blob: d8ce529e0fb0ce950cd23f37f23f27b325876670 [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
Andi Kleenbaca7e42012-05-08 13:32:02 +0930400 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;
Andi Kleenbaca7e42012-05-08 13:32:02 +0930434 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{
Andi Kleenbaca7e42012-05-08 13:32:02 +0930440 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500441 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Andi Kleenbaca7e42012-05-08 13:32:02 +0930442 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
Andi Kleenbaca7e42012-05-08 13:32:02 +0930454 br_write_lock(&vfsmount_lock);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100455 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 }
Andi Kleenbaca7e42012-05-08 13:32:02 +0930476 br_write_unlock(&vfsmount_lock);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100477
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/*
David Howells64399b22012-06-25 12:55:28 +0100518 * lookup_mnt - Return the first child mount mounted at path
519 *
520 * "First" means first mounted chronologically. If you create the
521 * following mounts:
522 *
523 * mount /dev/sda1 /mnt
524 * mount /dev/sda2 /mnt
525 * mount /dev/sda3 /mnt
526 *
527 * Then lookup_mnt() on the base /mnt dentry in the root mount will
528 * return successively the root dentry and vfsmount of /dev/sda1, then
529 * /dev/sda2, then /dev/sda3, then NULL.
530 *
531 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500532 */
Al Viro1c755af2009-04-18 14:06:57 -0400533struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500534{
Al Viroc7105362011-11-24 18:22:03 -0500535 struct mount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000536
Andi Kleenbaca7e42012-05-08 13:32:02 +0930537 br_read_lock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500538 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
539 if (child_mnt) {
540 mnt_add_count(child_mnt, 1);
Andi Kleenbaca7e42012-05-08 13:32:02 +0930541 br_read_unlock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500542 return &child_mnt->mnt;
543 } else {
Andi Kleenbaca7e42012-05-08 13:32:02 +0930544 br_read_unlock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500545 return NULL;
546 }
Ram Paia05964f2005-11-07 17:20:17 -0500547}
548
Al Viro143c8c92011-11-25 00:46:35 -0500549static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800551 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552}
553
Nick Piggin99b7db72010-08-18 04:37:39 +1000554/*
555 * vfsmount lock must be held for write
556 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800557static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500558{
559 if (ns) {
560 ns->event = ++event;
561 wake_up_interruptible(&ns->poll);
562 }
563}
564
Nick Piggin99b7db72010-08-18 04:37:39 +1000565/*
566 * vfsmount lock must be held for write
567 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800568static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500569{
570 if (ns && ns->event != event) {
571 ns->event = event;
572 wake_up_interruptible(&ns->poll);
573 }
574}
575
Nick Piggin99b7db72010-08-18 04:37:39 +1000576/*
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100577 * Clear dentry's mounted state if it has no remaining mounts.
578 * vfsmount_lock must be held for write.
579 */
Al Viroaa0a4cf2011-11-24 19:31:36 -0500580static void dentry_reset_mounted(struct dentry *dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100581{
582 unsigned u;
583
584 for (u = 0; u < HASH_SIZE; u++) {
Al Virod5e50f72011-11-24 20:58:57 -0500585 struct mount *p;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100586
Al Viro1b8e5562011-11-24 21:01:32 -0500587 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
Al Viroa73324d2011-11-24 22:25:07 -0500588 if (p->mnt_mountpoint == dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100589 return;
590 }
591 }
592 spin_lock(&dentry->d_lock);
593 dentry->d_flags &= ~DCACHE_MOUNTED;
594 spin_unlock(&dentry->d_lock);
595}
596
597/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000598 * vfsmount lock must be held for write
599 */
Al Viro419148d2011-11-24 19:41:16 -0500600static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
Al Viroa73324d2011-11-24 22:25:07 -0500602 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500603 old_path->mnt = &mnt->mnt_parent->mnt;
604 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500605 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500606 list_del_init(&mnt->mnt_child);
Al Viro1b8e5562011-11-24 21:01:32 -0500607 list_del_init(&mnt->mnt_hash);
Al Viroaa0a4cf2011-11-24 19:31:36 -0500608 dentry_reset_mounted(old_path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
610
Nick Piggin99b7db72010-08-18 04:37:39 +1000611/*
612 * vfsmount lock must be held for write
613 */
Al Viro14cf1fa2011-11-25 00:01:17 -0500614void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
Al Viro44d964d62011-11-24 21:28:22 -0500615 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500616{
Al Viro3a2393d2011-11-25 03:19:09 -0500617 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viroa73324d2011-11-24 22:25:07 -0500618 child_mnt->mnt_mountpoint = dget(dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500619 child_mnt->mnt_parent = mnt;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100620 spin_lock(&dentry->d_lock);
621 dentry->d_flags |= DCACHE_MOUNTED;
622 spin_unlock(&dentry->d_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500623}
624
Nick Piggin99b7db72010-08-18 04:37:39 +1000625/*
626 * vfsmount lock must be held for write
627 */
Al Viro419148d2011-11-24 19:41:16 -0500628static void attach_mnt(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Al Viro14cf1fa2011-11-25 00:01:17 -0500630 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
Al Viro1b8e5562011-11-24 21:01:32 -0500631 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400632 hash(path->mnt, path->dentry));
Al Viro6b41d532011-11-24 23:24:33 -0500633 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500634}
635
636/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000637 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500638 */
Al Viro4b2619a2011-11-24 19:47:15 -0500639static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500640{
Al Viro0714a532011-11-24 22:19:58 -0500641 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500642 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500643 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500644 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500645
Al Viro0714a532011-11-24 22:19:58 -0500646 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500647
Al Viro1a4eeaf2011-11-25 02:19:55 -0500648 list_add_tail(&head, &mnt->mnt_list);
Al Viro5296c2c2012-06-09 00:59:08 -0400649 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500650 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500651
Ram Paib90fa9a2005-11-07 17:19:50 -0500652 list_splice(&head, n->list.prev);
653
Al Viro1b8e5562011-11-24 21:01:32 -0500654 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viroa73324d2011-11-24 22:25:07 -0500655 hash(&parent->mnt, mnt->mnt_mountpoint));
Al Viro6b41d532011-11-24 23:24:33 -0500656 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800657 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
Al Viro909b0a82011-11-25 03:06:56 -0500660static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661{
Al Viro6b41d532011-11-24 23:24:33 -0500662 struct list_head *next = p->mnt_mounts.next;
663 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500665 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500667 next = p->mnt_child.next;
668 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 break;
Al Viro0714a532011-11-24 22:19:58 -0500670 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 }
672 }
Al Viro6b41d532011-11-24 23:24:33 -0500673 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
Al Viro315fc832011-11-24 18:57:30 -0500676static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500677{
Al Viro6b41d532011-11-24 23:24:33 -0500678 struct list_head *prev = p->mnt_mounts.prev;
679 while (prev != &p->mnt_mounts) {
680 p = list_entry(prev, struct mount, mnt_child);
681 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500682 }
683 return p;
684}
685
Al Viro9d412a42011-03-17 22:08:28 -0400686struct vfsmount *
687vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
688{
Al Virob105e272011-11-24 20:38:33 -0500689 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400690 struct dentry *root;
691
692 if (!type)
693 return ERR_PTR(-ENODEV);
694
695 mnt = alloc_vfsmnt(name);
696 if (!mnt)
697 return ERR_PTR(-ENOMEM);
698
699 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500700 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400701
702 root = mount_fs(type, flags, name, data);
703 if (IS_ERR(root)) {
704 free_vfsmnt(mnt);
705 return ERR_CAST(root);
706 }
707
Al Virob105e272011-11-24 20:38:33 -0500708 mnt->mnt.mnt_root = root;
709 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500710 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500711 mnt->mnt_parent = mnt;
Andi Kleenbaca7e42012-05-08 13:32:02 +0930712 br_write_lock(&vfsmount_lock);
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100713 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Andi Kleenbaca7e42012-05-08 13:32:02 +0930714 br_write_unlock(&vfsmount_lock);
Al Virob105e272011-11-24 20:38:33 -0500715 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400716}
717EXPORT_SYMBOL_GPL(vfs_kern_mount);
718
Al Viro87129cc2011-11-24 21:24:27 -0500719static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500720 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
Al Viro87129cc2011-11-24 21:24:27 -0500722 struct super_block *sb = old->mnt.mnt_sb;
David Howells210cff12012-06-25 12:55:18 +0100723 struct mount *mnt;
724 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
David Howells210cff12012-06-25 12:55:18 +0100726 mnt = alloc_vfsmnt(old->mnt_devname);
727 if (!mnt)
728 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100729
David Howells210cff12012-06-25 12:55:18 +0100730 if (flag & (CL_SLAVE | CL_PRIVATE))
731 mnt->mnt_group_id = 0; /* not a peer of original */
732 else
733 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100734
David Howells210cff12012-06-25 12:55:18 +0100735 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
736 err = mnt_alloc_group_id(mnt);
737 if (err)
738 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 }
David Howells210cff12012-06-25 12:55:18 +0100740
741 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
742 atomic_inc(&sb->s_active);
743 mnt->mnt.mnt_sb = sb;
744 mnt->mnt.mnt_root = dget(root);
745 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
746 mnt->mnt_parent = mnt;
747 br_write_lock(&vfsmount_lock);
748 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
749 br_write_unlock(&vfsmount_lock);
750
751 if (flag & CL_SLAVE) {
752 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
753 mnt->mnt_master = old;
754 CLEAR_MNT_SHARED(mnt);
755 } else if (!(flag & CL_PRIVATE)) {
756 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
757 list_add(&mnt->mnt_share, &old->mnt_share);
758 if (IS_MNT_SLAVE(old))
759 list_add(&mnt->mnt_slave, &old->mnt_slave);
760 mnt->mnt_master = old->mnt_master;
761 }
762 if (flag & CL_MAKE_SHARED)
763 set_mnt_shared(mnt);
764
765 /* stick the duplicate mount on the same expiry list
766 * as the original if that was on one */
767 if (flag & CL_EXPIRE) {
768 if (!list_empty(&old->mnt_expire))
769 list_add(&mnt->mnt_expire, &old->mnt_expire);
770 }
771
Al Virocb338d02011-11-24 20:55:08 -0500772 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100773
774 out_free:
775 free_vfsmnt(mnt);
David Howells210cff12012-06-25 12:55:18 +0100776 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777}
778
Al Viro83adc752011-11-24 22:37:54 -0500779static inline void mntfree(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780{
Al Viro83adc752011-11-24 22:37:54 -0500781 struct vfsmount *m = &mnt->mnt;
782 struct super_block *sb = m->mnt_sb;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100783
Dave Hansen3d733632008-02-15 14:37:59 -0800784 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800785 * This probably indicates that somebody messed
786 * up a mnt_want/drop_write() pair. If this
787 * happens, the filesystem was probably unable
788 * to make r/w->r/o transitions.
789 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000790 /*
Nick Pigginb3e19d92011-01-07 17:50:11 +1100791 * The locking used to deal with mnt_count decrement provides barriers,
792 * so mnt_get_writers() below is safe.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000793 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100794 WARN_ON(mnt_get_writers(mnt));
Al Viro83adc752011-11-24 22:37:54 -0500795 fsnotify_vfsmount_delete(m);
796 dput(m->mnt_root);
797 free_vfsmnt(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 deactivate_super(sb);
799}
800
Al Viro900148d2011-11-25 00:33:11 -0500801static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500802{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100803put_again:
Al Virof03c6592011-01-14 22:30:21 -0500804#ifdef CONFIG_SMP
Andi Kleenbaca7e42012-05-08 13:32:02 +0930805 br_read_lock(&vfsmount_lock);
Al Viro5296c2c2012-06-09 00:59:08 -0400806 if (likely(mnt->mnt_ns)) {
807 /* shouldn't be the last one */
Al Viroaa9c0e02011-11-23 19:04:52 -0500808 mnt_add_count(mnt, -1);
Andi Kleenbaca7e42012-05-08 13:32:02 +0930809 br_read_unlock(&vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500810 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100811 }
Andi Kleenbaca7e42012-05-08 13:32:02 +0930812 br_read_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100813
Andi Kleenbaca7e42012-05-08 13:32:02 +0930814 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)) {
Andi Kleenbaca7e42012-05-08 13:32:02 +0930817 br_write_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100818 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;
Andi Kleenbaca7e42012-05-08 13:32:02 +0930824 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;
Andi Kleenbaca7e42012-05-08 13:32:02 +0930829 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 }
Andi Kleenbaca7e42012-05-08 13:32:02 +0930833
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100834 list_del(&mnt->mnt_instance);
Andi Kleenbaca7e42012-05-08 13:32:02 +0930835 br_write_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100836 mntfree(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500837}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100838
839void mntput(struct vfsmount *mnt)
840{
841 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -0500842 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100843 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -0500844 if (unlikely(m->mnt_expiry_mark))
845 m->mnt_expiry_mark = 0;
846 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100847 }
848}
849EXPORT_SYMBOL(mntput);
850
851struct vfsmount *mntget(struct vfsmount *mnt)
852{
853 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -0500854 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100855 return mnt;
856}
857EXPORT_SYMBOL(mntget);
858
Al Viro7b7b1ac2005-11-07 17:13:39 -0500859void mnt_pin(struct vfsmount *mnt)
860{
Andi Kleenbaca7e42012-05-08 13:32:02 +0930861 br_write_lock(&vfsmount_lock);
Al Viro863d6842011-11-25 00:57:42 -0500862 real_mount(mnt)->mnt_pinned++;
Andi Kleenbaca7e42012-05-08 13:32:02 +0930863 br_write_unlock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500864}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500865EXPORT_SYMBOL(mnt_pin);
866
Al Viro863d6842011-11-25 00:57:42 -0500867void mnt_unpin(struct vfsmount *m)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500868{
Al Viro863d6842011-11-25 00:57:42 -0500869 struct mount *mnt = real_mount(m);
Andi Kleenbaca7e42012-05-08 13:32:02 +0930870 br_write_lock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500871 if (mnt->mnt_pinned) {
Al Viro863d6842011-11-25 00:57:42 -0500872 mnt_add_count(mnt, 1);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500873 mnt->mnt_pinned--;
874 }
Andi Kleenbaca7e42012-05-08 13:32:02 +0930875 br_write_unlock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500876}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500877EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800879static inline void mangle(struct seq_file *m, const char *s)
880{
881 seq_escape(m, s, " \t\n\\");
882}
883
884/*
885 * Simple .show_options callback for filesystems which don't want to
886 * implement more complex mount option showing.
887 *
888 * See also save_mount_options().
889 */
Al Viro34c80b12011-12-08 21:32:45 -0500890int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800891{
Al Viro2a32ceb2009-05-08 16:05:57 -0400892 const char *options;
893
894 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -0500895 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800896
897 if (options != NULL && options[0]) {
898 seq_putc(m, ',');
899 mangle(m, options);
900 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400901 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800902
903 return 0;
904}
905EXPORT_SYMBOL(generic_show_options);
906
907/*
908 * If filesystem uses generic_show_options(), this function should be
909 * called from the fill_super() callback.
910 *
911 * The .remount_fs callback usually needs to be handled in a special
912 * way, to make sure, that previous options are not overwritten if the
913 * remount fails.
914 *
915 * Also note, that if the filesystem's .remount_fs function doesn't
916 * reset all options to their default value, but changes only newly
917 * given options, then the displayed options will not reflect reality
918 * any more.
919 */
920void save_mount_options(struct super_block *sb, char *options)
921{
Al Viro2a32ceb2009-05-08 16:05:57 -0400922 BUG_ON(sb->s_options);
923 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800924}
925EXPORT_SYMBOL(save_mount_options);
926
Al Viro2a32ceb2009-05-08 16:05:57 -0400927void replace_mount_options(struct super_block *sb, char *options)
928{
929 char *old = sb->s_options;
930 rcu_assign_pointer(sb->s_options, options);
931 if (old) {
932 synchronize_rcu();
933 kfree(old);
934 }
935}
936EXPORT_SYMBOL(replace_mount_options);
937
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100938#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -0500939/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940static void *m_start(struct seq_file *m, loff_t *pos)
941{
Al Viro0bcf8f92012-06-09 01:16:59 -0400942 struct proc_mounts *p = proc_mounts(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
Ram Pai390c6842005-11-07 17:17:51 -0500944 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100945 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946}
947
948static void *m_next(struct seq_file *m, void *v, loff_t *pos)
949{
Al Viro0bcf8f92012-06-09 01:16:59 -0400950 struct proc_mounts *p = proc_mounts(m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700951
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100952 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953}
954
955static void m_stop(struct seq_file *m, void *v)
956{
Ram Pai390c6842005-11-07 17:17:51 -0500957 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958}
959
Al Viro0226f492011-12-06 12:21:54 -0500960static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -0500961{
Al Viro0bcf8f92012-06-09 01:16:59 -0400962 struct proc_mounts *p = proc_mounts(m);
Al Viro1a4eeaf2011-11-25 02:19:55 -0500963 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -0500964 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965}
966
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100967const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 .start = m_start,
969 .next = m_next,
970 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -0500971 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -0500972};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100973#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975/**
976 * may_umount_tree - check if a mount tree is busy
977 * @mnt: root of mount tree
978 *
979 * This is called to check if a tree of mounts has any
980 * open files, pwds, chroots or sub mounts that are
981 * busy.
982 */
Al Viro909b0a82011-11-25 03:06:56 -0500983int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Al Viro909b0a82011-11-25 03:06:56 -0500985 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -0500986 int actual_refs = 0;
987 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -0500988 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -0500989 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
Nick Pigginb3e19d92011-01-07 17:50:11 +1100991 /* write lock needed for mnt_get_count */
Andi Kleenbaca7e42012-05-08 13:32:02 +0930992 br_write_lock(&vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -0500993 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -0500994 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 }
Andi Kleenbaca7e42012-05-08 13:32:02 +0930997 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
999 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001000 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
Ian Kente3474a82006-03-27 01:14:51 -08001002 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003}
1004
1005EXPORT_SYMBOL(may_umount_tree);
1006
1007/**
1008 * may_umount - check if a mount point is busy
1009 * @mnt: root of mount
1010 *
1011 * This is called to check if a mount point has any
1012 * open files, pwds, chroots or sub mounts. If the
1013 * mount has sub mounts this will return busy
1014 * regardless of whether the sub mounts are busy.
1015 *
1016 * Doesn't take quota and stuff into account. IOW, in some cases it will
1017 * give false negatives. The main reason why it's here is that we need
1018 * a non-destructive way to look for easily umountable filesystems.
1019 */
1020int may_umount(struct vfsmount *mnt)
1021{
Ian Kente3474a82006-03-27 01:14:51 -08001022 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001023 down_read(&namespace_sem);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301024 br_write_lock(&vfsmount_lock);
Al Viro1ab59732011-11-24 21:35:16 -05001025 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001026 ret = 0;
Andi Kleenbaca7e42012-05-08 13:32:02 +09301027 br_write_unlock(&vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001028 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001029 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030}
1031
1032EXPORT_SYMBOL(may_umount);
1033
Ram Paib90fa9a2005-11-07 17:19:50 -05001034void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035{
Al Virod5e50f72011-11-24 20:58:57 -05001036 struct mount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001037 while (!list_empty(head)) {
Al Viro1b8e5562011-11-24 21:01:32 -05001038 mnt = list_first_entry(head, struct mount, mnt_hash);
1039 list_del_init(&mnt->mnt_hash);
Al Viro676da582011-11-24 21:47:05 -05001040 if (mnt_has_parent(mnt)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001041 struct dentry *dentry;
Al Viro863d6842011-11-25 00:57:42 -05001042 struct mount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001043
Andi Kleenbaca7e42012-05-08 13:32:02 +09301044 br_write_lock(&vfsmount_lock);
Al Viroa73324d2011-11-24 22:25:07 -05001045 dentry = mnt->mnt_mountpoint;
Al Viro863d6842011-11-25 00:57:42 -05001046 m = mnt->mnt_parent;
Al Viroa73324d2011-11-24 22:25:07 -05001047 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -05001048 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001049 m->mnt_ghosts--;
Andi Kleenbaca7e42012-05-08 13:32:02 +09301050 br_write_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001051 dput(dentry);
Al Viro863d6842011-11-25 00:57:42 -05001052 mntput(&m->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 }
Al Virod5e50f72011-11-24 20:58:57 -05001054 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001055 }
1056}
1057
Nick Piggin99b7db72010-08-18 04:37:39 +10001058/*
1059 * vfsmount lock must be held for write
1060 * namespace_sem must be held for write
1061 */
Al Viro761d5c32011-11-24 21:07:43 -05001062void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001063{
Al Viro7b8a53f2011-01-15 20:08:44 -05001064 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001065 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001066
Al Viro909b0a82011-11-25 03:06:56 -05001067 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro1b8e5562011-11-24 21:01:32 -05001068 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001069
Al Virof9c34842014-08-18 15:09:26 -04001070 list_for_each_entry(p, &tmp_list, mnt_hash)
1071 list_del_init(&p->mnt_child);
1072
Ram Paia05964f2005-11-07 17:20:17 -05001073 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001074 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001075
Al Viro1b8e5562011-11-24 21:01:32 -05001076 list_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001077 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001078 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001079 __touch_mnt_namespace(p->mnt_ns);
1080 p->mnt_ns = NULL;
Al Viro676da582011-11-24 21:47:05 -05001081 if (mnt_has_parent(p)) {
Al Viro863d6842011-11-25 00:57:42 -05001082 p->mnt_parent->mnt_ghosts++;
Al Viroa73324d2011-11-24 22:25:07 -05001083 dentry_reset_mounted(p->mnt_mountpoint);
Al Viro7c4b93d2008-03-21 23:59:49 -04001084 }
Al Viro0f0afb12011-11-24 20:43:10 -05001085 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 }
Al Viro7b8a53f2011-01-15 20:08:44 -05001087 list_splice(&tmp_list, kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088}
1089
Al Viro692afc32011-11-24 21:15:14 -05001090static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
Al Viroc35038b2008-03-22 00:46:23 -04001091
Al Viro1ab59732011-11-24 21:35:16 -05001092static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093{
Al Viro1ab59732011-11-24 21:35:16 -05001094 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001096 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
Al Viro1ab59732011-11-24 21:35:16 -05001098 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 if (retval)
1100 return retval;
1101
1102 /*
1103 * Allow userspace to request a mountpoint be expired rather than
1104 * unmounting unconditionally. Unmount only happens if:
1105 * (1) the mark is already set (the mark is cleared by mntput())
1106 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1107 */
1108 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001109 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 flags & (MNT_FORCE | MNT_DETACH))
1111 return -EINVAL;
1112
Nick Pigginb3e19d92011-01-07 17:50:11 +11001113 /*
1114 * probably don't strictly need the lock here if we examined
1115 * all race cases, but it's a slowpath.
1116 */
Andi Kleenbaca7e42012-05-08 13:32:02 +09301117 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -05001118 if (mnt_get_count(mnt) != 2) {
Andi Kleenbaca7e42012-05-08 13:32:02 +09301119 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001121 }
Andi Kleenbaca7e42012-05-08 13:32:02 +09301122 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Al Viro863d6842011-11-25 00:57:42 -05001124 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 return -EAGAIN;
1126 }
1127
1128 /*
1129 * If we may have to abort operations to get out of this
1130 * mount, and they will themselves hold resources we must
1131 * allow the fs to do things. In the Unix tradition of
1132 * 'Gee thats tricky lets do it in userspace' the umount_begin
1133 * might fail to complete on the first run through as other tasks
1134 * must return, and the like. Thats for the mount program to worry
1135 * about for the moment.
1136 */
1137
Al Viro42faad92008-04-24 07:21:56 -04001138 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001139 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
1142 /*
1143 * No sense to grab the lock for this test, but test itself looks
1144 * somewhat bogus. Suggestions for better replacement?
1145 * Ho-hum... In principle, we might treat that as umount + switch
1146 * to rootfs. GC would eventually take care of the old vfsmount.
1147 * Actually it makes sense, especially if rootfs would contain a
1148 * /reboot - static binary that would close all descriptors and
1149 * call reboot(9). Then init(8) could umount root and exec /reboot.
1150 */
Al Viro1ab59732011-11-24 21:35:16 -05001151 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 /*
1153 * Special case for "unmounting" root ...
1154 * we just try to remount it readonly.
1155 */
1156 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001157 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 up_write(&sb->s_umount);
1160 return retval;
1161 }
1162
Ram Pai390c6842005-11-07 17:17:51 -05001163 down_write(&namespace_sem);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301164 br_write_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001165 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
Al Viroc35038b2008-03-22 00:46:23 -04001167 if (!(flags & MNT_DETACH))
Al Viro1ab59732011-11-24 21:35:16 -05001168 shrink_submounts(mnt, &umount_list);
Al Viroc35038b2008-03-22 00:46:23 -04001169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001171 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001172 if (!list_empty(&mnt->mnt_list))
Al Viro1ab59732011-11-24 21:35:16 -05001173 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 retval = 0;
1175 }
Andi Kleenbaca7e42012-05-08 13:32:02 +09301176 br_write_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001177 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001178 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 return retval;
1180}
1181
1182/*
1183 * Now umount can handle mount points as well as block devices.
1184 * This is important for filesystems which use unnamed block devices.
1185 *
1186 * We now support a flag for forced unmount like the other 'big iron'
1187 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1188 */
1189
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001190SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191{
Al Viro2d8f3032008-07-22 09:59:21 -04001192 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001193 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001195 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001197 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1198 return -EINVAL;
1199
1200 if (!(flags & UMOUNT_NOFOLLOW))
1201 lookup_flags |= LOOKUP_FOLLOW;
1202
1203 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 if (retval)
1205 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001206 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001208 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001210 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 goto dput_and_out;
1212
1213 retval = -EPERM;
1214 if (!capable(CAP_SYS_ADMIN))
1215 goto dput_and_out;
1216
Al Viro900148d2011-11-25 00:33:11 -05001217 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001219 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001220 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001221 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222out:
1223 return retval;
1224}
1225
1226#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1227
1228/*
Ram Paib58fed82005-11-07 17:16:09 -05001229 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001231SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232{
Ram Paib58fed82005-11-07 17:16:09 -05001233 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234}
1235
1236#endif
1237
Al Viro2d92ab32008-08-02 00:51:11 -04001238static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239{
1240 if (capable(CAP_SYS_ADMIN))
1241 return 0;
1242 return -EPERM;
1243#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001244 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001246 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001247 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 return -EPERM;
1249 }
Al Viro2d92ab32008-08-02 00:51:11 -04001250 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 return -EPERM;
1252 return 0;
1253#endif
1254}
1255
Al Viro87129cc2011-11-24 21:24:27 -05001256struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001257 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258{
Al Viroa73324d2011-11-24 22:25:07 -05001259 struct mount *res, *p, *q, *r;
Al Viro1a390682008-03-21 20:48:19 -04001260 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
Al Virofc7be132011-11-25 01:05:37 -05001262 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
David Howells210cff12012-06-25 12:55:18 +01001263 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001264
Ram Pai36341f62005-11-07 17:17:22 -05001265 res = q = clone_mnt(mnt, dentry, flag);
David Howells210cff12012-06-25 12:55:18 +01001266 if (IS_ERR(q))
1267 return q;
1268
Al Viroa73324d2011-11-24 22:25:07 -05001269 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
1271 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001272 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001273 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001274 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 continue;
1276
Al Viro909b0a82011-11-25 03:06:56 -05001277 for (s = r; s; s = next_mnt(s, r)) {
Al Virofc7be132011-11-25 01:05:37 -05001278 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001279 s = skip_mnt_tree(s);
1280 continue;
1281 }
Al Viro0714a532011-11-24 22:19:58 -05001282 while (p != s->mnt_parent) {
1283 p = p->mnt_parent;
1284 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 }
Al Viro87129cc2011-11-24 21:24:27 -05001286 p = s;
Al Virocb338d02011-11-24 20:55:08 -05001287 path.mnt = &q->mnt;
Al Viroa73324d2011-11-24 22:25:07 -05001288 path.dentry = p->mnt_mountpoint;
Al Viro87129cc2011-11-24 21:24:27 -05001289 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howells210cff12012-06-25 12:55:18 +01001290 if (IS_ERR(q))
1291 goto out;
Andi Kleenbaca7e42012-05-08 13:32:02 +09301292 br_write_lock(&vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001293 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Virocb338d02011-11-24 20:55:08 -05001294 attach_mnt(q, &path);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301295 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 }
1297 }
1298 return res;
David Howells210cff12012-06-25 12:55:18 +01001299out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001301 LIST_HEAD(umount_list);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301302 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001303 umount_tree(res, 0, &umount_list);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301304 br_write_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001305 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 }
David Howells210cff12012-06-25 12:55:18 +01001307 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308}
1309
David Howells210cff12012-06-25 12:55:18 +01001310/* Caller should check returned pointer for errors */
1311
Al Viro589ff872009-04-18 03:28:19 -04001312struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001313{
Al Virocb338d02011-11-24 20:55:08 -05001314 struct mount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001315 down_write(&namespace_sem);
Al Viro87129cc2011-11-24 21:24:27 -05001316 tree = copy_tree(real_mount(path->mnt), path->dentry,
1317 CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001318 up_write(&namespace_sem);
David Howells210cff12012-06-25 12:55:18 +01001319 if (IS_ERR(tree))
1320 return NULL;
1321 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001322}
1323
1324void drop_collected_mounts(struct vfsmount *mnt)
1325{
1326 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001327 down_write(&namespace_sem);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301328 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001329 umount_tree(real_mount(mnt), 0, &umount_list);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301330 br_write_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001331 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001332 release_mounts(&umount_list);
1333}
1334
Al Viro1f707132010-01-30 22:51:25 -05001335int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1336 struct vfsmount *root)
1337{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001338 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001339 int res = f(root, arg);
1340 if (res)
1341 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001342 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1343 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001344 if (res)
1345 return res;
1346 }
1347 return 0;
1348}
1349
Al Viro4b8b21f2011-11-24 19:54:23 -05001350static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001351{
Al Viro315fc832011-11-24 18:57:30 -05001352 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001353
Al Viro909b0a82011-11-25 03:06:56 -05001354 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001355 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001356 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001357 }
1358}
1359
Al Viro4b8b21f2011-11-24 19:54:23 -05001360static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001361{
Al Viro315fc832011-11-24 18:57:30 -05001362 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001363
Al Viro909b0a82011-11-25 03:06:56 -05001364 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001365 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001366 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001367 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001368 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001369 return err;
1370 }
1371 }
1372 }
1373
1374 return 0;
1375}
1376
Ram Paib90fa9a2005-11-07 17:19:50 -05001377/*
1378 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001379 * @nd : place the mount tree @source_mnt is attached
1380 * @parent_nd : if non-null, detach the source_mnt from its parent and
1381 * store the parent mount and mountpoint dentry.
1382 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001383 *
1384 * NOTE: in the table below explains the semantics when a source mount
1385 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001386 * ---------------------------------------------------------------------------
1387 * | BIND MOUNT OPERATION |
1388 * |**************************************************************************
1389 * | source-->| shared | private | slave | unbindable |
1390 * | dest | | | | |
1391 * | | | | | | |
1392 * | v | | | | |
1393 * |**************************************************************************
1394 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1395 * | | | | | |
1396 * |non-shared| shared (+) | private | slave (*) | invalid |
1397 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001398 * A bind operation clones the source mount and mounts the clone on the
1399 * destination mount.
1400 *
1401 * (++) the cloned mount is propagated to all the mounts in the propagation
1402 * tree of the destination mount and the cloned mount is added to
1403 * the peer group of the source mount.
1404 * (+) the cloned mount is created under the destination mount and is marked
1405 * as shared. The cloned mount is added to the peer group of the source
1406 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001407 * (+++) the mount is propagated to all the mounts in the propagation tree
1408 * of the destination mount and the cloned mount is made slave
1409 * of the same master as that of the source mount. The cloned mount
1410 * is marked as 'shared and slave'.
1411 * (*) the cloned mount is made a slave of the same master as that of the
1412 * source mount.
1413 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001414 * ---------------------------------------------------------------------------
1415 * | MOVE MOUNT OPERATION |
1416 * |**************************************************************************
1417 * | source-->| shared | private | slave | unbindable |
1418 * | dest | | | | |
1419 * | | | | | | |
1420 * | v | | | | |
1421 * |**************************************************************************
1422 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1423 * | | | | | |
1424 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1425 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001426 *
1427 * (+) the mount is moved to the destination. And is then propagated to
1428 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001429 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001430 * (+++) the mount is moved to the destination and is then propagated to
1431 * all the mounts belonging to the destination mount's propagation tree.
1432 * the mount is marked as 'shared and slave'.
1433 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001434 *
1435 * if the source mount is a tree, the operations explained above is
1436 * applied to each mount in the tree.
1437 * Must be called without spinlocks held, since this function can sleep
1438 * in allocations.
1439 */
Al Viro0fb54e52011-11-24 19:59:16 -05001440static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001441 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001442{
1443 LIST_HEAD(tree_list);
Al Viroa8d56d82011-11-24 23:59:29 -05001444 struct mount *dest_mnt = real_mount(path->mnt);
Al Viro1a390682008-03-21 20:48:19 -04001445 struct dentry *dest_dentry = path->dentry;
Al Viro315fc832011-11-24 18:57:30 -05001446 struct mount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001447 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001448
Al Virofc7be132011-11-25 01:05:37 -05001449 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001450 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001451 if (err)
1452 goto out;
1453 }
Al Viroa8d56d82011-11-24 23:59:29 -05001454 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001455 if (err)
1456 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001457
Andi Kleenbaca7e42012-05-08 13:32:02 +09301458 br_write_lock(&vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001459
Al Virofc7be132011-11-25 01:05:37 -05001460 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro909b0a82011-11-25 03:06:56 -05001461 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001462 set_mnt_shared(p);
Ram Paib90fa9a2005-11-07 17:19:50 -05001463 }
Al Viro1a390682008-03-21 20:48:19 -04001464 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001465 detach_mnt(source_mnt, parent_path);
1466 attach_mnt(source_mnt, path);
Al Viro143c8c92011-11-25 00:46:35 -05001467 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001468 } else {
Al Viro14cf1fa2011-11-25 00:01:17 -05001469 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
Al Viro0fb54e52011-11-24 19:59:16 -05001470 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05001471 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001472
Al Viro1b8e5562011-11-24 21:01:32 -05001473 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1474 list_del_init(&child->mnt_hash);
Al Viro4b2619a2011-11-24 19:47:15 -05001475 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05001476 }
Andi Kleenbaca7e42012-05-08 13:32:02 +09301477 br_write_unlock(&vfsmount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +10001478
Ram Paib90fa9a2005-11-07 17:19:50 -05001479 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001480
1481 out_cleanup_ids:
Al Virofc7be132011-11-25 01:05:37 -05001482 if (IS_MNT_SHARED(dest_mnt))
Al Viro0fb54e52011-11-24 19:59:16 -05001483 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001484 out:
1485 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001486}
1487
Al Virob12cea92011-03-18 08:55:38 -04001488static int lock_mount(struct path *path)
1489{
1490 struct vfsmount *mnt;
1491retry:
1492 mutex_lock(&path->dentry->d_inode->i_mutex);
1493 if (unlikely(cant_mount(path->dentry))) {
1494 mutex_unlock(&path->dentry->d_inode->i_mutex);
1495 return -ENOENT;
1496 }
1497 down_write(&namespace_sem);
1498 mnt = lookup_mnt(path);
1499 if (likely(!mnt))
1500 return 0;
1501 up_write(&namespace_sem);
1502 mutex_unlock(&path->dentry->d_inode->i_mutex);
1503 path_put(path);
1504 path->mnt = mnt;
1505 path->dentry = dget(mnt->mnt_root);
1506 goto retry;
1507}
1508
1509static void unlock_mount(struct path *path)
1510{
1511 up_write(&namespace_sem);
1512 mutex_unlock(&path->dentry->d_inode->i_mutex);
1513}
1514
Al Viro95bc5f22011-11-25 00:30:56 -05001515static int graft_tree(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516{
Al Viro95bc5f22011-11-25 00:30:56 -05001517 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 return -EINVAL;
1519
Al Viro8c3ee422008-03-22 18:00:39 -04001520 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001521 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 return -ENOTDIR;
1523
Al Virob12cea92011-03-18 08:55:38 -04001524 if (d_unlinked(path->dentry))
1525 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526
Al Viro95bc5f22011-11-25 00:30:56 -05001527 return attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528}
1529
1530/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001531 * Sanity check the flags to change_mnt_propagation.
1532 */
1533
1534static int flags_to_propagation_type(int flags)
1535{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001536 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001537
1538 /* Fail if any non-propagation flags are set */
1539 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1540 return 0;
1541 /* Only one propagation flag should be set */
1542 if (!is_power_of_2(type))
1543 return 0;
1544 return type;
1545}
1546
1547/*
Ram Pai07b20882005-11-07 17:19:07 -05001548 * recursively change the type of the mountpoint.
1549 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001550static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001551{
Al Viro315fc832011-11-24 18:57:30 -05001552 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001553 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001554 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001555 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001556 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001557
Miklos Szerediee6f9582007-05-08 00:30:40 -07001558 if (!capable(CAP_SYS_ADMIN))
1559 return -EPERM;
1560
Al Viro2d92ab32008-08-02 00:51:11 -04001561 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001562 return -EINVAL;
1563
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001564 type = flags_to_propagation_type(flag);
1565 if (!type)
1566 return -EINVAL;
1567
Ram Pai07b20882005-11-07 17:19:07 -05001568 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001569 if (type == MS_SHARED) {
1570 err = invent_group_ids(mnt, recurse);
1571 if (err)
1572 goto out_unlock;
1573 }
1574
Andi Kleenbaca7e42012-05-08 13:32:02 +09301575 br_write_lock(&vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -05001576 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001577 change_mnt_propagation(m, type);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301578 br_write_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001579
1580 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001581 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001582 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001583}
1584
1585/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 * do loopback mount.
1587 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001588static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001589 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590{
Al Virob12cea92011-03-18 08:55:38 -04001591 LIST_HEAD(umount_list);
Al Viro2d92ab32008-08-02 00:51:11 -04001592 struct path old_path;
Al Viro87129cc2011-11-24 21:24:27 -05001593 struct mount *mnt = NULL, *old;
Al Viro2d92ab32008-08-02 00:51:11 -04001594 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 if (err)
1596 return err;
1597 if (!old_name || !*old_name)
1598 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001599 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 if (err)
1601 return err;
1602
Al Virob12cea92011-03-18 08:55:38 -04001603 err = lock_mount(path);
1604 if (err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001605 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001606
Al Viro87129cc2011-11-24 21:24:27 -05001607 old = real_mount(old_path.mnt);
1608
Al Virob12cea92011-03-18 08:55:38 -04001609 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001610 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04001611 goto out2;
1612
Al Viro143c8c92011-11-25 00:46:35 -05001613 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
Al Virob12cea92011-03-18 08:55:38 -04001614 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615
Al Viroccd48bc2005-11-07 17:15:04 -05001616 if (recurse)
Al Viro87129cc2011-11-24 21:24:27 -05001617 mnt = copy_tree(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001618 else
Al Viro87129cc2011-11-24 21:24:27 -05001619 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001620
David Howells210cff12012-06-25 12:55:18 +01001621 if (IS_ERR(mnt)) {
1622 err = PTR_ERR(mnt);
1623 goto out;
1624 }
Al Viroccd48bc2005-11-07 17:15:04 -05001625
Al Viro95bc5f22011-11-25 00:30:56 -05001626 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001627 if (err) {
Andi Kleenbaca7e42012-05-08 13:32:02 +09301628 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001629 umount_tree(mnt, 0, &umount_list);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301630 br_write_unlock(&vfsmount_lock);
Ram Pai5b83d2c2005-11-07 17:16:29 -05001631 }
Al Virob12cea92011-03-18 08:55:38 -04001632out2:
1633 unlock_mount(path);
1634 release_mounts(&umount_list);
Al Viroccd48bc2005-11-07 17:15:04 -05001635out:
Al Viro2d92ab32008-08-02 00:51:11 -04001636 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 return err;
1638}
1639
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001640static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1641{
1642 int error = 0;
1643 int readonly_request = 0;
1644
1645 if (ms_flags & MS_RDONLY)
1646 readonly_request = 1;
1647 if (readonly_request == __mnt_is_readonly(mnt))
1648 return 0;
1649
1650 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05001651 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001652 else
Al Viro83adc752011-11-24 22:37:54 -05001653 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001654 return error;
1655}
1656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657/*
1658 * change filesystem flags. dir should be a physical root of filesystem.
1659 * If you've mounted a non-root directory somewhere and want to do remount
1660 * on it - tough luck.
1661 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001662static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 void *data)
1664{
1665 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001666 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05001667 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
1669 if (!capable(CAP_SYS_ADMIN))
1670 return -EPERM;
1671
Al Viro143c8c92011-11-25 00:46:35 -05001672 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 return -EINVAL;
1674
Al Viro2d92ab32008-08-02 00:51:11 -04001675 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 return -EINVAL;
1677
Eric Parisff36fe22011-03-03 16:09:14 -05001678 err = security_sb_remount(sb, data);
1679 if (err)
1680 return err;
1681
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001683 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001684 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001685 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001686 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001687 if (!err) {
Andi Kleenbaca7e42012-05-08 13:32:02 +09301688 br_write_lock(&vfsmount_lock);
Eric W. Biedermanb47d65d2014-07-28 16:26:53 -07001689 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
Al Viro143c8c92011-11-25 00:46:35 -05001690 mnt->mnt.mnt_flags = mnt_flags;
Andi Kleenbaca7e42012-05-08 13:32:02 +09301691 br_write_unlock(&vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001694 if (!err) {
Andi Kleenbaca7e42012-05-08 13:32:02 +09301695 br_write_lock(&vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001696 touch_mnt_namespace(mnt->mnt_ns);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301697 br_write_unlock(&vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001698 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 return err;
1700}
1701
Al Virocbbe3622011-11-24 20:01:19 -05001702static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05001703{
Al Viro315fc832011-11-24 18:57:30 -05001704 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001705 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001706 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05001707 return 1;
1708 }
1709 return 0;
1710}
1711
Al Viro0a0d8a42008-08-02 00:55:27 -04001712static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713{
Al Viro2d92ab32008-08-02 00:51:11 -04001714 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05001715 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05001716 struct mount *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 int err = 0;
1718 if (!capable(CAP_SYS_ADMIN))
1719 return -EPERM;
1720 if (!old_name || !*old_name)
1721 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001722 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 if (err)
1724 return err;
1725
Al Virob12cea92011-03-18 08:55:38 -04001726 err = lock_mount(path);
David Howellscc53ce52011-01-14 18:45:26 +00001727 if (err < 0)
1728 goto out;
1729
Al Viro143c8c92011-11-25 00:46:35 -05001730 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05001731 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05001732
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001734 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 goto out1;
1736
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001737 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001738 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
1740 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001741 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001742 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
Al Viro676da582011-11-24 21:47:05 -05001744 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05001745 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
Al Viro2d92ab32008-08-02 00:51:11 -04001747 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1748 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001749 goto out1;
1750 /*
1751 * Don't move a mount residing in a shared parent.
1752 */
Al Virofc7be132011-11-25 01:05:37 -05001753 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001754 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001755 /*
1756 * Don't move a mount tree containing unbindable mounts to a destination
1757 * mount which is shared.
1758 */
Al Virofc7be132011-11-25 01:05:37 -05001759 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05001760 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05001762 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05001763 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05001764 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
Al Viro0fb54e52011-11-24 19:59:16 -05001766 err = attach_recursive_mnt(old, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001767 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001768 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 /* if the mount is moved, it should no longer be expire
1771 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05001772 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773out1:
Al Virob12cea92011-03-18 08:55:38 -04001774 unlock_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001777 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001778 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 return err;
1780}
1781
Al Viro9d412a42011-03-17 22:08:28 -04001782static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1783{
1784 int err;
1785 const char *subtype = strchr(fstype, '.');
1786 if (subtype) {
1787 subtype++;
1788 err = -EINVAL;
1789 if (!subtype[0])
1790 goto err;
1791 } else
1792 subtype = "";
1793
1794 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1795 err = -ENOMEM;
1796 if (!mnt->mnt_sb->s_subtype)
1797 goto err;
1798 return mnt;
1799
1800 err:
1801 mntput(mnt);
1802 return ERR_PTR(err);
1803}
1804
Al Viro79e801a2011-12-12 23:07:05 -05001805static struct vfsmount *
Al Viro9d412a42011-03-17 22:08:28 -04001806do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1807{
1808 struct file_system_type *type = get_fs_type(fstype);
1809 struct vfsmount *mnt;
1810 if (!type)
1811 return ERR_PTR(-ENODEV);
1812 mnt = vfs_kern_mount(type, flags, name, data);
1813 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1814 !mnt->mnt_sb->s_subtype)
1815 mnt = fs_set_subtype(mnt, fstype);
1816 put_filesystem(type);
1817 return mnt;
1818}
Al Viro9d412a42011-03-17 22:08:28 -04001819
1820/*
1821 * add a mount into a namespace's mount tree
1822 */
Al Viro95bc5f22011-11-25 00:30:56 -05001823static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04001824{
1825 int err;
1826
1827 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1828
Al Virob12cea92011-03-18 08:55:38 -04001829 err = lock_mount(path);
1830 if (err)
1831 return err;
Al Viro9d412a42011-03-17 22:08:28 -04001832
1833 err = -EINVAL;
Al Viro56f9e252012-09-21 08:19:02 -04001834 if (unlikely(!check_mnt(real_mount(path->mnt)))) {
1835 /* that's acceptable only for automounts done in private ns */
1836 if (!(mnt_flags & MNT_SHRINKABLE))
1837 goto unlock;
1838 /* ... and for those we'd better have mountpoint still alive */
1839 if (!real_mount(path->mnt)->mnt_ns)
1840 goto unlock;
1841 }
Al Viro9d412a42011-03-17 22:08:28 -04001842
1843 /* Refuse the same filesystem on the same mount point */
1844 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05001845 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04001846 path->mnt->mnt_root == path->dentry)
1847 goto unlock;
1848
1849 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05001850 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04001851 goto unlock;
1852
Al Viro95bc5f22011-11-25 00:30:56 -05001853 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro9d412a42011-03-17 22:08:28 -04001854 err = graft_tree(newmnt, path);
1855
1856unlock:
Al Virob12cea92011-03-18 08:55:38 -04001857 unlock_mount(path);
Al Viro9d412a42011-03-17 22:08:28 -04001858 return err;
1859}
Al Virob1e75df2011-01-17 01:47:59 -05001860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861/*
1862 * create a new mount for userspace and request it to be added into the
1863 * namespace's tree
1864 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001865static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 int mnt_flags, char *name, void *data)
1867{
1868 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05001869 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
Vegard Nossumeca6f532009-09-18 13:05:45 -07001871 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 return -EINVAL;
1873
1874 /* we need capabilities... */
1875 if (!capable(CAP_SYS_ADMIN))
1876 return -EPERM;
1877
1878 mnt = do_kern_mount(type, flags, name, data);
1879 if (IS_ERR(mnt))
1880 return PTR_ERR(mnt);
1881
Al Viro95bc5f22011-11-25 00:30:56 -05001882 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05001883 if (err)
1884 mntput(mnt);
1885 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886}
1887
Al Viro19a167a2011-01-17 01:35:23 -05001888int finish_automount(struct vfsmount *m, struct path *path)
1889{
Al Viro6776db3d2011-11-25 00:22:05 -05001890 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05001891 int err;
1892 /* The new mount record should have at least 2 refs to prevent it being
1893 * expired before we get a chance to add it
1894 */
Al Viro6776db3d2011-11-25 00:22:05 -05001895 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05001896
1897 if (m->mnt_sb == path->mnt->mnt_sb &&
1898 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05001899 err = -ELOOP;
1900 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05001901 }
1902
Al Viro95bc5f22011-11-25 00:30:56 -05001903 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05001904 if (!err)
1905 return 0;
1906fail:
1907 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05001908 if (!list_empty(&mnt->mnt_expire)) {
Al Virob1e75df2011-01-17 01:47:59 -05001909 down_write(&namespace_sem);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301910 br_write_lock(&vfsmount_lock);
Al Viro6776db3d2011-11-25 00:22:05 -05001911 list_del_init(&mnt->mnt_expire);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301912 br_write_unlock(&vfsmount_lock);
Al Virob1e75df2011-01-17 01:47:59 -05001913 up_write(&namespace_sem);
Al Viro19a167a2011-01-17 01:35:23 -05001914 }
Al Virob1e75df2011-01-17 01:47:59 -05001915 mntput(m);
1916 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05001917 return err;
1918}
1919
David Howellsea5b7782011-01-14 19:10:03 +00001920/**
1921 * mnt_set_expiry - Put a mount on an expiration list
1922 * @mnt: The mount to list.
1923 * @expiry_list: The list to add the mount to.
1924 */
1925void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1926{
1927 down_write(&namespace_sem);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301928 br_write_lock(&vfsmount_lock);
David Howellsea5b7782011-01-14 19:10:03 +00001929
Al Viro6776db3d2011-11-25 00:22:05 -05001930 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00001931
Andi Kleenbaca7e42012-05-08 13:32:02 +09301932 br_write_unlock(&vfsmount_lock);
David Howellsea5b7782011-01-14 19:10:03 +00001933 up_write(&namespace_sem);
1934}
1935EXPORT_SYMBOL(mnt_set_expiry);
1936
1937/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 * process a list of expirable mountpoints with the intent of discarding any
1939 * mountpoints that aren't in use and haven't been touched since last we came
1940 * here
1941 */
1942void mark_mounts_for_expiry(struct list_head *mounts)
1943{
Al Viro761d5c32011-11-24 21:07:43 -05001944 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001946 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
1948 if (list_empty(mounts))
1949 return;
1950
Al Virobcc5c7d2008-03-22 00:21:53 -04001951 down_write(&namespace_sem);
Andi Kleenbaca7e42012-05-08 13:32:02 +09301952 br_write_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
1954 /* extract from the expiration list every vfsmount that matches the
1955 * following criteria:
1956 * - only referenced by its parent vfsmount
1957 * - still marked for expiry (marked on the last call here; marks are
1958 * cleared by mntput())
1959 */
Al Viro6776db3d2011-11-25 00:22:05 -05001960 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05001961 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05001962 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05001964 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001966 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05001967 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05001968 touch_mnt_namespace(mnt->mnt_ns);
Al Virobcc5c7d2008-03-22 00:21:53 -04001969 umount_tree(mnt, 1, &umounts);
1970 }
Andi Kleenbaca7e42012-05-08 13:32:02 +09301971 br_write_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001972 up_write(&namespace_sem);
1973
1974 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975}
1976
1977EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1978
1979/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001980 * Ripoff of 'select_parent()'
1981 *
1982 * search the list of submounts for a given mountpoint, and move any
1983 * shrinkable submounts to the 'graveyard' list.
1984 */
Al Viro692afc32011-11-24 21:15:14 -05001985static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001986{
Al Viro692afc32011-11-24 21:15:14 -05001987 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001988 struct list_head *next;
1989 int found = 0;
1990
1991repeat:
Al Viro6b41d532011-11-24 23:24:33 -05001992 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001993resume:
Al Viro6b41d532011-11-24 23:24:33 -05001994 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001995 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05001996 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001997
1998 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05001999 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002000 continue;
2001 /*
2002 * Descend a level if the d_mounts list is non-empty.
2003 */
Al Viro6b41d532011-11-24 23:24:33 -05002004 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002005 this_parent = mnt;
2006 goto repeat;
2007 }
2008
Al Viro1ab59732011-11-24 21:35:16 -05002009 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002010 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002011 found++;
2012 }
2013 }
2014 /*
2015 * All done at this level ... ascend and resume the search
2016 */
2017 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002018 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002019 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002020 goto resume;
2021 }
2022 return found;
2023}
2024
2025/*
2026 * process a list of expirable mountpoints with the intent of discarding any
2027 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002028 *
2029 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002030 */
Al Viro692afc32011-11-24 21:15:14 -05002031static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002032{
2033 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002034 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002035
Trond Myklebust5528f9112006-06-09 09:34:17 -04002036 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002037 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002038 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002039 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002040 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002041 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08002042 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04002043 }
2044 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002045}
2046
Trond Myklebust5528f9112006-06-09 09:34:17 -04002047/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 * Some copy_from_user() implementations do not return the exact number of
2049 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2050 * Note that this function differs from copy_from_user() in that it will oops
2051 * on bad values of `to', rather than returning a short copy.
2052 */
Ram Paib58fed82005-11-07 17:16:09 -05002053static long exact_copy_from_user(void *to, const void __user * from,
2054 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055{
2056 char *t = to;
2057 const char __user *f = from;
2058 char c;
2059
2060 if (!access_ok(VERIFY_READ, from, n))
2061 return n;
2062
2063 while (n) {
2064 if (__get_user(c, f)) {
2065 memset(t, 0, n);
2066 break;
2067 }
2068 *t++ = c;
2069 f++;
2070 n--;
2071 }
2072 return n;
2073}
2074
Ram Paib58fed82005-11-07 17:16:09 -05002075int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076{
2077 int i;
2078 unsigned long page;
2079 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002080
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 *where = 0;
2082 if (!data)
2083 return 0;
2084
2085 if (!(page = __get_free_page(GFP_KERNEL)))
2086 return -ENOMEM;
2087
2088 /* We only care that *some* data at the address the user
2089 * gave us is valid. Just in case, we'll zero
2090 * the remainder of the page.
2091 */
2092 /* copy_from_user cannot cross TASK_SIZE ! */
2093 size = TASK_SIZE - (unsigned long)data;
2094 if (size > PAGE_SIZE)
2095 size = PAGE_SIZE;
2096
2097 i = size - exact_copy_from_user((void *)page, data, size);
2098 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002099 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 return -EFAULT;
2101 }
2102 if (i != PAGE_SIZE)
2103 memset((char *)page + i, 0, PAGE_SIZE - i);
2104 *where = page;
2105 return 0;
2106}
2107
Vegard Nossumeca6f532009-09-18 13:05:45 -07002108int copy_mount_string(const void __user *data, char **where)
2109{
2110 char *tmp;
2111
2112 if (!data) {
2113 *where = NULL;
2114 return 0;
2115 }
2116
2117 tmp = strndup_user(data, PAGE_SIZE);
2118 if (IS_ERR(tmp))
2119 return PTR_ERR(tmp);
2120
2121 *where = tmp;
2122 return 0;
2123}
2124
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125/*
2126 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2127 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2128 *
2129 * data is a (void *) that can point to any structure up to
2130 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2131 * information (or be NULL).
2132 *
2133 * Pre-0.97 versions of mount() didn't have a flags word.
2134 * When the flags word was introduced its top half was required
2135 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2136 * Therefore, if this magic number is present, it carries no information
2137 * and must be discarded.
2138 */
Ram Paib58fed82005-11-07 17:16:09 -05002139long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 unsigned long flags, void *data_page)
2141{
Al Viro2d92ab32008-08-02 00:51:11 -04002142 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 int retval = 0;
2144 int mnt_flags = 0;
2145
2146 /* Discard magic */
2147 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2148 flags &= ~MS_MGC_MSK;
2149
2150 /* Basic sanity checks */
2151
2152 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2153 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154
2155 if (data_page)
2156 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2157
Tetsuo Handaa27ab9f2009-10-04 21:49:49 +09002158 /* ... and get the mountpoint */
2159 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2160 if (retval)
2161 return retval;
2162
2163 retval = security_sb_mount(dev_name, &path,
2164 type_page, flags, data_page);
2165 if (retval)
2166 goto dput_out;
2167
Andi Kleen613cbe32009-04-19 18:40:43 +02002168 /* Default to relatime unless overriden */
2169 if (!(flags & MS_NOATIME))
2170 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002171
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 /* Separate the per-mountpoint flags */
2173 if (flags & MS_NOSUID)
2174 mnt_flags |= MNT_NOSUID;
2175 if (flags & MS_NODEV)
2176 mnt_flags |= MNT_NODEV;
2177 if (flags & MS_NOEXEC)
2178 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002179 if (flags & MS_NOATIME)
2180 mnt_flags |= MNT_NOATIME;
2181 if (flags & MS_NODIRATIME)
2182 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002183 if (flags & MS_STRICTATIME)
2184 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002185 if (flags & MS_RDONLY)
2186 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002187
Al Viro7a4dec52010-08-09 12:05:43 -04002188 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002189 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2190 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002193 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 data_page);
2195 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002196 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002197 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002198 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002200 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 else
Al Viro2d92ab32008-08-02 00:51:11 -04002202 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 dev_name, data_page);
2204dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002205 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 return retval;
2207}
2208
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002209static struct mnt_namespace *alloc_mnt_ns(void)
2210{
2211 struct mnt_namespace *new_ns;
2212
2213 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2214 if (!new_ns)
2215 return ERR_PTR(-ENOMEM);
2216 atomic_set(&new_ns->count, 1);
2217 new_ns->root = NULL;
2218 INIT_LIST_HEAD(&new_ns->list);
2219 init_waitqueue_head(&new_ns->poll);
2220 new_ns->event = 0;
2221 return new_ns;
2222}
2223
JANAK DESAI741a2952006-02-07 12:59:00 -08002224/*
2225 * Allocate a new namespace structure and populate it with contents
2226 * copied from the namespace of the passed in task structure.
2227 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002228static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002229 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002231 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002232 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002233 struct mount *p, *q;
Al Virobe08d6d2011-12-06 13:32:36 -05002234 struct mount *old = mnt_ns->root;
Al Virocb338d02011-11-24 20:55:08 -05002235 struct mount *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002237 new_ns = alloc_mnt_ns();
2238 if (IS_ERR(new_ns))
2239 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240
Ram Pai390c6842005-11-07 17:17:51 -05002241 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 /* First pass: copy the tree topology */
Al Viro909b0a82011-11-25 03:06:56 -05002243 new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE);
David Howells210cff12012-06-25 12:55:18 +01002244 if (IS_ERR(new)) {
Ram Pai390c6842005-11-07 17:17:51 -05002245 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 kfree(new_ns);
David Howells210cff12012-06-25 12:55:18 +01002247 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 }
Al Virobe08d6d2011-12-06 13:32:36 -05002249 new_ns->root = new;
Andi Kleenbaca7e42012-05-08 13:32:02 +09302250 br_write_lock(&vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05002251 list_add_tail(&new_ns->list, &new->mnt_list);
Andi Kleenbaca7e42012-05-08 13:32:02 +09302252 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253
2254 /*
2255 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2256 * as belonging to new namespace. We have already acquired a private
2257 * fs_struct, so tsk->fs->lock is not needed.
2258 */
Al Viro909b0a82011-11-25 03:06:56 -05002259 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002260 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002262 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 if (fs) {
Al Viro315fc832011-11-24 18:57:30 -05002264 if (&p->mnt == fs->root.mnt) {
2265 fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002266 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 }
Al Viro315fc832011-11-24 18:57:30 -05002268 if (&p->mnt == fs->pwd.mnt) {
2269 fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002270 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 }
Al Viro909b0a82011-11-25 03:06:56 -05002273 p = next_mnt(p, old);
2274 q = next_mnt(q, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 }
Ram Pai390c6842005-11-07 17:17:51 -05002276 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002279 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002281 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
JANAK DESAI741a2952006-02-07 12:59:00 -08002283 return new_ns;
2284}
2285
Eric W. Biederman213dd262007-07-15 23:41:15 -07002286struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002287 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002288{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002289 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002290
Badari Pulavartye3222c42007-05-08 00:25:21 -07002291 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002292 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002293
2294 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002295 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002296
Badari Pulavartye3222c42007-05-08 00:25:21 -07002297 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002298
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002299 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002300 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301}
2302
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002303/**
2304 * create_mnt_ns - creates a private namespace and adds a root filesystem
2305 * @mnt: pointer to the new root filesystem mountpoint
2306 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002307static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002308{
Al Viro1a4eeaf2011-11-25 02:19:55 -05002309 struct mnt_namespace *new_ns = alloc_mnt_ns();
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002310 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002311 struct mount *mnt = real_mount(m);
2312 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002313 new_ns->root = mnt;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002314 list_add(&new_ns->list, &mnt->mnt_list);
Al Viroc1334492011-11-16 16:12:14 -05002315 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002316 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002317 }
2318 return new_ns;
2319}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002320
Al Viroea441d12011-11-16 21:43:59 -05002321struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2322{
2323 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002324 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002325 struct path path;
2326 int err;
2327
2328 ns = create_mnt_ns(mnt);
2329 if (IS_ERR(ns))
2330 return ERR_CAST(ns);
2331
2332 err = vfs_path_lookup(mnt->mnt_root, mnt,
2333 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2334
2335 put_mnt_ns(ns);
2336
2337 if (err)
2338 return ERR_PTR(err);
2339
2340 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002341 s = path.mnt->mnt_sb;
2342 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002343 mntput(path.mnt);
2344 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002345 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002346 /* ... and return the root of (sub)tree on it */
2347 return path.dentry;
2348}
2349EXPORT_SYMBOL(mount_subtree);
2350
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002351SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2352 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002354 int ret;
2355 char *kernel_type;
2356 char *kernel_dir;
2357 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
Vegard Nossumeca6f532009-09-18 13:05:45 -07002360 ret = copy_mount_string(type, &kernel_type);
2361 if (ret < 0)
2362 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363
Vegard Nossumeca6f532009-09-18 13:05:45 -07002364 kernel_dir = getname(dir_name);
2365 if (IS_ERR(kernel_dir)) {
2366 ret = PTR_ERR(kernel_dir);
2367 goto out_dir;
2368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
Vegard Nossumeca6f532009-09-18 13:05:45 -07002370 ret = copy_mount_string(dev_name, &kernel_dev);
2371 if (ret < 0)
2372 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
Vegard Nossumeca6f532009-09-18 13:05:45 -07002374 ret = copy_mount_options(data, &data_page);
2375 if (ret < 0)
2376 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
Vegard Nossumeca6f532009-09-18 13:05:45 -07002378 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2379 (void *) data_page);
2380
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002382out_data:
2383 kfree(kernel_dev);
2384out_dev:
2385 putname(kernel_dir);
2386out_dir:
2387 kfree(kernel_type);
2388out_type:
2389 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390}
2391
2392/*
Al Viroafac7cb2011-11-23 19:34:49 -05002393 * Return true if path is reachable from root
2394 *
2395 * namespace_sem or vfsmount_lock is held
2396 */
Al Viro643822b2011-11-24 22:00:28 -05002397bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002398 const struct path *root)
2399{
Al Viro643822b2011-11-24 22:00:28 -05002400 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002401 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002402 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002403 }
Al Viro643822b2011-11-24 22:00:28 -05002404 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002405}
2406
2407int path_is_under(struct path *path1, struct path *path2)
2408{
2409 int res;
Andi Kleenbaca7e42012-05-08 13:32:02 +09302410 br_read_lock(&vfsmount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002411 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Andi Kleenbaca7e42012-05-08 13:32:02 +09302412 br_read_unlock(&vfsmount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002413 return res;
2414}
2415EXPORT_SYMBOL(path_is_under);
2416
2417/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 * pivot_root Semantics:
2419 * Moves the root file system of the current process to the directory put_old,
2420 * makes new_root as the new root file system of the current process, and sets
2421 * root/cwd of all processes which had them on the current root to new_root.
2422 *
2423 * Restrictions:
2424 * The new_root and put_old must be directories, and must not be on the
2425 * same file system as the current process root. The put_old must be
2426 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2427 * pointed to by put_old must yield the same directory as new_root. No other
2428 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2429 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002430 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2431 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2432 * in this situation.
2433 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 * Notes:
2435 * - we don't move root/cwd if they are not at the root (reason: if something
2436 * cared enough to change them, it's probably wrong to force them elsewhere)
2437 * - it's okay to pick a root that isn't the root of a file system, e.g.
2438 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2439 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2440 * first.
2441 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002442SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2443 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444{
Al Viro2d8f3032008-07-22 09:59:21 -04002445 struct path new, old, parent_path, root_parent, root;
Al Viro419148d2011-11-24 19:41:16 -05002446 struct mount *new_mnt, *root_mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 int error;
2448
2449 if (!capable(CAP_SYS_ADMIN))
2450 return -EPERM;
2451
Al Viro2d8f3032008-07-22 09:59:21 -04002452 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 if (error)
2454 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455
Al Viro2d8f3032008-07-22 09:59:21 -04002456 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 if (error)
2458 goto out1;
2459
Al Viro2d8f3032008-07-22 09:59:21 -04002460 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002461 if (error)
2462 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002464 get_fs_root(current->fs, &root);
Al Virob12cea92011-03-18 08:55:38 -04002465 error = lock_mount(&old);
2466 if (error)
2467 goto out3;
2468
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002470 new_mnt = real_mount(new.mnt);
2471 root_mnt = real_mount(root.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002472 if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2473 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2474 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002475 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002476 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002477 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002479 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002480 goto out4;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002481 if (d_unlinked(old.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002482 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002484 if (new.mnt == root.mnt ||
2485 old.mnt == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002486 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002488 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002489 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002490 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002491 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002492 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002493 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002494 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002495 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002496 /* make sure we can reach put_old from new_root */
Al Viro643822b2011-11-24 22:00:28 -05002497 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002498 goto out4;
Eric W. Biederman87f5b922014-10-08 10:42:27 -07002499 /* make certain new is below the root */
2500 if (!is_path_reachable(new_mnt, new.dentry, &root))
2501 goto out4;
Andi Kleenbaca7e42012-05-08 13:32:02 +09302502 br_write_lock(&vfsmount_lock);
Al Viro419148d2011-11-24 19:41:16 -05002503 detach_mnt(new_mnt, &parent_path);
2504 detach_mnt(root_mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002505 /* mount old root on put_old */
Al Viro419148d2011-11-24 19:41:16 -05002506 attach_mnt(root_mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002507 /* mount new_root on / */
Al Viro419148d2011-11-24 19:41:16 -05002508 attach_mnt(new_mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002509 touch_mnt_namespace(current->nsproxy->mnt_ns);
Andi Kleenbaca7e42012-05-08 13:32:02 +09302510 br_write_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002511 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002513out4:
2514 unlock_mount(&old);
2515 if (!error) {
2516 path_put(&root_parent);
2517 path_put(&parent_path);
2518 }
2519out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002520 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002521out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002522 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002524 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527}
2528
2529static void __init init_mount_tree(void)
2530{
2531 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002532 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002533 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534
2535 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2536 if (IS_ERR(mnt))
2537 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002538
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002539 ns = create_mnt_ns(mnt);
2540 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002543 init_task.nsproxy->mnt_ns = ns;
2544 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Al Virobe08d6d2011-12-06 13:32:36 -05002546 root.mnt = mnt;
2547 root.dentry = mnt->mnt_root;
Jan Blunckac748a02008-02-14 19:34:39 -08002548
2549 set_fs_pwd(current->fs, &root);
2550 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551}
2552
Denis Cheng74bf17c2007-10-16 23:26:30 -07002553void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002555 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002556 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557
Ram Pai390c6842005-11-07 17:17:51 -05002558 init_rwsem(&namespace_sem);
2559
Al Viro7d6fec42011-11-23 12:14:10 -05002560 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002561 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
Ram Paib58fed82005-11-07 17:16:09 -05002563 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564
2565 if (!mount_hashtable)
2566 panic("Failed to allocate mount hash table\n");
2567
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002568 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569
Eric Dumazet13f14b42008-02-06 01:37:57 -08002570 for (u = 0; u < HASH_SIZE; u++)
2571 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572
Andi Kleenbaca7e42012-05-08 13:32:02 +09302573 br_lock_init(&vfsmount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +10002574
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002575 err = sysfs_init();
2576 if (err)
2577 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002578 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002579 fs_kobj = kobject_create_and_add("fs", NULL);
2580 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002581 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 init_rootfs();
2583 init_mount_tree();
2584}
2585
Trond Myklebust616511d2009-06-22 15:09:13 -04002586void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002588 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002589
Al Virod498b252010-02-05 02:21:06 -05002590 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002591 return;
Ram Pai390c6842005-11-07 17:17:51 -05002592 down_write(&namespace_sem);
Andi Kleenbaca7e42012-05-08 13:32:02 +09302593 br_write_lock(&vfsmount_lock);
Al Virobe08d6d2011-12-06 13:32:36 -05002594 umount_tree(ns->root, 0, &umount_list);
Andi Kleenbaca7e42012-05-08 13:32:02 +09302595 br_write_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002596 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002597 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002598 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599}
Al Viro9d412a42011-03-17 22:08:28 -04002600
2601struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2602{
Tim Chen423e0ab2011-07-19 09:32:38 -07002603 struct vfsmount *mnt;
2604 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2605 if (!IS_ERR(mnt)) {
2606 /*
2607 * it is a longterm mount, don't release mnt until
2608 * we unmount before file sys is unregistered
2609 */
Al Viro5296c2c2012-06-09 00:59:08 -04002610 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07002611 }
2612 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002613}
2614EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002615
2616void kern_unmount(struct vfsmount *mnt)
2617{
2618 /* release long term mount so mount point can be released */
2619 if (!IS_ERR_OR_NULL(mnt)) {
Al Viro5296c2c2012-06-09 00:59:08 -04002620 br_write_lock(&vfsmount_lock);
2621 real_mount(mnt)->mnt_ns = NULL;
2622 br_write_unlock(&vfsmount_lock);
Tim Chen423e0ab2011-07-19 09:32:38 -07002623 mntput(mnt);
2624 }
2625}
2626EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002627
2628bool our_mnt(struct vfsmount *mnt)
2629{
Al Viro143c8c92011-11-25 00:46:35 -05002630 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05002631}