blob: 02c6875dd9457624bfced03bf1721bb5d494944f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
Al Virod10577a2011-12-07 13:06:11 -050012#include <linux/export.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080013#include <linux/capability.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080014#include <linux/mnt_namespace.h>
Eric W. Biederman771b1372012-07-26 21:08:32 -070015#include <linux/user_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/namei.h>
17#include <linux/security.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010018#include <linux/idr.h>
Rob Landley57f150a2013-09-11 14:26:10 -070019#include <linux/init.h> /* init_rootfs */
Al Virod10577a2011-12-07 13:06:11 -050020#include <linux/fs_struct.h> /* get_fs_root et.al. */
21#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22#include <linux/uaccess.h>
David Howells0bb80f22013-04-12 01:50:06 +010023#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070024#include <linux/magic.h>
Al Viro0818bf22014-02-28 13:46:44 -050025#include <linux/bootmem.h>
Al Viro9ea459e2014-08-08 13:08:20 -040026#include <linux/task_work.h>
Ram Pai07b20882005-11-07 17:19:07 -050027#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070028#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Al Viro0818bf22014-02-28 13:46:44 -050030static unsigned int m_hash_mask __read_mostly;
31static unsigned int m_hash_shift __read_mostly;
32static unsigned int mp_hash_mask __read_mostly;
33static unsigned int mp_hash_shift __read_mostly;
34
35static __initdata unsigned long mhash_entries;
36static int __init set_mhash_entries(char *str)
37{
38 if (!str)
39 return 0;
40 mhash_entries = simple_strtoul(str, &str, 0);
41 return 1;
42}
43__setup("mhash_entries=", set_mhash_entries);
44
45static __initdata unsigned long mphash_entries;
46static int __init set_mphash_entries(char *str)
47{
48 if (!str)
49 return 0;
50 mphash_entries = simple_strtoul(str, &str, 0);
51 return 1;
52}
53__setup("mphash_entries=", set_mphash_entries);
Eric Dumazet13f14b42008-02-06 01:37:57 -080054
Al Viroc7999c32014-02-27 14:40:10 -050055static u64 event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010056static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010057static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100058static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040059static int mnt_id_start = 0;
60static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Al Viro38129a12014-03-20 21:10:51 -040062static struct hlist_head *mount_hashtable __read_mostly;
Al Viro0818bf22014-02-28 13:46:44 -050063static struct hlist_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080064static struct kmem_cache *mnt_cache __read_mostly;
Al Viro59aa0da2013-09-16 21:34:53 -040065static DECLARE_RWSEM(namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080067/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060068struct kobject *fs_kobj;
69EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080070
Nick Piggin99b7db72010-08-18 04:37:39 +100071/*
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
74 * up the tree.
75 *
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
78 */
Al Viro48a066e2013-09-29 22:06:07 -040079__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +100080
Al Viro38129a12014-03-20 21:10:51 -040081static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082{
Ram Paib58fed82005-11-07 17:16:09 -050083 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Al Viro0818bf22014-02-28 13:46:44 -050085 tmp = tmp + (tmp >> m_hash_shift);
86 return &mount_hashtable[tmp & m_hash_mask];
87}
88
89static inline struct hlist_head *mp_hash(struct dentry *dentry)
90{
91 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 tmp = tmp + (tmp >> mp_hash_shift);
93 return &mountpoint_hashtable[tmp & mp_hash_mask];
Linus Torvalds1da177e2005-04-16 15:20:36 -070094}
95
Nick Piggin99b7db72010-08-18 04:37:39 +100096/*
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
99 */
Al Virob105e272011-11-24 20:38:33 -0500100static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100101{
102 int res;
103
104retry:
105 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +1000106 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -0500107 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -0400108 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500109 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +1000110 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100111 if (res == -EAGAIN)
112 goto retry;
113
114 return res;
115}
116
Al Virob105e272011-11-24 20:38:33 -0500117static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100118{
Al Viro15169fe2011-11-25 00:50:41 -0500119 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000120 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -0400121 ida_remove(&mnt_id_ida, id);
122 if (mnt_id_start > id)
123 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000124 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100125}
126
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100127/*
128 * Allocate a new peer group ID
129 *
130 * mnt_group_ida is protected by namespace_sem
131 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500132static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100133{
Al Virof21f6222009-06-24 03:12:00 -0400134 int res;
135
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100136 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
137 return -ENOMEM;
138
Al Virof21f6222009-06-24 03:12:00 -0400139 res = ida_get_new_above(&mnt_group_ida,
140 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500141 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400142 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500143 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400144
145 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100146}
147
148/*
149 * Release a peer group ID
150 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500151void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100152{
Al Viro15169fe2011-11-25 00:50:41 -0500153 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400154 ida_remove(&mnt_group_ida, id);
155 if (mnt_group_start > id)
156 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500157 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100158}
159
Nick Pigginb3e19d92011-01-07 17:50:11 +1100160/*
161 * vfsmount lock must be held for read
162 */
Al Viro83adc752011-11-24 22:37:54 -0500163static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100164{
165#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500166 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100167#else
168 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500169 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100170 preempt_enable();
171#endif
172}
173
Nick Pigginb3e19d92011-01-07 17:50:11 +1100174/*
175 * vfsmount lock must be held for write
176 */
Al Viro83adc752011-11-24 22:37:54 -0500177unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100178{
179#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500180 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100181 int cpu;
182
183 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500184 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100185 }
186
187 return count;
188#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500189 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100190#endif
191}
192
Al Viro87b95ce2015-01-10 19:01:08 -0500193static void drop_mountpoint(struct fs_pin *p)
194{
195 struct mount *m = container_of(p, struct mount, mnt_umount);
196 dput(m->mnt_ex_mountpoint);
197 pin_remove(p);
198 mntput(&m->mnt);
199}
200
Al Virob105e272011-11-24 20:38:33 -0500201static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
Al Viroc63181e2011-11-25 02:35:16 -0500203 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
204 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100205 int err;
206
Al Viroc63181e2011-11-25 02:35:16 -0500207 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800208 if (err)
209 goto out_free_cache;
210
211 if (name) {
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800212 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
Al Viroc63181e2011-11-25 02:35:16 -0500213 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800214 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100215 }
216
Nick Pigginb3e19d92011-01-07 17:50:11 +1100217#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500218 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
219 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100220 goto out_free_devname;
221
Al Viroc63181e2011-11-25 02:35:16 -0500222 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100223#else
Al Viroc63181e2011-11-25 02:35:16 -0500224 mnt->mnt_count = 1;
225 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100226#endif
227
Al Viro38129a12014-03-20 21:10:51 -0400228 INIT_HLIST_NODE(&mnt->mnt_hash);
Al Viroc63181e2011-11-25 02:35:16 -0500229 INIT_LIST_HEAD(&mnt->mnt_child);
230 INIT_LIST_HEAD(&mnt->mnt_mounts);
231 INIT_LIST_HEAD(&mnt->mnt_list);
232 INIT_LIST_HEAD(&mnt->mnt_expire);
233 INIT_LIST_HEAD(&mnt->mnt_share);
234 INIT_LIST_HEAD(&mnt->mnt_slave_list);
235 INIT_LIST_HEAD(&mnt->mnt_slave);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700236 INIT_HLIST_NODE(&mnt->mnt_mp_list);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500237#ifdef CONFIG_FSNOTIFY
238 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
239#endif
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 Szeredi4ed5e82f2011-11-21 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 Szeredi4ed5e82f2011-11-21 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 Szeredi4ed5e82f2011-11-21 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 Szeredi4ed5e82f2011-11-21 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 Szeredi4ed5e82f2011-11-21 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 Szeredi4ed5e82f2011-11-21 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 */
593bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
594{
595 struct mount *mnt;
596 if (read_seqretry(&mount_lock, seq))
597 return false;
598 if (bastard == NULL)
599 return true;
600 mnt = real_mount(bastard);
601 mnt_add_count(mnt, 1);
602 if (likely(!read_seqretry(&mount_lock, seq)))
603 return true;
604 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605 mnt_add_count(mnt, -1);
606 return false;
607 }
608 rcu_read_unlock();
609 mntput(bastard);
610 rcu_read_lock();
611 return false;
612}
613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614/*
Al Viro474279d2013-10-01 16:11:26 -0400615 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400616 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 */
Al Viro474279d2013-10-01 16:11:26 -0400618struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Al Viro38129a12014-03-20 21:10:51 -0400620 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400621 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
Al Viro38129a12014-03-20 21:10:51 -0400623 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400624 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
625 return p;
626 return NULL;
627}
628
629/*
630 * find the last mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400631 * mount_lock must be held.
Al Viro474279d2013-10-01 16:11:26 -0400632 */
633struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
634{
Eric W. Biederman411a9382014-12-22 19:12:07 -0600635 struct mount *p, *res = NULL;
636 p = __lookup_mnt(mnt, dentry);
Al Viro38129a12014-03-20 21:10:51 -0400637 if (!p)
638 goto out;
Eric W. Biederman411a9382014-12-22 19:12:07 -0600639 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
640 res = p;
Al Viro38129a12014-03-20 21:10:51 -0400641 hlist_for_each_entry_continue(p, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -0400642 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
643 break;
Eric W. Biederman411a9382014-12-22 19:12:07 -0600644 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
645 res = p;
Al Viro1d6a32a2014-03-20 20:34:43 -0400646 }
Al Viro38129a12014-03-20 21:10:51 -0400647out:
Al Viro1d6a32a2014-03-20 20:34:43 -0400648 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
Ram Paia05964f2005-11-07 17:20:17 -0500651/*
David Howellsf015f1262012-06-25 12:55:28 +0100652 * lookup_mnt - Return the first child mount mounted at path
653 *
654 * "First" means first mounted chronologically. If you create the
655 * following mounts:
656 *
657 * mount /dev/sda1 /mnt
658 * mount /dev/sda2 /mnt
659 * mount /dev/sda3 /mnt
660 *
661 * Then lookup_mnt() on the base /mnt dentry in the root mount will
662 * return successively the root dentry and vfsmount of /dev/sda1, then
663 * /dev/sda2, then /dev/sda3, then NULL.
664 *
665 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500666 */
Al Viro1c755af2009-04-18 14:06:57 -0400667struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500668{
Al Viroc7105362011-11-24 18:22:03 -0500669 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400670 struct vfsmount *m;
671 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000672
Al Viro48a066e2013-09-29 22:06:07 -0400673 rcu_read_lock();
674 do {
675 seq = read_seqbegin(&mount_lock);
676 child_mnt = __lookup_mnt(path->mnt, path->dentry);
677 m = child_mnt ? &child_mnt->mnt : NULL;
678 } while (!legitimize_mnt(m, seq));
679 rcu_read_unlock();
680 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500681}
682
Eric W. Biederman7af13642013-10-04 19:15:13 -0700683/*
684 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
685 * current mount namespace.
686 *
687 * The common case is dentries are not mountpoints at all and that
688 * test is handled inline. For the slow case when we are actually
689 * dealing with a mountpoint of some kind, walk through all of the
690 * mounts in the current mount namespace and test to see if the dentry
691 * is a mountpoint.
692 *
693 * The mount_hashtable is not usable in the context because we
694 * need to identify all mounts that may be in the current mount
695 * namespace not just a mount that happens to have some specified
696 * parent mount.
697 */
698bool __is_local_mountpoint(struct dentry *dentry)
699{
700 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
701 struct mount *mnt;
702 bool is_covered = false;
703
704 if (!d_mountpoint(dentry))
705 goto out;
706
707 down_read(&namespace_sem);
708 list_for_each_entry(mnt, &ns->list, mnt_list) {
709 is_covered = (mnt->mnt_mountpoint == dentry);
710 if (is_covered)
711 break;
712 }
713 up_read(&namespace_sem);
714out:
715 return is_covered;
716}
717
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800718static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
Al Viro84d17192013-03-15 10:53:28 -0400719{
Al Viro0818bf22014-02-28 13:46:44 -0500720 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d17192013-03-15 10:53:28 -0400721 struct mountpoint *mp;
722
Al Viro0818bf22014-02-28 13:46:44 -0500723 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d17192013-03-15 10:53:28 -0400724 if (mp->m_dentry == dentry) {
725 /* might be worth a WARN_ON() */
726 if (d_unlinked(dentry))
727 return ERR_PTR(-ENOENT);
728 mp->m_count++;
729 return mp;
730 }
731 }
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800732 return NULL;
733}
734
735static struct mountpoint *new_mountpoint(struct dentry *dentry)
736{
737 struct hlist_head *chain = mp_hash(dentry);
738 struct mountpoint *mp;
739 int ret;
Al Viro84d17192013-03-15 10:53:28 -0400740
741 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
742 if (!mp)
743 return ERR_PTR(-ENOMEM);
744
Miklos Szeredieed81002013-09-05 14:39:11 +0200745 ret = d_set_mounted(dentry);
746 if (ret) {
Al Viro84d17192013-03-15 10:53:28 -0400747 kfree(mp);
Miklos Szeredieed81002013-09-05 14:39:11 +0200748 return ERR_PTR(ret);
Al Viro84d17192013-03-15 10:53:28 -0400749 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200750
Al Viro84d17192013-03-15 10:53:28 -0400751 mp->m_dentry = dentry;
752 mp->m_count = 1;
Al Viro0818bf22014-02-28 13:46:44 -0500753 hlist_add_head(&mp->m_hash, chain);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700754 INIT_HLIST_HEAD(&mp->m_list);
Al Viro84d17192013-03-15 10:53:28 -0400755 return mp;
756}
757
758static void put_mountpoint(struct mountpoint *mp)
759{
760 if (!--mp->m_count) {
761 struct dentry *dentry = mp->m_dentry;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700762 BUG_ON(!hlist_empty(&mp->m_list));
Al Viro84d17192013-03-15 10:53:28 -0400763 spin_lock(&dentry->d_lock);
764 dentry->d_flags &= ~DCACHE_MOUNTED;
765 spin_unlock(&dentry->d_lock);
Al Viro0818bf22014-02-28 13:46:44 -0500766 hlist_del(&mp->m_hash);
Al Viro84d17192013-03-15 10:53:28 -0400767 kfree(mp);
768 }
769}
770
Al Viro143c8c92011-11-25 00:46:35 -0500771static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800773 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774}
775
Nick Piggin99b7db72010-08-18 04:37:39 +1000776/*
777 * vfsmount lock must be held for write
778 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800779static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500780{
781 if (ns) {
782 ns->event = ++event;
783 wake_up_interruptible(&ns->poll);
784 }
785}
786
Nick Piggin99b7db72010-08-18 04:37:39 +1000787/*
788 * vfsmount lock must be held for write
789 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800790static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500791{
792 if (ns && ns->event != event) {
793 ns->event = event;
794 wake_up_interruptible(&ns->poll);
795 }
796}
797
Nick Piggin99b7db72010-08-18 04:37:39 +1000798/*
799 * vfsmount lock must be held for write
800 */
Eric W. Biederman7bdb11de2014-12-29 13:03:41 -0600801static void unhash_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Al Viro0714a532011-11-24 22:19:58 -0500803 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500804 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500805 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400806 hlist_del_init_rcu(&mnt->mnt_hash);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700807 hlist_del_init(&mnt->mnt_mp_list);
Al Viro84d17192013-03-15 10:53:28 -0400808 put_mountpoint(mnt->mnt_mp);
809 mnt->mnt_mp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810}
811
Nick Piggin99b7db72010-08-18 04:37:39 +1000812/*
813 * vfsmount lock must be held for write
814 */
Eric W. Biederman7bdb11de2014-12-29 13:03:41 -0600815static void detach_mnt(struct mount *mnt, struct path *old_path)
816{
817 old_path->dentry = mnt->mnt_mountpoint;
818 old_path->mnt = &mnt->mnt_parent->mnt;
819 unhash_mnt(mnt);
820}
821
822/*
823 * vfsmount lock must be held for write
824 */
Eric W. Biederman6a46c572015-01-15 22:58:33 -0600825static void umount_mnt(struct mount *mnt)
826{
827 /* old mountpoint will be dropped when we can do that */
828 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
829 unhash_mnt(mnt);
830}
831
832/*
833 * vfsmount lock must be held for write
834 */
Al Viro84d17192013-03-15 10:53:28 -0400835void mnt_set_mountpoint(struct mount *mnt,
836 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500837 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500838{
Al Viro84d17192013-03-15 10:53:28 -0400839 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500840 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro84d17192013-03-15 10:53:28 -0400841 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500842 child_mnt->mnt_parent = mnt;
Al Viro84d17192013-03-15 10:53:28 -0400843 child_mnt->mnt_mp = mp;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700844 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500845}
846
Nick Piggin99b7db72010-08-18 04:37:39 +1000847/*
848 * vfsmount lock must be held for write
849 */
Al Viro84d17192013-03-15 10:53:28 -0400850static void attach_mnt(struct mount *mnt,
851 struct mount *parent,
852 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853{
Al Viro84d17192013-03-15 10:53:28 -0400854 mnt_set_mountpoint(parent, mp, mnt);
Al Viro38129a12014-03-20 21:10:51 -0400855 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
Al Viro84d17192013-03-15 10:53:28 -0400856 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500857}
858
Al Viro12a5b522014-08-10 03:44:55 -0400859static void attach_shadowed(struct mount *mnt,
860 struct mount *parent,
861 struct mount *shadows)
862{
863 if (shadows) {
Linus Torvaldsf6f99332014-08-11 11:44:11 -0700864 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
Al Viro12a5b522014-08-10 03:44:55 -0400865 list_add(&mnt->mnt_child, &shadows->mnt_child);
866 } else {
867 hlist_add_head_rcu(&mnt->mnt_hash,
868 m_hash(&parent->mnt, mnt->mnt_mountpoint));
869 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
870 }
871}
872
Ram Paib90fa9a2005-11-07 17:19:50 -0500873/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000874 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500875 */
Al Viro1d6a32a2014-03-20 20:34:43 -0400876static void commit_tree(struct mount *mnt, struct mount *shadows)
Ram Paib90fa9a2005-11-07 17:19:50 -0500877{
Al Viro0714a532011-11-24 22:19:58 -0500878 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500879 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500880 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500881 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500882
Al Viro0714a532011-11-24 22:19:58 -0500883 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500884
Al Viro1a4eeaf2011-11-25 02:19:55 -0500885 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400886 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500887 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500888
Ram Paib90fa9a2005-11-07 17:19:50 -0500889 list_splice(&head, n->list.prev);
890
Al Viro12a5b522014-08-10 03:44:55 -0400891 attach_shadowed(mnt, parent, shadows);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800892 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893}
894
Al Viro909b0a82011-11-25 03:06:56 -0500895static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896{
Al Viro6b41d532011-11-24 23:24:33 -0500897 struct list_head *next = p->mnt_mounts.next;
898 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500900 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500902 next = p->mnt_child.next;
903 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 break;
Al Viro0714a532011-11-24 22:19:58 -0500905 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 }
907 }
Al Viro6b41d532011-11-24 23:24:33 -0500908 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909}
910
Al Viro315fc832011-11-24 18:57:30 -0500911static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500912{
Al Viro6b41d532011-11-24 23:24:33 -0500913 struct list_head *prev = p->mnt_mounts.prev;
914 while (prev != &p->mnt_mounts) {
915 p = list_entry(prev, struct mount, mnt_child);
916 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500917 }
918 return p;
919}
920
Al Viro9d412a42011-03-17 22:08:28 -0400921struct vfsmount *
922vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
923{
Al Virob105e272011-11-24 20:38:33 -0500924 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400925 struct dentry *root;
926
927 if (!type)
928 return ERR_PTR(-ENODEV);
929
930 mnt = alloc_vfsmnt(name);
931 if (!mnt)
932 return ERR_PTR(-ENOMEM);
933
934 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500935 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400936
937 root = mount_fs(type, flags, name, data);
938 if (IS_ERR(root)) {
David Howells8ffcb322014-01-24 12:17:54 +0000939 mnt_free_id(mnt);
Al Viro9d412a42011-03-17 22:08:28 -0400940 free_vfsmnt(mnt);
941 return ERR_CAST(root);
942 }
943
Al Virob105e272011-11-24 20:38:33 -0500944 mnt->mnt.mnt_root = root;
945 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500946 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500947 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400948 lock_mount_hash();
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100949 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400950 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -0500951 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400952}
953EXPORT_SYMBOL_GPL(vfs_kern_mount);
954
Al Viro87129cc2011-11-24 21:24:27 -0500955static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500956 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957{
Al Viro87129cc2011-11-24 21:24:27 -0500958 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100959 struct mount *mnt;
960 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
David Howellsbe34d1a2012-06-25 12:55:18 +0100962 mnt = alloc_vfsmnt(old->mnt_devname);
963 if (!mnt)
964 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100965
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700966 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +0100967 mnt->mnt_group_id = 0; /* not a peer of original */
968 else
969 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100970
David Howellsbe34d1a2012-06-25 12:55:18 +0100971 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
972 err = mnt_alloc_group_id(mnt);
973 if (err)
974 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100976
Al Virof2ebb3a2014-02-27 09:35:45 -0500977 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700978 /* Don't allow unprivileged users to change mount flags */
Eric W. Biederman9566d672014-07-28 17:26:07 -0700979 if (flag & CL_UNPRIVILEGED) {
980 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
981
982 if (mnt->mnt.mnt_flags & MNT_READONLY)
983 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
984
985 if (mnt->mnt.mnt_flags & MNT_NODEV)
986 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
987
988 if (mnt->mnt.mnt_flags & MNT_NOSUID)
989 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
990
991 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
992 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
993 }
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700994
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -0700995 /* Don't allow unprivileged users to reveal what is under a mount */
Eric W. Biederman381cacb2014-10-07 17:11:46 -0700996 if ((flag & CL_UNPRIVILEGED) &&
997 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -0700998 mnt->mnt.mnt_flags |= MNT_LOCKED;
999
David Howellsbe34d1a2012-06-25 12:55:18 +01001000 atomic_inc(&sb->s_active);
1001 mnt->mnt.mnt_sb = sb;
1002 mnt->mnt.mnt_root = dget(root);
1003 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1004 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -04001005 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001006 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001007 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001008
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001009 if ((flag & CL_SLAVE) ||
1010 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +01001011 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1012 mnt->mnt_master = old;
1013 CLEAR_MNT_SHARED(mnt);
1014 } else if (!(flag & CL_PRIVATE)) {
1015 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1016 list_add(&mnt->mnt_share, &old->mnt_share);
1017 if (IS_MNT_SLAVE(old))
1018 list_add(&mnt->mnt_slave, &old->mnt_slave);
1019 mnt->mnt_master = old->mnt_master;
1020 }
1021 if (flag & CL_MAKE_SHARED)
1022 set_mnt_shared(mnt);
1023
1024 /* stick the duplicate mount on the same expiry list
1025 * as the original if that was on one */
1026 if (flag & CL_EXPIRE) {
1027 if (!list_empty(&old->mnt_expire))
1028 list_add(&mnt->mnt_expire, &old->mnt_expire);
1029 }
1030
Al Virocb338d02011-11-24 20:55:08 -05001031 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001032
1033 out_free:
David Howells8ffcb322014-01-24 12:17:54 +00001034 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001035 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001036 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037}
1038
Al Viro9ea459e2014-08-08 13:08:20 -04001039static void cleanup_mnt(struct mount *mnt)
1040{
1041 /*
1042 * This probably indicates that somebody messed
1043 * up a mnt_want/drop_write() pair. If this
1044 * happens, the filesystem was probably unable
1045 * to make r/w->r/o transitions.
1046 */
1047 /*
1048 * The locking used to deal with mnt_count decrement provides barriers,
1049 * so mnt_get_writers() below is safe.
1050 */
1051 WARN_ON(mnt_get_writers(mnt));
1052 if (unlikely(mnt->mnt_pins.first))
1053 mnt_pin_kill(mnt);
1054 fsnotify_vfsmount_delete(&mnt->mnt);
1055 dput(mnt->mnt.mnt_root);
1056 deactivate_super(mnt->mnt.mnt_sb);
1057 mnt_free_id(mnt);
1058 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1059}
1060
1061static void __cleanup_mnt(struct rcu_head *head)
1062{
1063 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1064}
1065
1066static LLIST_HEAD(delayed_mntput_list);
1067static void delayed_mntput(struct work_struct *unused)
1068{
1069 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1070 struct llist_node *next;
1071
1072 for (; node; node = next) {
1073 next = llist_next(node);
1074 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1075 }
1076}
1077static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1078
Al Viro900148d2011-11-25 00:33:11 -05001079static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001080{
Al Viro48a066e2013-09-29 22:06:07 -04001081 rcu_read_lock();
1082 mnt_add_count(mnt, -1);
1083 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1084 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -05001085 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001086 }
Al Viro719ea2f2013-09-29 11:24:49 -04001087 lock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001088 if (mnt_get_count(mnt)) {
Al Viro48a066e2013-09-29 22:06:07 -04001089 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -04001090 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001091 return;
1092 }
Al Viro48a066e2013-09-29 22:06:07 -04001093 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1094 rcu_read_unlock();
1095 unlock_mount_hash();
1096 return;
1097 }
1098 mnt->mnt.mnt_flags |= MNT_DOOMED;
1099 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301100
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001101 list_del(&mnt->mnt_instance);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001102
1103 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1104 struct mount *p, *tmp;
1105 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1106 umount_mnt(p);
1107 }
1108 }
Al Viro719ea2f2013-09-29 11:24:49 -04001109 unlock_mount_hash();
Al Viro649a7952013-09-28 12:41:25 -04001110
Al Viro9ea459e2014-08-08 13:08:20 -04001111 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1112 struct task_struct *task = current;
1113 if (likely(!(task->flags & PF_KTHREAD))) {
1114 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1115 if (!task_work_add(task, &mnt->mnt_rcu, true))
1116 return;
1117 }
1118 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1119 schedule_delayed_work(&delayed_mntput_work, 1);
1120 return;
1121 }
1122 cleanup_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001123}
Nick Pigginb3e19d92011-01-07 17:50:11 +11001124
1125void mntput(struct vfsmount *mnt)
1126{
1127 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -05001128 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001129 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -05001130 if (unlikely(m->mnt_expiry_mark))
1131 m->mnt_expiry_mark = 0;
1132 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001133 }
1134}
1135EXPORT_SYMBOL(mntput);
1136
1137struct vfsmount *mntget(struct vfsmount *mnt)
1138{
1139 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001140 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001141 return mnt;
1142}
1143EXPORT_SYMBOL(mntget);
1144
Al Viro3064c352014-08-07 09:12:31 -04001145struct vfsmount *mnt_clone_internal(struct path *path)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001146{
Al Viro3064c352014-08-07 09:12:31 -04001147 struct mount *p;
1148 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1149 if (IS_ERR(p))
1150 return ERR_CAST(p);
1151 p->mnt.mnt_flags |= MNT_INTERNAL;
1152 return &p->mnt;
Al Viro7b7b1ac2005-11-07 17:13:39 -05001153}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001155static inline void mangle(struct seq_file *m, const char *s)
1156{
1157 seq_escape(m, s, " \t\n\\");
1158}
1159
1160/*
1161 * Simple .show_options callback for filesystems which don't want to
1162 * implement more complex mount option showing.
1163 *
1164 * See also save_mount_options().
1165 */
Al Viro34c80b12011-12-08 21:32:45 -05001166int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001167{
Al Viro2a32ceb2009-05-08 16:05:57 -04001168 const char *options;
1169
1170 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -05001171 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001172
1173 if (options != NULL && options[0]) {
1174 seq_putc(m, ',');
1175 mangle(m, options);
1176 }
Al Viro2a32ceb2009-05-08 16:05:57 -04001177 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001178
1179 return 0;
1180}
1181EXPORT_SYMBOL(generic_show_options);
1182
1183/*
1184 * If filesystem uses generic_show_options(), this function should be
1185 * called from the fill_super() callback.
1186 *
1187 * The .remount_fs callback usually needs to be handled in a special
1188 * way, to make sure, that previous options are not overwritten if the
1189 * remount fails.
1190 *
1191 * Also note, that if the filesystem's .remount_fs function doesn't
1192 * reset all options to their default value, but changes only newly
1193 * given options, then the displayed options will not reflect reality
1194 * any more.
1195 */
1196void save_mount_options(struct super_block *sb, char *options)
1197{
Al Viro2a32ceb2009-05-08 16:05:57 -04001198 BUG_ON(sb->s_options);
1199 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001200}
1201EXPORT_SYMBOL(save_mount_options);
1202
Al Viro2a32ceb2009-05-08 16:05:57 -04001203void replace_mount_options(struct super_block *sb, char *options)
1204{
1205 char *old = sb->s_options;
1206 rcu_assign_pointer(sb->s_options, options);
1207 if (old) {
1208 synchronize_rcu();
1209 kfree(old);
1210 }
1211}
1212EXPORT_SYMBOL(replace_mount_options);
1213
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001214#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001215/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216static void *m_start(struct seq_file *m, loff_t *pos)
1217{
Al Viro6ce6e242012-06-09 01:16:59 -04001218 struct proc_mounts *p = proc_mounts(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
Ram Pai390c6842005-11-07 17:17:51 -05001220 down_read(&namespace_sem);
Al Viroc7999c32014-02-27 14:40:10 -05001221 if (p->cached_event == p->ns->event) {
1222 void *v = p->cached_mount;
1223 if (*pos == p->cached_index)
1224 return v;
1225 if (*pos == p->cached_index + 1) {
1226 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1227 return p->cached_mount = v;
1228 }
1229 }
1230
1231 p->cached_event = p->ns->event;
1232 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1233 p->cached_index = *pos;
1234 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
1236
1237static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1238{
Al Viro6ce6e242012-06-09 01:16:59 -04001239 struct proc_mounts *p = proc_mounts(m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001240
Al Viroc7999c32014-02-27 14:40:10 -05001241 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1242 p->cached_index = *pos;
1243 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244}
1245
1246static void m_stop(struct seq_file *m, void *v)
1247{
Ram Pai390c6842005-11-07 17:17:51 -05001248 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249}
1250
Al Viro0226f492011-12-06 12:21:54 -05001251static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001252{
Al Viro6ce6e242012-06-09 01:16:59 -04001253 struct proc_mounts *p = proc_mounts(m);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001254 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001255 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256}
1257
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001258const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 .start = m_start,
1260 .next = m_next,
1261 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001262 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001263};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001264#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266/**
1267 * may_umount_tree - check if a mount tree is busy
1268 * @mnt: root of mount tree
1269 *
1270 * This is called to check if a tree of mounts has any
1271 * open files, pwds, chroots or sub mounts that are
1272 * busy.
1273 */
Al Viro909b0a82011-11-25 03:06:56 -05001274int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275{
Al Viro909b0a82011-11-25 03:06:56 -05001276 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001277 int actual_refs = 0;
1278 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001279 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001280 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281
Nick Pigginb3e19d92011-01-07 17:50:11 +11001282 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001283 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001284 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001285 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 }
Al Viro719ea2f2013-09-29 11:24:49 -04001288 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
1290 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001291 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
Ian Kente3474a82006-03-27 01:14:51 -08001293 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294}
1295
1296EXPORT_SYMBOL(may_umount_tree);
1297
1298/**
1299 * may_umount - check if a mount point is busy
1300 * @mnt: root of mount
1301 *
1302 * This is called to check if a mount point has any
1303 * open files, pwds, chroots or sub mounts. If the
1304 * mount has sub mounts this will return busy
1305 * regardless of whether the sub mounts are busy.
1306 *
1307 * Doesn't take quota and stuff into account. IOW, in some cases it will
1308 * give false negatives. The main reason why it's here is that we need
1309 * a non-destructive way to look for easily umountable filesystems.
1310 */
1311int may_umount(struct vfsmount *mnt)
1312{
Ian Kente3474a82006-03-27 01:14:51 -08001313 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001314 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001315 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001316 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001317 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001318 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001319 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001320 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321}
1322
1323EXPORT_SYMBOL(may_umount);
1324
Al Viro38129a12014-03-20 21:10:51 -04001325static HLIST_HEAD(unmounted); /* protected by namespace_sem */
Al Viroe3197d82013-03-16 14:35:16 -04001326
Al Viro97216be2013-03-16 15:12:40 -04001327static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001329 struct hlist_head head;
Al Viro97216be2013-03-16 15:12:40 -04001330
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001331 hlist_move_list(&unmounted, &head);
Al Viro97216be2013-03-16 15:12:40 -04001332
Al Viro97216be2013-03-16 15:12:40 -04001333 up_write(&namespace_sem);
1334
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001335 if (likely(hlist_empty(&head)))
1336 return;
1337
Al Viro48a066e2013-09-29 22:06:07 -04001338 synchronize_rcu();
1339
Al Viro87b95ce2015-01-10 19:01:08 -05001340 group_pin_kill(&head);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001341}
1342
Al Viro97216be2013-03-16 15:12:40 -04001343static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001344{
Al Viro97216be2013-03-16 15:12:40 -04001345 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001346}
1347
Eric W. Biedermane819f152014-12-24 07:20:01 -06001348enum umount_tree_flags {
1349 UMOUNT_SYNC = 1,
1350 UMOUNT_PROPAGATE = 2,
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001351 UMOUNT_CONNECTED = 4,
Eric W. Biedermane819f152014-12-24 07:20:01 -06001352};
Nick Piggin99b7db72010-08-18 04:37:39 +10001353/*
Al Viro48a066e2013-09-29 22:06:07 -04001354 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001355 * namespace_sem must be held for write
1356 */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001357static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001358{
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001359 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001360 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001361
Eric W. Biederman5d884572015-01-03 05:39:35 -06001362 if (how & UMOUNT_PROPAGATE)
1363 propagate_mount_unlock(mnt);
1364
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001365 /* Gather the mounts to umount */
Eric W. Biederman590ce4bc2014-12-22 18:30:08 -06001366 for (p = mnt; p; p = next_mnt(p, mnt)) {
1367 p->mnt.mnt_flags |= MNT_UMOUNT;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001368 list_move(&p->mnt_list, &tmp_list);
Eric W. Biederman590ce4bc2014-12-22 18:30:08 -06001369 }
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001370
Eric W. Biederman411a9382014-12-22 19:12:07 -06001371 /* Hide the mounts from mnt_mounts */
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001372 list_for_each_entry(p, &tmp_list, mnt_list) {
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001373 list_del_init(&p->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -04001374 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001375
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001376 /* Add propogated mounts to the tmp_list */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001377 if (how & UMOUNT_PROPAGATE)
Al Viro7b8a53f2011-01-15 20:08:44 -05001378 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001379
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001380 while (!list_empty(&tmp_list)) {
Eric W. Biedermance07d892014-12-23 21:37:03 -06001381 bool disconnect;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001382 p = list_first_entry(&tmp_list, struct mount, mnt_list);
Al Viro6776db3d2011-11-25 00:22:05 -05001383 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001384 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001385 __touch_mnt_namespace(p->mnt_ns);
1386 p->mnt_ns = NULL;
Eric W. Biedermane819f152014-12-24 07:20:01 -06001387 if (how & UMOUNT_SYNC)
Al Viro48a066e2013-09-29 22:06:07 -04001388 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro87b95ce2015-01-10 19:01:08 -05001389
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001390 disconnect = !(((how & UMOUNT_CONNECTED) &&
1391 mnt_has_parent(p) &&
1392 (p->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) ||
1393 IS_MNT_LOCKED_AND_LAZY(p));
Eric W. Biedermance07d892014-12-23 21:37:03 -06001394
1395 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1396 disconnect ? &unmounted : NULL);
Al Viro676da582011-11-24 21:47:05 -05001397 if (mnt_has_parent(p)) {
Al Viro81b6b062014-08-30 18:32:05 -04001398 mnt_add_count(p->mnt_parent, -1);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001399 if (!disconnect) {
1400 /* Don't forget about p */
1401 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1402 } else {
1403 umount_mnt(p);
1404 }
Al Viro7c4b93d2008-03-21 23:59:49 -04001405 }
Al Viro0f0afb12011-11-24 20:43:10 -05001406 change_mnt_propagation(p, MS_PRIVATE);
Al Viro38129a12014-03-20 21:10:51 -04001407 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408}
1409
Al Virob54b9be2013-03-16 14:39:34 -04001410static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001411
Al Viro1ab59732011-11-24 21:35:16 -05001412static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413{
Al Viro1ab59732011-11-24 21:35:16 -05001414 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 int retval;
1416
Al Viro1ab59732011-11-24 21:35:16 -05001417 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 if (retval)
1419 return retval;
1420
1421 /*
1422 * Allow userspace to request a mountpoint be expired rather than
1423 * unmounting unconditionally. Unmount only happens if:
1424 * (1) the mark is already set (the mark is cleared by mntput())
1425 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1426 */
1427 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001428 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 flags & (MNT_FORCE | MNT_DETACH))
1430 return -EINVAL;
1431
Nick Pigginb3e19d92011-01-07 17:50:11 +11001432 /*
1433 * probably don't strictly need the lock here if we examined
1434 * all race cases, but it's a slowpath.
1435 */
Al Viro719ea2f2013-09-29 11:24:49 -04001436 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001437 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001438 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001440 }
Al Viro719ea2f2013-09-29 11:24:49 -04001441 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Al Viro863d6842011-11-25 00:57:42 -05001443 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 return -EAGAIN;
1445 }
1446
1447 /*
1448 * If we may have to abort operations to get out of this
1449 * mount, and they will themselves hold resources we must
1450 * allow the fs to do things. In the Unix tradition of
1451 * 'Gee thats tricky lets do it in userspace' the umount_begin
1452 * might fail to complete on the first run through as other tasks
1453 * must return, and the like. Thats for the mount program to worry
1454 * about for the moment.
1455 */
1456
Al Viro42faad92008-04-24 07:21:56 -04001457 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001458 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
1461 /*
1462 * No sense to grab the lock for this test, but test itself looks
1463 * somewhat bogus. Suggestions for better replacement?
1464 * Ho-hum... In principle, we might treat that as umount + switch
1465 * to rootfs. GC would eventually take care of the old vfsmount.
1466 * Actually it makes sense, especially if rootfs would contain a
1467 * /reboot - static binary that would close all descriptors and
1468 * call reboot(9). Then init(8) could umount root and exec /reboot.
1469 */
Al Viro1ab59732011-11-24 21:35:16 -05001470 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 /*
1472 * Special case for "unmounting" root ...
1473 * we just try to remount it readonly.
1474 */
Andy Lutomirskia1480dc2014-10-08 12:32:47 -07001475 if (!capable(CAP_SYS_ADMIN))
1476 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001478 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 up_write(&sb->s_umount);
1481 return retval;
1482 }
1483
Al Viro97216be2013-03-16 15:12:40 -04001484 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001485 lock_mount_hash();
Al Viro5addc5d2005-11-07 17:15:49 -05001486 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Al Viro48a066e2013-09-29 22:06:07 -04001488 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001489 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001490 umount_tree(mnt, UMOUNT_PROPAGATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001492 } else {
1493 shrink_submounts(mnt);
1494 retval = -EBUSY;
1495 if (!propagate_mount_busy(mnt, 2)) {
1496 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001497 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Viro48a066e2013-09-29 22:06:07 -04001498 retval = 0;
1499 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 }
Al Viro719ea2f2013-09-29 11:24:49 -04001501 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001502 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 return retval;
1504}
1505
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001506/*
1507 * __detach_mounts - lazily unmount all mounts on the specified dentry
1508 *
1509 * During unlink, rmdir, and d_drop it is possible to loose the path
1510 * to an existing mountpoint, and wind up leaking the mount.
1511 * detach_mounts allows lazily unmounting those mounts instead of
1512 * leaking them.
1513 *
1514 * The caller may hold dentry->d_inode->i_mutex.
1515 */
1516void __detach_mounts(struct dentry *dentry)
1517{
1518 struct mountpoint *mp;
1519 struct mount *mnt;
1520
1521 namespace_lock();
1522 mp = lookup_mountpoint(dentry);
Eric W. Biedermanf53e5792015-01-19 11:48:45 -06001523 if (IS_ERR_OR_NULL(mp))
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001524 goto out_unlock;
1525
1526 lock_mount_hash();
1527 while (!hlist_empty(&mp->m_list)) {
1528 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001529 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1530 struct mount *p, *tmp;
1531 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1532 hlist_add_head(&p->mnt_umount.s_list, &unmounted);
1533 umount_mnt(p);
1534 }
1535 }
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001536 else umount_tree(mnt, UMOUNT_CONNECTED);
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001537 }
1538 unlock_mount_hash();
1539 put_mountpoint(mp);
1540out_unlock:
1541 namespace_unlock();
1542}
1543
Al Viro9b40bc92013-02-22 22:45:42 -05001544/*
1545 * Is the caller allowed to modify his namespace?
1546 */
1547static inline bool may_mount(void)
1548{
1549 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1550}
1551
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552/*
1553 * Now umount can handle mount points as well as block devices.
1554 * This is important for filesystems which use unnamed block devices.
1555 *
1556 * We now support a flag for forced unmount like the other 'big iron'
1557 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1558 */
1559
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001560SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
Al Viro2d8f3032008-07-22 09:59:21 -04001562 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001563 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001565 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001567 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1568 return -EINVAL;
1569
Al Viro9b40bc92013-02-22 22:45:42 -05001570 if (!may_mount())
1571 return -EPERM;
1572
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001573 if (!(flags & UMOUNT_NOFOLLOW))
1574 lookup_flags |= LOOKUP_FOLLOW;
1575
Al Viro197df042013-09-08 14:03:27 -04001576 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 if (retval)
1578 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001579 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001581 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001583 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 goto dput_and_out;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001585 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1586 goto dput_and_out;
Eric W. Biedermanb2f5d4d2014-10-04 14:44:03 -07001587 retval = -EPERM;
1588 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1589 goto dput_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
Al Viro900148d2011-11-25 00:33:11 -05001591 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001593 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001594 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001595 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596out:
1597 return retval;
1598}
1599
1600#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1601
1602/*
Ram Paib58fed82005-11-07 17:16:09 -05001603 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001605SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606{
Ram Paib58fed82005-11-07 17:16:09 -05001607 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608}
1609
1610#endif
1611
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001612static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001613{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001614 /* Is this a proxy for a mount namespace? */
Al Viroe149ed22014-11-01 10:57:28 -04001615 return dentry->d_op == &ns_dentry_operations &&
1616 dentry->d_fsdata == &mntns_operations;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001617}
1618
Al Viro58be28252014-11-01 00:00:23 -04001619struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1620{
1621 return container_of(ns, struct mnt_namespace, ns);
1622}
1623
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001624static bool mnt_ns_loop(struct dentry *dentry)
1625{
1626 /* Could bind mounting the mount namespace inode cause a
1627 * mount namespace loop?
1628 */
1629 struct mnt_namespace *mnt_ns;
1630 if (!is_mnt_ns_file(dentry))
1631 return false;
1632
Al Virof77c8012014-11-01 03:13:17 -04001633 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
Eric W. Biederman8823c072010-03-07 18:49:36 -08001634 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1635}
1636
Al Viro87129cc2011-11-24 21:24:27 -05001637struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001638 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639{
Al Viro84d17192013-03-15 10:53:28 -04001640 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001642 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1643 return ERR_PTR(-EINVAL);
1644
1645 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001646 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001647
Ram Pai36341f62005-11-07 17:17:22 -05001648 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001649 if (IS_ERR(q))
1650 return q;
1651
Al Viroa73324d2011-11-24 22:25:07 -05001652 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
1654 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001655 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001656 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001657 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 continue;
1659
Al Viro909b0a82011-11-25 03:06:56 -05001660 for (s = r; s; s = next_mnt(s, r)) {
Al Viro12a5b522014-08-10 03:44:55 -04001661 struct mount *t = NULL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001662 if (!(flag & CL_COPY_UNBINDABLE) &&
1663 IS_MNT_UNBINDABLE(s)) {
1664 s = skip_mnt_tree(s);
1665 continue;
1666 }
1667 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1668 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001669 s = skip_mnt_tree(s);
1670 continue;
1671 }
Al Viro0714a532011-11-24 22:19:58 -05001672 while (p != s->mnt_parent) {
1673 p = p->mnt_parent;
1674 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 }
Al Viro87129cc2011-11-24 21:24:27 -05001676 p = s;
Al Viro84d17192013-03-15 10:53:28 -04001677 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001678 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001679 if (IS_ERR(q))
1680 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04001681 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05001682 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro12a5b522014-08-10 03:44:55 -04001683 mnt_set_mountpoint(parent, p->mnt_mp, q);
1684 if (!list_empty(&parent->mnt_mounts)) {
1685 t = list_last_entry(&parent->mnt_mounts,
1686 struct mount, mnt_child);
1687 if (t->mnt_mp != p->mnt_mp)
1688 t = NULL;
1689 }
1690 attach_shadowed(q, parent, t);
Al Viro719ea2f2013-09-29 11:24:49 -04001691 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 }
1693 }
1694 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001695out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04001697 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001698 umount_tree(res, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001699 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001701 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702}
1703
David Howellsbe34d1a2012-06-25 12:55:18 +01001704/* Caller should check returned pointer for errors */
1705
Al Viro589ff872009-04-18 03:28:19 -04001706struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001707{
Al Virocb338d02011-11-24 20:55:08 -05001708 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001709 namespace_lock();
Eric W. Biedermancd4a4012015-01-07 14:28:26 -06001710 if (!check_mnt(real_mount(path->mnt)))
1711 tree = ERR_PTR(-EINVAL);
1712 else
1713 tree = copy_tree(real_mount(path->mnt), path->dentry,
1714 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001715 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001716 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001717 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001718 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001719}
1720
1721void drop_collected_mounts(struct vfsmount *mnt)
1722{
Al Viro97216be2013-03-16 15:12:40 -04001723 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001724 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001725 umount_tree(real_mount(mnt), UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001726 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04001727 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001728}
1729
Miklos Szeredic771d682014-10-24 00:14:36 +02001730/**
1731 * clone_private_mount - create a private clone of a path
1732 *
1733 * This creates a new vfsmount, which will be the clone of @path. The new will
1734 * not be attached anywhere in the namespace and will be private (i.e. changes
1735 * to the originating mount won't be propagated into this).
1736 *
1737 * Release with mntput().
1738 */
1739struct vfsmount *clone_private_mount(struct path *path)
1740{
1741 struct mount *old_mnt = real_mount(path->mnt);
1742 struct mount *new_mnt;
1743
1744 if (IS_MNT_UNBINDABLE(old_mnt))
1745 return ERR_PTR(-EINVAL);
1746
1747 down_read(&namespace_sem);
1748 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1749 up_read(&namespace_sem);
1750 if (IS_ERR(new_mnt))
1751 return ERR_CAST(new_mnt);
1752
1753 return &new_mnt->mnt;
1754}
1755EXPORT_SYMBOL_GPL(clone_private_mount);
1756
Al Viro1f707132010-01-30 22:51:25 -05001757int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1758 struct vfsmount *root)
1759{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001760 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001761 int res = f(root, arg);
1762 if (res)
1763 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001764 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1765 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001766 if (res)
1767 return res;
1768 }
1769 return 0;
1770}
1771
Al Viro4b8b21f2011-11-24 19:54:23 -05001772static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001773{
Al Viro315fc832011-11-24 18:57:30 -05001774 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001775
Al Viro909b0a82011-11-25 03:06:56 -05001776 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001777 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001778 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001779 }
1780}
1781
Al Viro4b8b21f2011-11-24 19:54:23 -05001782static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001783{
Al Viro315fc832011-11-24 18:57:30 -05001784 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001785
Al Viro909b0a82011-11-25 03:06:56 -05001786 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001787 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001788 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001789 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001790 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001791 return err;
1792 }
1793 }
1794 }
1795
1796 return 0;
1797}
1798
Ram Paib90fa9a2005-11-07 17:19:50 -05001799/*
1800 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001801 * @nd : place the mount tree @source_mnt is attached
1802 * @parent_nd : if non-null, detach the source_mnt from its parent and
1803 * store the parent mount and mountpoint dentry.
1804 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001805 *
1806 * NOTE: in the table below explains the semantics when a source mount
1807 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001808 * ---------------------------------------------------------------------------
1809 * | BIND MOUNT OPERATION |
1810 * |**************************************************************************
1811 * | source-->| shared | private | slave | unbindable |
1812 * | dest | | | | |
1813 * | | | | | | |
1814 * | v | | | | |
1815 * |**************************************************************************
1816 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1817 * | | | | | |
1818 * |non-shared| shared (+) | private | slave (*) | invalid |
1819 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001820 * A bind operation clones the source mount and mounts the clone on the
1821 * destination mount.
1822 *
1823 * (++) the cloned mount is propagated to all the mounts in the propagation
1824 * tree of the destination mount and the cloned mount is added to
1825 * the peer group of the source mount.
1826 * (+) the cloned mount is created under the destination mount and is marked
1827 * as shared. The cloned mount is added to the peer group of the source
1828 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001829 * (+++) the mount is propagated to all the mounts in the propagation tree
1830 * of the destination mount and the cloned mount is made slave
1831 * of the same master as that of the source mount. The cloned mount
1832 * is marked as 'shared and slave'.
1833 * (*) the cloned mount is made a slave of the same master as that of the
1834 * source mount.
1835 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001836 * ---------------------------------------------------------------------------
1837 * | MOVE MOUNT OPERATION |
1838 * |**************************************************************************
1839 * | source-->| shared | private | slave | unbindable |
1840 * | dest | | | | |
1841 * | | | | | | |
1842 * | v | | | | |
1843 * |**************************************************************************
1844 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1845 * | | | | | |
1846 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1847 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001848 *
1849 * (+) the mount is moved to the destination. And is then propagated to
1850 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001851 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001852 * (+++) the mount is moved to the destination and is then propagated to
1853 * all the mounts belonging to the destination mount's propagation tree.
1854 * the mount is marked as 'shared and slave'.
1855 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001856 *
1857 * if the source mount is a tree, the operations explained above is
1858 * applied to each mount in the tree.
1859 * Must be called without spinlocks held, since this function can sleep
1860 * in allocations.
1861 */
Al Viro0fb54e52011-11-24 19:59:16 -05001862static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d17192013-03-15 10:53:28 -04001863 struct mount *dest_mnt,
1864 struct mountpoint *dest_mp,
1865 struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001866{
Al Viro38129a12014-03-20 21:10:51 -04001867 HLIST_HEAD(tree_list);
Al Viro315fc832011-11-24 18:57:30 -05001868 struct mount *child, *p;
Al Viro38129a12014-03-20 21:10:51 -04001869 struct hlist_node *n;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001870 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001871
Al Virofc7be132011-11-25 01:05:37 -05001872 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001873 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001874 if (err)
1875 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04001876 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Al Virof2ebb3a2014-02-27 09:35:45 -05001877 lock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001878 if (err)
1879 goto out_cleanup_ids;
Al Viro909b0a82011-11-25 03:06:56 -05001880 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001881 set_mnt_shared(p);
Al Viro0b1b9012014-03-21 10:14:08 -04001882 } else {
1883 lock_mount_hash();
Ram Paib90fa9a2005-11-07 17:19:50 -05001884 }
Al Viro1a390682008-03-21 20:48:19 -04001885 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001886 detach_mnt(source_mnt, parent_path);
Al Viro84d17192013-03-15 10:53:28 -04001887 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05001888 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001889 } else {
Al Viro84d17192013-03-15 10:53:28 -04001890 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Al Viro1d6a32a2014-03-20 20:34:43 -04001891 commit_tree(source_mnt, NULL);
Ram Pai21444402005-11-07 17:20:03 -05001892 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001893
Al Viro38129a12014-03-20 21:10:51 -04001894 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04001895 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04001896 hlist_del_init(&child->mnt_hash);
Al Viro1d6a32a2014-03-20 20:34:43 -04001897 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1898 child->mnt_mountpoint);
1899 commit_tree(child, q);
Ram Paib90fa9a2005-11-07 17:19:50 -05001900 }
Al Viro719ea2f2013-09-29 11:24:49 -04001901 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10001902
Ram Paib90fa9a2005-11-07 17:19:50 -05001903 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001904
1905 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05001906 while (!hlist_empty(&tree_list)) {
1907 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
Eric W. Biedermane819f152014-12-24 07:20:01 -06001908 umount_tree(child, UMOUNT_SYNC);
Al Virof2ebb3a2014-02-27 09:35:45 -05001909 }
1910 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001911 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001912 out:
1913 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001914}
1915
Al Viro84d17192013-03-15 10:53:28 -04001916static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04001917{
1918 struct vfsmount *mnt;
Al Viro84d17192013-03-15 10:53:28 -04001919 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04001920retry:
Al Viro84d17192013-03-15 10:53:28 -04001921 mutex_lock(&dentry->d_inode->i_mutex);
1922 if (unlikely(cant_mount(dentry))) {
1923 mutex_unlock(&dentry->d_inode->i_mutex);
1924 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04001925 }
Al Viro97216be2013-03-16 15:12:40 -04001926 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04001927 mnt = lookup_mnt(path);
Al Viro84d17192013-03-15 10:53:28 -04001928 if (likely(!mnt)) {
Eric W. Biedermane2dfa932014-02-24 17:32:34 -08001929 struct mountpoint *mp = lookup_mountpoint(dentry);
1930 if (!mp)
1931 mp = new_mountpoint(dentry);
Al Viro84d17192013-03-15 10:53:28 -04001932 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04001933 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001934 mutex_unlock(&dentry->d_inode->i_mutex);
1935 return mp;
1936 }
1937 return mp;
1938 }
Al Viro97216be2013-03-16 15:12:40 -04001939 namespace_unlock();
Al Virob12cea92011-03-18 08:55:38 -04001940 mutex_unlock(&path->dentry->d_inode->i_mutex);
1941 path_put(path);
1942 path->mnt = mnt;
Al Viro84d17192013-03-15 10:53:28 -04001943 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04001944 goto retry;
1945}
1946
Al Viro84d17192013-03-15 10:53:28 -04001947static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04001948{
Al Viro84d17192013-03-15 10:53:28 -04001949 struct dentry *dentry = where->m_dentry;
1950 put_mountpoint(where);
Al Viro328e6d92013-03-16 14:49:45 -04001951 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001952 mutex_unlock(&dentry->d_inode->i_mutex);
Al Virob12cea92011-03-18 08:55:38 -04001953}
1954
Al Viro84d17192013-03-15 10:53:28 -04001955static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956{
Al Viro95bc5f22011-11-25 00:30:56 -05001957 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 return -EINVAL;
1959
David Howellse36cb0b2015-01-29 12:02:35 +00001960 if (d_is_dir(mp->m_dentry) !=
1961 d_is_dir(mnt->mnt.mnt_root))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 return -ENOTDIR;
1963
Al Viro84d17192013-03-15 10:53:28 -04001964 return attach_recursive_mnt(mnt, p, mp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965}
1966
1967/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001968 * Sanity check the flags to change_mnt_propagation.
1969 */
1970
1971static int flags_to_propagation_type(int flags)
1972{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001973 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001974
1975 /* Fail if any non-propagation flags are set */
1976 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1977 return 0;
1978 /* Only one propagation flag should be set */
1979 if (!is_power_of_2(type))
1980 return 0;
1981 return type;
1982}
1983
1984/*
Ram Pai07b20882005-11-07 17:19:07 -05001985 * recursively change the type of the mountpoint.
1986 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001987static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001988{
Al Viro315fc832011-11-24 18:57:30 -05001989 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001990 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001991 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001992 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001993 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001994
Al Viro2d92ab32008-08-02 00:51:11 -04001995 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001996 return -EINVAL;
1997
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001998 type = flags_to_propagation_type(flag);
1999 if (!type)
2000 return -EINVAL;
2001
Al Viro97216be2013-03-16 15:12:40 -04002002 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002003 if (type == MS_SHARED) {
2004 err = invent_group_ids(mnt, recurse);
2005 if (err)
2006 goto out_unlock;
2007 }
2008
Al Viro719ea2f2013-09-29 11:24:49 -04002009 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05002010 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05002011 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04002012 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002013
2014 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04002015 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002016 return err;
Ram Pai07b20882005-11-07 17:19:07 -05002017}
2018
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002019static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2020{
2021 struct mount *child;
2022 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2023 if (!is_subdir(child->mnt_mountpoint, dentry))
2024 continue;
2025
2026 if (child->mnt.mnt_flags & MNT_LOCKED)
2027 return true;
2028 }
2029 return false;
2030}
2031
Ram Pai07b20882005-11-07 17:19:07 -05002032/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 * do loopback mount.
2034 */
Al Viro808d4e32012-10-11 11:42:01 -04002035static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08002036 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037{
Al Viro2d92ab32008-08-02 00:51:11 -04002038 struct path old_path;
Al Viro84d17192013-03-15 10:53:28 -04002039 struct mount *mnt = NULL, *old, *parent;
2040 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002041 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 if (!old_name || !*old_name)
2043 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04002044 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 if (err)
2046 return err;
2047
Eric W. Biederman8823c072010-03-07 18:49:36 -08002048 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002049 if (mnt_ns_loop(old_path.dentry))
Eric W. Biederman8823c072010-03-07 18:49:36 -08002050 goto out;
2051
Al Viro84d17192013-03-15 10:53:28 -04002052 mp = lock_mount(path);
2053 err = PTR_ERR(mp);
2054 if (IS_ERR(mp))
Jan Blunck4ac91372008-02-14 19:34:32 -08002055 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05002056
Al Viro87129cc2011-11-24 21:24:27 -05002057 old = real_mount(old_path.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002058 parent = real_mount(path->mnt);
Al Viro87129cc2011-11-24 21:24:27 -05002059
Al Virob12cea92011-03-18 08:55:38 -04002060 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002061 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04002062 goto out2;
2063
Al Viroe149ed22014-11-01 10:57:28 -04002064 if (!check_mnt(parent))
2065 goto out2;
2066
2067 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
Al Virob12cea92011-03-18 08:55:38 -04002068 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002070 if (!recurse && has_locked_children(old, old_path.dentry))
2071 goto out2;
2072
Al Viroccd48bc2005-11-07 17:15:04 -05002073 if (recurse)
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002074 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
Al Viroccd48bc2005-11-07 17:15:04 -05002075 else
Al Viro87129cc2011-11-24 21:24:27 -05002076 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05002077
David Howellsbe34d1a2012-06-25 12:55:18 +01002078 if (IS_ERR(mnt)) {
2079 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04002080 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01002081 }
Al Viroccd48bc2005-11-07 17:15:04 -05002082
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002083 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2084
Al Viro84d17192013-03-15 10:53:28 -04002085 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002086 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002087 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06002088 umount_tree(mnt, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04002089 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05002090 }
Al Virob12cea92011-03-18 08:55:38 -04002091out2:
Al Viro84d17192013-03-15 10:53:28 -04002092 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002093out:
Al Viro2d92ab32008-08-02 00:51:11 -04002094 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 return err;
2096}
2097
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002098static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2099{
2100 int error = 0;
2101 int readonly_request = 0;
2102
2103 if (ms_flags & MS_RDONLY)
2104 readonly_request = 1;
2105 if (readonly_request == __mnt_is_readonly(mnt))
2106 return 0;
2107
2108 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05002109 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002110 else
Al Viro83adc752011-11-24 22:37:54 -05002111 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002112 return error;
2113}
2114
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115/*
2116 * change filesystem flags. dir should be a physical root of filesystem.
2117 * If you've mounted a non-root directory somewhere and want to do remount
2118 * on it - tough luck.
2119 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002120static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 void *data)
2122{
2123 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04002124 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05002125 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
Al Viro143c8c92011-11-25 00:46:35 -05002127 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 return -EINVAL;
2129
Al Viro2d92ab32008-08-02 00:51:11 -04002130 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 return -EINVAL;
2132
Eric W. Biederman07b64552014-07-28 17:10:56 -07002133 /* Don't allow changing of locked mnt flags.
2134 *
2135 * No locks need to be held here while testing the various
2136 * MNT_LOCK flags because those flags can never be cleared
2137 * once they are set.
2138 */
2139 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2140 !(mnt_flags & MNT_READONLY)) {
2141 return -EPERM;
2142 }
Eric W. Biederman9566d672014-07-28 17:26:07 -07002143 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2144 !(mnt_flags & MNT_NODEV)) {
Eric W. Biederman3e186642014-08-13 01:33:38 -07002145 /* Was the nodev implicitly added in mount? */
2146 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2147 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2148 mnt_flags |= MNT_NODEV;
2149 } else {
2150 return -EPERM;
2151 }
Eric W. Biederman9566d672014-07-28 17:26:07 -07002152 }
2153 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2154 !(mnt_flags & MNT_NOSUID)) {
2155 return -EPERM;
2156 }
2157 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2158 !(mnt_flags & MNT_NOEXEC)) {
2159 return -EPERM;
2160 }
2161 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2162 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2163 return -EPERM;
2164 }
2165
Eric Parisff36fe22011-03-03 16:09:14 -05002166 err = security_sb_remount(sb, data);
2167 if (err)
2168 return err;
2169
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002171 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002172 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05002173 else if (!capable(CAP_SYS_ADMIN))
2174 err = -EPERM;
Al Viro4aa98cf2009-05-08 13:36:58 -04002175 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002176 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05002177 if (!err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002178 lock_mount_hash();
Eric W. Biedermana6138db2014-07-28 16:26:53 -07002179 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
Al Viro143c8c92011-11-25 00:46:35 -05002180 mnt->mnt.mnt_flags = mnt_flags;
Al Viro143c8c92011-11-25 00:46:35 -05002181 touch_mnt_namespace(mnt->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04002182 unlock_mount_hash();
Dan Williams0e55a7c2008-09-26 19:01:20 -07002183 }
Al Viro6339dab2013-09-16 22:41:01 -04002184 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 return err;
2186}
2187
Al Virocbbe3622011-11-24 20:01:19 -05002188static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05002189{
Al Viro315fc832011-11-24 18:57:30 -05002190 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05002191 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002192 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05002193 return 1;
2194 }
2195 return 0;
2196}
2197
Al Viro808d4e32012-10-11 11:42:01 -04002198static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199{
Al Viro2d92ab32008-08-02 00:51:11 -04002200 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05002201 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05002202 struct mount *old;
Al Viro84d17192013-03-15 10:53:28 -04002203 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002204 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 if (!old_name || !*old_name)
2206 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002207 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 if (err)
2209 return err;
2210
Al Viro84d17192013-03-15 10:53:28 -04002211 mp = lock_mount(path);
2212 err = PTR_ERR(mp);
2213 if (IS_ERR(mp))
David Howellscc53ce52011-01-14 18:45:26 +00002214 goto out;
2215
Al Viro143c8c92011-11-25 00:46:35 -05002216 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002217 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05002218
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002220 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 goto out1;
2222
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002223 if (old->mnt.mnt_flags & MNT_LOCKED)
2224 goto out1;
2225
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002227 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05002228 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
Al Viro676da582011-11-24 21:47:05 -05002230 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05002231 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232
David Howellse36cb0b2015-01-29 12:02:35 +00002233 if (d_is_dir(path->dentry) !=
2234 d_is_dir(old_path.dentry))
Ram Pai21444402005-11-07 17:20:03 -05002235 goto out1;
2236 /*
2237 * Don't move a mount residing in a shared parent.
2238 */
Al Virofc7be132011-11-25 01:05:37 -05002239 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002240 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05002241 /*
2242 * Don't move a mount tree containing unbindable mounts to a destination
2243 * mount which is shared.
2244 */
Al Virofc7be132011-11-25 01:05:37 -05002245 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05002246 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05002248 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05002249 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05002250 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251
Al Viro84d17192013-03-15 10:53:28 -04002252 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08002253 if (err)
Ram Pai21444402005-11-07 17:20:03 -05002254 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255
2256 /* if the mount is moved, it should no longer be expire
2257 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05002258 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259out1:
Al Viro84d17192013-03-15 10:53:28 -04002260 unlock_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04002263 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04002264 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 return err;
2266}
2267
Al Viro9d412a42011-03-17 22:08:28 -04002268static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2269{
2270 int err;
2271 const char *subtype = strchr(fstype, '.');
2272 if (subtype) {
2273 subtype++;
2274 err = -EINVAL;
2275 if (!subtype[0])
2276 goto err;
2277 } else
2278 subtype = "";
2279
2280 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2281 err = -ENOMEM;
2282 if (!mnt->mnt_sb->s_subtype)
2283 goto err;
2284 return mnt;
2285
2286 err:
2287 mntput(mnt);
2288 return ERR_PTR(err);
2289}
2290
Al Viro9d412a42011-03-17 22:08:28 -04002291/*
2292 * add a mount into a namespace's mount tree
2293 */
Al Viro95bc5f22011-11-25 00:30:56 -05002294static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04002295{
Al Viro84d17192013-03-15 10:53:28 -04002296 struct mountpoint *mp;
2297 struct mount *parent;
Al Viro9d412a42011-03-17 22:08:28 -04002298 int err;
2299
Al Virof2ebb3a2014-02-27 09:35:45 -05002300 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04002301
Al Viro84d17192013-03-15 10:53:28 -04002302 mp = lock_mount(path);
2303 if (IS_ERR(mp))
2304 return PTR_ERR(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002305
Al Viro84d17192013-03-15 10:53:28 -04002306 parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04002307 err = -EINVAL;
Al Viro84d17192013-03-15 10:53:28 -04002308 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04002309 /* that's acceptable only for automounts done in private ns */
2310 if (!(mnt_flags & MNT_SHRINKABLE))
2311 goto unlock;
2312 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d17192013-03-15 10:53:28 -04002313 if (!parent->mnt_ns)
Al Viro156cacb2012-09-21 08:19:02 -04002314 goto unlock;
2315 }
Al Viro9d412a42011-03-17 22:08:28 -04002316
2317 /* Refuse the same filesystem on the same mount point */
2318 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05002319 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04002320 path->mnt->mnt_root == path->dentry)
2321 goto unlock;
2322
2323 err = -EINVAL;
David Howellse36cb0b2015-01-29 12:02:35 +00002324 if (d_is_symlink(newmnt->mnt.mnt_root))
Al Viro9d412a42011-03-17 22:08:28 -04002325 goto unlock;
2326
Al Viro95bc5f22011-11-25 00:30:56 -05002327 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro84d17192013-03-15 10:53:28 -04002328 err = graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04002329
2330unlock:
Al Viro84d17192013-03-15 10:53:28 -04002331 unlock_mount(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002332 return err;
2333}
Al Virob1e75df2011-01-17 01:47:59 -05002334
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05002335static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05002336
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337/*
2338 * create a new mount for userspace and request it to be added into the
2339 * namespace's tree
2340 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002341static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04002342 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002344 struct file_system_type *type;
Al Viro9b40bc92013-02-22 22:45:42 -05002345 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002347 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002349 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 return -EINVAL;
2351
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002352 type = get_fs_type(fstype);
2353 if (!type)
2354 return -ENODEV;
2355
2356 if (user_ns != &init_user_ns) {
2357 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2358 put_filesystem(type);
2359 return -EPERM;
2360 }
2361 /* Only in special cases allow devices from mounts
2362 * created outside the initial user namespace.
2363 */
2364 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2365 flags |= MS_NODEV;
Eric W. Biederman9566d672014-07-28 17:26:07 -07002366 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002367 }
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05002368 if (type->fs_flags & FS_USERNS_VISIBLE) {
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05002369 if (!fs_fully_visible(type, &mnt_flags))
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05002370 return -EPERM;
2371 }
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002372 }
2373
2374 mnt = vfs_kern_mount(type, flags, name, data);
2375 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2376 !mnt->mnt_sb->s_subtype)
2377 mnt = fs_set_subtype(mnt, fstype);
2378
2379 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 if (IS_ERR(mnt))
2381 return PTR_ERR(mnt);
2382
Al Viro95bc5f22011-11-25 00:30:56 -05002383 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002384 if (err)
2385 mntput(mnt);
2386 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387}
2388
Al Viro19a167a2011-01-17 01:35:23 -05002389int finish_automount(struct vfsmount *m, struct path *path)
2390{
Al Viro6776db3d2011-11-25 00:22:05 -05002391 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002392 int err;
2393 /* The new mount record should have at least 2 refs to prevent it being
2394 * expired before we get a chance to add it
2395 */
Al Viro6776db3d2011-11-25 00:22:05 -05002396 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002397
2398 if (m->mnt_sb == path->mnt->mnt_sb &&
2399 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002400 err = -ELOOP;
2401 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002402 }
2403
Al Viro95bc5f22011-11-25 00:30:56 -05002404 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002405 if (!err)
2406 return 0;
2407fail:
2408 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002409 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002410 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05002411 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04002412 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002413 }
Al Virob1e75df2011-01-17 01:47:59 -05002414 mntput(m);
2415 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002416 return err;
2417}
2418
David Howellsea5b7782011-01-14 19:10:03 +00002419/**
2420 * mnt_set_expiry - Put a mount on an expiration list
2421 * @mnt: The mount to list.
2422 * @expiry_list: The list to add the mount to.
2423 */
2424void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2425{
Al Viro97216be2013-03-16 15:12:40 -04002426 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00002427
Al Viro6776db3d2011-11-25 00:22:05 -05002428 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002429
Al Viro97216be2013-03-16 15:12:40 -04002430 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002431}
2432EXPORT_SYMBOL(mnt_set_expiry);
2433
2434/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 * process a list of expirable mountpoints with the intent of discarding any
2436 * mountpoints that aren't in use and haven't been touched since last we came
2437 * here
2438 */
2439void mark_mounts_for_expiry(struct list_head *mounts)
2440{
Al Viro761d5c32011-11-24 21:07:43 -05002441 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 LIST_HEAD(graveyard);
2443
2444 if (list_empty(mounts))
2445 return;
2446
Al Viro97216be2013-03-16 15:12:40 -04002447 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002448 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
2450 /* extract from the expiration list every vfsmount that matches the
2451 * following criteria:
2452 * - only referenced by its parent vfsmount
2453 * - still marked for expiry (marked on the last call here; marks are
2454 * cleared by mntput())
2455 */
Al Viro6776db3d2011-11-25 00:22:05 -05002456 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002457 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002458 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002460 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002462 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002463 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002464 touch_mnt_namespace(mnt->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06002465 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04002466 }
Al Viro719ea2f2013-09-29 11:24:49 -04002467 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002468 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469}
2470
2471EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2472
2473/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002474 * Ripoff of 'select_parent()'
2475 *
2476 * search the list of submounts for a given mountpoint, and move any
2477 * shrinkable submounts to the 'graveyard' list.
2478 */
Al Viro692afc32011-11-24 21:15:14 -05002479static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002480{
Al Viro692afc32011-11-24 21:15:14 -05002481 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002482 struct list_head *next;
2483 int found = 0;
2484
2485repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002486 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002487resume:
Al Viro6b41d532011-11-24 23:24:33 -05002488 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002489 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002490 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002491
2492 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002493 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002494 continue;
2495 /*
2496 * Descend a level if the d_mounts list is non-empty.
2497 */
Al Viro6b41d532011-11-24 23:24:33 -05002498 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002499 this_parent = mnt;
2500 goto repeat;
2501 }
2502
Al Viro1ab59732011-11-24 21:35:16 -05002503 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002504 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002505 found++;
2506 }
2507 }
2508 /*
2509 * All done at this level ... ascend and resume the search
2510 */
2511 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002512 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002513 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002514 goto resume;
2515 }
2516 return found;
2517}
2518
2519/*
2520 * process a list of expirable mountpoints with the intent of discarding any
2521 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002522 *
Al Viro48a066e2013-09-29 22:06:07 -04002523 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002524 */
Al Virob54b9be2013-03-16 14:39:34 -04002525static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002526{
2527 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002528 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002529
Trond Myklebust5528f9112006-06-09 09:34:17 -04002530 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002531 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002532 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002533 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002534 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002535 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06002536 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04002537 }
2538 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002539}
2540
Trond Myklebust5528f9112006-06-09 09:34:17 -04002541/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 * Some copy_from_user() implementations do not return the exact number of
2543 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2544 * Note that this function differs from copy_from_user() in that it will oops
2545 * on bad values of `to', rather than returning a short copy.
2546 */
Ram Paib58fed82005-11-07 17:16:09 -05002547static long exact_copy_from_user(void *to, const void __user * from,
2548 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549{
2550 char *t = to;
2551 const char __user *f = from;
2552 char c;
2553
2554 if (!access_ok(VERIFY_READ, from, n))
2555 return n;
2556
2557 while (n) {
2558 if (__get_user(c, f)) {
2559 memset(t, 0, n);
2560 break;
2561 }
2562 *t++ = c;
2563 f++;
2564 n--;
2565 }
2566 return n;
2567}
2568
Ram Paib58fed82005-11-07 17:16:09 -05002569int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570{
2571 int i;
2572 unsigned long page;
2573 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002574
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 *where = 0;
2576 if (!data)
2577 return 0;
2578
2579 if (!(page = __get_free_page(GFP_KERNEL)))
2580 return -ENOMEM;
2581
2582 /* We only care that *some* data at the address the user
2583 * gave us is valid. Just in case, we'll zero
2584 * the remainder of the page.
2585 */
2586 /* copy_from_user cannot cross TASK_SIZE ! */
2587 size = TASK_SIZE - (unsigned long)data;
2588 if (size > PAGE_SIZE)
2589 size = PAGE_SIZE;
2590
2591 i = size - exact_copy_from_user((void *)page, data, size);
2592 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002593 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 return -EFAULT;
2595 }
2596 if (i != PAGE_SIZE)
2597 memset((char *)page + i, 0, PAGE_SIZE - i);
2598 *where = page;
2599 return 0;
2600}
2601
Tim Gardnerb8850d12014-08-28 11:26:03 -06002602char *copy_mount_string(const void __user *data)
Vegard Nossumeca6f532009-09-18 13:05:45 -07002603{
Tim Gardnerb8850d12014-08-28 11:26:03 -06002604 return data ? strndup_user(data, PAGE_SIZE) : NULL;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002605}
2606
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607/*
2608 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2609 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2610 *
2611 * data is a (void *) that can point to any structure up to
2612 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2613 * information (or be NULL).
2614 *
2615 * Pre-0.97 versions of mount() didn't have a flags word.
2616 * When the flags word was introduced its top half was required
2617 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2618 * Therefore, if this magic number is present, it carries no information
2619 * and must be discarded.
2620 */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002621long do_mount(const char *dev_name, const char __user *dir_name,
Al Viro808d4e32012-10-11 11:42:01 -04002622 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623{
Al Viro2d92ab32008-08-02 00:51:11 -04002624 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 int retval = 0;
2626 int mnt_flags = 0;
2627
2628 /* Discard magic */
2629 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2630 flags &= ~MS_MGC_MSK;
2631
2632 /* Basic sanity checks */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 if (data_page)
2634 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2635
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002636 /* ... and get the mountpoint */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002637 retval = user_path(dir_name, &path);
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002638 if (retval)
2639 return retval;
2640
2641 retval = security_sb_mount(dev_name, &path,
2642 type_page, flags, data_page);
Al Viro0d5cadb2013-05-04 14:40:51 -04002643 if (!retval && !may_mount())
2644 retval = -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002645 if (retval)
2646 goto dput_out;
2647
Andi Kleen613cbe32009-04-19 18:40:43 +02002648 /* Default to relatime unless overriden */
2649 if (!(flags & MS_NOATIME))
2650 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002651
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 /* Separate the per-mountpoint flags */
2653 if (flags & MS_NOSUID)
2654 mnt_flags |= MNT_NOSUID;
2655 if (flags & MS_NODEV)
2656 mnt_flags |= MNT_NODEV;
2657 if (flags & MS_NOEXEC)
2658 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002659 if (flags & MS_NOATIME)
2660 mnt_flags |= MNT_NOATIME;
2661 if (flags & MS_NODIRATIME)
2662 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002663 if (flags & MS_STRICTATIME)
2664 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002665 if (flags & MS_RDONLY)
2666 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002667
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07002668 /* The default atime for remount is preservation */
2669 if ((flags & MS_REMOUNT) &&
2670 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2671 MS_STRICTATIME)) == 0)) {
2672 mnt_flags &= ~MNT_ATIME_MASK;
2673 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2674 }
2675
Al Viro7a4dec52010-08-09 12:05:43 -04002676 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002677 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2678 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002681 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 data_page);
2683 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002684 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002685 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002686 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002688 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 else
Al Viro2d92ab32008-08-02 00:51:11 -04002690 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 dev_name, data_page);
2692dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002693 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 return retval;
2695}
2696
Eric W. Biederman771b1372012-07-26 21:08:32 -07002697static void free_mnt_ns(struct mnt_namespace *ns)
2698{
Al Viro6344c432014-11-01 00:45:45 -04002699 ns_free_inum(&ns->ns);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002700 put_user_ns(ns->user_ns);
2701 kfree(ns);
2702}
2703
Eric W. Biederman8823c072010-03-07 18:49:36 -08002704/*
2705 * Assign a sequence number so we can detect when we attempt to bind
2706 * mount a reference to an older mount namespace into the current
2707 * mount namespace, preventing reference counting loops. A 64bit
2708 * number incrementing at 10Ghz will take 12,427 years to wrap which
2709 * is effectively never, so we can ignore the possibility.
2710 */
2711static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2712
Eric W. Biederman771b1372012-07-26 21:08:32 -07002713static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002714{
2715 struct mnt_namespace *new_ns;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002716 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002717
2718 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2719 if (!new_ns)
2720 return ERR_PTR(-ENOMEM);
Al Viro6344c432014-11-01 00:45:45 -04002721 ret = ns_alloc_inum(&new_ns->ns);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002722 if (ret) {
2723 kfree(new_ns);
2724 return ERR_PTR(ret);
2725 }
Al Viro33c42942014-11-01 02:32:53 -04002726 new_ns->ns.ops = &mntns_operations;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002727 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002728 atomic_set(&new_ns->count, 1);
2729 new_ns->root = NULL;
2730 INIT_LIST_HEAD(&new_ns->list);
2731 init_waitqueue_head(&new_ns->poll);
2732 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002733 new_ns->user_ns = get_user_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002734 return new_ns;
2735}
2736
Al Viro9559f682013-09-28 20:47:57 -04002737struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2738 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002740 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002741 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002742 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04002743 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05002744 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002745 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746
Al Viro9559f682013-09-28 20:47:57 -04002747 BUG_ON(!ns);
2748
2749 if (likely(!(flags & CLONE_NEWNS))) {
2750 get_mnt_ns(ns);
2751 return ns;
2752 }
2753
2754 old = ns->root;
2755
Eric W. Biederman771b1372012-07-26 21:08:32 -07002756 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002757 if (IS_ERR(new_ns))
2758 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759
Al Viro97216be2013-03-16 15:12:40 -04002760 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002762 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04002763 if (user_ns != ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002764 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002765 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002766 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04002767 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002768 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002769 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 }
Al Virobe08d6d2011-12-06 13:32:36 -05002771 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002772 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
2774 /*
2775 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2776 * as belonging to new namespace. We have already acquired a private
2777 * fs_struct, so tsk->fs->lock is not needed.
2778 */
Al Viro909b0a82011-11-25 03:06:56 -05002779 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002780 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002782 q->mnt_ns = new_ns;
Al Viro9559f682013-09-28 20:47:57 -04002783 if (new_fs) {
2784 if (&p->mnt == new_fs->root.mnt) {
2785 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002786 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 }
Al Viro9559f682013-09-28 20:47:57 -04002788 if (&p->mnt == new_fs->pwd.mnt) {
2789 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002790 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 }
Al Viro909b0a82011-11-25 03:06:56 -05002793 p = next_mnt(p, old);
2794 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002795 if (!q)
2796 break;
2797 while (p->mnt.mnt_root != q->mnt.mnt_root)
2798 p = next_mnt(p, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 }
Al Viro328e6d92013-03-16 14:49:45 -04002800 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002803 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002805 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806
JANAK DESAI741a2952006-02-07 12:59:00 -08002807 return new_ns;
2808}
2809
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002810/**
2811 * create_mnt_ns - creates a private namespace and adds a root filesystem
2812 * @mnt: pointer to the new root filesystem mountpoint
2813 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002814static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002815{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002816 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002817 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002818 struct mount *mnt = real_mount(m);
2819 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002820 new_ns->root = mnt;
Al Virob1983cd2013-05-04 15:18:53 -04002821 list_add(&mnt->mnt_list, &new_ns->list);
Al Viroc1334492011-11-16 16:12:14 -05002822 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002823 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002824 }
2825 return new_ns;
2826}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002827
Al Viroea441d12011-11-16 21:43:59 -05002828struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2829{
2830 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002831 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002832 struct path path;
2833 int err;
2834
2835 ns = create_mnt_ns(mnt);
2836 if (IS_ERR(ns))
2837 return ERR_CAST(ns);
2838
2839 err = vfs_path_lookup(mnt->mnt_root, mnt,
2840 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2841
2842 put_mnt_ns(ns);
2843
2844 if (err)
2845 return ERR_PTR(err);
2846
2847 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002848 s = path.mnt->mnt_sb;
2849 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002850 mntput(path.mnt);
2851 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002852 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002853 /* ... and return the root of (sub)tree on it */
2854 return path.dentry;
2855}
2856EXPORT_SYMBOL(mount_subtree);
2857
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002858SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2859 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002861 int ret;
2862 char *kernel_type;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002863 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865
Tim Gardnerb8850d12014-08-28 11:26:03 -06002866 kernel_type = copy_mount_string(type);
2867 ret = PTR_ERR(kernel_type);
2868 if (IS_ERR(kernel_type))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002869 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870
Tim Gardnerb8850d12014-08-28 11:26:03 -06002871 kernel_dev = copy_mount_string(dev_name);
2872 ret = PTR_ERR(kernel_dev);
2873 if (IS_ERR(kernel_dev))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002874 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875
Vegard Nossumeca6f532009-09-18 13:05:45 -07002876 ret = copy_mount_options(data, &data_page);
2877 if (ret < 0)
2878 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002880 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
Vegard Nossumeca6f532009-09-18 13:05:45 -07002881 (void *) data_page);
2882
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002884out_data:
2885 kfree(kernel_dev);
2886out_dev:
Vegard Nossumeca6f532009-09-18 13:05:45 -07002887 kfree(kernel_type);
2888out_type:
2889 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890}
2891
2892/*
Al Viroafac7cb2011-11-23 19:34:49 -05002893 * Return true if path is reachable from root
2894 *
Al Viro48a066e2013-09-29 22:06:07 -04002895 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05002896 */
Al Viro643822b2011-11-24 22:00:28 -05002897bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002898 const struct path *root)
2899{
Al Viro643822b2011-11-24 22:00:28 -05002900 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002901 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002902 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002903 }
Al Viro643822b2011-11-24 22:00:28 -05002904 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002905}
2906
2907int path_is_under(struct path *path1, struct path *path2)
2908{
2909 int res;
Al Viro48a066e2013-09-29 22:06:07 -04002910 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002911 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04002912 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002913 return res;
2914}
2915EXPORT_SYMBOL(path_is_under);
2916
2917/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 * pivot_root Semantics:
2919 * Moves the root file system of the current process to the directory put_old,
2920 * makes new_root as the new root file system of the current process, and sets
2921 * root/cwd of all processes which had them on the current root to new_root.
2922 *
2923 * Restrictions:
2924 * The new_root and put_old must be directories, and must not be on the
2925 * same file system as the current process root. The put_old must be
2926 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2927 * pointed to by put_old must yield the same directory as new_root. No other
2928 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2929 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002930 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2931 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2932 * in this situation.
2933 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 * Notes:
2935 * - we don't move root/cwd if they are not at the root (reason: if something
2936 * cared enough to change them, it's probably wrong to force them elsewhere)
2937 * - it's okay to pick a root that isn't the root of a file system, e.g.
2938 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2939 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2940 * first.
2941 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002942SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2943 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944{
Al Viro2d8f3032008-07-22 09:59:21 -04002945 struct path new, old, parent_path, root_parent, root;
Al Viro84d17192013-03-15 10:53:28 -04002946 struct mount *new_mnt, *root_mnt, *old_mnt;
2947 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 int error;
2949
Al Viro9b40bc92013-02-22 22:45:42 -05002950 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 return -EPERM;
2952
Al Viro2d8f3032008-07-22 09:59:21 -04002953 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954 if (error)
2955 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956
Al Viro2d8f3032008-07-22 09:59:21 -04002957 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 if (error)
2959 goto out1;
2960
Al Viro2d8f3032008-07-22 09:59:21 -04002961 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002962 if (error)
2963 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002965 get_fs_root(current->fs, &root);
Al Viro84d17192013-03-15 10:53:28 -04002966 old_mp = lock_mount(&old);
2967 error = PTR_ERR(old_mp);
2968 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04002969 goto out3;
2970
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002972 new_mnt = real_mount(new.mnt);
2973 root_mnt = real_mount(root.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002974 old_mnt = real_mount(old.mnt);
2975 if (IS_MNT_SHARED(old_mnt) ||
Al Virofc7be132011-11-25 01:05:37 -05002976 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2977 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002978 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002979 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002980 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002981 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2982 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002984 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002985 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 error = -EBUSY;
Al Viro84d17192013-03-15 10:53:28 -04002987 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04002988 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002990 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002991 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002992 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002993 goto out4; /* not attached */
Al Viro84d17192013-03-15 10:53:28 -04002994 root_mp = root_mnt->mnt_mp;
Al Viro2d8f3032008-07-22 09:59:21 -04002995 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002996 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002997 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002998 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002999 /* make sure we can reach put_old from new_root */
Al Viro84d17192013-03-15 10:53:28 -04003000 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04003001 goto out4;
Eric W. Biederman0d082602014-10-08 10:42:27 -07003002 /* make certain new is below the root */
3003 if (!is_path_reachable(new_mnt, new.dentry, &root))
3004 goto out4;
Al Viro84d17192013-03-15 10:53:28 -04003005 root_mp->m_count++; /* pin it so it won't go away */
Al Viro719ea2f2013-09-29 11:24:49 -04003006 lock_mount_hash();
Al Viro419148d2011-11-24 19:41:16 -05003007 detach_mnt(new_mnt, &parent_path);
3008 detach_mnt(root_mnt, &root_parent);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003009 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3010 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3011 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3012 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003013 /* mount old root on put_old */
Al Viro84d17192013-03-15 10:53:28 -04003014 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08003015 /* mount new_root on / */
Al Viro84d17192013-03-15 10:53:28 -04003016 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003017 touch_mnt_namespace(current->nsproxy->mnt_ns);
Eric W. Biederman4fed6552014-10-08 10:42:57 -07003018 /* A moved mount should not expire automatically */
3019 list_del_init(&new_mnt->mnt_expire);
Al Viro719ea2f2013-09-29 11:24:49 -04003020 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04003021 chroot_fs_refs(&root, &new);
Al Viro84d17192013-03-15 10:53:28 -04003022 put_mountpoint(root_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04003024out4:
Al Viro84d17192013-03-15 10:53:28 -04003025 unlock_mount(old_mp);
Al Virob12cea92011-03-18 08:55:38 -04003026 if (!error) {
3027 path_put(&root_parent);
3028 path_put(&parent_path);
3029 }
3030out3:
Al Viro8c3ee422008-03-22 18:00:39 -04003031 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04003032out2:
Al Viro2d8f3032008-07-22 09:59:21 -04003033 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034out1:
Al Viro2d8f3032008-07-22 09:59:21 -04003035 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038}
3039
3040static void __init init_mount_tree(void)
3041{
3042 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003043 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08003044 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003045 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003047 type = get_fs_type("rootfs");
3048 if (!type)
3049 panic("Can't find rootfs type");
3050 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3051 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 if (IS_ERR(mnt))
3053 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11003054
Trond Myklebust3b22edc2009-06-23 17:29:49 -04003055 ns = create_mnt_ns(mnt);
3056 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003059 init_task.nsproxy->mnt_ns = ns;
3060 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061
Al Virobe08d6d2011-12-06 13:32:36 -05003062 root.mnt = mnt;
3063 root.dentry = mnt->mnt_root;
Eric W. Biedermanda362b02014-10-07 12:19:53 -07003064 mnt->mnt_flags |= MNT_LOCKED;
Jan Blunckac748a02008-02-14 19:34:39 -08003065
3066 set_fs_pwd(current->fs, &root);
3067 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068}
3069
Denis Cheng74bf17c2007-10-16 23:26:30 -07003070void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071{
Eric Dumazet13f14b42008-02-06 01:37:57 -08003072 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003073 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074
Al Viro7d6fec42011-11-23 12:14:10 -05003075 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09003076 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077
Al Viro0818bf22014-02-28 13:46:44 -05003078 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04003079 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05003080 mhash_entries, 19,
3081 0,
3082 &m_hash_shift, &m_hash_mask, 0, 0);
3083 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3084 sizeof(struct hlist_head),
3085 mphash_entries, 19,
3086 0,
3087 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088
Al Viro84d17192013-03-15 10:53:28 -04003089 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 panic("Failed to allocate mount hash table\n");
3091
Al Viro0818bf22014-02-28 13:46:44 -05003092 for (u = 0; u <= m_hash_mask; u++)
Al Viro38129a12014-03-20 21:10:51 -04003093 INIT_HLIST_HEAD(&mount_hashtable[u]);
Al Viro0818bf22014-02-28 13:46:44 -05003094 for (u = 0; u <= mp_hash_mask; u++)
3095 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096
Tejun Heo4b93dc92013-11-28 14:54:43 -05003097 kernfs_init();
3098
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003099 err = sysfs_init();
3100 if (err)
3101 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003102 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06003103 fs_kobj = kobject_create_and_add("fs", NULL);
3104 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003105 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 init_rootfs();
3107 init_mount_tree();
3108}
3109
Trond Myklebust616511d2009-06-22 15:09:13 -04003110void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111{
Al Virod498b252010-02-05 02:21:06 -05003112 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04003113 return;
Al Viro7b00ed62013-09-16 21:19:20 -04003114 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003115 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116}
Al Viro9d412a42011-03-17 22:08:28 -04003117
3118struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3119{
Tim Chen423e0ab2011-07-19 09:32:38 -07003120 struct vfsmount *mnt;
3121 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3122 if (!IS_ERR(mnt)) {
3123 /*
3124 * it is a longterm mount, don't release mnt until
3125 * we unmount before file sys is unregistered
3126 */
Al Virof7a99c52012-06-09 00:59:08 -04003127 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07003128 }
3129 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04003130}
3131EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07003132
3133void kern_unmount(struct vfsmount *mnt)
3134{
3135 /* release long term mount so mount point can be released */
3136 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04003137 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04003138 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07003139 mntput(mnt);
3140 }
3141}
3142EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05003143
3144bool our_mnt(struct vfsmount *mnt)
3145{
Al Viro143c8c92011-11-25 00:46:35 -05003146 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05003147}
Eric W. Biederman8823c072010-03-07 18:49:36 -08003148
Eric W. Biederman31515272013-03-15 01:45:51 -07003149bool current_chrooted(void)
3150{
3151 /* Does the current process have a non-standard root */
3152 struct path ns_root;
3153 struct path fs_root;
3154 bool chrooted;
3155
3156 /* Find the namespace root */
3157 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3158 ns_root.dentry = ns_root.mnt->mnt_root;
3159 path_get(&ns_root);
3160 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3161 ;
3162
3163 get_fs_root(current->fs, &fs_root);
3164
3165 chrooted = !path_equal(&fs_root, &ns_root);
3166
3167 path_put(&fs_root);
3168 path_put(&ns_root);
3169
3170 return chrooted;
3171}
3172
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003173static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003174{
3175 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003176 int new_flags = *new_mnt_flags;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003177 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07003178 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003179
Eric W. Biedermane51db732013-03-30 19:57:41 -07003180 if (unlikely(!ns))
3181 return false;
3182
Al Viro44bb4382013-09-16 21:37:36 -04003183 down_read(&namespace_sem);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003184 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07003185 struct mount *child;
3186 if (mnt->mnt.mnt_sb->s_type != type)
3187 continue;
3188
Eric W. Biederman7e96c1b2015-05-08 16:36:50 -05003189 /* This mount is not fully visible if it's root directory
3190 * is not the root directory of the filesystem.
3191 */
3192 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3193 continue;
3194
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003195 /* Verify the mount flags are equal to or more permissive
3196 * than the proposed new mount.
3197 */
3198 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
3199 !(new_flags & MNT_READONLY))
3200 continue;
3201 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
3202 !(new_flags & MNT_NODEV))
3203 continue;
3204 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
3205 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3206 continue;
3207
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003208 /* This mount is not fully visible if there are any
3209 * locked child mounts that cover anything except for
3210 * empty directories.
Eric W. Biedermane51db732013-03-30 19:57:41 -07003211 */
3212 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3213 struct inode *inode = child->mnt_mountpoint->d_inode;
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003214 /* Only worry about locked mounts */
3215 if (!(mnt->mnt.mnt_flags & MNT_LOCKED))
3216 continue;
Eric W. Biederman7236c852015-05-13 20:51:09 -05003217 /* Is the directory permanetly empty? */
3218 if (!is_empty_dir_inode(inode))
Eric W. Biedermane51db732013-03-30 19:57:41 -07003219 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003220 }
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003221 /* Preserve the locked attributes */
3222 *new_mnt_flags |= mnt->mnt.mnt_flags & (MNT_LOCK_READONLY | \
3223 MNT_LOCK_NODEV | \
3224 MNT_LOCK_ATIME);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003225 visible = true;
3226 goto found;
3227 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003228 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07003229found:
Al Viro44bb4382013-09-16 21:37:36 -04003230 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003231 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003232}
3233
Al Viro64964522014-11-01 00:37:32 -04003234static struct ns_common *mntns_get(struct task_struct *task)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003235{
Al Viro58be28252014-11-01 00:00:23 -04003236 struct ns_common *ns = NULL;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003237 struct nsproxy *nsproxy;
3238
Eric W. Biederman728dba32014-02-03 19:13:49 -08003239 task_lock(task);
3240 nsproxy = task->nsproxy;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003241 if (nsproxy) {
Al Viro58be28252014-11-01 00:00:23 -04003242 ns = &nsproxy->mnt_ns->ns;
3243 get_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003244 }
Eric W. Biederman728dba32014-02-03 19:13:49 -08003245 task_unlock(task);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003246
3247 return ns;
3248}
3249
Al Viro64964522014-11-01 00:37:32 -04003250static void mntns_put(struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003251{
Al Viro58be28252014-11-01 00:00:23 -04003252 put_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003253}
3254
Al Viro64964522014-11-01 00:37:32 -04003255static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003256{
3257 struct fs_struct *fs = current->fs;
Al Viro58be28252014-11-01 00:00:23 -04003258 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003259 struct path root;
3260
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003261 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -07003262 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3263 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08003264 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003265
3266 if (fs->users != 1)
3267 return -EINVAL;
3268
3269 get_mnt_ns(mnt_ns);
3270 put_mnt_ns(nsproxy->mnt_ns);
3271 nsproxy->mnt_ns = mnt_ns;
3272
3273 /* Find the root */
3274 root.mnt = &mnt_ns->root->mnt;
3275 root.dentry = mnt_ns->root->mnt.mnt_root;
3276 path_get(&root);
3277 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3278 ;
3279
3280 /* Update the pwd and root */
3281 set_fs_pwd(fs, &root);
3282 set_fs_root(fs, &root);
3283
3284 path_put(&root);
3285 return 0;
3286}
3287
3288const struct proc_ns_operations mntns_operations = {
3289 .name = "mnt",
3290 .type = CLONE_NEWNS,
3291 .get = mntns_get,
3292 .put = mntns_put,
3293 .install = mntns_install,
3294};