blob: 22e536705c45d128ab8651759cea74d83f481ab1 [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>
Eric W. Biederman771b1372012-07-26 21:08:32 -070015#include <linux/user_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/namei.h>
17#include <linux/security.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010018#include <linux/idr.h>
Al Virod10577a2011-12-07 13:06:11 -050019#include <linux/acct.h> /* acct_auto_close_mnt */
Rob Landley57f150a2013-09-11 14:26:10 -070020#include <linux/init.h> /* init_rootfs */
Al Virod10577a2011-12-07 13:06:11 -050021#include <linux/fs_struct.h> /* get_fs_root et.al. */
22#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23#include <linux/uaccess.h>
David Howells0bb80f22013-04-12 01:50:06 +010024#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070025#include <linux/magic.h>
Ram Pai07b20882005-11-07 17:19:07 -050026#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070027#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
Eric Dumazet13f14b42008-02-06 01:37:57 -080029#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
30#define HASH_SIZE (1UL << HASH_SHIFT)
31
Al Viro5addc5d2005-11-07 17:15:49 -050032static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010033static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010034static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100035static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040036static int mnt_id_start = 0;
37static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Eric Dumazetfa3536c2006-03-26 01:37:24 -080039static struct list_head *mount_hashtable __read_mostly;
Al Viro84d17192013-03-15 10:53:28 -040040static struct list_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080041static struct kmem_cache *mnt_cache __read_mostly;
Al Viro59aa0da2013-09-16 21:34:53 -040042static DECLARE_RWSEM(namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080044/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060045struct kobject *fs_kobj;
46EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080047
Nick Piggin99b7db72010-08-18 04:37:39 +100048/*
49 * vfsmount lock may be taken for read to prevent changes to the
50 * vfsmount hash, ie. during mountpoint lookups or walking back
51 * up the tree.
52 *
53 * It should be taken for write in all cases where the vfsmount
54 * tree or hash is modified or when a vfsmount structure is modified.
55 */
Al Viro48a066e2013-09-29 22:06:07 -040056__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +100057
Linus Torvalds1da177e2005-04-16 15:20:36 -070058static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
59{
Ram Paib58fed82005-11-07 17:16:09 -050060 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
61 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080062 tmp = tmp + (tmp >> HASH_SHIFT);
63 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064}
65
Nick Piggin99b7db72010-08-18 04:37:39 +100066/*
67 * allocation is serialized by namespace_sem, but we need the spinlock to
68 * serialize with freeing.
69 */
Al Virob105e272011-11-24 20:38:33 -050070static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010071{
72 int res;
73
74retry:
75 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +100076 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -050077 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -040078 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -050079 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +100080 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010081 if (res == -EAGAIN)
82 goto retry;
83
84 return res;
85}
86
Al Virob105e272011-11-24 20:38:33 -050087static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010088{
Al Viro15169fe2011-11-25 00:50:41 -050089 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +100090 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040091 ida_remove(&mnt_id_ida, id);
92 if (mnt_id_start > id)
93 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +100094 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010095}
96
Miklos Szeredi719f5d72008-03-27 13:06:23 +010097/*
98 * Allocate a new peer group ID
99 *
100 * mnt_group_ida is protected by namespace_sem
101 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500102static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100103{
Al Virof21f6222009-06-24 03:12:00 -0400104 int res;
105
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100106 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
107 return -ENOMEM;
108
Al Virof21f6222009-06-24 03:12:00 -0400109 res = ida_get_new_above(&mnt_group_ida,
110 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500111 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400112 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500113 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400114
115 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100116}
117
118/*
119 * Release a peer group ID
120 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500121void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100122{
Al Viro15169fe2011-11-25 00:50:41 -0500123 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400124 ida_remove(&mnt_group_ida, id);
125 if (mnt_group_start > id)
126 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500127 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100128}
129
Nick Pigginb3e19d92011-01-07 17:50:11 +1100130/*
131 * vfsmount lock must be held for read
132 */
Al Viro83adc752011-11-24 22:37:54 -0500133static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100134{
135#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500136 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100137#else
138 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500139 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100140 preempt_enable();
141#endif
142}
143
Nick Pigginb3e19d92011-01-07 17:50:11 +1100144/*
145 * vfsmount lock must be held for write
146 */
Al Viro83adc752011-11-24 22:37:54 -0500147unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100148{
149#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500150 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100151 int cpu;
152
153 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500154 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100155 }
156
157 return count;
158#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500159 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100160#endif
161}
162
Al Virob105e272011-11-24 20:38:33 -0500163static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
Al Viroc63181e2011-11-25 02:35:16 -0500165 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
166 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100167 int err;
168
Al Viroc63181e2011-11-25 02:35:16 -0500169 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800170 if (err)
171 goto out_free_cache;
172
173 if (name) {
Al Viroc63181e2011-11-25 02:35:16 -0500174 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
175 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800176 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100177 }
178
Nick Pigginb3e19d92011-01-07 17:50:11 +1100179#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500180 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
181 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100182 goto out_free_devname;
183
Al Viroc63181e2011-11-25 02:35:16 -0500184 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100185#else
Al Viroc63181e2011-11-25 02:35:16 -0500186 mnt->mnt_count = 1;
187 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100188#endif
189
Al Viroc63181e2011-11-25 02:35:16 -0500190 INIT_LIST_HEAD(&mnt->mnt_hash);
191 INIT_LIST_HEAD(&mnt->mnt_child);
192 INIT_LIST_HEAD(&mnt->mnt_mounts);
193 INIT_LIST_HEAD(&mnt->mnt_list);
194 INIT_LIST_HEAD(&mnt->mnt_expire);
195 INIT_LIST_HEAD(&mnt->mnt_share);
196 INIT_LIST_HEAD(&mnt->mnt_slave_list);
197 INIT_LIST_HEAD(&mnt->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500198#ifdef CONFIG_FSNOTIFY
199 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
200#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 }
Al Viroc63181e2011-11-25 02:35:16 -0500202 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800203
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000204#ifdef CONFIG_SMP
205out_free_devname:
Al Viroc63181e2011-11-25 02:35:16 -0500206 kfree(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000207#endif
Li Zefan88b38782008-07-21 18:06:36 +0800208out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500209 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800210out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500211 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800212 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
Dave Hansen83660252008-02-15 14:37:30 -0800215/*
216 * Most r/o checks on a fs are for operations that take
217 * discrete amounts of time, like a write() or unlink().
218 * We must keep track of when those operations start
219 * (for permission checks) and when they end, so that
220 * we can determine when writes are able to occur to
221 * a filesystem.
222 */
Dave Hansen3d733632008-02-15 14:37:59 -0800223/*
224 * __mnt_is_readonly: check whether a mount is read-only
225 * @mnt: the mount to check for its write status
226 *
227 * This shouldn't be used directly ouside of the VFS.
228 * It does not guarantee that the filesystem will stay
229 * r/w, just that it is right *now*. This can not and
230 * should not be used in place of IS_RDONLY(inode).
231 * mnt_want/drop_write() will _keep_ the filesystem
232 * r/w.
233 */
234int __mnt_is_readonly(struct vfsmount *mnt)
235{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800236 if (mnt->mnt_flags & MNT_READONLY)
237 return 1;
238 if (mnt->mnt_sb->s_flags & MS_RDONLY)
239 return 1;
240 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800241}
242EXPORT_SYMBOL_GPL(__mnt_is_readonly);
243
Al Viro83adc752011-11-24 22:37:54 -0500244static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800245{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000246#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500247 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000248#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500249 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000250#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800251}
Dave Hansen3d733632008-02-15 14:37:59 -0800252
Al Viro83adc752011-11-24 22:37:54 -0500253static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800254{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000255#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500256 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000257#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500258 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000259#endif
260}
261
Al Viro83adc752011-11-24 22:37:54 -0500262static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000263{
264#ifdef CONFIG_SMP
265 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800266 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800267
268 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500269 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800270 }
Dave Hansen3d733632008-02-15 14:37:59 -0800271
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000272 return count;
273#else
274 return mnt->mnt_writers;
275#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800276}
277
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100278static int mnt_is_readonly(struct vfsmount *mnt)
279{
280 if (mnt->mnt_sb->s_readonly_remount)
281 return 1;
282 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
283 smp_rmb();
284 return __mnt_is_readonly(mnt);
285}
286
Dave Hansen3d733632008-02-15 14:37:59 -0800287/*
Jan Karaeb04c282012-06-12 16:20:35 +0200288 * Most r/o & frozen checks on a fs are for operations that take discrete
289 * amounts of time, like a write() or unlink(). We must keep track of when
290 * those operations start (for permission checks) and when they end, so that we
291 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800292 */
Dave Hansen83660252008-02-15 14:37:30 -0800293/**
Jan Karaeb04c282012-06-12 16:20:35 +0200294 * __mnt_want_write - get write access to a mount without freeze protection
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 *
Jan Karaeb04c282012-06-12 16:20:35 +0200297 * This tells the low-level filesystem that a write is about to be performed to
298 * it, and makes sure that writes are allowed (mnt it read-write) before
299 * returning success. This operation does not protect against filesystem being
300 * frozen. When the write operation is finished, __mnt_drop_write() must be
301 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800302 */
Jan Karaeb04c282012-06-12 16:20:35 +0200303int __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();
Miao Xie1e755292012-11-16 17:23:50 +0800316 while (ACCESS_ONCE(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();
Jan Karaeb04c282012-06-12 16:20:35 +0200329
330 return ret;
331}
332
333/**
334 * mnt_want_write - get write access to a mount
335 * @m: the mount on which to take a write
336 *
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mount is read-write, filesystem
339 * is not frozen) before returning success. When the write operation is
340 * finished, mnt_drop_write() must be called. This is effectively a refcount.
341 */
342int mnt_want_write(struct vfsmount *m)
343{
344 int ret;
345
346 sb_start_write(m->mnt_sb);
347 ret = __mnt_want_write(m);
348 if (ret)
349 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800350 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800351}
352EXPORT_SYMBOL_GPL(mnt_want_write);
353
354/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000355 * mnt_clone_write - get write access to a mount
356 * @mnt: the mount on which to take a write
357 *
358 * This is effectively like mnt_want_write, except
359 * it must only be used to take an extra write reference
360 * on a mountpoint that we already know has a write reference
361 * on it. This allows some optimisation.
362 *
363 * After finished, mnt_drop_write must be called as usual to
364 * drop the reference.
365 */
366int mnt_clone_write(struct vfsmount *mnt)
367{
368 /* superblock may be r/o */
369 if (__mnt_is_readonly(mnt))
370 return -EROFS;
371 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500372 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000373 preempt_enable();
374 return 0;
375}
376EXPORT_SYMBOL_GPL(mnt_clone_write);
377
378/**
Jan Karaeb04c282012-06-12 16:20:35 +0200379 * __mnt_want_write_file - get write access to a file's mount
380 * @file: the file who's mount on which to take a write
381 *
382 * This is like __mnt_want_write, but it takes a file and can
383 * do some optimisations if the file is open for write already
384 */
385int __mnt_want_write_file(struct file *file)
386{
Al Viro496ad9a2013-01-23 17:07:38 -0500387 struct inode *inode = file_inode(file);
Jan Karaeb04c282012-06-12 16:20:35 +0200388
389 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
390 return __mnt_want_write(file->f_path.mnt);
391 else
392 return mnt_clone_write(file->f_path.mnt);
393}
394
395/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000396 * mnt_want_write_file - get write access to a file's mount
397 * @file: the file who's mount on which to take a write
398 *
399 * This is like mnt_want_write, but it takes a file and can
400 * do some optimisations if the file is open for write already
401 */
402int mnt_want_write_file(struct file *file)
403{
Jan Karaeb04c282012-06-12 16:20:35 +0200404 int ret;
405
406 sb_start_write(file->f_path.mnt->mnt_sb);
407 ret = __mnt_want_write_file(file);
408 if (ret)
409 sb_end_write(file->f_path.mnt->mnt_sb);
410 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000411}
412EXPORT_SYMBOL_GPL(mnt_want_write_file);
413
414/**
Jan Karaeb04c282012-06-12 16:20:35 +0200415 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800416 * @mnt: the mount on which to give up write access
417 *
418 * Tells the low-level filesystem that we are done
419 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200420 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800421 */
Jan Karaeb04c282012-06-12 16:20:35 +0200422void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800423{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000424 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500425 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000426 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800427}
Jan Karaeb04c282012-06-12 16:20:35 +0200428
429/**
430 * mnt_drop_write - give up write access to a mount
431 * @mnt: the mount on which to give up write access
432 *
433 * Tells the low-level filesystem that we are done performing writes to it and
434 * also allows filesystem to be frozen again. Must be matched with
435 * mnt_want_write() call above.
436 */
437void mnt_drop_write(struct vfsmount *mnt)
438{
439 __mnt_drop_write(mnt);
440 sb_end_write(mnt->mnt_sb);
441}
Dave Hansen83660252008-02-15 14:37:30 -0800442EXPORT_SYMBOL_GPL(mnt_drop_write);
443
Jan Karaeb04c282012-06-12 16:20:35 +0200444void __mnt_drop_write_file(struct file *file)
445{
446 __mnt_drop_write(file->f_path.mnt);
447}
448
Al Viro2a79f172011-12-09 08:06:57 -0500449void mnt_drop_write_file(struct file *file)
450{
451 mnt_drop_write(file->f_path.mnt);
452}
453EXPORT_SYMBOL(mnt_drop_write_file);
454
Al Viro83adc752011-11-24 22:37:54 -0500455static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800456{
Dave Hansen3d733632008-02-15 14:37:59 -0800457 int ret = 0;
458
Al Viro719ea2f2013-09-29 11:24:49 -0400459 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500460 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000461 /*
462 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
463 * should be visible before we do.
464 */
465 smp_mb();
466
467 /*
468 * With writers on hold, if this value is zero, then there are
469 * definitely no active writers (although held writers may subsequently
470 * increment the count, they'll have to wait, and decrement it after
471 * seeing MNT_READONLY).
472 *
473 * It is OK to have counter incremented on one CPU and decremented on
474 * another: the sum will add up correctly. The danger would be when we
475 * sum up each counter, if we read a counter before it is incremented,
476 * but then read another CPU's count which it has been subsequently
477 * decremented from -- we would see more decrements than we should.
478 * MNT_WRITE_HOLD protects against this scenario, because
479 * mnt_want_write first increments count, then smp_mb, then spins on
480 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
481 * we're counting up here.
482 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100483 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000484 ret = -EBUSY;
485 else
Al Viro83adc752011-11-24 22:37:54 -0500486 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000487 /*
488 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
489 * that become unheld will see MNT_READONLY.
490 */
491 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500492 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Al Viro719ea2f2013-09-29 11:24:49 -0400493 unlock_mount_hash();
Dave Hansen3d733632008-02-15 14:37:59 -0800494 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800495}
Dave Hansen83660252008-02-15 14:37:30 -0800496
Al Viro83adc752011-11-24 22:37:54 -0500497static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800498{
Al Viro719ea2f2013-09-29 11:24:49 -0400499 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500500 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Al Viro719ea2f2013-09-29 11:24:49 -0400501 unlock_mount_hash();
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800502}
503
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100504int sb_prepare_remount_readonly(struct super_block *sb)
505{
506 struct mount *mnt;
507 int err = 0;
508
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100509 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
510 if (atomic_long_read(&sb->s_remove_count))
511 return -EBUSY;
512
Al Viro719ea2f2013-09-29 11:24:49 -0400513 lock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100514 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
515 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
516 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
517 smp_mb();
518 if (mnt_get_writers(mnt) > 0) {
519 err = -EBUSY;
520 break;
521 }
522 }
523 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100524 if (!err && atomic_long_read(&sb->s_remove_count))
525 err = -EBUSY;
526
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100527 if (!err) {
528 sb->s_readonly_remount = 1;
529 smp_wmb();
530 }
531 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
532 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
533 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
534 }
Al Viro719ea2f2013-09-29 11:24:49 -0400535 unlock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100536
537 return err;
538}
539
Al Virob105e272011-11-24 20:38:33 -0500540static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541{
Al Viro52ba1622011-11-25 02:25:17 -0500542 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100543 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000544#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500545 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000546#endif
Al Virob105e272011-11-24 20:38:33 -0500547 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548}
549
Al Viro48a066e2013-09-29 22:06:07 -0400550/* call under rcu_read_lock */
551bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
552{
553 struct mount *mnt;
554 if (read_seqretry(&mount_lock, seq))
555 return false;
556 if (bastard == NULL)
557 return true;
558 mnt = real_mount(bastard);
559 mnt_add_count(mnt, 1);
560 if (likely(!read_seqretry(&mount_lock, seq)))
561 return true;
562 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
563 mnt_add_count(mnt, -1);
564 return false;
565 }
566 rcu_read_unlock();
567 mntput(bastard);
568 rcu_read_lock();
569 return false;
570}
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572/*
Al Viro474279d2013-10-01 16:11:26 -0400573 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400574 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 */
Al Viro474279d2013-10-01 16:11:26 -0400576struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
Ram Paib58fed82005-11-07 17:16:09 -0500578 struct list_head *head = mount_hashtable + hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400579 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
Al Viro48a066e2013-09-29 22:06:07 -0400581 list_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400582 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
583 return p;
584 return NULL;
585}
586
587/*
588 * find the last mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400589 * mount_lock must be held.
Al Viro474279d2013-10-01 16:11:26 -0400590 */
591struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
592{
593 struct list_head *head = mount_hashtable + hash(mnt, dentry);
594 struct mount *p;
595
596 list_for_each_entry_reverse(p, head, mnt_hash)
597 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
598 return p;
599 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}
601
Ram Paia05964f2005-11-07 17:20:17 -0500602/*
David Howellsf015f1262012-06-25 12:55:28 +0100603 * lookup_mnt - Return the first child mount mounted at path
604 *
605 * "First" means first mounted chronologically. If you create the
606 * following mounts:
607 *
608 * mount /dev/sda1 /mnt
609 * mount /dev/sda2 /mnt
610 * mount /dev/sda3 /mnt
611 *
612 * Then lookup_mnt() on the base /mnt dentry in the root mount will
613 * return successively the root dentry and vfsmount of /dev/sda1, then
614 * /dev/sda2, then /dev/sda3, then NULL.
615 *
616 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500617 */
Al Viro1c755af2009-04-18 14:06:57 -0400618struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500619{
Al Viroc7105362011-11-24 18:22:03 -0500620 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400621 struct vfsmount *m;
622 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000623
Al Viro48a066e2013-09-29 22:06:07 -0400624 rcu_read_lock();
625 do {
626 seq = read_seqbegin(&mount_lock);
627 child_mnt = __lookup_mnt(path->mnt, path->dentry);
628 m = child_mnt ? &child_mnt->mnt : NULL;
629 } while (!legitimize_mnt(m, seq));
630 rcu_read_unlock();
631 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500632}
633
Al Viro84d17192013-03-15 10:53:28 -0400634static struct mountpoint *new_mountpoint(struct dentry *dentry)
635{
636 struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry);
637 struct mountpoint *mp;
Miklos Szeredieed81002013-09-05 14:39:11 +0200638 int ret;
Al Viro84d17192013-03-15 10:53:28 -0400639
640 list_for_each_entry(mp, chain, m_hash) {
641 if (mp->m_dentry == dentry) {
642 /* might be worth a WARN_ON() */
643 if (d_unlinked(dentry))
644 return ERR_PTR(-ENOENT);
645 mp->m_count++;
646 return mp;
647 }
648 }
649
650 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
651 if (!mp)
652 return ERR_PTR(-ENOMEM);
653
Miklos Szeredieed81002013-09-05 14:39:11 +0200654 ret = d_set_mounted(dentry);
655 if (ret) {
Al Viro84d17192013-03-15 10:53:28 -0400656 kfree(mp);
Miklos Szeredieed81002013-09-05 14:39:11 +0200657 return ERR_PTR(ret);
Al Viro84d17192013-03-15 10:53:28 -0400658 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200659
Al Viro84d17192013-03-15 10:53:28 -0400660 mp->m_dentry = dentry;
661 mp->m_count = 1;
662 list_add(&mp->m_hash, chain);
663 return mp;
664}
665
666static void put_mountpoint(struct mountpoint *mp)
667{
668 if (!--mp->m_count) {
669 struct dentry *dentry = mp->m_dentry;
670 spin_lock(&dentry->d_lock);
671 dentry->d_flags &= ~DCACHE_MOUNTED;
672 spin_unlock(&dentry->d_lock);
673 list_del(&mp->m_hash);
674 kfree(mp);
675 }
676}
677
Al Viro143c8c92011-11-25 00:46:35 -0500678static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800680 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
Nick Piggin99b7db72010-08-18 04:37:39 +1000683/*
684 * vfsmount lock must be held for write
685 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800686static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500687{
688 if (ns) {
689 ns->event = ++event;
690 wake_up_interruptible(&ns->poll);
691 }
692}
693
Nick Piggin99b7db72010-08-18 04:37:39 +1000694/*
695 * vfsmount lock must be held for write
696 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800697static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500698{
699 if (ns && ns->event != event) {
700 ns->event = event;
701 wake_up_interruptible(&ns->poll);
702 }
703}
704
Nick Piggin99b7db72010-08-18 04:37:39 +1000705/*
706 * vfsmount lock must be held for write
707 */
Al Viro419148d2011-11-24 19:41:16 -0500708static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709{
Al Viroa73324d2011-11-24 22:25:07 -0500710 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500711 old_path->mnt = &mnt->mnt_parent->mnt;
712 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500713 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500714 list_del_init(&mnt->mnt_child);
Al Viro1b8e5562011-11-24 21:01:32 -0500715 list_del_init(&mnt->mnt_hash);
Al Viro84d17192013-03-15 10:53:28 -0400716 put_mountpoint(mnt->mnt_mp);
717 mnt->mnt_mp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718}
719
Nick Piggin99b7db72010-08-18 04:37:39 +1000720/*
721 * vfsmount lock must be held for write
722 */
Al Viro84d17192013-03-15 10:53:28 -0400723void mnt_set_mountpoint(struct mount *mnt,
724 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500725 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500726{
Al Viro84d17192013-03-15 10:53:28 -0400727 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500728 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro84d17192013-03-15 10:53:28 -0400729 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500730 child_mnt->mnt_parent = mnt;
Al Viro84d17192013-03-15 10:53:28 -0400731 child_mnt->mnt_mp = mp;
Ram Paib90fa9a2005-11-07 17:19:50 -0500732}
733
Nick Piggin99b7db72010-08-18 04:37:39 +1000734/*
735 * vfsmount lock must be held for write
736 */
Al Viro84d17192013-03-15 10:53:28 -0400737static void attach_mnt(struct mount *mnt,
738 struct mount *parent,
739 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740{
Al Viro84d17192013-03-15 10:53:28 -0400741 mnt_set_mountpoint(parent, mp, mnt);
Al Viro1b8e5562011-11-24 21:01:32 -0500742 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro84d17192013-03-15 10:53:28 -0400743 hash(&parent->mnt, mp->m_dentry));
744 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500745}
746
747/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000748 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500749 */
Al Viro4b2619a2011-11-24 19:47:15 -0500750static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500751{
Al Viro0714a532011-11-24 22:19:58 -0500752 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500753 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500754 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500755 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500756
Al Viro0714a532011-11-24 22:19:58 -0500757 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500758
Al Viro1a4eeaf2011-11-25 02:19:55 -0500759 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400760 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500761 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500762
Ram Paib90fa9a2005-11-07 17:19:50 -0500763 list_splice(&head, n->list.prev);
764
Al Viro1b8e5562011-11-24 21:01:32 -0500765 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viroa73324d2011-11-24 22:25:07 -0500766 hash(&parent->mnt, mnt->mnt_mountpoint));
Al Viro6b41d532011-11-24 23:24:33 -0500767 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800768 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769}
770
Al Viro909b0a82011-11-25 03:06:56 -0500771static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
Al Viro6b41d532011-11-24 23:24:33 -0500773 struct list_head *next = p->mnt_mounts.next;
774 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500776 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500778 next = p->mnt_child.next;
779 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 break;
Al Viro0714a532011-11-24 22:19:58 -0500781 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
783 }
Al Viro6b41d532011-11-24 23:24:33 -0500784 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785}
786
Al Viro315fc832011-11-24 18:57:30 -0500787static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500788{
Al Viro6b41d532011-11-24 23:24:33 -0500789 struct list_head *prev = p->mnt_mounts.prev;
790 while (prev != &p->mnt_mounts) {
791 p = list_entry(prev, struct mount, mnt_child);
792 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500793 }
794 return p;
795}
796
Al Viro9d412a42011-03-17 22:08:28 -0400797struct vfsmount *
798vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
799{
Al Virob105e272011-11-24 20:38:33 -0500800 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400801 struct dentry *root;
802
803 if (!type)
804 return ERR_PTR(-ENODEV);
805
806 mnt = alloc_vfsmnt(name);
807 if (!mnt)
808 return ERR_PTR(-ENOMEM);
809
810 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500811 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400812
813 root = mount_fs(type, flags, name, data);
814 if (IS_ERR(root)) {
815 free_vfsmnt(mnt);
816 return ERR_CAST(root);
817 }
818
Al Virob105e272011-11-24 20:38:33 -0500819 mnt->mnt.mnt_root = root;
820 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500821 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500822 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400823 lock_mount_hash();
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100824 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400825 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -0500826 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400827}
828EXPORT_SYMBOL_GPL(vfs_kern_mount);
829
Al Viro87129cc2011-11-24 21:24:27 -0500830static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500831 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
Al Viro87129cc2011-11-24 21:24:27 -0500833 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100834 struct mount *mnt;
835 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
David Howellsbe34d1a2012-06-25 12:55:18 +0100837 mnt = alloc_vfsmnt(old->mnt_devname);
838 if (!mnt)
839 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100840
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700841 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +0100842 mnt->mnt_group_id = 0; /* not a peer of original */
843 else
844 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100845
David Howellsbe34d1a2012-06-25 12:55:18 +0100846 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
847 err = mnt_alloc_group_id(mnt);
848 if (err)
849 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100851
852 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700853 /* Don't allow unprivileged users to change mount flags */
854 if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
855 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
856
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -0700857 /* Don't allow unprivileged users to reveal what is under a mount */
858 if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
859 mnt->mnt.mnt_flags |= MNT_LOCKED;
860
David Howellsbe34d1a2012-06-25 12:55:18 +0100861 atomic_inc(&sb->s_active);
862 mnt->mnt.mnt_sb = sb;
863 mnt->mnt.mnt_root = dget(root);
864 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
865 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400866 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +0100867 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400868 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +0100869
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700870 if ((flag & CL_SLAVE) ||
871 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +0100872 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
873 mnt->mnt_master = old;
874 CLEAR_MNT_SHARED(mnt);
875 } else if (!(flag & CL_PRIVATE)) {
876 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
877 list_add(&mnt->mnt_share, &old->mnt_share);
878 if (IS_MNT_SLAVE(old))
879 list_add(&mnt->mnt_slave, &old->mnt_slave);
880 mnt->mnt_master = old->mnt_master;
881 }
882 if (flag & CL_MAKE_SHARED)
883 set_mnt_shared(mnt);
884
885 /* stick the duplicate mount on the same expiry list
886 * as the original if that was on one */
887 if (flag & CL_EXPIRE) {
888 if (!list_empty(&old->mnt_expire))
889 list_add(&mnt->mnt_expire, &old->mnt_expire);
890 }
891
Al Virocb338d02011-11-24 20:55:08 -0500892 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100893
894 out_free:
895 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +0100896 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
Al Viro48a066e2013-09-29 22:06:07 -0400899static void delayed_free(struct rcu_head *head)
900{
901 struct mount *mnt = container_of(head, struct mount, mnt_rcu);
902 kfree(mnt->mnt_devname);
903#ifdef CONFIG_SMP
904 free_percpu(mnt->mnt_pcp);
905#endif
906 kmem_cache_free(mnt_cache, mnt);
907}
908
Al Viro900148d2011-11-25 00:33:11 -0500909static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500910{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100911put_again:
Al Viro48a066e2013-09-29 22:06:07 -0400912 rcu_read_lock();
913 mnt_add_count(mnt, -1);
914 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
915 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -0500916 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100917 }
Al Viro719ea2f2013-09-29 11:24:49 -0400918 lock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +1100919 if (mnt_get_count(mnt)) {
Al Viro48a066e2013-09-29 22:06:07 -0400920 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -0400921 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +1100922 return;
923 }
Al Viro863d6842011-11-25 00:57:42 -0500924 if (unlikely(mnt->mnt_pinned)) {
925 mnt_add_count(mnt, mnt->mnt_pinned + 1);
926 mnt->mnt_pinned = 0;
Al Viro48a066e2013-09-29 22:06:07 -0400927 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -0400928 unlock_mount_hash();
Al Viro900148d2011-11-25 00:33:11 -0500929 acct_auto_close_mnt(&mnt->mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100930 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500931 }
Al Viro48a066e2013-09-29 22:06:07 -0400932 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
933 rcu_read_unlock();
934 unlock_mount_hash();
935 return;
936 }
937 mnt->mnt.mnt_flags |= MNT_DOOMED;
938 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +0930939
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100940 list_del(&mnt->mnt_instance);
Al Viro719ea2f2013-09-29 11:24:49 -0400941 unlock_mount_hash();
Al Viro649a7952013-09-28 12:41:25 -0400942
943 /*
944 * This probably indicates that somebody messed
945 * up a mnt_want/drop_write() pair. If this
946 * happens, the filesystem was probably unable
947 * to make r/w->r/o transitions.
948 */
949 /*
950 * The locking used to deal with mnt_count decrement provides barriers,
951 * so mnt_get_writers() below is safe.
952 */
953 WARN_ON(mnt_get_writers(mnt));
954 fsnotify_vfsmount_delete(&mnt->mnt);
955 dput(mnt->mnt.mnt_root);
956 deactivate_super(mnt->mnt.mnt_sb);
Al Viro48a066e2013-09-29 22:06:07 -0400957 mnt_free_id(mnt);
958 call_rcu(&mnt->mnt_rcu, delayed_free);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500959}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100960
961void mntput(struct vfsmount *mnt)
962{
963 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -0500964 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100965 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -0500966 if (unlikely(m->mnt_expiry_mark))
967 m->mnt_expiry_mark = 0;
968 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100969 }
970}
971EXPORT_SYMBOL(mntput);
972
973struct vfsmount *mntget(struct vfsmount *mnt)
974{
975 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -0500976 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100977 return mnt;
978}
979EXPORT_SYMBOL(mntget);
980
Al Viro7b7b1ac2005-11-07 17:13:39 -0500981void mnt_pin(struct vfsmount *mnt)
982{
Al Viro719ea2f2013-09-29 11:24:49 -0400983 lock_mount_hash();
Al Viro863d6842011-11-25 00:57:42 -0500984 real_mount(mnt)->mnt_pinned++;
Al Viro719ea2f2013-09-29 11:24:49 -0400985 unlock_mount_hash();
Al Viro7b7b1ac2005-11-07 17:13:39 -0500986}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500987EXPORT_SYMBOL(mnt_pin);
988
Al Viro863d6842011-11-25 00:57:42 -0500989void mnt_unpin(struct vfsmount *m)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500990{
Al Viro863d6842011-11-25 00:57:42 -0500991 struct mount *mnt = real_mount(m);
Al Viro719ea2f2013-09-29 11:24:49 -0400992 lock_mount_hash();
Al Viro7b7b1ac2005-11-07 17:13:39 -0500993 if (mnt->mnt_pinned) {
Al Viro863d6842011-11-25 00:57:42 -0500994 mnt_add_count(mnt, 1);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500995 mnt->mnt_pinned--;
996 }
Al Viro719ea2f2013-09-29 11:24:49 -0400997 unlock_mount_hash();
Al Viro7b7b1ac2005-11-07 17:13:39 -0500998}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500999EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001001static inline void mangle(struct seq_file *m, const char *s)
1002{
1003 seq_escape(m, s, " \t\n\\");
1004}
1005
1006/*
1007 * Simple .show_options callback for filesystems which don't want to
1008 * implement more complex mount option showing.
1009 *
1010 * See also save_mount_options().
1011 */
Al Viro34c80b12011-12-08 21:32:45 -05001012int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001013{
Al Viro2a32ceb2009-05-08 16:05:57 -04001014 const char *options;
1015
1016 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -05001017 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001018
1019 if (options != NULL && options[0]) {
1020 seq_putc(m, ',');
1021 mangle(m, options);
1022 }
Al Viro2a32ceb2009-05-08 16:05:57 -04001023 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001024
1025 return 0;
1026}
1027EXPORT_SYMBOL(generic_show_options);
1028
1029/*
1030 * If filesystem uses generic_show_options(), this function should be
1031 * called from the fill_super() callback.
1032 *
1033 * The .remount_fs callback usually needs to be handled in a special
1034 * way, to make sure, that previous options are not overwritten if the
1035 * remount fails.
1036 *
1037 * Also note, that if the filesystem's .remount_fs function doesn't
1038 * reset all options to their default value, but changes only newly
1039 * given options, then the displayed options will not reflect reality
1040 * any more.
1041 */
1042void save_mount_options(struct super_block *sb, char *options)
1043{
Al Viro2a32ceb2009-05-08 16:05:57 -04001044 BUG_ON(sb->s_options);
1045 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001046}
1047EXPORT_SYMBOL(save_mount_options);
1048
Al Viro2a32ceb2009-05-08 16:05:57 -04001049void replace_mount_options(struct super_block *sb, char *options)
1050{
1051 char *old = sb->s_options;
1052 rcu_assign_pointer(sb->s_options, options);
1053 if (old) {
1054 synchronize_rcu();
1055 kfree(old);
1056 }
1057}
1058EXPORT_SYMBOL(replace_mount_options);
1059
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001060#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001061/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062static void *m_start(struct seq_file *m, loff_t *pos)
1063{
Al Viro6ce6e242012-06-09 01:16:59 -04001064 struct proc_mounts *p = proc_mounts(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
Ram Pai390c6842005-11-07 17:17:51 -05001066 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001067 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068}
1069
1070static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1071{
Al Viro6ce6e242012-06-09 01:16:59 -04001072 struct proc_mounts *p = proc_mounts(m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001073
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001074 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075}
1076
1077static void m_stop(struct seq_file *m, void *v)
1078{
Ram Pai390c6842005-11-07 17:17:51 -05001079 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080}
1081
Al Viro0226f492011-12-06 12:21:54 -05001082static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001083{
Al Viro6ce6e242012-06-09 01:16:59 -04001084 struct proc_mounts *p = proc_mounts(m);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001085 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001086 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087}
1088
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001089const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 .start = m_start,
1091 .next = m_next,
1092 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001093 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001094};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001095#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001096
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097/**
1098 * may_umount_tree - check if a mount tree is busy
1099 * @mnt: root of mount tree
1100 *
1101 * This is called to check if a tree of mounts has any
1102 * open files, pwds, chroots or sub mounts that are
1103 * busy.
1104 */
Al Viro909b0a82011-11-25 03:06:56 -05001105int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106{
Al Viro909b0a82011-11-25 03:06:56 -05001107 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001108 int actual_refs = 0;
1109 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001110 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001111 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
Nick Pigginb3e19d92011-01-07 17:50:11 +11001113 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001114 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001115 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001116 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 }
Al Viro719ea2f2013-09-29 11:24:49 -04001119 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
1121 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001122 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Ian Kente3474a82006-03-27 01:14:51 -08001124 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125}
1126
1127EXPORT_SYMBOL(may_umount_tree);
1128
1129/**
1130 * may_umount - check if a mount point is busy
1131 * @mnt: root of mount
1132 *
1133 * This is called to check if a mount point has any
1134 * open files, pwds, chroots or sub mounts. If the
1135 * mount has sub mounts this will return busy
1136 * regardless of whether the sub mounts are busy.
1137 *
1138 * Doesn't take quota and stuff into account. IOW, in some cases it will
1139 * give false negatives. The main reason why it's here is that we need
1140 * a non-destructive way to look for easily umountable filesystems.
1141 */
1142int may_umount(struct vfsmount *mnt)
1143{
Ian Kente3474a82006-03-27 01:14:51 -08001144 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001145 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001146 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001147 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001148 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001149 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001150 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001151 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152}
1153
1154EXPORT_SYMBOL(may_umount);
1155
Al Viroe3197d82013-03-16 14:35:16 -04001156static LIST_HEAD(unmounted); /* protected by namespace_sem */
1157
Al Viro97216be2013-03-16 15:12:40 -04001158static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159{
Al Virod5e50f72011-11-24 20:58:57 -05001160 struct mount *mnt;
Al Viro97216be2013-03-16 15:12:40 -04001161 LIST_HEAD(head);
1162
1163 if (likely(list_empty(&unmounted))) {
1164 up_write(&namespace_sem);
1165 return;
1166 }
1167
1168 list_splice_init(&unmounted, &head);
1169 up_write(&namespace_sem);
1170
Al Viro48a066e2013-09-29 22:06:07 -04001171 synchronize_rcu();
1172
Al Viro97216be2013-03-16 15:12:40 -04001173 while (!list_empty(&head)) {
1174 mnt = list_first_entry(&head, struct mount, mnt_hash);
Al Viro1b8e5562011-11-24 21:01:32 -05001175 list_del_init(&mnt->mnt_hash);
Al Viroaba809cf2013-09-28 23:10:55 -04001176 if (mnt->mnt_ex_mountpoint.mnt)
1177 path_put(&mnt->mnt_ex_mountpoint);
Al Virod5e50f72011-11-24 20:58:57 -05001178 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001179 }
1180}
1181
Al Viro97216be2013-03-16 15:12:40 -04001182static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001183{
Al Viro97216be2013-03-16 15:12:40 -04001184 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001185}
1186
Nick Piggin99b7db72010-08-18 04:37:39 +10001187/*
Al Viro48a066e2013-09-29 22:06:07 -04001188 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001189 * namespace_sem must be held for write
Al Viro48a066e2013-09-29 22:06:07 -04001190 * how = 0 => just this tree, don't propagate
1191 * how = 1 => propagate; we know that nobody else has reference to any victims
1192 * how = 2 => lazy umount
Nick Piggin99b7db72010-08-18 04:37:39 +10001193 */
Al Viro48a066e2013-09-29 22:06:07 -04001194void umount_tree(struct mount *mnt, int how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001195{
Al Viro7b8a53f2011-01-15 20:08:44 -05001196 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001197 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001198
Al Viro909b0a82011-11-25 03:06:56 -05001199 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro1b8e5562011-11-24 21:01:32 -05001200 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001201
Al Viro48a066e2013-09-29 22:06:07 -04001202 if (how)
Al Viro7b8a53f2011-01-15 20:08:44 -05001203 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001204
Al Viro1b8e5562011-11-24 21:01:32 -05001205 list_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001206 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001207 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001208 __touch_mnt_namespace(p->mnt_ns);
1209 p->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04001210 if (how < 2)
1211 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro6b41d532011-11-24 23:24:33 -05001212 list_del_init(&p->mnt_child);
Al Viro676da582011-11-24 21:47:05 -05001213 if (mnt_has_parent(p)) {
Al Viro84d17192013-03-15 10:53:28 -04001214 put_mountpoint(p->mnt_mp);
Al Viroaba809cf2013-09-28 23:10:55 -04001215 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1216 p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint;
1217 p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt;
1218 p->mnt_mountpoint = p->mnt.mnt_root;
1219 p->mnt_parent = p;
Al Viro84d17192013-03-15 10:53:28 -04001220 p->mnt_mp = NULL;
Al Viro7c4b93d2008-03-21 23:59:49 -04001221 }
Al Viro0f0afb12011-11-24 20:43:10 -05001222 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 }
Al Viro328e6d92013-03-16 14:49:45 -04001224 list_splice(&tmp_list, &unmounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225}
1226
Al Virob54b9be2013-03-16 14:39:34 -04001227static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001228
Al Viro1ab59732011-11-24 21:35:16 -05001229static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230{
Al Viro1ab59732011-11-24 21:35:16 -05001231 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 int retval;
1233
Al Viro1ab59732011-11-24 21:35:16 -05001234 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 if (retval)
1236 return retval;
1237
1238 /*
1239 * Allow userspace to request a mountpoint be expired rather than
1240 * unmounting unconditionally. Unmount only happens if:
1241 * (1) the mark is already set (the mark is cleared by mntput())
1242 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1243 */
1244 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001245 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 flags & (MNT_FORCE | MNT_DETACH))
1247 return -EINVAL;
1248
Nick Pigginb3e19d92011-01-07 17:50:11 +11001249 /*
1250 * probably don't strictly need the lock here if we examined
1251 * all race cases, but it's a slowpath.
1252 */
Al Viro719ea2f2013-09-29 11:24:49 -04001253 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001254 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001255 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001257 }
Al Viro719ea2f2013-09-29 11:24:49 -04001258 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
Al Viro863d6842011-11-25 00:57:42 -05001260 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 return -EAGAIN;
1262 }
1263
1264 /*
1265 * If we may have to abort operations to get out of this
1266 * mount, and they will themselves hold resources we must
1267 * allow the fs to do things. In the Unix tradition of
1268 * 'Gee thats tricky lets do it in userspace' the umount_begin
1269 * might fail to complete on the first run through as other tasks
1270 * must return, and the like. Thats for the mount program to worry
1271 * about for the moment.
1272 */
1273
Al Viro42faad92008-04-24 07:21:56 -04001274 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001275 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
1278 /*
1279 * No sense to grab the lock for this test, but test itself looks
1280 * somewhat bogus. Suggestions for better replacement?
1281 * Ho-hum... In principle, we might treat that as umount + switch
1282 * to rootfs. GC would eventually take care of the old vfsmount.
1283 * Actually it makes sense, especially if rootfs would contain a
1284 * /reboot - static binary that would close all descriptors and
1285 * call reboot(9). Then init(8) could umount root and exec /reboot.
1286 */
Al Viro1ab59732011-11-24 21:35:16 -05001287 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 /*
1289 * Special case for "unmounting" root ...
1290 * we just try to remount it readonly.
1291 */
1292 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001293 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 up_write(&sb->s_umount);
1296 return retval;
1297 }
1298
Al Viro97216be2013-03-16 15:12:40 -04001299 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001300 lock_mount_hash();
Al Viro5addc5d2005-11-07 17:15:49 -05001301 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Al Viro48a066e2013-09-29 22:06:07 -04001303 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001304 if (!list_empty(&mnt->mnt_list))
Al Viro48a066e2013-09-29 22:06:07 -04001305 umount_tree(mnt, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001307 } else {
1308 shrink_submounts(mnt);
1309 retval = -EBUSY;
1310 if (!propagate_mount_busy(mnt, 2)) {
1311 if (!list_empty(&mnt->mnt_list))
1312 umount_tree(mnt, 1);
1313 retval = 0;
1314 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 }
Al Viro719ea2f2013-09-29 11:24:49 -04001316 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001317 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 return retval;
1319}
1320
Al Viro9b40bc92013-02-22 22:45:42 -05001321/*
1322 * Is the caller allowed to modify his namespace?
1323 */
1324static inline bool may_mount(void)
1325{
1326 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1327}
1328
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329/*
1330 * Now umount can handle mount points as well as block devices.
1331 * This is important for filesystems which use unnamed block devices.
1332 *
1333 * We now support a flag for forced unmount like the other 'big iron'
1334 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1335 */
1336
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001337SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338{
Al Viro2d8f3032008-07-22 09:59:21 -04001339 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001340 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001342 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001344 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1345 return -EINVAL;
1346
Al Viro9b40bc92013-02-22 22:45:42 -05001347 if (!may_mount())
1348 return -EPERM;
1349
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001350 if (!(flags & UMOUNT_NOFOLLOW))
1351 lookup_flags |= LOOKUP_FOLLOW;
1352
Al Viro197df042013-09-08 14:03:27 -04001353 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 if (retval)
1355 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001356 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001358 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001360 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 goto dput_and_out;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001362 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1363 goto dput_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Al Viro900148d2011-11-25 00:33:11 -05001365 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001367 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001368 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001369 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370out:
1371 return retval;
1372}
1373
1374#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1375
1376/*
Ram Paib58fed82005-11-07 17:16:09 -05001377 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001379SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380{
Ram Paib58fed82005-11-07 17:16:09 -05001381 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382}
1383
1384#endif
1385
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001386static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001387{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001388 /* Is this a proxy for a mount namespace? */
1389 struct inode *inode = dentry->d_inode;
David Howells0bb80f22013-04-12 01:50:06 +01001390 struct proc_ns *ei;
Eric W. Biederman8823c072010-03-07 18:49:36 -08001391
1392 if (!proc_ns_inode(inode))
1393 return false;
1394
David Howells0bb80f22013-04-12 01:50:06 +01001395 ei = get_proc_ns(inode);
Eric W. Biederman8823c072010-03-07 18:49:36 -08001396 if (ei->ns_ops != &mntns_operations)
1397 return false;
1398
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001399 return true;
1400}
1401
1402static bool mnt_ns_loop(struct dentry *dentry)
1403{
1404 /* Could bind mounting the mount namespace inode cause a
1405 * mount namespace loop?
1406 */
1407 struct mnt_namespace *mnt_ns;
1408 if (!is_mnt_ns_file(dentry))
1409 return false;
1410
1411 mnt_ns = get_proc_ns(dentry->d_inode)->ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08001412 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1413}
1414
Al Viro87129cc2011-11-24 21:24:27 -05001415struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001416 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
Al Viro84d17192013-03-15 10:53:28 -04001418 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001420 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1421 return ERR_PTR(-EINVAL);
1422
1423 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001424 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001425
Ram Pai36341f62005-11-07 17:17:22 -05001426 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001427 if (IS_ERR(q))
1428 return q;
1429
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001430 q->mnt.mnt_flags &= ~MNT_LOCKED;
Al Viroa73324d2011-11-24 22:25:07 -05001431 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
1433 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001434 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001435 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001436 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 continue;
1438
Al Viro909b0a82011-11-25 03:06:56 -05001439 for (s = r; s; s = next_mnt(s, r)) {
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001440 if (!(flag & CL_COPY_UNBINDABLE) &&
1441 IS_MNT_UNBINDABLE(s)) {
1442 s = skip_mnt_tree(s);
1443 continue;
1444 }
1445 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1446 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001447 s = skip_mnt_tree(s);
1448 continue;
1449 }
Al Viro0714a532011-11-24 22:19:58 -05001450 while (p != s->mnt_parent) {
1451 p = p->mnt_parent;
1452 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 }
Al Viro87129cc2011-11-24 21:24:27 -05001454 p = s;
Al Viro84d17192013-03-15 10:53:28 -04001455 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001456 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001457 if (IS_ERR(q))
1458 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04001459 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05001460 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro84d17192013-03-15 10:53:28 -04001461 attach_mnt(q, parent, p->mnt_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04001462 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 }
1464 }
1465 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001466out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04001468 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001469 umount_tree(res, 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001470 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001472 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473}
1474
David Howellsbe34d1a2012-06-25 12:55:18 +01001475/* Caller should check returned pointer for errors */
1476
Al Viro589ff872009-04-18 03:28:19 -04001477struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001478{
Al Virocb338d02011-11-24 20:55:08 -05001479 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001480 namespace_lock();
Al Viro87129cc2011-11-24 21:24:27 -05001481 tree = copy_tree(real_mount(path->mnt), path->dentry,
1482 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001483 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001484 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001485 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001486 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001487}
1488
1489void drop_collected_mounts(struct vfsmount *mnt)
1490{
Al Viro97216be2013-03-16 15:12:40 -04001491 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001492 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001493 umount_tree(real_mount(mnt), 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001494 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04001495 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001496}
1497
Al Viro1f707132010-01-30 22:51:25 -05001498int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1499 struct vfsmount *root)
1500{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001501 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001502 int res = f(root, arg);
1503 if (res)
1504 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001505 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1506 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001507 if (res)
1508 return res;
1509 }
1510 return 0;
1511}
1512
Al Viro4b8b21f2011-11-24 19:54:23 -05001513static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001514{
Al Viro315fc832011-11-24 18:57:30 -05001515 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001516
Al Viro909b0a82011-11-25 03:06:56 -05001517 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001518 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001519 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001520 }
1521}
1522
Al Viro4b8b21f2011-11-24 19:54:23 -05001523static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001524{
Al Viro315fc832011-11-24 18:57:30 -05001525 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001526
Al Viro909b0a82011-11-25 03:06:56 -05001527 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001528 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001529 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001530 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001531 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001532 return err;
1533 }
1534 }
1535 }
1536
1537 return 0;
1538}
1539
Ram Paib90fa9a2005-11-07 17:19:50 -05001540/*
1541 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001542 * @nd : place the mount tree @source_mnt is attached
1543 * @parent_nd : if non-null, detach the source_mnt from its parent and
1544 * store the parent mount and mountpoint dentry.
1545 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001546 *
1547 * NOTE: in the table below explains the semantics when a source mount
1548 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001549 * ---------------------------------------------------------------------------
1550 * | BIND MOUNT OPERATION |
1551 * |**************************************************************************
1552 * | source-->| shared | private | slave | unbindable |
1553 * | dest | | | | |
1554 * | | | | | | |
1555 * | v | | | | |
1556 * |**************************************************************************
1557 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1558 * | | | | | |
1559 * |non-shared| shared (+) | private | slave (*) | invalid |
1560 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001561 * A bind operation clones the source mount and mounts the clone on the
1562 * destination mount.
1563 *
1564 * (++) the cloned mount is propagated to all the mounts in the propagation
1565 * tree of the destination mount and the cloned mount is added to
1566 * the peer group of the source mount.
1567 * (+) the cloned mount is created under the destination mount and is marked
1568 * as shared. The cloned mount is added to the peer group of the source
1569 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001570 * (+++) the mount is propagated to all the mounts in the propagation tree
1571 * of the destination mount and the cloned mount is made slave
1572 * of the same master as that of the source mount. The cloned mount
1573 * is marked as 'shared and slave'.
1574 * (*) the cloned mount is made a slave of the same master as that of the
1575 * source mount.
1576 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001577 * ---------------------------------------------------------------------------
1578 * | MOVE MOUNT OPERATION |
1579 * |**************************************************************************
1580 * | source-->| shared | private | slave | unbindable |
1581 * | dest | | | | |
1582 * | | | | | | |
1583 * | v | | | | |
1584 * |**************************************************************************
1585 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1586 * | | | | | |
1587 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1588 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001589 *
1590 * (+) the mount is moved to the destination. And is then propagated to
1591 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001592 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001593 * (+++) the mount is moved to the destination and is then propagated to
1594 * all the mounts belonging to the destination mount's propagation tree.
1595 * the mount is marked as 'shared and slave'.
1596 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001597 *
1598 * if the source mount is a tree, the operations explained above is
1599 * applied to each mount in the tree.
1600 * Must be called without spinlocks held, since this function can sleep
1601 * in allocations.
1602 */
Al Viro0fb54e52011-11-24 19:59:16 -05001603static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d17192013-03-15 10:53:28 -04001604 struct mount *dest_mnt,
1605 struct mountpoint *dest_mp,
1606 struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001607{
1608 LIST_HEAD(tree_list);
Al Viro315fc832011-11-24 18:57:30 -05001609 struct mount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001610 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001611
Al Virofc7be132011-11-25 01:05:37 -05001612 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001613 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001614 if (err)
1615 goto out;
1616 }
Al Viro84d17192013-03-15 10:53:28 -04001617 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001618 if (err)
1619 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001620
Al Viro719ea2f2013-09-29 11:24:49 -04001621 lock_mount_hash();
Al Virodf1a1ad2010-01-16 12:57:40 -05001622
Al Virofc7be132011-11-25 01:05:37 -05001623 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro909b0a82011-11-25 03:06:56 -05001624 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001625 set_mnt_shared(p);
Ram Paib90fa9a2005-11-07 17:19:50 -05001626 }
Al Viro1a390682008-03-21 20:48:19 -04001627 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001628 detach_mnt(source_mnt, parent_path);
Al Viro84d17192013-03-15 10:53:28 -04001629 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05001630 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001631 } else {
Al Viro84d17192013-03-15 10:53:28 -04001632 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Al Viro0fb54e52011-11-24 19:59:16 -05001633 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05001634 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001635
Al Viro1b8e5562011-11-24 21:01:32 -05001636 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1637 list_del_init(&child->mnt_hash);
Al Viro4b2619a2011-11-24 19:47:15 -05001638 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05001639 }
Al Viro719ea2f2013-09-29 11:24:49 -04001640 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10001641
Ram Paib90fa9a2005-11-07 17:19:50 -05001642 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001643
1644 out_cleanup_ids:
Al Virofc7be132011-11-25 01:05:37 -05001645 if (IS_MNT_SHARED(dest_mnt))
Al Viro0fb54e52011-11-24 19:59:16 -05001646 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001647 out:
1648 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001649}
1650
Al Viro84d17192013-03-15 10:53:28 -04001651static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04001652{
1653 struct vfsmount *mnt;
Al Viro84d17192013-03-15 10:53:28 -04001654 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04001655retry:
Al Viro84d17192013-03-15 10:53:28 -04001656 mutex_lock(&dentry->d_inode->i_mutex);
1657 if (unlikely(cant_mount(dentry))) {
1658 mutex_unlock(&dentry->d_inode->i_mutex);
1659 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04001660 }
Al Viro97216be2013-03-16 15:12:40 -04001661 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04001662 mnt = lookup_mnt(path);
Al Viro84d17192013-03-15 10:53:28 -04001663 if (likely(!mnt)) {
1664 struct mountpoint *mp = new_mountpoint(dentry);
1665 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04001666 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001667 mutex_unlock(&dentry->d_inode->i_mutex);
1668 return mp;
1669 }
1670 return mp;
1671 }
Al Viro97216be2013-03-16 15:12:40 -04001672 namespace_unlock();
Al Virob12cea92011-03-18 08:55:38 -04001673 mutex_unlock(&path->dentry->d_inode->i_mutex);
1674 path_put(path);
1675 path->mnt = mnt;
Al Viro84d17192013-03-15 10:53:28 -04001676 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04001677 goto retry;
1678}
1679
Al Viro84d17192013-03-15 10:53:28 -04001680static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04001681{
Al Viro84d17192013-03-15 10:53:28 -04001682 struct dentry *dentry = where->m_dentry;
1683 put_mountpoint(where);
Al Viro328e6d92013-03-16 14:49:45 -04001684 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001685 mutex_unlock(&dentry->d_inode->i_mutex);
Al Virob12cea92011-03-18 08:55:38 -04001686}
1687
Al Viro84d17192013-03-15 10:53:28 -04001688static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689{
Al Viro95bc5f22011-11-25 00:30:56 -05001690 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 return -EINVAL;
1692
Al Viro84d17192013-03-15 10:53:28 -04001693 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001694 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 return -ENOTDIR;
1696
Al Viro84d17192013-03-15 10:53:28 -04001697 return attach_recursive_mnt(mnt, p, mp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698}
1699
1700/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001701 * Sanity check the flags to change_mnt_propagation.
1702 */
1703
1704static int flags_to_propagation_type(int flags)
1705{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001706 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001707
1708 /* Fail if any non-propagation flags are set */
1709 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1710 return 0;
1711 /* Only one propagation flag should be set */
1712 if (!is_power_of_2(type))
1713 return 0;
1714 return type;
1715}
1716
1717/*
Ram Pai07b20882005-11-07 17:19:07 -05001718 * recursively change the type of the mountpoint.
1719 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001720static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001721{
Al Viro315fc832011-11-24 18:57:30 -05001722 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001723 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001724 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001725 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001726 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001727
Al Viro2d92ab32008-08-02 00:51:11 -04001728 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001729 return -EINVAL;
1730
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001731 type = flags_to_propagation_type(flag);
1732 if (!type)
1733 return -EINVAL;
1734
Al Viro97216be2013-03-16 15:12:40 -04001735 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001736 if (type == MS_SHARED) {
1737 err = invent_group_ids(mnt, recurse);
1738 if (err)
1739 goto out_unlock;
1740 }
1741
Al Viro719ea2f2013-09-29 11:24:49 -04001742 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001743 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001744 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04001745 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001746
1747 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04001748 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001749 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001750}
1751
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001752static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1753{
1754 struct mount *child;
1755 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1756 if (!is_subdir(child->mnt_mountpoint, dentry))
1757 continue;
1758
1759 if (child->mnt.mnt_flags & MNT_LOCKED)
1760 return true;
1761 }
1762 return false;
1763}
1764
Ram Pai07b20882005-11-07 17:19:07 -05001765/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 * do loopback mount.
1767 */
Al Viro808d4e32012-10-11 11:42:01 -04001768static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001769 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770{
Al Viro2d92ab32008-08-02 00:51:11 -04001771 struct path old_path;
Al Viro84d17192013-03-15 10:53:28 -04001772 struct mount *mnt = NULL, *old, *parent;
1773 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05001774 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 if (!old_name || !*old_name)
1776 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001777 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 if (err)
1779 return err;
1780
Eric W. Biederman8823c072010-03-07 18:49:36 -08001781 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001782 if (mnt_ns_loop(old_path.dentry))
Eric W. Biederman8823c072010-03-07 18:49:36 -08001783 goto out;
1784
Al Viro84d17192013-03-15 10:53:28 -04001785 mp = lock_mount(path);
1786 err = PTR_ERR(mp);
1787 if (IS_ERR(mp))
Jan Blunck4ac91372008-02-14 19:34:32 -08001788 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001789
Al Viro87129cc2011-11-24 21:24:27 -05001790 old = real_mount(old_path.mnt);
Al Viro84d17192013-03-15 10:53:28 -04001791 parent = real_mount(path->mnt);
Al Viro87129cc2011-11-24 21:24:27 -05001792
Al Virob12cea92011-03-18 08:55:38 -04001793 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001794 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04001795 goto out2;
1796
Al Viro84d17192013-03-15 10:53:28 -04001797 if (!check_mnt(parent) || !check_mnt(old))
Al Virob12cea92011-03-18 08:55:38 -04001798 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001800 if (!recurse && has_locked_children(old, old_path.dentry))
1801 goto out2;
1802
Al Viroccd48bc2005-11-07 17:15:04 -05001803 if (recurse)
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001804 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
Al Viroccd48bc2005-11-07 17:15:04 -05001805 else
Al Viro87129cc2011-11-24 21:24:27 -05001806 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001807
David Howellsbe34d1a2012-06-25 12:55:18 +01001808 if (IS_ERR(mnt)) {
1809 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04001810 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01001811 }
Al Viroccd48bc2005-11-07 17:15:04 -05001812
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001813 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
1814
Al Viro84d17192013-03-15 10:53:28 -04001815 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05001816 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04001817 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001818 umount_tree(mnt, 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001819 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001820 }
Al Virob12cea92011-03-18 08:55:38 -04001821out2:
Al Viro84d17192013-03-15 10:53:28 -04001822 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05001823out:
Al Viro2d92ab32008-08-02 00:51:11 -04001824 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 return err;
1826}
1827
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001828static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1829{
1830 int error = 0;
1831 int readonly_request = 0;
1832
1833 if (ms_flags & MS_RDONLY)
1834 readonly_request = 1;
1835 if (readonly_request == __mnt_is_readonly(mnt))
1836 return 0;
1837
Eric W. Biederman90563b12013-03-22 03:10:15 -07001838 if (mnt->mnt_flags & MNT_LOCK_READONLY)
1839 return -EPERM;
1840
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001841 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05001842 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001843 else
Al Viro83adc752011-11-24 22:37:54 -05001844 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001845 return error;
1846}
1847
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848/*
1849 * change filesystem flags. dir should be a physical root of filesystem.
1850 * If you've mounted a non-root directory somewhere and want to do remount
1851 * on it - tough luck.
1852 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001853static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 void *data)
1855{
1856 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001857 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05001858 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
Al Viro143c8c92011-11-25 00:46:35 -05001860 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 return -EINVAL;
1862
Al Viro2d92ab32008-08-02 00:51:11 -04001863 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 return -EINVAL;
1865
Eric Parisff36fe22011-03-03 16:09:14 -05001866 err = security_sb_remount(sb, data);
1867 if (err)
1868 return err;
1869
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001871 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001872 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05001873 else if (!capable(CAP_SYS_ADMIN))
1874 err = -EPERM;
Al Viro4aa98cf2009-05-08 13:36:58 -04001875 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001876 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001877 if (!err) {
Al Viro719ea2f2013-09-29 11:24:49 -04001878 lock_mount_hash();
Al Viro143c8c92011-11-25 00:46:35 -05001879 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1880 mnt->mnt.mnt_flags = mnt_flags;
Al Viro143c8c92011-11-25 00:46:35 -05001881 touch_mnt_namespace(mnt->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04001882 unlock_mount_hash();
Dan Williams0e55a7c2008-09-26 19:01:20 -07001883 }
Al Viro6339dab2013-09-16 22:41:01 -04001884 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 return err;
1886}
1887
Al Virocbbe3622011-11-24 20:01:19 -05001888static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05001889{
Al Viro315fc832011-11-24 18:57:30 -05001890 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001891 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001892 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05001893 return 1;
1894 }
1895 return 0;
1896}
1897
Al Viro808d4e32012-10-11 11:42:01 -04001898static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899{
Al Viro2d92ab32008-08-02 00:51:11 -04001900 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05001901 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05001902 struct mount *old;
Al Viro84d17192013-03-15 10:53:28 -04001903 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05001904 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 if (!old_name || !*old_name)
1906 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001907 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 if (err)
1909 return err;
1910
Al Viro84d17192013-03-15 10:53:28 -04001911 mp = lock_mount(path);
1912 err = PTR_ERR(mp);
1913 if (IS_ERR(mp))
David Howellscc53ce52011-01-14 18:45:26 +00001914 goto out;
1915
Al Viro143c8c92011-11-25 00:46:35 -05001916 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05001917 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05001918
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001920 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 goto out1;
1922
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001923 if (old->mnt.mnt_flags & MNT_LOCKED)
1924 goto out1;
1925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001927 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001928 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929
Al Viro676da582011-11-24 21:47:05 -05001930 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05001931 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
Al Viro2d92ab32008-08-02 00:51:11 -04001933 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1934 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001935 goto out1;
1936 /*
1937 * Don't move a mount residing in a shared parent.
1938 */
Al Virofc7be132011-11-25 01:05:37 -05001939 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001940 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001941 /*
1942 * Don't move a mount tree containing unbindable mounts to a destination
1943 * mount which is shared.
1944 */
Al Virofc7be132011-11-25 01:05:37 -05001945 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05001946 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05001948 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05001949 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05001950 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951
Al Viro84d17192013-03-15 10:53:28 -04001952 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001953 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001954 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
1956 /* if the mount is moved, it should no longer be expire
1957 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05001958 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959out1:
Al Viro84d17192013-03-15 10:53:28 -04001960 unlock_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001963 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001964 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 return err;
1966}
1967
Al Viro9d412a42011-03-17 22:08:28 -04001968static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1969{
1970 int err;
1971 const char *subtype = strchr(fstype, '.');
1972 if (subtype) {
1973 subtype++;
1974 err = -EINVAL;
1975 if (!subtype[0])
1976 goto err;
1977 } else
1978 subtype = "";
1979
1980 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1981 err = -ENOMEM;
1982 if (!mnt->mnt_sb->s_subtype)
1983 goto err;
1984 return mnt;
1985
1986 err:
1987 mntput(mnt);
1988 return ERR_PTR(err);
1989}
1990
Al Viro9d412a42011-03-17 22:08:28 -04001991/*
1992 * add a mount into a namespace's mount tree
1993 */
Al Viro95bc5f22011-11-25 00:30:56 -05001994static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04001995{
Al Viro84d17192013-03-15 10:53:28 -04001996 struct mountpoint *mp;
1997 struct mount *parent;
Al Viro9d412a42011-03-17 22:08:28 -04001998 int err;
1999
Al Viro48a066e2013-09-29 22:06:07 -04002000 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL | MNT_DOOMED | MNT_SYNC_UMOUNT);
Al Viro9d412a42011-03-17 22:08:28 -04002001
Al Viro84d17192013-03-15 10:53:28 -04002002 mp = lock_mount(path);
2003 if (IS_ERR(mp))
2004 return PTR_ERR(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002005
Al Viro84d17192013-03-15 10:53:28 -04002006 parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04002007 err = -EINVAL;
Al Viro84d17192013-03-15 10:53:28 -04002008 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04002009 /* that's acceptable only for automounts done in private ns */
2010 if (!(mnt_flags & MNT_SHRINKABLE))
2011 goto unlock;
2012 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d17192013-03-15 10:53:28 -04002013 if (!parent->mnt_ns)
Al Viro156cacb2012-09-21 08:19:02 -04002014 goto unlock;
2015 }
Al Viro9d412a42011-03-17 22:08:28 -04002016
2017 /* Refuse the same filesystem on the same mount point */
2018 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05002019 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04002020 path->mnt->mnt_root == path->dentry)
2021 goto unlock;
2022
2023 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05002024 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04002025 goto unlock;
2026
Al Viro95bc5f22011-11-25 00:30:56 -05002027 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro84d17192013-03-15 10:53:28 -04002028 err = graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04002029
2030unlock:
Al Viro84d17192013-03-15 10:53:28 -04002031 unlock_mount(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002032 return err;
2033}
Al Virob1e75df2011-01-17 01:47:59 -05002034
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035/*
2036 * create a new mount for userspace and request it to be added into the
2037 * namespace's tree
2038 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002039static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04002040 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002042 struct file_system_type *type;
Al Viro9b40bc92013-02-22 22:45:42 -05002043 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002045 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002047 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return -EINVAL;
2049
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002050 type = get_fs_type(fstype);
2051 if (!type)
2052 return -ENODEV;
2053
2054 if (user_ns != &init_user_ns) {
2055 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2056 put_filesystem(type);
2057 return -EPERM;
2058 }
2059 /* Only in special cases allow devices from mounts
2060 * created outside the initial user namespace.
2061 */
2062 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2063 flags |= MS_NODEV;
2064 mnt_flags |= MNT_NODEV;
2065 }
2066 }
2067
2068 mnt = vfs_kern_mount(type, flags, name, data);
2069 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2070 !mnt->mnt_sb->s_subtype)
2071 mnt = fs_set_subtype(mnt, fstype);
2072
2073 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 if (IS_ERR(mnt))
2075 return PTR_ERR(mnt);
2076
Al Viro95bc5f22011-11-25 00:30:56 -05002077 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002078 if (err)
2079 mntput(mnt);
2080 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081}
2082
Al Viro19a167a2011-01-17 01:35:23 -05002083int finish_automount(struct vfsmount *m, struct path *path)
2084{
Al Viro6776db3d2011-11-25 00:22:05 -05002085 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002086 int err;
2087 /* The new mount record should have at least 2 refs to prevent it being
2088 * expired before we get a chance to add it
2089 */
Al Viro6776db3d2011-11-25 00:22:05 -05002090 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002091
2092 if (m->mnt_sb == path->mnt->mnt_sb &&
2093 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002094 err = -ELOOP;
2095 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002096 }
2097
Al Viro95bc5f22011-11-25 00:30:56 -05002098 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002099 if (!err)
2100 return 0;
2101fail:
2102 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002103 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002104 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05002105 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04002106 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002107 }
Al Virob1e75df2011-01-17 01:47:59 -05002108 mntput(m);
2109 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002110 return err;
2111}
2112
David Howellsea5b7782011-01-14 19:10:03 +00002113/**
2114 * mnt_set_expiry - Put a mount on an expiration list
2115 * @mnt: The mount to list.
2116 * @expiry_list: The list to add the mount to.
2117 */
2118void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2119{
Al Viro97216be2013-03-16 15:12:40 -04002120 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00002121
Al Viro6776db3d2011-11-25 00:22:05 -05002122 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002123
Al Viro97216be2013-03-16 15:12:40 -04002124 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002125}
2126EXPORT_SYMBOL(mnt_set_expiry);
2127
2128/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 * process a list of expirable mountpoints with the intent of discarding any
2130 * mountpoints that aren't in use and haven't been touched since last we came
2131 * here
2132 */
2133void mark_mounts_for_expiry(struct list_head *mounts)
2134{
Al Viro761d5c32011-11-24 21:07:43 -05002135 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 LIST_HEAD(graveyard);
2137
2138 if (list_empty(mounts))
2139 return;
2140
Al Viro97216be2013-03-16 15:12:40 -04002141 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002142 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143
2144 /* extract from the expiration list every vfsmount that matches the
2145 * following criteria:
2146 * - only referenced by its parent vfsmount
2147 * - still marked for expiry (marked on the last call here; marks are
2148 * cleared by mntput())
2149 */
Al Viro6776db3d2011-11-25 00:22:05 -05002150 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002151 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002152 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002154 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002156 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002157 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002158 touch_mnt_namespace(mnt->mnt_ns);
Al Viro328e6d92013-03-16 14:49:45 -04002159 umount_tree(mnt, 1);
Al Virobcc5c7d2008-03-22 00:21:53 -04002160 }
Al Viro719ea2f2013-09-29 11:24:49 -04002161 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002162 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163}
2164
2165EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2166
2167/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002168 * Ripoff of 'select_parent()'
2169 *
2170 * search the list of submounts for a given mountpoint, and move any
2171 * shrinkable submounts to the 'graveyard' list.
2172 */
Al Viro692afc32011-11-24 21:15:14 -05002173static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002174{
Al Viro692afc32011-11-24 21:15:14 -05002175 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002176 struct list_head *next;
2177 int found = 0;
2178
2179repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002180 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002181resume:
Al Viro6b41d532011-11-24 23:24:33 -05002182 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002183 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002184 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002185
2186 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002187 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002188 continue;
2189 /*
2190 * Descend a level if the d_mounts list is non-empty.
2191 */
Al Viro6b41d532011-11-24 23:24:33 -05002192 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002193 this_parent = mnt;
2194 goto repeat;
2195 }
2196
Al Viro1ab59732011-11-24 21:35:16 -05002197 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002198 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002199 found++;
2200 }
2201 }
2202 /*
2203 * All done at this level ... ascend and resume the search
2204 */
2205 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002206 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002207 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002208 goto resume;
2209 }
2210 return found;
2211}
2212
2213/*
2214 * process a list of expirable mountpoints with the intent of discarding any
2215 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002216 *
Al Viro48a066e2013-09-29 22:06:07 -04002217 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002218 */
Al Virob54b9be2013-03-16 14:39:34 -04002219static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002220{
2221 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002222 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002223
Trond Myklebust5528f9112006-06-09 09:34:17 -04002224 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002225 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002226 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002227 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002228 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002229 touch_mnt_namespace(m->mnt_ns);
Al Viro328e6d92013-03-16 14:49:45 -04002230 umount_tree(m, 1);
Al Virobcc5c7d2008-03-22 00:21:53 -04002231 }
2232 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002233}
2234
Trond Myklebust5528f9112006-06-09 09:34:17 -04002235/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 * Some copy_from_user() implementations do not return the exact number of
2237 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2238 * Note that this function differs from copy_from_user() in that it will oops
2239 * on bad values of `to', rather than returning a short copy.
2240 */
Ram Paib58fed82005-11-07 17:16:09 -05002241static long exact_copy_from_user(void *to, const void __user * from,
2242 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243{
2244 char *t = to;
2245 const char __user *f = from;
2246 char c;
2247
2248 if (!access_ok(VERIFY_READ, from, n))
2249 return n;
2250
2251 while (n) {
2252 if (__get_user(c, f)) {
2253 memset(t, 0, n);
2254 break;
2255 }
2256 *t++ = c;
2257 f++;
2258 n--;
2259 }
2260 return n;
2261}
2262
Ram Paib58fed82005-11-07 17:16:09 -05002263int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264{
2265 int i;
2266 unsigned long page;
2267 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002268
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 *where = 0;
2270 if (!data)
2271 return 0;
2272
2273 if (!(page = __get_free_page(GFP_KERNEL)))
2274 return -ENOMEM;
2275
2276 /* We only care that *some* data at the address the user
2277 * gave us is valid. Just in case, we'll zero
2278 * the remainder of the page.
2279 */
2280 /* copy_from_user cannot cross TASK_SIZE ! */
2281 size = TASK_SIZE - (unsigned long)data;
2282 if (size > PAGE_SIZE)
2283 size = PAGE_SIZE;
2284
2285 i = size - exact_copy_from_user((void *)page, data, size);
2286 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002287 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 return -EFAULT;
2289 }
2290 if (i != PAGE_SIZE)
2291 memset((char *)page + i, 0, PAGE_SIZE - i);
2292 *where = page;
2293 return 0;
2294}
2295
Vegard Nossumeca6f532009-09-18 13:05:45 -07002296int copy_mount_string(const void __user *data, char **where)
2297{
2298 char *tmp;
2299
2300 if (!data) {
2301 *where = NULL;
2302 return 0;
2303 }
2304
2305 tmp = strndup_user(data, PAGE_SIZE);
2306 if (IS_ERR(tmp))
2307 return PTR_ERR(tmp);
2308
2309 *where = tmp;
2310 return 0;
2311}
2312
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313/*
2314 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2315 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2316 *
2317 * data is a (void *) that can point to any structure up to
2318 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2319 * information (or be NULL).
2320 *
2321 * Pre-0.97 versions of mount() didn't have a flags word.
2322 * When the flags word was introduced its top half was required
2323 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2324 * Therefore, if this magic number is present, it carries no information
2325 * and must be discarded.
2326 */
Al Viro808d4e32012-10-11 11:42:01 -04002327long do_mount(const char *dev_name, const char *dir_name,
2328 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329{
Al Viro2d92ab32008-08-02 00:51:11 -04002330 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 int retval = 0;
2332 int mnt_flags = 0;
2333
2334 /* Discard magic */
2335 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2336 flags &= ~MS_MGC_MSK;
2337
2338 /* Basic sanity checks */
2339
2340 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2341 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342
2343 if (data_page)
2344 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2345
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002346 /* ... and get the mountpoint */
2347 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2348 if (retval)
2349 return retval;
2350
2351 retval = security_sb_mount(dev_name, &path,
2352 type_page, flags, data_page);
Al Viro0d5cadb2013-05-04 14:40:51 -04002353 if (!retval && !may_mount())
2354 retval = -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002355 if (retval)
2356 goto dput_out;
2357
Andi Kleen613cbe32009-04-19 18:40:43 +02002358 /* Default to relatime unless overriden */
2359 if (!(flags & MS_NOATIME))
2360 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002361
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 /* Separate the per-mountpoint flags */
2363 if (flags & MS_NOSUID)
2364 mnt_flags |= MNT_NOSUID;
2365 if (flags & MS_NODEV)
2366 mnt_flags |= MNT_NODEV;
2367 if (flags & MS_NOEXEC)
2368 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002369 if (flags & MS_NOATIME)
2370 mnt_flags |= MNT_NOATIME;
2371 if (flags & MS_NODIRATIME)
2372 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002373 if (flags & MS_STRICTATIME)
2374 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002375 if (flags & MS_RDONLY)
2376 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002377
Al Viro7a4dec52010-08-09 12:05:43 -04002378 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002379 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2380 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002383 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 data_page);
2385 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002386 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002387 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002388 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002390 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 else
Al Viro2d92ab32008-08-02 00:51:11 -04002392 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 dev_name, data_page);
2394dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002395 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 return retval;
2397}
2398
Eric W. Biederman771b1372012-07-26 21:08:32 -07002399static void free_mnt_ns(struct mnt_namespace *ns)
2400{
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002401 proc_free_inum(ns->proc_inum);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002402 put_user_ns(ns->user_ns);
2403 kfree(ns);
2404}
2405
Eric W. Biederman8823c072010-03-07 18:49:36 -08002406/*
2407 * Assign a sequence number so we can detect when we attempt to bind
2408 * mount a reference to an older mount namespace into the current
2409 * mount namespace, preventing reference counting loops. A 64bit
2410 * number incrementing at 10Ghz will take 12,427 years to wrap which
2411 * is effectively never, so we can ignore the possibility.
2412 */
2413static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2414
Eric W. Biederman771b1372012-07-26 21:08:32 -07002415static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002416{
2417 struct mnt_namespace *new_ns;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002418 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002419
2420 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2421 if (!new_ns)
2422 return ERR_PTR(-ENOMEM);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002423 ret = proc_alloc_inum(&new_ns->proc_inum);
2424 if (ret) {
2425 kfree(new_ns);
2426 return ERR_PTR(ret);
2427 }
Eric W. Biederman8823c072010-03-07 18:49:36 -08002428 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002429 atomic_set(&new_ns->count, 1);
2430 new_ns->root = NULL;
2431 INIT_LIST_HEAD(&new_ns->list);
2432 init_waitqueue_head(&new_ns->poll);
2433 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002434 new_ns->user_ns = get_user_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002435 return new_ns;
2436}
2437
Al Viro9559f682013-09-28 20:47:57 -04002438struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2439 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002441 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002442 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002443 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04002444 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05002445 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002446 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447
Al Viro9559f682013-09-28 20:47:57 -04002448 BUG_ON(!ns);
2449
2450 if (likely(!(flags & CLONE_NEWNS))) {
2451 get_mnt_ns(ns);
2452 return ns;
2453 }
2454
2455 old = ns->root;
2456
Eric W. Biederman771b1372012-07-26 21:08:32 -07002457 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002458 if (IS_ERR(new_ns))
2459 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460
Al Viro97216be2013-03-16 15:12:40 -04002461 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002463 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04002464 if (user_ns != ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002465 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002466 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002467 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04002468 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002469 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002470 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 }
Al Virobe08d6d2011-12-06 13:32:36 -05002472 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002473 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474
2475 /*
2476 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2477 * as belonging to new namespace. We have already acquired a private
2478 * fs_struct, so tsk->fs->lock is not needed.
2479 */
Al Viro909b0a82011-11-25 03:06:56 -05002480 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002481 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002483 q->mnt_ns = new_ns;
Al Viro9559f682013-09-28 20:47:57 -04002484 if (new_fs) {
2485 if (&p->mnt == new_fs->root.mnt) {
2486 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002487 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 }
Al Viro9559f682013-09-28 20:47:57 -04002489 if (&p->mnt == new_fs->pwd.mnt) {
2490 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002491 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 }
Al Viro909b0a82011-11-25 03:06:56 -05002494 p = next_mnt(p, old);
2495 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002496 if (!q)
2497 break;
2498 while (p->mnt.mnt_root != q->mnt.mnt_root)
2499 p = next_mnt(p, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 }
Al Viro328e6d92013-03-16 14:49:45 -04002501 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002504 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002506 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507
JANAK DESAI741a2952006-02-07 12:59:00 -08002508 return new_ns;
2509}
2510
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002511/**
2512 * create_mnt_ns - creates a private namespace and adds a root filesystem
2513 * @mnt: pointer to the new root filesystem mountpoint
2514 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002515static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002516{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002517 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002518 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002519 struct mount *mnt = real_mount(m);
2520 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002521 new_ns->root = mnt;
Al Virob1983cd2013-05-04 15:18:53 -04002522 list_add(&mnt->mnt_list, &new_ns->list);
Al Viroc1334492011-11-16 16:12:14 -05002523 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002524 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002525 }
2526 return new_ns;
2527}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002528
Al Viroea441d12011-11-16 21:43:59 -05002529struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2530{
2531 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002532 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002533 struct path path;
2534 int err;
2535
2536 ns = create_mnt_ns(mnt);
2537 if (IS_ERR(ns))
2538 return ERR_CAST(ns);
2539
2540 err = vfs_path_lookup(mnt->mnt_root, mnt,
2541 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2542
2543 put_mnt_ns(ns);
2544
2545 if (err)
2546 return ERR_PTR(err);
2547
2548 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002549 s = path.mnt->mnt_sb;
2550 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002551 mntput(path.mnt);
2552 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002553 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002554 /* ... and return the root of (sub)tree on it */
2555 return path.dentry;
2556}
2557EXPORT_SYMBOL(mount_subtree);
2558
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002559SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2560 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002562 int ret;
2563 char *kernel_type;
Jeff Layton91a27b22012-10-10 15:25:28 -04002564 struct filename *kernel_dir;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002565 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567
Vegard Nossumeca6f532009-09-18 13:05:45 -07002568 ret = copy_mount_string(type, &kernel_type);
2569 if (ret < 0)
2570 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571
Vegard Nossumeca6f532009-09-18 13:05:45 -07002572 kernel_dir = getname(dir_name);
2573 if (IS_ERR(kernel_dir)) {
2574 ret = PTR_ERR(kernel_dir);
2575 goto out_dir;
2576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577
Vegard Nossumeca6f532009-09-18 13:05:45 -07002578 ret = copy_mount_string(dev_name, &kernel_dev);
2579 if (ret < 0)
2580 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581
Vegard Nossumeca6f532009-09-18 13:05:45 -07002582 ret = copy_mount_options(data, &data_page);
2583 if (ret < 0)
2584 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585
Jeff Layton91a27b22012-10-10 15:25:28 -04002586 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
Vegard Nossumeca6f532009-09-18 13:05:45 -07002587 (void *) data_page);
2588
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002590out_data:
2591 kfree(kernel_dev);
2592out_dev:
2593 putname(kernel_dir);
2594out_dir:
2595 kfree(kernel_type);
2596out_type:
2597 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598}
2599
2600/*
Al Viroafac7cb2011-11-23 19:34:49 -05002601 * Return true if path is reachable from root
2602 *
Al Viro48a066e2013-09-29 22:06:07 -04002603 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05002604 */
Al Viro643822b2011-11-24 22:00:28 -05002605bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002606 const struct path *root)
2607{
Al Viro643822b2011-11-24 22:00:28 -05002608 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002609 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002610 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002611 }
Al Viro643822b2011-11-24 22:00:28 -05002612 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002613}
2614
2615int path_is_under(struct path *path1, struct path *path2)
2616{
2617 int res;
Al Viro48a066e2013-09-29 22:06:07 -04002618 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002619 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04002620 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002621 return res;
2622}
2623EXPORT_SYMBOL(path_is_under);
2624
2625/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 * pivot_root Semantics:
2627 * Moves the root file system of the current process to the directory put_old,
2628 * makes new_root as the new root file system of the current process, and sets
2629 * root/cwd of all processes which had them on the current root to new_root.
2630 *
2631 * Restrictions:
2632 * The new_root and put_old must be directories, and must not be on the
2633 * same file system as the current process root. The put_old must be
2634 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2635 * pointed to by put_old must yield the same directory as new_root. No other
2636 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2637 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002638 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2639 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2640 * in this situation.
2641 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 * Notes:
2643 * - we don't move root/cwd if they are not at the root (reason: if something
2644 * cared enough to change them, it's probably wrong to force them elsewhere)
2645 * - it's okay to pick a root that isn't the root of a file system, e.g.
2646 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2647 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2648 * first.
2649 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002650SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2651 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652{
Al Viro2d8f3032008-07-22 09:59:21 -04002653 struct path new, old, parent_path, root_parent, root;
Al Viro84d17192013-03-15 10:53:28 -04002654 struct mount *new_mnt, *root_mnt, *old_mnt;
2655 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 int error;
2657
Al Viro9b40bc92013-02-22 22:45:42 -05002658 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 return -EPERM;
2660
Al Viro2d8f3032008-07-22 09:59:21 -04002661 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 if (error)
2663 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664
Al Viro2d8f3032008-07-22 09:59:21 -04002665 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 if (error)
2667 goto out1;
2668
Al Viro2d8f3032008-07-22 09:59:21 -04002669 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002670 if (error)
2671 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002673 get_fs_root(current->fs, &root);
Al Viro84d17192013-03-15 10:53:28 -04002674 old_mp = lock_mount(&old);
2675 error = PTR_ERR(old_mp);
2676 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04002677 goto out3;
2678
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002680 new_mnt = real_mount(new.mnt);
2681 root_mnt = real_mount(root.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002682 old_mnt = real_mount(old.mnt);
2683 if (IS_MNT_SHARED(old_mnt) ||
Al Virofc7be132011-11-25 01:05:37 -05002684 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2685 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002686 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002687 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002688 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002689 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2690 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002692 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002693 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 error = -EBUSY;
Al Viro84d17192013-03-15 10:53:28 -04002695 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04002696 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002698 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002699 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002700 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002701 goto out4; /* not attached */
Al Viro84d17192013-03-15 10:53:28 -04002702 root_mp = root_mnt->mnt_mp;
Al Viro2d8f3032008-07-22 09:59:21 -04002703 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002704 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002705 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002706 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002707 /* make sure we can reach put_old from new_root */
Al Viro84d17192013-03-15 10:53:28 -04002708 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002709 goto out4;
Al Viro84d17192013-03-15 10:53:28 -04002710 root_mp->m_count++; /* pin it so it won't go away */
Al Viro719ea2f2013-09-29 11:24:49 -04002711 lock_mount_hash();
Al Viro419148d2011-11-24 19:41:16 -05002712 detach_mnt(new_mnt, &parent_path);
2713 detach_mnt(root_mnt, &root_parent);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002714 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2715 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2716 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2717 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002718 /* mount old root on put_old */
Al Viro84d17192013-03-15 10:53:28 -04002719 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08002720 /* mount new_root on / */
Al Viro84d17192013-03-15 10:53:28 -04002721 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002722 touch_mnt_namespace(current->nsproxy->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04002723 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04002724 chroot_fs_refs(&root, &new);
Al Viro84d17192013-03-15 10:53:28 -04002725 put_mountpoint(root_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002727out4:
Al Viro84d17192013-03-15 10:53:28 -04002728 unlock_mount(old_mp);
Al Virob12cea92011-03-18 08:55:38 -04002729 if (!error) {
2730 path_put(&root_parent);
2731 path_put(&parent_path);
2732 }
2733out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002734 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002735out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002736 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002738 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741}
2742
2743static void __init init_mount_tree(void)
2744{
2745 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002746 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002747 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002748 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002750 type = get_fs_type("rootfs");
2751 if (!type)
2752 panic("Can't find rootfs type");
2753 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2754 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 if (IS_ERR(mnt))
2756 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002757
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002758 ns = create_mnt_ns(mnt);
2759 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002762 init_task.nsproxy->mnt_ns = ns;
2763 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764
Al Virobe08d6d2011-12-06 13:32:36 -05002765 root.mnt = mnt;
2766 root.dentry = mnt->mnt_root;
Jan Blunckac748a02008-02-14 19:34:39 -08002767
2768 set_fs_pwd(current->fs, &root);
2769 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770}
2771
Denis Cheng74bf17c2007-10-16 23:26:30 -07002772void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002774 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002775 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
Al Viro7d6fec42011-11-23 12:14:10 -05002777 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002778 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779
Ram Paib58fed82005-11-07 17:16:09 -05002780 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Al Viro84d17192013-03-15 10:53:28 -04002781 mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782
Al Viro84d17192013-03-15 10:53:28 -04002783 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 panic("Failed to allocate mount hash table\n");
2785
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002786 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787
Eric Dumazet13f14b42008-02-06 01:37:57 -08002788 for (u = 0; u < HASH_SIZE; u++)
2789 INIT_LIST_HEAD(&mount_hashtable[u]);
Al Viro84d17192013-03-15 10:53:28 -04002790 for (u = 0; u < HASH_SIZE; u++)
2791 INIT_LIST_HEAD(&mountpoint_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792
Tejun Heo4b93dc92013-11-28 14:54:43 -05002793 kernfs_init();
2794
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002795 err = sysfs_init();
2796 if (err)
2797 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002798 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002799 fs_kobj = kobject_create_and_add("fs", NULL);
2800 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002801 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 init_rootfs();
2803 init_mount_tree();
2804}
2805
Trond Myklebust616511d2009-06-22 15:09:13 -04002806void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807{
Al Virod498b252010-02-05 02:21:06 -05002808 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002809 return;
Al Viro7b00ed62013-09-16 21:19:20 -04002810 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002811 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812}
Al Viro9d412a42011-03-17 22:08:28 -04002813
2814struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2815{
Tim Chen423e0ab2011-07-19 09:32:38 -07002816 struct vfsmount *mnt;
2817 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2818 if (!IS_ERR(mnt)) {
2819 /*
2820 * it is a longterm mount, don't release mnt until
2821 * we unmount before file sys is unregistered
2822 */
Al Virof7a99c52012-06-09 00:59:08 -04002823 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07002824 }
2825 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002826}
2827EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002828
2829void kern_unmount(struct vfsmount *mnt)
2830{
2831 /* release long term mount so mount point can be released */
2832 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04002833 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04002834 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07002835 mntput(mnt);
2836 }
2837}
2838EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002839
2840bool our_mnt(struct vfsmount *mnt)
2841{
Al Viro143c8c92011-11-25 00:46:35 -05002842 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05002843}
Eric W. Biederman8823c072010-03-07 18:49:36 -08002844
Eric W. Biederman31515272013-03-15 01:45:51 -07002845bool current_chrooted(void)
2846{
2847 /* Does the current process have a non-standard root */
2848 struct path ns_root;
2849 struct path fs_root;
2850 bool chrooted;
2851
2852 /* Find the namespace root */
2853 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
2854 ns_root.dentry = ns_root.mnt->mnt_root;
2855 path_get(&ns_root);
2856 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2857 ;
2858
2859 get_fs_root(current->fs, &fs_root);
2860
2861 chrooted = !path_equal(&fs_root, &ns_root);
2862
2863 path_put(&fs_root);
2864 path_put(&ns_root);
2865
2866 return chrooted;
2867}
2868
Eric W. Biedermane51db732013-03-30 19:57:41 -07002869bool fs_fully_visible(struct file_system_type *type)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002870{
2871 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2872 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07002873 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002874
Eric W. Biedermane51db732013-03-30 19:57:41 -07002875 if (unlikely(!ns))
2876 return false;
2877
Al Viro44bb4382013-09-16 21:37:36 -04002878 down_read(&namespace_sem);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002879 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07002880 struct mount *child;
2881 if (mnt->mnt.mnt_sb->s_type != type)
2882 continue;
2883
2884 /* This mount is not fully visible if there are any child mounts
2885 * that cover anything except for empty directories.
2886 */
2887 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2888 struct inode *inode = child->mnt_mountpoint->d_inode;
2889 if (!S_ISDIR(inode->i_mode))
2890 goto next;
Eric W. Biederman41301ae2013-11-14 21:22:25 -08002891 if (inode->i_nlink > 2)
Eric W. Biedermane51db732013-03-30 19:57:41 -07002892 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002893 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07002894 visible = true;
2895 goto found;
2896 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002897 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07002898found:
Al Viro44bb4382013-09-16 21:37:36 -04002899 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07002900 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002901}
2902
Eric W. Biederman8823c072010-03-07 18:49:36 -08002903static void *mntns_get(struct task_struct *task)
2904{
2905 struct mnt_namespace *ns = NULL;
2906 struct nsproxy *nsproxy;
2907
2908 rcu_read_lock();
2909 nsproxy = task_nsproxy(task);
2910 if (nsproxy) {
2911 ns = nsproxy->mnt_ns;
2912 get_mnt_ns(ns);
2913 }
2914 rcu_read_unlock();
2915
2916 return ns;
2917}
2918
2919static void mntns_put(void *ns)
2920{
2921 put_mnt_ns(ns);
2922}
2923
2924static int mntns_install(struct nsproxy *nsproxy, void *ns)
2925{
2926 struct fs_struct *fs = current->fs;
2927 struct mnt_namespace *mnt_ns = ns;
2928 struct path root;
2929
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002930 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -07002931 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
2932 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08002933 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002934
2935 if (fs->users != 1)
2936 return -EINVAL;
2937
2938 get_mnt_ns(mnt_ns);
2939 put_mnt_ns(nsproxy->mnt_ns);
2940 nsproxy->mnt_ns = mnt_ns;
2941
2942 /* Find the root */
2943 root.mnt = &mnt_ns->root->mnt;
2944 root.dentry = mnt_ns->root->mnt.mnt_root;
2945 path_get(&root);
2946 while(d_mountpoint(root.dentry) && follow_down_one(&root))
2947 ;
2948
2949 /* Update the pwd and root */
2950 set_fs_pwd(fs, &root);
2951 set_fs_root(fs, &root);
2952
2953 path_put(&root);
2954 return 0;
2955}
2956
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002957static unsigned int mntns_inum(void *ns)
2958{
2959 struct mnt_namespace *mnt_ns = ns;
2960 return mnt_ns->proc_inum;
2961}
2962
Eric W. Biederman8823c072010-03-07 18:49:36 -08002963const struct proc_ns_operations mntns_operations = {
2964 .name = "mnt",
2965 .type = CLONE_NEWNS,
2966 .get = mntns_get,
2967 .put = mntns_put,
2968 .install = mntns_install,
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002969 .inum = mntns_inum,
Eric W. Biederman8823c072010-03-07 18:49:36 -08002970};