13:32:02 +09302307 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308
2309 /*
2310 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2311 * as belonging to new namespace. We have already acquired a private
2312 * fs_struct, so tsk->fs->lock is not needed.
2313 */
Al Viro909b0a82011-11-25 03:06:56 -05002314 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002315 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002317 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 if (fs) {
Al Viro315fc832011-11-24 18:57:30 -05002319 if (&p->mnt == fs->root.mnt) {
2320 fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002321 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 }
Al Viro315fc832011-11-24 18:57:30 -05002323 if (&p->mnt == fs->pwd.mnt) {
2324 fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002325 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 }
Al Viro909b0a82011-11-25 03:06:56 -05002328 p = next_mnt(p, old);
2329 q = next_mnt(q, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 }
Ram Pai390c6842005-11-07 17:17:51 -05002331 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002334 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002336 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337
JANAK DESAI741a2952006-02-07 12:59:00 -08002338 return new_ns;
2339}
2340
Eric W. Biederman213dd262007-07-15 23:41:15 -07002341struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002342 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002343{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002344 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002345
Badari Pulavartye3222c42007-05-08 00:25:21 -07002346 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002347 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002348
2349 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002350 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002351
Badari Pulavartye3222c42007-05-08 00:25:21 -07002352 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002353
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002354 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002355 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356}
2357
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002358/**
2359 * create_mnt_ns - creates a private namespace and adds a root filesystem
2360 * @mnt: pointer to the new root filesystem mountpoint
2361 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002362static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002363{
Al Viro1a4eeaf2011-11-25 02:19:55 -05002364 struct mnt_namespace *new_ns = alloc_mnt_ns();
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002365 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002366 struct mount *mnt = real_mount(m);
2367 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002368 new_ns->root = mnt;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002369 list_add(&new_ns->list, &mnt->mnt_list);
Al Viroc1334492011-11-16 16:12:14 -05002370 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002371 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002372 }
2373 return new_ns;
2374}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002375
Al Viroea441d12011-11-16 21:43:59 -05002376struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2377{
2378 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002379 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002380 struct path path;
2381 int err;
2382
2383 ns = create_mnt_ns(mnt);
2384 if (IS_ERR(ns))
2385 return ERR_CAST(ns);
2386
2387 err = vfs_path_lookup(mnt->mnt_root, mnt,
2388 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2389
2390 put_mnt_ns(ns);
2391
2392 if (err)
2393 return ERR_PTR(err);
2394
2395 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002396 s = path.mnt->mnt_sb;
2397 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002398 mntput(path.mnt);
2399 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002400 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002401 /* ... and return the root of (sub)tree on it */
2402 return path.dentry;
2403}
2404EXPORT_SYMBOL(mount_subtree);
2405
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002406SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2407 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002409 int ret;
2410 char *kernel_type;
Jeff Layton91a27b22012-10-10 15:25:28 -04002411 struct filename *kernel_dir;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002412 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414
Vegard Nossumeca6f532009-09-18 13:05:45 -07002415 ret = copy_mount_string(type, &kernel_type);
2416 if (ret < 0)
2417 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418
Vegard Nossumeca6f532009-09-18 13:05:45 -07002419 kernel_dir = getname(dir_name);
2420 if (IS_ERR(kernel_dir)) {
2421 ret = PTR_ERR(kernel_dir);
2422 goto out_dir;
2423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424
Vegard Nossumeca6f532009-09-18 13:05:45 -07002425 ret = copy_mount_string(dev_name, &kernel_dev);
2426 if (ret < 0)
2427 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428
Vegard Nossumeca6f532009-09-18 13:05:45 -07002429 ret = copy_mount_options(data, &data_page);
2430 if (ret < 0)
2431 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432
Jeff Layton91a27b22012-10-10 15:25:28 -04002433 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
Vegard Nossumeca6f532009-09-18 13:05:45 -07002434 (void *) data_page);
2435
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002437out_data:
2438 kfree(kernel_dev);
2439out_dev:
2440 putname(kernel_dir);
2441out_dir:
2442 kfree(kernel_type);
2443out_type:
2444 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445}
2446
2447/*
Al Viroafac7cb2011-11-23 19:34:49 -05002448 * Return true if path is reachable from root
2449 *
2450 * namespace_sem or vfsmount_lock is held
2451 */
Al Viro643822b2011-11-24 22:00:28 -05002452bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002453 const struct path *root)
2454{
Al Viro643822b2011-11-24 22:00:28 -05002455 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002456 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002457 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002458 }
Al Viro643822b2011-11-24 22:00:28 -05002459 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002460}
2461
2462int path_is_under(struct path *path1, struct path *path2)
2463{
2464 int res;
Andi Kleen962830d2012-05-08 13:32:02 +09302465 br_read_lock(&vfsmount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002466 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Andi Kleen962830d2012-05-08 13:32:02 +09302467 br_read_unlock(&vfsmount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002468 return res;
2469}
2470EXPORT_SYMBOL(path_is_under);
2471
2472/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 * pivot_root Semantics:
2474 * Moves the root file system of the current process to the directory put_old,
2475 * makes new_root as the new root file system of the current process, and sets
2476 * root/cwd of all processes which had them on the current root to new_root.
2477 *
2478 * Restrictions:
2479 * The new_root and put_old must be directories, and must not be on the
2480 * same file system as the current process root. The put_old must be
2481 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2482 * pointed to by put_old must yield the same directory as new_root. No other
2483 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2484 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002485 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2486 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2487 * in this situation.
2488 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 * Notes:
2490 * - we don't move root/cwd if they are not at the root (reason: if something
2491 * cared enough to change them, it's probably wrong to force them elsewhere)
2492 * - it's okay to pick a root that isn't the root of a file system, e.g.
2493 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2494 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2495 * first.
2496 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002497SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2498 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499{
Al Viro2d8f3032008-07-22 09:59:21 -04002500 struct path new, old, parent_path, root_parent, root;
Al Viro419148d2011-11-24 19:41:16 -05002501 struct mount *new_mnt, *root_mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 int error;
2503
2504 if (!capable(CAP_SYS_ADMIN))
2505 return -EPERM;
2506
Al Viro2d8f3032008-07-22 09:59:21 -04002507 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 if (error)
2509 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
Al Viro2d8f3032008-07-22 09:59:21 -04002511 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 if (error)
2513 goto out1;
2514
Al Viro2d8f3032008-07-22 09:59:21 -04002515 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002516 if (error)
2517 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002519 get_fs_root(current->fs, &root);
Al Virob12cea92011-03-18 08:55:38 -04002520 error = lock_mount(&old);
2521 if (error)
2522 goto out3;
2523
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002525 new_mnt = real_mount(new.mnt);
2526 root_mnt = real_mount(root.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002527 if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2528 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2529 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002530 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002531 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002532 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002534 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002535 goto out4;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002536 if (d_unlinked(old.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002537 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002539 if (new.mnt == root.mnt ||
2540 old.mnt == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002541 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002543 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002544 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002545 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002546 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002547 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002548 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002549 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002550 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002551 /* make sure we can reach put_old from new_root */
Al Viro643822b2011-11-24 22:00:28 -05002552 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002553 goto out4;
Andi Kleen962830d2012-05-08 13:32:02 +09302554 br_write_lock(&vfsmount_lock);
Al Viro419148d2011-11-24 19:41:16 -05002555 detach_mnt(new_mnt, &parent_path);
2556 detach_mnt(root_mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002557 /* mount old root on put_old */
Al Viro419148d2011-11-24 19:41:16 -05002558 attach_mnt(root_mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002559 /* mount new_root on / */
Al Viro419148d2011-11-24 19:41:16 -05002560 attach_mnt(new_mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002561 touch_mnt_namespace(current->nsproxy->mnt_ns);
Andi Kleen962830d2012-05-08 13:32:02 +09302562 br_write_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002563 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002565out4:
2566 unlock_mount(&old);
2567 if (!error) {
2568 path_put(&root_parent);
2569 path_put(&parent_path);
2570 }
2571out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002572 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002573out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002574 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002576 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579}
2580
2581static void __init init_mount_tree(void)
2582{
2583 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002584 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002585 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586
2587 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2588 if (IS_ERR(mnt))
2589 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002590
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002591 ns = create_mnt_ns(mnt);
2592 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002595 init_task.nsproxy->mnt_ns = ns;
2596 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597
Al Virobe08d6d2011-12-06 13:32:36 -05002598 root.mnt = mnt;
2599 root.dentry = mnt->mnt_root;
Jan Blunckac748a02008-02-14 19:34:39 -08002600
2601 set_fs_pwd(current->fs, &root);
2602 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603}
2604
Denis Cheng74bf17c2007-10-16 23:26:30 -07002605void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002607 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002608 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
Ram Pai390c6842005-11-07 17:17:51 -05002610 init_rwsem(&namespace_sem);
2611
Al Viro7d6fec42011-11-23 12:14:10 -05002612 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002613 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614
Ram Paib58fed82005-11-07 17:16:09 -05002615 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
2617 if (!mount_hashtable)
2618 panic("Failed to allocate mount hash table\n");
2619
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002620 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621
Eric Dumazet13f14b42008-02-06 01:37:57 -08002622 for (u = 0; u < HASH_SIZE; u++)
2623 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624
Andi Kleen962830d2012-05-08 13:32:02 +09302625 br_lock_init(&vfsmount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +10002626
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002627 err = sysfs_init();
2628 if (err)
2629 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002630 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002631 fs_kobj = kobject_create_and_add("fs", NULL);
2632 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002633 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 init_rootfs();
2635 init_mount_tree();
2636}
2637
Trond Myklebust616511d2009-06-22 15:09:13 -04002638void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002640 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002641
Al Virod498b252010-02-05 02:21:06 -05002642 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002643 return;
Ram Pai390c6842005-11-07 17:17:51 -05002644 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09302645 br_write_lock(&vfsmount_lock);
Al Virobe08d6d2011-12-06 13:32:36 -05002646 umount_tree(ns->root, 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09302647 br_write_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002648 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002649 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002650 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651}
Al Viro9d412a42011-03-17 22:08:28 -04002652
2653struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2654{
Tim Chen423e0ab2011-07-19 09:32:38 -07002655 struct vfsmount *mnt;
2656 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2657 if (!IS_ERR(mnt)) {
2658 /*
2659 * it is a longterm mount, don't release mnt until
2660 * we unmount before file sys is unregistered
2661 */
Al Virof7a99c52012-06-09 00:59:08 -04002662 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07002663 }
2664 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002665}
2666EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002667
2668void kern_unmount(struct vfsmount *mnt)
2669{
2670 /* release long term mount so mount point can be released */
2671 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04002672 br_write_lock(&vfsmount_lock);
2673 real_mount(mnt)->mnt_ns = NULL;
2674 br_write_unlock(&vfsmount_lock);
Tim Chen423e0ab2011-07-19 09:32:38 -07002675 mntput(mnt);
2676 }
2677}
2678EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002679
2680bool our_mnt(struct vfsmount *mnt)
2681{
Al Viro143c8c92011-11-25 00:46:35 -05002682 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05002683}