blob: d581e45c0a9fd6ada94edd2bb6ea4fe4c7fc9e23 [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 */
20#include <linux/ramfs.h> /* init_rootfs */
21#include <linux/fs_struct.h> /* get_fs_root et.al. */
22#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23#include <linux/uaccess.h>
Eric W. Biederman8823c072010-03-07 18:49:36 -080024#include <linux/proc_fs.h>
Ram Pai07b20882005-11-07 17:19:07 -050025#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070026#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
Eric Dumazet13f14b42008-02-06 01:37:57 -080028#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
29#define HASH_SIZE (1UL << HASH_SHIFT)
30
Al Viro5addc5d2005-11-07 17:15:49 -050031static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010032static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010033static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100034static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040035static int mnt_id_start = 0;
36static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Eric Dumazetfa3536c2006-03-26 01:37:24 -080038static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080039static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050040static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080042/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060043struct kobject *fs_kobj;
44EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080045
Nick Piggin99b7db72010-08-18 04:37:39 +100046/*
47 * vfsmount lock may be taken for read to prevent changes to the
48 * vfsmount hash, ie. during mountpoint lookups or walking back
49 * up the tree.
50 *
51 * It should be taken for write in all cases where the vfsmount
52 * tree or hash is modified or when a vfsmount structure is modified.
53 */
54DEFINE_BRLOCK(vfsmount_lock);
55
Linus Torvalds1da177e2005-04-16 15:20:36 -070056static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
57{
Ram Paib58fed82005-11-07 17:16:09 -050058 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
59 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080060 tmp = tmp + (tmp >> HASH_SHIFT);
61 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062}
63
Dave Hansen3d733632008-02-15 14:37:59 -080064#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
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
Andi Kleen962830d2012-05-08 13:32:02 +0930459 br_write_lock(&vfsmount_lock);
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;
Andi Kleen962830d2012-05-08 13:32:02 +0930493 br_write_unlock(&vfsmount_lock);
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{
Andi Kleen962830d2012-05-08 13:32:02 +0930499 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500500 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Andi Kleen962830d2012-05-08 13:32:02 +0930501 br_write_unlock(&vfsmount_lock);
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
Andi Kleen962830d2012-05-08 13:32:02 +0930513 br_write_lock(&vfsmount_lock);
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 }
Andi Kleen962830d2012-05-08 13:32:02 +0930535 br_write_unlock(&vfsmount_lock);
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
550/*
Ram Paia05964f2005-11-07 17:20:17 -0500551 * find the first or last mount at @dentry on vfsmount @mnt depending on
552 * @dir. If @dir is set return the first mount else return the last mount.
Nick Piggin99b7db72010-08-18 04:37:39 +1000553 * vfsmount_lock must be held for read or write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 */
Al Viroc7105362011-11-24 18:22:03 -0500555struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
Ram Paia05964f2005-11-07 17:20:17 -0500556 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557{
Ram Paib58fed82005-11-07 17:16:09 -0500558 struct list_head *head = mount_hashtable + hash(mnt, dentry);
559 struct list_head *tmp = head;
Al Viroc7105362011-11-24 18:22:03 -0500560 struct mount *p, *found = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500563 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 p = NULL;
565 if (tmp == head)
566 break;
Al Viro1b8e5562011-11-24 21:01:32 -0500567 p = list_entry(tmp, struct mount, mnt_hash);
Al Viroa73324d2011-11-24 22:25:07 -0500568 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500569 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 break;
571 }
572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 return found;
574}
575
Ram Paia05964f2005-11-07 17:20:17 -0500576/*
David Howellsf015f1262012-06-25 12:55:28 +0100577 * lookup_mnt - Return the first child mount mounted at path
578 *
579 * "First" means first mounted chronologically. If you create the
580 * following mounts:
581 *
582 * mount /dev/sda1 /mnt
583 * mount /dev/sda2 /mnt
584 * mount /dev/sda3 /mnt
585 *
586 * Then lookup_mnt() on the base /mnt dentry in the root mount will
587 * return successively the root dentry and vfsmount of /dev/sda1, then
588 * /dev/sda2, then /dev/sda3, then NULL.
589 *
590 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500591 */
Al Viro1c755af2009-04-18 14:06:57 -0400592struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500593{
Al Viroc7105362011-11-24 18:22:03 -0500594 struct mount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000595
Andi Kleen962830d2012-05-08 13:32:02 +0930596 br_read_lock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500597 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
598 if (child_mnt) {
599 mnt_add_count(child_mnt, 1);
Andi Kleen962830d2012-05-08 13:32:02 +0930600 br_read_unlock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500601 return &child_mnt->mnt;
602 } else {
Andi Kleen962830d2012-05-08 13:32:02 +0930603 br_read_unlock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500604 return NULL;
605 }
Ram Paia05964f2005-11-07 17:20:17 -0500606}
607
Al Viro143c8c92011-11-25 00:46:35 -0500608static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800610 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611}
612
Nick Piggin99b7db72010-08-18 04:37:39 +1000613/*
614 * vfsmount lock must be held for write
615 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800616static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500617{
618 if (ns) {
619 ns->event = ++event;
620 wake_up_interruptible(&ns->poll);
621 }
622}
623
Nick Piggin99b7db72010-08-18 04:37:39 +1000624/*
625 * vfsmount lock must be held for write
626 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800627static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500628{
629 if (ns && ns->event != event) {
630 ns->event = event;
631 wake_up_interruptible(&ns->poll);
632 }
633}
634
Nick Piggin99b7db72010-08-18 04:37:39 +1000635/*
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100636 * Clear dentry's mounted state if it has no remaining mounts.
637 * vfsmount_lock must be held for write.
638 */
Al Viroaa0a4cf2011-11-24 19:31:36 -0500639static void dentry_reset_mounted(struct dentry *dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100640{
641 unsigned u;
642
643 for (u = 0; u < HASH_SIZE; u++) {
Al Virod5e50f72011-11-24 20:58:57 -0500644 struct mount *p;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100645
Al Viro1b8e5562011-11-24 21:01:32 -0500646 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
Al Viroa73324d2011-11-24 22:25:07 -0500647 if (p->mnt_mountpoint == dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100648 return;
649 }
650 }
651 spin_lock(&dentry->d_lock);
652 dentry->d_flags &= ~DCACHE_MOUNTED;
653 spin_unlock(&dentry->d_lock);
654}
655
656/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000657 * vfsmount lock must be held for write
658 */
Al Viro419148d2011-11-24 19:41:16 -0500659static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
Al Viroa73324d2011-11-24 22:25:07 -0500661 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500662 old_path->mnt = &mnt->mnt_parent->mnt;
663 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500664 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500665 list_del_init(&mnt->mnt_child);
Al Viro1b8e5562011-11-24 21:01:32 -0500666 list_del_init(&mnt->mnt_hash);
Al Viroaa0a4cf2011-11-24 19:31:36 -0500667 dentry_reset_mounted(old_path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668}
669
Nick Piggin99b7db72010-08-18 04:37:39 +1000670/*
671 * vfsmount lock must be held for write
672 */
Al Viro14cf1fa2011-11-25 00:01:17 -0500673void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
Al Viro44d964d62011-11-24 21:28:22 -0500674 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500675{
Al Viro3a2393d2011-11-25 03:19:09 -0500676 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viroa73324d2011-11-24 22:25:07 -0500677 child_mnt->mnt_mountpoint = dget(dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500678 child_mnt->mnt_parent = mnt;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100679 spin_lock(&dentry->d_lock);
680 dentry->d_flags |= DCACHE_MOUNTED;
681 spin_unlock(&dentry->d_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500682}
683
Nick Piggin99b7db72010-08-18 04:37:39 +1000684/*
685 * vfsmount lock must be held for write
686 */
Al Viro419148d2011-11-24 19:41:16 -0500687static void attach_mnt(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688{
Al Viro14cf1fa2011-11-25 00:01:17 -0500689 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
Al Viro1b8e5562011-11-24 21:01:32 -0500690 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400691 hash(path->mnt, path->dentry));
Al Viro6b41d532011-11-24 23:24:33 -0500692 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500693}
694
695/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000696 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500697 */
Al Viro4b2619a2011-11-24 19:47:15 -0500698static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500699{
Al Viro0714a532011-11-24 22:19:58 -0500700 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500701 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500702 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500703 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500704
Al Viro0714a532011-11-24 22:19:58 -0500705 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500706
Al Viro1a4eeaf2011-11-25 02:19:55 -0500707 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400708 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500709 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500710
Ram Paib90fa9a2005-11-07 17:19:50 -0500711 list_splice(&head, n->list.prev);
712
Al Viro1b8e5562011-11-24 21:01:32 -0500713 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viroa73324d2011-11-24 22:25:07 -0500714 hash(&parent->mnt, mnt->mnt_mountpoint));
Al Viro6b41d532011-11-24 23:24:33 -0500715 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800716 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
Al Viro909b0a82011-11-25 03:06:56 -0500719static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720{
Al Viro6b41d532011-11-24 23:24:33 -0500721 struct list_head *next = p->mnt_mounts.next;
722 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500724 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500726 next = p->mnt_child.next;
727 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 break;
Al Viro0714a532011-11-24 22:19:58 -0500729 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 }
731 }
Al Viro6b41d532011-11-24 23:24:33 -0500732 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
734
Al Viro315fc832011-11-24 18:57:30 -0500735static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500736{
Al Viro6b41d532011-11-24 23:24:33 -0500737 struct list_head *prev = p->mnt_mounts.prev;
738 while (prev != &p->mnt_mounts) {
739 p = list_entry(prev, struct mount, mnt_child);
740 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500741 }
742 return p;
743}
744
Al Viro9d412a42011-03-17 22:08:28 -0400745struct vfsmount *
746vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
747{
Al Virob105e272011-11-24 20:38:33 -0500748 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400749 struct dentry *root;
750
751 if (!type)
752 return ERR_PTR(-ENODEV);
753
754 mnt = alloc_vfsmnt(name);
755 if (!mnt)
756 return ERR_PTR(-ENOMEM);
757
758 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500759 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400760
761 root = mount_fs(type, flags, name, data);
762 if (IS_ERR(root)) {
763 free_vfsmnt(mnt);
764 return ERR_CAST(root);
765 }
766
Al Virob105e272011-11-24 20:38:33 -0500767 mnt->mnt.mnt_root = root;
768 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500769 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500770 mnt->mnt_parent = mnt;
Andi Kleen962830d2012-05-08 13:32:02 +0930771 br_write_lock(&vfsmount_lock);
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100772 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Andi Kleen962830d2012-05-08 13:32:02 +0930773 br_write_unlock(&vfsmount_lock);
Al Virob105e272011-11-24 20:38:33 -0500774 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400775}
776EXPORT_SYMBOL_GPL(vfs_kern_mount);
777
Al Viro87129cc2011-11-24 21:24:27 -0500778static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500779 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780{
Al Viro87129cc2011-11-24 21:24:27 -0500781 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100782 struct mount *mnt;
783 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
David Howellsbe34d1a2012-06-25 12:55:18 +0100785 mnt = alloc_vfsmnt(old->mnt_devname);
786 if (!mnt)
787 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100788
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700789 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +0100790 mnt->mnt_group_id = 0; /* not a peer of original */
791 else
792 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100793
David Howellsbe34d1a2012-06-25 12:55:18 +0100794 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
795 err = mnt_alloc_group_id(mnt);
796 if (err)
797 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100799
800 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700801 /* Don't allow unprivileged users to change mount flags */
802 if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
803 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
804
David Howellsbe34d1a2012-06-25 12:55:18 +0100805 atomic_inc(&sb->s_active);
806 mnt->mnt.mnt_sb = sb;
807 mnt->mnt.mnt_root = dget(root);
808 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
809 mnt->mnt_parent = mnt;
810 br_write_lock(&vfsmount_lock);
811 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
812 br_write_unlock(&vfsmount_lock);
813
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700814 if ((flag & CL_SLAVE) ||
815 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +0100816 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
817 mnt->mnt_master = old;
818 CLEAR_MNT_SHARED(mnt);
819 } else if (!(flag & CL_PRIVATE)) {
820 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
821 list_add(&mnt->mnt_share, &old->mnt_share);
822 if (IS_MNT_SLAVE(old))
823 list_add(&mnt->mnt_slave, &old->mnt_slave);
824 mnt->mnt_master = old->mnt_master;
825 }
826 if (flag & CL_MAKE_SHARED)
827 set_mnt_shared(mnt);
828
829 /* stick the duplicate mount on the same expiry list
830 * as the original if that was on one */
831 if (flag & CL_EXPIRE) {
832 if (!list_empty(&old->mnt_expire))
833 list_add(&mnt->mnt_expire, &old->mnt_expire);
834 }
835
Al Virocb338d02011-11-24 20:55:08 -0500836 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100837
838 out_free:
839 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +0100840 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841}
842
Al Viro83adc752011-11-24 22:37:54 -0500843static inline void mntfree(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844{
Al Viro83adc752011-11-24 22:37:54 -0500845 struct vfsmount *m = &mnt->mnt;
846 struct super_block *sb = m->mnt_sb;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100847
Dave Hansen3d733632008-02-15 14:37:59 -0800848 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800849 * This probably indicates that somebody messed
850 * up a mnt_want/drop_write() pair. If this
851 * happens, the filesystem was probably unable
852 * to make r/w->r/o transitions.
853 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000854 /*
Nick Pigginb3e19d92011-01-07 17:50:11 +1100855 * The locking used to deal with mnt_count decrement provides barriers,
856 * so mnt_get_writers() below is safe.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000857 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100858 WARN_ON(mnt_get_writers(mnt));
Al Viro83adc752011-11-24 22:37:54 -0500859 fsnotify_vfsmount_delete(m);
860 dput(m->mnt_root);
861 free_vfsmnt(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 deactivate_super(sb);
863}
864
Al Viro900148d2011-11-25 00:33:11 -0500865static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500866{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100867put_again:
Al Virof03c6592011-01-14 22:30:21 -0500868#ifdef CONFIG_SMP
Andi Kleen962830d2012-05-08 13:32:02 +0930869 br_read_lock(&vfsmount_lock);
Al Virof7a99c52012-06-09 00:59:08 -0400870 if (likely(mnt->mnt_ns)) {
871 /* shouldn't be the last one */
Al Viroaa9c0e02011-11-23 19:04:52 -0500872 mnt_add_count(mnt, -1);
Andi Kleen962830d2012-05-08 13:32:02 +0930873 br_read_unlock(&vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500874 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100875 }
Andi Kleen962830d2012-05-08 13:32:02 +0930876 br_read_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100877
Andi Kleen962830d2012-05-08 13:32:02 +0930878 br_write_lock(&vfsmount_lock);
Al Viroaa9c0e02011-11-23 19:04:52 -0500879 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100880 if (mnt_get_count(mnt)) {
Andi Kleen962830d2012-05-08 13:32:02 +0930881 br_write_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100882 return;
883 }
Nick Pigginb3e19d92011-01-07 17:50:11 +1100884#else
Al Viroaa9c0e02011-11-23 19:04:52 -0500885 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100886 if (likely(mnt_get_count(mnt)))
Nick Piggin99b7db72010-08-18 04:37:39 +1000887 return;
Andi Kleen962830d2012-05-08 13:32:02 +0930888 br_write_lock(&vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500889#endif
Al Viro863d6842011-11-25 00:57:42 -0500890 if (unlikely(mnt->mnt_pinned)) {
891 mnt_add_count(mnt, mnt->mnt_pinned + 1);
892 mnt->mnt_pinned = 0;
Andi Kleen962830d2012-05-08 13:32:02 +0930893 br_write_unlock(&vfsmount_lock);
Al Viro900148d2011-11-25 00:33:11 -0500894 acct_auto_close_mnt(&mnt->mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100895 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500896 }
Andi Kleen962830d2012-05-08 13:32:02 +0930897
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100898 list_del(&mnt->mnt_instance);
Andi Kleen962830d2012-05-08 13:32:02 +0930899 br_write_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100900 mntfree(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500901}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100902
903void mntput(struct vfsmount *mnt)
904{
905 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -0500906 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100907 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -0500908 if (unlikely(m->mnt_expiry_mark))
909 m->mnt_expiry_mark = 0;
910 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100911 }
912}
913EXPORT_SYMBOL(mntput);
914
915struct vfsmount *mntget(struct vfsmount *mnt)
916{
917 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -0500918 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100919 return mnt;
920}
921EXPORT_SYMBOL(mntget);
922
Al Viro7b7b1ac2005-11-07 17:13:39 -0500923void mnt_pin(struct vfsmount *mnt)
924{
Andi Kleen962830d2012-05-08 13:32:02 +0930925 br_write_lock(&vfsmount_lock);
Al Viro863d6842011-11-25 00:57:42 -0500926 real_mount(mnt)->mnt_pinned++;
Andi Kleen962830d2012-05-08 13:32:02 +0930927 br_write_unlock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500928}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500929EXPORT_SYMBOL(mnt_pin);
930
Al Viro863d6842011-11-25 00:57:42 -0500931void mnt_unpin(struct vfsmount *m)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500932{
Al Viro863d6842011-11-25 00:57:42 -0500933 struct mount *mnt = real_mount(m);
Andi Kleen962830d2012-05-08 13:32:02 +0930934 br_write_lock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500935 if (mnt->mnt_pinned) {
Al Viro863d6842011-11-25 00:57:42 -0500936 mnt_add_count(mnt, 1);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500937 mnt->mnt_pinned--;
938 }
Andi Kleen962830d2012-05-08 13:32:02 +0930939 br_write_unlock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500940}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500941EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800943static inline void mangle(struct seq_file *m, const char *s)
944{
945 seq_escape(m, s, " \t\n\\");
946}
947
948/*
949 * Simple .show_options callback for filesystems which don't want to
950 * implement more complex mount option showing.
951 *
952 * See also save_mount_options().
953 */
Al Viro34c80b12011-12-08 21:32:45 -0500954int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800955{
Al Viro2a32ceb2009-05-08 16:05:57 -0400956 const char *options;
957
958 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -0500959 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800960
961 if (options != NULL && options[0]) {
962 seq_putc(m, ',');
963 mangle(m, options);
964 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400965 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800966
967 return 0;
968}
969EXPORT_SYMBOL(generic_show_options);
970
971/*
972 * If filesystem uses generic_show_options(), this function should be
973 * called from the fill_super() callback.
974 *
975 * The .remount_fs callback usually needs to be handled in a special
976 * way, to make sure, that previous options are not overwritten if the
977 * remount fails.
978 *
979 * Also note, that if the filesystem's .remount_fs function doesn't
980 * reset all options to their default value, but changes only newly
981 * given options, then the displayed options will not reflect reality
982 * any more.
983 */
984void save_mount_options(struct super_block *sb, char *options)
985{
Al Viro2a32ceb2009-05-08 16:05:57 -0400986 BUG_ON(sb->s_options);
987 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800988}
989EXPORT_SYMBOL(save_mount_options);
990
Al Viro2a32ceb2009-05-08 16:05:57 -0400991void replace_mount_options(struct super_block *sb, char *options)
992{
993 char *old = sb->s_options;
994 rcu_assign_pointer(sb->s_options, options);
995 if (old) {
996 synchronize_rcu();
997 kfree(old);
998 }
999}
1000EXPORT_SYMBOL(replace_mount_options);
1001
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001002#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001003/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004static void *m_start(struct seq_file *m, loff_t *pos)
1005{
Al Viro6ce6e242012-06-09 01:16:59 -04001006 struct proc_mounts *p = proc_mounts(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
Ram Pai390c6842005-11-07 17:17:51 -05001008 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001009 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010}
1011
1012static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1013{
Al Viro6ce6e242012-06-09 01:16:59 -04001014 struct proc_mounts *p = proc_mounts(m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001015
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001016 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
1018
1019static void m_stop(struct seq_file *m, void *v)
1020{
Ram Pai390c6842005-11-07 17:17:51 -05001021 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022}
1023
Al Viro0226f492011-12-06 12:21:54 -05001024static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001025{
Al Viro6ce6e242012-06-09 01:16:59 -04001026 struct proc_mounts *p = proc_mounts(m);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001027 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001028 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029}
1030
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001031const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 .start = m_start,
1033 .next = m_next,
1034 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001035 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001036};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001037#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039/**
1040 * may_umount_tree - check if a mount tree is busy
1041 * @mnt: root of mount tree
1042 *
1043 * This is called to check if a tree of mounts has any
1044 * open files, pwds, chroots or sub mounts that are
1045 * busy.
1046 */
Al Viro909b0a82011-11-25 03:06:56 -05001047int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048{
Al Viro909b0a82011-11-25 03:06:56 -05001049 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001050 int actual_refs = 0;
1051 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001052 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001053 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Nick Pigginb3e19d92011-01-07 17:50:11 +11001055 /* write lock needed for mnt_get_count */
Andi Kleen962830d2012-05-08 13:32:02 +09301056 br_write_lock(&vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -05001057 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001058 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 }
Andi Kleen962830d2012-05-08 13:32:02 +09301061 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
1063 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001064 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
Ian Kente3474a82006-03-27 01:14:51 -08001066 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067}
1068
1069EXPORT_SYMBOL(may_umount_tree);
1070
1071/**
1072 * may_umount - check if a mount point is busy
1073 * @mnt: root of mount
1074 *
1075 * This is called to check if a mount point has any
1076 * open files, pwds, chroots or sub mounts. If the
1077 * mount has sub mounts this will return busy
1078 * regardless of whether the sub mounts are busy.
1079 *
1080 * Doesn't take quota and stuff into account. IOW, in some cases it will
1081 * give false negatives. The main reason why it's here is that we need
1082 * a non-destructive way to look for easily umountable filesystems.
1083 */
1084int may_umount(struct vfsmount *mnt)
1085{
Ian Kente3474a82006-03-27 01:14:51 -08001086 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001087 down_read(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301088 br_write_lock(&vfsmount_lock);
Al Viro1ab59732011-11-24 21:35:16 -05001089 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001090 ret = 0;
Andi Kleen962830d2012-05-08 13:32:02 +09301091 br_write_unlock(&vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001092 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001093 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094}
1095
1096EXPORT_SYMBOL(may_umount);
1097
Ram Paib90fa9a2005-11-07 17:19:50 -05001098void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099{
Al Virod5e50f72011-11-24 20:58:57 -05001100 struct mount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001101 while (!list_empty(head)) {
Al Viro1b8e5562011-11-24 21:01:32 -05001102 mnt = list_first_entry(head, struct mount, mnt_hash);
1103 list_del_init(&mnt->mnt_hash);
Al Viro676da582011-11-24 21:47:05 -05001104 if (mnt_has_parent(mnt)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001105 struct dentry *dentry;
Al Viro863d6842011-11-25 00:57:42 -05001106 struct mount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001107
Andi Kleen962830d2012-05-08 13:32:02 +09301108 br_write_lock(&vfsmount_lock);
Al Viroa73324d2011-11-24 22:25:07 -05001109 dentry = mnt->mnt_mountpoint;
Al Viro863d6842011-11-25 00:57:42 -05001110 m = mnt->mnt_parent;
Al Viroa73324d2011-11-24 22:25:07 -05001111 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -05001112 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001113 m->mnt_ghosts--;
Andi Kleen962830d2012-05-08 13:32:02 +09301114 br_write_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001115 dput(dentry);
Al Viro863d6842011-11-25 00:57:42 -05001116 mntput(&m->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 }
Al Virod5e50f72011-11-24 20:58:57 -05001118 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001119 }
1120}
1121
Nick Piggin99b7db72010-08-18 04:37:39 +10001122/*
1123 * vfsmount lock must be held for write
1124 * namespace_sem must be held for write
1125 */
Al Viro761d5c32011-11-24 21:07:43 -05001126void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001127{
Al Viro7b8a53f2011-01-15 20:08:44 -05001128 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001129 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001130
Al Viro909b0a82011-11-25 03:06:56 -05001131 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro1b8e5562011-11-24 21:01:32 -05001132 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001133
Ram Paia05964f2005-11-07 17:20:17 -05001134 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001135 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001136
Al Viro1b8e5562011-11-24 21:01:32 -05001137 list_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001138 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001139 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001140 __touch_mnt_namespace(p->mnt_ns);
1141 p->mnt_ns = NULL;
Al Viro6b41d532011-11-24 23:24:33 -05001142 list_del_init(&p->mnt_child);
Al Viro676da582011-11-24 21:47:05 -05001143 if (mnt_has_parent(p)) {
Al Viro863d6842011-11-25 00:57:42 -05001144 p->mnt_parent->mnt_ghosts++;
Al Viroa73324d2011-11-24 22:25:07 -05001145 dentry_reset_mounted(p->mnt_mountpoint);
Al Viro7c4b93d2008-03-21 23:59:49 -04001146 }
Al Viro0f0afb12011-11-24 20:43:10 -05001147 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 }
Al Viro7b8a53f2011-01-15 20:08:44 -05001149 list_splice(&tmp_list, kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150}
1151
Al Viro692afc32011-11-24 21:15:14 -05001152static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
Al Viroc35038b2008-03-22 00:46:23 -04001153
Al Viro1ab59732011-11-24 21:35:16 -05001154static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155{
Al Viro1ab59732011-11-24 21:35:16 -05001156 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001158 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
Al Viro1ab59732011-11-24 21:35:16 -05001160 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 if (retval)
1162 return retval;
1163
1164 /*
1165 * Allow userspace to request a mountpoint be expired rather than
1166 * unmounting unconditionally. Unmount only happens if:
1167 * (1) the mark is already set (the mark is cleared by mntput())
1168 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1169 */
1170 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001171 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 flags & (MNT_FORCE | MNT_DETACH))
1173 return -EINVAL;
1174
Nick Pigginb3e19d92011-01-07 17:50:11 +11001175 /*
1176 * probably don't strictly need the lock here if we examined
1177 * all race cases, but it's a slowpath.
1178 */
Andi Kleen962830d2012-05-08 13:32:02 +09301179 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -05001180 if (mnt_get_count(mnt) != 2) {
Andi Kleen962830d2012-05-08 13:32:02 +09301181 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001183 }
Andi Kleen962830d2012-05-08 13:32:02 +09301184 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Al Viro863d6842011-11-25 00:57:42 -05001186 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 return -EAGAIN;
1188 }
1189
1190 /*
1191 * If we may have to abort operations to get out of this
1192 * mount, and they will themselves hold resources we must
1193 * allow the fs to do things. In the Unix tradition of
1194 * 'Gee thats tricky lets do it in userspace' the umount_begin
1195 * might fail to complete on the first run through as other tasks
1196 * must return, and the like. Thats for the mount program to worry
1197 * about for the moment.
1198 */
1199
Al Viro42faad92008-04-24 07:21:56 -04001200 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001201 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001202 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
1204 /*
1205 * No sense to grab the lock for this test, but test itself looks
1206 * somewhat bogus. Suggestions for better replacement?
1207 * Ho-hum... In principle, we might treat that as umount + switch
1208 * to rootfs. GC would eventually take care of the old vfsmount.
1209 * Actually it makes sense, especially if rootfs would contain a
1210 * /reboot - static binary that would close all descriptors and
1211 * call reboot(9). Then init(8) could umount root and exec /reboot.
1212 */
Al Viro1ab59732011-11-24 21:35:16 -05001213 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 /*
1215 * Special case for "unmounting" root ...
1216 * we just try to remount it readonly.
1217 */
1218 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001219 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 up_write(&sb->s_umount);
1222 return retval;
1223 }
1224
Ram Pai390c6842005-11-07 17:17:51 -05001225 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301226 br_write_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001227 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
Al Viroc35038b2008-03-22 00:46:23 -04001229 if (!(flags & MNT_DETACH))
Al Viro1ab59732011-11-24 21:35:16 -05001230 shrink_submounts(mnt, &umount_list);
Al Viroc35038b2008-03-22 00:46:23 -04001231
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001233 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001234 if (!list_empty(&mnt->mnt_list))
Al Viro1ab59732011-11-24 21:35:16 -05001235 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 retval = 0;
1237 }
Andi Kleen962830d2012-05-08 13:32:02 +09301238 br_write_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001239 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001240 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 return retval;
1242}
1243
Al Viro9b40bc92013-02-22 22:45:42 -05001244/*
1245 * Is the caller allowed to modify his namespace?
1246 */
1247static inline bool may_mount(void)
1248{
1249 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1250}
1251
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252/*
1253 * Now umount can handle mount points as well as block devices.
1254 * This is important for filesystems which use unnamed block devices.
1255 *
1256 * We now support a flag for forced unmount like the other 'big iron'
1257 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1258 */
1259
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001260SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
Al Viro2d8f3032008-07-22 09:59:21 -04001262 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001263 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001265 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001267 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1268 return -EINVAL;
1269
Al Viro9b40bc92013-02-22 22:45:42 -05001270 if (!may_mount())
1271 return -EPERM;
1272
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001273 if (!(flags & UMOUNT_NOFOLLOW))
1274 lookup_flags |= LOOKUP_FOLLOW;
1275
1276 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 if (retval)
1278 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001279 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001281 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001283 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 goto dput_and_out;
1285
Al Viro900148d2011-11-25 00:33:11 -05001286 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001288 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001289 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001290 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291out:
1292 return retval;
1293}
1294
1295#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1296
1297/*
Ram Paib58fed82005-11-07 17:16:09 -05001298 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001300SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301{
Ram Paib58fed82005-11-07 17:16:09 -05001302 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303}
1304
1305#endif
1306
Eric W. Biederman8823c072010-03-07 18:49:36 -08001307static bool mnt_ns_loop(struct path *path)
1308{
1309 /* Could bind mounting the mount namespace inode cause a
1310 * mount namespace loop?
1311 */
1312 struct inode *inode = path->dentry->d_inode;
1313 struct proc_inode *ei;
1314 struct mnt_namespace *mnt_ns;
1315
1316 if (!proc_ns_inode(inode))
1317 return false;
1318
1319 ei = PROC_I(inode);
1320 if (ei->ns_ops != &mntns_operations)
1321 return false;
1322
1323 mnt_ns = ei->ns;
1324 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1325}
1326
Al Viro87129cc2011-11-24 21:24:27 -05001327struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001328 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329{
Al Viroa73324d2011-11-24 22:25:07 -05001330 struct mount *res, *p, *q, *r;
Al Viro1a390682008-03-21 20:48:19 -04001331 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
Al Virofc7be132011-11-25 01:05:37 -05001333 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
David Howellsbe34d1a2012-06-25 12:55:18 +01001334 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001335
Ram Pai36341f62005-11-07 17:17:22 -05001336 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001337 if (IS_ERR(q))
1338 return q;
1339
Al Viroa73324d2011-11-24 22:25:07 -05001340 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
1342 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001343 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001344 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001345 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 continue;
1347
Al Viro909b0a82011-11-25 03:06:56 -05001348 for (s = r; s; s = next_mnt(s, r)) {
Al Virofc7be132011-11-25 01:05:37 -05001349 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001350 s = skip_mnt_tree(s);
1351 continue;
1352 }
Al Viro0714a532011-11-24 22:19:58 -05001353 while (p != s->mnt_parent) {
1354 p = p->mnt_parent;
1355 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
Al Viro87129cc2011-11-24 21:24:27 -05001357 p = s;
Al Virocb338d02011-11-24 20:55:08 -05001358 path.mnt = &q->mnt;
Al Viroa73324d2011-11-24 22:25:07 -05001359 path.dentry = p->mnt_mountpoint;
Al Viro87129cc2011-11-24 21:24:27 -05001360 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001361 if (IS_ERR(q))
1362 goto out;
Andi Kleen962830d2012-05-08 13:32:02 +09301363 br_write_lock(&vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001364 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Virocb338d02011-11-24 20:55:08 -05001365 attach_mnt(q, &path);
Andi Kleen962830d2012-05-08 13:32:02 +09301366 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 }
1368 }
1369 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001370out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001372 LIST_HEAD(umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301373 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001374 umount_tree(res, 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301375 br_write_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001376 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001378 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379}
1380
David Howellsbe34d1a2012-06-25 12:55:18 +01001381/* Caller should check returned pointer for errors */
1382
Al Viro589ff872009-04-18 03:28:19 -04001383struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001384{
Al Virocb338d02011-11-24 20:55:08 -05001385 struct mount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001386 down_write(&namespace_sem);
Al Viro87129cc2011-11-24 21:24:27 -05001387 tree = copy_tree(real_mount(path->mnt), path->dentry,
1388 CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001389 up_write(&namespace_sem);
David Howellsbe34d1a2012-06-25 12:55:18 +01001390 if (IS_ERR(tree))
1391 return NULL;
1392 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001393}
1394
1395void drop_collected_mounts(struct vfsmount *mnt)
1396{
1397 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001398 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301399 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001400 umount_tree(real_mount(mnt), 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301401 br_write_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001402 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001403 release_mounts(&umount_list);
1404}
1405
Al Viro1f707132010-01-30 22:51:25 -05001406int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1407 struct vfsmount *root)
1408{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001409 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001410 int res = f(root, arg);
1411 if (res)
1412 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001413 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1414 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001415 if (res)
1416 return res;
1417 }
1418 return 0;
1419}
1420
Al Viro4b8b21f2011-11-24 19:54:23 -05001421static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001422{
Al Viro315fc832011-11-24 18:57:30 -05001423 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001424
Al Viro909b0a82011-11-25 03:06:56 -05001425 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001426 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001427 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001428 }
1429}
1430
Al Viro4b8b21f2011-11-24 19:54:23 -05001431static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001432{
Al Viro315fc832011-11-24 18:57:30 -05001433 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001434
Al Viro909b0a82011-11-25 03:06:56 -05001435 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001436 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001437 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001438 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001439 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001440 return err;
1441 }
1442 }
1443 }
1444
1445 return 0;
1446}
1447
Ram Paib90fa9a2005-11-07 17:19:50 -05001448/*
1449 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001450 * @nd : place the mount tree @source_mnt is attached
1451 * @parent_nd : if non-null, detach the source_mnt from its parent and
1452 * store the parent mount and mountpoint dentry.
1453 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001454 *
1455 * NOTE: in the table below explains the semantics when a source mount
1456 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001457 * ---------------------------------------------------------------------------
1458 * | BIND MOUNT OPERATION |
1459 * |**************************************************************************
1460 * | source-->| shared | private | slave | unbindable |
1461 * | dest | | | | |
1462 * | | | | | | |
1463 * | v | | | | |
1464 * |**************************************************************************
1465 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1466 * | | | | | |
1467 * |non-shared| shared (+) | private | slave (*) | invalid |
1468 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001469 * A bind operation clones the source mount and mounts the clone on the
1470 * destination mount.
1471 *
1472 * (++) the cloned mount is propagated to all the mounts in the propagation
1473 * tree of the destination mount and the cloned mount is added to
1474 * the peer group of the source mount.
1475 * (+) the cloned mount is created under the destination mount and is marked
1476 * as shared. The cloned mount is added to the peer group of the source
1477 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001478 * (+++) the mount is propagated to all the mounts in the propagation tree
1479 * of the destination mount and the cloned mount is made slave
1480 * of the same master as that of the source mount. The cloned mount
1481 * is marked as 'shared and slave'.
1482 * (*) the cloned mount is made a slave of the same master as that of the
1483 * source mount.
1484 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001485 * ---------------------------------------------------------------------------
1486 * | MOVE MOUNT OPERATION |
1487 * |**************************************************************************
1488 * | source-->| shared | private | slave | unbindable |
1489 * | dest | | | | |
1490 * | | | | | | |
1491 * | v | | | | |
1492 * |**************************************************************************
1493 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1494 * | | | | | |
1495 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1496 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001497 *
1498 * (+) the mount is moved to the destination. And is then propagated to
1499 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001500 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001501 * (+++) the mount is moved to the destination and is then propagated to
1502 * all the mounts belonging to the destination mount's propagation tree.
1503 * the mount is marked as 'shared and slave'.
1504 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001505 *
1506 * if the source mount is a tree, the operations explained above is
1507 * applied to each mount in the tree.
1508 * Must be called without spinlocks held, since this function can sleep
1509 * in allocations.
1510 */
Al Viro0fb54e52011-11-24 19:59:16 -05001511static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001512 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001513{
1514 LIST_HEAD(tree_list);
Al Viroa8d56d82011-11-24 23:59:29 -05001515 struct mount *dest_mnt = real_mount(path->mnt);
Al Viro1a390682008-03-21 20:48:19 -04001516 struct dentry *dest_dentry = path->dentry;
Al Viro315fc832011-11-24 18:57:30 -05001517 struct mount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001518 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001519
Al Virofc7be132011-11-25 01:05:37 -05001520 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001521 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001522 if (err)
1523 goto out;
1524 }
Al Viroa8d56d82011-11-24 23:59:29 -05001525 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001526 if (err)
1527 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001528
Andi Kleen962830d2012-05-08 13:32:02 +09301529 br_write_lock(&vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001530
Al Virofc7be132011-11-25 01:05:37 -05001531 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro909b0a82011-11-25 03:06:56 -05001532 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001533 set_mnt_shared(p);
Ram Paib90fa9a2005-11-07 17:19:50 -05001534 }
Al Viro1a390682008-03-21 20:48:19 -04001535 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001536 detach_mnt(source_mnt, parent_path);
1537 attach_mnt(source_mnt, path);
Al Viro143c8c92011-11-25 00:46:35 -05001538 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001539 } else {
Al Viro14cf1fa2011-11-25 00:01:17 -05001540 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
Al Viro0fb54e52011-11-24 19:59:16 -05001541 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05001542 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001543
Al Viro1b8e5562011-11-24 21:01:32 -05001544 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1545 list_del_init(&child->mnt_hash);
Al Viro4b2619a2011-11-24 19:47:15 -05001546 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05001547 }
Andi Kleen962830d2012-05-08 13:32:02 +09301548 br_write_unlock(&vfsmount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +10001549
Ram Paib90fa9a2005-11-07 17:19:50 -05001550 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001551
1552 out_cleanup_ids:
Al Virofc7be132011-11-25 01:05:37 -05001553 if (IS_MNT_SHARED(dest_mnt))
Al Viro0fb54e52011-11-24 19:59:16 -05001554 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001555 out:
1556 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001557}
1558
Al Virob12cea92011-03-18 08:55:38 -04001559static int lock_mount(struct path *path)
1560{
1561 struct vfsmount *mnt;
1562retry:
1563 mutex_lock(&path->dentry->d_inode->i_mutex);
1564 if (unlikely(cant_mount(path->dentry))) {
1565 mutex_unlock(&path->dentry->d_inode->i_mutex);
1566 return -ENOENT;
1567 }
1568 down_write(&namespace_sem);
1569 mnt = lookup_mnt(path);
1570 if (likely(!mnt))
1571 return 0;
1572 up_write(&namespace_sem);
1573 mutex_unlock(&path->dentry->d_inode->i_mutex);
1574 path_put(path);
1575 path->mnt = mnt;
1576 path->dentry = dget(mnt->mnt_root);
1577 goto retry;
1578}
1579
1580static void unlock_mount(struct path *path)
1581{
1582 up_write(&namespace_sem);
1583 mutex_unlock(&path->dentry->d_inode->i_mutex);
1584}
1585
Al Viro95bc5f22011-11-25 00:30:56 -05001586static int graft_tree(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587{
Al Viro95bc5f22011-11-25 00:30:56 -05001588 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 return -EINVAL;
1590
Al Viro8c3ee422008-03-22 18:00:39 -04001591 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001592 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 return -ENOTDIR;
1594
Al Virob12cea92011-03-18 08:55:38 -04001595 if (d_unlinked(path->dentry))
1596 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
Al Viro95bc5f22011-11-25 00:30:56 -05001598 return attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599}
1600
1601/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001602 * Sanity check the flags to change_mnt_propagation.
1603 */
1604
1605static int flags_to_propagation_type(int flags)
1606{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001607 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001608
1609 /* Fail if any non-propagation flags are set */
1610 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1611 return 0;
1612 /* Only one propagation flag should be set */
1613 if (!is_power_of_2(type))
1614 return 0;
1615 return type;
1616}
1617
1618/*
Ram Pai07b20882005-11-07 17:19:07 -05001619 * recursively change the type of the mountpoint.
1620 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001621static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001622{
Al Viro315fc832011-11-24 18:57:30 -05001623 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001624 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001625 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001626 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001627 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001628
Al Viro2d92ab32008-08-02 00:51:11 -04001629 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001630 return -EINVAL;
1631
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001632 type = flags_to_propagation_type(flag);
1633 if (!type)
1634 return -EINVAL;
1635
Ram Pai07b20882005-11-07 17:19:07 -05001636 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001637 if (type == MS_SHARED) {
1638 err = invent_group_ids(mnt, recurse);
1639 if (err)
1640 goto out_unlock;
1641 }
1642
Andi Kleen962830d2012-05-08 13:32:02 +09301643 br_write_lock(&vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -05001644 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001645 change_mnt_propagation(m, type);
Andi Kleen962830d2012-05-08 13:32:02 +09301646 br_write_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001647
1648 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001649 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001650 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001651}
1652
1653/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 * do loopback mount.
1655 */
Al Viro808d4e32012-10-11 11:42:01 -04001656static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001657 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658{
Al Virob12cea92011-03-18 08:55:38 -04001659 LIST_HEAD(umount_list);
Al Viro2d92ab32008-08-02 00:51:11 -04001660 struct path old_path;
Al Viro87129cc2011-11-24 21:24:27 -05001661 struct mount *mnt = NULL, *old;
Al Viro57eccb82013-02-22 22:49:10 -05001662 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 if (!old_name || !*old_name)
1664 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001665 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 if (err)
1667 return err;
1668
Eric W. Biederman8823c072010-03-07 18:49:36 -08001669 err = -EINVAL;
1670 if (mnt_ns_loop(&old_path))
1671 goto out;
1672
Al Virob12cea92011-03-18 08:55:38 -04001673 err = lock_mount(path);
1674 if (err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001675 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001676
Al Viro87129cc2011-11-24 21:24:27 -05001677 old = real_mount(old_path.mnt);
1678
Al Virob12cea92011-03-18 08:55:38 -04001679 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001680 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04001681 goto out2;
1682
Al Viro143c8c92011-11-25 00:46:35 -05001683 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
Al Virob12cea92011-03-18 08:55:38 -04001684 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
Al Viroccd48bc2005-11-07 17:15:04 -05001686 if (recurse)
Al Viro87129cc2011-11-24 21:24:27 -05001687 mnt = copy_tree(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001688 else
Al Viro87129cc2011-11-24 21:24:27 -05001689 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001690
David Howellsbe34d1a2012-06-25 12:55:18 +01001691 if (IS_ERR(mnt)) {
1692 err = PTR_ERR(mnt);
1693 goto out;
1694 }
Al Viroccd48bc2005-11-07 17:15:04 -05001695
Al Viro95bc5f22011-11-25 00:30:56 -05001696 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001697 if (err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301698 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001699 umount_tree(mnt, 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301700 br_write_unlock(&vfsmount_lock);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001701 }
Al Virob12cea92011-03-18 08:55:38 -04001702out2:
1703 unlock_mount(path);
1704 release_mounts(&umount_list);
Al Viroccd48bc2005-11-07 17:15:04 -05001705out:
Al Viro2d92ab32008-08-02 00:51:11 -04001706 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 return err;
1708}
1709
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001710static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1711{
1712 int error = 0;
1713 int readonly_request = 0;
1714
1715 if (ms_flags & MS_RDONLY)
1716 readonly_request = 1;
1717 if (readonly_request == __mnt_is_readonly(mnt))
1718 return 0;
1719
Eric W. Biederman90563b12013-03-22 03:10:15 -07001720 if (mnt->mnt_flags & MNT_LOCK_READONLY)
1721 return -EPERM;
1722
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001723 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05001724 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001725 else
Al Viro83adc752011-11-24 22:37:54 -05001726 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001727 return error;
1728}
1729
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730/*
1731 * change filesystem flags. dir should be a physical root of filesystem.
1732 * If you've mounted a non-root directory somewhere and want to do remount
1733 * on it - tough luck.
1734 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001735static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 void *data)
1737{
1738 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001739 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05001740 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
Al Viro143c8c92011-11-25 00:46:35 -05001742 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 return -EINVAL;
1744
Al Viro2d92ab32008-08-02 00:51:11 -04001745 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 return -EINVAL;
1747
Eric Parisff36fe22011-03-03 16:09:14 -05001748 err = security_sb_remount(sb, data);
1749 if (err)
1750 return err;
1751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001753 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001754 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05001755 else if (!capable(CAP_SYS_ADMIN))
1756 err = -EPERM;
Al Viro4aa98cf2009-05-08 13:36:58 -04001757 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001758 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001759 if (!err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301760 br_write_lock(&vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001761 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1762 mnt->mnt.mnt_flags = mnt_flags;
Andi Kleen962830d2012-05-08 13:32:02 +09301763 br_write_unlock(&vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001766 if (!err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301767 br_write_lock(&vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001768 touch_mnt_namespace(mnt->mnt_ns);
Andi Kleen962830d2012-05-08 13:32:02 +09301769 br_write_unlock(&vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001770 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 return err;
1772}
1773
Al Virocbbe3622011-11-24 20:01:19 -05001774static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05001775{
Al Viro315fc832011-11-24 18:57:30 -05001776 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001777 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001778 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05001779 return 1;
1780 }
1781 return 0;
1782}
1783
Al Viro808d4e32012-10-11 11:42:01 -04001784static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785{
Al Viro2d92ab32008-08-02 00:51:11 -04001786 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05001787 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05001788 struct mount *old;
Al Viro57eccb82013-02-22 22:49:10 -05001789 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 if (!old_name || !*old_name)
1791 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001792 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 if (err)
1794 return err;
1795
Al Virob12cea92011-03-18 08:55:38 -04001796 err = lock_mount(path);
David Howellscc53ce52011-01-14 18:45:26 +00001797 if (err < 0)
1798 goto out;
1799
Al Viro143c8c92011-11-25 00:46:35 -05001800 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05001801 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05001802
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001804 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 goto out1;
1806
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001807 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001808 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809
1810 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001811 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001812 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
Al Viro676da582011-11-24 21:47:05 -05001814 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05001815 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816
Al Viro2d92ab32008-08-02 00:51:11 -04001817 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1818 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001819 goto out1;
1820 /*
1821 * Don't move a mount residing in a shared parent.
1822 */
Al Virofc7be132011-11-25 01:05:37 -05001823 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001824 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001825 /*
1826 * Don't move a mount tree containing unbindable mounts to a destination
1827 * mount which is shared.
1828 */
Al Virofc7be132011-11-25 01:05:37 -05001829 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05001830 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05001832 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05001833 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05001834 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
Al Viro0fb54e52011-11-24 19:59:16 -05001836 err = attach_recursive_mnt(old, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001837 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001838 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
1840 /* if the mount is moved, it should no longer be expire
1841 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05001842 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843out1:
Al Virob12cea92011-03-18 08:55:38 -04001844 unlock_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001847 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001848 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 return err;
1850}
1851
Al Viro9d412a42011-03-17 22:08:28 -04001852static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1853{
1854 int err;
1855 const char *subtype = strchr(fstype, '.');
1856 if (subtype) {
1857 subtype++;
1858 err = -EINVAL;
1859 if (!subtype[0])
1860 goto err;
1861 } else
1862 subtype = "";
1863
1864 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1865 err = -ENOMEM;
1866 if (!mnt->mnt_sb->s_subtype)
1867 goto err;
1868 return mnt;
1869
1870 err:
1871 mntput(mnt);
1872 return ERR_PTR(err);
1873}
1874
Al Viro9d412a42011-03-17 22:08:28 -04001875/*
1876 * add a mount into a namespace's mount tree
1877 */
Al Viro95bc5f22011-11-25 00:30:56 -05001878static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04001879{
1880 int err;
1881
1882 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1883
Al Virob12cea92011-03-18 08:55:38 -04001884 err = lock_mount(path);
1885 if (err)
1886 return err;
Al Viro9d412a42011-03-17 22:08:28 -04001887
1888 err = -EINVAL;
Al Viro156cacb2012-09-21 08:19:02 -04001889 if (unlikely(!check_mnt(real_mount(path->mnt)))) {
1890 /* that's acceptable only for automounts done in private ns */
1891 if (!(mnt_flags & MNT_SHRINKABLE))
1892 goto unlock;
1893 /* ... and for those we'd better have mountpoint still alive */
1894 if (!real_mount(path->mnt)->mnt_ns)
1895 goto unlock;
1896 }
Al Viro9d412a42011-03-17 22:08:28 -04001897
1898 /* Refuse the same filesystem on the same mount point */
1899 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05001900 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04001901 path->mnt->mnt_root == path->dentry)
1902 goto unlock;
1903
1904 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05001905 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04001906 goto unlock;
1907
Al Viro95bc5f22011-11-25 00:30:56 -05001908 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro9d412a42011-03-17 22:08:28 -04001909 err = graft_tree(newmnt, path);
1910
1911unlock:
Al Virob12cea92011-03-18 08:55:38 -04001912 unlock_mount(path);
Al Viro9d412a42011-03-17 22:08:28 -04001913 return err;
1914}
Al Virob1e75df2011-01-17 01:47:59 -05001915
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916/*
1917 * create a new mount for userspace and request it to be added into the
1918 * namespace's tree
1919 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001920static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04001921 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001923 struct file_system_type *type;
Al Viro9b40bc92013-02-22 22:45:42 -05001924 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05001926 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001928 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 return -EINVAL;
1930
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001931 type = get_fs_type(fstype);
1932 if (!type)
1933 return -ENODEV;
1934
1935 if (user_ns != &init_user_ns) {
1936 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
1937 put_filesystem(type);
1938 return -EPERM;
1939 }
1940 /* Only in special cases allow devices from mounts
1941 * created outside the initial user namespace.
1942 */
1943 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
1944 flags |= MS_NODEV;
1945 mnt_flags |= MNT_NODEV;
1946 }
1947 }
1948
1949 mnt = vfs_kern_mount(type, flags, name, data);
1950 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1951 !mnt->mnt_sb->s_subtype)
1952 mnt = fs_set_subtype(mnt, fstype);
1953
1954 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 if (IS_ERR(mnt))
1956 return PTR_ERR(mnt);
1957
Al Viro95bc5f22011-11-25 00:30:56 -05001958 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05001959 if (err)
1960 mntput(mnt);
1961 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962}
1963
Al Viro19a167a2011-01-17 01:35:23 -05001964int finish_automount(struct vfsmount *m, struct path *path)
1965{
Al Viro6776db3d2011-11-25 00:22:05 -05001966 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05001967 int err;
1968 /* The new mount record should have at least 2 refs to prevent it being
1969 * expired before we get a chance to add it
1970 */
Al Viro6776db3d2011-11-25 00:22:05 -05001971 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05001972
1973 if (m->mnt_sb == path->mnt->mnt_sb &&
1974 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05001975 err = -ELOOP;
1976 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05001977 }
1978
Al Viro95bc5f22011-11-25 00:30:56 -05001979 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05001980 if (!err)
1981 return 0;
1982fail:
1983 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05001984 if (!list_empty(&mnt->mnt_expire)) {
Al Virob1e75df2011-01-17 01:47:59 -05001985 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301986 br_write_lock(&vfsmount_lock);
Al Viro6776db3d2011-11-25 00:22:05 -05001987 list_del_init(&mnt->mnt_expire);
Andi Kleen962830d2012-05-08 13:32:02 +09301988 br_write_unlock(&vfsmount_lock);
Al Virob1e75df2011-01-17 01:47:59 -05001989 up_write(&namespace_sem);
Al Viro19a167a2011-01-17 01:35:23 -05001990 }
Al Virob1e75df2011-01-17 01:47:59 -05001991 mntput(m);
1992 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05001993 return err;
1994}
1995
David Howellsea5b7782011-01-14 19:10:03 +00001996/**
1997 * mnt_set_expiry - Put a mount on an expiration list
1998 * @mnt: The mount to list.
1999 * @expiry_list: The list to add the mount to.
2000 */
2001void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2002{
2003 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09302004 br_write_lock(&vfsmount_lock);
David Howellsea5b7782011-01-14 19:10:03 +00002005
Al Viro6776db3d2011-11-25 00:22:05 -05002006 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002007
Andi Kleen962830d2012-05-08 13:32:02 +09302008 br_write_unlock(&vfsmount_lock);
David Howellsea5b7782011-01-14 19:10:03 +00002009 up_write(&namespace_sem);
2010}
2011EXPORT_SYMBOL(mnt_set_expiry);
2012
2013/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 * process a list of expirable mountpoints with the intent of discarding any
2015 * mountpoints that aren't in use and haven't been touched since last we came
2016 * here
2017 */
2018void mark_mounts_for_expiry(struct list_head *mounts)
2019{
Al Viro761d5c32011-11-24 21:07:43 -05002020 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04002022 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
2024 if (list_empty(mounts))
2025 return;
2026
Al Virobcc5c7d2008-03-22 00:21:53 -04002027 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09302028 br_write_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
2030 /* extract from the expiration list every vfsmount that matches the
2031 * following criteria:
2032 * - only referenced by its parent vfsmount
2033 * - still marked for expiry (marked on the last call here; marks are
2034 * cleared by mntput())
2035 */
Al Viro6776db3d2011-11-25 00:22:05 -05002036 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002037 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002038 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002040 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002042 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002043 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002044 touch_mnt_namespace(mnt->mnt_ns);
Al Virobcc5c7d2008-03-22 00:21:53 -04002045 umount_tree(mnt, 1, &umounts);
2046 }
Andi Kleen962830d2012-05-08 13:32:02 +09302047 br_write_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04002048 up_write(&namespace_sem);
2049
2050 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051}
2052
2053EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2054
2055/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002056 * Ripoff of 'select_parent()'
2057 *
2058 * search the list of submounts for a given mountpoint, and move any
2059 * shrinkable submounts to the 'graveyard' list.
2060 */
Al Viro692afc32011-11-24 21:15:14 -05002061static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002062{
Al Viro692afc32011-11-24 21:15:14 -05002063 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002064 struct list_head *next;
2065 int found = 0;
2066
2067repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002068 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002069resume:
Al Viro6b41d532011-11-24 23:24:33 -05002070 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002071 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002072 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002073
2074 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002075 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002076 continue;
2077 /*
2078 * Descend a level if the d_mounts list is non-empty.
2079 */
Al Viro6b41d532011-11-24 23:24:33 -05002080 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002081 this_parent = mnt;
2082 goto repeat;
2083 }
2084
Al Viro1ab59732011-11-24 21:35:16 -05002085 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002086 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002087 found++;
2088 }
2089 }
2090 /*
2091 * All done at this level ... ascend and resume the search
2092 */
2093 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002094 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002095 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002096 goto resume;
2097 }
2098 return found;
2099}
2100
2101/*
2102 * process a list of expirable mountpoints with the intent of discarding any
2103 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002104 *
2105 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002106 */
Al Viro692afc32011-11-24 21:15:14 -05002107static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002108{
2109 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002110 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002111
Trond Myklebust5528f9112006-06-09 09:34:17 -04002112 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002113 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002114 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002115 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002116 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002117 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08002118 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04002119 }
2120 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002121}
2122
Trond Myklebust5528f9112006-06-09 09:34:17 -04002123/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 * Some copy_from_user() implementations do not return the exact number of
2125 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2126 * Note that this function differs from copy_from_user() in that it will oops
2127 * on bad values of `to', rather than returning a short copy.
2128 */
Ram Paib58fed82005-11-07 17:16:09 -05002129static long exact_copy_from_user(void *to, const void __user * from,
2130 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131{
2132 char *t = to;
2133 const char __user *f = from;
2134 char c;
2135
2136 if (!access_ok(VERIFY_READ, from, n))
2137 return n;
2138
2139 while (n) {
2140 if (__get_user(c, f)) {
2141 memset(t, 0, n);
2142 break;
2143 }
2144 *t++ = c;
2145 f++;
2146 n--;
2147 }
2148 return n;
2149}
2150
Ram Paib58fed82005-11-07 17:16:09 -05002151int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152{
2153 int i;
2154 unsigned long page;
2155 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002156
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 *where = 0;
2158 if (!data)
2159 return 0;
2160
2161 if (!(page = __get_free_page(GFP_KERNEL)))
2162 return -ENOMEM;
2163
2164 /* We only care that *some* data at the address the user
2165 * gave us is valid. Just in case, we'll zero
2166 * the remainder of the page.
2167 */
2168 /* copy_from_user cannot cross TASK_SIZE ! */
2169 size = TASK_SIZE - (unsigned long)data;
2170 if (size > PAGE_SIZE)
2171 size = PAGE_SIZE;
2172
2173 i = size - exact_copy_from_user((void *)page, data, size);
2174 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002175 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 return -EFAULT;
2177 }
2178 if (i != PAGE_SIZE)
2179 memset((char *)page + i, 0, PAGE_SIZE - i);
2180 *where = page;
2181 return 0;
2182}
2183
Vegard Nossumeca6f532009-09-18 13:05:45 -07002184int copy_mount_string(const void __user *data, char **where)
2185{
2186 char *tmp;
2187
2188 if (!data) {
2189 *where = NULL;
2190 return 0;
2191 }
2192
2193 tmp = strndup_user(data, PAGE_SIZE);
2194 if (IS_ERR(tmp))
2195 return PTR_ERR(tmp);
2196
2197 *where = tmp;
2198 return 0;
2199}
2200
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201/*
2202 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2203 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2204 *
2205 * data is a (void *) that can point to any structure up to
2206 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2207 * information (or be NULL).
2208 *
2209 * Pre-0.97 versions of mount() didn't have a flags word.
2210 * When the flags word was introduced its top half was required
2211 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2212 * Therefore, if this magic number is present, it carries no information
2213 * and must be discarded.
2214 */
Al Viro808d4e32012-10-11 11:42:01 -04002215long do_mount(const char *dev_name, const char *dir_name,
2216 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217{
Al Viro2d92ab32008-08-02 00:51:11 -04002218 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 int retval = 0;
2220 int mnt_flags = 0;
2221
2222 /* Discard magic */
2223 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2224 flags &= ~MS_MGC_MSK;
2225
2226 /* Basic sanity checks */
2227
2228 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2229 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230
2231 if (data_page)
2232 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2233
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002234 /* ... and get the mountpoint */
2235 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2236 if (retval)
2237 return retval;
2238
2239 retval = security_sb_mount(dev_name, &path,
2240 type_page, flags, data_page);
2241 if (retval)
2242 goto dput_out;
2243
Al Viro57eccb82013-02-22 22:49:10 -05002244 if (!may_mount())
2245 return -EPERM;
2246
Andi Kleen613cbe32009-04-19 18:40:43 +02002247 /* Default to relatime unless overriden */
2248 if (!(flags & MS_NOATIME))
2249 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002250
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 /* Separate the per-mountpoint flags */
2252 if (flags & MS_NOSUID)
2253 mnt_flags |= MNT_NOSUID;
2254 if (flags & MS_NODEV)
2255 mnt_flags |= MNT_NODEV;
2256 if (flags & MS_NOEXEC)
2257 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002258 if (flags & MS_NOATIME)
2259 mnt_flags |= MNT_NOATIME;
2260 if (flags & MS_NODIRATIME)
2261 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002262 if (flags & MS_STRICTATIME)
2263 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002264 if (flags & MS_RDONLY)
2265 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002266
Al Viro7a4dec52010-08-09 12:05:43 -04002267 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002268 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2269 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002272 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 data_page);
2274 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002275 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002276 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002277 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002279 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 else
Al Viro2d92ab32008-08-02 00:51:11 -04002281 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 dev_name, data_page);
2283dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002284 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 return retval;
2286}
2287
Eric W. Biederman771b1372012-07-26 21:08:32 -07002288static void free_mnt_ns(struct mnt_namespace *ns)
2289{
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002290 proc_free_inum(ns->proc_inum);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002291 put_user_ns(ns->user_ns);
2292 kfree(ns);
2293}
2294
Eric W. Biederman8823c072010-03-07 18:49:36 -08002295/*
2296 * Assign a sequence number so we can detect when we attempt to bind
2297 * mount a reference to an older mount namespace into the current
2298 * mount namespace, preventing reference counting loops. A 64bit
2299 * number incrementing at 10Ghz will take 12,427 years to wrap which
2300 * is effectively never, so we can ignore the possibility.
2301 */
2302static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2303
Eric W. Biederman771b1372012-07-26 21:08:32 -07002304static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002305{
2306 struct mnt_namespace *new_ns;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002307 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002308
2309 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2310 if (!new_ns)
2311 return ERR_PTR(-ENOMEM);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002312 ret = proc_alloc_inum(&new_ns->proc_inum);
2313 if (ret) {
2314 kfree(new_ns);
2315 return ERR_PTR(ret);
2316 }
Eric W. Biederman8823c072010-03-07 18:49:36 -08002317 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002318 atomic_set(&new_ns->count, 1);
2319 new_ns->root = NULL;
2320 INIT_LIST_HEAD(&new_ns->list);
2321 init_waitqueue_head(&new_ns->poll);
2322 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002323 new_ns->user_ns = get_user_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002324 return new_ns;
2325}
2326
JANAK DESAI741a2952006-02-07 12:59:00 -08002327/*
2328 * Allocate a new namespace structure and populate it with contents
2329 * copied from the namespace of the passed in task structure.
2330 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002331static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Eric W. Biederman771b1372012-07-26 21:08:32 -07002332 struct user_namespace *user_ns, struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002334 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002335 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002336 struct mount *p, *q;
Al Virobe08d6d2011-12-06 13:32:36 -05002337 struct mount *old = mnt_ns->root;
Al Virocb338d02011-11-24 20:55:08 -05002338 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002339 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340
Eric W. Biederman771b1372012-07-26 21:08:32 -07002341 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002342 if (IS_ERR(new_ns))
2343 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344
Ram Pai390c6842005-11-07 17:17:51 -05002345 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 /* First pass: copy the tree topology */
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002347 copy_flags = CL_COPY_ALL | CL_EXPIRE;
2348 if (user_ns != mnt_ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002349 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002350 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002351 if (IS_ERR(new)) {
Ram Pai390c6842005-11-07 17:17:51 -05002352 up_write(&namespace_sem);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002353 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002354 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 }
Al Virobe08d6d2011-12-06 13:32:36 -05002356 new_ns->root = new;
Andi Kleen962830d2012-05-08 13:32:02 +09302357 br_write_lock(&vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05002358 list_add_tail(&new_ns->list, &new->mnt_list);
Andi Kleen962830d2012-05-08 13:32:02 +09302359 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
2361 /*
2362 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2363 * as belonging to new namespace. We have already acquired a private
2364 * fs_struct, so tsk->fs->lock is not needed.
2365 */
Al Viro909b0a82011-11-25 03:06:56 -05002366 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002367 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002369 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 if (fs) {
Al Viro315fc832011-11-24 18:57:30 -05002371 if (&p->mnt == fs->root.mnt) {
2372 fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002373 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 }
Al Viro315fc832011-11-24 18:57:30 -05002375 if (&p->mnt == fs->pwd.mnt) {
2376 fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002377 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 }
Al Viro909b0a82011-11-25 03:06:56 -05002380 p = next_mnt(p, old);
2381 q = next_mnt(q, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 }
Ram Pai390c6842005-11-07 17:17:51 -05002383 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002386 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002388 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389
JANAK DESAI741a2952006-02-07 12:59:00 -08002390 return new_ns;
2391}
2392
Eric W. Biederman213dd262007-07-15 23:41:15 -07002393struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Eric W. Biederman771b1372012-07-26 21:08:32 -07002394 struct user_namespace *user_ns, struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002395{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002396 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002397
Badari Pulavartye3222c42007-05-08 00:25:21 -07002398 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002399 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002400
2401 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002402 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002403
Eric W. Biederman771b1372012-07-26 21:08:32 -07002404 new_ns = dup_mnt_ns(ns, user_ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002405
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002406 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002407 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408}
2409
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002410/**
2411 * create_mnt_ns - creates a private namespace and adds a root filesystem
2412 * @mnt: pointer to the new root filesystem mountpoint
2413 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002414static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002415{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002416 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002417 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002418 struct mount *mnt = real_mount(m);
2419 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002420 new_ns->root = mnt;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002421 list_add(&new_ns->list, &mnt->mnt_list);
Al Viroc1334492011-11-16 16:12:14 -05002422 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002423 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002424 }
2425 return new_ns;
2426}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002427
Al Viroea441d12011-11-16 21:43:59 -05002428struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2429{
2430 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002431 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002432 struct path path;
2433 int err;
2434
2435 ns = create_mnt_ns(mnt);
2436 if (IS_ERR(ns))
2437 return ERR_CAST(ns);
2438
2439 err = vfs_path_lookup(mnt->mnt_root, mnt,
2440 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2441
2442 put_mnt_ns(ns);
2443
2444 if (err)
2445 return ERR_PTR(err);
2446
2447 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002448 s = path.mnt->mnt_sb;
2449 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002450 mntput(path.mnt);
2451 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002452 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002453 /* ... and return the root of (sub)tree on it */
2454 return path.dentry;
2455}
2456EXPORT_SYMBOL(mount_subtree);
2457
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002458SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2459 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002461 int ret;
2462 char *kernel_type;
Jeff Layton91a27b22012-10-10 15:25:28 -04002463 struct filename *kernel_dir;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002464 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
Vegard Nossumeca6f532009-09-18 13:05:45 -07002467 ret = copy_mount_string(type, &kernel_type);
2468 if (ret < 0)
2469 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470
Vegard Nossumeca6f532009-09-18 13:05:45 -07002471 kernel_dir = getname(dir_name);
2472 if (IS_ERR(kernel_dir)) {
2473 ret = PTR_ERR(kernel_dir);
2474 goto out_dir;
2475 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
Vegard Nossumeca6f532009-09-18 13:05:45 -07002477 ret = copy_mount_string(dev_name, &kernel_dev);
2478 if (ret < 0)
2479 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480
Vegard Nossumeca6f532009-09-18 13:05:45 -07002481 ret = copy_mount_options(data, &data_page);
2482 if (ret < 0)
2483 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
Jeff Layton91a27b22012-10-10 15:25:28 -04002485 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
Vegard Nossumeca6f532009-09-18 13:05:45 -07002486 (void *) data_page);
2487
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002489out_data:
2490 kfree(kernel_dev);
2491out_dev:
2492 putname(kernel_dir);
2493out_dir:
2494 kfree(kernel_type);
2495out_type:
2496 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497}
2498
2499/*
Al Viroafac7cb2011-11-23 19:34:49 -05002500 * Return true if path is reachable from root
2501 *
2502 * namespace_sem or vfsmount_lock is held
2503 */
Al Viro643822b2011-11-24 22:00:28 -05002504bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002505 const struct path *root)
2506{
Al Viro643822b2011-11-24 22:00:28 -05002507 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002508 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002509 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002510 }
Al Viro643822b2011-11-24 22:00:28 -05002511 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002512}
2513
2514int path_is_under(struct path *path1, struct path *path2)
2515{
2516 int res;
Andi Kleen962830d2012-05-08 13:32:02 +09302517 br_read_lock(&vfsmount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002518 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Andi Kleen962830d2012-05-08 13:32:02 +09302519 br_read_unlock(&vfsmount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002520 return res;
2521}
2522EXPORT_SYMBOL(path_is_under);
2523
2524/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 * pivot_root Semantics:
2526 * Moves the root file system of the current process to the directory put_old,
2527 * makes new_root as the new root file system of the current process, and sets
2528 * root/cwd of all processes which had them on the current root to new_root.
2529 *
2530 * Restrictions:
2531 * The new_root and put_old must be directories, and must not be on the
2532 * same file system as the current process root. The put_old must be
2533 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2534 * pointed to by put_old must yield the same directory as new_root. No other
2535 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2536 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002537 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2538 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2539 * in this situation.
2540 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 * Notes:
2542 * - we don't move root/cwd if they are not at the root (reason: if something
2543 * cared enough to change them, it's probably wrong to force them elsewhere)
2544 * - it's okay to pick a root that isn't the root of a file system, e.g.
2545 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2546 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2547 * first.
2548 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002549SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2550 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551{
Al Viro2d8f3032008-07-22 09:59:21 -04002552 struct path new, old, parent_path, root_parent, root;
Al Viro419148d2011-11-24 19:41:16 -05002553 struct mount *new_mnt, *root_mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 int error;
2555
Al Viro9b40bc92013-02-22 22:45:42 -05002556 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 return -EPERM;
2558
Al Viro2d8f3032008-07-22 09:59:21 -04002559 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 if (error)
2561 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
Al Viro2d8f3032008-07-22 09:59:21 -04002563 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 if (error)
2565 goto out1;
2566
Al Viro2d8f3032008-07-22 09:59:21 -04002567 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002568 if (error)
2569 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002571 get_fs_root(current->fs, &root);
Al Virob12cea92011-03-18 08:55:38 -04002572 error = lock_mount(&old);
2573 if (error)
2574 goto out3;
2575
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002577 new_mnt = real_mount(new.mnt);
2578 root_mnt = real_mount(root.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002579 if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2580 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2581 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002582 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002583 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002584 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002586 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002587 goto out4;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002588 if (d_unlinked(old.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002589 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002591 if (new.mnt == root.mnt ||
2592 old.mnt == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002593 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002595 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002596 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002597 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002598 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002599 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002600 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002601 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002602 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002603 /* make sure we can reach put_old from new_root */
Al Viro643822b2011-11-24 22:00:28 -05002604 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002605 goto out4;
Andi Kleen962830d2012-05-08 13:32:02 +09302606 br_write_lock(&vfsmount_lock);
Al Viro419148d2011-11-24 19:41:16 -05002607 detach_mnt(new_mnt, &parent_path);
2608 detach_mnt(root_mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002609 /* mount old root on put_old */
Al Viro419148d2011-11-24 19:41:16 -05002610 attach_mnt(root_mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002611 /* mount new_root on / */
Al Viro419148d2011-11-24 19:41:16 -05002612 attach_mnt(new_mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002613 touch_mnt_namespace(current->nsproxy->mnt_ns);
Andi Kleen962830d2012-05-08 13:32:02 +09302614 br_write_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002615 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002617out4:
2618 unlock_mount(&old);
2619 if (!error) {
2620 path_put(&root_parent);
2621 path_put(&parent_path);
2622 }
2623out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002624 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002625out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002626 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002628 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631}
2632
2633static void __init init_mount_tree(void)
2634{
2635 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002636 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002637 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002638 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002640 type = get_fs_type("rootfs");
2641 if (!type)
2642 panic("Can't find rootfs type");
2643 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2644 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 if (IS_ERR(mnt))
2646 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002647
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002648 ns = create_mnt_ns(mnt);
2649 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002652 init_task.nsproxy->mnt_ns = ns;
2653 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654
Al Virobe08d6d2011-12-06 13:32:36 -05002655 root.mnt = mnt;
2656 root.dentry = mnt->mnt_root;
Jan Blunckac748a02008-02-14 19:34:39 -08002657
2658 set_fs_pwd(current->fs, &root);
2659 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660}
2661
Denis Cheng74bf17c2007-10-16 23:26:30 -07002662void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002664 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002665 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666
Ram Pai390c6842005-11-07 17:17:51 -05002667 init_rwsem(&namespace_sem);
2668
Al Viro7d6fec42011-11-23 12:14:10 -05002669 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002670 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671
Ram Paib58fed82005-11-07 17:16:09 -05002672 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673
2674 if (!mount_hashtable)
2675 panic("Failed to allocate mount hash table\n");
2676
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002677 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678
Eric Dumazet13f14b42008-02-06 01:37:57 -08002679 for (u = 0; u < HASH_SIZE; u++)
2680 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
Andi Kleen962830d2012-05-08 13:32:02 +09302682 br_lock_init(&vfsmount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +10002683
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002684 err = sysfs_init();
2685 if (err)
2686 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002687 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002688 fs_kobj = kobject_create_and_add("fs", NULL);
2689 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002690 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 init_rootfs();
2692 init_mount_tree();
2693}
2694
Trond Myklebust616511d2009-06-22 15:09:13 -04002695void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002697 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002698
Al Virod498b252010-02-05 02:21:06 -05002699 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002700 return;
Ram Pai390c6842005-11-07 17:17:51 -05002701 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09302702 br_write_lock(&vfsmount_lock);
Al Virobe08d6d2011-12-06 13:32:36 -05002703 umount_tree(ns->root, 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09302704 br_write_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002705 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002706 release_mounts(&umount_list);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002707 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708}
Al Viro9d412a42011-03-17 22:08:28 -04002709
2710struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2711{
Tim Chen423e0ab2011-07-19 09:32:38 -07002712 struct vfsmount *mnt;
2713 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2714 if (!IS_ERR(mnt)) {
2715 /*
2716 * it is a longterm mount, don't release mnt until
2717 * we unmount before file sys is unregistered
2718 */
Al Virof7a99c52012-06-09 00:59:08 -04002719 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07002720 }
2721 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002722}
2723EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002724
2725void kern_unmount(struct vfsmount *mnt)
2726{
2727 /* release long term mount so mount point can be released */
2728 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04002729 br_write_lock(&vfsmount_lock);
2730 real_mount(mnt)->mnt_ns = NULL;
2731 br_write_unlock(&vfsmount_lock);
Tim Chen423e0ab2011-07-19 09:32:38 -07002732 mntput(mnt);
2733 }
2734}
2735EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002736
2737bool our_mnt(struct vfsmount *mnt)
2738{
Al Viro143c8c92011-11-25 00:46:35 -05002739 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05002740}
Eric W. Biederman8823c072010-03-07 18:49:36 -08002741
Eric W. Biederman31515272013-03-15 01:45:51 -07002742bool current_chrooted(void)
2743{
2744 /* Does the current process have a non-standard root */
2745 struct path ns_root;
2746 struct path fs_root;
2747 bool chrooted;
2748
2749 /* Find the namespace root */
2750 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
2751 ns_root.dentry = ns_root.mnt->mnt_root;
2752 path_get(&ns_root);
2753 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2754 ;
2755
2756 get_fs_root(current->fs, &fs_root);
2757
2758 chrooted = !path_equal(&fs_root, &ns_root);
2759
2760 path_put(&fs_root);
2761 path_put(&ns_root);
2762
2763 return chrooted;
2764}
2765
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002766void update_mnt_policy(struct user_namespace *userns)
2767{
2768 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2769 struct mount *mnt;
2770
2771 down_read(&namespace_sem);
2772 list_for_each_entry(mnt, &ns->list, mnt_list) {
2773 switch (mnt->mnt.mnt_sb->s_magic) {
2774 case SYSFS_MAGIC:
2775 userns->may_mount_sysfs = true;
2776 break;
2777 case PROC_SUPER_MAGIC:
2778 userns->may_mount_proc = true;
2779 break;
2780 }
2781 if (userns->may_mount_sysfs && userns->may_mount_proc)
2782 break;
2783 }
2784 up_read(&namespace_sem);
2785}
2786
Eric W. Biederman8823c072010-03-07 18:49:36 -08002787static void *mntns_get(struct task_struct *task)
2788{
2789 struct mnt_namespace *ns = NULL;
2790 struct nsproxy *nsproxy;
2791
2792 rcu_read_lock();
2793 nsproxy = task_nsproxy(task);
2794 if (nsproxy) {
2795 ns = nsproxy->mnt_ns;
2796 get_mnt_ns(ns);
2797 }
2798 rcu_read_unlock();
2799
2800 return ns;
2801}
2802
2803static void mntns_put(void *ns)
2804{
2805 put_mnt_ns(ns);
2806}
2807
2808static int mntns_install(struct nsproxy *nsproxy, void *ns)
2809{
2810 struct fs_struct *fs = current->fs;
2811 struct mnt_namespace *mnt_ns = ns;
2812 struct path root;
2813
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002814 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biederman5e4a0842012-12-14 07:55:36 -08002815 !nsown_capable(CAP_SYS_CHROOT) ||
2816 !nsown_capable(CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08002817 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002818
2819 if (fs->users != 1)
2820 return -EINVAL;
2821
2822 get_mnt_ns(mnt_ns);
2823 put_mnt_ns(nsproxy->mnt_ns);
2824 nsproxy->mnt_ns = mnt_ns;
2825
2826 /* Find the root */
2827 root.mnt = &mnt_ns->root->mnt;
2828 root.dentry = mnt_ns->root->mnt.mnt_root;
2829 path_get(&root);
2830 while(d_mountpoint(root.dentry) && follow_down_one(&root))
2831 ;
2832
2833 /* Update the pwd and root */
2834 set_fs_pwd(fs, &root);
2835 set_fs_root(fs, &root);
2836
2837 path_put(&root);
2838 return 0;
2839}
2840
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002841static unsigned int mntns_inum(void *ns)
2842{
2843 struct mnt_namespace *mnt_ns = ns;
2844 return mnt_ns->proc_inum;
2845}
2846
Eric W. Biederman8823c072010-03-07 18:49:36 -08002847const struct proc_ns_operations mntns_operations = {
2848 .name = "mnt",
2849 .type = CLONE_NEWNS,
2850 .get = mntns_get,
2851 .put = mntns_put,
2852 .install = mntns_install,
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002853 .inum = mntns_inum,
Eric W. Biederman8823c072010-03-07 18:49:36 -08002854};