blob: 5147334eb3e47da35dc7bbcb011213071b3bd6e9 [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>
Rob Landley57f150a2013-09-11 14:26:10 -070019#include <linux/init.h> /* init_rootfs */
Al Virod10577a2011-12-07 13:06:11 -050020#include <linux/fs_struct.h> /* get_fs_root et.al. */
21#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22#include <linux/uaccess.h>
David Howells0bb80f22013-04-12 01:50:06 +010023#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070024#include <linux/magic.h>
Al Viro0818bf22014-02-28 13:46:44 -050025#include <linux/bootmem.h>
Al Viro9ea459e2014-08-08 13:08:20 -040026#include <linux/task_work.h>
Ram Pai07b20882005-11-07 17:19:07 -050027#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070028#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Eric W. Biedermand2921682016-09-28 00:27:17 -050030/* Maximum number of mounts in a mount namespace */
31unsigned int sysctl_mount_max __read_mostly = 100000;
32
Al Viro0818bf22014-02-28 13:46:44 -050033static unsigned int m_hash_mask __read_mostly;
34static unsigned int m_hash_shift __read_mostly;
35static unsigned int mp_hash_mask __read_mostly;
36static unsigned int mp_hash_shift __read_mostly;
37
38static __initdata unsigned long mhash_entries;
39static int __init set_mhash_entries(char *str)
40{
41 if (!str)
42 return 0;
43 mhash_entries = simple_strtoul(str, &str, 0);
44 return 1;
45}
46__setup("mhash_entries=", set_mhash_entries);
47
48static __initdata unsigned long mphash_entries;
49static int __init set_mphash_entries(char *str)
50{
51 if (!str)
52 return 0;
53 mphash_entries = simple_strtoul(str, &str, 0);
54 return 1;
55}
56__setup("mphash_entries=", set_mphash_entries);
Eric Dumazet13f14b42008-02-06 01:37:57 -080057
Al Viroc7999c32014-02-27 14:40:10 -050058static u64 event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010059static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010060static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100061static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040062static int mnt_id_start = 0;
63static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Al Viro38129a12014-03-20 21:10:51 -040065static struct hlist_head *mount_hashtable __read_mostly;
Al Viro0818bf22014-02-28 13:46:44 -050066static struct hlist_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080067static struct kmem_cache *mnt_cache __read_mostly;
Al Viro59aa0da2013-09-16 21:34:53 -040068static DECLARE_RWSEM(namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080070/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060071struct kobject *fs_kobj;
72EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080073
Nick Piggin99b7db72010-08-18 04:37:39 +100074/*
75 * vfsmount lock may be taken for read to prevent changes to the
76 * vfsmount hash, ie. during mountpoint lookups or walking back
77 * up the tree.
78 *
79 * It should be taken for write in all cases where the vfsmount
80 * tree or hash is modified or when a vfsmount structure is modified.
81 */
Al Viro48a066e2013-09-29 22:06:07 -040082__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +100083
Al Viro38129a12014-03-20 21:10:51 -040084static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085{
Ram Paib58fed82005-11-07 17:16:09 -050086 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
87 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Al Viro0818bf22014-02-28 13:46:44 -050088 tmp = tmp + (tmp >> m_hash_shift);
89 return &mount_hashtable[tmp & m_hash_mask];
90}
91
92static inline struct hlist_head *mp_hash(struct dentry *dentry)
93{
94 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
95 tmp = tmp + (tmp >> mp_hash_shift);
96 return &mountpoint_hashtable[tmp & mp_hash_mask];
Linus Torvalds1da177e2005-04-16 15:20:36 -070097}
98
Nick Piggin99b7db72010-08-18 04:37:39 +100099/*
100 * allocation is serialized by namespace_sem, but we need the spinlock to
101 * serialize with freeing.
102 */
Al Virob105e272011-11-24 20:38:33 -0500103static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100104{
105 int res;
106
107retry:
108 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +1000109 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -0500110 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -0400111 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500112 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +1000113 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100114 if (res == -EAGAIN)
115 goto retry;
116
117 return res;
118}
119
Al Virob105e272011-11-24 20:38:33 -0500120static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100121{
Al Viro15169fe2011-11-25 00:50:41 -0500122 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000123 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -0400124 ida_remove(&mnt_id_ida, id);
125 if (mnt_id_start > id)
126 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000127 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100128}
129
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100130/*
131 * Allocate a new peer group ID
132 *
133 * mnt_group_ida is protected by namespace_sem
134 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500135static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100136{
Al Virof21f6222009-06-24 03:12:00 -0400137 int res;
138
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100139 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
140 return -ENOMEM;
141
Al Virof21f6222009-06-24 03:12:00 -0400142 res = ida_get_new_above(&mnt_group_ida,
143 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500144 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400145 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500146 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400147
148 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100149}
150
151/*
152 * Release a peer group ID
153 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500154void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100155{
Al Viro15169fe2011-11-25 00:50:41 -0500156 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400157 ida_remove(&mnt_group_ida, id);
158 if (mnt_group_start > id)
159 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500160 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100161}
162
Nick Pigginb3e19d92011-01-07 17:50:11 +1100163/*
164 * vfsmount lock must be held for read
165 */
Al Viro83adc752011-11-24 22:37:54 -0500166static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100167{
168#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500169 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100170#else
171 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500172 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100173 preempt_enable();
174#endif
175}
176
Nick Pigginb3e19d92011-01-07 17:50:11 +1100177/*
178 * vfsmount lock must be held for write
179 */
Al Viro83adc752011-11-24 22:37:54 -0500180unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100181{
182#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500183 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100184 int cpu;
185
186 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500187 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100188 }
189
190 return count;
191#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500192 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100193#endif
194}
195
Al Viro87b95ce2015-01-10 19:01:08 -0500196static void drop_mountpoint(struct fs_pin *p)
197{
198 struct mount *m = container_of(p, struct mount, mnt_umount);
199 dput(m->mnt_ex_mountpoint);
200 pin_remove(p);
201 mntput(&m->mnt);
202}
203
Al Virob105e272011-11-24 20:38:33 -0500204static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205{
Al Viroc63181e2011-11-25 02:35:16 -0500206 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
207 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100208 int err;
209
Al Viroc63181e2011-11-25 02:35:16 -0500210 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800211 if (err)
212 goto out_free_cache;
213
214 if (name) {
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800215 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
Al Viroc63181e2011-11-25 02:35:16 -0500216 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800217 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100218 }
219
Nick Pigginb3e19d92011-01-07 17:50:11 +1100220#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500221 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
222 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100223 goto out_free_devname;
224
Al Viroc63181e2011-11-25 02:35:16 -0500225 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100226#else
Al Viroc63181e2011-11-25 02:35:16 -0500227 mnt->mnt_count = 1;
228 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100229#endif
230
Al Viro38129a12014-03-20 21:10:51 -0400231 INIT_HLIST_NODE(&mnt->mnt_hash);
Al Viroc63181e2011-11-25 02:35:16 -0500232 INIT_LIST_HEAD(&mnt->mnt_child);
233 INIT_LIST_HEAD(&mnt->mnt_mounts);
234 INIT_LIST_HEAD(&mnt->mnt_list);
235 INIT_LIST_HEAD(&mnt->mnt_expire);
236 INIT_LIST_HEAD(&mnt->mnt_share);
237 INIT_LIST_HEAD(&mnt->mnt_slave_list);
238 INIT_LIST_HEAD(&mnt->mnt_slave);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700239 INIT_HLIST_NODE(&mnt->mnt_mp_list);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500240#ifdef CONFIG_FSNOTIFY
241 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
242#endif
Al Viro87b95ce2015-01-10 19:01:08 -0500243 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 }
Al Viroc63181e2011-11-25 02:35:16 -0500245 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800246
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000247#ifdef CONFIG_SMP
248out_free_devname:
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800249 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000250#endif
Li Zefan88b38782008-07-21 18:06:36 +0800251out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500252 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800253out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500254 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800255 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
257
Dave Hansen83660252008-02-15 14:37:30 -0800258/*
259 * Most r/o checks on a fs are for operations that take
260 * discrete amounts of time, like a write() or unlink().
261 * We must keep track of when those operations start
262 * (for permission checks) and when they end, so that
263 * we can determine when writes are able to occur to
264 * a filesystem.
265 */
Dave Hansen3d733632008-02-15 14:37:59 -0800266/*
267 * __mnt_is_readonly: check whether a mount is read-only
268 * @mnt: the mount to check for its write status
269 *
270 * This shouldn't be used directly ouside of the VFS.
271 * It does not guarantee that the filesystem will stay
272 * r/w, just that it is right *now*. This can not and
273 * should not be used in place of IS_RDONLY(inode).
274 * mnt_want/drop_write() will _keep_ the filesystem
275 * r/w.
276 */
277int __mnt_is_readonly(struct vfsmount *mnt)
278{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800279 if (mnt->mnt_flags & MNT_READONLY)
280 return 1;
281 if (mnt->mnt_sb->s_flags & MS_RDONLY)
282 return 1;
283 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800284}
285EXPORT_SYMBOL_GPL(__mnt_is_readonly);
286
Al Viro83adc752011-11-24 22:37:54 -0500287static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800288{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000289#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500290 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000291#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500292 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000293#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800294}
Dave Hansen3d733632008-02-15 14:37:59 -0800295
Al Viro83adc752011-11-24 22:37:54 -0500296static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800297{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000298#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500299 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000300#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500301 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000302#endif
303}
304
Al Viro83adc752011-11-24 22:37:54 -0500305static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000306{
307#ifdef CONFIG_SMP
308 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800309 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800310
311 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500312 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800313 }
Dave Hansen3d733632008-02-15 14:37:59 -0800314
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000315 return count;
316#else
317 return mnt->mnt_writers;
318#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800319}
320
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100321static int mnt_is_readonly(struct vfsmount *mnt)
322{
323 if (mnt->mnt_sb->s_readonly_remount)
324 return 1;
325 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
326 smp_rmb();
327 return __mnt_is_readonly(mnt);
328}
329
Dave Hansen3d733632008-02-15 14:37:59 -0800330/*
Jan Karaeb04c282012-06-12 16:20:35 +0200331 * Most r/o & frozen checks on a fs are for operations that take discrete
332 * amounts of time, like a write() or unlink(). We must keep track of when
333 * those operations start (for permission checks) and when they end, so that we
334 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800335 */
Dave Hansen83660252008-02-15 14:37:30 -0800336/**
Jan Karaeb04c282012-06-12 16:20:35 +0200337 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500338 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800339 *
Jan Karaeb04c282012-06-12 16:20:35 +0200340 * This tells the low-level filesystem that a write is about to be performed to
341 * it, and makes sure that writes are allowed (mnt it read-write) before
342 * returning success. This operation does not protect against filesystem being
343 * frozen. When the write operation is finished, __mnt_drop_write() must be
344 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800345 */
Jan Karaeb04c282012-06-12 16:20:35 +0200346int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800347{
Al Viro83adc752011-11-24 22:37:54 -0500348 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800349 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800350
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000351 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100352 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000353 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100354 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000355 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
356 * incremented count after it has set MNT_WRITE_HOLD.
357 */
358 smp_mb();
Miao Xie1e755292012-11-16 17:23:50 +0800359 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000360 cpu_relax();
361 /*
362 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
363 * be set to match its requirements. So we must not load that until
364 * MNT_WRITE_HOLD is cleared.
365 */
366 smp_rmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100367 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100368 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800369 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800370 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000371 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200372
373 return ret;
374}
375
376/**
377 * mnt_want_write - get write access to a mount
378 * @m: the mount on which to take a write
379 *
380 * This tells the low-level filesystem that a write is about to be performed to
381 * it, and makes sure that writes are allowed (mount is read-write, filesystem
382 * is not frozen) before returning success. When the write operation is
383 * finished, mnt_drop_write() must be called. This is effectively a refcount.
384 */
385int mnt_want_write(struct vfsmount *m)
386{
387 int ret;
388
389 sb_start_write(m->mnt_sb);
390 ret = __mnt_want_write(m);
391 if (ret)
392 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800393 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800394}
395EXPORT_SYMBOL_GPL(mnt_want_write);
396
397/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000398 * mnt_clone_write - get write access to a mount
399 * @mnt: the mount on which to take a write
400 *
401 * This is effectively like mnt_want_write, except
402 * it must only be used to take an extra write reference
403 * on a mountpoint that we already know has a write reference
404 * on it. This allows some optimisation.
405 *
406 * After finished, mnt_drop_write must be called as usual to
407 * drop the reference.
408 */
409int mnt_clone_write(struct vfsmount *mnt)
410{
411 /* superblock may be r/o */
412 if (__mnt_is_readonly(mnt))
413 return -EROFS;
414 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500415 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000416 preempt_enable();
417 return 0;
418}
419EXPORT_SYMBOL_GPL(mnt_clone_write);
420
421/**
Jan Karaeb04c282012-06-12 16:20:35 +0200422 * __mnt_want_write_file - get write access to a file's mount
423 * @file: the file who's mount on which to take a write
424 *
425 * This is like __mnt_want_write, but it takes a file and can
426 * do some optimisations if the file is open for write already
427 */
428int __mnt_want_write_file(struct file *file)
429{
Al Viro83f936c2014-03-14 12:02:47 -0400430 if (!(file->f_mode & FMODE_WRITER))
Jan Karaeb04c282012-06-12 16:20:35 +0200431 return __mnt_want_write(file->f_path.mnt);
432 else
433 return mnt_clone_write(file->f_path.mnt);
434}
435
436/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000437 * mnt_want_write_file - get write access to a file's mount
438 * @file: the file who's mount on which to take a write
439 *
440 * This is like mnt_want_write, but it takes a file and can
441 * do some optimisations if the file is open for write already
442 */
443int mnt_want_write_file(struct file *file)
444{
Jan Karaeb04c282012-06-12 16:20:35 +0200445 int ret;
446
447 sb_start_write(file->f_path.mnt->mnt_sb);
448 ret = __mnt_want_write_file(file);
449 if (ret)
450 sb_end_write(file->f_path.mnt->mnt_sb);
451 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000452}
453EXPORT_SYMBOL_GPL(mnt_want_write_file);
454
455/**
Jan Karaeb04c282012-06-12 16:20:35 +0200456 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800457 * @mnt: the mount on which to give up write access
458 *
459 * Tells the low-level filesystem that we are done
460 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200461 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800462 */
Jan Karaeb04c282012-06-12 16:20:35 +0200463void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800464{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000465 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500466 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000467 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800468}
Jan Karaeb04c282012-06-12 16:20:35 +0200469
470/**
471 * mnt_drop_write - give up write access to a mount
472 * @mnt: the mount on which to give up write access
473 *
474 * Tells the low-level filesystem that we are done performing writes to it and
475 * also allows filesystem to be frozen again. Must be matched with
476 * mnt_want_write() call above.
477 */
478void mnt_drop_write(struct vfsmount *mnt)
479{
480 __mnt_drop_write(mnt);
481 sb_end_write(mnt->mnt_sb);
482}
Dave Hansen83660252008-02-15 14:37:30 -0800483EXPORT_SYMBOL_GPL(mnt_drop_write);
484
Jan Karaeb04c282012-06-12 16:20:35 +0200485void __mnt_drop_write_file(struct file *file)
486{
487 __mnt_drop_write(file->f_path.mnt);
488}
489
Al Viro2a79f172011-12-09 08:06:57 -0500490void mnt_drop_write_file(struct file *file)
491{
492 mnt_drop_write(file->f_path.mnt);
493}
494EXPORT_SYMBOL(mnt_drop_write_file);
495
Al Viro83adc752011-11-24 22:37:54 -0500496static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800497{
Dave Hansen3d733632008-02-15 14:37:59 -0800498 int ret = 0;
499
Al Viro719ea2f2013-09-29 11:24:49 -0400500 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500501 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000502 /*
503 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
504 * should be visible before we do.
505 */
506 smp_mb();
507
508 /*
509 * With writers on hold, if this value is zero, then there are
510 * definitely no active writers (although held writers may subsequently
511 * increment the count, they'll have to wait, and decrement it after
512 * seeing MNT_READONLY).
513 *
514 * It is OK to have counter incremented on one CPU and decremented on
515 * another: the sum will add up correctly. The danger would be when we
516 * sum up each counter, if we read a counter before it is incremented,
517 * but then read another CPU's count which it has been subsequently
518 * decremented from -- we would see more decrements than we should.
519 * MNT_WRITE_HOLD protects against this scenario, because
520 * mnt_want_write first increments count, then smp_mb, then spins on
521 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
522 * we're counting up here.
523 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100524 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000525 ret = -EBUSY;
526 else
Al Viro83adc752011-11-24 22:37:54 -0500527 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000528 /*
529 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
530 * that become unheld will see MNT_READONLY.
531 */
532 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500533 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Al Viro719ea2f2013-09-29 11:24:49 -0400534 unlock_mount_hash();
Dave Hansen3d733632008-02-15 14:37:59 -0800535 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800536}
Dave Hansen83660252008-02-15 14:37:30 -0800537
Al Viro83adc752011-11-24 22:37:54 -0500538static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800539{
Al Viro719ea2f2013-09-29 11:24:49 -0400540 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500541 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Al Viro719ea2f2013-09-29 11:24:49 -0400542 unlock_mount_hash();
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800543}
544
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100545int sb_prepare_remount_readonly(struct super_block *sb)
546{
547 struct mount *mnt;
548 int err = 0;
549
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100550 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
551 if (atomic_long_read(&sb->s_remove_count))
552 return -EBUSY;
553
Al Viro719ea2f2013-09-29 11:24:49 -0400554 lock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100555 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
556 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
557 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
558 smp_mb();
559 if (mnt_get_writers(mnt) > 0) {
560 err = -EBUSY;
561 break;
562 }
563 }
564 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100565 if (!err && atomic_long_read(&sb->s_remove_count))
566 err = -EBUSY;
567
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100568 if (!err) {
569 sb->s_readonly_remount = 1;
570 smp_wmb();
571 }
572 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
573 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
574 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
575 }
Al Viro719ea2f2013-09-29 11:24:49 -0400576 unlock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100577
578 return err;
579}
580
Al Virob105e272011-11-24 20:38:33 -0500581static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
Daniel Rosenberg1aaf05b2016-10-26 15:29:51 -0700583 kfree(mnt->mnt.data);
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800584 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000585#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500586 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000587#endif
Al Virob105e272011-11-24 20:38:33 -0500588 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589}
590
David Howells8ffcb322014-01-24 12:17:54 +0000591static void delayed_free_vfsmnt(struct rcu_head *head)
592{
593 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
594}
595
Al Viro48a066e2013-09-29 22:06:07 -0400596/* call under rcu_read_lock */
Al Viro294d71f2015-05-08 11:43:53 -0400597int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
Al Viro48a066e2013-09-29 22:06:07 -0400598{
599 struct mount *mnt;
600 if (read_seqretry(&mount_lock, seq))
Al Viro294d71f2015-05-08 11:43:53 -0400601 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400602 if (bastard == NULL)
Al Viro294d71f2015-05-08 11:43:53 -0400603 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400604 mnt = real_mount(bastard);
605 mnt_add_count(mnt, 1);
606 if (likely(!read_seqretry(&mount_lock, seq)))
Al Viro294d71f2015-05-08 11:43:53 -0400607 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400608 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
609 mnt_add_count(mnt, -1);
Al Viro294d71f2015-05-08 11:43:53 -0400610 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400611 }
Al Viro294d71f2015-05-08 11:43:53 -0400612 return -1;
613}
614
615/* call under rcu_read_lock */
616bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
617{
618 int res = __legitimize_mnt(bastard, seq);
619 if (likely(!res))
620 return true;
621 if (unlikely(res < 0)) {
622 rcu_read_unlock();
623 mntput(bastard);
624 rcu_read_lock();
625 }
Al Viro48a066e2013-09-29 22:06:07 -0400626 return false;
627}
628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629/*
Al Viro474279d2013-10-01 16:11:26 -0400630 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400631 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 */
Al Viro474279d2013-10-01 16:11:26 -0400633struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
Al Viro38129a12014-03-20 21:10:51 -0400635 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400636 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
Al Viro38129a12014-03-20 21:10:51 -0400638 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400639 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
640 return p;
641 return NULL;
642}
643
644/*
David Howellsf015f1262012-06-25 12:55:28 +0100645 * lookup_mnt - Return the first child mount mounted at path
646 *
647 * "First" means first mounted chronologically. If you create the
648 * following mounts:
649 *
650 * mount /dev/sda1 /mnt
651 * mount /dev/sda2 /mnt
652 * mount /dev/sda3 /mnt
653 *
654 * Then lookup_mnt() on the base /mnt dentry in the root mount will
655 * return successively the root dentry and vfsmount of /dev/sda1, then
656 * /dev/sda2, then /dev/sda3, then NULL.
657 *
658 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500659 */
Al Viro1c755af2009-04-18 14:06:57 -0400660struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500661{
Al Viroc7105362011-11-24 18:22:03 -0500662 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400663 struct vfsmount *m;
664 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000665
Al Viro48a066e2013-09-29 22:06:07 -0400666 rcu_read_lock();
667 do {
668 seq = read_seqbegin(&mount_lock);
669 child_mnt = __lookup_mnt(path->mnt, path->dentry);
670 m = child_mnt ? &child_mnt->mnt : NULL;
671 } while (!legitimize_mnt(m, seq));
672 rcu_read_unlock();
673 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500674}
675
Eric W. Biederman7af13642013-10-04 19:15:13 -0700676/*
677 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
678 * current mount namespace.
679 *
680 * The common case is dentries are not mountpoints at all and that
681 * test is handled inline. For the slow case when we are actually
682 * dealing with a mountpoint of some kind, walk through all of the
683 * mounts in the current mount namespace and test to see if the dentry
684 * is a mountpoint.
685 *
686 * The mount_hashtable is not usable in the context because we
687 * need to identify all mounts that may be in the current mount
688 * namespace not just a mount that happens to have some specified
689 * parent mount.
690 */
691bool __is_local_mountpoint(struct dentry *dentry)
692{
693 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
694 struct mount *mnt;
695 bool is_covered = false;
696
697 if (!d_mountpoint(dentry))
698 goto out;
699
700 down_read(&namespace_sem);
701 list_for_each_entry(mnt, &ns->list, mnt_list) {
702 is_covered = (mnt->mnt_mountpoint == dentry);
703 if (is_covered)
704 break;
705 }
706 up_read(&namespace_sem);
707out:
708 return is_covered;
709}
710
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800711static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
Al Viro84d17192013-03-15 10:53:28 -0400712{
Al Viro0818bf22014-02-28 13:46:44 -0500713 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d17192013-03-15 10:53:28 -0400714 struct mountpoint *mp;
715
Al Viro0818bf22014-02-28 13:46:44 -0500716 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d17192013-03-15 10:53:28 -0400717 if (mp->m_dentry == dentry) {
718 /* might be worth a WARN_ON() */
719 if (d_unlinked(dentry))
720 return ERR_PTR(-ENOENT);
721 mp->m_count++;
722 return mp;
723 }
724 }
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800725 return NULL;
726}
727
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +1300728static struct mountpoint *get_mountpoint(struct dentry *dentry)
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800729{
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +1300730 struct mountpoint *mp, *new = NULL;
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800731 int ret;
Al Viro84d17192013-03-15 10:53:28 -0400732
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +1300733 if (d_mountpoint(dentry)) {
734mountpoint:
735 read_seqlock_excl(&mount_lock);
736 mp = lookup_mountpoint(dentry);
737 read_sequnlock_excl(&mount_lock);
738 if (mp)
739 goto done;
Al Viro84d17192013-03-15 10:53:28 -0400740 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200741
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +1300742 if (!new)
743 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
744 if (!new)
745 return ERR_PTR(-ENOMEM);
746
747
748 /* Exactly one processes may set d_mounted */
749 ret = d_set_mounted(dentry);
750
751 /* Someone else set d_mounted? */
752 if (ret == -EBUSY)
753 goto mountpoint;
754
755 /* The dentry is not available as a mountpoint? */
756 mp = ERR_PTR(ret);
757 if (ret)
758 goto done;
759
760 /* Add the new mountpoint to the hash table */
761 read_seqlock_excl(&mount_lock);
762 new->m_dentry = dentry;
763 new->m_count = 1;
764 hlist_add_head(&new->m_hash, mp_hash(dentry));
765 INIT_HLIST_HEAD(&new->m_list);
766 read_sequnlock_excl(&mount_lock);
767
768 mp = new;
769 new = NULL;
770done:
771 kfree(new);
Al Viro84d17192013-03-15 10:53:28 -0400772 return mp;
773}
774
775static void put_mountpoint(struct mountpoint *mp)
776{
777 if (!--mp->m_count) {
778 struct dentry *dentry = mp->m_dentry;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700779 BUG_ON(!hlist_empty(&mp->m_list));
Al Viro84d17192013-03-15 10:53:28 -0400780 spin_lock(&dentry->d_lock);
781 dentry->d_flags &= ~DCACHE_MOUNTED;
782 spin_unlock(&dentry->d_lock);
Al Viro0818bf22014-02-28 13:46:44 -0500783 hlist_del(&mp->m_hash);
Al Viro84d17192013-03-15 10:53:28 -0400784 kfree(mp);
785 }
786}
787
Al Viro143c8c92011-11-25 00:46:35 -0500788static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800790 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
792
Nick Piggin99b7db72010-08-18 04:37:39 +1000793/*
794 * vfsmount lock must be held for write
795 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800796static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500797{
798 if (ns) {
799 ns->event = ++event;
800 wake_up_interruptible(&ns->poll);
801 }
802}
803
Nick Piggin99b7db72010-08-18 04:37:39 +1000804/*
805 * vfsmount lock must be held for write
806 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800807static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500808{
809 if (ns && ns->event != event) {
810 ns->event = event;
811 wake_up_interruptible(&ns->poll);
812 }
813}
814
Nick Piggin99b7db72010-08-18 04:37:39 +1000815/*
816 * vfsmount lock must be held for write
817 */
Eric W. Biederman7bdb11d2014-12-29 13:03:41 -0600818static void unhash_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819{
Al Viro0714a532011-11-24 22:19:58 -0500820 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500821 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500822 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400823 hlist_del_init_rcu(&mnt->mnt_hash);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700824 hlist_del_init(&mnt->mnt_mp_list);
Al Viro84d17192013-03-15 10:53:28 -0400825 put_mountpoint(mnt->mnt_mp);
826 mnt->mnt_mp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827}
828
Nick Piggin99b7db72010-08-18 04:37:39 +1000829/*
830 * vfsmount lock must be held for write
831 */
Eric W. Biederman7bdb11d2014-12-29 13:03:41 -0600832static void detach_mnt(struct mount *mnt, struct path *old_path)
833{
834 old_path->dentry = mnt->mnt_mountpoint;
835 old_path->mnt = &mnt->mnt_parent->mnt;
836 unhash_mnt(mnt);
837}
838
839/*
840 * vfsmount lock must be held for write
841 */
Eric W. Biederman6a46c572015-01-15 22:58:33 -0600842static void umount_mnt(struct mount *mnt)
843{
844 /* old mountpoint will be dropped when we can do that */
845 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
846 unhash_mnt(mnt);
847}
848
849/*
850 * vfsmount lock must be held for write
851 */
Al Viro84d17192013-03-15 10:53:28 -0400852void mnt_set_mountpoint(struct mount *mnt,
853 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500854 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500855{
Al Viro84d17192013-03-15 10:53:28 -0400856 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500857 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro84d17192013-03-15 10:53:28 -0400858 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500859 child_mnt->mnt_parent = mnt;
Al Viro84d17192013-03-15 10:53:28 -0400860 child_mnt->mnt_mp = mp;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700861 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500862}
863
Eric W. Biederman808e83e2017-01-20 18:28:35 +1300864static void __attach_mnt(struct mount *mnt, struct mount *parent)
865{
866 hlist_add_head_rcu(&mnt->mnt_hash,
867 m_hash(&parent->mnt, mnt->mnt_mountpoint));
868 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
869}
870
Nick Piggin99b7db72010-08-18 04:37:39 +1000871/*
872 * vfsmount lock must be held for write
873 */
Al Viro84d17192013-03-15 10:53:28 -0400874static void attach_mnt(struct mount *mnt,
875 struct mount *parent,
876 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877{
Al Viro84d17192013-03-15 10:53:28 -0400878 mnt_set_mountpoint(parent, mp, mnt);
Eric W. Biederman808e83e2017-01-20 18:28:35 +1300879 __attach_mnt(mnt, parent);
Ram Paib90fa9a2005-11-07 17:19:50 -0500880}
881
Eric W. Biederman808e83e2017-01-20 18:28:35 +1300882void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
Al Viro12a5b522014-08-10 03:44:55 -0400883{
Eric W. Biederman808e83e2017-01-20 18:28:35 +1300884 struct mountpoint *old_mp = mnt->mnt_mp;
885 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
886 struct mount *old_parent = mnt->mnt_parent;
887
888 list_del_init(&mnt->mnt_child);
889 hlist_del_init(&mnt->mnt_mp_list);
890 hlist_del_init_rcu(&mnt->mnt_hash);
891
892 attach_mnt(mnt, parent, mp);
893
894 put_mountpoint(old_mp);
895
896 /*
897 * Safely avoid even the suggestion this code might sleep or
898 * lock the mount hash by taking advantage of the knowledge that
899 * mnt_change_mountpoint will not release the final reference
900 * to a mountpoint.
901 *
902 * During mounting, the mount passed in as the parent mount will
903 * continue to use the old mountpoint and during unmounting, the
904 * old mountpoint will continue to exist until namespace_unlock,
905 * which happens well after mnt_change_mountpoint.
906 */
907 spin_lock(&old_mountpoint->d_lock);
908 old_mountpoint->d_lockref.count--;
909 spin_unlock(&old_mountpoint->d_lock);
910
911 mnt_add_count(old_parent, -1);
Al Viro12a5b522014-08-10 03:44:55 -0400912}
913
Ram Paib90fa9a2005-11-07 17:19:50 -0500914/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000915 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500916 */
Eric W. Biederman808e83e2017-01-20 18:28:35 +1300917static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500918{
Al Viro0714a532011-11-24 22:19:58 -0500919 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500920 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500921 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500922 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500923
Al Viro0714a532011-11-24 22:19:58 -0500924 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500925
Al Viro1a4eeaf2011-11-25 02:19:55 -0500926 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400927 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500928 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500929
Ram Paib90fa9a2005-11-07 17:19:50 -0500930 list_splice(&head, n->list.prev);
931
Eric W. Biedermand2921682016-09-28 00:27:17 -0500932 n->mounts += n->pending_mounts;
933 n->pending_mounts = 0;
934
Eric W. Biederman808e83e2017-01-20 18:28:35 +1300935 __attach_mnt(mnt, parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800936 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937}
938
Al Viro909b0a82011-11-25 03:06:56 -0500939static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
Al Viro6b41d532011-11-24 23:24:33 -0500941 struct list_head *next = p->mnt_mounts.next;
942 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500944 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500946 next = p->mnt_child.next;
947 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 break;
Al Viro0714a532011-11-24 22:19:58 -0500949 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 }
951 }
Al Viro6b41d532011-11-24 23:24:33 -0500952 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953}
954
Al Viro315fc832011-11-24 18:57:30 -0500955static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500956{
Al Viro6b41d532011-11-24 23:24:33 -0500957 struct list_head *prev = p->mnt_mounts.prev;
958 while (prev != &p->mnt_mounts) {
959 p = list_entry(prev, struct mount, mnt_child);
960 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500961 }
962 return p;
963}
964
Al Viro9d412a42011-03-17 22:08:28 -0400965struct vfsmount *
966vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
967{
Al Virob105e272011-11-24 20:38:33 -0500968 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400969 struct dentry *root;
970
971 if (!type)
972 return ERR_PTR(-ENODEV);
973
974 mnt = alloc_vfsmnt(name);
975 if (!mnt)
976 return ERR_PTR(-ENOMEM);
977
Daniel Rosenberg1aaf05b2016-10-26 15:29:51 -0700978 mnt->mnt.data = NULL;
979 if (type->alloc_mnt_data) {
980 mnt->mnt.data = type->alloc_mnt_data();
981 if (!mnt->mnt.data) {
982 mnt_free_id(mnt);
983 free_vfsmnt(mnt);
984 return ERR_PTR(-ENOMEM);
985 }
986 }
Al Viro9d412a42011-03-17 22:08:28 -0400987 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500988 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400989
Daniel Rosenberg29f888e2016-10-26 15:58:22 -0700990 root = mount_fs(type, flags, name, &mnt->mnt, data);
Al Viro9d412a42011-03-17 22:08:28 -0400991 if (IS_ERR(root)) {
Daniel Rosenberg1aaf05b2016-10-26 15:29:51 -0700992 kfree(mnt->mnt.data);
David Howells8ffcb322014-01-24 12:17:54 +0000993 mnt_free_id(mnt);
Al Viro9d412a42011-03-17 22:08:28 -0400994 free_vfsmnt(mnt);
995 return ERR_CAST(root);
996 }
997
Al Virob105e272011-11-24 20:38:33 -0500998 mnt->mnt.mnt_root = root;
999 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -05001000 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -05001001 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -04001002 lock_mount_hash();
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001003 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001004 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -05001005 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -04001006}
1007EXPORT_SYMBOL_GPL(vfs_kern_mount);
1008
Eric W. Biedermand3381fa2017-02-01 06:06:16 +13001009struct vfsmount *
1010vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1011 const char *name, void *data)
1012{
1013 /* Until it is worked out how to pass the user namespace
1014 * through from the parent mount to the submount don't support
1015 * unprivileged mounts with submounts.
1016 */
1017 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1018 return ERR_PTR(-EPERM);
1019
1020 return vfs_kern_mount(type, MS_SUBMOUNT, name, data);
1021}
1022EXPORT_SYMBOL_GPL(vfs_submount);
1023
Al Viro87129cc2011-11-24 21:24:27 -05001024static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -05001025 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
Al Viro87129cc2011-11-24 21:24:27 -05001027 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +01001028 struct mount *mnt;
1029 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
David Howellsbe34d1a2012-06-25 12:55:18 +01001031 mnt = alloc_vfsmnt(old->mnt_devname);
1032 if (!mnt)
1033 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001034
Daniel Rosenberg1aaf05b2016-10-26 15:29:51 -07001035 if (sb->s_op->clone_mnt_data) {
1036 mnt->mnt.data = sb->s_op->clone_mnt_data(old->mnt.data);
1037 if (!mnt->mnt.data) {
1038 err = -ENOMEM;
1039 goto out_free;
1040 }
1041 }
1042
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001043 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +01001044 mnt->mnt_group_id = 0; /* not a peer of original */
1045 else
1046 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001047
David Howellsbe34d1a2012-06-25 12:55:18 +01001048 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1049 err = mnt_alloc_group_id(mnt);
1050 if (err)
1051 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001053
Al Virof2ebb3a2014-02-27 09:35:45 -05001054 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
Eric W. Biederman132c94e2013-03-22 04:08:05 -07001055 /* Don't allow unprivileged users to change mount flags */
Eric W. Biederman9566d672014-07-28 17:26:07 -07001056 if (flag & CL_UNPRIVILEGED) {
1057 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1058
1059 if (mnt->mnt.mnt_flags & MNT_READONLY)
1060 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1061
1062 if (mnt->mnt.mnt_flags & MNT_NODEV)
1063 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1064
1065 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1066 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1067
1068 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1069 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1070 }
Eric W. Biederman132c94e2013-03-22 04:08:05 -07001071
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001072 /* Don't allow unprivileged users to reveal what is under a mount */
Eric W. Biederman381cacb2014-10-07 17:11:46 -07001073 if ((flag & CL_UNPRIVILEGED) &&
1074 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001075 mnt->mnt.mnt_flags |= MNT_LOCKED;
1076
David Howellsbe34d1a2012-06-25 12:55:18 +01001077 atomic_inc(&sb->s_active);
1078 mnt->mnt.mnt_sb = sb;
1079 mnt->mnt.mnt_root = dget(root);
1080 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1081 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -04001082 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001083 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001084 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001085
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001086 if ((flag & CL_SLAVE) ||
1087 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +01001088 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1089 mnt->mnt_master = old;
1090 CLEAR_MNT_SHARED(mnt);
1091 } else if (!(flag & CL_PRIVATE)) {
1092 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1093 list_add(&mnt->mnt_share, &old->mnt_share);
1094 if (IS_MNT_SLAVE(old))
1095 list_add(&mnt->mnt_slave, &old->mnt_slave);
1096 mnt->mnt_master = old->mnt_master;
1097 }
1098 if (flag & CL_MAKE_SHARED)
1099 set_mnt_shared(mnt);
1100
1101 /* stick the duplicate mount on the same expiry list
1102 * as the original if that was on one */
1103 if (flag & CL_EXPIRE) {
1104 if (!list_empty(&old->mnt_expire))
1105 list_add(&mnt->mnt_expire, &old->mnt_expire);
1106 }
1107
Al Virocb338d02011-11-24 20:55:08 -05001108 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001109
1110 out_free:
Daniel Rosenberg1aaf05b2016-10-26 15:29:51 -07001111 kfree(mnt->mnt.data);
David Howells8ffcb322014-01-24 12:17:54 +00001112 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001113 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001114 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115}
1116
Al Viro9ea459e2014-08-08 13:08:20 -04001117static void cleanup_mnt(struct mount *mnt)
1118{
1119 /*
1120 * This probably indicates that somebody messed
1121 * up a mnt_want/drop_write() pair. If this
1122 * happens, the filesystem was probably unable
1123 * to make r/w->r/o transitions.
1124 */
1125 /*
1126 * The locking used to deal with mnt_count decrement provides barriers,
1127 * so mnt_get_writers() below is safe.
1128 */
1129 WARN_ON(mnt_get_writers(mnt));
1130 if (unlikely(mnt->mnt_pins.first))
1131 mnt_pin_kill(mnt);
1132 fsnotify_vfsmount_delete(&mnt->mnt);
1133 dput(mnt->mnt.mnt_root);
1134 deactivate_super(mnt->mnt.mnt_sb);
1135 mnt_free_id(mnt);
1136 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1137}
1138
1139static void __cleanup_mnt(struct rcu_head *head)
1140{
1141 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1142}
1143
1144static LLIST_HEAD(delayed_mntput_list);
1145static void delayed_mntput(struct work_struct *unused)
1146{
1147 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1148 struct llist_node *next;
1149
1150 for (; node; node = next) {
1151 next = llist_next(node);
1152 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1153 }
1154}
1155static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1156
Al Viro900148d2011-11-25 00:33:11 -05001157static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001158{
Al Viro48a066e2013-09-29 22:06:07 -04001159 rcu_read_lock();
1160 mnt_add_count(mnt, -1);
1161 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1162 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -05001163 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001164 }
Al Viro719ea2f2013-09-29 11:24:49 -04001165 lock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001166 if (mnt_get_count(mnt)) {
Al Viro48a066e2013-09-29 22:06:07 -04001167 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -04001168 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001169 return;
1170 }
Al Viro48a066e2013-09-29 22:06:07 -04001171 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1172 rcu_read_unlock();
1173 unlock_mount_hash();
1174 return;
1175 }
1176 mnt->mnt.mnt_flags |= MNT_DOOMED;
1177 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301178
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001179 list_del(&mnt->mnt_instance);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001180
1181 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1182 struct mount *p, *tmp;
1183 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1184 umount_mnt(p);
1185 }
1186 }
Al Viro719ea2f2013-09-29 11:24:49 -04001187 unlock_mount_hash();
Al Viro649a7952013-09-28 12:41:25 -04001188
Al Viro9ea459e2014-08-08 13:08:20 -04001189 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1190 struct task_struct *task = current;
1191 if (likely(!(task->flags & PF_KTHREAD))) {
1192 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1193 if (!task_work_add(task, &mnt->mnt_rcu, true))
1194 return;
1195 }
1196 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1197 schedule_delayed_work(&delayed_mntput_work, 1);
1198 return;
1199 }
1200 cleanup_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001201}
Nick Pigginb3e19d92011-01-07 17:50:11 +11001202
1203void mntput(struct vfsmount *mnt)
1204{
1205 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -05001206 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001207 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -05001208 if (unlikely(m->mnt_expiry_mark))
1209 m->mnt_expiry_mark = 0;
1210 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001211 }
1212}
1213EXPORT_SYMBOL(mntput);
1214
1215struct vfsmount *mntget(struct vfsmount *mnt)
1216{
1217 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001218 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001219 return mnt;
1220}
1221EXPORT_SYMBOL(mntget);
1222
Al Viro3064c352014-08-07 09:12:31 -04001223struct vfsmount *mnt_clone_internal(struct path *path)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001224{
Al Viro3064c352014-08-07 09:12:31 -04001225 struct mount *p;
1226 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1227 if (IS_ERR(p))
1228 return ERR_CAST(p);
1229 p->mnt.mnt_flags |= MNT_INTERNAL;
1230 return &p->mnt;
Al Viro7b7b1ac2005-11-07 17:13:39 -05001231}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001233static inline void mangle(struct seq_file *m, const char *s)
1234{
1235 seq_escape(m, s, " \t\n\\");
1236}
1237
1238/*
1239 * Simple .show_options callback for filesystems which don't want to
1240 * implement more complex mount option showing.
1241 *
1242 * See also save_mount_options().
1243 */
Al Viro34c80b12011-12-08 21:32:45 -05001244int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001245{
Al Viro2a32ceb2009-05-08 16:05:57 -04001246 const char *options;
1247
1248 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -05001249 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001250
1251 if (options != NULL && options[0]) {
1252 seq_putc(m, ',');
1253 mangle(m, options);
1254 }
Al Viro2a32ceb2009-05-08 16:05:57 -04001255 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001256
1257 return 0;
1258}
1259EXPORT_SYMBOL(generic_show_options);
1260
1261/*
1262 * If filesystem uses generic_show_options(), this function should be
1263 * called from the fill_super() callback.
1264 *
1265 * The .remount_fs callback usually needs to be handled in a special
1266 * way, to make sure, that previous options are not overwritten if the
1267 * remount fails.
1268 *
1269 * Also note, that if the filesystem's .remount_fs function doesn't
1270 * reset all options to their default value, but changes only newly
1271 * given options, then the displayed options will not reflect reality
1272 * any more.
1273 */
1274void save_mount_options(struct super_block *sb, char *options)
1275{
Al Viro2a32ceb2009-05-08 16:05:57 -04001276 BUG_ON(sb->s_options);
1277 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001278}
1279EXPORT_SYMBOL(save_mount_options);
1280
Al Viro2a32ceb2009-05-08 16:05:57 -04001281void replace_mount_options(struct super_block *sb, char *options)
1282{
1283 char *old = sb->s_options;
1284 rcu_assign_pointer(sb->s_options, options);
1285 if (old) {
1286 synchronize_rcu();
1287 kfree(old);
1288 }
1289}
1290EXPORT_SYMBOL(replace_mount_options);
1291
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001292#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001293/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294static void *m_start(struct seq_file *m, loff_t *pos)
1295{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001296 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Ram Pai390c6842005-11-07 17:17:51 -05001298 down_read(&namespace_sem);
Al Viroc7999c32014-02-27 14:40:10 -05001299 if (p->cached_event == p->ns->event) {
1300 void *v = p->cached_mount;
1301 if (*pos == p->cached_index)
1302 return v;
1303 if (*pos == p->cached_index + 1) {
1304 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1305 return p->cached_mount = v;
1306 }
1307 }
1308
1309 p->cached_event = p->ns->event;
1310 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1311 p->cached_index = *pos;
1312 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313}
1314
1315static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1316{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001317 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001318
Al Viroc7999c32014-02-27 14:40:10 -05001319 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1320 p->cached_index = *pos;
1321 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322}
1323
1324static void m_stop(struct seq_file *m, void *v)
1325{
Ram Pai390c6842005-11-07 17:17:51 -05001326 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327}
1328
Al Viro0226f492011-12-06 12:21:54 -05001329static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001330{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001331 struct proc_mounts *p = m->private;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001332 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001333 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334}
1335
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001336const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 .start = m_start,
1338 .next = m_next,
1339 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001340 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001341};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001342#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344/**
1345 * may_umount_tree - check if a mount tree is busy
1346 * @mnt: root of mount tree
1347 *
1348 * This is called to check if a tree of mounts has any
1349 * open files, pwds, chroots or sub mounts that are
1350 * busy.
1351 */
Al Viro909b0a82011-11-25 03:06:56 -05001352int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353{
Al Viro909b0a82011-11-25 03:06:56 -05001354 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001355 int actual_refs = 0;
1356 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001357 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001358 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
Nick Pigginb3e19d92011-01-07 17:50:11 +11001360 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001361 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001362 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001363 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 }
Al Viro719ea2f2013-09-29 11:24:49 -04001366 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
1368 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001369 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Ian Kente3474a82006-03-27 01:14:51 -08001371 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372}
1373
1374EXPORT_SYMBOL(may_umount_tree);
1375
1376/**
1377 * may_umount - check if a mount point is busy
1378 * @mnt: root of mount
1379 *
1380 * This is called to check if a mount point has any
1381 * open files, pwds, chroots or sub mounts. If the
1382 * mount has sub mounts this will return busy
1383 * regardless of whether the sub mounts are busy.
1384 *
1385 * Doesn't take quota and stuff into account. IOW, in some cases it will
1386 * give false negatives. The main reason why it's here is that we need
1387 * a non-destructive way to look for easily umountable filesystems.
1388 */
1389int may_umount(struct vfsmount *mnt)
1390{
Ian Kente3474a82006-03-27 01:14:51 -08001391 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001392 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001393 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001394 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001395 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001396 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001397 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001398 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399}
1400
1401EXPORT_SYMBOL(may_umount);
1402
Al Viro38129a12014-03-20 21:10:51 -04001403static HLIST_HEAD(unmounted); /* protected by namespace_sem */
Al Viroe3197d82013-03-16 14:35:16 -04001404
Al Viro97216be2013-03-16 15:12:40 -04001405static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406{
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001407 struct hlist_head head;
Al Viro97216be2013-03-16 15:12:40 -04001408
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001409 hlist_move_list(&unmounted, &head);
Al Viro97216be2013-03-16 15:12:40 -04001410
Al Viro97216be2013-03-16 15:12:40 -04001411 up_write(&namespace_sem);
1412
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001413 if (likely(hlist_empty(&head)))
1414 return;
1415
Al Viro48a066e2013-09-29 22:06:07 -04001416 synchronize_rcu();
1417
Al Viro87b95ce2015-01-10 19:01:08 -05001418 group_pin_kill(&head);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001419}
1420
Al Viro97216be2013-03-16 15:12:40 -04001421static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001422{
Al Viro97216be2013-03-16 15:12:40 -04001423 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001424}
1425
Eric W. Biedermane819f152014-12-24 07:20:01 -06001426enum umount_tree_flags {
1427 UMOUNT_SYNC = 1,
1428 UMOUNT_PROPAGATE = 2,
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001429 UMOUNT_CONNECTED = 4,
Eric W. Biedermane819f152014-12-24 07:20:01 -06001430};
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001431
1432static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1433{
1434 /* Leaving mounts connected is only valid for lazy umounts */
1435 if (how & UMOUNT_SYNC)
1436 return true;
1437
1438 /* A mount without a parent has nothing to be connected to */
1439 if (!mnt_has_parent(mnt))
1440 return true;
1441
1442 /* Because the reference counting rules change when mounts are
1443 * unmounted and connected, umounted mounts may not be
1444 * connected to mounted mounts.
1445 */
1446 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1447 return true;
1448
1449 /* Has it been requested that the mount remain connected? */
1450 if (how & UMOUNT_CONNECTED)
1451 return false;
1452
1453 /* Is the mount locked such that it needs to remain connected? */
1454 if (IS_MNT_LOCKED(mnt))
1455 return false;
1456
1457 /* By default disconnect the mount */
1458 return true;
1459}
1460
Nick Piggin99b7db72010-08-18 04:37:39 +10001461/*
Al Viro48a066e2013-09-29 22:06:07 -04001462 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001463 * namespace_sem must be held for write
1464 */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001465static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001466{
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001467 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001468 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001469
Eric W. Biederman5d884572015-01-03 05:39:35 -06001470 if (how & UMOUNT_PROPAGATE)
1471 propagate_mount_unlock(mnt);
1472
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001473 /* Gather the mounts to umount */
Eric W. Biederman590ce4b2014-12-22 18:30:08 -06001474 for (p = mnt; p; p = next_mnt(p, mnt)) {
1475 p->mnt.mnt_flags |= MNT_UMOUNT;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001476 list_move(&p->mnt_list, &tmp_list);
Eric W. Biederman590ce4b2014-12-22 18:30:08 -06001477 }
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001478
Eric W. Biederman411a938b2014-12-22 19:12:07 -06001479 /* Hide the mounts from mnt_mounts */
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001480 list_for_each_entry(p, &tmp_list, mnt_list) {
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001481 list_del_init(&p->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -04001482 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001483
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001484 /* Add propogated mounts to the tmp_list */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001485 if (how & UMOUNT_PROPAGATE)
Al Viro7b8a53f2011-01-15 20:08:44 -05001486 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001487
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001488 while (!list_empty(&tmp_list)) {
Eric W. Biedermand2921682016-09-28 00:27:17 -05001489 struct mnt_namespace *ns;
Eric W. Biedermance07d892014-12-23 21:37:03 -06001490 bool disconnect;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001491 p = list_first_entry(&tmp_list, struct mount, mnt_list);
Al Viro6776db3d2011-11-25 00:22:05 -05001492 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001493 list_del_init(&p->mnt_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05001494 ns = p->mnt_ns;
1495 if (ns) {
1496 ns->mounts--;
1497 __touch_mnt_namespace(ns);
1498 }
Al Viro143c8c92011-11-25 00:46:35 -05001499 p->mnt_ns = NULL;
Eric W. Biedermane819f152014-12-24 07:20:01 -06001500 if (how & UMOUNT_SYNC)
Al Viro48a066e2013-09-29 22:06:07 -04001501 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro87b95ce2015-01-10 19:01:08 -05001502
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001503 disconnect = disconnect_mount(p, how);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001504
1505 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1506 disconnect ? &unmounted : NULL);
Al Viro676da582011-11-24 21:47:05 -05001507 if (mnt_has_parent(p)) {
Al Viro81b6b062014-08-30 18:32:05 -04001508 mnt_add_count(p->mnt_parent, -1);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001509 if (!disconnect) {
1510 /* Don't forget about p */
1511 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1512 } else {
1513 umount_mnt(p);
1514 }
Al Viro7c4b93d2008-03-21 23:59:49 -04001515 }
Al Viro0f0afb12011-11-24 20:43:10 -05001516 change_mnt_propagation(p, MS_PRIVATE);
Al Viro38129a12014-03-20 21:10:51 -04001517 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518}
1519
Al Virob54b9be2013-03-16 14:39:34 -04001520static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001521
Al Viro1ab59732011-11-24 21:35:16 -05001522static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523{
Al Viro1ab59732011-11-24 21:35:16 -05001524 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 int retval;
1526
Al Viro1ab59732011-11-24 21:35:16 -05001527 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 if (retval)
1529 return retval;
1530
1531 /*
1532 * Allow userspace to request a mountpoint be expired rather than
1533 * unmounting unconditionally. Unmount only happens if:
1534 * (1) the mark is already set (the mark is cleared by mntput())
1535 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1536 */
1537 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001538 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 flags & (MNT_FORCE | MNT_DETACH))
1540 return -EINVAL;
1541
Nick Pigginb3e19d92011-01-07 17:50:11 +11001542 /*
1543 * probably don't strictly need the lock here if we examined
1544 * all race cases, but it's a slowpath.
1545 */
Al Viro719ea2f2013-09-29 11:24:49 -04001546 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001547 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001548 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001550 }
Al Viro719ea2f2013-09-29 11:24:49 -04001551 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Al Viro863d6842011-11-25 00:57:42 -05001553 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 return -EAGAIN;
1555 }
1556
1557 /*
1558 * If we may have to abort operations to get out of this
1559 * mount, and they will themselves hold resources we must
1560 * allow the fs to do things. In the Unix tradition of
1561 * 'Gee thats tricky lets do it in userspace' the umount_begin
1562 * might fail to complete on the first run through as other tasks
1563 * must return, and the like. Thats for the mount program to worry
1564 * about for the moment.
1565 */
1566
Al Viro42faad92008-04-24 07:21:56 -04001567 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001568 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
1571 /*
1572 * No sense to grab the lock for this test, but test itself looks
1573 * somewhat bogus. Suggestions for better replacement?
1574 * Ho-hum... In principle, we might treat that as umount + switch
1575 * to rootfs. GC would eventually take care of the old vfsmount.
1576 * Actually it makes sense, especially if rootfs would contain a
1577 * /reboot - static binary that would close all descriptors and
1578 * call reboot(9). Then init(8) could umount root and exec /reboot.
1579 */
Al Viro1ab59732011-11-24 21:35:16 -05001580 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 /*
1582 * Special case for "unmounting" root ...
1583 * we just try to remount it readonly.
1584 */
Andy Lutomirskia1480dc2014-10-08 12:32:47 -07001585 if (!capable(CAP_SYS_ADMIN))
1586 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001588 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 up_write(&sb->s_umount);
1591 return retval;
1592 }
1593
Al Viro97216be2013-03-16 15:12:40 -04001594 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001595 lock_mount_hash();
Al Viro5addc5d2005-11-07 17:15:49 -05001596 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
Al Viro48a066e2013-09-29 22:06:07 -04001598 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001599 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001600 umount_tree(mnt, UMOUNT_PROPAGATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001602 } else {
1603 shrink_submounts(mnt);
1604 retval = -EBUSY;
1605 if (!propagate_mount_busy(mnt, 2)) {
1606 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001607 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Viro48a066e2013-09-29 22:06:07 -04001608 retval = 0;
1609 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 }
Al Viro719ea2f2013-09-29 11:24:49 -04001611 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001612 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 return retval;
1614}
1615
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001616/*
1617 * __detach_mounts - lazily unmount all mounts on the specified dentry
1618 *
1619 * During unlink, rmdir, and d_drop it is possible to loose the path
1620 * to an existing mountpoint, and wind up leaking the mount.
1621 * detach_mounts allows lazily unmounting those mounts instead of
1622 * leaking them.
1623 *
1624 * The caller may hold dentry->d_inode->i_mutex.
1625 */
1626void __detach_mounts(struct dentry *dentry)
1627{
1628 struct mountpoint *mp;
1629 struct mount *mnt;
1630
1631 namespace_lock();
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +13001632 lock_mount_hash();
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001633 mp = lookup_mountpoint(dentry);
Eric W. Biedermanf53e5792015-01-19 11:48:45 -06001634 if (IS_ERR_OR_NULL(mp))
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001635 goto out_unlock;
1636
Andrey Ulanove06b9332016-04-15 14:24:41 -07001637 event++;
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001638 while (!hlist_empty(&mp->m_list)) {
1639 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001640 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
Eric W. Biedermanfe78fcc2015-07-17 14:54:27 -05001641 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1642 umount_mnt(mnt);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001643 }
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001644 else umount_tree(mnt, UMOUNT_CONNECTED);
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001645 }
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001646 put_mountpoint(mp);
1647out_unlock:
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +13001648 unlock_mount_hash();
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001649 namespace_unlock();
1650}
1651
Al Viro9b40bc92013-02-22 22:45:42 -05001652/*
1653 * Is the caller allowed to modify his namespace?
1654 */
1655static inline bool may_mount(void)
1656{
1657 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1658}
1659
Jeff Layton9e8925b2015-11-16 09:49:34 -05001660static inline bool may_mandlock(void)
1661{
1662#ifndef CONFIG_MANDATORY_FILE_LOCKING
1663 return false;
1664#endif
Eric W. Biederman95ace752015-11-11 17:22:33 -06001665 return capable(CAP_SYS_ADMIN);
Jeff Layton9e8925b2015-11-16 09:49:34 -05001666}
1667
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668/*
1669 * Now umount can handle mount points as well as block devices.
1670 * This is important for filesystems which use unnamed block devices.
1671 *
1672 * We now support a flag for forced unmount like the other 'big iron'
1673 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1674 */
1675
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001676SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677{
Al Viro2d8f3032008-07-22 09:59:21 -04001678 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001679 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001681 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001683 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1684 return -EINVAL;
1685
Al Viro9b40bc92013-02-22 22:45:42 -05001686 if (!may_mount())
1687 return -EPERM;
1688
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001689 if (!(flags & UMOUNT_NOFOLLOW))
1690 lookup_flags |= LOOKUP_FOLLOW;
1691
Al Viro197df042013-09-08 14:03:27 -04001692 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 if (retval)
1694 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001695 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001697 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001699 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 goto dput_and_out;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001701 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1702 goto dput_and_out;
Eric W. Biedermanb2f5d4d2014-10-04 14:44:03 -07001703 retval = -EPERM;
1704 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1705 goto dput_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706
Al Viro900148d2011-11-25 00:33:11 -05001707 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001709 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001710 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001711 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712out:
1713 return retval;
1714}
1715
1716#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1717
1718/*
Ram Paib58fed82005-11-07 17:16:09 -05001719 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001721SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722{
Ram Paib58fed82005-11-07 17:16:09 -05001723 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724}
1725
1726#endif
1727
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001728static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001729{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001730 /* Is this a proxy for a mount namespace? */
Al Viroe149ed22014-11-01 10:57:28 -04001731 return dentry->d_op == &ns_dentry_operations &&
1732 dentry->d_fsdata == &mntns_operations;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001733}
1734
Al Viro58be28252014-11-01 00:00:23 -04001735struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1736{
1737 return container_of(ns, struct mnt_namespace, ns);
1738}
1739
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001740static bool mnt_ns_loop(struct dentry *dentry)
1741{
1742 /* Could bind mounting the mount namespace inode cause a
1743 * mount namespace loop?
1744 */
1745 struct mnt_namespace *mnt_ns;
1746 if (!is_mnt_ns_file(dentry))
1747 return false;
1748
Al Virof77c8012014-11-01 03:13:17 -04001749 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
Eric W. Biederman8823c072010-03-07 18:49:36 -08001750 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1751}
1752
Al Viro87129cc2011-11-24 21:24:27 -05001753struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001754 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755{
Al Viro84d17192013-03-15 10:53:28 -04001756 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001758 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1759 return ERR_PTR(-EINVAL);
1760
1761 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001762 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001763
Ram Pai36341f62005-11-07 17:17:22 -05001764 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001765 if (IS_ERR(q))
1766 return q;
1767
Al Viroa73324d2011-11-24 22:25:07 -05001768 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001771 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001772 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001773 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 continue;
1775
Al Viro909b0a82011-11-25 03:06:56 -05001776 for (s = r; s; s = next_mnt(s, r)) {
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001777 if (!(flag & CL_COPY_UNBINDABLE) &&
1778 IS_MNT_UNBINDABLE(s)) {
1779 s = skip_mnt_tree(s);
1780 continue;
1781 }
1782 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1783 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001784 s = skip_mnt_tree(s);
1785 continue;
1786 }
Al Viro0714a532011-11-24 22:19:58 -05001787 while (p != s->mnt_parent) {
1788 p = p->mnt_parent;
1789 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 }
Al Viro87129cc2011-11-24 21:24:27 -05001791 p = s;
Al Viro84d17192013-03-15 10:53:28 -04001792 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001793 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001794 if (IS_ERR(q))
1795 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04001796 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05001797 list_add_tail(&q->mnt_list, &res->mnt_list);
Eric W. Biederman808e83e2017-01-20 18:28:35 +13001798 attach_mnt(q, parent, p->mnt_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04001799 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 }
1801 }
1802 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001803out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04001805 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001806 umount_tree(res, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001807 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001809 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810}
1811
David Howellsbe34d1a2012-06-25 12:55:18 +01001812/* Caller should check returned pointer for errors */
1813
Al Viro589ff872009-04-18 03:28:19 -04001814struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001815{
Al Virocb338d02011-11-24 20:55:08 -05001816 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001817 namespace_lock();
Eric W. Biedermancd4a4012015-01-07 14:28:26 -06001818 if (!check_mnt(real_mount(path->mnt)))
1819 tree = ERR_PTR(-EINVAL);
1820 else
1821 tree = copy_tree(real_mount(path->mnt), path->dentry,
1822 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001823 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001824 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001825 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001826 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001827}
1828
1829void drop_collected_mounts(struct vfsmount *mnt)
1830{
Al Viro97216be2013-03-16 15:12:40 -04001831 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001832 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001833 umount_tree(real_mount(mnt), UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001834 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04001835 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001836}
1837
Miklos Szeredic771d682014-10-24 00:14:36 +02001838/**
1839 * clone_private_mount - create a private clone of a path
1840 *
1841 * This creates a new vfsmount, which will be the clone of @path. The new will
1842 * not be attached anywhere in the namespace and will be private (i.e. changes
1843 * to the originating mount won't be propagated into this).
1844 *
1845 * Release with mntput().
1846 */
1847struct vfsmount *clone_private_mount(struct path *path)
1848{
1849 struct mount *old_mnt = real_mount(path->mnt);
1850 struct mount *new_mnt;
1851
1852 if (IS_MNT_UNBINDABLE(old_mnt))
1853 return ERR_PTR(-EINVAL);
1854
1855 down_read(&namespace_sem);
1856 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1857 up_read(&namespace_sem);
1858 if (IS_ERR(new_mnt))
1859 return ERR_CAST(new_mnt);
1860
1861 return &new_mnt->mnt;
1862}
1863EXPORT_SYMBOL_GPL(clone_private_mount);
1864
Al Viro1f707132010-01-30 22:51:25 -05001865int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1866 struct vfsmount *root)
1867{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001868 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001869 int res = f(root, arg);
1870 if (res)
1871 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001872 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1873 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001874 if (res)
1875 return res;
1876 }
1877 return 0;
1878}
1879
Al Viro4b8b21f2011-11-24 19:54:23 -05001880static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001881{
Al Viro315fc832011-11-24 18:57:30 -05001882 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001883
Al Viro909b0a82011-11-25 03:06:56 -05001884 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001885 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001886 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001887 }
1888}
1889
Al Viro4b8b21f2011-11-24 19:54:23 -05001890static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001891{
Al Viro315fc832011-11-24 18:57:30 -05001892 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001893
Al Viro909b0a82011-11-25 03:06:56 -05001894 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001895 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001896 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001897 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001898 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001899 return err;
1900 }
1901 }
1902 }
1903
1904 return 0;
1905}
1906
Eric W. Biedermand2921682016-09-28 00:27:17 -05001907int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1908{
1909 unsigned int max = READ_ONCE(sysctl_mount_max);
1910 unsigned int mounts = 0, old, pending, sum;
1911 struct mount *p;
1912
1913 for (p = mnt; p; p = next_mnt(p, mnt))
1914 mounts++;
1915
1916 old = ns->mounts;
1917 pending = ns->pending_mounts;
1918 sum = old + pending;
1919 if ((old > sum) ||
1920 (pending > sum) ||
1921 (max < sum) ||
1922 (mounts > (max - sum)))
1923 return -ENOSPC;
1924
1925 ns->pending_mounts = pending + mounts;
1926 return 0;
1927}
1928
Ram Paib90fa9a2005-11-07 17:19:50 -05001929/*
1930 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001931 * @nd : place the mount tree @source_mnt is attached
1932 * @parent_nd : if non-null, detach the source_mnt from its parent and
1933 * store the parent mount and mountpoint dentry.
1934 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001935 *
1936 * NOTE: in the table below explains the semantics when a source mount
1937 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001938 * ---------------------------------------------------------------------------
1939 * | BIND MOUNT OPERATION |
1940 * |**************************************************************************
1941 * | source-->| shared | private | slave | unbindable |
1942 * | dest | | | | |
1943 * | | | | | | |
1944 * | v | | | | |
1945 * |**************************************************************************
1946 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1947 * | | | | | |
1948 * |non-shared| shared (+) | private | slave (*) | invalid |
1949 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001950 * A bind operation clones the source mount and mounts the clone on the
1951 * destination mount.
1952 *
1953 * (++) the cloned mount is propagated to all the mounts in the propagation
1954 * tree of the destination mount and the cloned mount is added to
1955 * the peer group of the source mount.
1956 * (+) the cloned mount is created under the destination mount and is marked
1957 * as shared. The cloned mount is added to the peer group of the source
1958 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001959 * (+++) the mount is propagated to all the mounts in the propagation tree
1960 * of the destination mount and the cloned mount is made slave
1961 * of the same master as that of the source mount. The cloned mount
1962 * is marked as 'shared and slave'.
1963 * (*) the cloned mount is made a slave of the same master as that of the
1964 * source mount.
1965 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001966 * ---------------------------------------------------------------------------
1967 * | MOVE MOUNT OPERATION |
1968 * |**************************************************************************
1969 * | source-->| shared | private | slave | unbindable |
1970 * | dest | | | | |
1971 * | | | | | | |
1972 * | v | | | | |
1973 * |**************************************************************************
1974 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1975 * | | | | | |
1976 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1977 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001978 *
1979 * (+) the mount is moved to the destination. And is then propagated to
1980 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001981 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001982 * (+++) the mount is moved to the destination and is then propagated to
1983 * all the mounts belonging to the destination mount's propagation tree.
1984 * the mount is marked as 'shared and slave'.
1985 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001986 *
1987 * if the source mount is a tree, the operations explained above is
1988 * applied to each mount in the tree.
1989 * Must be called without spinlocks held, since this function can sleep
1990 * in allocations.
1991 */
Al Viro0fb54e52011-11-24 19:59:16 -05001992static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d17192013-03-15 10:53:28 -04001993 struct mount *dest_mnt,
1994 struct mountpoint *dest_mp,
1995 struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001996{
Al Viro38129a12014-03-20 21:10:51 -04001997 HLIST_HEAD(tree_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05001998 struct mnt_namespace *ns = dest_mnt->mnt_ns;
Eric W. Biederman808e83e2017-01-20 18:28:35 +13001999 struct mountpoint *smp;
Al Viro315fc832011-11-24 18:57:30 -05002000 struct mount *child, *p;
Al Viro38129a12014-03-20 21:10:51 -04002001 struct hlist_node *n;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002002 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05002003
Eric W. Biederman808e83e2017-01-20 18:28:35 +13002004 /* Preallocate a mountpoint in case the new mounts need
2005 * to be tucked under other mounts.
2006 */
2007 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2008 if (IS_ERR(smp))
2009 return PTR_ERR(smp);
2010
Eric W. Biedermand2921682016-09-28 00:27:17 -05002011 /* Is there space to add these mounts to the mount namespace? */
2012 if (!parent_path) {
2013 err = count_mounts(ns, source_mnt);
2014 if (err)
2015 goto out;
2016 }
2017
Al Virofc7be132011-11-25 01:05:37 -05002018 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05002019 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002020 if (err)
2021 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04002022 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Al Virof2ebb3a2014-02-27 09:35:45 -05002023 lock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04002024 if (err)
2025 goto out_cleanup_ids;
Al Viro909b0a82011-11-25 03:06:56 -05002026 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05002027 set_mnt_shared(p);
Al Viro0b1b9012014-03-21 10:14:08 -04002028 } else {
2029 lock_mount_hash();
Ram Paib90fa9a2005-11-07 17:19:50 -05002030 }
Al Viro1a390682008-03-21 20:48:19 -04002031 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05002032 detach_mnt(source_mnt, parent_path);
Al Viro84d17192013-03-15 10:53:28 -04002033 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05002034 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05002035 } else {
Al Viro84d17192013-03-15 10:53:28 -04002036 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Eric W. Biederman808e83e2017-01-20 18:28:35 +13002037 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05002038 }
Ram Paib90fa9a2005-11-07 17:19:50 -05002039
Al Viro38129a12014-03-20 21:10:51 -04002040 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04002041 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04002042 hlist_del_init(&child->mnt_hash);
Eric W. Biederman808e83e2017-01-20 18:28:35 +13002043 q = __lookup_mnt(&child->mnt_parent->mnt,
2044 child->mnt_mountpoint);
2045 if (q)
2046 mnt_change_mountpoint(child, smp, q);
2047 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05002048 }
Eric W. Biederman808e83e2017-01-20 18:28:35 +13002049 put_mountpoint(smp);
Al Viro719ea2f2013-09-29 11:24:49 -04002050 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10002051
Ram Paib90fa9a2005-11-07 17:19:50 -05002052 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002053
2054 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05002055 while (!hlist_empty(&tree_list)) {
2056 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
Eric W. Biedermand2921682016-09-28 00:27:17 -05002057 child->mnt_parent->mnt_ns->pending_mounts = 0;
Eric W. Biedermane819f152014-12-24 07:20:01 -06002058 umount_tree(child, UMOUNT_SYNC);
Al Virof2ebb3a2014-02-27 09:35:45 -05002059 }
2060 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04002061 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002062 out:
Eric W. Biedermand2921682016-09-28 00:27:17 -05002063 ns->pending_mounts = 0;
Eric W. Biederman808e83e2017-01-20 18:28:35 +13002064
2065 read_seqlock_excl(&mount_lock);
2066 put_mountpoint(smp);
2067 read_sequnlock_excl(&mount_lock);
2068
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002069 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05002070}
2071
Al Viro84d17192013-03-15 10:53:28 -04002072static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04002073{
2074 struct vfsmount *mnt;
Al Viro84d17192013-03-15 10:53:28 -04002075 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04002076retry:
Al Viro59551022016-01-22 15:40:57 -05002077 inode_lock(dentry->d_inode);
Al Viro84d17192013-03-15 10:53:28 -04002078 if (unlikely(cant_mount(dentry))) {
Al Viro59551022016-01-22 15:40:57 -05002079 inode_unlock(dentry->d_inode);
Al Viro84d17192013-03-15 10:53:28 -04002080 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04002081 }
Al Viro97216be2013-03-16 15:12:40 -04002082 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04002083 mnt = lookup_mnt(path);
Al Viro84d17192013-03-15 10:53:28 -04002084 if (likely(!mnt)) {
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +13002085 struct mountpoint *mp = get_mountpoint(dentry);
Al Viro84d17192013-03-15 10:53:28 -04002086 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04002087 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002088 inode_unlock(dentry->d_inode);
Al Viro84d17192013-03-15 10:53:28 -04002089 return mp;
2090 }
2091 return mp;
2092 }
Al Viro97216be2013-03-16 15:12:40 -04002093 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002094 inode_unlock(path->dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002095 path_put(path);
2096 path->mnt = mnt;
Al Viro84d17192013-03-15 10:53:28 -04002097 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04002098 goto retry;
2099}
2100
Al Viro84d17192013-03-15 10:53:28 -04002101static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04002102{
Al Viro84d17192013-03-15 10:53:28 -04002103 struct dentry *dentry = where->m_dentry;
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +13002104
2105 read_seqlock_excl(&mount_lock);
Al Viro84d17192013-03-15 10:53:28 -04002106 put_mountpoint(where);
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +13002107 read_sequnlock_excl(&mount_lock);
2108
Al Viro328e6d92013-03-16 14:49:45 -04002109 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002110 inode_unlock(dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002111}
2112
Al Viro84d17192013-03-15 10:53:28 -04002113static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114{
Al Viro95bc5f22011-11-25 00:30:56 -05002115 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 return -EINVAL;
2117
David Howellse36cb0b2015-01-29 12:02:35 +00002118 if (d_is_dir(mp->m_dentry) !=
2119 d_is_dir(mnt->mnt.mnt_root))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 return -ENOTDIR;
2121
Al Viro84d17192013-03-15 10:53:28 -04002122 return attach_recursive_mnt(mnt, p, mp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123}
2124
2125/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002126 * Sanity check the flags to change_mnt_propagation.
2127 */
2128
2129static int flags_to_propagation_type(int flags)
2130{
Roman Borisov7c6e9842011-05-25 16:26:48 -07002131 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002132
2133 /* Fail if any non-propagation flags are set */
2134 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2135 return 0;
2136 /* Only one propagation flag should be set */
2137 if (!is_power_of_2(type))
2138 return 0;
2139 return type;
2140}
2141
2142/*
Ram Pai07b20882005-11-07 17:19:07 -05002143 * recursively change the type of the mountpoint.
2144 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002145static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05002146{
Al Viro315fc832011-11-24 18:57:30 -05002147 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05002148 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05002149 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002150 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002151 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05002152
Al Viro2d92ab32008-08-02 00:51:11 -04002153 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05002154 return -EINVAL;
2155
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002156 type = flags_to_propagation_type(flag);
2157 if (!type)
2158 return -EINVAL;
2159
Al Viro97216be2013-03-16 15:12:40 -04002160 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002161 if (type == MS_SHARED) {
2162 err = invent_group_ids(mnt, recurse);
2163 if (err)
2164 goto out_unlock;
2165 }
2166
Al Viro719ea2f2013-09-29 11:24:49 -04002167 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05002168 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05002169 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04002170 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002171
2172 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04002173 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002174 return err;
Ram Pai07b20882005-11-07 17:19:07 -05002175}
2176
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002177static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2178{
2179 struct mount *child;
2180 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2181 if (!is_subdir(child->mnt_mountpoint, dentry))
2182 continue;
2183
2184 if (child->mnt.mnt_flags & MNT_LOCKED)
2185 return true;
2186 }
2187 return false;
2188}
2189
Ram Pai07b20882005-11-07 17:19:07 -05002190/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 * do loopback mount.
2192 */
Al Viro808d4e32012-10-11 11:42:01 -04002193static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08002194 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195{
Al Viro2d92ab32008-08-02 00:51:11 -04002196 struct path old_path;
Al Viro84d17192013-03-15 10:53:28 -04002197 struct mount *mnt = NULL, *old, *parent;
2198 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002199 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 if (!old_name || !*old_name)
2201 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04002202 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 if (err)
2204 return err;
2205
Eric W. Biederman8823c072010-03-07 18:49:36 -08002206 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002207 if (mnt_ns_loop(old_path.dentry))
Eric W. Biederman8823c072010-03-07 18:49:36 -08002208 goto out;
2209
Al Viro84d17192013-03-15 10:53:28 -04002210 mp = lock_mount(path);
2211 err = PTR_ERR(mp);
2212 if (IS_ERR(mp))
Jan Blunck4ac91372008-02-14 19:34:32 -08002213 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05002214
Al Viro87129cc2011-11-24 21:24:27 -05002215 old = real_mount(old_path.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002216 parent = real_mount(path->mnt);
Al Viro87129cc2011-11-24 21:24:27 -05002217
Al Virob12cea92011-03-18 08:55:38 -04002218 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002219 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04002220 goto out2;
2221
Al Viroe149ed22014-11-01 10:57:28 -04002222 if (!check_mnt(parent))
2223 goto out2;
2224
2225 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
Al Virob12cea92011-03-18 08:55:38 -04002226 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002228 if (!recurse && has_locked_children(old, old_path.dentry))
2229 goto out2;
2230
Al Viroccd48bc2005-11-07 17:15:04 -05002231 if (recurse)
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002232 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
Al Viroccd48bc2005-11-07 17:15:04 -05002233 else
Al Viro87129cc2011-11-24 21:24:27 -05002234 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05002235
David Howellsbe34d1a2012-06-25 12:55:18 +01002236 if (IS_ERR(mnt)) {
2237 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04002238 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01002239 }
Al Viroccd48bc2005-11-07 17:15:04 -05002240
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002241 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2242
Al Viro84d17192013-03-15 10:53:28 -04002243 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002244 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002245 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06002246 umount_tree(mnt, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04002247 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05002248 }
Al Virob12cea92011-03-18 08:55:38 -04002249out2:
Al Viro84d17192013-03-15 10:53:28 -04002250 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002251out:
Al Viro2d92ab32008-08-02 00:51:11 -04002252 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 return err;
2254}
2255
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002256static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2257{
2258 int error = 0;
2259 int readonly_request = 0;
2260
2261 if (ms_flags & MS_RDONLY)
2262 readonly_request = 1;
2263 if (readonly_request == __mnt_is_readonly(mnt))
2264 return 0;
2265
2266 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05002267 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002268 else
Al Viro83adc752011-11-24 22:37:54 -05002269 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002270 return error;
2271}
2272
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273/*
2274 * change filesystem flags. dir should be a physical root of filesystem.
2275 * If you've mounted a non-root directory somewhere and want to do remount
2276 * on it - tough luck.
2277 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002278static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 void *data)
2280{
2281 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04002282 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05002283 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
Al Viro143c8c92011-11-25 00:46:35 -05002285 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 return -EINVAL;
2287
Al Viro2d92ab32008-08-02 00:51:11 -04002288 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 return -EINVAL;
2290
Eric W. Biederman07b64552014-07-28 17:10:56 -07002291 /* Don't allow changing of locked mnt flags.
2292 *
2293 * No locks need to be held here while testing the various
2294 * MNT_LOCK flags because those flags can never be cleared
2295 * once they are set.
2296 */
2297 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2298 !(mnt_flags & MNT_READONLY)) {
2299 return -EPERM;
2300 }
Eric W. Biederman9566d672014-07-28 17:26:07 -07002301 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2302 !(mnt_flags & MNT_NODEV)) {
Eric W. Biederman67690f92016-05-18 13:50:06 -05002303 return -EPERM;
Eric W. Biederman9566d672014-07-28 17:26:07 -07002304 }
2305 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2306 !(mnt_flags & MNT_NOSUID)) {
2307 return -EPERM;
2308 }
2309 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2310 !(mnt_flags & MNT_NOEXEC)) {
2311 return -EPERM;
2312 }
2313 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2314 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2315 return -EPERM;
2316 }
2317
Eric Parisff36fe22011-03-03 16:09:14 -05002318 err = security_sb_remount(sb, data);
2319 if (err)
2320 return err;
2321
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002323 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002324 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05002325 else if (!capable(CAP_SYS_ADMIN))
2326 err = -EPERM;
Daniel Rosenberg1aaf05b2016-10-26 15:29:51 -07002327 else {
Daniel Rosenberg29f888e2016-10-26 15:58:22 -07002328 err = do_remount_sb2(path->mnt, sb, flags, data, 0);
Daniel Rosenberg1aaf05b2016-10-26 15:29:51 -07002329 namespace_lock();
2330 lock_mount_hash();
2331 propagate_remount(mnt);
2332 unlock_mount_hash();
2333 namespace_unlock();
2334 }
Al Viro7b43a792010-01-16 13:01:26 -05002335 if (!err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002336 lock_mount_hash();
Eric W. Biedermana6138db2014-07-28 16:26:53 -07002337 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
Al Viro143c8c92011-11-25 00:46:35 -05002338 mnt->mnt.mnt_flags = mnt_flags;
Al Viro143c8c92011-11-25 00:46:35 -05002339 touch_mnt_namespace(mnt->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04002340 unlock_mount_hash();
Dan Williams0e55a7c2008-09-26 19:01:20 -07002341 }
Al Viro6339dab2013-09-16 22:41:01 -04002342 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 return err;
2344}
2345
Al Virocbbe3622011-11-24 20:01:19 -05002346static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05002347{
Al Viro315fc832011-11-24 18:57:30 -05002348 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05002349 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002350 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05002351 return 1;
2352 }
2353 return 0;
2354}
2355
Al Viro808d4e32012-10-11 11:42:01 -04002356static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357{
Al Viro2d92ab32008-08-02 00:51:11 -04002358 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05002359 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05002360 struct mount *old;
Al Viro84d17192013-03-15 10:53:28 -04002361 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002362 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 if (!old_name || !*old_name)
2364 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002365 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 if (err)
2367 return err;
2368
Al Viro84d17192013-03-15 10:53:28 -04002369 mp = lock_mount(path);
2370 err = PTR_ERR(mp);
2371 if (IS_ERR(mp))
David Howellscc53ce52011-01-14 18:45:26 +00002372 goto out;
2373
Al Viro143c8c92011-11-25 00:46:35 -05002374 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002375 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05002376
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002378 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 goto out1;
2380
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002381 if (old->mnt.mnt_flags & MNT_LOCKED)
2382 goto out1;
2383
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002385 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05002386 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387
Al Viro676da582011-11-24 21:47:05 -05002388 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05002389 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390
David Howellse36cb0b2015-01-29 12:02:35 +00002391 if (d_is_dir(path->dentry) !=
2392 d_is_dir(old_path.dentry))
Ram Pai21444402005-11-07 17:20:03 -05002393 goto out1;
2394 /*
2395 * Don't move a mount residing in a shared parent.
2396 */
Al Virofc7be132011-11-25 01:05:37 -05002397 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002398 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05002399 /*
2400 * Don't move a mount tree containing unbindable mounts to a destination
2401 * mount which is shared.
2402 */
Al Virofc7be132011-11-25 01:05:37 -05002403 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05002404 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05002406 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05002407 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05002408 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
Al Viro84d17192013-03-15 10:53:28 -04002410 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08002411 if (err)
Ram Pai21444402005-11-07 17:20:03 -05002412 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413
2414 /* if the mount is moved, it should no longer be expire
2415 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05002416 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417out1:
Al Viro84d17192013-03-15 10:53:28 -04002418 unlock_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04002421 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04002422 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 return err;
2424}
2425
Al Viro9d412a42011-03-17 22:08:28 -04002426static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2427{
2428 int err;
2429 const char *subtype = strchr(fstype, '.');
2430 if (subtype) {
2431 subtype++;
2432 err = -EINVAL;
2433 if (!subtype[0])
2434 goto err;
2435 } else
2436 subtype = "";
2437
2438 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2439 err = -ENOMEM;
2440 if (!mnt->mnt_sb->s_subtype)
2441 goto err;
2442 return mnt;
2443
2444 err:
2445 mntput(mnt);
2446 return ERR_PTR(err);
2447}
2448
Al Viro9d412a42011-03-17 22:08:28 -04002449/*
2450 * add a mount into a namespace's mount tree
2451 */
Al Viro95bc5f22011-11-25 00:30:56 -05002452static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04002453{
Al Viro84d17192013-03-15 10:53:28 -04002454 struct mountpoint *mp;
2455 struct mount *parent;
Al Viro9d412a42011-03-17 22:08:28 -04002456 int err;
2457
Al Virof2ebb3a2014-02-27 09:35:45 -05002458 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04002459
Al Viro84d17192013-03-15 10:53:28 -04002460 mp = lock_mount(path);
2461 if (IS_ERR(mp))
2462 return PTR_ERR(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002463
Al Viro84d17192013-03-15 10:53:28 -04002464 parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04002465 err = -EINVAL;
Al Viro84d17192013-03-15 10:53:28 -04002466 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04002467 /* that's acceptable only for automounts done in private ns */
2468 if (!(mnt_flags & MNT_SHRINKABLE))
2469 goto unlock;
2470 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d17192013-03-15 10:53:28 -04002471 if (!parent->mnt_ns)
Al Viro156cacb2012-09-21 08:19:02 -04002472 goto unlock;
2473 }
Al Viro9d412a42011-03-17 22:08:28 -04002474
2475 /* Refuse the same filesystem on the same mount point */
2476 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05002477 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04002478 path->mnt->mnt_root == path->dentry)
2479 goto unlock;
2480
2481 err = -EINVAL;
David Howellse36cb0b2015-01-29 12:02:35 +00002482 if (d_is_symlink(newmnt->mnt.mnt_root))
Al Viro9d412a42011-03-17 22:08:28 -04002483 goto unlock;
2484
Al Viro95bc5f22011-11-25 00:30:56 -05002485 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro84d17192013-03-15 10:53:28 -04002486 err = graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04002487
2488unlock:
Al Viro84d17192013-03-15 10:53:28 -04002489 unlock_mount(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002490 return err;
2491}
Al Virob1e75df2011-01-17 01:47:59 -05002492
Eric W. Biederman8654df42016-06-09 16:06:06 -05002493static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05002494
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495/*
2496 * create a new mount for userspace and request it to be added into the
2497 * namespace's tree
2498 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002499static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04002500 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002502 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002504 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002506 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 return -EINVAL;
2508
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002509 type = get_fs_type(fstype);
2510 if (!type)
2511 return -ENODEV;
2512
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002513 mnt = vfs_kern_mount(type, flags, name, data);
2514 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2515 !mnt->mnt_sb->s_subtype)
2516 mnt = fs_set_subtype(mnt, fstype);
2517
2518 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 if (IS_ERR(mnt))
2520 return PTR_ERR(mnt);
2521
Eric W. Biederman8654df42016-06-09 16:06:06 -05002522 if (mount_too_revealing(mnt, &mnt_flags)) {
2523 mntput(mnt);
2524 return -EPERM;
2525 }
2526
Al Viro95bc5f22011-11-25 00:30:56 -05002527 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002528 if (err)
2529 mntput(mnt);
2530 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531}
2532
Al Viro19a167a2011-01-17 01:35:23 -05002533int finish_automount(struct vfsmount *m, struct path *path)
2534{
Al Viro6776db3d2011-11-25 00:22:05 -05002535 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002536 int err;
2537 /* The new mount record should have at least 2 refs to prevent it being
2538 * expired before we get a chance to add it
2539 */
Al Viro6776db3d2011-11-25 00:22:05 -05002540 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002541
2542 if (m->mnt_sb == path->mnt->mnt_sb &&
2543 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002544 err = -ELOOP;
2545 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002546 }
2547
Al Viro95bc5f22011-11-25 00:30:56 -05002548 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002549 if (!err)
2550 return 0;
2551fail:
2552 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002553 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002554 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05002555 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04002556 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002557 }
Al Virob1e75df2011-01-17 01:47:59 -05002558 mntput(m);
2559 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002560 return err;
2561}
2562
David Howellsea5b7782011-01-14 19:10:03 +00002563/**
2564 * mnt_set_expiry - Put a mount on an expiration list
2565 * @mnt: The mount to list.
2566 * @expiry_list: The list to add the mount to.
2567 */
2568void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2569{
Al Viro97216be2013-03-16 15:12:40 -04002570 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00002571
Al Viro6776db3d2011-11-25 00:22:05 -05002572 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002573
Al Viro97216be2013-03-16 15:12:40 -04002574 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002575}
2576EXPORT_SYMBOL(mnt_set_expiry);
2577
2578/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 * process a list of expirable mountpoints with the intent of discarding any
2580 * mountpoints that aren't in use and haven't been touched since last we came
2581 * here
2582 */
2583void mark_mounts_for_expiry(struct list_head *mounts)
2584{
Al Viro761d5c32011-11-24 21:07:43 -05002585 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 LIST_HEAD(graveyard);
2587
2588 if (list_empty(mounts))
2589 return;
2590
Al Viro97216be2013-03-16 15:12:40 -04002591 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002592 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593
2594 /* extract from the expiration list every vfsmount that matches the
2595 * following criteria:
2596 * - only referenced by its parent vfsmount
2597 * - still marked for expiry (marked on the last call here; marks are
2598 * cleared by mntput())
2599 */
Al Viro6776db3d2011-11-25 00:22:05 -05002600 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002601 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002602 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002604 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002606 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002607 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002608 touch_mnt_namespace(mnt->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06002609 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04002610 }
Al Viro719ea2f2013-09-29 11:24:49 -04002611 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002612 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613}
2614
2615EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2616
2617/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002618 * Ripoff of 'select_parent()'
2619 *
2620 * search the list of submounts for a given mountpoint, and move any
2621 * shrinkable submounts to the 'graveyard' list.
2622 */
Al Viro692afc32011-11-24 21:15:14 -05002623static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002624{
Al Viro692afc32011-11-24 21:15:14 -05002625 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002626 struct list_head *next;
2627 int found = 0;
2628
2629repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002630 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002631resume:
Al Viro6b41d532011-11-24 23:24:33 -05002632 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002633 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002634 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002635
2636 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002637 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002638 continue;
2639 /*
2640 * Descend a level if the d_mounts list is non-empty.
2641 */
Al Viro6b41d532011-11-24 23:24:33 -05002642 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002643 this_parent = mnt;
2644 goto repeat;
2645 }
2646
Al Viro1ab59732011-11-24 21:35:16 -05002647 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002648 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002649 found++;
2650 }
2651 }
2652 /*
2653 * All done at this level ... ascend and resume the search
2654 */
2655 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002656 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002657 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002658 goto resume;
2659 }
2660 return found;
2661}
2662
2663/*
2664 * process a list of expirable mountpoints with the intent of discarding any
2665 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002666 *
Al Viro48a066e2013-09-29 22:06:07 -04002667 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002668 */
Al Virob54b9be2013-03-16 14:39:34 -04002669static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002670{
2671 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002672 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002673
Trond Myklebust5528f9112006-06-09 09:34:17 -04002674 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002675 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002676 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002677 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002678 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002679 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06002680 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04002681 }
2682 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002683}
2684
Trond Myklebust5528f9112006-06-09 09:34:17 -04002685/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 * Some copy_from_user() implementations do not return the exact number of
2687 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2688 * Note that this function differs from copy_from_user() in that it will oops
2689 * on bad values of `to', rather than returning a short copy.
2690 */
Ram Paib58fed82005-11-07 17:16:09 -05002691static long exact_copy_from_user(void *to, const void __user * from,
2692 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693{
2694 char *t = to;
2695 const char __user *f = from;
2696 char c;
2697
2698 if (!access_ok(VERIFY_READ, from, n))
2699 return n;
2700
2701 while (n) {
2702 if (__get_user(c, f)) {
2703 memset(t, 0, n);
2704 break;
2705 }
2706 *t++ = c;
2707 f++;
2708 n--;
2709 }
2710 return n;
2711}
2712
Al Virob40ef862015-12-14 18:44:44 -05002713void *copy_mount_options(const void __user * data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714{
2715 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 unsigned long size;
Al Virob40ef862015-12-14 18:44:44 -05002717 char *copy;
Ram Paib58fed82005-11-07 17:16:09 -05002718
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 if (!data)
Al Virob40ef862015-12-14 18:44:44 -05002720 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721
Al Virob40ef862015-12-14 18:44:44 -05002722 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2723 if (!copy)
2724 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725
2726 /* We only care that *some* data at the address the user
2727 * gave us is valid. Just in case, we'll zero
2728 * the remainder of the page.
2729 */
2730 /* copy_from_user cannot cross TASK_SIZE ! */
2731 size = TASK_SIZE - (unsigned long)data;
2732 if (size > PAGE_SIZE)
2733 size = PAGE_SIZE;
2734
Al Virob40ef862015-12-14 18:44:44 -05002735 i = size - exact_copy_from_user(copy, data, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 if (!i) {
Al Virob40ef862015-12-14 18:44:44 -05002737 kfree(copy);
2738 return ERR_PTR(-EFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 }
2740 if (i != PAGE_SIZE)
Al Virob40ef862015-12-14 18:44:44 -05002741 memset(copy + i, 0, PAGE_SIZE - i);
2742 return copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743}
2744
Tim Gardnerb8850d12014-08-28 11:26:03 -06002745char *copy_mount_string(const void __user *data)
Vegard Nossumeca6f532009-09-18 13:05:45 -07002746{
Tim Gardnerb8850d12014-08-28 11:26:03 -06002747 return data ? strndup_user(data, PAGE_SIZE) : NULL;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002748}
2749
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750/*
2751 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2752 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2753 *
2754 * data is a (void *) that can point to any structure up to
2755 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2756 * information (or be NULL).
2757 *
2758 * Pre-0.97 versions of mount() didn't have a flags word.
2759 * When the flags word was introduced its top half was required
2760 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2761 * Therefore, if this magic number is present, it carries no information
2762 * and must be discarded.
2763 */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002764long do_mount(const char *dev_name, const char __user *dir_name,
Al Viro808d4e32012-10-11 11:42:01 -04002765 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766{
Al Viro2d92ab32008-08-02 00:51:11 -04002767 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 int retval = 0;
2769 int mnt_flags = 0;
2770
2771 /* Discard magic */
2772 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2773 flags &= ~MS_MGC_MSK;
2774
2775 /* Basic sanity checks */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 if (data_page)
2777 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2778
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002779 /* ... and get the mountpoint */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002780 retval = user_path(dir_name, &path);
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002781 if (retval)
2782 return retval;
2783
2784 retval = security_sb_mount(dev_name, &path,
2785 type_page, flags, data_page);
Al Viro0d5cadb2013-05-04 14:40:51 -04002786 if (!retval && !may_mount())
2787 retval = -EPERM;
Jeff Layton9e8925b2015-11-16 09:49:34 -05002788 if (!retval && (flags & MS_MANDLOCK) && !may_mandlock())
2789 retval = -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002790 if (retval)
2791 goto dput_out;
2792
Andi Kleen613cbe32009-04-19 18:40:43 +02002793 /* Default to relatime unless overriden */
2794 if (!(flags & MS_NOATIME))
2795 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002796
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 /* Separate the per-mountpoint flags */
2798 if (flags & MS_NOSUID)
2799 mnt_flags |= MNT_NOSUID;
2800 if (flags & MS_NODEV)
2801 mnt_flags |= MNT_NODEV;
2802 if (flags & MS_NOEXEC)
2803 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002804 if (flags & MS_NOATIME)
2805 mnt_flags |= MNT_NOATIME;
2806 if (flags & MS_NODIRATIME)
2807 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002808 if (flags & MS_STRICTATIME)
2809 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002810 if (flags & MS_RDONLY)
2811 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002812
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07002813 /* The default atime for remount is preservation */
2814 if ((flags & MS_REMOUNT) &&
2815 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2816 MS_STRICTATIME)) == 0)) {
2817 mnt_flags &= ~MNT_ATIME_MASK;
2818 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2819 }
2820
Al Viro7a4dec52010-08-09 12:05:43 -04002821 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002822 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
Eric W. Biedermand3381fa2017-02-01 06:06:16 +13002823 MS_STRICTATIME | MS_NOREMOTELOCK | MS_SUBMOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002826 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 data_page);
2828 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002829 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002830 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002831 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002833 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 else
Al Viro2d92ab32008-08-02 00:51:11 -04002835 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 dev_name, data_page);
2837dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002838 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 return retval;
2840}
2841
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002842static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2843{
2844 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2845}
2846
2847static void dec_mnt_namespaces(struct ucounts *ucounts)
2848{
2849 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2850}
2851
Eric W. Biederman771b1372012-07-26 21:08:32 -07002852static void free_mnt_ns(struct mnt_namespace *ns)
2853{
Al Viro6344c432014-11-01 00:45:45 -04002854 ns_free_inum(&ns->ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002855 dec_mnt_namespaces(ns->ucounts);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002856 put_user_ns(ns->user_ns);
2857 kfree(ns);
2858}
2859
Eric W. Biederman8823c072010-03-07 18:49:36 -08002860/*
2861 * Assign a sequence number so we can detect when we attempt to bind
2862 * mount a reference to an older mount namespace into the current
2863 * mount namespace, preventing reference counting loops. A 64bit
2864 * number incrementing at 10Ghz will take 12,427 years to wrap which
2865 * is effectively never, so we can ignore the possibility.
2866 */
2867static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2868
Eric W. Biederman771b1372012-07-26 21:08:32 -07002869static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002870{
2871 struct mnt_namespace *new_ns;
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002872 struct ucounts *ucounts;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002873 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002874
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002875 ucounts = inc_mnt_namespaces(user_ns);
2876 if (!ucounts)
Eric W. Biedermandf75e772016-09-22 13:08:36 -05002877 return ERR_PTR(-ENOSPC);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002878
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002879 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002880 if (!new_ns) {
2881 dec_mnt_namespaces(ucounts);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002882 return ERR_PTR(-ENOMEM);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002883 }
Al Viro6344c432014-11-01 00:45:45 -04002884 ret = ns_alloc_inum(&new_ns->ns);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002885 if (ret) {
2886 kfree(new_ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002887 dec_mnt_namespaces(ucounts);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002888 return ERR_PTR(ret);
2889 }
Al Viro33c42942014-11-01 02:32:53 -04002890 new_ns->ns.ops = &mntns_operations;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002891 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002892 atomic_set(&new_ns->count, 1);
2893 new_ns->root = NULL;
2894 INIT_LIST_HEAD(&new_ns->list);
2895 init_waitqueue_head(&new_ns->poll);
2896 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002897 new_ns->user_ns = get_user_ns(user_ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002898 new_ns->ucounts = ucounts;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002899 new_ns->mounts = 0;
2900 new_ns->pending_mounts = 0;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002901 return new_ns;
2902}
2903
Emese Revfy0766f782016-06-20 20:42:34 +02002904__latent_entropy
Al Viro9559f682013-09-28 20:47:57 -04002905struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2906 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002908 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002909 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002910 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04002911 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05002912 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002913 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914
Al Viro9559f682013-09-28 20:47:57 -04002915 BUG_ON(!ns);
2916
2917 if (likely(!(flags & CLONE_NEWNS))) {
2918 get_mnt_ns(ns);
2919 return ns;
2920 }
2921
2922 old = ns->root;
2923
Eric W. Biederman771b1372012-07-26 21:08:32 -07002924 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002925 if (IS_ERR(new_ns))
2926 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927
Al Viro97216be2013-03-16 15:12:40 -04002928 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002930 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04002931 if (user_ns != ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002932 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002933 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002934 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04002935 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002936 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002937 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 }
Al Virobe08d6d2011-12-06 13:32:36 -05002939 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002940 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941
2942 /*
2943 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2944 * as belonging to new namespace. We have already acquired a private
2945 * fs_struct, so tsk->fs->lock is not needed.
2946 */
Al Viro909b0a82011-11-25 03:06:56 -05002947 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002948 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002950 q->mnt_ns = new_ns;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002951 new_ns->mounts++;
Al Viro9559f682013-09-28 20:47:57 -04002952 if (new_fs) {
2953 if (&p->mnt == new_fs->root.mnt) {
2954 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002955 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 }
Al Viro9559f682013-09-28 20:47:57 -04002957 if (&p->mnt == new_fs->pwd.mnt) {
2958 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002959 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 }
Al Viro909b0a82011-11-25 03:06:56 -05002962 p = next_mnt(p, old);
2963 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002964 if (!q)
2965 break;
2966 while (p->mnt.mnt_root != q->mnt.mnt_root)
2967 p = next_mnt(p, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968 }
Al Viro328e6d92013-03-16 14:49:45 -04002969 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002972 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002974 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
JANAK DESAI741a2952006-02-07 12:59:00 -08002976 return new_ns;
2977}
2978
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002979/**
2980 * create_mnt_ns - creates a private namespace and adds a root filesystem
2981 * @mnt: pointer to the new root filesystem mountpoint
2982 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002983static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002984{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002985 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002986 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002987 struct mount *mnt = real_mount(m);
2988 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002989 new_ns->root = mnt;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002990 new_ns->mounts++;
Al Virob1983cd2013-05-04 15:18:53 -04002991 list_add(&mnt->mnt_list, &new_ns->list);
Al Viroc1334492011-11-16 16:12:14 -05002992 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002993 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002994 }
2995 return new_ns;
2996}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002997
Al Viroea441d12011-11-16 21:43:59 -05002998struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2999{
3000 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05003001 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05003002 struct path path;
3003 int err;
3004
3005 ns = create_mnt_ns(mnt);
3006 if (IS_ERR(ns))
3007 return ERR_CAST(ns);
3008
3009 err = vfs_path_lookup(mnt->mnt_root, mnt,
3010 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3011
3012 put_mnt_ns(ns);
3013
3014 if (err)
3015 return ERR_PTR(err);
3016
3017 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05003018 s = path.mnt->mnt_sb;
3019 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05003020 mntput(path.mnt);
3021 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05003022 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05003023 /* ... and return the root of (sub)tree on it */
3024 return path.dentry;
3025}
3026EXPORT_SYMBOL(mount_subtree);
3027
Heiko Carstensbdc480e2009-01-14 14:14:12 +01003028SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3029 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030{
Vegard Nossumeca6f532009-09-18 13:05:45 -07003031 int ret;
3032 char *kernel_type;
Vegard Nossumeca6f532009-09-18 13:05:45 -07003033 char *kernel_dev;
Al Virob40ef862015-12-14 18:44:44 -05003034 void *options;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035
Tim Gardnerb8850d12014-08-28 11:26:03 -06003036 kernel_type = copy_mount_string(type);
3037 ret = PTR_ERR(kernel_type);
3038 if (IS_ERR(kernel_type))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003039 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040
Tim Gardnerb8850d12014-08-28 11:26:03 -06003041 kernel_dev = copy_mount_string(dev_name);
3042 ret = PTR_ERR(kernel_dev);
3043 if (IS_ERR(kernel_dev))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003044 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045
Al Virob40ef862015-12-14 18:44:44 -05003046 options = copy_mount_options(data);
3047 ret = PTR_ERR(options);
3048 if (IS_ERR(options))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003049 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050
Al Virob40ef862015-12-14 18:44:44 -05003051 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07003052
Al Virob40ef862015-12-14 18:44:44 -05003053 kfree(options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07003054out_data:
3055 kfree(kernel_dev);
3056out_dev:
Vegard Nossumeca6f532009-09-18 13:05:45 -07003057 kfree(kernel_type);
3058out_type:
3059 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060}
3061
3062/*
Al Viroafac7cb2011-11-23 19:34:49 -05003063 * Return true if path is reachable from root
3064 *
Al Viro48a066e2013-09-29 22:06:07 -04003065 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05003066 */
Al Viro643822b2011-11-24 22:00:28 -05003067bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05003068 const struct path *root)
3069{
Al Viro643822b2011-11-24 22:00:28 -05003070 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05003071 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05003072 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05003073 }
Al Viro643822b2011-11-24 22:00:28 -05003074 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05003075}
3076
Yaowei Bai25ab4c92015-11-17 14:40:10 +08003077bool path_is_under(struct path *path1, struct path *path2)
Al Viroafac7cb2011-11-23 19:34:49 -05003078{
Yaowei Bai25ab4c92015-11-17 14:40:10 +08003079 bool res;
Al Viro48a066e2013-09-29 22:06:07 -04003080 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05003081 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04003082 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05003083 return res;
3084}
3085EXPORT_SYMBOL(path_is_under);
3086
3087/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088 * pivot_root Semantics:
3089 * Moves the root file system of the current process to the directory put_old,
3090 * makes new_root as the new root file system of the current process, and sets
3091 * root/cwd of all processes which had them on the current root to new_root.
3092 *
3093 * Restrictions:
3094 * The new_root and put_old must be directories, and must not be on the
3095 * same file system as the current process root. The put_old must be
3096 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3097 * pointed to by put_old must yield the same directory as new_root. No other
3098 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3099 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08003100 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3101 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3102 * in this situation.
3103 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104 * Notes:
3105 * - we don't move root/cwd if they are not at the root (reason: if something
3106 * cared enough to change them, it's probably wrong to force them elsewhere)
3107 * - it's okay to pick a root that isn't the root of a file system, e.g.
3108 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3109 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3110 * first.
3111 */
Heiko Carstens3480b252009-01-14 14:14:16 +01003112SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3113 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114{
Al Viro2d8f3032008-07-22 09:59:21 -04003115 struct path new, old, parent_path, root_parent, root;
Al Viro84d17192013-03-15 10:53:28 -04003116 struct mount *new_mnt, *root_mnt, *old_mnt;
3117 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 int error;
3119
Al Viro9b40bc92013-02-22 22:45:42 -05003120 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121 return -EPERM;
3122
Al Viro2d8f3032008-07-22 09:59:21 -04003123 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124 if (error)
3125 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126
Al Viro2d8f3032008-07-22 09:59:21 -04003127 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 if (error)
3129 goto out1;
3130
Al Viro2d8f3032008-07-22 09:59:21 -04003131 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04003132 if (error)
3133 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02003135 get_fs_root(current->fs, &root);
Al Viro84d17192013-03-15 10:53:28 -04003136 old_mp = lock_mount(&old);
3137 error = PTR_ERR(old_mp);
3138 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04003139 goto out3;
3140
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05003142 new_mnt = real_mount(new.mnt);
3143 root_mnt = real_mount(root.mnt);
Al Viro84d17192013-03-15 10:53:28 -04003144 old_mnt = real_mount(old.mnt);
3145 if (IS_MNT_SHARED(old_mnt) ||
Al Virofc7be132011-11-25 01:05:37 -05003146 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3147 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04003148 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05003149 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003150 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003151 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3152 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003154 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04003155 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003156 error = -EBUSY;
Al Viro84d17192013-03-15 10:53:28 -04003157 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04003158 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04003160 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04003161 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05003162 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003163 goto out4; /* not attached */
Al Viro84d17192013-03-15 10:53:28 -04003164 root_mp = root_mnt->mnt_mp;
Al Viro2d8f3032008-07-22 09:59:21 -04003165 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04003166 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05003167 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003168 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08003169 /* make sure we can reach put_old from new_root */
Al Viro84d17192013-03-15 10:53:28 -04003170 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04003171 goto out4;
Eric W. Biederman0d082602014-10-08 10:42:27 -07003172 /* make certain new is below the root */
3173 if (!is_path_reachable(new_mnt, new.dentry, &root))
3174 goto out4;
Al Viro84d17192013-03-15 10:53:28 -04003175 root_mp->m_count++; /* pin it so it won't go away */
Al Viro719ea2f2013-09-29 11:24:49 -04003176 lock_mount_hash();
Al Viro419148d2011-11-24 19:41:16 -05003177 detach_mnt(new_mnt, &parent_path);
3178 detach_mnt(root_mnt, &root_parent);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003179 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3180 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3181 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3182 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003183 /* mount old root on put_old */
Al Viro84d17192013-03-15 10:53:28 -04003184 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08003185 /* mount new_root on / */
Al Viro84d17192013-03-15 10:53:28 -04003186 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003187 touch_mnt_namespace(current->nsproxy->mnt_ns);
Eric W. Biederman4fed6552014-10-08 10:42:57 -07003188 /* A moved mount should not expire automatically */
3189 list_del_init(&new_mnt->mnt_expire);
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +13003190 put_mountpoint(root_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04003191 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04003192 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04003194out4:
Al Viro84d17192013-03-15 10:53:28 -04003195 unlock_mount(old_mp);
Al Virob12cea92011-03-18 08:55:38 -04003196 if (!error) {
3197 path_put(&root_parent);
3198 path_put(&parent_path);
3199 }
3200out3:
Al Viro8c3ee422008-03-22 18:00:39 -04003201 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04003202out2:
Al Viro2d8f3032008-07-22 09:59:21 -04003203 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204out1:
Al Viro2d8f3032008-07-22 09:59:21 -04003205 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208}
3209
3210static void __init init_mount_tree(void)
3211{
3212 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003213 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08003214 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003215 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003217 type = get_fs_type("rootfs");
3218 if (!type)
3219 panic("Can't find rootfs type");
3220 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3221 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222 if (IS_ERR(mnt))
3223 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11003224
Trond Myklebust3b22edc2009-06-23 17:29:49 -04003225 ns = create_mnt_ns(mnt);
3226 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003229 init_task.nsproxy->mnt_ns = ns;
3230 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231
Al Virobe08d6d2011-12-06 13:32:36 -05003232 root.mnt = mnt;
3233 root.dentry = mnt->mnt_root;
Eric W. Biedermanda362b02014-10-07 12:19:53 -07003234 mnt->mnt_flags |= MNT_LOCKED;
Jan Blunckac748a02008-02-14 19:34:39 -08003235
3236 set_fs_pwd(current->fs, &root);
3237 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238}
3239
Denis Cheng74bf17c2007-10-16 23:26:30 -07003240void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241{
Eric Dumazet13f14b42008-02-06 01:37:57 -08003242 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003243 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244
Al Viro7d6fec42011-11-23 12:14:10 -05003245 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09003246 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247
Al Viro0818bf22014-02-28 13:46:44 -05003248 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04003249 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05003250 mhash_entries, 19,
3251 0,
3252 &m_hash_shift, &m_hash_mask, 0, 0);
3253 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3254 sizeof(struct hlist_head),
3255 mphash_entries, 19,
3256 0,
3257 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258
Al Viro84d17192013-03-15 10:53:28 -04003259 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260 panic("Failed to allocate mount hash table\n");
3261
Al Viro0818bf22014-02-28 13:46:44 -05003262 for (u = 0; u <= m_hash_mask; u++)
Al Viro38129a12014-03-20 21:10:51 -04003263 INIT_HLIST_HEAD(&mount_hashtable[u]);
Al Viro0818bf22014-02-28 13:46:44 -05003264 for (u = 0; u <= mp_hash_mask; u++)
3265 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266
Tejun Heo4b93dc92013-11-28 14:54:43 -05003267 kernfs_init();
3268
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003269 err = sysfs_init();
3270 if (err)
3271 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003272 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06003273 fs_kobj = kobject_create_and_add("fs", NULL);
3274 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003275 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276 init_rootfs();
3277 init_mount_tree();
3278}
3279
Trond Myklebust616511d2009-06-22 15:09:13 -04003280void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281{
Al Virod498b252010-02-05 02:21:06 -05003282 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04003283 return;
Al Viro7b00ed62013-09-16 21:19:20 -04003284 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003285 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286}
Al Viro9d412a42011-03-17 22:08:28 -04003287
3288struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3289{
Tim Chen423e0ab2011-07-19 09:32:38 -07003290 struct vfsmount *mnt;
3291 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3292 if (!IS_ERR(mnt)) {
3293 /*
3294 * it is a longterm mount, don't release mnt until
3295 * we unmount before file sys is unregistered
3296 */
Al Virof7a99c52012-06-09 00:59:08 -04003297 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07003298 }
3299 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04003300}
3301EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07003302
3303void kern_unmount(struct vfsmount *mnt)
3304{
3305 /* release long term mount so mount point can be released */
3306 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04003307 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04003308 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07003309 mntput(mnt);
3310 }
3311}
3312EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05003313
3314bool our_mnt(struct vfsmount *mnt)
3315{
Al Viro143c8c92011-11-25 00:46:35 -05003316 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05003317}
Eric W. Biederman8823c072010-03-07 18:49:36 -08003318
Eric W. Biederman31515272013-03-15 01:45:51 -07003319bool current_chrooted(void)
3320{
3321 /* Does the current process have a non-standard root */
3322 struct path ns_root;
3323 struct path fs_root;
3324 bool chrooted;
3325
3326 /* Find the namespace root */
3327 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3328 ns_root.dentry = ns_root.mnt->mnt_root;
3329 path_get(&ns_root);
3330 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3331 ;
3332
3333 get_fs_root(current->fs, &fs_root);
3334
3335 chrooted = !path_equal(&fs_root, &ns_root);
3336
3337 path_put(&fs_root);
3338 path_put(&ns_root);
3339
3340 return chrooted;
3341}
3342
Eric W. Biederman8654df42016-06-09 16:06:06 -05003343static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3344 int *new_mnt_flags)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003345{
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003346 int new_flags = *new_mnt_flags;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003347 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07003348 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003349
Al Viro44bb4382013-09-16 21:37:36 -04003350 down_read(&namespace_sem);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003351 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07003352 struct mount *child;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003353 int mnt_flags;
3354
Eric W. Biederman8654df42016-06-09 16:06:06 -05003355 if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
Eric W. Biedermane51db732013-03-30 19:57:41 -07003356 continue;
3357
Eric W. Biederman7e96c1b2015-05-08 16:36:50 -05003358 /* This mount is not fully visible if it's root directory
3359 * is not the root directory of the filesystem.
3360 */
3361 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3362 continue;
3363
Eric W. Biedermana1935c12016-06-15 06:59:49 -05003364 /* A local view of the mount flags */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003365 mnt_flags = mnt->mnt.mnt_flags;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003366
Eric W. Biederman695e9df2016-06-10 12:21:40 -05003367 /* Don't miss readonly hidden in the superblock flags */
3368 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3369 mnt_flags |= MNT_LOCK_READONLY;
3370
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003371 /* Verify the mount flags are equal to or more permissive
3372 * than the proposed new mount.
3373 */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003374 if ((mnt_flags & MNT_LOCK_READONLY) &&
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003375 !(new_flags & MNT_READONLY))
3376 continue;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003377 if ((mnt_flags & MNT_LOCK_ATIME) &&
3378 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003379 continue;
3380
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003381 /* This mount is not fully visible if there are any
3382 * locked child mounts that cover anything except for
3383 * empty directories.
Eric W. Biedermane51db732013-03-30 19:57:41 -07003384 */
3385 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3386 struct inode *inode = child->mnt_mountpoint->d_inode;
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003387 /* Only worry about locked mounts */
Eric W. Biedermand71ed6c2016-05-27 14:50:05 -05003388 if (!(child->mnt.mnt_flags & MNT_LOCKED))
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003389 continue;
Eric W. Biederman7236c852015-05-13 20:51:09 -05003390 /* Is the directory permanetly empty? */
3391 if (!is_empty_dir_inode(inode))
Eric W. Biedermane51db732013-03-30 19:57:41 -07003392 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003393 }
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003394 /* Preserve the locked attributes */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003395 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003396 MNT_LOCK_ATIME);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003397 visible = true;
3398 goto found;
3399 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003400 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07003401found:
Al Viro44bb4382013-09-16 21:37:36 -04003402 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003403 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003404}
3405
Eric W. Biederman8654df42016-06-09 16:06:06 -05003406static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3407{
Eric W. Biedermana1935c12016-06-15 06:59:49 -05003408 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
Eric W. Biederman8654df42016-06-09 16:06:06 -05003409 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3410 unsigned long s_iflags;
3411
3412 if (ns->user_ns == &init_user_ns)
3413 return false;
3414
3415 /* Can this filesystem be too revealing? */
3416 s_iflags = mnt->mnt_sb->s_iflags;
3417 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3418 return false;
3419
Eric W. Biedermana1935c12016-06-15 06:59:49 -05003420 if ((s_iflags & required_iflags) != required_iflags) {
3421 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3422 required_iflags);
3423 return true;
3424 }
3425
Eric W. Biederman8654df42016-06-09 16:06:06 -05003426 return !mnt_already_visible(ns, mnt, new_mnt_flags);
3427}
3428
Andy Lutomirski380cf5b2016-06-23 16:41:05 -05003429bool mnt_may_suid(struct vfsmount *mnt)
3430{
3431 /*
3432 * Foreign mounts (accessed via fchdir or through /proc
3433 * symlinks) are always treated as if they are nosuid. This
3434 * prevents namespaces from trusting potentially unsafe
3435 * suid/sgid bits, file caps, or security labels that originate
3436 * in other namespaces.
3437 */
3438 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3439 current_in_userns(mnt->mnt_sb->s_user_ns);
3440}
3441
Al Viro64964522014-11-01 00:37:32 -04003442static struct ns_common *mntns_get(struct task_struct *task)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003443{
Al Viro58be28252014-11-01 00:00:23 -04003444 struct ns_common *ns = NULL;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003445 struct nsproxy *nsproxy;
3446
Eric W. Biederman728dba32014-02-03 19:13:49 -08003447 task_lock(task);
3448 nsproxy = task->nsproxy;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003449 if (nsproxy) {
Al Viro58be28252014-11-01 00:00:23 -04003450 ns = &nsproxy->mnt_ns->ns;
3451 get_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003452 }
Eric W. Biederman728dba32014-02-03 19:13:49 -08003453 task_unlock(task);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003454
3455 return ns;
3456}
3457
Al Viro64964522014-11-01 00:37:32 -04003458static void mntns_put(struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003459{
Al Viro58be28252014-11-01 00:00:23 -04003460 put_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003461}
3462
Al Viro64964522014-11-01 00:37:32 -04003463static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003464{
3465 struct fs_struct *fs = current->fs;
Al Viro58be28252014-11-01 00:00:23 -04003466 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003467 struct path root;
3468
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003469 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -07003470 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3471 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08003472 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003473
3474 if (fs->users != 1)
3475 return -EINVAL;
3476
3477 get_mnt_ns(mnt_ns);
3478 put_mnt_ns(nsproxy->mnt_ns);
3479 nsproxy->mnt_ns = mnt_ns;
3480
3481 /* Find the root */
3482 root.mnt = &mnt_ns->root->mnt;
3483 root.dentry = mnt_ns->root->mnt.mnt_root;
3484 path_get(&root);
3485 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3486 ;
3487
3488 /* Update the pwd and root */
3489 set_fs_pwd(fs, &root);
3490 set_fs_root(fs, &root);
3491
3492 path_put(&root);
3493 return 0;
3494}
3495
Andrey Vaginbcac25a2016-09-06 00:47:13 -07003496static struct user_namespace *mntns_owner(struct ns_common *ns)
3497{
3498 return to_mnt_ns(ns)->user_ns;
3499}
3500
Eric W. Biederman8823c072010-03-07 18:49:36 -08003501const struct proc_ns_operations mntns_operations = {
3502 .name = "mnt",
3503 .type = CLONE_NEWNS,
3504 .get = mntns_get,
3505 .put = mntns_put,
3506 .install = mntns_install,
Andrey Vaginbcac25a2016-09-06 00:47:13 -07003507 .owner = mntns_owner,
Eric W. Biederman8823c072010-03-07 18:49:36 -08003508};