blob: f8893dc6a989dc6c9b0f4e8daba2e985a64976e6 [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>
Ingo Molnar5b825c32017-02-02 17:54:15 +010018#include <linux/cred.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010019#include <linux/idr.h>
Rob Landley57f150a2013-09-11 14:26:10 -070020#include <linux/init.h> /* init_rootfs */
Al Virod10577a2011-12-07 13:06:11 -050021#include <linux/fs_struct.h> /* get_fs_root et.al. */
22#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23#include <linux/uaccess.h>
David Howells0bb80f22013-04-12 01:50:06 +010024#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070025#include <linux/magic.h>
Al Viro0818bf22014-02-28 13:46:44 -050026#include <linux/bootmem.h>
Al Viro9ea459e2014-08-08 13:08:20 -040027#include <linux/task_work.h>
Ingo Molnar9164bb42017-02-04 01:20:53 +010028#include <linux/sched/task.h>
29
Ram Pai07b20882005-11-07 17:19:07 -050030#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070031#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Eric W. Biedermand2921682016-09-28 00:27:17 -050033/* Maximum number of mounts in a mount namespace */
34unsigned int sysctl_mount_max __read_mostly = 100000;
35
Al Viro0818bf22014-02-28 13:46:44 -050036static unsigned int m_hash_mask __read_mostly;
37static unsigned int m_hash_shift __read_mostly;
38static unsigned int mp_hash_mask __read_mostly;
39static unsigned int mp_hash_shift __read_mostly;
40
41static __initdata unsigned long mhash_entries;
42static int __init set_mhash_entries(char *str)
43{
44 if (!str)
45 return 0;
46 mhash_entries = simple_strtoul(str, &str, 0);
47 return 1;
48}
49__setup("mhash_entries=", set_mhash_entries);
50
51static __initdata unsigned long mphash_entries;
52static int __init set_mphash_entries(char *str)
53{
54 if (!str)
55 return 0;
56 mphash_entries = simple_strtoul(str, &str, 0);
57 return 1;
58}
59__setup("mphash_entries=", set_mphash_entries);
Eric Dumazet13f14b42008-02-06 01:37:57 -080060
Al Viroc7999c32014-02-27 14:40:10 -050061static u64 event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010062static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010063static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100064static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040065static int mnt_id_start = 0;
66static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Al Viro38129a12014-03-20 21:10:51 -040068static struct hlist_head *mount_hashtable __read_mostly;
Al Viro0818bf22014-02-28 13:46:44 -050069static struct hlist_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080070static struct kmem_cache *mnt_cache __read_mostly;
Al Viro59aa0da2013-09-16 21:34:53 -040071static DECLARE_RWSEM(namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080073/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060074struct kobject *fs_kobj;
75EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080076
Nick Piggin99b7db72010-08-18 04:37:39 +100077/*
78 * vfsmount lock may be taken for read to prevent changes to the
79 * vfsmount hash, ie. during mountpoint lookups or walking back
80 * up the tree.
81 *
82 * It should be taken for write in all cases where the vfsmount
83 * tree or hash is modified or when a vfsmount structure is modified.
84 */
Al Viro48a066e2013-09-29 22:06:07 -040085__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +100086
Al Viro38129a12014-03-20 21:10:51 -040087static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
Ram Paib58fed82005-11-07 17:16:09 -050089 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
90 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Al Viro0818bf22014-02-28 13:46:44 -050091 tmp = tmp + (tmp >> m_hash_shift);
92 return &mount_hashtable[tmp & m_hash_mask];
93}
94
95static inline struct hlist_head *mp_hash(struct dentry *dentry)
96{
97 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
98 tmp = tmp + (tmp >> mp_hash_shift);
99 return &mountpoint_hashtable[tmp & mp_hash_mask];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100}
101
Al Virob105e272011-11-24 20:38:33 -0500102static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100103{
104 int res;
105
106retry:
107 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +1000108 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -0500109 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -0400110 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500111 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +1000112 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100113 if (res == -EAGAIN)
114 goto retry;
115
116 return res;
117}
118
Al Virob105e272011-11-24 20:38:33 -0500119static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100120{
Al Viro15169fe2011-11-25 00:50:41 -0500121 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000122 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -0400123 ida_remove(&mnt_id_ida, id);
124 if (mnt_id_start > id)
125 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000126 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100127}
128
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100129/*
130 * Allocate a new peer group ID
131 *
132 * mnt_group_ida is protected by namespace_sem
133 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500134static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100135{
Al Virof21f6222009-06-24 03:12:00 -0400136 int res;
137
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100138 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
139 return -ENOMEM;
140
Al Virof21f6222009-06-24 03:12:00 -0400141 res = ida_get_new_above(&mnt_group_ida,
142 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500143 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400144 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500145 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400146
147 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100148}
149
150/*
151 * Release a peer group ID
152 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500153void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100154{
Al Viro15169fe2011-11-25 00:50:41 -0500155 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400156 ida_remove(&mnt_group_ida, id);
157 if (mnt_group_start > id)
158 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500159 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100160}
161
Nick Pigginb3e19d92011-01-07 17:50:11 +1100162/*
163 * vfsmount lock must be held for read
164 */
Al Viro83adc752011-11-24 22:37:54 -0500165static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100166{
167#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500168 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100169#else
170 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500171 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100172 preempt_enable();
173#endif
174}
175
Nick Pigginb3e19d92011-01-07 17:50:11 +1100176/*
177 * vfsmount lock must be held for write
178 */
Al Viro83adc752011-11-24 22:37:54 -0500179unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100180{
181#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500182 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100183 int cpu;
184
185 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500186 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100187 }
188
189 return count;
190#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500191 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100192#endif
193}
194
Al Viro87b95ce2015-01-10 19:01:08 -0500195static void drop_mountpoint(struct fs_pin *p)
196{
197 struct mount *m = container_of(p, struct mount, mnt_umount);
198 dput(m->mnt_ex_mountpoint);
199 pin_remove(p);
200 mntput(&m->mnt);
201}
202
Al Virob105e272011-11-24 20:38:33 -0500203static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
Al Viroc63181e2011-11-25 02:35:16 -0500205 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
206 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100207 int err;
208
Al Viroc63181e2011-11-25 02:35:16 -0500209 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800210 if (err)
211 goto out_free_cache;
212
213 if (name) {
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800214 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
Al Viroc63181e2011-11-25 02:35:16 -0500215 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800216 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100217 }
218
Nick Pigginb3e19d92011-01-07 17:50:11 +1100219#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500220 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
221 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100222 goto out_free_devname;
223
Al Viroc63181e2011-11-25 02:35:16 -0500224 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100225#else
Al Viroc63181e2011-11-25 02:35:16 -0500226 mnt->mnt_count = 1;
227 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100228#endif
229
Al Viro38129a12014-03-20 21:10:51 -0400230 INIT_HLIST_NODE(&mnt->mnt_hash);
Al Viroc63181e2011-11-25 02:35:16 -0500231 INIT_LIST_HEAD(&mnt->mnt_child);
232 INIT_LIST_HEAD(&mnt->mnt_mounts);
233 INIT_LIST_HEAD(&mnt->mnt_list);
234 INIT_LIST_HEAD(&mnt->mnt_expire);
235 INIT_LIST_HEAD(&mnt->mnt_share);
236 INIT_LIST_HEAD(&mnt->mnt_slave_list);
237 INIT_LIST_HEAD(&mnt->mnt_slave);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700238 INIT_HLIST_NODE(&mnt->mnt_mp_list);
Eric W. Biederman99b19d12016-10-24 16:16:13 -0500239 INIT_LIST_HEAD(&mnt->mnt_umounting);
Al Viro87b95ce2015-01-10 19:01:08 -0500240 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 }
Al Viroc63181e2011-11-25 02:35:16 -0500242 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800243
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000244#ifdef CONFIG_SMP
245out_free_devname:
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800246 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000247#endif
Li Zefan88b38782008-07-21 18:06:36 +0800248out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500249 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800250out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500251 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800252 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253}
254
Dave Hansen83660252008-02-15 14:37:30 -0800255/*
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
261 * a filesystem.
262 */
Dave Hansen3d733632008-02-15 14:37:59 -0800263/*
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
266 *
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*. This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
272 * r/w.
273 */
274int __mnt_is_readonly(struct vfsmount *mnt)
275{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800276 if (mnt->mnt_flags & MNT_READONLY)
277 return 1;
278 if (mnt->mnt_sb->s_flags & MS_RDONLY)
279 return 1;
280 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800281}
282EXPORT_SYMBOL_GPL(__mnt_is_readonly);
283
Al Viro83adc752011-11-24 22:37:54 -0500284static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800285{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000286#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500287 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000288#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500289 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000290#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800291}
Dave Hansen3d733632008-02-15 14:37:59 -0800292
Al Viro83adc752011-11-24 22:37:54 -0500293static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800294{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000295#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500296 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000297#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500298 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000299#endif
300}
301
Al Viro83adc752011-11-24 22:37:54 -0500302static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000303{
304#ifdef CONFIG_SMP
305 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800306 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800307
308 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500309 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800310 }
Dave Hansen3d733632008-02-15 14:37:59 -0800311
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000312 return count;
313#else
314 return mnt->mnt_writers;
315#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800316}
317
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100318static int mnt_is_readonly(struct vfsmount *mnt)
319{
320 if (mnt->mnt_sb->s_readonly_remount)
321 return 1;
322 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
323 smp_rmb();
324 return __mnt_is_readonly(mnt);
325}
326
Dave Hansen3d733632008-02-15 14:37:59 -0800327/*
Jan Karaeb04c282012-06-12 16:20:35 +0200328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink(). We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800332 */
Dave Hansen83660252008-02-15 14:37:30 -0800333/**
Jan Karaeb04c282012-06-12 16:20:35 +0200334 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500335 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800336 *
Jan Karaeb04c282012-06-12 16:20:35 +0200337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800342 */
Jan Karaeb04c282012-06-12 16:20:35 +0200343int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800344{
Al Viro83adc752011-11-24 22:37:54 -0500345 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800346 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800347
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000348 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100349 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000350 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100351 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000352 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 * incremented count after it has set MNT_WRITE_HOLD.
354 */
355 smp_mb();
Miao Xie1e755292012-11-16 17:23:50 +0800356 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000357 cpu_relax();
358 /*
359 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 * be set to match its requirements. So we must not load that until
361 * MNT_WRITE_HOLD is cleared.
362 */
363 smp_rmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100364 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100365 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800366 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800367 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000368 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200369
370 return ret;
371}
372
373/**
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
376 *
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success. When the write operation is
380 * finished, mnt_drop_write() must be called. This is effectively a refcount.
381 */
382int mnt_want_write(struct vfsmount *m)
383{
384 int ret;
385
386 sb_start_write(m->mnt_sb);
387 ret = __mnt_want_write(m);
388 if (ret)
389 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800390 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800391}
392EXPORT_SYMBOL_GPL(mnt_want_write);
393
394/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
397 *
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
402 *
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
405 */
406int mnt_clone_write(struct vfsmount *mnt)
407{
408 /* superblock may be r/o */
409 if (__mnt_is_readonly(mnt))
410 return -EROFS;
411 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500412 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000413 preempt_enable();
414 return 0;
415}
416EXPORT_SYMBOL_GPL(mnt_clone_write);
417
418/**
Jan Karaeb04c282012-06-12 16:20:35 +0200419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
421 *
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
424 */
425int __mnt_want_write_file(struct file *file)
426{
Al Viro83f936c2014-03-14 12:02:47 -0400427 if (!(file->f_mode & FMODE_WRITER))
Jan Karaeb04c282012-06-12 16:20:35 +0200428 return __mnt_want_write(file->f_path.mnt);
429 else
430 return mnt_clone_write(file->f_path.mnt);
431}
432
433/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000434 * mnt_want_write_file - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
436 *
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
439 */
440int mnt_want_write_file(struct file *file)
441{
Jan Karaeb04c282012-06-12 16:20:35 +0200442 int ret;
443
444 sb_start_write(file->f_path.mnt->mnt_sb);
445 ret = __mnt_want_write_file(file);
446 if (ret)
447 sb_end_write(file->f_path.mnt->mnt_sb);
448 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000449}
450EXPORT_SYMBOL_GPL(mnt_want_write_file);
451
452/**
Jan Karaeb04c282012-06-12 16:20:35 +0200453 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800454 * @mnt: the mount on which to give up write access
455 *
456 * Tells the low-level filesystem that we are done
457 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200458 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800459 */
Jan Karaeb04c282012-06-12 16:20:35 +0200460void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800461{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000462 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500463 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000464 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800465}
Jan Karaeb04c282012-06-12 16:20:35 +0200466
467/**
468 * mnt_drop_write - give up write access to a mount
469 * @mnt: the mount on which to give up write access
470 *
471 * Tells the low-level filesystem that we are done performing writes to it and
472 * also allows filesystem to be frozen again. Must be matched with
473 * mnt_want_write() call above.
474 */
475void mnt_drop_write(struct vfsmount *mnt)
476{
477 __mnt_drop_write(mnt);
478 sb_end_write(mnt->mnt_sb);
479}
Dave Hansen83660252008-02-15 14:37:30 -0800480EXPORT_SYMBOL_GPL(mnt_drop_write);
481
Jan Karaeb04c282012-06-12 16:20:35 +0200482void __mnt_drop_write_file(struct file *file)
483{
484 __mnt_drop_write(file->f_path.mnt);
485}
486
Al Viro2a79f172011-12-09 08:06:57 -0500487void mnt_drop_write_file(struct file *file)
488{
489 mnt_drop_write(file->f_path.mnt);
490}
491EXPORT_SYMBOL(mnt_drop_write_file);
492
Al Viro83adc752011-11-24 22:37:54 -0500493static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800494{
Dave Hansen3d733632008-02-15 14:37:59 -0800495 int ret = 0;
496
Al Viro719ea2f2013-09-29 11:24:49 -0400497 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500498 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000499 /*
500 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501 * should be visible before we do.
502 */
503 smp_mb();
504
505 /*
506 * With writers on hold, if this value is zero, then there are
507 * definitely no active writers (although held writers may subsequently
508 * increment the count, they'll have to wait, and decrement it after
509 * seeing MNT_READONLY).
510 *
511 * It is OK to have counter incremented on one CPU and decremented on
512 * another: the sum will add up correctly. The danger would be when we
513 * sum up each counter, if we read a counter before it is incremented,
514 * but then read another CPU's count which it has been subsequently
515 * decremented from -- we would see more decrements than we should.
516 * MNT_WRITE_HOLD protects against this scenario, because
517 * mnt_want_write first increments count, then smp_mb, then spins on
518 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519 * we're counting up here.
520 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100521 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000522 ret = -EBUSY;
523 else
Al Viro83adc752011-11-24 22:37:54 -0500524 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000525 /*
526 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527 * that become unheld will see MNT_READONLY.
528 */
529 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500530 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Al Viro719ea2f2013-09-29 11:24:49 -0400531 unlock_mount_hash();
Dave Hansen3d733632008-02-15 14:37:59 -0800532 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800533}
Dave Hansen83660252008-02-15 14:37:30 -0800534
Al Viro83adc752011-11-24 22:37:54 -0500535static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800536{
Al Viro719ea2f2013-09-29 11:24:49 -0400537 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500538 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Al Viro719ea2f2013-09-29 11:24:49 -0400539 unlock_mount_hash();
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800540}
541
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100542int sb_prepare_remount_readonly(struct super_block *sb)
543{
544 struct mount *mnt;
545 int err = 0;
546
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100547 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
548 if (atomic_long_read(&sb->s_remove_count))
549 return -EBUSY;
550
Al Viro719ea2f2013-09-29 11:24:49 -0400551 lock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100552 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
553 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
554 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
555 smp_mb();
556 if (mnt_get_writers(mnt) > 0) {
557 err = -EBUSY;
558 break;
559 }
560 }
561 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100562 if (!err && atomic_long_read(&sb->s_remove_count))
563 err = -EBUSY;
564
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100565 if (!err) {
566 sb->s_readonly_remount = 1;
567 smp_wmb();
568 }
569 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
570 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
571 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
572 }
Al Viro719ea2f2013-09-29 11:24:49 -0400573 unlock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100574
575 return err;
576}
577
Al Virob105e272011-11-24 20:38:33 -0500578static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800580 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000581#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500582 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000583#endif
Al Virob105e272011-11-24 20:38:33 -0500584 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585}
586
David Howells8ffcb322014-01-24 12:17:54 +0000587static void delayed_free_vfsmnt(struct rcu_head *head)
588{
589 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
590}
591
Al Viro48a066e2013-09-29 22:06:07 -0400592/* call under rcu_read_lock */
Al Viro294d71f2015-05-08 11:43:53 -0400593int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
Al Viro48a066e2013-09-29 22:06:07 -0400594{
595 struct mount *mnt;
596 if (read_seqretry(&mount_lock, seq))
Al Viro294d71f2015-05-08 11:43:53 -0400597 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400598 if (bastard == NULL)
Al Viro294d71f2015-05-08 11:43:53 -0400599 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400600 mnt = real_mount(bastard);
601 mnt_add_count(mnt, 1);
602 if (likely(!read_seqretry(&mount_lock, seq)))
Al Viro294d71f2015-05-08 11:43:53 -0400603 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400604 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605 mnt_add_count(mnt, -1);
Al Viro294d71f2015-05-08 11:43:53 -0400606 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400607 }
Al Viro294d71f2015-05-08 11:43:53 -0400608 return -1;
609}
610
611/* call under rcu_read_lock */
612bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
613{
614 int res = __legitimize_mnt(bastard, seq);
615 if (likely(!res))
616 return true;
617 if (unlikely(res < 0)) {
618 rcu_read_unlock();
619 mntput(bastard);
620 rcu_read_lock();
621 }
Al Viro48a066e2013-09-29 22:06:07 -0400622 return false;
623}
624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625/*
Al Viro474279d2013-10-01 16:11:26 -0400626 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400627 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 */
Al Viro474279d2013-10-01 16:11:26 -0400629struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Al Viro38129a12014-03-20 21:10:51 -0400631 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400632 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Al Viro38129a12014-03-20 21:10:51 -0400634 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400635 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
636 return p;
637 return NULL;
638}
639
640/*
David Howellsf015f1262012-06-25 12:55:28 +0100641 * lookup_mnt - Return the first child mount mounted at path
642 *
643 * "First" means first mounted chronologically. If you create the
644 * following mounts:
645 *
646 * mount /dev/sda1 /mnt
647 * mount /dev/sda2 /mnt
648 * mount /dev/sda3 /mnt
649 *
650 * Then lookup_mnt() on the base /mnt dentry in the root mount will
651 * return successively the root dentry and vfsmount of /dev/sda1, then
652 * /dev/sda2, then /dev/sda3, then NULL.
653 *
654 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500655 */
Al Viroca71cf72016-11-20 19:45:28 -0500656struct vfsmount *lookup_mnt(const struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500657{
Al Viroc7105362011-11-24 18:22:03 -0500658 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400659 struct vfsmount *m;
660 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000661
Al Viro48a066e2013-09-29 22:06:07 -0400662 rcu_read_lock();
663 do {
664 seq = read_seqbegin(&mount_lock);
665 child_mnt = __lookup_mnt(path->mnt, path->dentry);
666 m = child_mnt ? &child_mnt->mnt : NULL;
667 } while (!legitimize_mnt(m, seq));
668 rcu_read_unlock();
669 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500670}
671
Eric W. Biederman7af13642013-10-04 19:15:13 -0700672/*
673 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
674 * current mount namespace.
675 *
676 * The common case is dentries are not mountpoints at all and that
677 * test is handled inline. For the slow case when we are actually
678 * dealing with a mountpoint of some kind, walk through all of the
679 * mounts in the current mount namespace and test to see if the dentry
680 * is a mountpoint.
681 *
682 * The mount_hashtable is not usable in the context because we
683 * need to identify all mounts that may be in the current mount
684 * namespace not just a mount that happens to have some specified
685 * parent mount.
686 */
687bool __is_local_mountpoint(struct dentry *dentry)
688{
689 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
690 struct mount *mnt;
691 bool is_covered = false;
692
693 if (!d_mountpoint(dentry))
694 goto out;
695
696 down_read(&namespace_sem);
697 list_for_each_entry(mnt, &ns->list, mnt_list) {
698 is_covered = (mnt->mnt_mountpoint == dentry);
699 if (is_covered)
700 break;
701 }
702 up_read(&namespace_sem);
703out:
704 return is_covered;
705}
706
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800707static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
Al Viro84d17192013-03-15 10:53:28 -0400708{
Al Viro0818bf22014-02-28 13:46:44 -0500709 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d17192013-03-15 10:53:28 -0400710 struct mountpoint *mp;
711
Al Viro0818bf22014-02-28 13:46:44 -0500712 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d17192013-03-15 10:53:28 -0400713 if (mp->m_dentry == dentry) {
714 /* might be worth a WARN_ON() */
715 if (d_unlinked(dentry))
716 return ERR_PTR(-ENOENT);
717 mp->m_count++;
718 return mp;
719 }
720 }
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800721 return NULL;
722}
723
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300724static struct mountpoint *get_mountpoint(struct dentry *dentry)
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800725{
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300726 struct mountpoint *mp, *new = NULL;
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800727 int ret;
Al Viro84d17192013-03-15 10:53:28 -0400728
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300729 if (d_mountpoint(dentry)) {
730mountpoint:
731 read_seqlock_excl(&mount_lock);
732 mp = lookup_mountpoint(dentry);
733 read_sequnlock_excl(&mount_lock);
734 if (mp)
735 goto done;
Al Viro84d17192013-03-15 10:53:28 -0400736 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200737
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300738 if (!new)
739 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
740 if (!new)
741 return ERR_PTR(-ENOMEM);
742
743
744 /* Exactly one processes may set d_mounted */
745 ret = d_set_mounted(dentry);
746
747 /* Someone else set d_mounted? */
748 if (ret == -EBUSY)
749 goto mountpoint;
750
751 /* The dentry is not available as a mountpoint? */
752 mp = ERR_PTR(ret);
753 if (ret)
754 goto done;
755
756 /* Add the new mountpoint to the hash table */
757 read_seqlock_excl(&mount_lock);
758 new->m_dentry = dentry;
759 new->m_count = 1;
760 hlist_add_head(&new->m_hash, mp_hash(dentry));
761 INIT_HLIST_HEAD(&new->m_list);
762 read_sequnlock_excl(&mount_lock);
763
764 mp = new;
765 new = NULL;
766done:
767 kfree(new);
Al Viro84d17192013-03-15 10:53:28 -0400768 return mp;
769}
770
771static void put_mountpoint(struct mountpoint *mp)
772{
773 if (!--mp->m_count) {
774 struct dentry *dentry = mp->m_dentry;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700775 BUG_ON(!hlist_empty(&mp->m_list));
Al Viro84d17192013-03-15 10:53:28 -0400776 spin_lock(&dentry->d_lock);
777 dentry->d_flags &= ~DCACHE_MOUNTED;
778 spin_unlock(&dentry->d_lock);
Al Viro0818bf22014-02-28 13:46:44 -0500779 hlist_del(&mp->m_hash);
Al Viro84d17192013-03-15 10:53:28 -0400780 kfree(mp);
781 }
782}
783
Al Viro143c8c92011-11-25 00:46:35 -0500784static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800786 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787}
788
Nick Piggin99b7db72010-08-18 04:37:39 +1000789/*
790 * vfsmount lock must be held for write
791 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800792static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500793{
794 if (ns) {
795 ns->event = ++event;
796 wake_up_interruptible(&ns->poll);
797 }
798}
799
Nick Piggin99b7db72010-08-18 04:37:39 +1000800/*
801 * vfsmount lock must be held for write
802 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800803static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500804{
805 if (ns && ns->event != event) {
806 ns->event = event;
807 wake_up_interruptible(&ns->poll);
808 }
809}
810
Nick Piggin99b7db72010-08-18 04:37:39 +1000811/*
812 * vfsmount lock must be held for write
813 */
Eric W. Biederman7bdb11d2014-12-29 13:03:41 -0600814static void unhash_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
Al Viro0714a532011-11-24 22:19:58 -0500816 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500817 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500818 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400819 hlist_del_init_rcu(&mnt->mnt_hash);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700820 hlist_del_init(&mnt->mnt_mp_list);
Al Viro84d17192013-03-15 10:53:28 -0400821 put_mountpoint(mnt->mnt_mp);
822 mnt->mnt_mp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823}
824
Nick Piggin99b7db72010-08-18 04:37:39 +1000825/*
826 * vfsmount lock must be held for write
827 */
Eric W. Biederman7bdb11d2014-12-29 13:03:41 -0600828static void detach_mnt(struct mount *mnt, struct path *old_path)
829{
830 old_path->dentry = mnt->mnt_mountpoint;
831 old_path->mnt = &mnt->mnt_parent->mnt;
832 unhash_mnt(mnt);
833}
834
835/*
836 * vfsmount lock must be held for write
837 */
Eric W. Biederman6a46c572015-01-15 22:58:33 -0600838static void umount_mnt(struct mount *mnt)
839{
840 /* old mountpoint will be dropped when we can do that */
841 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
842 unhash_mnt(mnt);
843}
844
845/*
846 * vfsmount lock must be held for write
847 */
Al Viro84d17192013-03-15 10:53:28 -0400848void mnt_set_mountpoint(struct mount *mnt,
849 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500850 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500851{
Al Viro84d17192013-03-15 10:53:28 -0400852 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500853 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro84d17192013-03-15 10:53:28 -0400854 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500855 child_mnt->mnt_parent = mnt;
Al Viro84d17192013-03-15 10:53:28 -0400856 child_mnt->mnt_mp = mp;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700857 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500858}
859
Eric W. Biederman1064f872017-01-20 18:28:35 +1300860static void __attach_mnt(struct mount *mnt, struct mount *parent)
861{
862 hlist_add_head_rcu(&mnt->mnt_hash,
863 m_hash(&parent->mnt, mnt->mnt_mountpoint));
864 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
865}
866
Nick Piggin99b7db72010-08-18 04:37:39 +1000867/*
868 * vfsmount lock must be held for write
869 */
Al Viro84d17192013-03-15 10:53:28 -0400870static void attach_mnt(struct mount *mnt,
871 struct mount *parent,
872 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873{
Al Viro84d17192013-03-15 10:53:28 -0400874 mnt_set_mountpoint(parent, mp, mnt);
Eric W. Biederman1064f872017-01-20 18:28:35 +1300875 __attach_mnt(mnt, parent);
Ram Paib90fa9a2005-11-07 17:19:50 -0500876}
877
Eric W. Biederman1064f872017-01-20 18:28:35 +1300878void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
Al Viro12a5b522014-08-10 03:44:55 -0400879{
Eric W. Biederman1064f872017-01-20 18:28:35 +1300880 struct mountpoint *old_mp = mnt->mnt_mp;
881 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
882 struct mount *old_parent = mnt->mnt_parent;
883
884 list_del_init(&mnt->mnt_child);
885 hlist_del_init(&mnt->mnt_mp_list);
886 hlist_del_init_rcu(&mnt->mnt_hash);
887
888 attach_mnt(mnt, parent, mp);
889
890 put_mountpoint(old_mp);
891
892 /*
893 * Safely avoid even the suggestion this code might sleep or
894 * lock the mount hash by taking advantage of the knowledge that
895 * mnt_change_mountpoint will not release the final reference
896 * to a mountpoint.
897 *
898 * During mounting, the mount passed in as the parent mount will
899 * continue to use the old mountpoint and during unmounting, the
900 * old mountpoint will continue to exist until namespace_unlock,
901 * which happens well after mnt_change_mountpoint.
902 */
903 spin_lock(&old_mountpoint->d_lock);
904 old_mountpoint->d_lockref.count--;
905 spin_unlock(&old_mountpoint->d_lock);
906
907 mnt_add_count(old_parent, -1);
Al Viro12a5b522014-08-10 03:44:55 -0400908}
909
Ram Paib90fa9a2005-11-07 17:19:50 -0500910/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000911 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500912 */
Eric W. Biederman1064f872017-01-20 18:28:35 +1300913static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500914{
Al Viro0714a532011-11-24 22:19:58 -0500915 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500916 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500917 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500918 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500919
Al Viro0714a532011-11-24 22:19:58 -0500920 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500921
Al Viro1a4eeaf2011-11-25 02:19:55 -0500922 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400923 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500924 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500925
Ram Paib90fa9a2005-11-07 17:19:50 -0500926 list_splice(&head, n->list.prev);
927
Eric W. Biedermand2921682016-09-28 00:27:17 -0500928 n->mounts += n->pending_mounts;
929 n->pending_mounts = 0;
930
Eric W. Biederman1064f872017-01-20 18:28:35 +1300931 __attach_mnt(mnt, parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800932 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933}
934
Al Viro909b0a82011-11-25 03:06:56 -0500935static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936{
Al Viro6b41d532011-11-24 23:24:33 -0500937 struct list_head *next = p->mnt_mounts.next;
938 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500940 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500942 next = p->mnt_child.next;
943 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 break;
Al Viro0714a532011-11-24 22:19:58 -0500945 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 }
947 }
Al Viro6b41d532011-11-24 23:24:33 -0500948 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949}
950
Al Viro315fc832011-11-24 18:57:30 -0500951static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500952{
Al Viro6b41d532011-11-24 23:24:33 -0500953 struct list_head *prev = p->mnt_mounts.prev;
954 while (prev != &p->mnt_mounts) {
955 p = list_entry(prev, struct mount, mnt_child);
956 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500957 }
958 return p;
959}
960
Al Viro9d412a42011-03-17 22:08:28 -0400961struct vfsmount *
962vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
963{
Al Virob105e272011-11-24 20:38:33 -0500964 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400965 struct dentry *root;
966
967 if (!type)
968 return ERR_PTR(-ENODEV);
969
970 mnt = alloc_vfsmnt(name);
971 if (!mnt)
972 return ERR_PTR(-ENOMEM);
973
974 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500975 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400976
977 root = mount_fs(type, flags, name, data);
978 if (IS_ERR(root)) {
David Howells8ffcb322014-01-24 12:17:54 +0000979 mnt_free_id(mnt);
Al Viro9d412a42011-03-17 22:08:28 -0400980 free_vfsmnt(mnt);
981 return ERR_CAST(root);
982 }
983
Al Virob105e272011-11-24 20:38:33 -0500984 mnt->mnt.mnt_root = root;
985 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500986 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500987 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400988 lock_mount_hash();
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100989 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400990 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -0500991 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400992}
993EXPORT_SYMBOL_GPL(vfs_kern_mount);
994
Eric W. Biederman93faccbb2017-02-01 06:06:16 +1300995struct vfsmount *
996vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
997 const char *name, void *data)
998{
999 /* Until it is worked out how to pass the user namespace
1000 * through from the parent mount to the submount don't support
1001 * unprivileged mounts with submounts.
1002 */
1003 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1004 return ERR_PTR(-EPERM);
1005
1006 return vfs_kern_mount(type, MS_SUBMOUNT, name, data);
1007}
1008EXPORT_SYMBOL_GPL(vfs_submount);
1009
Al Viro87129cc2011-11-24 21:24:27 -05001010static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -05001011 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012{
Al Viro87129cc2011-11-24 21:24:27 -05001013 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +01001014 struct mount *mnt;
1015 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
David Howellsbe34d1a2012-06-25 12:55:18 +01001017 mnt = alloc_vfsmnt(old->mnt_devname);
1018 if (!mnt)
1019 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001020
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001021 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +01001022 mnt->mnt_group_id = 0; /* not a peer of original */
1023 else
1024 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001025
David Howellsbe34d1a2012-06-25 12:55:18 +01001026 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1027 err = mnt_alloc_group_id(mnt);
1028 if (err)
1029 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001031
Al Virof2ebb3a2014-02-27 09:35:45 -05001032 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
Eric W. Biederman132c94e2013-03-22 04:08:05 -07001033 /* Don't allow unprivileged users to change mount flags */
Eric W. Biederman9566d672014-07-28 17:26:07 -07001034 if (flag & CL_UNPRIVILEGED) {
1035 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1036
1037 if (mnt->mnt.mnt_flags & MNT_READONLY)
1038 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1039
1040 if (mnt->mnt.mnt_flags & MNT_NODEV)
1041 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1042
1043 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1044 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1045
1046 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1047 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1048 }
Eric W. Biederman132c94e2013-03-22 04:08:05 -07001049
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001050 /* Don't allow unprivileged users to reveal what is under a mount */
Eric W. Biederman381cacb2014-10-07 17:11:46 -07001051 if ((flag & CL_UNPRIVILEGED) &&
1052 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001053 mnt->mnt.mnt_flags |= MNT_LOCKED;
1054
David Howellsbe34d1a2012-06-25 12:55:18 +01001055 atomic_inc(&sb->s_active);
1056 mnt->mnt.mnt_sb = sb;
1057 mnt->mnt.mnt_root = dget(root);
1058 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1059 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -04001060 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001061 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001062 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001063
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001064 if ((flag & CL_SLAVE) ||
1065 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +01001066 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1067 mnt->mnt_master = old;
1068 CLEAR_MNT_SHARED(mnt);
1069 } else if (!(flag & CL_PRIVATE)) {
1070 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1071 list_add(&mnt->mnt_share, &old->mnt_share);
1072 if (IS_MNT_SLAVE(old))
1073 list_add(&mnt->mnt_slave, &old->mnt_slave);
1074 mnt->mnt_master = old->mnt_master;
Al Viro5235d442016-11-20 19:33:09 -05001075 } else {
1076 CLEAR_MNT_SHARED(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001077 }
1078 if (flag & CL_MAKE_SHARED)
1079 set_mnt_shared(mnt);
1080
1081 /* stick the duplicate mount on the same expiry list
1082 * as the original if that was on one */
1083 if (flag & CL_EXPIRE) {
1084 if (!list_empty(&old->mnt_expire))
1085 list_add(&mnt->mnt_expire, &old->mnt_expire);
1086 }
1087
Al Virocb338d02011-11-24 20:55:08 -05001088 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001089
1090 out_free:
David Howells8ffcb322014-01-24 12:17:54 +00001091 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001092 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001093 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094}
1095
Al Viro9ea459e2014-08-08 13:08:20 -04001096static void cleanup_mnt(struct mount *mnt)
1097{
1098 /*
1099 * This probably indicates that somebody messed
1100 * up a mnt_want/drop_write() pair. If this
1101 * happens, the filesystem was probably unable
1102 * to make r/w->r/o transitions.
1103 */
1104 /*
1105 * The locking used to deal with mnt_count decrement provides barriers,
1106 * so mnt_get_writers() below is safe.
1107 */
1108 WARN_ON(mnt_get_writers(mnt));
1109 if (unlikely(mnt->mnt_pins.first))
1110 mnt_pin_kill(mnt);
1111 fsnotify_vfsmount_delete(&mnt->mnt);
1112 dput(mnt->mnt.mnt_root);
1113 deactivate_super(mnt->mnt.mnt_sb);
1114 mnt_free_id(mnt);
1115 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1116}
1117
1118static void __cleanup_mnt(struct rcu_head *head)
1119{
1120 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1121}
1122
1123static LLIST_HEAD(delayed_mntput_list);
1124static void delayed_mntput(struct work_struct *unused)
1125{
1126 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1127 struct llist_node *next;
1128
1129 for (; node; node = next) {
1130 next = llist_next(node);
1131 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1132 }
1133}
1134static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1135
Al Viro900148d2011-11-25 00:33:11 -05001136static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001137{
Al Viro48a066e2013-09-29 22:06:07 -04001138 rcu_read_lock();
1139 mnt_add_count(mnt, -1);
1140 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1141 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -05001142 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001143 }
Al Viro719ea2f2013-09-29 11:24:49 -04001144 lock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001145 if (mnt_get_count(mnt)) {
Al Viro48a066e2013-09-29 22:06:07 -04001146 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -04001147 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001148 return;
1149 }
Al Viro48a066e2013-09-29 22:06:07 -04001150 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1151 rcu_read_unlock();
1152 unlock_mount_hash();
1153 return;
1154 }
1155 mnt->mnt.mnt_flags |= MNT_DOOMED;
1156 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301157
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001158 list_del(&mnt->mnt_instance);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001159
1160 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1161 struct mount *p, *tmp;
1162 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1163 umount_mnt(p);
1164 }
1165 }
Al Viro719ea2f2013-09-29 11:24:49 -04001166 unlock_mount_hash();
Al Viro649a7952013-09-28 12:41:25 -04001167
Al Viro9ea459e2014-08-08 13:08:20 -04001168 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1169 struct task_struct *task = current;
1170 if (likely(!(task->flags & PF_KTHREAD))) {
1171 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1172 if (!task_work_add(task, &mnt->mnt_rcu, true))
1173 return;
1174 }
1175 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1176 schedule_delayed_work(&delayed_mntput_work, 1);
1177 return;
1178 }
1179 cleanup_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001180}
Nick Pigginb3e19d92011-01-07 17:50:11 +11001181
1182void mntput(struct vfsmount *mnt)
1183{
1184 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -05001185 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001186 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -05001187 if (unlikely(m->mnt_expiry_mark))
1188 m->mnt_expiry_mark = 0;
1189 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001190 }
1191}
1192EXPORT_SYMBOL(mntput);
1193
1194struct vfsmount *mntget(struct vfsmount *mnt)
1195{
1196 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001197 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001198 return mnt;
1199}
1200EXPORT_SYMBOL(mntget);
1201
Ian Kentc6609c02016-11-24 08:03:41 +11001202/* path_is_mountpoint() - Check if path is a mount in the current
1203 * namespace.
1204 *
1205 * d_mountpoint() can only be used reliably to establish if a dentry is
1206 * not mounted in any namespace and that common case is handled inline.
1207 * d_mountpoint() isn't aware of the possibility there may be multiple
1208 * mounts using a given dentry in a different namespace. This function
1209 * checks if the passed in path is a mountpoint rather than the dentry
1210 * alone.
1211 */
1212bool path_is_mountpoint(const struct path *path)
1213{
1214 unsigned seq;
1215 bool res;
1216
1217 if (!d_mountpoint(path->dentry))
1218 return false;
1219
1220 rcu_read_lock();
1221 do {
1222 seq = read_seqbegin(&mount_lock);
1223 res = __path_is_mountpoint(path);
1224 } while (read_seqretry(&mount_lock, seq));
1225 rcu_read_unlock();
1226
1227 return res;
1228}
1229EXPORT_SYMBOL(path_is_mountpoint);
1230
Al Viroca71cf72016-11-20 19:45:28 -05001231struct vfsmount *mnt_clone_internal(const struct path *path)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001232{
Al Viro3064c352014-08-07 09:12:31 -04001233 struct mount *p;
1234 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1235 if (IS_ERR(p))
1236 return ERR_CAST(p);
1237 p->mnt.mnt_flags |= MNT_INTERNAL;
1238 return &p->mnt;
Al Viro7b7b1ac2005-11-07 17:13:39 -05001239}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001241#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001242/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243static void *m_start(struct seq_file *m, loff_t *pos)
1244{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001245 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Ram Pai390c6842005-11-07 17:17:51 -05001247 down_read(&namespace_sem);
Al Viroc7999c32014-02-27 14:40:10 -05001248 if (p->cached_event == p->ns->event) {
1249 void *v = p->cached_mount;
1250 if (*pos == p->cached_index)
1251 return v;
1252 if (*pos == p->cached_index + 1) {
1253 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1254 return p->cached_mount = v;
1255 }
1256 }
1257
1258 p->cached_event = p->ns->event;
1259 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1260 p->cached_index = *pos;
1261 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262}
1263
1264static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1265{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001266 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001267
Al Viroc7999c32014-02-27 14:40:10 -05001268 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1269 p->cached_index = *pos;
1270 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271}
1272
1273static void m_stop(struct seq_file *m, void *v)
1274{
Ram Pai390c6842005-11-07 17:17:51 -05001275 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276}
1277
Al Viro0226f492011-12-06 12:21:54 -05001278static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001279{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001280 struct proc_mounts *p = m->private;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001281 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001282 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283}
1284
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001285const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 .start = m_start,
1287 .next = m_next,
1288 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001289 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001290};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001291#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293/**
1294 * may_umount_tree - check if a mount tree is busy
1295 * @mnt: root of mount tree
1296 *
1297 * This is called to check if a tree of mounts has any
1298 * open files, pwds, chroots or sub mounts that are
1299 * busy.
1300 */
Al Viro909b0a82011-11-25 03:06:56 -05001301int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302{
Al Viro909b0a82011-11-25 03:06:56 -05001303 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001304 int actual_refs = 0;
1305 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001306 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001307 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
Nick Pigginb3e19d92011-01-07 17:50:11 +11001309 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001310 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001311 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001312 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 }
Al Viro719ea2f2013-09-29 11:24:49 -04001315 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
1317 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001318 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
Ian Kente3474a82006-03-27 01:14:51 -08001320 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321}
1322
1323EXPORT_SYMBOL(may_umount_tree);
1324
1325/**
1326 * may_umount - check if a mount point is busy
1327 * @mnt: root of mount
1328 *
1329 * This is called to check if a mount point has any
1330 * open files, pwds, chroots or sub mounts. If the
1331 * mount has sub mounts this will return busy
1332 * regardless of whether the sub mounts are busy.
1333 *
1334 * Doesn't take quota and stuff into account. IOW, in some cases it will
1335 * give false negatives. The main reason why it's here is that we need
1336 * a non-destructive way to look for easily umountable filesystems.
1337 */
1338int may_umount(struct vfsmount *mnt)
1339{
Ian Kente3474a82006-03-27 01:14:51 -08001340 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001341 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001342 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001343 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001344 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001345 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001346 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001347 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348}
1349
1350EXPORT_SYMBOL(may_umount);
1351
Al Viro38129a12014-03-20 21:10:51 -04001352static HLIST_HEAD(unmounted); /* protected by namespace_sem */
Al Viroe3197d82013-03-16 14:35:16 -04001353
Al Viro97216be2013-03-16 15:12:40 -04001354static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355{
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001356 struct hlist_head head;
Al Viro97216be2013-03-16 15:12:40 -04001357
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001358 hlist_move_list(&unmounted, &head);
Al Viro97216be2013-03-16 15:12:40 -04001359
Al Viro97216be2013-03-16 15:12:40 -04001360 up_write(&namespace_sem);
1361
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001362 if (likely(hlist_empty(&head)))
1363 return;
1364
Al Viro48a066e2013-09-29 22:06:07 -04001365 synchronize_rcu();
1366
Al Viro87b95ce2015-01-10 19:01:08 -05001367 group_pin_kill(&head);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001368}
1369
Al Viro97216be2013-03-16 15:12:40 -04001370static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001371{
Al Viro97216be2013-03-16 15:12:40 -04001372 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001373}
1374
Eric W. Biedermane819f152014-12-24 07:20:01 -06001375enum umount_tree_flags {
1376 UMOUNT_SYNC = 1,
1377 UMOUNT_PROPAGATE = 2,
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001378 UMOUNT_CONNECTED = 4,
Eric W. Biedermane819f152014-12-24 07:20:01 -06001379};
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001380
1381static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1382{
1383 /* Leaving mounts connected is only valid for lazy umounts */
1384 if (how & UMOUNT_SYNC)
1385 return true;
1386
1387 /* A mount without a parent has nothing to be connected to */
1388 if (!mnt_has_parent(mnt))
1389 return true;
1390
1391 /* Because the reference counting rules change when mounts are
1392 * unmounted and connected, umounted mounts may not be
1393 * connected to mounted mounts.
1394 */
1395 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1396 return true;
1397
1398 /* Has it been requested that the mount remain connected? */
1399 if (how & UMOUNT_CONNECTED)
1400 return false;
1401
1402 /* Is the mount locked such that it needs to remain connected? */
1403 if (IS_MNT_LOCKED(mnt))
1404 return false;
1405
1406 /* By default disconnect the mount */
1407 return true;
1408}
1409
Nick Piggin99b7db72010-08-18 04:37:39 +10001410/*
Al Viro48a066e2013-09-29 22:06:07 -04001411 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001412 * namespace_sem must be held for write
1413 */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001414static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001415{
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001416 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001417 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001418
Eric W. Biederman5d884572015-01-03 05:39:35 -06001419 if (how & UMOUNT_PROPAGATE)
1420 propagate_mount_unlock(mnt);
1421
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001422 /* Gather the mounts to umount */
Eric W. Biederman590ce4b2014-12-22 18:30:08 -06001423 for (p = mnt; p; p = next_mnt(p, mnt)) {
1424 p->mnt.mnt_flags |= MNT_UMOUNT;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001425 list_move(&p->mnt_list, &tmp_list);
Eric W. Biederman590ce4b2014-12-22 18:30:08 -06001426 }
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001427
Eric W. Biederman411a9382014-12-22 19:12:07 -06001428 /* Hide the mounts from mnt_mounts */
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001429 list_for_each_entry(p, &tmp_list, mnt_list) {
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001430 list_del_init(&p->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -04001431 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001432
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001433 /* Add propogated mounts to the tmp_list */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001434 if (how & UMOUNT_PROPAGATE)
Al Viro7b8a53f2011-01-15 20:08:44 -05001435 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001436
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001437 while (!list_empty(&tmp_list)) {
Eric W. Biedermand2921682016-09-28 00:27:17 -05001438 struct mnt_namespace *ns;
Eric W. Biedermance07d892014-12-23 21:37:03 -06001439 bool disconnect;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001440 p = list_first_entry(&tmp_list, struct mount, mnt_list);
Al Viro6776db3d2011-11-25 00:22:05 -05001441 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001442 list_del_init(&p->mnt_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05001443 ns = p->mnt_ns;
1444 if (ns) {
1445 ns->mounts--;
1446 __touch_mnt_namespace(ns);
1447 }
Al Viro143c8c92011-11-25 00:46:35 -05001448 p->mnt_ns = NULL;
Eric W. Biedermane819f152014-12-24 07:20:01 -06001449 if (how & UMOUNT_SYNC)
Al Viro48a066e2013-09-29 22:06:07 -04001450 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro87b95ce2015-01-10 19:01:08 -05001451
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001452 disconnect = disconnect_mount(p, how);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001453
1454 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1455 disconnect ? &unmounted : NULL);
Al Viro676da582011-11-24 21:47:05 -05001456 if (mnt_has_parent(p)) {
Al Viro81b6b062014-08-30 18:32:05 -04001457 mnt_add_count(p->mnt_parent, -1);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001458 if (!disconnect) {
1459 /* Don't forget about p */
1460 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1461 } else {
1462 umount_mnt(p);
1463 }
Al Viro7c4b93d2008-03-21 23:59:49 -04001464 }
Al Viro0f0afb12011-11-24 20:43:10 -05001465 change_mnt_propagation(p, MS_PRIVATE);
Al Viro38129a12014-03-20 21:10:51 -04001466 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467}
1468
Al Virob54b9be2013-03-16 14:39:34 -04001469static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001470
Al Viro1ab59732011-11-24 21:35:16 -05001471static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
Al Viro1ab59732011-11-24 21:35:16 -05001473 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 int retval;
1475
Al Viro1ab59732011-11-24 21:35:16 -05001476 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 if (retval)
1478 return retval;
1479
1480 /*
1481 * Allow userspace to request a mountpoint be expired rather than
1482 * unmounting unconditionally. Unmount only happens if:
1483 * (1) the mark is already set (the mark is cleared by mntput())
1484 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1485 */
1486 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001487 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 flags & (MNT_FORCE | MNT_DETACH))
1489 return -EINVAL;
1490
Nick Pigginb3e19d92011-01-07 17:50:11 +11001491 /*
1492 * probably don't strictly need the lock here if we examined
1493 * all race cases, but it's a slowpath.
1494 */
Al Viro719ea2f2013-09-29 11:24:49 -04001495 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001496 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001497 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001499 }
Al Viro719ea2f2013-09-29 11:24:49 -04001500 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
Al Viro863d6842011-11-25 00:57:42 -05001502 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 return -EAGAIN;
1504 }
1505
1506 /*
1507 * If we may have to abort operations to get out of this
1508 * mount, and they will themselves hold resources we must
1509 * allow the fs to do things. In the Unix tradition of
1510 * 'Gee thats tricky lets do it in userspace' the umount_begin
1511 * might fail to complete on the first run through as other tasks
1512 * must return, and the like. Thats for the mount program to worry
1513 * about for the moment.
1514 */
1515
Al Viro42faad92008-04-24 07:21:56 -04001516 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001517 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
1520 /*
1521 * No sense to grab the lock for this test, but test itself looks
1522 * somewhat bogus. Suggestions for better replacement?
1523 * Ho-hum... In principle, we might treat that as umount + switch
1524 * to rootfs. GC would eventually take care of the old vfsmount.
1525 * Actually it makes sense, especially if rootfs would contain a
1526 * /reboot - static binary that would close all descriptors and
1527 * call reboot(9). Then init(8) could umount root and exec /reboot.
1528 */
Al Viro1ab59732011-11-24 21:35:16 -05001529 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 /*
1531 * Special case for "unmounting" root ...
1532 * we just try to remount it readonly.
1533 */
Andy Lutomirskia1480dc2014-10-08 12:32:47 -07001534 if (!capable(CAP_SYS_ADMIN))
1535 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001537 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 up_write(&sb->s_umount);
1540 return retval;
1541 }
1542
Al Viro97216be2013-03-16 15:12:40 -04001543 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001544 lock_mount_hash();
Al Viro5addc5d2005-11-07 17:15:49 -05001545 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
Al Viro48a066e2013-09-29 22:06:07 -04001547 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001548 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001549 umount_tree(mnt, UMOUNT_PROPAGATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001551 } else {
1552 shrink_submounts(mnt);
1553 retval = -EBUSY;
1554 if (!propagate_mount_busy(mnt, 2)) {
1555 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001556 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Viro48a066e2013-09-29 22:06:07 -04001557 retval = 0;
1558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 }
Al Viro719ea2f2013-09-29 11:24:49 -04001560 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001561 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 return retval;
1563}
1564
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001565/*
1566 * __detach_mounts - lazily unmount all mounts on the specified dentry
1567 *
1568 * During unlink, rmdir, and d_drop it is possible to loose the path
1569 * to an existing mountpoint, and wind up leaking the mount.
1570 * detach_mounts allows lazily unmounting those mounts instead of
1571 * leaking them.
1572 *
1573 * The caller may hold dentry->d_inode->i_mutex.
1574 */
1575void __detach_mounts(struct dentry *dentry)
1576{
1577 struct mountpoint *mp;
1578 struct mount *mnt;
1579
1580 namespace_lock();
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13001581 lock_mount_hash();
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001582 mp = lookup_mountpoint(dentry);
Eric W. Biedermanf53e5792015-01-19 11:48:45 -06001583 if (IS_ERR_OR_NULL(mp))
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001584 goto out_unlock;
1585
Andrey Ulanove06b9332016-04-15 14:24:41 -07001586 event++;
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001587 while (!hlist_empty(&mp->m_list)) {
1588 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001589 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
Eric W. Biedermanfe78fcc2015-07-17 14:54:27 -05001590 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1591 umount_mnt(mnt);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001592 }
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001593 else umount_tree(mnt, UMOUNT_CONNECTED);
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001594 }
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001595 put_mountpoint(mp);
1596out_unlock:
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13001597 unlock_mount_hash();
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001598 namespace_unlock();
1599}
1600
David Howellsdd111b32017-07-04 17:25:09 +01001601/*
Al Viro9b40bc92013-02-22 22:45:42 -05001602 * Is the caller allowed to modify his namespace?
1603 */
1604static inline bool may_mount(void)
1605{
1606 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1607}
1608
Jeff Layton9e8925b2015-11-16 09:49:34 -05001609static inline bool may_mandlock(void)
1610{
1611#ifndef CONFIG_MANDATORY_FILE_LOCKING
1612 return false;
1613#endif
Eric W. Biederman95ace752015-11-11 17:22:33 -06001614 return capable(CAP_SYS_ADMIN);
Jeff Layton9e8925b2015-11-16 09:49:34 -05001615}
1616
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617/*
1618 * Now umount can handle mount points as well as block devices.
1619 * This is important for filesystems which use unnamed block devices.
1620 *
1621 * We now support a flag for forced unmount like the other 'big iron'
1622 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1623 */
1624
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001625SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
Al Viro2d8f3032008-07-22 09:59:21 -04001627 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001628 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001630 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001632 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1633 return -EINVAL;
1634
Al Viro9b40bc92013-02-22 22:45:42 -05001635 if (!may_mount())
1636 return -EPERM;
1637
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001638 if (!(flags & UMOUNT_NOFOLLOW))
1639 lookup_flags |= LOOKUP_FOLLOW;
1640
Al Viro197df042013-09-08 14:03:27 -04001641 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 if (retval)
1643 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001644 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001646 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001648 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 goto dput_and_out;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001650 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1651 goto dput_and_out;
Eric W. Biedermanb2f5d4d2014-10-04 14:44:03 -07001652 retval = -EPERM;
1653 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1654 goto dput_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Al Viro900148d2011-11-25 00:33:11 -05001656 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001658 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001659 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001660 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661out:
1662 return retval;
1663}
1664
1665#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1666
1667/*
Ram Paib58fed82005-11-07 17:16:09 -05001668 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001670SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
Ram Paib58fed82005-11-07 17:16:09 -05001672 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673}
1674
1675#endif
1676
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001677static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001678{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001679 /* Is this a proxy for a mount namespace? */
Al Viroe149ed22014-11-01 10:57:28 -04001680 return dentry->d_op == &ns_dentry_operations &&
1681 dentry->d_fsdata == &mntns_operations;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001682}
1683
Al Viro58be28252014-11-01 00:00:23 -04001684struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1685{
1686 return container_of(ns, struct mnt_namespace, ns);
1687}
1688
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001689static bool mnt_ns_loop(struct dentry *dentry)
1690{
1691 /* Could bind mounting the mount namespace inode cause a
1692 * mount namespace loop?
1693 */
1694 struct mnt_namespace *mnt_ns;
1695 if (!is_mnt_ns_file(dentry))
1696 return false;
1697
Al Virof77c8012014-11-01 03:13:17 -04001698 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
Eric W. Biederman8823c072010-03-07 18:49:36 -08001699 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1700}
1701
Al Viro87129cc2011-11-24 21:24:27 -05001702struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001703 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704{
Al Viro84d17192013-03-15 10:53:28 -04001705 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001707 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1708 return ERR_PTR(-EINVAL);
1709
1710 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001711 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001712
Ram Pai36341f62005-11-07 17:17:22 -05001713 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001714 if (IS_ERR(q))
1715 return q;
1716
Al Viroa73324d2011-11-24 22:25:07 -05001717 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
1719 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001720 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001721 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001722 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 continue;
1724
Al Viro909b0a82011-11-25 03:06:56 -05001725 for (s = r; s; s = next_mnt(s, r)) {
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001726 if (!(flag & CL_COPY_UNBINDABLE) &&
1727 IS_MNT_UNBINDABLE(s)) {
1728 s = skip_mnt_tree(s);
1729 continue;
1730 }
1731 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1732 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001733 s = skip_mnt_tree(s);
1734 continue;
1735 }
Al Viro0714a532011-11-24 22:19:58 -05001736 while (p != s->mnt_parent) {
1737 p = p->mnt_parent;
1738 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 }
Al Viro87129cc2011-11-24 21:24:27 -05001740 p = s;
Al Viro84d17192013-03-15 10:53:28 -04001741 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001742 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001743 if (IS_ERR(q))
1744 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04001745 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05001746 list_add_tail(&q->mnt_list, &res->mnt_list);
Eric W. Biederman1064f872017-01-20 18:28:35 +13001747 attach_mnt(q, parent, p->mnt_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04001748 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 }
1750 }
1751 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001752out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04001754 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001755 umount_tree(res, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001756 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001758 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759}
1760
David Howellsbe34d1a2012-06-25 12:55:18 +01001761/* Caller should check returned pointer for errors */
1762
Al Viroca71cf72016-11-20 19:45:28 -05001763struct vfsmount *collect_mounts(const struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001764{
Al Virocb338d02011-11-24 20:55:08 -05001765 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001766 namespace_lock();
Eric W. Biedermancd4a4012015-01-07 14:28:26 -06001767 if (!check_mnt(real_mount(path->mnt)))
1768 tree = ERR_PTR(-EINVAL);
1769 else
1770 tree = copy_tree(real_mount(path->mnt), path->dentry,
1771 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001772 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001773 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001774 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001775 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001776}
1777
1778void drop_collected_mounts(struct vfsmount *mnt)
1779{
Al Viro97216be2013-03-16 15:12:40 -04001780 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001781 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001782 umount_tree(real_mount(mnt), UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001783 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04001784 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001785}
1786
Miklos Szeredic771d682014-10-24 00:14:36 +02001787/**
1788 * clone_private_mount - create a private clone of a path
1789 *
1790 * This creates a new vfsmount, which will be the clone of @path. The new will
1791 * not be attached anywhere in the namespace and will be private (i.e. changes
1792 * to the originating mount won't be propagated into this).
1793 *
1794 * Release with mntput().
1795 */
Al Viroca71cf72016-11-20 19:45:28 -05001796struct vfsmount *clone_private_mount(const struct path *path)
Miklos Szeredic771d682014-10-24 00:14:36 +02001797{
1798 struct mount *old_mnt = real_mount(path->mnt);
1799 struct mount *new_mnt;
1800
1801 if (IS_MNT_UNBINDABLE(old_mnt))
1802 return ERR_PTR(-EINVAL);
1803
Miklos Szeredic771d682014-10-24 00:14:36 +02001804 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
Miklos Szeredic771d682014-10-24 00:14:36 +02001805 if (IS_ERR(new_mnt))
1806 return ERR_CAST(new_mnt);
1807
1808 return &new_mnt->mnt;
1809}
1810EXPORT_SYMBOL_GPL(clone_private_mount);
1811
Al Viro1f707132010-01-30 22:51:25 -05001812int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1813 struct vfsmount *root)
1814{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001815 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001816 int res = f(root, arg);
1817 if (res)
1818 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001819 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1820 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001821 if (res)
1822 return res;
1823 }
1824 return 0;
1825}
1826
Al Viro4b8b21f2011-11-24 19:54:23 -05001827static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001828{
Al Viro315fc832011-11-24 18:57:30 -05001829 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001830
Al Viro909b0a82011-11-25 03:06:56 -05001831 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001832 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001833 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001834 }
1835}
1836
Al Viro4b8b21f2011-11-24 19:54:23 -05001837static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001838{
Al Viro315fc832011-11-24 18:57:30 -05001839 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001840
Al Viro909b0a82011-11-25 03:06:56 -05001841 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001842 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001843 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001844 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001845 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001846 return err;
1847 }
1848 }
1849 }
1850
1851 return 0;
1852}
1853
Eric W. Biedermand2921682016-09-28 00:27:17 -05001854int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1855{
1856 unsigned int max = READ_ONCE(sysctl_mount_max);
1857 unsigned int mounts = 0, old, pending, sum;
1858 struct mount *p;
1859
1860 for (p = mnt; p; p = next_mnt(p, mnt))
1861 mounts++;
1862
1863 old = ns->mounts;
1864 pending = ns->pending_mounts;
1865 sum = old + pending;
1866 if ((old > sum) ||
1867 (pending > sum) ||
1868 (max < sum) ||
1869 (mounts > (max - sum)))
1870 return -ENOSPC;
1871
1872 ns->pending_mounts = pending + mounts;
1873 return 0;
1874}
1875
Ram Paib90fa9a2005-11-07 17:19:50 -05001876/*
1877 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001878 * @nd : place the mount tree @source_mnt is attached
1879 * @parent_nd : if non-null, detach the source_mnt from its parent and
1880 * store the parent mount and mountpoint dentry.
1881 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001882 *
1883 * NOTE: in the table below explains the semantics when a source mount
1884 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001885 * ---------------------------------------------------------------------------
1886 * | BIND MOUNT OPERATION |
1887 * |**************************************************************************
1888 * | source-->| shared | private | slave | unbindable |
1889 * | dest | | | | |
1890 * | | | | | | |
1891 * | v | | | | |
1892 * |**************************************************************************
1893 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1894 * | | | | | |
1895 * |non-shared| shared (+) | private | slave (*) | invalid |
1896 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001897 * A bind operation clones the source mount and mounts the clone on the
1898 * destination mount.
1899 *
1900 * (++) the cloned mount is propagated to all the mounts in the propagation
1901 * tree of the destination mount and the cloned mount is added to
1902 * the peer group of the source mount.
1903 * (+) the cloned mount is created under the destination mount and is marked
1904 * as shared. The cloned mount is added to the peer group of the source
1905 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001906 * (+++) the mount is propagated to all the mounts in the propagation tree
1907 * of the destination mount and the cloned mount is made slave
1908 * of the same master as that of the source mount. The cloned mount
1909 * is marked as 'shared and slave'.
1910 * (*) the cloned mount is made a slave of the same master as that of the
1911 * source mount.
1912 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001913 * ---------------------------------------------------------------------------
1914 * | MOVE MOUNT OPERATION |
1915 * |**************************************************************************
1916 * | source-->| shared | private | slave | unbindable |
1917 * | dest | | | | |
1918 * | | | | | | |
1919 * | v | | | | |
1920 * |**************************************************************************
1921 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1922 * | | | | | |
1923 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1924 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001925 *
1926 * (+) the mount is moved to the destination. And is then propagated to
1927 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001928 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001929 * (+++) the mount is moved to the destination and is then propagated to
1930 * all the mounts belonging to the destination mount's propagation tree.
1931 * the mount is marked as 'shared and slave'.
1932 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001933 *
1934 * if the source mount is a tree, the operations explained above is
1935 * applied to each mount in the tree.
1936 * Must be called without spinlocks held, since this function can sleep
1937 * in allocations.
1938 */
Al Viro0fb54e52011-11-24 19:59:16 -05001939static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d17192013-03-15 10:53:28 -04001940 struct mount *dest_mnt,
1941 struct mountpoint *dest_mp,
1942 struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001943{
Al Viro38129a12014-03-20 21:10:51 -04001944 HLIST_HEAD(tree_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05001945 struct mnt_namespace *ns = dest_mnt->mnt_ns;
Eric W. Biederman1064f872017-01-20 18:28:35 +13001946 struct mountpoint *smp;
Al Viro315fc832011-11-24 18:57:30 -05001947 struct mount *child, *p;
Al Viro38129a12014-03-20 21:10:51 -04001948 struct hlist_node *n;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001949 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001950
Eric W. Biederman1064f872017-01-20 18:28:35 +13001951 /* Preallocate a mountpoint in case the new mounts need
1952 * to be tucked under other mounts.
1953 */
1954 smp = get_mountpoint(source_mnt->mnt.mnt_root);
1955 if (IS_ERR(smp))
1956 return PTR_ERR(smp);
1957
Eric W. Biedermand2921682016-09-28 00:27:17 -05001958 /* Is there space to add these mounts to the mount namespace? */
1959 if (!parent_path) {
1960 err = count_mounts(ns, source_mnt);
1961 if (err)
1962 goto out;
1963 }
1964
Al Virofc7be132011-11-25 01:05:37 -05001965 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001966 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001967 if (err)
1968 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04001969 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Al Virof2ebb3a2014-02-27 09:35:45 -05001970 lock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001971 if (err)
1972 goto out_cleanup_ids;
Al Viro909b0a82011-11-25 03:06:56 -05001973 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001974 set_mnt_shared(p);
Al Viro0b1b9012014-03-21 10:14:08 -04001975 } else {
1976 lock_mount_hash();
Ram Paib90fa9a2005-11-07 17:19:50 -05001977 }
Al Viro1a390682008-03-21 20:48:19 -04001978 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001979 detach_mnt(source_mnt, parent_path);
Al Viro84d17192013-03-15 10:53:28 -04001980 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05001981 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001982 } else {
Al Viro84d17192013-03-15 10:53:28 -04001983 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Eric W. Biederman1064f872017-01-20 18:28:35 +13001984 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05001985 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001986
Al Viro38129a12014-03-20 21:10:51 -04001987 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04001988 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04001989 hlist_del_init(&child->mnt_hash);
Eric W. Biederman1064f872017-01-20 18:28:35 +13001990 q = __lookup_mnt(&child->mnt_parent->mnt,
1991 child->mnt_mountpoint);
1992 if (q)
1993 mnt_change_mountpoint(child, smp, q);
1994 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05001995 }
Eric W. Biederman1064f872017-01-20 18:28:35 +13001996 put_mountpoint(smp);
Al Viro719ea2f2013-09-29 11:24:49 -04001997 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10001998
Ram Paib90fa9a2005-11-07 17:19:50 -05001999 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002000
2001 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05002002 while (!hlist_empty(&tree_list)) {
2003 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
Eric W. Biedermand2921682016-09-28 00:27:17 -05002004 child->mnt_parent->mnt_ns->pending_mounts = 0;
Eric W. Biedermane819f152014-12-24 07:20:01 -06002005 umount_tree(child, UMOUNT_SYNC);
Al Virof2ebb3a2014-02-27 09:35:45 -05002006 }
2007 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04002008 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002009 out:
Eric W. Biedermand2921682016-09-28 00:27:17 -05002010 ns->pending_mounts = 0;
Eric W. Biederman1064f872017-01-20 18:28:35 +13002011
2012 read_seqlock_excl(&mount_lock);
2013 put_mountpoint(smp);
2014 read_sequnlock_excl(&mount_lock);
2015
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002016 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05002017}
2018
Al Viro84d17192013-03-15 10:53:28 -04002019static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04002020{
2021 struct vfsmount *mnt;
Al Viro84d17192013-03-15 10:53:28 -04002022 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04002023retry:
Al Viro59551022016-01-22 15:40:57 -05002024 inode_lock(dentry->d_inode);
Al Viro84d17192013-03-15 10:53:28 -04002025 if (unlikely(cant_mount(dentry))) {
Al Viro59551022016-01-22 15:40:57 -05002026 inode_unlock(dentry->d_inode);
Al Viro84d17192013-03-15 10:53:28 -04002027 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04002028 }
Al Viro97216be2013-03-16 15:12:40 -04002029 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04002030 mnt = lookup_mnt(path);
Al Viro84d17192013-03-15 10:53:28 -04002031 if (likely(!mnt)) {
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002032 struct mountpoint *mp = get_mountpoint(dentry);
Al Viro84d17192013-03-15 10:53:28 -04002033 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04002034 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002035 inode_unlock(dentry->d_inode);
Al Viro84d17192013-03-15 10:53:28 -04002036 return mp;
2037 }
2038 return mp;
2039 }
Al Viro97216be2013-03-16 15:12:40 -04002040 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002041 inode_unlock(path->dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002042 path_put(path);
2043 path->mnt = mnt;
Al Viro84d17192013-03-15 10:53:28 -04002044 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04002045 goto retry;
2046}
2047
Al Viro84d17192013-03-15 10:53:28 -04002048static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04002049{
Al Viro84d17192013-03-15 10:53:28 -04002050 struct dentry *dentry = where->m_dentry;
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002051
2052 read_seqlock_excl(&mount_lock);
Al Viro84d17192013-03-15 10:53:28 -04002053 put_mountpoint(where);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002054 read_sequnlock_excl(&mount_lock);
2055
Al Viro328e6d92013-03-16 14:49:45 -04002056 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002057 inode_unlock(dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002058}
2059
Al Viro84d17192013-03-15 10:53:28 -04002060static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061{
Al Viro95bc5f22011-11-25 00:30:56 -05002062 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 return -EINVAL;
2064
David Howellse36cb0b2015-01-29 12:02:35 +00002065 if (d_is_dir(mp->m_dentry) !=
2066 d_is_dir(mnt->mnt.mnt_root))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 return -ENOTDIR;
2068
Al Viro84d17192013-03-15 10:53:28 -04002069 return attach_recursive_mnt(mnt, p, mp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070}
2071
2072/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002073 * Sanity check the flags to change_mnt_propagation.
2074 */
2075
2076static int flags_to_propagation_type(int flags)
2077{
Roman Borisov7c6e9842011-05-25 16:26:48 -07002078 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002079
2080 /* Fail if any non-propagation flags are set */
2081 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2082 return 0;
2083 /* Only one propagation flag should be set */
2084 if (!is_power_of_2(type))
2085 return 0;
2086 return type;
2087}
2088
2089/*
Ram Pai07b20882005-11-07 17:19:07 -05002090 * recursively change the type of the mountpoint.
2091 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002092static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05002093{
Al Viro315fc832011-11-24 18:57:30 -05002094 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05002095 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05002096 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002097 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002098 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05002099
Al Viro2d92ab32008-08-02 00:51:11 -04002100 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05002101 return -EINVAL;
2102
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002103 type = flags_to_propagation_type(flag);
2104 if (!type)
2105 return -EINVAL;
2106
Al Viro97216be2013-03-16 15:12:40 -04002107 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002108 if (type == MS_SHARED) {
2109 err = invent_group_ids(mnt, recurse);
2110 if (err)
2111 goto out_unlock;
2112 }
2113
Al Viro719ea2f2013-09-29 11:24:49 -04002114 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05002115 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05002116 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04002117 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002118
2119 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04002120 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002121 return err;
Ram Pai07b20882005-11-07 17:19:07 -05002122}
2123
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002124static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2125{
2126 struct mount *child;
2127 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2128 if (!is_subdir(child->mnt_mountpoint, dentry))
2129 continue;
2130
2131 if (child->mnt.mnt_flags & MNT_LOCKED)
2132 return true;
2133 }
2134 return false;
2135}
2136
Ram Pai07b20882005-11-07 17:19:07 -05002137/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 * do loopback mount.
2139 */
Al Viro808d4e32012-10-11 11:42:01 -04002140static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08002141 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142{
Al Viro2d92ab32008-08-02 00:51:11 -04002143 struct path old_path;
Al Viro84d17192013-03-15 10:53:28 -04002144 struct mount *mnt = NULL, *old, *parent;
2145 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002146 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 if (!old_name || !*old_name)
2148 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04002149 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 if (err)
2151 return err;
2152
Eric W. Biederman8823c072010-03-07 18:49:36 -08002153 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002154 if (mnt_ns_loop(old_path.dentry))
David Howellsdd111b32017-07-04 17:25:09 +01002155 goto out;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002156
Al Viro84d17192013-03-15 10:53:28 -04002157 mp = lock_mount(path);
2158 err = PTR_ERR(mp);
2159 if (IS_ERR(mp))
Jan Blunck4ac91372008-02-14 19:34:32 -08002160 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05002161
Al Viro87129cc2011-11-24 21:24:27 -05002162 old = real_mount(old_path.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002163 parent = real_mount(path->mnt);
Al Viro87129cc2011-11-24 21:24:27 -05002164
Al Virob12cea92011-03-18 08:55:38 -04002165 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002166 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04002167 goto out2;
2168
Al Viroe149ed22014-11-01 10:57:28 -04002169 if (!check_mnt(parent))
2170 goto out2;
2171
2172 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
Al Virob12cea92011-03-18 08:55:38 -04002173 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002175 if (!recurse && has_locked_children(old, old_path.dentry))
2176 goto out2;
2177
Al Viroccd48bc2005-11-07 17:15:04 -05002178 if (recurse)
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002179 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
Al Viroccd48bc2005-11-07 17:15:04 -05002180 else
Al Viro87129cc2011-11-24 21:24:27 -05002181 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05002182
David Howellsbe34d1a2012-06-25 12:55:18 +01002183 if (IS_ERR(mnt)) {
2184 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04002185 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01002186 }
Al Viroccd48bc2005-11-07 17:15:04 -05002187
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002188 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2189
Al Viro84d17192013-03-15 10:53:28 -04002190 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002191 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002192 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06002193 umount_tree(mnt, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04002194 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05002195 }
Al Virob12cea92011-03-18 08:55:38 -04002196out2:
Al Viro84d17192013-03-15 10:53:28 -04002197 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002198out:
Al Viro2d92ab32008-08-02 00:51:11 -04002199 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 return err;
2201}
2202
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002203static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2204{
2205 int error = 0;
2206 int readonly_request = 0;
2207
2208 if (ms_flags & MS_RDONLY)
2209 readonly_request = 1;
2210 if (readonly_request == __mnt_is_readonly(mnt))
2211 return 0;
2212
2213 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05002214 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002215 else
Al Viro83adc752011-11-24 22:37:54 -05002216 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002217 return error;
2218}
2219
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220/*
2221 * change filesystem flags. dir should be a physical root of filesystem.
2222 * If you've mounted a non-root directory somewhere and want to do remount
2223 * on it - tough luck.
2224 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002225static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 void *data)
2227{
2228 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04002229 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05002230 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231
Al Viro143c8c92011-11-25 00:46:35 -05002232 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 return -EINVAL;
2234
Al Viro2d92ab32008-08-02 00:51:11 -04002235 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 return -EINVAL;
2237
Eric W. Biederman07b64552014-07-28 17:10:56 -07002238 /* Don't allow changing of locked mnt flags.
2239 *
2240 * No locks need to be held here while testing the various
2241 * MNT_LOCK flags because those flags can never be cleared
2242 * once they are set.
2243 */
2244 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2245 !(mnt_flags & MNT_READONLY)) {
2246 return -EPERM;
2247 }
Eric W. Biederman9566d672014-07-28 17:26:07 -07002248 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2249 !(mnt_flags & MNT_NODEV)) {
Eric W. Biederman67690f92016-05-18 13:50:06 -05002250 return -EPERM;
Eric W. Biederman9566d672014-07-28 17:26:07 -07002251 }
2252 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2253 !(mnt_flags & MNT_NOSUID)) {
2254 return -EPERM;
2255 }
2256 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2257 !(mnt_flags & MNT_NOEXEC)) {
2258 return -EPERM;
2259 }
2260 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2261 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2262 return -EPERM;
2263 }
2264
Eric Parisff36fe22011-03-03 16:09:14 -05002265 err = security_sb_remount(sb, data);
2266 if (err)
2267 return err;
2268
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002270 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002271 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05002272 else if (!capable(CAP_SYS_ADMIN))
2273 err = -EPERM;
Al Viro4aa98cf2009-05-08 13:36:58 -04002274 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002275 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05002276 if (!err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002277 lock_mount_hash();
Eric W. Biedermana6138db2014-07-28 16:26:53 -07002278 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
Al Viro143c8c92011-11-25 00:46:35 -05002279 mnt->mnt.mnt_flags = mnt_flags;
Al Viro143c8c92011-11-25 00:46:35 -05002280 touch_mnt_namespace(mnt->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04002281 unlock_mount_hash();
Dan Williams0e55a7c2008-09-26 19:01:20 -07002282 }
Al Viro6339dab2013-09-16 22:41:01 -04002283 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 return err;
2285}
2286
Al Virocbbe3622011-11-24 20:01:19 -05002287static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05002288{
Al Viro315fc832011-11-24 18:57:30 -05002289 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05002290 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002291 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05002292 return 1;
2293 }
2294 return 0;
2295}
2296
Al Viro808d4e32012-10-11 11:42:01 -04002297static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298{
Al Viro2d92ab32008-08-02 00:51:11 -04002299 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05002300 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05002301 struct mount *old;
Al Viro84d17192013-03-15 10:53:28 -04002302 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002303 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 if (!old_name || !*old_name)
2305 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002306 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 if (err)
2308 return err;
2309
Al Viro84d17192013-03-15 10:53:28 -04002310 mp = lock_mount(path);
2311 err = PTR_ERR(mp);
2312 if (IS_ERR(mp))
David Howellscc53ce52011-01-14 18:45:26 +00002313 goto out;
2314
Al Viro143c8c92011-11-25 00:46:35 -05002315 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002316 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05002317
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002319 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 goto out1;
2321
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002322 if (old->mnt.mnt_flags & MNT_LOCKED)
2323 goto out1;
2324
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002326 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05002327 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328
Al Viro676da582011-11-24 21:47:05 -05002329 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05002330 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
David Howellse36cb0b2015-01-29 12:02:35 +00002332 if (d_is_dir(path->dentry) !=
2333 d_is_dir(old_path.dentry))
Ram Pai21444402005-11-07 17:20:03 -05002334 goto out1;
2335 /*
2336 * Don't move a mount residing in a shared parent.
2337 */
Al Virofc7be132011-11-25 01:05:37 -05002338 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002339 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05002340 /*
2341 * Don't move a mount tree containing unbindable mounts to a destination
2342 * mount which is shared.
2343 */
Al Virofc7be132011-11-25 01:05:37 -05002344 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05002345 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05002347 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05002348 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05002349 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350
Al Viro84d17192013-03-15 10:53:28 -04002351 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08002352 if (err)
Ram Pai21444402005-11-07 17:20:03 -05002353 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354
2355 /* if the mount is moved, it should no longer be expire
2356 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05002357 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358out1:
Al Viro84d17192013-03-15 10:53:28 -04002359 unlock_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04002362 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04002363 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 return err;
2365}
2366
Al Viro9d412a42011-03-17 22:08:28 -04002367static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2368{
2369 int err;
2370 const char *subtype = strchr(fstype, '.');
2371 if (subtype) {
2372 subtype++;
2373 err = -EINVAL;
2374 if (!subtype[0])
2375 goto err;
2376 } else
2377 subtype = "";
2378
2379 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2380 err = -ENOMEM;
2381 if (!mnt->mnt_sb->s_subtype)
2382 goto err;
2383 return mnt;
2384
2385 err:
2386 mntput(mnt);
2387 return ERR_PTR(err);
2388}
2389
Al Viro9d412a42011-03-17 22:08:28 -04002390/*
2391 * add a mount into a namespace's mount tree
2392 */
Al Viro95bc5f22011-11-25 00:30:56 -05002393static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04002394{
Al Viro84d17192013-03-15 10:53:28 -04002395 struct mountpoint *mp;
2396 struct mount *parent;
Al Viro9d412a42011-03-17 22:08:28 -04002397 int err;
2398
Al Virof2ebb3a2014-02-27 09:35:45 -05002399 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04002400
Al Viro84d17192013-03-15 10:53:28 -04002401 mp = lock_mount(path);
2402 if (IS_ERR(mp))
2403 return PTR_ERR(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002404
Al Viro84d17192013-03-15 10:53:28 -04002405 parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04002406 err = -EINVAL;
Al Viro84d17192013-03-15 10:53:28 -04002407 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04002408 /* that's acceptable only for automounts done in private ns */
2409 if (!(mnt_flags & MNT_SHRINKABLE))
2410 goto unlock;
2411 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d17192013-03-15 10:53:28 -04002412 if (!parent->mnt_ns)
Al Viro156cacb2012-09-21 08:19:02 -04002413 goto unlock;
2414 }
Al Viro9d412a42011-03-17 22:08:28 -04002415
2416 /* Refuse the same filesystem on the same mount point */
2417 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05002418 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04002419 path->mnt->mnt_root == path->dentry)
2420 goto unlock;
2421
2422 err = -EINVAL;
David Howellse36cb0b2015-01-29 12:02:35 +00002423 if (d_is_symlink(newmnt->mnt.mnt_root))
Al Viro9d412a42011-03-17 22:08:28 -04002424 goto unlock;
2425
Al Viro95bc5f22011-11-25 00:30:56 -05002426 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro84d17192013-03-15 10:53:28 -04002427 err = graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04002428
2429unlock:
Al Viro84d17192013-03-15 10:53:28 -04002430 unlock_mount(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002431 return err;
2432}
Al Virob1e75df2011-01-17 01:47:59 -05002433
Eric W. Biederman8654df42016-06-09 16:06:06 -05002434static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05002435
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436/*
2437 * create a new mount for userspace and request it to be added into the
2438 * namespace's tree
2439 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002440static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04002441 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002443 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002445 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002447 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 return -EINVAL;
2449
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002450 type = get_fs_type(fstype);
2451 if (!type)
2452 return -ENODEV;
2453
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002454 mnt = vfs_kern_mount(type, flags, name, data);
2455 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2456 !mnt->mnt_sb->s_subtype)
2457 mnt = fs_set_subtype(mnt, fstype);
2458
2459 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 if (IS_ERR(mnt))
2461 return PTR_ERR(mnt);
2462
Eric W. Biederman8654df42016-06-09 16:06:06 -05002463 if (mount_too_revealing(mnt, &mnt_flags)) {
2464 mntput(mnt);
2465 return -EPERM;
2466 }
2467
Al Viro95bc5f22011-11-25 00:30:56 -05002468 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002469 if (err)
2470 mntput(mnt);
2471 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472}
2473
Al Viro19a167a2011-01-17 01:35:23 -05002474int finish_automount(struct vfsmount *m, struct path *path)
2475{
Al Viro6776db3d2011-11-25 00:22:05 -05002476 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002477 int err;
2478 /* The new mount record should have at least 2 refs to prevent it being
2479 * expired before we get a chance to add it
2480 */
Al Viro6776db3d2011-11-25 00:22:05 -05002481 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002482
2483 if (m->mnt_sb == path->mnt->mnt_sb &&
2484 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002485 err = -ELOOP;
2486 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002487 }
2488
Al Viro95bc5f22011-11-25 00:30:56 -05002489 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002490 if (!err)
2491 return 0;
2492fail:
2493 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002494 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002495 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05002496 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04002497 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002498 }
Al Virob1e75df2011-01-17 01:47:59 -05002499 mntput(m);
2500 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002501 return err;
2502}
2503
David Howellsea5b7782011-01-14 19:10:03 +00002504/**
2505 * mnt_set_expiry - Put a mount on an expiration list
2506 * @mnt: The mount to list.
2507 * @expiry_list: The list to add the mount to.
2508 */
2509void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2510{
Al Viro97216be2013-03-16 15:12:40 -04002511 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00002512
Al Viro6776db3d2011-11-25 00:22:05 -05002513 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002514
Al Viro97216be2013-03-16 15:12:40 -04002515 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002516}
2517EXPORT_SYMBOL(mnt_set_expiry);
2518
2519/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 * process a list of expirable mountpoints with the intent of discarding any
2521 * mountpoints that aren't in use and haven't been touched since last we came
2522 * here
2523 */
2524void mark_mounts_for_expiry(struct list_head *mounts)
2525{
Al Viro761d5c32011-11-24 21:07:43 -05002526 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 LIST_HEAD(graveyard);
2528
2529 if (list_empty(mounts))
2530 return;
2531
Al Viro97216be2013-03-16 15:12:40 -04002532 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002533 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534
2535 /* extract from the expiration list every vfsmount that matches the
2536 * following criteria:
2537 * - only referenced by its parent vfsmount
2538 * - still marked for expiry (marked on the last call here; marks are
2539 * cleared by mntput())
2540 */
Al Viro6776db3d2011-11-25 00:22:05 -05002541 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002542 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002543 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002545 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002547 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002548 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002549 touch_mnt_namespace(mnt->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06002550 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04002551 }
Al Viro719ea2f2013-09-29 11:24:49 -04002552 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002553 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554}
2555
2556EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2557
2558/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002559 * Ripoff of 'select_parent()'
2560 *
2561 * search the list of submounts for a given mountpoint, and move any
2562 * shrinkable submounts to the 'graveyard' list.
2563 */
Al Viro692afc32011-11-24 21:15:14 -05002564static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002565{
Al Viro692afc32011-11-24 21:15:14 -05002566 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002567 struct list_head *next;
2568 int found = 0;
2569
2570repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002571 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002572resume:
Al Viro6b41d532011-11-24 23:24:33 -05002573 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002574 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002575 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002576
2577 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002578 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002579 continue;
2580 /*
2581 * Descend a level if the d_mounts list is non-empty.
2582 */
Al Viro6b41d532011-11-24 23:24:33 -05002583 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002584 this_parent = mnt;
2585 goto repeat;
2586 }
2587
Al Viro1ab59732011-11-24 21:35:16 -05002588 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002589 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002590 found++;
2591 }
2592 }
2593 /*
2594 * All done at this level ... ascend and resume the search
2595 */
2596 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002597 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002598 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002599 goto resume;
2600 }
2601 return found;
2602}
2603
2604/*
2605 * process a list of expirable mountpoints with the intent of discarding any
2606 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002607 *
Al Viro48a066e2013-09-29 22:06:07 -04002608 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002609 */
Al Virob54b9be2013-03-16 14:39:34 -04002610static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002611{
2612 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002613 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002614
Trond Myklebust5528f9112006-06-09 09:34:17 -04002615 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002616 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002617 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002618 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002619 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002620 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06002621 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04002622 }
2623 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002624}
2625
Trond Myklebust5528f9112006-06-09 09:34:17 -04002626/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 * Some copy_from_user() implementations do not return the exact number of
2628 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2629 * Note that this function differs from copy_from_user() in that it will oops
2630 * on bad values of `to', rather than returning a short copy.
2631 */
Ram Paib58fed82005-11-07 17:16:09 -05002632static long exact_copy_from_user(void *to, const void __user * from,
2633 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634{
2635 char *t = to;
2636 const char __user *f = from;
2637 char c;
2638
2639 if (!access_ok(VERIFY_READ, from, n))
2640 return n;
2641
2642 while (n) {
2643 if (__get_user(c, f)) {
2644 memset(t, 0, n);
2645 break;
2646 }
2647 *t++ = c;
2648 f++;
2649 n--;
2650 }
2651 return n;
2652}
2653
Al Virob40ef862015-12-14 18:44:44 -05002654void *copy_mount_options(const void __user * data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655{
2656 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 unsigned long size;
Al Virob40ef862015-12-14 18:44:44 -05002658 char *copy;
Ram Paib58fed82005-11-07 17:16:09 -05002659
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 if (!data)
Al Virob40ef862015-12-14 18:44:44 -05002661 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662
Al Virob40ef862015-12-14 18:44:44 -05002663 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2664 if (!copy)
2665 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666
2667 /* We only care that *some* data at the address the user
2668 * gave us is valid. Just in case, we'll zero
2669 * the remainder of the page.
2670 */
2671 /* copy_from_user cannot cross TASK_SIZE ! */
2672 size = TASK_SIZE - (unsigned long)data;
2673 if (size > PAGE_SIZE)
2674 size = PAGE_SIZE;
2675
Al Virob40ef862015-12-14 18:44:44 -05002676 i = size - exact_copy_from_user(copy, data, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 if (!i) {
Al Virob40ef862015-12-14 18:44:44 -05002678 kfree(copy);
2679 return ERR_PTR(-EFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 }
2681 if (i != PAGE_SIZE)
Al Virob40ef862015-12-14 18:44:44 -05002682 memset(copy + i, 0, PAGE_SIZE - i);
2683 return copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684}
2685
Tim Gardnerb8850d12014-08-28 11:26:03 -06002686char *copy_mount_string(const void __user *data)
Vegard Nossumeca6f532009-09-18 13:05:45 -07002687{
Tim Gardnerb8850d12014-08-28 11:26:03 -06002688 return data ? strndup_user(data, PAGE_SIZE) : NULL;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002689}
2690
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691/*
2692 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2693 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2694 *
2695 * data is a (void *) that can point to any structure up to
2696 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2697 * information (or be NULL).
2698 *
2699 * Pre-0.97 versions of mount() didn't have a flags word.
2700 * When the flags word was introduced its top half was required
2701 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2702 * Therefore, if this magic number is present, it carries no information
2703 * and must be discarded.
2704 */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002705long do_mount(const char *dev_name, const char __user *dir_name,
Al Viro808d4e32012-10-11 11:42:01 -04002706 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707{
Al Viro2d92ab32008-08-02 00:51:11 -04002708 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 int retval = 0;
2710 int mnt_flags = 0;
2711
2712 /* Discard magic */
2713 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2714 flags &= ~MS_MGC_MSK;
2715
2716 /* Basic sanity checks */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 if (data_page)
2718 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2719
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002720 /* ... and get the mountpoint */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002721 retval = user_path(dir_name, &path);
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002722 if (retval)
2723 return retval;
2724
2725 retval = security_sb_mount(dev_name, &path,
2726 type_page, flags, data_page);
Al Viro0d5cadb2013-05-04 14:40:51 -04002727 if (!retval && !may_mount())
2728 retval = -EPERM;
Jeff Layton9e8925b2015-11-16 09:49:34 -05002729 if (!retval && (flags & MS_MANDLOCK) && !may_mandlock())
2730 retval = -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002731 if (retval)
2732 goto dput_out;
2733
Andi Kleen613cbe32009-04-19 18:40:43 +02002734 /* Default to relatime unless overriden */
2735 if (!(flags & MS_NOATIME))
2736 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002737
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 /* Separate the per-mountpoint flags */
2739 if (flags & MS_NOSUID)
2740 mnt_flags |= MNT_NOSUID;
2741 if (flags & MS_NODEV)
2742 mnt_flags |= MNT_NODEV;
2743 if (flags & MS_NOEXEC)
2744 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002745 if (flags & MS_NOATIME)
2746 mnt_flags |= MNT_NOATIME;
2747 if (flags & MS_NODIRATIME)
2748 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002749 if (flags & MS_STRICTATIME)
2750 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002751 if (flags & MS_RDONLY)
2752 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002753
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07002754 /* The default atime for remount is preservation */
2755 if ((flags & MS_REMOUNT) &&
2756 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2757 MS_STRICTATIME)) == 0)) {
2758 mnt_flags &= ~MNT_ATIME_MASK;
2759 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2760 }
2761
Al Viro7a4dec52010-08-09 12:05:43 -04002762 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002763 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
Eric W. Biederman93faccbb2017-02-01 06:06:16 +13002764 MS_STRICTATIME | MS_NOREMOTELOCK | MS_SUBMOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002767 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 data_page);
2769 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002770 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002771 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002772 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002774 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 else
Al Viro2d92ab32008-08-02 00:51:11 -04002776 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 dev_name, data_page);
2778dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002779 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 return retval;
2781}
2782
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002783static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2784{
2785 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2786}
2787
2788static void dec_mnt_namespaces(struct ucounts *ucounts)
2789{
2790 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2791}
2792
Eric W. Biederman771b1372012-07-26 21:08:32 -07002793static void free_mnt_ns(struct mnt_namespace *ns)
2794{
Al Viro6344c432014-11-01 00:45:45 -04002795 ns_free_inum(&ns->ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002796 dec_mnt_namespaces(ns->ucounts);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002797 put_user_ns(ns->user_ns);
2798 kfree(ns);
2799}
2800
Eric W. Biederman8823c072010-03-07 18:49:36 -08002801/*
2802 * Assign a sequence number so we can detect when we attempt to bind
2803 * mount a reference to an older mount namespace into the current
2804 * mount namespace, preventing reference counting loops. A 64bit
2805 * number incrementing at 10Ghz will take 12,427 years to wrap which
2806 * is effectively never, so we can ignore the possibility.
2807 */
2808static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2809
Eric W. Biederman771b1372012-07-26 21:08:32 -07002810static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002811{
2812 struct mnt_namespace *new_ns;
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002813 struct ucounts *ucounts;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002814 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002815
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002816 ucounts = inc_mnt_namespaces(user_ns);
2817 if (!ucounts)
Eric W. Biedermandf75e772016-09-22 13:08:36 -05002818 return ERR_PTR(-ENOSPC);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002819
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002820 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002821 if (!new_ns) {
2822 dec_mnt_namespaces(ucounts);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002823 return ERR_PTR(-ENOMEM);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002824 }
Al Viro6344c432014-11-01 00:45:45 -04002825 ret = ns_alloc_inum(&new_ns->ns);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002826 if (ret) {
2827 kfree(new_ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002828 dec_mnt_namespaces(ucounts);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002829 return ERR_PTR(ret);
2830 }
Al Viro33c42942014-11-01 02:32:53 -04002831 new_ns->ns.ops = &mntns_operations;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002832 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002833 atomic_set(&new_ns->count, 1);
2834 new_ns->root = NULL;
2835 INIT_LIST_HEAD(&new_ns->list);
2836 init_waitqueue_head(&new_ns->poll);
2837 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002838 new_ns->user_ns = get_user_ns(user_ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002839 new_ns->ucounts = ucounts;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002840 new_ns->mounts = 0;
2841 new_ns->pending_mounts = 0;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002842 return new_ns;
2843}
2844
Emese Revfy0766f782016-06-20 20:42:34 +02002845__latent_entropy
Al Viro9559f682013-09-28 20:47:57 -04002846struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2847 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002849 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002850 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002851 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04002852 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05002853 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002854 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855
Al Viro9559f682013-09-28 20:47:57 -04002856 BUG_ON(!ns);
2857
2858 if (likely(!(flags & CLONE_NEWNS))) {
2859 get_mnt_ns(ns);
2860 return ns;
2861 }
2862
2863 old = ns->root;
2864
Eric W. Biederman771b1372012-07-26 21:08:32 -07002865 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002866 if (IS_ERR(new_ns))
2867 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868
Al Viro97216be2013-03-16 15:12:40 -04002869 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002871 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04002872 if (user_ns != ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002873 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002874 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002875 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04002876 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002877 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002878 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 }
Al Virobe08d6d2011-12-06 13:32:36 -05002880 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002881 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882
2883 /*
2884 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2885 * as belonging to new namespace. We have already acquired a private
2886 * fs_struct, so tsk->fs->lock is not needed.
2887 */
Al Viro909b0a82011-11-25 03:06:56 -05002888 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002889 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002891 q->mnt_ns = new_ns;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002892 new_ns->mounts++;
Al Viro9559f682013-09-28 20:47:57 -04002893 if (new_fs) {
2894 if (&p->mnt == new_fs->root.mnt) {
2895 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002896 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 }
Al Viro9559f682013-09-28 20:47:57 -04002898 if (&p->mnt == new_fs->pwd.mnt) {
2899 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002900 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 }
Al Viro909b0a82011-11-25 03:06:56 -05002903 p = next_mnt(p, old);
2904 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002905 if (!q)
2906 break;
2907 while (p->mnt.mnt_root != q->mnt.mnt_root)
2908 p = next_mnt(p, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 }
Al Viro328e6d92013-03-16 14:49:45 -04002910 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002913 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002915 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916
JANAK DESAI741a2952006-02-07 12:59:00 -08002917 return new_ns;
2918}
2919
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002920/**
2921 * create_mnt_ns - creates a private namespace and adds a root filesystem
2922 * @mnt: pointer to the new root filesystem mountpoint
2923 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002924static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002925{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002926 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002927 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002928 struct mount *mnt = real_mount(m);
2929 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002930 new_ns->root = mnt;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002931 new_ns->mounts++;
Al Virob1983cd2013-05-04 15:18:53 -04002932 list_add(&mnt->mnt_list, &new_ns->list);
Al Viroc1334492011-11-16 16:12:14 -05002933 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002934 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002935 }
2936 return new_ns;
2937}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002938
Al Viroea441d12011-11-16 21:43:59 -05002939struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2940{
2941 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002942 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002943 struct path path;
2944 int err;
2945
2946 ns = create_mnt_ns(mnt);
2947 if (IS_ERR(ns))
2948 return ERR_CAST(ns);
2949
2950 err = vfs_path_lookup(mnt->mnt_root, mnt,
2951 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2952
2953 put_mnt_ns(ns);
2954
2955 if (err)
2956 return ERR_PTR(err);
2957
2958 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002959 s = path.mnt->mnt_sb;
2960 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002961 mntput(path.mnt);
2962 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002963 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002964 /* ... and return the root of (sub)tree on it */
2965 return path.dentry;
2966}
2967EXPORT_SYMBOL(mount_subtree);
2968
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002969SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2970 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002972 int ret;
2973 char *kernel_type;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002974 char *kernel_dev;
Al Virob40ef862015-12-14 18:44:44 -05002975 void *options;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976
Tim Gardnerb8850d12014-08-28 11:26:03 -06002977 kernel_type = copy_mount_string(type);
2978 ret = PTR_ERR(kernel_type);
2979 if (IS_ERR(kernel_type))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002980 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981
Tim Gardnerb8850d12014-08-28 11:26:03 -06002982 kernel_dev = copy_mount_string(dev_name);
2983 ret = PTR_ERR(kernel_dev);
2984 if (IS_ERR(kernel_dev))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002985 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986
Al Virob40ef862015-12-14 18:44:44 -05002987 options = copy_mount_options(data);
2988 ret = PTR_ERR(options);
2989 if (IS_ERR(options))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002990 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991
Al Virob40ef862015-12-14 18:44:44 -05002992 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002993
Al Virob40ef862015-12-14 18:44:44 -05002994 kfree(options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002995out_data:
2996 kfree(kernel_dev);
2997out_dev:
Vegard Nossumeca6f532009-09-18 13:05:45 -07002998 kfree(kernel_type);
2999out_type:
3000 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001}
3002
3003/*
Al Viroafac7cb2011-11-23 19:34:49 -05003004 * Return true if path is reachable from root
3005 *
Al Viro48a066e2013-09-29 22:06:07 -04003006 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05003007 */
Al Viro643822b2011-11-24 22:00:28 -05003008bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05003009 const struct path *root)
3010{
Al Viro643822b2011-11-24 22:00:28 -05003011 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05003012 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05003013 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05003014 }
Al Viro643822b2011-11-24 22:00:28 -05003015 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05003016}
3017
Mickaël Salaün640eb7e2016-11-14 22:14:35 +01003018bool path_is_under(const struct path *path1, const struct path *path2)
Al Viroafac7cb2011-11-23 19:34:49 -05003019{
Yaowei Bai25ab4c92015-11-17 14:40:10 +08003020 bool res;
Al Viro48a066e2013-09-29 22:06:07 -04003021 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05003022 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04003023 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05003024 return res;
3025}
3026EXPORT_SYMBOL(path_is_under);
3027
3028/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 * pivot_root Semantics:
3030 * Moves the root file system of the current process to the directory put_old,
3031 * makes new_root as the new root file system of the current process, and sets
3032 * root/cwd of all processes which had them on the current root to new_root.
3033 *
3034 * Restrictions:
3035 * The new_root and put_old must be directories, and must not be on the
3036 * same file system as the current process root. The put_old must be
3037 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3038 * pointed to by put_old must yield the same directory as new_root. No other
3039 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3040 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08003041 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3042 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3043 * in this situation.
3044 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 * Notes:
3046 * - we don't move root/cwd if they are not at the root (reason: if something
3047 * cared enough to change them, it's probably wrong to force them elsewhere)
3048 * - it's okay to pick a root that isn't the root of a file system, e.g.
3049 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3050 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3051 * first.
3052 */
Heiko Carstens3480b252009-01-14 14:14:16 +01003053SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3054 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055{
Al Viro2d8f3032008-07-22 09:59:21 -04003056 struct path new, old, parent_path, root_parent, root;
Al Viro84d17192013-03-15 10:53:28 -04003057 struct mount *new_mnt, *root_mnt, *old_mnt;
3058 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059 int error;
3060
Al Viro9b40bc92013-02-22 22:45:42 -05003061 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062 return -EPERM;
3063
Al Viro2d8f3032008-07-22 09:59:21 -04003064 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 if (error)
3066 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067
Al Viro2d8f3032008-07-22 09:59:21 -04003068 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 if (error)
3070 goto out1;
3071
Al Viro2d8f3032008-07-22 09:59:21 -04003072 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04003073 if (error)
3074 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02003076 get_fs_root(current->fs, &root);
Al Viro84d17192013-03-15 10:53:28 -04003077 old_mp = lock_mount(&old);
3078 error = PTR_ERR(old_mp);
3079 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04003080 goto out3;
3081
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05003083 new_mnt = real_mount(new.mnt);
3084 root_mnt = real_mount(root.mnt);
Al Viro84d17192013-03-15 10:53:28 -04003085 old_mnt = real_mount(old.mnt);
3086 if (IS_MNT_SHARED(old_mnt) ||
Al Virofc7be132011-11-25 01:05:37 -05003087 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3088 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04003089 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05003090 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003091 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003092 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3093 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003095 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04003096 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 error = -EBUSY;
Al Viro84d17192013-03-15 10:53:28 -04003098 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04003099 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04003101 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04003102 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05003103 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003104 goto out4; /* not attached */
Al Viro84d17192013-03-15 10:53:28 -04003105 root_mp = root_mnt->mnt_mp;
Al Viro2d8f3032008-07-22 09:59:21 -04003106 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04003107 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05003108 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003109 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08003110 /* make sure we can reach put_old from new_root */
Al Viro84d17192013-03-15 10:53:28 -04003111 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04003112 goto out4;
Eric W. Biederman0d082602014-10-08 10:42:27 -07003113 /* make certain new is below the root */
3114 if (!is_path_reachable(new_mnt, new.dentry, &root))
3115 goto out4;
Al Viro84d17192013-03-15 10:53:28 -04003116 root_mp->m_count++; /* pin it so it won't go away */
Al Viro719ea2f2013-09-29 11:24:49 -04003117 lock_mount_hash();
Al Viro419148d2011-11-24 19:41:16 -05003118 detach_mnt(new_mnt, &parent_path);
3119 detach_mnt(root_mnt, &root_parent);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003120 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3121 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3122 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3123 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003124 /* mount old root on put_old */
Al Viro84d17192013-03-15 10:53:28 -04003125 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08003126 /* mount new_root on / */
Al Viro84d17192013-03-15 10:53:28 -04003127 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003128 touch_mnt_namespace(current->nsproxy->mnt_ns);
Eric W. Biederman4fed6552014-10-08 10:42:57 -07003129 /* A moved mount should not expire automatically */
3130 list_del_init(&new_mnt->mnt_expire);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13003131 put_mountpoint(root_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04003132 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04003133 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04003135out4:
Al Viro84d17192013-03-15 10:53:28 -04003136 unlock_mount(old_mp);
Al Virob12cea92011-03-18 08:55:38 -04003137 if (!error) {
3138 path_put(&root_parent);
3139 path_put(&parent_path);
3140 }
3141out3:
Al Viro8c3ee422008-03-22 18:00:39 -04003142 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04003143out2:
Al Viro2d8f3032008-07-22 09:59:21 -04003144 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145out1:
Al Viro2d8f3032008-07-22 09:59:21 -04003146 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149}
3150
3151static void __init init_mount_tree(void)
3152{
3153 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003154 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08003155 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003156 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003158 type = get_fs_type("rootfs");
3159 if (!type)
3160 panic("Can't find rootfs type");
3161 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3162 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 if (IS_ERR(mnt))
3164 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11003165
Trond Myklebust3b22edc2009-06-23 17:29:49 -04003166 ns = create_mnt_ns(mnt);
3167 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003170 init_task.nsproxy->mnt_ns = ns;
3171 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172
Al Virobe08d6d2011-12-06 13:32:36 -05003173 root.mnt = mnt;
3174 root.dentry = mnt->mnt_root;
Eric W. Biedermanda362b02014-10-07 12:19:53 -07003175 mnt->mnt_flags |= MNT_LOCKED;
Jan Blunckac748a02008-02-14 19:34:39 -08003176
3177 set_fs_pwd(current->fs, &root);
3178 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179}
3180
Denis Cheng74bf17c2007-10-16 23:26:30 -07003181void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182{
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003183 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184
Al Viro7d6fec42011-11-23 12:14:10 -05003185 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09003186 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187
Al Viro0818bf22014-02-28 13:46:44 -05003188 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04003189 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05003190 mhash_entries, 19,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003191 HASH_ZERO,
Al Viro0818bf22014-02-28 13:46:44 -05003192 &m_hash_shift, &m_hash_mask, 0, 0);
3193 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3194 sizeof(struct hlist_head),
3195 mphash_entries, 19,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003196 HASH_ZERO,
Al Viro0818bf22014-02-28 13:46:44 -05003197 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198
Al Viro84d17192013-03-15 10:53:28 -04003199 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200 panic("Failed to allocate mount hash table\n");
3201
Tejun Heo4b93dc92013-11-28 14:54:43 -05003202 kernfs_init();
3203
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003204 err = sysfs_init();
3205 if (err)
3206 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003207 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06003208 fs_kobj = kobject_create_and_add("fs", NULL);
3209 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003210 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 init_rootfs();
3212 init_mount_tree();
3213}
3214
Trond Myklebust616511d2009-06-22 15:09:13 -04003215void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216{
Al Virod498b252010-02-05 02:21:06 -05003217 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04003218 return;
Al Viro7b00ed62013-09-16 21:19:20 -04003219 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003220 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221}
Al Viro9d412a42011-03-17 22:08:28 -04003222
3223struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3224{
Tim Chen423e0ab2011-07-19 09:32:38 -07003225 struct vfsmount *mnt;
3226 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3227 if (!IS_ERR(mnt)) {
3228 /*
3229 * it is a longterm mount, don't release mnt until
3230 * we unmount before file sys is unregistered
3231 */
Al Virof7a99c52012-06-09 00:59:08 -04003232 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07003233 }
3234 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04003235}
3236EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07003237
3238void kern_unmount(struct vfsmount *mnt)
3239{
3240 /* release long term mount so mount point can be released */
3241 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04003242 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04003243 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07003244 mntput(mnt);
3245 }
3246}
3247EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05003248
3249bool our_mnt(struct vfsmount *mnt)
3250{
Al Viro143c8c92011-11-25 00:46:35 -05003251 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05003252}
Eric W. Biederman8823c072010-03-07 18:49:36 -08003253
Eric W. Biederman31515272013-03-15 01:45:51 -07003254bool current_chrooted(void)
3255{
3256 /* Does the current process have a non-standard root */
3257 struct path ns_root;
3258 struct path fs_root;
3259 bool chrooted;
3260
3261 /* Find the namespace root */
3262 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3263 ns_root.dentry = ns_root.mnt->mnt_root;
3264 path_get(&ns_root);
3265 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3266 ;
3267
3268 get_fs_root(current->fs, &fs_root);
3269
3270 chrooted = !path_equal(&fs_root, &ns_root);
3271
3272 path_put(&fs_root);
3273 path_put(&ns_root);
3274
3275 return chrooted;
3276}
3277
Eric W. Biederman8654df42016-06-09 16:06:06 -05003278static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3279 int *new_mnt_flags)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003280{
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003281 int new_flags = *new_mnt_flags;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003282 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07003283 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003284
Al Viro44bb4382013-09-16 21:37:36 -04003285 down_read(&namespace_sem);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003286 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07003287 struct mount *child;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003288 int mnt_flags;
3289
Eric W. Biederman8654df42016-06-09 16:06:06 -05003290 if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
Eric W. Biedermane51db732013-03-30 19:57:41 -07003291 continue;
3292
Eric W. Biederman7e96c1b2015-05-08 16:36:50 -05003293 /* This mount is not fully visible if it's root directory
3294 * is not the root directory of the filesystem.
3295 */
3296 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3297 continue;
3298
Eric W. Biedermana1935c12016-06-15 06:59:49 -05003299 /* A local view of the mount flags */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003300 mnt_flags = mnt->mnt.mnt_flags;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003301
Eric W. Biederman695e9df2016-06-10 12:21:40 -05003302 /* Don't miss readonly hidden in the superblock flags */
3303 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3304 mnt_flags |= MNT_LOCK_READONLY;
3305
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003306 /* Verify the mount flags are equal to or more permissive
3307 * than the proposed new mount.
3308 */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003309 if ((mnt_flags & MNT_LOCK_READONLY) &&
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003310 !(new_flags & MNT_READONLY))
3311 continue;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003312 if ((mnt_flags & MNT_LOCK_ATIME) &&
3313 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003314 continue;
3315
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003316 /* This mount is not fully visible if there are any
3317 * locked child mounts that cover anything except for
3318 * empty directories.
Eric W. Biedermane51db732013-03-30 19:57:41 -07003319 */
3320 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3321 struct inode *inode = child->mnt_mountpoint->d_inode;
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003322 /* Only worry about locked mounts */
Eric W. Biedermand71ed6c2016-05-27 14:50:05 -05003323 if (!(child->mnt.mnt_flags & MNT_LOCKED))
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003324 continue;
Eric W. Biederman7236c852015-05-13 20:51:09 -05003325 /* Is the directory permanetly empty? */
3326 if (!is_empty_dir_inode(inode))
Eric W. Biedermane51db732013-03-30 19:57:41 -07003327 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003328 }
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003329 /* Preserve the locked attributes */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003330 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003331 MNT_LOCK_ATIME);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003332 visible = true;
3333 goto found;
3334 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003335 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07003336found:
Al Viro44bb4382013-09-16 21:37:36 -04003337 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003338 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003339}
3340
Eric W. Biederman8654df42016-06-09 16:06:06 -05003341static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3342{
Eric W. Biedermana1935c12016-06-15 06:59:49 -05003343 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
Eric W. Biederman8654df42016-06-09 16:06:06 -05003344 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3345 unsigned long s_iflags;
3346
3347 if (ns->user_ns == &init_user_ns)
3348 return false;
3349
3350 /* Can this filesystem be too revealing? */
3351 s_iflags = mnt->mnt_sb->s_iflags;
3352 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3353 return false;
3354
Eric W. Biedermana1935c12016-06-15 06:59:49 -05003355 if ((s_iflags & required_iflags) != required_iflags) {
3356 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3357 required_iflags);
3358 return true;
3359 }
3360
Eric W. Biederman8654df42016-06-09 16:06:06 -05003361 return !mnt_already_visible(ns, mnt, new_mnt_flags);
3362}
3363
Andy Lutomirski380cf5b2016-06-23 16:41:05 -05003364bool mnt_may_suid(struct vfsmount *mnt)
3365{
3366 /*
3367 * Foreign mounts (accessed via fchdir or through /proc
3368 * symlinks) are always treated as if they are nosuid. This
3369 * prevents namespaces from trusting potentially unsafe
3370 * suid/sgid bits, file caps, or security labels that originate
3371 * in other namespaces.
3372 */
3373 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3374 current_in_userns(mnt->mnt_sb->s_user_ns);
3375}
3376
Al Viro64964522014-11-01 00:37:32 -04003377static struct ns_common *mntns_get(struct task_struct *task)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003378{
Al Viro58be28252014-11-01 00:00:23 -04003379 struct ns_common *ns = NULL;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003380 struct nsproxy *nsproxy;
3381
Eric W. Biederman728dba32014-02-03 19:13:49 -08003382 task_lock(task);
3383 nsproxy = task->nsproxy;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003384 if (nsproxy) {
Al Viro58be28252014-11-01 00:00:23 -04003385 ns = &nsproxy->mnt_ns->ns;
3386 get_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003387 }
Eric W. Biederman728dba32014-02-03 19:13:49 -08003388 task_unlock(task);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003389
3390 return ns;
3391}
3392
Al Viro64964522014-11-01 00:37:32 -04003393static void mntns_put(struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003394{
Al Viro58be28252014-11-01 00:00:23 -04003395 put_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003396}
3397
Al Viro64964522014-11-01 00:37:32 -04003398static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003399{
3400 struct fs_struct *fs = current->fs;
Al Viro4f757f32017-04-15 17:31:22 -04003401 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003402 struct path root;
Al Viro4f757f32017-04-15 17:31:22 -04003403 int err;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003404
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003405 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -07003406 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3407 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08003408 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003409
3410 if (fs->users != 1)
3411 return -EINVAL;
3412
3413 get_mnt_ns(mnt_ns);
Al Viro4f757f32017-04-15 17:31:22 -04003414 old_mnt_ns = nsproxy->mnt_ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003415 nsproxy->mnt_ns = mnt_ns;
3416
3417 /* Find the root */
Al Viro4f757f32017-04-15 17:31:22 -04003418 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3419 "/", LOOKUP_DOWN, &root);
3420 if (err) {
3421 /* revert to old namespace */
3422 nsproxy->mnt_ns = old_mnt_ns;
3423 put_mnt_ns(mnt_ns);
3424 return err;
3425 }
Eric W. Biederman8823c072010-03-07 18:49:36 -08003426
Andrei Vagin40683672017-06-08 17:32:29 -07003427 put_mnt_ns(old_mnt_ns);
3428
Eric W. Biederman8823c072010-03-07 18:49:36 -08003429 /* Update the pwd and root */
3430 set_fs_pwd(fs, &root);
3431 set_fs_root(fs, &root);
3432
3433 path_put(&root);
3434 return 0;
3435}
3436
Andrey Vaginbcac25a2016-09-06 00:47:13 -07003437static struct user_namespace *mntns_owner(struct ns_common *ns)
3438{
3439 return to_mnt_ns(ns)->user_ns;
3440}
3441
Eric W. Biederman8823c072010-03-07 18:49:36 -08003442const struct proc_ns_operations mntns_operations = {
3443 .name = "mnt",
3444 .type = CLONE_NEWNS,
3445 .get = mntns_get,
3446 .put = mntns_put,
3447 .install = mntns_install,
Andrey Vaginbcac25a2016-09-06 00:47:13 -07003448 .owner = mntns_owner,
Eric W. Biederman8823c072010-03-07 18:49:36 -08003449};