blob: e6081996c9a2f9d26525740545445630c4737583 [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/*
286 * Most r/o checks on a fs are for operations that take
287 * discrete amounts of time, like a write() or unlink().
288 * We must keep track of when those operations start
289 * (for permission checks) and when they end, so that
290 * we can determine when writes are able to occur to
291 * a filesystem.
292 */
Dave Hansen83660252008-02-15 14:37:30 -0800293/**
294 * mnt_want_write - get write access to a mount
Al Viro83adc752011-11-24 22:37:54 -0500295 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800296 *
297 * This tells the low-level filesystem that a write is
298 * about to be performed to it, and makes sure that
299 * writes are allowed before returning success. When
300 * the write operation is finished, mnt_drop_write()
301 * must be called. This is effectively a refcount.
302 */
Al Viro83adc752011-11-24 22:37:54 -0500303int mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800304{
Al Viro83adc752011-11-24 22:37:54 -0500305 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800306 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800307
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000308 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100309 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000310 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100311 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000312 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
313 * incremented count after it has set MNT_WRITE_HOLD.
314 */
315 smp_mb();
Al Viro83adc752011-11-24 22:37:54 -0500316 while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000317 cpu_relax();
318 /*
319 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
320 * be set to match its requirements. So we must not load that until
321 * MNT_WRITE_HOLD is cleared.
322 */
323 smp_rmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100324 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100325 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800326 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800327 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000328 preempt_enable();
Dave Hansen3d733632008-02-15 14:37:59 -0800329 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800330}
331EXPORT_SYMBOL_GPL(mnt_want_write);
332
333/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000334 * mnt_clone_write - get write access to a mount
335 * @mnt: the mount on which to take a write
336 *
337 * This is effectively like mnt_want_write, except
338 * it must only be used to take an extra write reference
339 * on a mountpoint that we already know has a write reference
340 * on it. This allows some optimisation.
341 *
342 * After finished, mnt_drop_write must be called as usual to
343 * drop the reference.
344 */
345int mnt_clone_write(struct vfsmount *mnt)
346{
347 /* superblock may be r/o */
348 if (__mnt_is_readonly(mnt))
349 return -EROFS;
350 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500351 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000352 preempt_enable();
353 return 0;
354}
355EXPORT_SYMBOL_GPL(mnt_clone_write);
356
357/**
358 * mnt_want_write_file - get write access to a file's mount
359 * @file: the file who's mount on which to take a write
360 *
361 * This is like mnt_want_write, but it takes a file and can
362 * do some optimisations if the file is open for write already
363 */
364int mnt_want_write_file(struct file *file)
365{
OGAWA Hirofumi2d8dd382009-08-06 15:07:39 -0700366 struct inode *inode = file->f_dentry->d_inode;
367 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
npiggin@suse.de96029c42009-04-26 20:25:55 +1000368 return mnt_want_write(file->f_path.mnt);
369 else
370 return mnt_clone_write(file->f_path.mnt);
371}
372EXPORT_SYMBOL_GPL(mnt_want_write_file);
373
374/**
Dave Hansen83660252008-02-15 14:37:30 -0800375 * mnt_drop_write - give up write access to a mount
376 * @mnt: the mount on which to give up write access
377 *
378 * Tells the low-level filesystem that we are done
379 * performing writes to it. Must be matched with
380 * mnt_want_write() call above.
381 */
382void mnt_drop_write(struct vfsmount *mnt)
383{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000384 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500385 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000386 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800387}
388EXPORT_SYMBOL_GPL(mnt_drop_write);
389
Al Viro2a79f172011-12-09 08:06:57 -0500390void mnt_drop_write_file(struct file *file)
391{
392 mnt_drop_write(file->f_path.mnt);
393}
394EXPORT_SYMBOL(mnt_drop_write_file);
395
Al Viro83adc752011-11-24 22:37:54 -0500396static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800397{
Dave Hansen3d733632008-02-15 14:37:59 -0800398 int ret = 0;
399
Nick Piggin99b7db72010-08-18 04:37:39 +1000400 br_write_lock(vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500401 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000402 /*
403 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
404 * should be visible before we do.
405 */
406 smp_mb();
407
408 /*
409 * With writers on hold, if this value is zero, then there are
410 * definitely no active writers (although held writers may subsequently
411 * increment the count, they'll have to wait, and decrement it after
412 * seeing MNT_READONLY).
413 *
414 * It is OK to have counter incremented on one CPU and decremented on
415 * another: the sum will add up correctly. The danger would be when we
416 * sum up each counter, if we read a counter before it is incremented,
417 * but then read another CPU's count which it has been subsequently
418 * decremented from -- we would see more decrements than we should.
419 * MNT_WRITE_HOLD protects against this scenario, because
420 * mnt_want_write first increments count, then smp_mb, then spins on
421 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
422 * we're counting up here.
423 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100424 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000425 ret = -EBUSY;
426 else
Al Viro83adc752011-11-24 22:37:54 -0500427 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000428 /*
429 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
430 * that become unheld will see MNT_READONLY.
431 */
432 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500433 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Nick Piggin99b7db72010-08-18 04:37:39 +1000434 br_write_unlock(vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800435 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800436}
Dave Hansen83660252008-02-15 14:37:30 -0800437
Al Viro83adc752011-11-24 22:37:54 -0500438static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800439{
Nick Piggin99b7db72010-08-18 04:37:39 +1000440 br_write_lock(vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500441 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Nick Piggin99b7db72010-08-18 04:37:39 +1000442 br_write_unlock(vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800443}
444
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100445int sb_prepare_remount_readonly(struct super_block *sb)
446{
447 struct mount *mnt;
448 int err = 0;
449
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100450 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
451 if (atomic_long_read(&sb->s_remove_count))
452 return -EBUSY;
453
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100454 br_write_lock(vfsmount_lock);
455 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
456 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
457 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
458 smp_mb();
459 if (mnt_get_writers(mnt) > 0) {
460 err = -EBUSY;
461 break;
462 }
463 }
464 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100465 if (!err && atomic_long_read(&sb->s_remove_count))
466 err = -EBUSY;
467
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100468 if (!err) {
469 sb->s_readonly_remount = 1;
470 smp_wmb();
471 }
472 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
473 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
474 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
475 }
476 br_write_unlock(vfsmount_lock);
477
478 return err;
479}
480
Al Virob105e272011-11-24 20:38:33 -0500481static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
Al Viro52ba1622011-11-25 02:25:17 -0500483 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100484 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000485#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500486 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000487#endif
Al Virob105e272011-11-24 20:38:33 -0500488 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489}
490
491/*
Ram Paia05964f2005-11-07 17:20:17 -0500492 * find the first or last mount at @dentry on vfsmount @mnt depending on
493 * @dir. If @dir is set return the first mount else return the last mount.
Nick Piggin99b7db72010-08-18 04:37:39 +1000494 * vfsmount_lock must be held for read or write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 */
Al Viroc7105362011-11-24 18:22:03 -0500496struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
Ram Paia05964f2005-11-07 17:20:17 -0500497 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498{
Ram Paib58fed82005-11-07 17:16:09 -0500499 struct list_head *head = mount_hashtable + hash(mnt, dentry);
500 struct list_head *tmp = head;
Al Viroc7105362011-11-24 18:22:03 -0500501 struct mount *p, *found = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500504 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 p = NULL;
506 if (tmp == head)
507 break;
Al Viro1b8e5562011-11-24 21:01:32 -0500508 p = list_entry(tmp, struct mount, mnt_hash);
Al Viroa73324d2011-11-24 22:25:07 -0500509 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500510 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 break;
512 }
513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 return found;
515}
516
Ram Paia05964f2005-11-07 17:20:17 -0500517/*
518 * lookup_mnt increments the ref count before returning
519 * the vfsmount struct.
520 */
Al Viro1c755af2009-04-18 14:06:57 -0400521struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500522{
Al Viroc7105362011-11-24 18:22:03 -0500523 struct mount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000524
525 br_read_lock(vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500526 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
527 if (child_mnt) {
528 mnt_add_count(child_mnt, 1);
529 br_read_unlock(vfsmount_lock);
530 return &child_mnt->mnt;
531 } else {
532 br_read_unlock(vfsmount_lock);
533 return NULL;
534 }
Ram Paia05964f2005-11-07 17:20:17 -0500535}
536
Al Viro143c8c92011-11-25 00:46:35 -0500537static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800539 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540}
541
Nick Piggin99b7db72010-08-18 04:37:39 +1000542/*
543 * vfsmount lock must be held for write
544 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800545static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500546{
547 if (ns) {
548 ns->event = ++event;
549 wake_up_interruptible(&ns->poll);
550 }
551}
552
Nick Piggin99b7db72010-08-18 04:37:39 +1000553/*
554 * vfsmount lock must be held for write
555 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800556static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500557{
558 if (ns && ns->event != event) {
559 ns->event = event;
560 wake_up_interruptible(&ns->poll);
561 }
562}
563
Nick Piggin99b7db72010-08-18 04:37:39 +1000564/*
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100565 * Clear dentry's mounted state if it has no remaining mounts.
566 * vfsmount_lock must be held for write.
567 */
Al Viroaa0a4cf2011-11-24 19:31:36 -0500568static void dentry_reset_mounted(struct dentry *dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100569{
570 unsigned u;
571
572 for (u = 0; u < HASH_SIZE; u++) {
Al Virod5e50f72011-11-24 20:58:57 -0500573 struct mount *p;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100574
Al Viro1b8e5562011-11-24 21:01:32 -0500575 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
Al Viroa73324d2011-11-24 22:25:07 -0500576 if (p->mnt_mountpoint == dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100577 return;
578 }
579 }
580 spin_lock(&dentry->d_lock);
581 dentry->d_flags &= ~DCACHE_MOUNTED;
582 spin_unlock(&dentry->d_lock);
583}
584
585/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000586 * vfsmount lock must be held for write
587 */
Al Viro419148d2011-11-24 19:41:16 -0500588static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
Al Viroa73324d2011-11-24 22:25:07 -0500590 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500591 old_path->mnt = &mnt->mnt_parent->mnt;
592 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500593 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500594 list_del_init(&mnt->mnt_child);
Al Viro1b8e5562011-11-24 21:01:32 -0500595 list_del_init(&mnt->mnt_hash);
Al Viroaa0a4cf2011-11-24 19:31:36 -0500596 dentry_reset_mounted(old_path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597}
598
Nick Piggin99b7db72010-08-18 04:37:39 +1000599/*
600 * vfsmount lock must be held for write
601 */
Al Viro14cf1fa2011-11-25 00:01:17 -0500602void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
Al Viro44d964d62011-11-24 21:28:22 -0500603 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500604{
Al Viro3a2393d2011-11-25 03:19:09 -0500605 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viroa73324d2011-11-24 22:25:07 -0500606 child_mnt->mnt_mountpoint = dget(dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500607 child_mnt->mnt_parent = mnt;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100608 spin_lock(&dentry->d_lock);
609 dentry->d_flags |= DCACHE_MOUNTED;
610 spin_unlock(&dentry->d_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500611}
612
Nick Piggin99b7db72010-08-18 04:37:39 +1000613/*
614 * vfsmount lock must be held for write
615 */
Al Viro419148d2011-11-24 19:41:16 -0500616static void attach_mnt(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
Al Viro14cf1fa2011-11-25 00:01:17 -0500618 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
Al Viro1b8e5562011-11-24 21:01:32 -0500619 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400620 hash(path->mnt, path->dentry));
Al Viro6b41d532011-11-24 23:24:33 -0500621 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500622}
623
Al Viro83adc752011-11-24 22:37:54 -0500624static inline void __mnt_make_longterm(struct mount *mnt)
Al Viro7e3d0eb2011-01-16 16:32:11 -0500625{
626#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500627 atomic_inc(&mnt->mnt_longterm);
Al Viro7e3d0eb2011-01-16 16:32:11 -0500628#endif
629}
630
631/* needs vfsmount lock for write */
Al Viro83adc752011-11-24 22:37:54 -0500632static inline void __mnt_make_shortterm(struct mount *mnt)
Al Viro7e3d0eb2011-01-16 16:32:11 -0500633{
634#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500635 atomic_dec(&mnt->mnt_longterm);
Al Viro7e3d0eb2011-01-16 16:32:11 -0500636#endif
637}
638
Ram Paib90fa9a2005-11-07 17:19:50 -0500639/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000640 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500641 */
Al Viro4b2619a2011-11-24 19:47:15 -0500642static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500643{
Al Viro0714a532011-11-24 22:19:58 -0500644 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500645 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500646 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500647 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500648
Al Viro0714a532011-11-24 22:19:58 -0500649 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500650
Al Viro1a4eeaf2011-11-25 02:19:55 -0500651 list_add_tail(&head, &mnt->mnt_list);
652 list_for_each_entry(m, &head, mnt_list) {
Al Viro143c8c92011-11-25 00:46:35 -0500653 m->mnt_ns = n;
Al Viro7e3d0eb2011-01-16 16:32:11 -0500654 __mnt_make_longterm(m);
Al Virof03c6592011-01-14 22:30:21 -0500655 }
656
Ram Paib90fa9a2005-11-07 17:19:50 -0500657 list_splice(&head, n->list.prev);
658
Al Viro1b8e5562011-11-24 21:01:32 -0500659 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viroa73324d2011-11-24 22:25:07 -0500660 hash(&parent->mnt, mnt->mnt_mountpoint));
Al Viro6b41d532011-11-24 23:24:33 -0500661 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800662 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
Al Viro909b0a82011-11-25 03:06:56 -0500665static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
Al Viro6b41d532011-11-24 23:24:33 -0500667 struct list_head *next = p->mnt_mounts.next;
668 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500670 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500672 next = p->mnt_child.next;
673 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 break;
Al Viro0714a532011-11-24 22:19:58 -0500675 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 }
677 }
Al Viro6b41d532011-11-24 23:24:33 -0500678 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679}
680
Al Viro315fc832011-11-24 18:57:30 -0500681static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500682{
Al Viro6b41d532011-11-24 23:24:33 -0500683 struct list_head *prev = p->mnt_mounts.prev;
684 while (prev != &p->mnt_mounts) {
685 p = list_entry(prev, struct mount, mnt_child);
686 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500687 }
688 return p;
689}
690
Al Viro9d412a42011-03-17 22:08:28 -0400691struct vfsmount *
692vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
693{
Al Virob105e272011-11-24 20:38:33 -0500694 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400695 struct dentry *root;
696
697 if (!type)
698 return ERR_PTR(-ENODEV);
699
700 mnt = alloc_vfsmnt(name);
701 if (!mnt)
702 return ERR_PTR(-ENOMEM);
703
704 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500705 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400706
707 root = mount_fs(type, flags, name, data);
708 if (IS_ERR(root)) {
709 free_vfsmnt(mnt);
710 return ERR_CAST(root);
711 }
712
Al Virob105e272011-11-24 20:38:33 -0500713 mnt->mnt.mnt_root = root;
714 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500715 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500716 mnt->mnt_parent = mnt;
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100717 br_write_lock(vfsmount_lock);
718 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
719 br_write_unlock(vfsmount_lock);
Al Virob105e272011-11-24 20:38:33 -0500720 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400721}
722EXPORT_SYMBOL_GPL(vfs_kern_mount);
723
Al Viro87129cc2011-11-24 21:24:27 -0500724static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500725 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
Al Viro87129cc2011-11-24 21:24:27 -0500727 struct super_block *sb = old->mnt.mnt_sb;
Al Viro52ba1622011-11-25 02:25:17 -0500728 struct mount *mnt = alloc_vfsmnt(old->mnt_devname);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
730 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100731 if (flag & (CL_SLAVE | CL_PRIVATE))
Al Viro15169fe2011-11-25 00:50:41 -0500732 mnt->mnt_group_id = 0; /* not a peer of original */
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100733 else
Al Viro15169fe2011-11-25 00:50:41 -0500734 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100735
Al Viro15169fe2011-11-25 00:50:41 -0500736 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
Al Virob105e272011-11-24 20:38:33 -0500737 int err = mnt_alloc_group_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100738 if (err)
739 goto out_free;
740 }
741
Al Viro87129cc2011-11-24 21:24:27 -0500742 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 atomic_inc(&sb->s_active);
Al Virob105e272011-11-24 20:38:33 -0500744 mnt->mnt.mnt_sb = sb;
745 mnt->mnt.mnt_root = dget(root);
Al Viroa73324d2011-11-24 22:25:07 -0500746 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500747 mnt->mnt_parent = mnt;
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100748 br_write_lock(vfsmount_lock);
749 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
750 br_write_unlock(vfsmount_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500751
Ram Pai5afe0022005-11-07 17:21:01 -0500752 if (flag & CL_SLAVE) {
Al Viro6776db3d2011-11-25 00:22:05 -0500753 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
Al Viro32301922011-11-25 00:10:28 -0500754 mnt->mnt_master = old;
Al Virofc7be132011-11-25 01:05:37 -0500755 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400756 } else if (!(flag & CL_PRIVATE)) {
Al Virofc7be132011-11-25 01:05:37 -0500757 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
Al Viro6776db3d2011-11-25 00:22:05 -0500758 list_add(&mnt->mnt_share, &old->mnt_share);
Al Virod10e8de2011-11-25 00:07:16 -0500759 if (IS_MNT_SLAVE(old))
Al Viro6776db3d2011-11-25 00:22:05 -0500760 list_add(&mnt->mnt_slave, &old->mnt_slave);
Al Virod10e8de2011-11-25 00:07:16 -0500761 mnt->mnt_master = old->mnt_master;
Ram Pai5afe0022005-11-07 17:21:01 -0500762 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500763 if (flag & CL_MAKE_SHARED)
Al Viro0f0afb12011-11-24 20:43:10 -0500764 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766 /* stick the duplicate mount on the same expiry list
767 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500768 if (flag & CL_EXPIRE) {
Al Viro6776db3d2011-11-25 00:22:05 -0500769 if (!list_empty(&old->mnt_expire))
770 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 }
Al Virocb338d02011-11-24 20:55:08 -0500773 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100774
775 out_free:
776 free_vfsmnt(mnt);
777 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
Al Viro83adc752011-11-24 22:37:54 -0500780static inline void mntfree(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781{
Al Viro83adc752011-11-24 22:37:54 -0500782 struct vfsmount *m = &mnt->mnt;
783 struct super_block *sb = m->mnt_sb;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100784
Dave Hansen3d733632008-02-15 14:37:59 -0800785 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800786 * This probably indicates that somebody messed
787 * up a mnt_want/drop_write() pair. If this
788 * happens, the filesystem was probably unable
789 * to make r/w->r/o transitions.
790 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000791 /*
Nick Pigginb3e19d92011-01-07 17:50:11 +1100792 * The locking used to deal with mnt_count decrement provides barriers,
793 * so mnt_get_writers() below is safe.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000794 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100795 WARN_ON(mnt_get_writers(mnt));
Al Viro83adc752011-11-24 22:37:54 -0500796 fsnotify_vfsmount_delete(m);
797 dput(m->mnt_root);
798 free_vfsmnt(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 deactivate_super(sb);
800}
801
Al Viro900148d2011-11-25 00:33:11 -0500802static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500803{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100804put_again:
Al Virof03c6592011-01-14 22:30:21 -0500805#ifdef CONFIG_SMP
806 br_read_lock(vfsmount_lock);
Al Viro68e8a9f2011-11-24 22:53:09 -0500807 if (likely(atomic_read(&mnt->mnt_longterm))) {
Al Viroaa9c0e02011-11-23 19:04:52 -0500808 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100809 br_read_unlock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500810 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100811 }
Al Virof03c6592011-01-14 22:30:21 -0500812 br_read_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100813
814 br_write_lock(vfsmount_lock);
Al Viroaa9c0e02011-11-23 19:04:52 -0500815 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100816 if (mnt_get_count(mnt)) {
817 br_write_unlock(vfsmount_lock);
818 return;
819 }
Nick Pigginb3e19d92011-01-07 17:50:11 +1100820#else
Al Viroaa9c0e02011-11-23 19:04:52 -0500821 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100822 if (likely(mnt_get_count(mnt)))
Nick Piggin99b7db72010-08-18 04:37:39 +1000823 return;
824 br_write_lock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500825#endif
Al Viro863d6842011-11-25 00:57:42 -0500826 if (unlikely(mnt->mnt_pinned)) {
827 mnt_add_count(mnt, mnt->mnt_pinned + 1);
828 mnt->mnt_pinned = 0;
Nick Piggin99b7db72010-08-18 04:37:39 +1000829 br_write_unlock(vfsmount_lock);
Al Viro900148d2011-11-25 00:33:11 -0500830 acct_auto_close_mnt(&mnt->mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100831 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500832 }
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100833 list_del(&mnt->mnt_instance);
Nick Piggin99b7db72010-08-18 04:37:39 +1000834 br_write_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100835 mntfree(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500836}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100837
838void mntput(struct vfsmount *mnt)
839{
840 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -0500841 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100842 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -0500843 if (unlikely(m->mnt_expiry_mark))
844 m->mnt_expiry_mark = 0;
845 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100846 }
847}
848EXPORT_SYMBOL(mntput);
849
850struct vfsmount *mntget(struct vfsmount *mnt)
851{
852 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -0500853 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100854 return mnt;
855}
856EXPORT_SYMBOL(mntget);
857
Al Viro7b7b1ac2005-11-07 17:13:39 -0500858void mnt_pin(struct vfsmount *mnt)
859{
Nick Piggin99b7db72010-08-18 04:37:39 +1000860 br_write_lock(vfsmount_lock);
Al Viro863d6842011-11-25 00:57:42 -0500861 real_mount(mnt)->mnt_pinned++;
Nick Piggin99b7db72010-08-18 04:37:39 +1000862 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500863}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500864EXPORT_SYMBOL(mnt_pin);
865
Al Viro863d6842011-11-25 00:57:42 -0500866void mnt_unpin(struct vfsmount *m)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500867{
Al Viro863d6842011-11-25 00:57:42 -0500868 struct mount *mnt = real_mount(m);
Nick Piggin99b7db72010-08-18 04:37:39 +1000869 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500870 if (mnt->mnt_pinned) {
Al Viro863d6842011-11-25 00:57:42 -0500871 mnt_add_count(mnt, 1);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500872 mnt->mnt_pinned--;
873 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000874 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500875}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500876EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800878static inline void mangle(struct seq_file *m, const char *s)
879{
880 seq_escape(m, s, " \t\n\\");
881}
882
883/*
884 * Simple .show_options callback for filesystems which don't want to
885 * implement more complex mount option showing.
886 *
887 * See also save_mount_options().
888 */
Al Viro34c80b12011-12-08 21:32:45 -0500889int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800890{
Al Viro2a32ceb2009-05-08 16:05:57 -0400891 const char *options;
892
893 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -0500894 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800895
896 if (options != NULL && options[0]) {
897 seq_putc(m, ',');
898 mangle(m, options);
899 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400900 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800901
902 return 0;
903}
904EXPORT_SYMBOL(generic_show_options);
905
906/*
907 * If filesystem uses generic_show_options(), this function should be
908 * called from the fill_super() callback.
909 *
910 * The .remount_fs callback usually needs to be handled in a special
911 * way, to make sure, that previous options are not overwritten if the
912 * remount fails.
913 *
914 * Also note, that if the filesystem's .remount_fs function doesn't
915 * reset all options to their default value, but changes only newly
916 * given options, then the displayed options will not reflect reality
917 * any more.
918 */
919void save_mount_options(struct super_block *sb, char *options)
920{
Al Viro2a32ceb2009-05-08 16:05:57 -0400921 BUG_ON(sb->s_options);
922 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800923}
924EXPORT_SYMBOL(save_mount_options);
925
Al Viro2a32ceb2009-05-08 16:05:57 -0400926void replace_mount_options(struct super_block *sb, char *options)
927{
928 char *old = sb->s_options;
929 rcu_assign_pointer(sb->s_options, options);
930 if (old) {
931 synchronize_rcu();
932 kfree(old);
933 }
934}
935EXPORT_SYMBOL(replace_mount_options);
936
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100937#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -0500938/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939static void *m_start(struct seq_file *m, loff_t *pos)
940{
Al Viro0226f492011-12-06 12:21:54 -0500941 struct proc_mounts *p = container_of(m, struct proc_mounts, m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Ram Pai390c6842005-11-07 17:17:51 -0500943 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100944 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945}
946
947static void *m_next(struct seq_file *m, void *v, loff_t *pos)
948{
Al Viro0226f492011-12-06 12:21:54 -0500949 struct proc_mounts *p = container_of(m, struct proc_mounts, m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700950
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100951 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952}
953
954static void m_stop(struct seq_file *m, void *v)
955{
Ram Pai390c6842005-11-07 17:17:51 -0500956 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
Al Viro0226f492011-12-06 12:21:54 -0500959static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -0500960{
Al Viro0226f492011-12-06 12:21:54 -0500961 struct proc_mounts *p = container_of(m, struct proc_mounts, m);
Al Viro1a4eeaf2011-11-25 02:19:55 -0500962 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -0500963 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964}
965
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100966const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 .start = m_start,
968 .next = m_next,
969 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -0500970 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -0500971};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100972#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974/**
975 * may_umount_tree - check if a mount tree is busy
976 * @mnt: root of mount tree
977 *
978 * This is called to check if a tree of mounts has any
979 * open files, pwds, chroots or sub mounts that are
980 * busy.
981 */
Al Viro909b0a82011-11-25 03:06:56 -0500982int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983{
Al Viro909b0a82011-11-25 03:06:56 -0500984 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -0500985 int actual_refs = 0;
986 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -0500987 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -0500988 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Nick Pigginb3e19d92011-01-07 17:50:11 +1100990 /* write lock needed for mnt_get_count */
991 br_write_lock(vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -0500992 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -0500993 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 }
Nick Pigginb3e19d92011-01-07 17:50:11 +1100996 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800999 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Ian Kente3474a82006-03-27 01:14:51 -08001001 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002}
1003
1004EXPORT_SYMBOL(may_umount_tree);
1005
1006/**
1007 * may_umount - check if a mount point is busy
1008 * @mnt: root of mount
1009 *
1010 * This is called to check if a mount point has any
1011 * open files, pwds, chroots or sub mounts. If the
1012 * mount has sub mounts this will return busy
1013 * regardless of whether the sub mounts are busy.
1014 *
1015 * Doesn't take quota and stuff into account. IOW, in some cases it will
1016 * give false negatives. The main reason why it's here is that we need
1017 * a non-destructive way to look for easily umountable filesystems.
1018 */
1019int may_umount(struct vfsmount *mnt)
1020{
Ian Kente3474a82006-03-27 01:14:51 -08001021 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001022 down_read(&namespace_sem);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001023 br_write_lock(vfsmount_lock);
Al Viro1ab59732011-11-24 21:35:16 -05001024 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001025 ret = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001026 br_write_unlock(vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001027 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001028 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029}
1030
1031EXPORT_SYMBOL(may_umount);
1032
Ram Paib90fa9a2005-11-07 17:19:50 -05001033void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
Al Virod5e50f72011-11-24 20:58:57 -05001035 struct mount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001036 while (!list_empty(head)) {
Al Viro1b8e5562011-11-24 21:01:32 -05001037 mnt = list_first_entry(head, struct mount, mnt_hash);
1038 list_del_init(&mnt->mnt_hash);
Al Viro676da582011-11-24 21:47:05 -05001039 if (mnt_has_parent(mnt)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001040 struct dentry *dentry;
Al Viro863d6842011-11-25 00:57:42 -05001041 struct mount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001042
1043 br_write_lock(vfsmount_lock);
Al Viroa73324d2011-11-24 22:25:07 -05001044 dentry = mnt->mnt_mountpoint;
Al Viro863d6842011-11-25 00:57:42 -05001045 m = mnt->mnt_parent;
Al Viroa73324d2011-11-24 22:25:07 -05001046 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -05001047 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001048 m->mnt_ghosts--;
Nick Piggin99b7db72010-08-18 04:37:39 +10001049 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001050 dput(dentry);
Al Viro863d6842011-11-25 00:57:42 -05001051 mntput(&m->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 }
Al Virod5e50f72011-11-24 20:58:57 -05001053 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001054 }
1055}
1056
Nick Piggin99b7db72010-08-18 04:37:39 +10001057/*
1058 * vfsmount lock must be held for write
1059 * namespace_sem must be held for write
1060 */
Al Viro761d5c32011-11-24 21:07:43 -05001061void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001062{
Al Viro7b8a53f2011-01-15 20:08:44 -05001063 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001064 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001065
Al Viro909b0a82011-11-25 03:06:56 -05001066 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro1b8e5562011-11-24 21:01:32 -05001067 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001068
Ram Paia05964f2005-11-07 17:20:17 -05001069 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001070 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001071
Al Viro1b8e5562011-11-24 21:01:32 -05001072 list_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001073 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001074 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001075 __touch_mnt_namespace(p->mnt_ns);
1076 p->mnt_ns = NULL;
Al Viro83adc752011-11-24 22:37:54 -05001077 __mnt_make_shortterm(p);
Al Viro6b41d532011-11-24 23:24:33 -05001078 list_del_init(&p->mnt_child);
Al Viro676da582011-11-24 21:47:05 -05001079 if (mnt_has_parent(p)) {
Al Viro863d6842011-11-25 00:57:42 -05001080 p->mnt_parent->mnt_ghosts++;
Al Viroa73324d2011-11-24 22:25:07 -05001081 dentry_reset_mounted(p->mnt_mountpoint);
Al Viro7c4b93d2008-03-21 23:59:49 -04001082 }
Al Viro0f0afb12011-11-24 20:43:10 -05001083 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 }
Al Viro7b8a53f2011-01-15 20:08:44 -05001085 list_splice(&tmp_list, kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086}
1087
Al Viro692afc32011-11-24 21:15:14 -05001088static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
Al Viroc35038b2008-03-22 00:46:23 -04001089
Al Viro1ab59732011-11-24 21:35:16 -05001090static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091{
Al Viro1ab59732011-11-24 21:35:16 -05001092 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001094 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Al Viro1ab59732011-11-24 21:35:16 -05001096 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 if (retval)
1098 return retval;
1099
1100 /*
1101 * Allow userspace to request a mountpoint be expired rather than
1102 * unmounting unconditionally. Unmount only happens if:
1103 * (1) the mark is already set (the mark is cleared by mntput())
1104 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1105 */
1106 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001107 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 flags & (MNT_FORCE | MNT_DETACH))
1109 return -EINVAL;
1110
Nick Pigginb3e19d92011-01-07 17:50:11 +11001111 /*
1112 * probably don't strictly need the lock here if we examined
1113 * all race cases, but it's a slowpath.
1114 */
1115 br_write_lock(vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -05001116 if (mnt_get_count(mnt) != 2) {
J. R. Okajimabf9faa22011-02-23 16:59:49 +09001117 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001119 }
1120 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
Al Viro863d6842011-11-25 00:57:42 -05001122 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 return -EAGAIN;
1124 }
1125
1126 /*
1127 * If we may have to abort operations to get out of this
1128 * mount, and they will themselves hold resources we must
1129 * allow the fs to do things. In the Unix tradition of
1130 * 'Gee thats tricky lets do it in userspace' the umount_begin
1131 * might fail to complete on the first run through as other tasks
1132 * must return, and the like. Thats for the mount program to worry
1133 * about for the moment.
1134 */
1135
Al Viro42faad92008-04-24 07:21:56 -04001136 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001137 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001138 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
1140 /*
1141 * No sense to grab the lock for this test, but test itself looks
1142 * somewhat bogus. Suggestions for better replacement?
1143 * Ho-hum... In principle, we might treat that as umount + switch
1144 * to rootfs. GC would eventually take care of the old vfsmount.
1145 * Actually it makes sense, especially if rootfs would contain a
1146 * /reboot - static binary that would close all descriptors and
1147 * call reboot(9). Then init(8) could umount root and exec /reboot.
1148 */
Al Viro1ab59732011-11-24 21:35:16 -05001149 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 /*
1151 * Special case for "unmounting" root ...
1152 * we just try to remount it readonly.
1153 */
1154 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001155 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 up_write(&sb->s_umount);
1158 return retval;
1159 }
1160
Ram Pai390c6842005-11-07 17:17:51 -05001161 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001162 br_write_lock(vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001163 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
Al Viroc35038b2008-03-22 00:46:23 -04001165 if (!(flags & MNT_DETACH))
Al Viro1ab59732011-11-24 21:35:16 -05001166 shrink_submounts(mnt, &umount_list);
Al Viroc35038b2008-03-22 00:46:23 -04001167
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001169 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001170 if (!list_empty(&mnt->mnt_list))
Al Viro1ab59732011-11-24 21:35:16 -05001171 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 retval = 0;
1173 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001174 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001175 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001176 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 return retval;
1178}
1179
1180/*
1181 * Now umount can handle mount points as well as block devices.
1182 * This is important for filesystems which use unnamed block devices.
1183 *
1184 * We now support a flag for forced unmount like the other 'big iron'
1185 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1186 */
1187
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001188SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
Al Viro2d8f3032008-07-22 09:59:21 -04001190 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001191 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001193 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001195 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1196 return -EINVAL;
1197
1198 if (!(flags & UMOUNT_NOFOLLOW))
1199 lookup_flags |= LOOKUP_FOLLOW;
1200
1201 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 if (retval)
1203 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001204 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001206 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001208 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 goto dput_and_out;
1210
1211 retval = -EPERM;
1212 if (!capable(CAP_SYS_ADMIN))
1213 goto dput_and_out;
1214
Al Viro900148d2011-11-25 00:33:11 -05001215 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001217 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001218 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001219 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220out:
1221 return retval;
1222}
1223
1224#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1225
1226/*
Ram Paib58fed82005-11-07 17:16:09 -05001227 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001229SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230{
Ram Paib58fed82005-11-07 17:16:09 -05001231 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232}
1233
1234#endif
1235
Al Viro2d92ab32008-08-02 00:51:11 -04001236static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237{
1238 if (capable(CAP_SYS_ADMIN))
1239 return 0;
1240 return -EPERM;
1241#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001242 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001244 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001245 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 return -EPERM;
1247 }
Al Viro2d92ab32008-08-02 00:51:11 -04001248 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 return -EPERM;
1250 return 0;
1251#endif
1252}
1253
Al Viro87129cc2011-11-24 21:24:27 -05001254struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001255 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256{
Al Viroa73324d2011-11-24 22:25:07 -05001257 struct mount *res, *p, *q, *r;
Al Viro1a390682008-03-21 20:48:19 -04001258 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
Al Virofc7be132011-11-25 01:05:37 -05001260 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001261 return NULL;
1262
Ram Pai36341f62005-11-07 17:17:22 -05001263 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 if (!q)
1265 goto Enomem;
Al Viroa73324d2011-11-24 22:25:07 -05001266 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
1268 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001269 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001270 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001271 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 continue;
1273
Al Viro909b0a82011-11-25 03:06:56 -05001274 for (s = r; s; s = next_mnt(s, r)) {
Al Virofc7be132011-11-25 01:05:37 -05001275 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001276 s = skip_mnt_tree(s);
1277 continue;
1278 }
Al Viro0714a532011-11-24 22:19:58 -05001279 while (p != s->mnt_parent) {
1280 p = p->mnt_parent;
1281 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 }
Al Viro87129cc2011-11-24 21:24:27 -05001283 p = s;
Al Virocb338d02011-11-24 20:55:08 -05001284 path.mnt = &q->mnt;
Al Viroa73324d2011-11-24 22:25:07 -05001285 path.dentry = p->mnt_mountpoint;
Al Viro87129cc2011-11-24 21:24:27 -05001286 q = clone_mnt(p, p->mnt.mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 if (!q)
1288 goto Enomem;
Nick Piggin99b7db72010-08-18 04:37:39 +10001289 br_write_lock(vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001290 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Virocb338d02011-11-24 20:55:08 -05001291 attach_mnt(q, &path);
Nick Piggin99b7db72010-08-18 04:37:39 +10001292 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 }
1294 }
1295 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001296Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001298 LIST_HEAD(umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001299 br_write_lock(vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001300 umount_tree(res, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001301 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001302 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 }
1304 return NULL;
1305}
1306
Al Viro589ff872009-04-18 03:28:19 -04001307struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001308{
Al Virocb338d02011-11-24 20:55:08 -05001309 struct mount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001310 down_write(&namespace_sem);
Al Viro87129cc2011-11-24 21:24:27 -05001311 tree = copy_tree(real_mount(path->mnt), path->dentry,
1312 CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001313 up_write(&namespace_sem);
Al Virocb338d02011-11-24 20:55:08 -05001314 return tree ? &tree->mnt : NULL;
Al Viro8aec0802007-06-07 12:20:32 -04001315}
1316
1317void drop_collected_mounts(struct vfsmount *mnt)
1318{
1319 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001320 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001321 br_write_lock(vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001322 umount_tree(real_mount(mnt), 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001323 br_write_unlock(vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001324 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001325 release_mounts(&umount_list);
1326}
1327
Al Viro1f707132010-01-30 22:51:25 -05001328int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1329 struct vfsmount *root)
1330{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001331 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001332 int res = f(root, arg);
1333 if (res)
1334 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001335 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1336 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001337 if (res)
1338 return res;
1339 }
1340 return 0;
1341}
1342
Al Viro4b8b21f2011-11-24 19:54:23 -05001343static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001344{
Al Viro315fc832011-11-24 18:57:30 -05001345 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001346
Al Viro909b0a82011-11-25 03:06:56 -05001347 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001348 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001349 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001350 }
1351}
1352
Al Viro4b8b21f2011-11-24 19:54:23 -05001353static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001354{
Al Viro315fc832011-11-24 18:57:30 -05001355 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001356
Al Viro909b0a82011-11-25 03:06:56 -05001357 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001358 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001359 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001360 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001361 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001362 return err;
1363 }
1364 }
1365 }
1366
1367 return 0;
1368}
1369
Ram Paib90fa9a2005-11-07 17:19:50 -05001370/*
1371 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001372 * @nd : place the mount tree @source_mnt is attached
1373 * @parent_nd : if non-null, detach the source_mnt from its parent and
1374 * store the parent mount and mountpoint dentry.
1375 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001376 *
1377 * NOTE: in the table below explains the semantics when a source mount
1378 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001379 * ---------------------------------------------------------------------------
1380 * | BIND MOUNT OPERATION |
1381 * |**************************************************************************
1382 * | source-->| shared | private | slave | unbindable |
1383 * | dest | | | | |
1384 * | | | | | | |
1385 * | v | | | | |
1386 * |**************************************************************************
1387 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1388 * | | | | | |
1389 * |non-shared| shared (+) | private | slave (*) | invalid |
1390 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001391 * A bind operation clones the source mount and mounts the clone on the
1392 * destination mount.
1393 *
1394 * (++) the cloned mount is propagated to all the mounts in the propagation
1395 * tree of the destination mount and the cloned mount is added to
1396 * the peer group of the source mount.
1397 * (+) the cloned mount is created under the destination mount and is marked
1398 * as shared. The cloned mount is added to the peer group of the source
1399 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001400 * (+++) the mount is propagated to all the mounts in the propagation tree
1401 * of the destination mount and the cloned mount is made slave
1402 * of the same master as that of the source mount. The cloned mount
1403 * is marked as 'shared and slave'.
1404 * (*) the cloned mount is made a slave of the same master as that of the
1405 * source mount.
1406 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001407 * ---------------------------------------------------------------------------
1408 * | MOVE MOUNT OPERATION |
1409 * |**************************************************************************
1410 * | source-->| shared | private | slave | unbindable |
1411 * | dest | | | | |
1412 * | | | | | | |
1413 * | v | | | | |
1414 * |**************************************************************************
1415 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1416 * | | | | | |
1417 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1418 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001419 *
1420 * (+) the mount is moved to the destination. And is then propagated to
1421 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001422 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001423 * (+++) the mount is moved to the destination and is then propagated to
1424 * all the mounts belonging to the destination mount's propagation tree.
1425 * the mount is marked as 'shared and slave'.
1426 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001427 *
1428 * if the source mount is a tree, the operations explained above is
1429 * applied to each mount in the tree.
1430 * Must be called without spinlocks held, since this function can sleep
1431 * in allocations.
1432 */
Al Viro0fb54e52011-11-24 19:59:16 -05001433static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001434 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001435{
1436 LIST_HEAD(tree_list);
Al Viroa8d56d82011-11-24 23:59:29 -05001437 struct mount *dest_mnt = real_mount(path->mnt);
Al Viro1a390682008-03-21 20:48:19 -04001438 struct dentry *dest_dentry = path->dentry;
Al Viro315fc832011-11-24 18:57:30 -05001439 struct mount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001440 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001441
Al Virofc7be132011-11-25 01:05:37 -05001442 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001443 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001444 if (err)
1445 goto out;
1446 }
Al Viroa8d56d82011-11-24 23:59:29 -05001447 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001448 if (err)
1449 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001450
Nick Piggin99b7db72010-08-18 04:37:39 +10001451 br_write_lock(vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001452
Al Virofc7be132011-11-25 01:05:37 -05001453 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro909b0a82011-11-25 03:06:56 -05001454 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001455 set_mnt_shared(p);
Ram Paib90fa9a2005-11-07 17:19:50 -05001456 }
Al Viro1a390682008-03-21 20:48:19 -04001457 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001458 detach_mnt(source_mnt, parent_path);
1459 attach_mnt(source_mnt, path);
Al Viro143c8c92011-11-25 00:46:35 -05001460 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001461 } else {
Al Viro14cf1fa2011-11-25 00:01:17 -05001462 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
Al Viro0fb54e52011-11-24 19:59:16 -05001463 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05001464 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001465
Al Viro1b8e5562011-11-24 21:01:32 -05001466 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1467 list_del_init(&child->mnt_hash);
Al Viro4b2619a2011-11-24 19:47:15 -05001468 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05001469 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001470 br_write_unlock(vfsmount_lock);
1471
Ram Paib90fa9a2005-11-07 17:19:50 -05001472 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001473
1474 out_cleanup_ids:
Al Virofc7be132011-11-25 01:05:37 -05001475 if (IS_MNT_SHARED(dest_mnt))
Al Viro0fb54e52011-11-24 19:59:16 -05001476 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001477 out:
1478 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001479}
1480
Al Virob12cea92011-03-18 08:55:38 -04001481static int lock_mount(struct path *path)
1482{
1483 struct vfsmount *mnt;
1484retry:
1485 mutex_lock(&path->dentry->d_inode->i_mutex);
1486 if (unlikely(cant_mount(path->dentry))) {
1487 mutex_unlock(&path->dentry->d_inode->i_mutex);
1488 return -ENOENT;
1489 }
1490 down_write(&namespace_sem);
1491 mnt = lookup_mnt(path);
1492 if (likely(!mnt))
1493 return 0;
1494 up_write(&namespace_sem);
1495 mutex_unlock(&path->dentry->d_inode->i_mutex);
1496 path_put(path);
1497 path->mnt = mnt;
1498 path->dentry = dget(mnt->mnt_root);
1499 goto retry;
1500}
1501
1502static void unlock_mount(struct path *path)
1503{
1504 up_write(&namespace_sem);
1505 mutex_unlock(&path->dentry->d_inode->i_mutex);
1506}
1507
Al Viro95bc5f22011-11-25 00:30:56 -05001508static int graft_tree(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509{
Al Viro95bc5f22011-11-25 00:30:56 -05001510 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 return -EINVAL;
1512
Al Viro8c3ee422008-03-22 18:00:39 -04001513 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001514 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 return -ENOTDIR;
1516
Al Virob12cea92011-03-18 08:55:38 -04001517 if (d_unlinked(path->dentry))
1518 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
Al Viro95bc5f22011-11-25 00:30:56 -05001520 return attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521}
1522
1523/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001524 * Sanity check the flags to change_mnt_propagation.
1525 */
1526
1527static int flags_to_propagation_type(int flags)
1528{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001529 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001530
1531 /* Fail if any non-propagation flags are set */
1532 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1533 return 0;
1534 /* Only one propagation flag should be set */
1535 if (!is_power_of_2(type))
1536 return 0;
1537 return type;
1538}
1539
1540/*
Ram Pai07b20882005-11-07 17:19:07 -05001541 * recursively change the type of the mountpoint.
1542 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001543static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001544{
Al Viro315fc832011-11-24 18:57:30 -05001545 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001546 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001547 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001548 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001549 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001550
Miklos Szerediee6f9582007-05-08 00:30:40 -07001551 if (!capable(CAP_SYS_ADMIN))
1552 return -EPERM;
1553
Al Viro2d92ab32008-08-02 00:51:11 -04001554 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001555 return -EINVAL;
1556
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001557 type = flags_to_propagation_type(flag);
1558 if (!type)
1559 return -EINVAL;
1560
Ram Pai07b20882005-11-07 17:19:07 -05001561 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001562 if (type == MS_SHARED) {
1563 err = invent_group_ids(mnt, recurse);
1564 if (err)
1565 goto out_unlock;
1566 }
1567
Nick Piggin99b7db72010-08-18 04:37:39 +10001568 br_write_lock(vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -05001569 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001570 change_mnt_propagation(m, type);
Nick Piggin99b7db72010-08-18 04:37:39 +10001571 br_write_unlock(vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001572
1573 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001574 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001575 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001576}
1577
1578/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 * do loopback mount.
1580 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001581static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001582 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583{
Al Virob12cea92011-03-18 08:55:38 -04001584 LIST_HEAD(umount_list);
Al Viro2d92ab32008-08-02 00:51:11 -04001585 struct path old_path;
Al Viro87129cc2011-11-24 21:24:27 -05001586 struct mount *mnt = NULL, *old;
Al Viro2d92ab32008-08-02 00:51:11 -04001587 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 if (err)
1589 return err;
1590 if (!old_name || !*old_name)
1591 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001592 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 if (err)
1594 return err;
1595
Al Virob12cea92011-03-18 08:55:38 -04001596 err = lock_mount(path);
1597 if (err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001598 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001599
Al Viro87129cc2011-11-24 21:24:27 -05001600 old = real_mount(old_path.mnt);
1601
Al Virob12cea92011-03-18 08:55:38 -04001602 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001603 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04001604 goto out2;
1605
Al Viro143c8c92011-11-25 00:46:35 -05001606 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
Al Virob12cea92011-03-18 08:55:38 -04001607 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
Al Viroccd48bc2005-11-07 17:15:04 -05001609 err = -ENOMEM;
1610 if (recurse)
Al Viro87129cc2011-11-24 21:24:27 -05001611 mnt = copy_tree(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001612 else
Al Viro87129cc2011-11-24 21:24:27 -05001613 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001614
1615 if (!mnt)
Al Virob12cea92011-03-18 08:55:38 -04001616 goto out2;
Al Viroccd48bc2005-11-07 17:15:04 -05001617
Al Viro95bc5f22011-11-25 00:30:56 -05001618 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001619 if (err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001620 br_write_lock(vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001621 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001622 br_write_unlock(vfsmount_lock);
Ram Pai5b83d2c2005-11-07 17:16:29 -05001623 }
Al Virob12cea92011-03-18 08:55:38 -04001624out2:
1625 unlock_mount(path);
1626 release_mounts(&umount_list);
Al Viroccd48bc2005-11-07 17:15:04 -05001627out:
Al Viro2d92ab32008-08-02 00:51:11 -04001628 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 return err;
1630}
1631
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001632static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1633{
1634 int error = 0;
1635 int readonly_request = 0;
1636
1637 if (ms_flags & MS_RDONLY)
1638 readonly_request = 1;
1639 if (readonly_request == __mnt_is_readonly(mnt))
1640 return 0;
1641
1642 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05001643 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001644 else
Al Viro83adc752011-11-24 22:37:54 -05001645 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001646 return error;
1647}
1648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649/*
1650 * change filesystem flags. dir should be a physical root of filesystem.
1651 * If you've mounted a non-root directory somewhere and want to do remount
1652 * on it - tough luck.
1653 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001654static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 void *data)
1656{
1657 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001658 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05001659 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
1661 if (!capable(CAP_SYS_ADMIN))
1662 return -EPERM;
1663
Al Viro143c8c92011-11-25 00:46:35 -05001664 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 return -EINVAL;
1666
Al Viro2d92ab32008-08-02 00:51:11 -04001667 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 return -EINVAL;
1669
Eric Parisff36fe22011-03-03 16:09:14 -05001670 err = security_sb_remount(sb, data);
1671 if (err)
1672 return err;
1673
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001675 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001676 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001677 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001678 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001679 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001680 br_write_lock(vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001681 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1682 mnt->mnt.mnt_flags = mnt_flags;
Nick Piggin99b7db72010-08-18 04:37:39 +10001683 br_write_unlock(vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001686 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001687 br_write_lock(vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001688 touch_mnt_namespace(mnt->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10001689 br_write_unlock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001690 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 return err;
1692}
1693
Al Virocbbe3622011-11-24 20:01:19 -05001694static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05001695{
Al Viro315fc832011-11-24 18:57:30 -05001696 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001697 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001698 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05001699 return 1;
1700 }
1701 return 0;
1702}
1703
Al Viro0a0d8a42008-08-02 00:55:27 -04001704static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705{
Al Viro2d92ab32008-08-02 00:51:11 -04001706 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05001707 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05001708 struct mount *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 int err = 0;
1710 if (!capable(CAP_SYS_ADMIN))
1711 return -EPERM;
1712 if (!old_name || !*old_name)
1713 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001714 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 if (err)
1716 return err;
1717
Al Virob12cea92011-03-18 08:55:38 -04001718 err = lock_mount(path);
David Howellscc53ce52011-01-14 18:45:26 +00001719 if (err < 0)
1720 goto out;
1721
Al Viro143c8c92011-11-25 00:46:35 -05001722 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05001723 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05001724
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001726 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 goto out1;
1728
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001729 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001730 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
1732 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001733 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001734 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
Al Viro676da582011-11-24 21:47:05 -05001736 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05001737 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738
Al Viro2d92ab32008-08-02 00:51:11 -04001739 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1740 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001741 goto out1;
1742 /*
1743 * Don't move a mount residing in a shared parent.
1744 */
Al Virofc7be132011-11-25 01:05:37 -05001745 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001746 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001747 /*
1748 * Don't move a mount tree containing unbindable mounts to a destination
1749 * mount which is shared.
1750 */
Al Virofc7be132011-11-25 01:05:37 -05001751 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05001752 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05001754 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05001755 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05001756 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
Al Viro0fb54e52011-11-24 19:59:16 -05001758 err = attach_recursive_mnt(old, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001759 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001760 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761
1762 /* if the mount is moved, it should no longer be expire
1763 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05001764 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765out1:
Al Virob12cea92011-03-18 08:55:38 -04001766 unlock_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001769 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001770 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 return err;
1772}
1773
Al Viro9d412a42011-03-17 22:08:28 -04001774static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1775{
1776 int err;
1777 const char *subtype = strchr(fstype, '.');
1778 if (subtype) {
1779 subtype++;
1780 err = -EINVAL;
1781 if (!subtype[0])
1782 goto err;
1783 } else
1784 subtype = "";
1785
1786 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1787 err = -ENOMEM;
1788 if (!mnt->mnt_sb->s_subtype)
1789 goto err;
1790 return mnt;
1791
1792 err:
1793 mntput(mnt);
1794 return ERR_PTR(err);
1795}
1796
Al Viro79e801a2011-12-12 23:07:05 -05001797static struct vfsmount *
Al Viro9d412a42011-03-17 22:08:28 -04001798do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1799{
1800 struct file_system_type *type = get_fs_type(fstype);
1801 struct vfsmount *mnt;
1802 if (!type)
1803 return ERR_PTR(-ENODEV);
1804 mnt = vfs_kern_mount(type, flags, name, data);
1805 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1806 !mnt->mnt_sb->s_subtype)
1807 mnt = fs_set_subtype(mnt, fstype);
1808 put_filesystem(type);
1809 return mnt;
1810}
Al Viro9d412a42011-03-17 22:08:28 -04001811
1812/*
1813 * add a mount into a namespace's mount tree
1814 */
Al Viro95bc5f22011-11-25 00:30:56 -05001815static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04001816{
1817 int err;
1818
1819 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1820
Al Virob12cea92011-03-18 08:55:38 -04001821 err = lock_mount(path);
1822 if (err)
1823 return err;
Al Viro9d412a42011-03-17 22:08:28 -04001824
1825 err = -EINVAL;
Al Viro143c8c92011-11-25 00:46:35 -05001826 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(real_mount(path->mnt)))
Al Viro9d412a42011-03-17 22:08:28 -04001827 goto unlock;
1828
1829 /* Refuse the same filesystem on the same mount point */
1830 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05001831 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04001832 path->mnt->mnt_root == path->dentry)
1833 goto unlock;
1834
1835 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05001836 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04001837 goto unlock;
1838
Al Viro95bc5f22011-11-25 00:30:56 -05001839 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro9d412a42011-03-17 22:08:28 -04001840 err = graft_tree(newmnt, path);
1841
1842unlock:
Al Virob12cea92011-03-18 08:55:38 -04001843 unlock_mount(path);
Al Viro9d412a42011-03-17 22:08:28 -04001844 return err;
1845}
Al Virob1e75df2011-01-17 01:47:59 -05001846
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847/*
1848 * create a new mount for userspace and request it to be added into the
1849 * namespace's tree
1850 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001851static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 int mnt_flags, char *name, void *data)
1853{
1854 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05001855 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
Vegard Nossumeca6f532009-09-18 13:05:45 -07001857 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 return -EINVAL;
1859
1860 /* we need capabilities... */
1861 if (!capable(CAP_SYS_ADMIN))
1862 return -EPERM;
1863
1864 mnt = do_kern_mount(type, flags, name, data);
1865 if (IS_ERR(mnt))
1866 return PTR_ERR(mnt);
1867
Al Viro95bc5f22011-11-25 00:30:56 -05001868 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05001869 if (err)
1870 mntput(mnt);
1871 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872}
1873
Al Viro19a167a2011-01-17 01:35:23 -05001874int finish_automount(struct vfsmount *m, struct path *path)
1875{
Al Viro6776db3d2011-11-25 00:22:05 -05001876 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05001877 int err;
1878 /* The new mount record should have at least 2 refs to prevent it being
1879 * expired before we get a chance to add it
1880 */
Al Viro6776db3d2011-11-25 00:22:05 -05001881 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05001882
1883 if (m->mnt_sb == path->mnt->mnt_sb &&
1884 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05001885 err = -ELOOP;
1886 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05001887 }
1888
Al Viro95bc5f22011-11-25 00:30:56 -05001889 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05001890 if (!err)
1891 return 0;
1892fail:
1893 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05001894 if (!list_empty(&mnt->mnt_expire)) {
Al Virob1e75df2011-01-17 01:47:59 -05001895 down_write(&namespace_sem);
1896 br_write_lock(vfsmount_lock);
Al Viro6776db3d2011-11-25 00:22:05 -05001897 list_del_init(&mnt->mnt_expire);
Al Virob1e75df2011-01-17 01:47:59 -05001898 br_write_unlock(vfsmount_lock);
1899 up_write(&namespace_sem);
Al Viro19a167a2011-01-17 01:35:23 -05001900 }
Al Virob1e75df2011-01-17 01:47:59 -05001901 mntput(m);
1902 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05001903 return err;
1904}
1905
David Howellsea5b7782011-01-14 19:10:03 +00001906/**
1907 * mnt_set_expiry - Put a mount on an expiration list
1908 * @mnt: The mount to list.
1909 * @expiry_list: The list to add the mount to.
1910 */
1911void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1912{
1913 down_write(&namespace_sem);
1914 br_write_lock(vfsmount_lock);
1915
Al Viro6776db3d2011-11-25 00:22:05 -05001916 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00001917
1918 br_write_unlock(vfsmount_lock);
1919 up_write(&namespace_sem);
1920}
1921EXPORT_SYMBOL(mnt_set_expiry);
1922
1923/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 * process a list of expirable mountpoints with the intent of discarding any
1925 * mountpoints that aren't in use and haven't been touched since last we came
1926 * here
1927 */
1928void mark_mounts_for_expiry(struct list_head *mounts)
1929{
Al Viro761d5c32011-11-24 21:07:43 -05001930 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001932 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933
1934 if (list_empty(mounts))
1935 return;
1936
Al Virobcc5c7d2008-03-22 00:21:53 -04001937 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001938 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939
1940 /* extract from the expiration list every vfsmount that matches the
1941 * following criteria:
1942 * - only referenced by its parent vfsmount
1943 * - still marked for expiry (marked on the last call here; marks are
1944 * cleared by mntput())
1945 */
Al Viro6776db3d2011-11-25 00:22:05 -05001946 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05001947 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05001948 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05001950 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001952 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05001953 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05001954 touch_mnt_namespace(mnt->mnt_ns);
Al Virobcc5c7d2008-03-22 00:21:53 -04001955 umount_tree(mnt, 1, &umounts);
1956 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001957 br_write_unlock(vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001958 up_write(&namespace_sem);
1959
1960 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961}
1962
1963EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1964
1965/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001966 * Ripoff of 'select_parent()'
1967 *
1968 * search the list of submounts for a given mountpoint, and move any
1969 * shrinkable submounts to the 'graveyard' list.
1970 */
Al Viro692afc32011-11-24 21:15:14 -05001971static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001972{
Al Viro692afc32011-11-24 21:15:14 -05001973 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001974 struct list_head *next;
1975 int found = 0;
1976
1977repeat:
Al Viro6b41d532011-11-24 23:24:33 -05001978 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001979resume:
Al Viro6b41d532011-11-24 23:24:33 -05001980 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001981 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05001982 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001983
1984 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05001985 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04001986 continue;
1987 /*
1988 * Descend a level if the d_mounts list is non-empty.
1989 */
Al Viro6b41d532011-11-24 23:24:33 -05001990 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001991 this_parent = mnt;
1992 goto repeat;
1993 }
1994
Al Viro1ab59732011-11-24 21:35:16 -05001995 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05001996 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001997 found++;
1998 }
1999 }
2000 /*
2001 * All done at this level ... ascend and resume the search
2002 */
2003 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002004 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002005 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002006 goto resume;
2007 }
2008 return found;
2009}
2010
2011/*
2012 * process a list of expirable mountpoints with the intent of discarding any
2013 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002014 *
2015 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002016 */
Al Viro692afc32011-11-24 21:15:14 -05002017static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002018{
2019 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002020 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002021
Trond Myklebust5528f9112006-06-09 09:34:17 -04002022 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002023 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002024 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002025 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002026 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002027 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08002028 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04002029 }
2030 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002031}
2032
Trond Myklebust5528f9112006-06-09 09:34:17 -04002033/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 * Some copy_from_user() implementations do not return the exact number of
2035 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2036 * Note that this function differs from copy_from_user() in that it will oops
2037 * on bad values of `to', rather than returning a short copy.
2038 */
Ram Paib58fed82005-11-07 17:16:09 -05002039static long exact_copy_from_user(void *to, const void __user * from,
2040 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041{
2042 char *t = to;
2043 const char __user *f = from;
2044 char c;
2045
2046 if (!access_ok(VERIFY_READ, from, n))
2047 return n;
2048
2049 while (n) {
2050 if (__get_user(c, f)) {
2051 memset(t, 0, n);
2052 break;
2053 }
2054 *t++ = c;
2055 f++;
2056 n--;
2057 }
2058 return n;
2059}
2060
Ram Paib58fed82005-11-07 17:16:09 -05002061int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062{
2063 int i;
2064 unsigned long page;
2065 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002066
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 *where = 0;
2068 if (!data)
2069 return 0;
2070
2071 if (!(page = __get_free_page(GFP_KERNEL)))
2072 return -ENOMEM;
2073
2074 /* We only care that *some* data at the address the user
2075 * gave us is valid. Just in case, we'll zero
2076 * the remainder of the page.
2077 */
2078 /* copy_from_user cannot cross TASK_SIZE ! */
2079 size = TASK_SIZE - (unsigned long)data;
2080 if (size > PAGE_SIZE)
2081 size = PAGE_SIZE;
2082
2083 i = size - exact_copy_from_user((void *)page, data, size);
2084 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002085 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 return -EFAULT;
2087 }
2088 if (i != PAGE_SIZE)
2089 memset((char *)page + i, 0, PAGE_SIZE - i);
2090 *where = page;
2091 return 0;
2092}
2093
Vegard Nossumeca6f532009-09-18 13:05:45 -07002094int copy_mount_string(const void __user *data, char **where)
2095{
2096 char *tmp;
2097
2098 if (!data) {
2099 *where = NULL;
2100 return 0;
2101 }
2102
2103 tmp = strndup_user(data, PAGE_SIZE);
2104 if (IS_ERR(tmp))
2105 return PTR_ERR(tmp);
2106
2107 *where = tmp;
2108 return 0;
2109}
2110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111/*
2112 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2113 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2114 *
2115 * data is a (void *) that can point to any structure up to
2116 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2117 * information (or be NULL).
2118 *
2119 * Pre-0.97 versions of mount() didn't have a flags word.
2120 * When the flags word was introduced its top half was required
2121 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2122 * Therefore, if this magic number is present, it carries no information
2123 * and must be discarded.
2124 */
Ram Paib58fed82005-11-07 17:16:09 -05002125long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 unsigned long flags, void *data_page)
2127{
Al Viro2d92ab32008-08-02 00:51:11 -04002128 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 int retval = 0;
2130 int mnt_flags = 0;
2131
2132 /* Discard magic */
2133 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2134 flags &= ~MS_MGC_MSK;
2135
2136 /* Basic sanity checks */
2137
2138 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2139 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
2141 if (data_page)
2142 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2143
Tetsuo Handaa27ab9f2009-10-04 21:49:49 +09002144 /* ... and get the mountpoint */
2145 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2146 if (retval)
2147 return retval;
2148
2149 retval = security_sb_mount(dev_name, &path,
2150 type_page, flags, data_page);
2151 if (retval)
2152 goto dput_out;
2153
Andi Kleen613cbe32009-04-19 18:40:43 +02002154 /* Default to relatime unless overriden */
2155 if (!(flags & MS_NOATIME))
2156 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002157
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 /* Separate the per-mountpoint flags */
2159 if (flags & MS_NOSUID)
2160 mnt_flags |= MNT_NOSUID;
2161 if (flags & MS_NODEV)
2162 mnt_flags |= MNT_NODEV;
2163 if (flags & MS_NOEXEC)
2164 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002165 if (flags & MS_NOATIME)
2166 mnt_flags |= MNT_NOATIME;
2167 if (flags & MS_NODIRATIME)
2168 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002169 if (flags & MS_STRICTATIME)
2170 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002171 if (flags & MS_RDONLY)
2172 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002173
Al Viro7a4dec52010-08-09 12:05:43 -04002174 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002175 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2176 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002179 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 data_page);
2181 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002182 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002183 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002184 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002186 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 else
Al Viro2d92ab32008-08-02 00:51:11 -04002188 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 dev_name, data_page);
2190dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002191 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 return retval;
2193}
2194
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002195static struct mnt_namespace *alloc_mnt_ns(void)
2196{
2197 struct mnt_namespace *new_ns;
2198
2199 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2200 if (!new_ns)
2201 return ERR_PTR(-ENOMEM);
2202 atomic_set(&new_ns->count, 1);
2203 new_ns->root = NULL;
2204 INIT_LIST_HEAD(&new_ns->list);
2205 init_waitqueue_head(&new_ns->poll);
2206 new_ns->event = 0;
2207 return new_ns;
2208}
2209
Al Virof03c6592011-01-14 22:30:21 -05002210void mnt_make_longterm(struct vfsmount *mnt)
2211{
Al Viro83adc752011-11-24 22:37:54 -05002212 __mnt_make_longterm(real_mount(mnt));
Al Virof03c6592011-01-14 22:30:21 -05002213}
2214
Al Viro83adc752011-11-24 22:37:54 -05002215void mnt_make_shortterm(struct vfsmount *m)
Al Virof03c6592011-01-14 22:30:21 -05002216{
Al Viro7e3d0eb2011-01-16 16:32:11 -05002217#ifdef CONFIG_SMP
Al Viro83adc752011-11-24 22:37:54 -05002218 struct mount *mnt = real_mount(m);
Al Viro68e8a9f2011-11-24 22:53:09 -05002219 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
Al Virof03c6592011-01-14 22:30:21 -05002220 return;
2221 br_write_lock(vfsmount_lock);
Al Viro68e8a9f2011-11-24 22:53:09 -05002222 atomic_dec(&mnt->mnt_longterm);
Al Virof03c6592011-01-14 22:30:21 -05002223 br_write_unlock(vfsmount_lock);
Al Viro7e3d0eb2011-01-16 16:32:11 -05002224#endif
Al Virof03c6592011-01-14 22:30:21 -05002225}
2226
JANAK DESAI741a2952006-02-07 12:59:00 -08002227/*
2228 * Allocate a new namespace structure and populate it with contents
2229 * copied from the namespace of the passed in task structure.
2230 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002231static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002232 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002234 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002235 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002236 struct mount *p, *q;
Al Virobe08d6d2011-12-06 13:32:36 -05002237 struct mount *old = mnt_ns->root;
Al Virocb338d02011-11-24 20:55:08 -05002238 struct mount *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002240 new_ns = alloc_mnt_ns();
2241 if (IS_ERR(new_ns))
2242 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243
Ram Pai390c6842005-11-07 17:17:51 -05002244 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 /* First pass: copy the tree topology */
Al Viro909b0a82011-11-25 03:06:56 -05002246 new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE);
Al Virocb338d02011-11-24 20:55:08 -05002247 if (!new) {
Ram Pai390c6842005-11-07 17:17:51 -05002248 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002250 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 }
Al Virobe08d6d2011-12-06 13:32:36 -05002252 new_ns->root = new;
Nick Piggin99b7db72010-08-18 04:37:39 +10002253 br_write_lock(vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05002254 list_add_tail(&new_ns->list, &new->mnt_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002255 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256
2257 /*
2258 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2259 * as belonging to new namespace. We have already acquired a private
2260 * fs_struct, so tsk->fs->lock is not needed.
2261 */
Al Viro909b0a82011-11-25 03:06:56 -05002262 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002263 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002265 q->mnt_ns = new_ns;
Al Viro83adc752011-11-24 22:37:54 -05002266 __mnt_make_longterm(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 if (fs) {
Al Viro315fc832011-11-24 18:57:30 -05002268 if (&p->mnt == fs->root.mnt) {
2269 fs->root.mnt = mntget(&q->mnt);
Al Viro83adc752011-11-24 22:37:54 -05002270 __mnt_make_longterm(q);
Al Viro315fc832011-11-24 18:57:30 -05002271 mnt_make_shortterm(&p->mnt);
2272 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 }
Al Viro315fc832011-11-24 18:57:30 -05002274 if (&p->mnt == fs->pwd.mnt) {
2275 fs->pwd.mnt = mntget(&q->mnt);
Al Viro83adc752011-11-24 22:37:54 -05002276 __mnt_make_longterm(q);
Al Viro315fc832011-11-24 18:57:30 -05002277 mnt_make_shortterm(&p->mnt);
2278 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 }
Al Viro909b0a82011-11-25 03:06:56 -05002281 p = next_mnt(p, old);
2282 q = next_mnt(q, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 }
Ram Pai390c6842005-11-07 17:17:51 -05002284 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002287 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002289 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
JANAK DESAI741a2952006-02-07 12:59:00 -08002291 return new_ns;
2292}
2293
Eric W. Biederman213dd262007-07-15 23:41:15 -07002294struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002295 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002296{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002297 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002298
Badari Pulavartye3222c42007-05-08 00:25:21 -07002299 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002300 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002301
2302 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002303 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002304
Badari Pulavartye3222c42007-05-08 00:25:21 -07002305 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002306
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002307 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002308 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309}
2310
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002311/**
2312 * create_mnt_ns - creates a private namespace and adds a root filesystem
2313 * @mnt: pointer to the new root filesystem mountpoint
2314 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002315static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002316{
Al Viro1a4eeaf2011-11-25 02:19:55 -05002317 struct mnt_namespace *new_ns = alloc_mnt_ns();
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002318 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002319 struct mount *mnt = real_mount(m);
2320 mnt->mnt_ns = new_ns;
2321 __mnt_make_longterm(mnt);
Al Virobe08d6d2011-12-06 13:32:36 -05002322 new_ns->root = mnt;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002323 list_add(&new_ns->list, &mnt->mnt_list);
Al Viroc1334492011-11-16 16:12:14 -05002324 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002325 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002326 }
2327 return new_ns;
2328}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002329
Al Viroea441d12011-11-16 21:43:59 -05002330struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2331{
2332 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002333 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002334 struct path path;
2335 int err;
2336
2337 ns = create_mnt_ns(mnt);
2338 if (IS_ERR(ns))
2339 return ERR_CAST(ns);
2340
2341 err = vfs_path_lookup(mnt->mnt_root, mnt,
2342 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2343
2344 put_mnt_ns(ns);
2345
2346 if (err)
2347 return ERR_PTR(err);
2348
2349 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002350 s = path.mnt->mnt_sb;
2351 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002352 mntput(path.mnt);
2353 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002354 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002355 /* ... and return the root of (sub)tree on it */
2356 return path.dentry;
2357}
2358EXPORT_SYMBOL(mount_subtree);
2359
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002360SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2361 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002363 int ret;
2364 char *kernel_type;
2365 char *kernel_dir;
2366 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368
Vegard Nossumeca6f532009-09-18 13:05:45 -07002369 ret = copy_mount_string(type, &kernel_type);
2370 if (ret < 0)
2371 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
Vegard Nossumeca6f532009-09-18 13:05:45 -07002373 kernel_dir = getname(dir_name);
2374 if (IS_ERR(kernel_dir)) {
2375 ret = PTR_ERR(kernel_dir);
2376 goto out_dir;
2377 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378
Vegard Nossumeca6f532009-09-18 13:05:45 -07002379 ret = copy_mount_string(dev_name, &kernel_dev);
2380 if (ret < 0)
2381 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
Vegard Nossumeca6f532009-09-18 13:05:45 -07002383 ret = copy_mount_options(data, &data_page);
2384 if (ret < 0)
2385 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386
Vegard Nossumeca6f532009-09-18 13:05:45 -07002387 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2388 (void *) data_page);
2389
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002391out_data:
2392 kfree(kernel_dev);
2393out_dev:
2394 putname(kernel_dir);
2395out_dir:
2396 kfree(kernel_type);
2397out_type:
2398 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399}
2400
2401/*
Al Viroafac7cb2011-11-23 19:34:49 -05002402 * Return true if path is reachable from root
2403 *
2404 * namespace_sem or vfsmount_lock is held
2405 */
Al Viro643822b2011-11-24 22:00:28 -05002406bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002407 const struct path *root)
2408{
Al Viro643822b2011-11-24 22:00:28 -05002409 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002410 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002411 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002412 }
Al Viro643822b2011-11-24 22:00:28 -05002413 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002414}
2415
2416int path_is_under(struct path *path1, struct path *path2)
2417{
2418 int res;
2419 br_read_lock(vfsmount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002420 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viroafac7cb2011-11-23 19:34:49 -05002421 br_read_unlock(vfsmount_lock);
2422 return res;
2423}
2424EXPORT_SYMBOL(path_is_under);
2425
2426/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 * pivot_root Semantics:
2428 * Moves the root file system of the current process to the directory put_old,
2429 * makes new_root as the new root file system of the current process, and sets
2430 * root/cwd of all processes which had them on the current root to new_root.
2431 *
2432 * Restrictions:
2433 * The new_root and put_old must be directories, and must not be on the
2434 * same file system as the current process root. The put_old must be
2435 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2436 * pointed to by put_old must yield the same directory as new_root. No other
2437 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2438 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002439 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2440 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2441 * in this situation.
2442 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 * Notes:
2444 * - we don't move root/cwd if they are not at the root (reason: if something
2445 * cared enough to change them, it's probably wrong to force them elsewhere)
2446 * - it's okay to pick a root that isn't the root of a file system, e.g.
2447 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2448 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2449 * first.
2450 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002451SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2452 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453{
Al Viro2d8f3032008-07-22 09:59:21 -04002454 struct path new, old, parent_path, root_parent, root;
Al Viro419148d2011-11-24 19:41:16 -05002455 struct mount *new_mnt, *root_mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 int error;
2457
2458 if (!capable(CAP_SYS_ADMIN))
2459 return -EPERM;
2460
Al Viro2d8f3032008-07-22 09:59:21 -04002461 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 if (error)
2463 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464
Al Viro2d8f3032008-07-22 09:59:21 -04002465 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 if (error)
2467 goto out1;
2468
Al Viro2d8f3032008-07-22 09:59:21 -04002469 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002470 if (error)
2471 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002473 get_fs_root(current->fs, &root);
Al Virob12cea92011-03-18 08:55:38 -04002474 error = lock_mount(&old);
2475 if (error)
2476 goto out3;
2477
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002479 new_mnt = real_mount(new.mnt);
2480 root_mnt = real_mount(root.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002481 if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2482 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2483 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002484 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002485 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002486 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002488 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002489 goto out4;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002490 if (d_unlinked(old.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002491 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002493 if (new.mnt == root.mnt ||
2494 old.mnt == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002495 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002497 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002498 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002499 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002500 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002501 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002502 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002503 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002504 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002505 /* make sure we can reach put_old from new_root */
Al Viro643822b2011-11-24 22:00:28 -05002506 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002507 goto out4;
Al Viro27cb1572011-03-18 08:29:36 -04002508 br_write_lock(vfsmount_lock);
Al Viro419148d2011-11-24 19:41:16 -05002509 detach_mnt(new_mnt, &parent_path);
2510 detach_mnt(root_mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002511 /* mount old root on put_old */
Al Viro419148d2011-11-24 19:41:16 -05002512 attach_mnt(root_mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002513 /* mount new_root on / */
Al Viro419148d2011-11-24 19:41:16 -05002514 attach_mnt(new_mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002515 touch_mnt_namespace(current->nsproxy->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10002516 br_write_unlock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002517 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002519out4:
2520 unlock_mount(&old);
2521 if (!error) {
2522 path_put(&root_parent);
2523 path_put(&parent_path);
2524 }
2525out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002526 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002527out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002528 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002530 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533}
2534
2535static void __init init_mount_tree(void)
2536{
2537 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002538 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002539 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540
2541 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2542 if (IS_ERR(mnt))
2543 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002544
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002545 ns = create_mnt_ns(mnt);
2546 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002549 init_task.nsproxy->mnt_ns = ns;
2550 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551
Al Virobe08d6d2011-12-06 13:32:36 -05002552 root.mnt = mnt;
2553 root.dentry = mnt->mnt_root;
Jan Blunckac748a02008-02-14 19:34:39 -08002554
2555 set_fs_pwd(current->fs, &root);
2556 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557}
2558
Denis Cheng74bf17c2007-10-16 23:26:30 -07002559void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002561 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002562 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
Ram Pai390c6842005-11-07 17:17:51 -05002564 init_rwsem(&namespace_sem);
2565
Al Viro7d6fec42011-11-23 12:14:10 -05002566 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002567 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568
Ram Paib58fed82005-11-07 17:16:09 -05002569 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
2571 if (!mount_hashtable)
2572 panic("Failed to allocate mount hash table\n");
2573
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002574 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575
Eric Dumazet13f14b42008-02-06 01:37:57 -08002576 for (u = 0; u < HASH_SIZE; u++)
2577 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578
Nick Piggin99b7db72010-08-18 04:37:39 +10002579 br_lock_init(vfsmount_lock);
2580
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002581 err = sysfs_init();
2582 if (err)
2583 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002584 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002585 fs_kobj = kobject_create_and_add("fs", NULL);
2586 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002587 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 init_rootfs();
2589 init_mount_tree();
2590}
2591
Trond Myklebust616511d2009-06-22 15:09:13 -04002592void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002594 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002595
Al Virod498b252010-02-05 02:21:06 -05002596 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002597 return;
Ram Pai390c6842005-11-07 17:17:51 -05002598 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002599 br_write_lock(vfsmount_lock);
Al Virobe08d6d2011-12-06 13:32:36 -05002600 umount_tree(ns->root, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002601 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002602 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002603 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002604 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605}
Al Viro9d412a42011-03-17 22:08:28 -04002606
2607struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2608{
Tim Chen423e0ab2011-07-19 09:32:38 -07002609 struct vfsmount *mnt;
2610 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2611 if (!IS_ERR(mnt)) {
2612 /*
2613 * it is a longterm mount, don't release mnt until
2614 * we unmount before file sys is unregistered
2615 */
2616 mnt_make_longterm(mnt);
2617 }
2618 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002619}
2620EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002621
2622void kern_unmount(struct vfsmount *mnt)
2623{
2624 /* release long term mount so mount point can be released */
2625 if (!IS_ERR_OR_NULL(mnt)) {
2626 mnt_make_shortterm(mnt);
2627 mntput(mnt);
2628 }
2629}
2630EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002631
2632bool our_mnt(struct vfsmount *mnt)
2633{
Al Viro143c8c92011-11-25 00:46:35 -05002634 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05002635}