blob: 2b8aa15fd6dfa755368ddf21136eaab5192f1e28 [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 Szeredi4ed5e822011-11-21 12:11:31 +0100318static int mnt_is_readonly(struct vfsmount *mnt)
319{
320 if (mnt->mnt_sb->s_readonly_remount)
321 return 1;
322 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
323 smp_rmb();
324 return __mnt_is_readonly(mnt);
325}
326
Dave Hansen3d733632008-02-15 14:37:59 -0800327/*
Jan Karaeb04c282012-06-12 16:20:35 +0200328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink(). We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800332 */
Dave Hansen83660252008-02-15 14:37:30 -0800333/**
Jan Karaeb04c282012-06-12 16:20:35 +0200334 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500335 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800336 *
Jan Karaeb04c282012-06-12 16:20:35 +0200337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800342 */
Jan Karaeb04c282012-06-12 16:20:35 +0200343int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800344{
Al Viro83adc752011-11-24 22:37:54 -0500345 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800346 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800347
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000348 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100349 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000350 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100351 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000352 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 * incremented count after it has set MNT_WRITE_HOLD.
354 */
355 smp_mb();
Miao Xie1e755292012-11-16 17:23:50 +0800356 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000357 cpu_relax();
358 /*
359 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 * be set to match its requirements. So we must not load that until
361 * MNT_WRITE_HOLD is cleared.
362 */
363 smp_rmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100364 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100365 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800366 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800367 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000368 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200369
370 return ret;
371}
372
373/**
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
376 *
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success. When the write operation is
380 * finished, mnt_drop_write() must be called. This is effectively a refcount.
381 */
382int mnt_want_write(struct vfsmount *m)
383{
384 int ret;
385
386 sb_start_write(m->mnt_sb);
387 ret = __mnt_want_write(m);
388 if (ret)
389 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800390 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800391}
392EXPORT_SYMBOL_GPL(mnt_want_write);
393
394/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
397 *
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
402 *
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
405 */
406int mnt_clone_write(struct vfsmount *mnt)
407{
408 /* superblock may be r/o */
409 if (__mnt_is_readonly(mnt))
410 return -EROFS;
411 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500412 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000413 preempt_enable();
414 return 0;
415}
416EXPORT_SYMBOL_GPL(mnt_clone_write);
417
418/**
Jan Karaeb04c282012-06-12 16:20:35 +0200419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
421 *
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
424 */
425int __mnt_want_write_file(struct file *file)
426{
Al Viro83f936c2014-03-14 12:02:47 -0400427 if (!(file->f_mode & FMODE_WRITER))
Jan Karaeb04c282012-06-12 16:20:35 +0200428 return __mnt_want_write(file->f_path.mnt);
429 else
430 return mnt_clone_write(file->f_path.mnt);
431}
432
433/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000434 * mnt_want_write_file - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
436 *
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
439 */
440int mnt_want_write_file(struct file *file)
441{
Jan Karaeb04c282012-06-12 16:20:35 +0200442 int ret;
443
444 sb_start_write(file->f_path.mnt->mnt_sb);
445 ret = __mnt_want_write_file(file);
446 if (ret)
447 sb_end_write(file->f_path.mnt->mnt_sb);
448 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000449}
450EXPORT_SYMBOL_GPL(mnt_want_write_file);
451
452/**
Jan Karaeb04c282012-06-12 16:20:35 +0200453 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800454 * @mnt: the mount on which to give up write access
455 *
456 * Tells the low-level filesystem that we are done
457 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200458 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800459 */
Jan Karaeb04c282012-06-12 16:20:35 +0200460void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800461{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000462 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500463 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000464 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800465}
Jan Karaeb04c282012-06-12 16:20:35 +0200466
467/**
468 * mnt_drop_write - give up write access to a mount
469 * @mnt: the mount on which to give up write access
470 *
471 * Tells the low-level filesystem that we are done performing writes to it and
472 * also allows filesystem to be frozen again. Must be matched with
473 * mnt_want_write() call above.
474 */
475void mnt_drop_write(struct vfsmount *mnt)
476{
477 __mnt_drop_write(mnt);
478 sb_end_write(mnt->mnt_sb);
479}
Dave Hansen83660252008-02-15 14:37:30 -0800480EXPORT_SYMBOL_GPL(mnt_drop_write);
481
Jan Karaeb04c282012-06-12 16:20:35 +0200482void __mnt_drop_write_file(struct file *file)
483{
484 __mnt_drop_write(file->f_path.mnt);
485}
486
Al Viro2a79f172011-12-09 08:06:57 -0500487void mnt_drop_write_file(struct file *file)
488{
489 mnt_drop_write(file->f_path.mnt);
490}
491EXPORT_SYMBOL(mnt_drop_write_file);
492
Al Viro83adc752011-11-24 22:37:54 -0500493static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800494{
Dave Hansen3d733632008-02-15 14:37:59 -0800495 int ret = 0;
496
Al Viro719ea2f2013-09-29 11:24:49 -0400497 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500498 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000499 /*
500 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501 * should be visible before we do.
502 */
503 smp_mb();
504
505 /*
506 * With writers on hold, if this value is zero, then there are
507 * definitely no active writers (although held writers may subsequently
508 * increment the count, they'll have to wait, and decrement it after
509 * seeing MNT_READONLY).
510 *
511 * It is OK to have counter incremented on one CPU and decremented on
512 * another: the sum will add up correctly. The danger would be when we
513 * sum up each counter, if we read a counter before it is incremented,
514 * but then read another CPU's count which it has been subsequently
515 * decremented from -- we would see more decrements than we should.
516 * MNT_WRITE_HOLD protects against this scenario, because
517 * mnt_want_write first increments count, then smp_mb, then spins on
518 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519 * we're counting up here.
520 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100521 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000522 ret = -EBUSY;
523 else
Al Viro83adc752011-11-24 22:37:54 -0500524 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000525 /*
526 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527 * that become unheld will see MNT_READONLY.
528 */
529 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500530 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Al Viro719ea2f2013-09-29 11:24:49 -0400531 unlock_mount_hash();
Dave Hansen3d733632008-02-15 14:37:59 -0800532 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800533}
Dave Hansen83660252008-02-15 14:37:30 -0800534
Al Viro83adc752011-11-24 22:37:54 -0500535static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800536{
Al Viro719ea2f2013-09-29 11:24:49 -0400537 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500538 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Al Viro719ea2f2013-09-29 11:24:49 -0400539 unlock_mount_hash();
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800540}
541
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100542int sb_prepare_remount_readonly(struct super_block *sb)
543{
544 struct mount *mnt;
545 int err = 0;
546
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100547 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
548 if (atomic_long_read(&sb->s_remove_count))
549 return -EBUSY;
550
Al Viro719ea2f2013-09-29 11:24:49 -0400551 lock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100552 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
553 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
554 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
555 smp_mb();
556 if (mnt_get_writers(mnt) > 0) {
557 err = -EBUSY;
558 break;
559 }
560 }
561 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100562 if (!err && atomic_long_read(&sb->s_remove_count))
563 err = -EBUSY;
564
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100565 if (!err) {
566 sb->s_readonly_remount = 1;
567 smp_wmb();
568 }
569 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
570 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
571 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
572 }
Al Viro719ea2f2013-09-29 11:24:49 -0400573 unlock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100574
575 return err;
576}
577
Al Virob105e272011-11-24 20:38:33 -0500578static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800580 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000581#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500582 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000583#endif
Al Virob105e272011-11-24 20:38:33 -0500584 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585}
586
David Howells8ffcb322014-01-24 12:17:54 +0000587static void delayed_free_vfsmnt(struct rcu_head *head)
588{
589 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
590}
591
Al Viro48a066e2013-09-29 22:06:07 -0400592/* call under rcu_read_lock */
Al Viro294d71f2015-05-08 11:43:53 -0400593int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
Al Viro48a066e2013-09-29 22:06:07 -0400594{
595 struct mount *mnt;
596 if (read_seqretry(&mount_lock, seq))
Al Viro294d71f2015-05-08 11:43:53 -0400597 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400598 if (bastard == NULL)
Al Viro294d71f2015-05-08 11:43:53 -0400599 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400600 mnt = real_mount(bastard);
601 mnt_add_count(mnt, 1);
602 if (likely(!read_seqretry(&mount_lock, seq)))
Al Viro294d71f2015-05-08 11:43:53 -0400603 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400604 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605 mnt_add_count(mnt, -1);
Al Viro294d71f2015-05-08 11:43:53 -0400606 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400607 }
Al Viro294d71f2015-05-08 11:43:53 -0400608 return -1;
609}
610
611/* call under rcu_read_lock */
612bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
613{
614 int res = __legitimize_mnt(bastard, seq);
615 if (likely(!res))
616 return true;
617 if (unlikely(res < 0)) {
618 rcu_read_unlock();
619 mntput(bastard);
620 rcu_read_lock();
621 }
Al Viro48a066e2013-09-29 22:06:07 -0400622 return false;
623}
624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625/*
Al Viro474279d2013-10-01 16:11:26 -0400626 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400627 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 */
Al Viro474279d2013-10-01 16:11:26 -0400629struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Al Viro38129a12014-03-20 21:10:51 -0400631 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400632 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Al Viro38129a12014-03-20 21:10:51 -0400634 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400635 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
636 return p;
637 return NULL;
638}
639
640/*
641 * find the last mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400642 * mount_lock must be held.
Al Viro474279d2013-10-01 16:11:26 -0400643 */
644struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
645{
Eric W. Biederman411a938b2014-12-22 19:12:07 -0600646 struct mount *p, *res = NULL;
647 p = __lookup_mnt(mnt, dentry);
Al Viro38129a12014-03-20 21:10:51 -0400648 if (!p)
649 goto out;
Eric W. Biederman411a938b2014-12-22 19:12:07 -0600650 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
651 res = p;
Al Viro38129a12014-03-20 21:10:51 -0400652 hlist_for_each_entry_continue(p, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -0400653 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
654 break;
Eric W. Biederman411a938b2014-12-22 19:12:07 -0600655 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
656 res = p;
Al Viro1d6a32a2014-03-20 20:34:43 -0400657 }
Al Viro38129a12014-03-20 21:10:51 -0400658out:
Al Viro1d6a32a2014-03-20 20:34:43 -0400659 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660}
661
Ram Paia05964f2005-11-07 17:20:17 -0500662/*
David Howellsf015f1262012-06-25 12:55:28 +0100663 * lookup_mnt - Return the first child mount mounted at path
664 *
665 * "First" means first mounted chronologically. If you create the
666 * following mounts:
667 *
668 * mount /dev/sda1 /mnt
669 * mount /dev/sda2 /mnt
670 * mount /dev/sda3 /mnt
671 *
672 * Then lookup_mnt() on the base /mnt dentry in the root mount will
673 * return successively the root dentry and vfsmount of /dev/sda1, then
674 * /dev/sda2, then /dev/sda3, then NULL.
675 *
676 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500677 */
Al Viro1c755af2009-04-18 14:06:57 -0400678struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500679{
Al Viroc7105362011-11-24 18:22:03 -0500680 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400681 struct vfsmount *m;
682 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000683
Al Viro48a066e2013-09-29 22:06:07 -0400684 rcu_read_lock();
685 do {
686 seq = read_seqbegin(&mount_lock);
687 child_mnt = __lookup_mnt(path->mnt, path->dentry);
688 m = child_mnt ? &child_mnt->mnt : NULL;
689 } while (!legitimize_mnt(m, seq));
690 rcu_read_unlock();
691 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500692}
693
Eric W. Biederman7af13642013-10-04 19:15:13 -0700694/*
695 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
696 * current mount namespace.
697 *
698 * The common case is dentries are not mountpoints at all and that
699 * test is handled inline. For the slow case when we are actually
700 * dealing with a mountpoint of some kind, walk through all of the
701 * mounts in the current mount namespace and test to see if the dentry
702 * is a mountpoint.
703 *
704 * The mount_hashtable is not usable in the context because we
705 * need to identify all mounts that may be in the current mount
706 * namespace not just a mount that happens to have some specified
707 * parent mount.
708 */
709bool __is_local_mountpoint(struct dentry *dentry)
710{
711 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
712 struct mount *mnt;
713 bool is_covered = false;
714
715 if (!d_mountpoint(dentry))
716 goto out;
717
718 down_read(&namespace_sem);
719 list_for_each_entry(mnt, &ns->list, mnt_list) {
720 is_covered = (mnt->mnt_mountpoint == dentry);
721 if (is_covered)
722 break;
723 }
724 up_read(&namespace_sem);
725out:
726 return is_covered;
727}
728
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800729static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
Al Viro84d17192013-03-15 10:53:28 -0400730{
Al Viro0818bf22014-02-28 13:46:44 -0500731 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d17192013-03-15 10:53:28 -0400732 struct mountpoint *mp;
733
Al Viro0818bf22014-02-28 13:46:44 -0500734 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d17192013-03-15 10:53:28 -0400735 if (mp->m_dentry == dentry) {
736 /* might be worth a WARN_ON() */
737 if (d_unlinked(dentry))
738 return ERR_PTR(-ENOENT);
739 mp->m_count++;
740 return mp;
741 }
742 }
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800743 return NULL;
744}
745
746static struct mountpoint *new_mountpoint(struct dentry *dentry)
747{
748 struct hlist_head *chain = mp_hash(dentry);
749 struct mountpoint *mp;
750 int ret;
Al Viro84d17192013-03-15 10:53:28 -0400751
752 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
753 if (!mp)
754 return ERR_PTR(-ENOMEM);
755
Miklos Szeredieed81002013-09-05 14:39:11 +0200756 ret = d_set_mounted(dentry);
757 if (ret) {
Al Viro84d17192013-03-15 10:53:28 -0400758 kfree(mp);
Miklos Szeredieed81002013-09-05 14:39:11 +0200759 return ERR_PTR(ret);
Al Viro84d17192013-03-15 10:53:28 -0400760 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200761
Al Viro84d17192013-03-15 10:53:28 -0400762 mp->m_dentry = dentry;
763 mp->m_count = 1;
Al Viro0818bf22014-02-28 13:46:44 -0500764 hlist_add_head(&mp->m_hash, chain);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700765 INIT_HLIST_HEAD(&mp->m_list);
Al Viro84d17192013-03-15 10:53:28 -0400766 return mp;
767}
768
769static void put_mountpoint(struct mountpoint *mp)
770{
771 if (!--mp->m_count) {
772 struct dentry *dentry = mp->m_dentry;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700773 BUG_ON(!hlist_empty(&mp->m_list));
Al Viro84d17192013-03-15 10:53:28 -0400774 spin_lock(&dentry->d_lock);
775 dentry->d_flags &= ~DCACHE_MOUNTED;
776 spin_unlock(&dentry->d_lock);
Al Viro0818bf22014-02-28 13:46:44 -0500777 hlist_del(&mp->m_hash);
Al Viro84d17192013-03-15 10:53:28 -0400778 kfree(mp);
779 }
780}
781
Al Viro143c8c92011-11-25 00:46:35 -0500782static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800784 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785}
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) {
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 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800801static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500802{
803 if (ns && ns->event != event) {
804 ns->event = event;
805 wake_up_interruptible(&ns->poll);
806 }
807}
808
Nick Piggin99b7db72010-08-18 04:37:39 +1000809/*
810 * vfsmount lock must be held for write
811 */
Eric W. Biederman7bdb11d2014-12-29 13:03:41 -0600812static void unhash_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
Al Viro0714a532011-11-24 22:19:58 -0500814 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500815 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500816 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400817 hlist_del_init_rcu(&mnt->mnt_hash);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700818 hlist_del_init(&mnt->mnt_mp_list);
Al Viro84d17192013-03-15 10:53:28 -0400819 put_mountpoint(mnt->mnt_mp);
820 mnt->mnt_mp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821}
822
Nick Piggin99b7db72010-08-18 04:37:39 +1000823/*
824 * vfsmount lock must be held for write
825 */
Eric W. Biederman7bdb11d2014-12-29 13:03:41 -0600826static void detach_mnt(struct mount *mnt, struct path *old_path)
827{
828 old_path->dentry = mnt->mnt_mountpoint;
829 old_path->mnt = &mnt->mnt_parent->mnt;
830 unhash_mnt(mnt);
831}
832
833/*
834 * vfsmount lock must be held for write
835 */
Eric W. Biederman6a46c572015-01-15 22:58:33 -0600836static void umount_mnt(struct mount *mnt)
837{
838 /* old mountpoint will be dropped when we can do that */
839 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
840 unhash_mnt(mnt);
841}
842
843/*
844 * vfsmount lock must be held for write
845 */
Al Viro84d17192013-03-15 10:53:28 -0400846void mnt_set_mountpoint(struct mount *mnt,
847 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500848 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500849{
Al Viro84d17192013-03-15 10:53:28 -0400850 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500851 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro84d17192013-03-15 10:53:28 -0400852 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500853 child_mnt->mnt_parent = mnt;
Al Viro84d17192013-03-15 10:53:28 -0400854 child_mnt->mnt_mp = mp;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700855 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500856}
857
Nick Piggin99b7db72010-08-18 04:37:39 +1000858/*
859 * vfsmount lock must be held for write
860 */
Al Viro84d17192013-03-15 10:53:28 -0400861static void attach_mnt(struct mount *mnt,
862 struct mount *parent,
863 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
Al Viro84d17192013-03-15 10:53:28 -0400865 mnt_set_mountpoint(parent, mp, mnt);
Al Viro38129a12014-03-20 21:10:51 -0400866 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
Al Viro84d17192013-03-15 10:53:28 -0400867 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500868}
869
Al Viro12a5b522014-08-10 03:44:55 -0400870static void attach_shadowed(struct mount *mnt,
871 struct mount *parent,
872 struct mount *shadows)
873{
874 if (shadows) {
Linus Torvaldsf6f99332014-08-11 11:44:11 -0700875 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
Al Viro12a5b522014-08-10 03:44:55 -0400876 list_add(&mnt->mnt_child, &shadows->mnt_child);
877 } else {
878 hlist_add_head_rcu(&mnt->mnt_hash,
879 m_hash(&parent->mnt, mnt->mnt_mountpoint));
880 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
881 }
882}
883
Ram Paib90fa9a2005-11-07 17:19:50 -0500884/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000885 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500886 */
Al Viro1d6a32a2014-03-20 20:34:43 -0400887static void commit_tree(struct mount *mnt, struct mount *shadows)
Ram Paib90fa9a2005-11-07 17:19:50 -0500888{
Al Viro0714a532011-11-24 22:19:58 -0500889 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500890 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500891 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500892 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500893
Al Viro0714a532011-11-24 22:19:58 -0500894 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500895
Al Viro1a4eeaf2011-11-25 02:19:55 -0500896 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400897 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500898 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500899
Ram Paib90fa9a2005-11-07 17:19:50 -0500900 list_splice(&head, n->list.prev);
901
Al Viro12a5b522014-08-10 03:44:55 -0400902 attach_shadowed(mnt, parent, shadows);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800903 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904}
905
Al Viro909b0a82011-11-25 03:06:56 -0500906static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
Al Viro6b41d532011-11-24 23:24:33 -0500908 struct list_head *next = p->mnt_mounts.next;
909 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500911 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500913 next = p->mnt_child.next;
914 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 break;
Al Viro0714a532011-11-24 22:19:58 -0500916 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 }
918 }
Al Viro6b41d532011-11-24 23:24:33 -0500919 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920}
921
Al Viro315fc832011-11-24 18:57:30 -0500922static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500923{
Al Viro6b41d532011-11-24 23:24:33 -0500924 struct list_head *prev = p->mnt_mounts.prev;
925 while (prev != &p->mnt_mounts) {
926 p = list_entry(prev, struct mount, mnt_child);
927 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500928 }
929 return p;
930}
931
Al Viro9d412a42011-03-17 22:08:28 -0400932struct vfsmount *
933vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
934{
Al Virob105e272011-11-24 20:38:33 -0500935 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400936 struct dentry *root;
937
938 if (!type)
939 return ERR_PTR(-ENODEV);
940
941 mnt = alloc_vfsmnt(name);
942 if (!mnt)
943 return ERR_PTR(-ENOMEM);
944
945 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500946 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400947
948 root = mount_fs(type, flags, name, data);
949 if (IS_ERR(root)) {
David Howells8ffcb322014-01-24 12:17:54 +0000950 mnt_free_id(mnt);
Al Viro9d412a42011-03-17 22:08:28 -0400951 free_vfsmnt(mnt);
952 return ERR_CAST(root);
953 }
954
Al Virob105e272011-11-24 20:38:33 -0500955 mnt->mnt.mnt_root = root;
956 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500957 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500958 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400959 lock_mount_hash();
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100960 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400961 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -0500962 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400963}
964EXPORT_SYMBOL_GPL(vfs_kern_mount);
965
Al Viro87129cc2011-11-24 21:24:27 -0500966static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500967 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968{
Al Viro87129cc2011-11-24 21:24:27 -0500969 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100970 struct mount *mnt;
971 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
David Howellsbe34d1a2012-06-25 12:55:18 +0100973 mnt = alloc_vfsmnt(old->mnt_devname);
974 if (!mnt)
975 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100976
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700977 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +0100978 mnt->mnt_group_id = 0; /* not a peer of original */
979 else
980 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100981
David Howellsbe34d1a2012-06-25 12:55:18 +0100982 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
983 err = mnt_alloc_group_id(mnt);
984 if (err)
985 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100987
Al Virof2ebb3a2014-02-27 09:35:45 -0500988 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700989 /* Don't allow unprivileged users to change mount flags */
Eric W. Biederman9566d672014-07-28 17:26:07 -0700990 if (flag & CL_UNPRIVILEGED) {
991 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
992
993 if (mnt->mnt.mnt_flags & MNT_READONLY)
994 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
995
996 if (mnt->mnt.mnt_flags & MNT_NODEV)
997 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
998
999 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1000 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1001
1002 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1003 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1004 }
Eric W. Biederman132c94e2013-03-22 04:08:05 -07001005
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001006 /* Don't allow unprivileged users to reveal what is under a mount */
Eric W. Biederman381cacb2014-10-07 17:11:46 -07001007 if ((flag & CL_UNPRIVILEGED) &&
1008 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001009 mnt->mnt.mnt_flags |= MNT_LOCKED;
1010
David Howellsbe34d1a2012-06-25 12:55:18 +01001011 atomic_inc(&sb->s_active);
1012 mnt->mnt.mnt_sb = sb;
1013 mnt->mnt.mnt_root = dget(root);
1014 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1015 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -04001016 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001017 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001018 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001019
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001020 if ((flag & CL_SLAVE) ||
1021 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +01001022 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1023 mnt->mnt_master = old;
1024 CLEAR_MNT_SHARED(mnt);
1025 } else if (!(flag & CL_PRIVATE)) {
1026 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1027 list_add(&mnt->mnt_share, &old->mnt_share);
1028 if (IS_MNT_SLAVE(old))
1029 list_add(&mnt->mnt_slave, &old->mnt_slave);
1030 mnt->mnt_master = old->mnt_master;
1031 }
1032 if (flag & CL_MAKE_SHARED)
1033 set_mnt_shared(mnt);
1034
1035 /* stick the duplicate mount on the same expiry list
1036 * as the original if that was on one */
1037 if (flag & CL_EXPIRE) {
1038 if (!list_empty(&old->mnt_expire))
1039 list_add(&mnt->mnt_expire, &old->mnt_expire);
1040 }
1041
Al Virocb338d02011-11-24 20:55:08 -05001042 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001043
1044 out_free:
David Howells8ffcb322014-01-24 12:17:54 +00001045 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001046 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001047 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048}
1049
Al Viro9ea459e2014-08-08 13:08:20 -04001050static void cleanup_mnt(struct mount *mnt)
1051{
1052 /*
1053 * This probably indicates that somebody messed
1054 * up a mnt_want/drop_write() pair. If this
1055 * happens, the filesystem was probably unable
1056 * to make r/w->r/o transitions.
1057 */
1058 /*
1059 * The locking used to deal with mnt_count decrement provides barriers,
1060 * so mnt_get_writers() below is safe.
1061 */
1062 WARN_ON(mnt_get_writers(mnt));
1063 if (unlikely(mnt->mnt_pins.first))
1064 mnt_pin_kill(mnt);
1065 fsnotify_vfsmount_delete(&mnt->mnt);
1066 dput(mnt->mnt.mnt_root);
1067 deactivate_super(mnt->mnt.mnt_sb);
1068 mnt_free_id(mnt);
1069 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1070}
1071
1072static void __cleanup_mnt(struct rcu_head *head)
1073{
1074 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1075}
1076
1077static LLIST_HEAD(delayed_mntput_list);
1078static void delayed_mntput(struct work_struct *unused)
1079{
1080 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1081 struct llist_node *next;
1082
1083 for (; node; node = next) {
1084 next = llist_next(node);
1085 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1086 }
1087}
1088static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1089
Al Viro900148d2011-11-25 00:33:11 -05001090static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001091{
Al Viro48a066e2013-09-29 22:06:07 -04001092 rcu_read_lock();
1093 mnt_add_count(mnt, -1);
1094 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1095 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -05001096 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001097 }
Al Viro719ea2f2013-09-29 11:24:49 -04001098 lock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001099 if (mnt_get_count(mnt)) {
Al Viro48a066e2013-09-29 22:06:07 -04001100 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -04001101 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001102 return;
1103 }
Al Viro48a066e2013-09-29 22:06:07 -04001104 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1105 rcu_read_unlock();
1106 unlock_mount_hash();
1107 return;
1108 }
1109 mnt->mnt.mnt_flags |= MNT_DOOMED;
1110 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301111
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001112 list_del(&mnt->mnt_instance);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001113
1114 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1115 struct mount *p, *tmp;
1116 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1117 umount_mnt(p);
1118 }
1119 }
Al Viro719ea2f2013-09-29 11:24:49 -04001120 unlock_mount_hash();
Al Viro649a7952013-09-28 12:41:25 -04001121
Al Viro9ea459e2014-08-08 13:08:20 -04001122 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1123 struct task_struct *task = current;
1124 if (likely(!(task->flags & PF_KTHREAD))) {
1125 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1126 if (!task_work_add(task, &mnt->mnt_rcu, true))
1127 return;
1128 }
1129 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1130 schedule_delayed_work(&delayed_mntput_work, 1);
1131 return;
1132 }
1133 cleanup_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001134}
Nick Pigginb3e19d92011-01-07 17:50:11 +11001135
1136void mntput(struct vfsmount *mnt)
1137{
1138 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -05001139 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001140 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -05001141 if (unlikely(m->mnt_expiry_mark))
1142 m->mnt_expiry_mark = 0;
1143 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001144 }
1145}
1146EXPORT_SYMBOL(mntput);
1147
1148struct vfsmount *mntget(struct vfsmount *mnt)
1149{
1150 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001151 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001152 return mnt;
1153}
1154EXPORT_SYMBOL(mntget);
1155
Al Viro3064c352014-08-07 09:12:31 -04001156struct vfsmount *mnt_clone_internal(struct path *path)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001157{
Al Viro3064c352014-08-07 09:12:31 -04001158 struct mount *p;
1159 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1160 if (IS_ERR(p))
1161 return ERR_CAST(p);
1162 p->mnt.mnt_flags |= MNT_INTERNAL;
1163 return &p->mnt;
Al Viro7b7b1ac2005-11-07 17:13:39 -05001164}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001166static inline void mangle(struct seq_file *m, const char *s)
1167{
1168 seq_escape(m, s, " \t\n\\");
1169}
1170
1171/*
1172 * Simple .show_options callback for filesystems which don't want to
1173 * implement more complex mount option showing.
1174 *
1175 * See also save_mount_options().
1176 */
Al Viro34c80b12011-12-08 21:32:45 -05001177int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001178{
Al Viro2a32ceb2009-05-08 16:05:57 -04001179 const char *options;
1180
1181 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -05001182 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001183
1184 if (options != NULL && options[0]) {
1185 seq_putc(m, ',');
1186 mangle(m, options);
1187 }
Al Viro2a32ceb2009-05-08 16:05:57 -04001188 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001189
1190 return 0;
1191}
1192EXPORT_SYMBOL(generic_show_options);
1193
1194/*
1195 * If filesystem uses generic_show_options(), this function should be
1196 * called from the fill_super() callback.
1197 *
1198 * The .remount_fs callback usually needs to be handled in a special
1199 * way, to make sure, that previous options are not overwritten if the
1200 * remount fails.
1201 *
1202 * Also note, that if the filesystem's .remount_fs function doesn't
1203 * reset all options to their default value, but changes only newly
1204 * given options, then the displayed options will not reflect reality
1205 * any more.
1206 */
1207void save_mount_options(struct super_block *sb, char *options)
1208{
Al Viro2a32ceb2009-05-08 16:05:57 -04001209 BUG_ON(sb->s_options);
1210 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001211}
1212EXPORT_SYMBOL(save_mount_options);
1213
Al Viro2a32ceb2009-05-08 16:05:57 -04001214void replace_mount_options(struct super_block *sb, char *options)
1215{
1216 char *old = sb->s_options;
1217 rcu_assign_pointer(sb->s_options, options);
1218 if (old) {
1219 synchronize_rcu();
1220 kfree(old);
1221 }
1222}
1223EXPORT_SYMBOL(replace_mount_options);
1224
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001225#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001226/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227static void *m_start(struct seq_file *m, loff_t *pos)
1228{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001229 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Ram Pai390c6842005-11-07 17:17:51 -05001231 down_read(&namespace_sem);
Al Viroc7999c32014-02-27 14:40:10 -05001232 if (p->cached_event == p->ns->event) {
1233 void *v = p->cached_mount;
1234 if (*pos == p->cached_index)
1235 return v;
1236 if (*pos == p->cached_index + 1) {
1237 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1238 return p->cached_mount = v;
1239 }
1240 }
1241
1242 p->cached_event = p->ns->event;
1243 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1244 p->cached_index = *pos;
1245 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246}
1247
1248static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1249{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001250 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001251
Al Viroc7999c32014-02-27 14:40:10 -05001252 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1253 p->cached_index = *pos;
1254 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255}
1256
1257static void m_stop(struct seq_file *m, void *v)
1258{
Ram Pai390c6842005-11-07 17:17:51 -05001259 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
1261
Al Viro0226f492011-12-06 12:21:54 -05001262static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001263{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001264 struct proc_mounts *p = m->private;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001265 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001266 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267}
1268
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001269const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 .start = m_start,
1271 .next = m_next,
1272 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001273 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001274};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001275#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277/**
1278 * may_umount_tree - check if a mount tree is busy
1279 * @mnt: root of mount tree
1280 *
1281 * This is called to check if a tree of mounts has any
1282 * open files, pwds, chroots or sub mounts that are
1283 * busy.
1284 */
Al Viro909b0a82011-11-25 03:06:56 -05001285int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286{
Al Viro909b0a82011-11-25 03:06:56 -05001287 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001288 int actual_refs = 0;
1289 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001290 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001291 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
Nick Pigginb3e19d92011-01-07 17:50:11 +11001293 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001294 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001295 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001296 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 }
Al Viro719ea2f2013-09-29 11:24:49 -04001299 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
1301 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001302 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Ian Kente3474a82006-03-27 01:14:51 -08001304 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305}
1306
1307EXPORT_SYMBOL(may_umount_tree);
1308
1309/**
1310 * may_umount - check if a mount point is busy
1311 * @mnt: root of mount
1312 *
1313 * This is called to check if a mount point has any
1314 * open files, pwds, chroots or sub mounts. If the
1315 * mount has sub mounts this will return busy
1316 * regardless of whether the sub mounts are busy.
1317 *
1318 * Doesn't take quota and stuff into account. IOW, in some cases it will
1319 * give false negatives. The main reason why it's here is that we need
1320 * a non-destructive way to look for easily umountable filesystems.
1321 */
1322int may_umount(struct vfsmount *mnt)
1323{
Ian Kente3474a82006-03-27 01:14:51 -08001324 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001325 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001326 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001327 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001328 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001329 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001330 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001331 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332}
1333
1334EXPORT_SYMBOL(may_umount);
1335
Al Viro38129a12014-03-20 21:10:51 -04001336static HLIST_HEAD(unmounted); /* protected by namespace_sem */
Al Viroe3197d82013-03-16 14:35:16 -04001337
Al Viro97216be2013-03-16 15:12:40 -04001338static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339{
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001340 struct hlist_head head;
Al Viro97216be2013-03-16 15:12:40 -04001341
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001342 hlist_move_list(&unmounted, &head);
Al Viro97216be2013-03-16 15:12:40 -04001343
Al Viro97216be2013-03-16 15:12:40 -04001344 up_write(&namespace_sem);
1345
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001346 if (likely(hlist_empty(&head)))
1347 return;
1348
Al Viro48a066e2013-09-29 22:06:07 -04001349 synchronize_rcu();
1350
Al Viro87b95ce2015-01-10 19:01:08 -05001351 group_pin_kill(&head);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001352}
1353
Al Viro97216be2013-03-16 15:12:40 -04001354static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001355{
Al Viro97216be2013-03-16 15:12:40 -04001356 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001357}
1358
Eric W. Biedermane819f152014-12-24 07:20:01 -06001359enum umount_tree_flags {
1360 UMOUNT_SYNC = 1,
1361 UMOUNT_PROPAGATE = 2,
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001362 UMOUNT_CONNECTED = 4,
Eric W. Biedermane819f152014-12-24 07:20:01 -06001363};
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001364
1365static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1366{
1367 /* Leaving mounts connected is only valid for lazy umounts */
1368 if (how & UMOUNT_SYNC)
1369 return true;
1370
1371 /* A mount without a parent has nothing to be connected to */
1372 if (!mnt_has_parent(mnt))
1373 return true;
1374
1375 /* Because the reference counting rules change when mounts are
1376 * unmounted and connected, umounted mounts may not be
1377 * connected to mounted mounts.
1378 */
1379 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1380 return true;
1381
1382 /* Has it been requested that the mount remain connected? */
1383 if (how & UMOUNT_CONNECTED)
1384 return false;
1385
1386 /* Is the mount locked such that it needs to remain connected? */
1387 if (IS_MNT_LOCKED(mnt))
1388 return false;
1389
1390 /* By default disconnect the mount */
1391 return true;
1392}
1393
Nick Piggin99b7db72010-08-18 04:37:39 +10001394/*
Al Viro48a066e2013-09-29 22:06:07 -04001395 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001396 * namespace_sem must be held for write
1397 */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001398static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001399{
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001400 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001401 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001402
Eric W. Biederman5d884572015-01-03 05:39:35 -06001403 if (how & UMOUNT_PROPAGATE)
1404 propagate_mount_unlock(mnt);
1405
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001406 /* Gather the mounts to umount */
Eric W. Biederman590ce4b2014-12-22 18:30:08 -06001407 for (p = mnt; p; p = next_mnt(p, mnt)) {
1408 p->mnt.mnt_flags |= MNT_UMOUNT;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001409 list_move(&p->mnt_list, &tmp_list);
Eric W. Biederman590ce4b2014-12-22 18:30:08 -06001410 }
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001411
Eric W. Biederman411a938b2014-12-22 19:12:07 -06001412 /* Hide the mounts from mnt_mounts */
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001413 list_for_each_entry(p, &tmp_list, mnt_list) {
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001414 list_del_init(&p->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -04001415 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001416
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001417 /* Add propogated mounts to the tmp_list */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001418 if (how & UMOUNT_PROPAGATE)
Al Viro7b8a53f2011-01-15 20:08:44 -05001419 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001420
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001421 while (!list_empty(&tmp_list)) {
Eric W. Biedermance07d892014-12-23 21:37:03 -06001422 bool disconnect;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001423 p = list_first_entry(&tmp_list, struct mount, mnt_list);
Al Viro6776db3d2011-11-25 00:22:05 -05001424 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001425 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001426 __touch_mnt_namespace(p->mnt_ns);
1427 p->mnt_ns = NULL;
Eric W. Biedermane819f152014-12-24 07:20:01 -06001428 if (how & UMOUNT_SYNC)
Al Viro48a066e2013-09-29 22:06:07 -04001429 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro87b95ce2015-01-10 19:01:08 -05001430
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001431 disconnect = disconnect_mount(p, how);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001432
1433 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1434 disconnect ? &unmounted : NULL);
Al Viro676da582011-11-24 21:47:05 -05001435 if (mnt_has_parent(p)) {
Al Viro81b6b062014-08-30 18:32:05 -04001436 mnt_add_count(p->mnt_parent, -1);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001437 if (!disconnect) {
1438 /* Don't forget about p */
1439 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1440 } else {
1441 umount_mnt(p);
1442 }
Al Viro7c4b93d2008-03-21 23:59:49 -04001443 }
Al Viro0f0afb12011-11-24 20:43:10 -05001444 change_mnt_propagation(p, MS_PRIVATE);
Al Viro38129a12014-03-20 21:10:51 -04001445 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446}
1447
Al Virob54b9be2013-03-16 14:39:34 -04001448static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001449
Al Viro1ab59732011-11-24 21:35:16 -05001450static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451{
Al Viro1ab59732011-11-24 21:35:16 -05001452 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 int retval;
1454
Al Viro1ab59732011-11-24 21:35:16 -05001455 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 if (retval)
1457 return retval;
1458
1459 /*
1460 * Allow userspace to request a mountpoint be expired rather than
1461 * unmounting unconditionally. Unmount only happens if:
1462 * (1) the mark is already set (the mark is cleared by mntput())
1463 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1464 */
1465 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001466 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 flags & (MNT_FORCE | MNT_DETACH))
1468 return -EINVAL;
1469
Nick Pigginb3e19d92011-01-07 17:50:11 +11001470 /*
1471 * probably don't strictly need the lock here if we examined
1472 * all race cases, but it's a slowpath.
1473 */
Al Viro719ea2f2013-09-29 11:24:49 -04001474 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001475 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001476 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001478 }
Al Viro719ea2f2013-09-29 11:24:49 -04001479 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Al Viro863d6842011-11-25 00:57:42 -05001481 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 return -EAGAIN;
1483 }
1484
1485 /*
1486 * If we may have to abort operations to get out of this
1487 * mount, and they will themselves hold resources we must
1488 * allow the fs to do things. In the Unix tradition of
1489 * 'Gee thats tricky lets do it in userspace' the umount_begin
1490 * might fail to complete on the first run through as other tasks
1491 * must return, and the like. Thats for the mount program to worry
1492 * about for the moment.
1493 */
1494
Al Viro42faad92008-04-24 07:21:56 -04001495 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001496 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001497 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498
1499 /*
1500 * No sense to grab the lock for this test, but test itself looks
1501 * somewhat bogus. Suggestions for better replacement?
1502 * Ho-hum... In principle, we might treat that as umount + switch
1503 * to rootfs. GC would eventually take care of the old vfsmount.
1504 * Actually it makes sense, especially if rootfs would contain a
1505 * /reboot - static binary that would close all descriptors and
1506 * call reboot(9). Then init(8) could umount root and exec /reboot.
1507 */
Al Viro1ab59732011-11-24 21:35:16 -05001508 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 /*
1510 * Special case for "unmounting" root ...
1511 * we just try to remount it readonly.
1512 */
Andy Lutomirskia1480dc2014-10-08 12:32:47 -07001513 if (!capable(CAP_SYS_ADMIN))
1514 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001516 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 up_write(&sb->s_umount);
1519 return retval;
1520 }
1521
Al Viro97216be2013-03-16 15:12:40 -04001522 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001523 lock_mount_hash();
Al Viro5addc5d2005-11-07 17:15:49 -05001524 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
Al Viro48a066e2013-09-29 22:06:07 -04001526 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001527 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001528 umount_tree(mnt, UMOUNT_PROPAGATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001530 } else {
1531 shrink_submounts(mnt);
1532 retval = -EBUSY;
1533 if (!propagate_mount_busy(mnt, 2)) {
1534 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001535 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Viro48a066e2013-09-29 22:06:07 -04001536 retval = 0;
1537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 }
Al Viro719ea2f2013-09-29 11:24:49 -04001539 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001540 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 return retval;
1542}
1543
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001544/*
1545 * __detach_mounts - lazily unmount all mounts on the specified dentry
1546 *
1547 * During unlink, rmdir, and d_drop it is possible to loose the path
1548 * to an existing mountpoint, and wind up leaking the mount.
1549 * detach_mounts allows lazily unmounting those mounts instead of
1550 * leaking them.
1551 *
1552 * The caller may hold dentry->d_inode->i_mutex.
1553 */
1554void __detach_mounts(struct dentry *dentry)
1555{
1556 struct mountpoint *mp;
1557 struct mount *mnt;
1558
1559 namespace_lock();
1560 mp = lookup_mountpoint(dentry);
Eric W. Biedermanf53e5792015-01-19 11:48:45 -06001561 if (IS_ERR_OR_NULL(mp))
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001562 goto out_unlock;
1563
1564 lock_mount_hash();
1565 while (!hlist_empty(&mp->m_list)) {
1566 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001567 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
Eric W. Biedermanfe78fcc2015-07-17 14:54:27 -05001568 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1569 umount_mnt(mnt);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001570 }
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001571 else umount_tree(mnt, UMOUNT_CONNECTED);
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001572 }
1573 unlock_mount_hash();
1574 put_mountpoint(mp);
1575out_unlock:
1576 namespace_unlock();
1577}
1578
Al Viro9b40bc92013-02-22 22:45:42 -05001579/*
1580 * Is the caller allowed to modify his namespace?
1581 */
1582static inline bool may_mount(void)
1583{
1584 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1585}
1586
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587/*
1588 * Now umount can handle mount points as well as block devices.
1589 * This is important for filesystems which use unnamed block devices.
1590 *
1591 * We now support a flag for forced unmount like the other 'big iron'
1592 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1593 */
1594
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001595SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596{
Al Viro2d8f3032008-07-22 09:59:21 -04001597 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001598 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001600 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001602 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1603 return -EINVAL;
1604
Al Viro9b40bc92013-02-22 22:45:42 -05001605 if (!may_mount())
1606 return -EPERM;
1607
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001608 if (!(flags & UMOUNT_NOFOLLOW))
1609 lookup_flags |= LOOKUP_FOLLOW;
1610
Al Viro197df042013-09-08 14:03:27 -04001611 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (retval)
1613 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001614 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001616 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001618 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 goto dput_and_out;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001620 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1621 goto dput_and_out;
Eric W. Biedermanb2f5d4d2014-10-04 14:44:03 -07001622 retval = -EPERM;
1623 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1624 goto dput_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
Al Viro900148d2011-11-25 00:33:11 -05001626 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001628 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001629 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001630 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631out:
1632 return retval;
1633}
1634
1635#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1636
1637/*
Ram Paib58fed82005-11-07 17:16:09 -05001638 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001640SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641{
Ram Paib58fed82005-11-07 17:16:09 -05001642 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643}
1644
1645#endif
1646
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001647static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001648{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001649 /* Is this a proxy for a mount namespace? */
Al Viroe149ed22014-11-01 10:57:28 -04001650 return dentry->d_op == &ns_dentry_operations &&
1651 dentry->d_fsdata == &mntns_operations;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001652}
1653
Al Viro58be28252014-11-01 00:00:23 -04001654struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1655{
1656 return container_of(ns, struct mnt_namespace, ns);
1657}
1658
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001659static bool mnt_ns_loop(struct dentry *dentry)
1660{
1661 /* Could bind mounting the mount namespace inode cause a
1662 * mount namespace loop?
1663 */
1664 struct mnt_namespace *mnt_ns;
1665 if (!is_mnt_ns_file(dentry))
1666 return false;
1667
Al Virof77c8012014-11-01 03:13:17 -04001668 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
Eric W. Biederman8823c072010-03-07 18:49:36 -08001669 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1670}
1671
Al Viro87129cc2011-11-24 21:24:27 -05001672struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001673 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674{
Al Viro84d17192013-03-15 10:53:28 -04001675 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001677 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1678 return ERR_PTR(-EINVAL);
1679
1680 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001681 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001682
Ram Pai36341f62005-11-07 17:17:22 -05001683 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001684 if (IS_ERR(q))
1685 return q;
1686
Al Viroa73324d2011-11-24 22:25:07 -05001687 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
1689 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001690 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001691 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001692 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 continue;
1694
Al Viro909b0a82011-11-25 03:06:56 -05001695 for (s = r; s; s = next_mnt(s, r)) {
Al Viro12a5b522014-08-10 03:44:55 -04001696 struct mount *t = NULL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001697 if (!(flag & CL_COPY_UNBINDABLE) &&
1698 IS_MNT_UNBINDABLE(s)) {
1699 s = skip_mnt_tree(s);
1700 continue;
1701 }
1702 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1703 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001704 s = skip_mnt_tree(s);
1705 continue;
1706 }
Al Viro0714a532011-11-24 22:19:58 -05001707 while (p != s->mnt_parent) {
1708 p = p->mnt_parent;
1709 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 }
Al Viro87129cc2011-11-24 21:24:27 -05001711 p = s;
Al Viro84d17192013-03-15 10:53:28 -04001712 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001713 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001714 if (IS_ERR(q))
1715 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04001716 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05001717 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro12a5b522014-08-10 03:44:55 -04001718 mnt_set_mountpoint(parent, p->mnt_mp, q);
1719 if (!list_empty(&parent->mnt_mounts)) {
1720 t = list_last_entry(&parent->mnt_mounts,
1721 struct mount, mnt_child);
1722 if (t->mnt_mp != p->mnt_mp)
1723 t = NULL;
1724 }
1725 attach_shadowed(q, parent, t);
Al Viro719ea2f2013-09-29 11:24:49 -04001726 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 }
1728 }
1729 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001730out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04001732 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001733 umount_tree(res, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001734 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001736 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737}
1738
David Howellsbe34d1a2012-06-25 12:55:18 +01001739/* Caller should check returned pointer for errors */
1740
Al Viro589ff872009-04-18 03:28:19 -04001741struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001742{
Al Virocb338d02011-11-24 20:55:08 -05001743 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001744 namespace_lock();
Eric W. Biedermancd4a4012015-01-07 14:28:26 -06001745 if (!check_mnt(real_mount(path->mnt)))
1746 tree = ERR_PTR(-EINVAL);
1747 else
1748 tree = copy_tree(real_mount(path->mnt), path->dentry,
1749 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001750 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001751 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001752 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001753 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001754}
1755
1756void drop_collected_mounts(struct vfsmount *mnt)
1757{
Al Viro97216be2013-03-16 15:12:40 -04001758 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001759 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001760 umount_tree(real_mount(mnt), UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001761 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04001762 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001763}
1764
Miklos Szeredic771d682014-10-24 00:14:36 +02001765/**
1766 * clone_private_mount - create a private clone of a path
1767 *
1768 * This creates a new vfsmount, which will be the clone of @path. The new will
1769 * not be attached anywhere in the namespace and will be private (i.e. changes
1770 * to the originating mount won't be propagated into this).
1771 *
1772 * Release with mntput().
1773 */
1774struct vfsmount *clone_private_mount(struct path *path)
1775{
1776 struct mount *old_mnt = real_mount(path->mnt);
1777 struct mount *new_mnt;
1778
1779 if (IS_MNT_UNBINDABLE(old_mnt))
1780 return ERR_PTR(-EINVAL);
1781
1782 down_read(&namespace_sem);
1783 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1784 up_read(&namespace_sem);
1785 if (IS_ERR(new_mnt))
1786 return ERR_CAST(new_mnt);
1787
1788 return &new_mnt->mnt;
1789}
1790EXPORT_SYMBOL_GPL(clone_private_mount);
1791
Al Viro1f707132010-01-30 22:51:25 -05001792int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1793 struct vfsmount *root)
1794{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001795 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001796 int res = f(root, arg);
1797 if (res)
1798 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001799 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1800 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001801 if (res)
1802 return res;
1803 }
1804 return 0;
1805}
1806
Al Viro4b8b21f2011-11-24 19:54:23 -05001807static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001808{
Al Viro315fc832011-11-24 18:57:30 -05001809 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001810
Al Viro909b0a82011-11-25 03:06:56 -05001811 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001812 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001813 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001814 }
1815}
1816
Al Viro4b8b21f2011-11-24 19:54:23 -05001817static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001818{
Al Viro315fc832011-11-24 18:57:30 -05001819 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001820
Al Viro909b0a82011-11-25 03:06:56 -05001821 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001822 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001823 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001824 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001825 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001826 return err;
1827 }
1828 }
1829 }
1830
1831 return 0;
1832}
1833
Ram Paib90fa9a2005-11-07 17:19:50 -05001834/*
1835 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001836 * @nd : place the mount tree @source_mnt is attached
1837 * @parent_nd : if non-null, detach the source_mnt from its parent and
1838 * store the parent mount and mountpoint dentry.
1839 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001840 *
1841 * NOTE: in the table below explains the semantics when a source mount
1842 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001843 * ---------------------------------------------------------------------------
1844 * | BIND MOUNT OPERATION |
1845 * |**************************************************************************
1846 * | source-->| shared | private | slave | unbindable |
1847 * | dest | | | | |
1848 * | | | | | | |
1849 * | v | | | | |
1850 * |**************************************************************************
1851 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1852 * | | | | | |
1853 * |non-shared| shared (+) | private | slave (*) | invalid |
1854 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001855 * A bind operation clones the source mount and mounts the clone on the
1856 * destination mount.
1857 *
1858 * (++) the cloned mount is propagated to all the mounts in the propagation
1859 * tree of the destination mount and the cloned mount is added to
1860 * the peer group of the source mount.
1861 * (+) the cloned mount is created under the destination mount and is marked
1862 * as shared. The cloned mount is added to the peer group of the source
1863 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001864 * (+++) the mount is propagated to all the mounts in the propagation tree
1865 * of the destination mount and the cloned mount is made slave
1866 * of the same master as that of the source mount. The cloned mount
1867 * is marked as 'shared and slave'.
1868 * (*) the cloned mount is made a slave of the same master as that of the
1869 * source mount.
1870 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001871 * ---------------------------------------------------------------------------
1872 * | MOVE MOUNT OPERATION |
1873 * |**************************************************************************
1874 * | source-->| shared | private | slave | unbindable |
1875 * | dest | | | | |
1876 * | | | | | | |
1877 * | v | | | | |
1878 * |**************************************************************************
1879 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1880 * | | | | | |
1881 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1882 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001883 *
1884 * (+) the mount is moved to the destination. And is then propagated to
1885 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001886 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001887 * (+++) the mount is moved to the destination and is then propagated to
1888 * all the mounts belonging to the destination mount's propagation tree.
1889 * the mount is marked as 'shared and slave'.
1890 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001891 *
1892 * if the source mount is a tree, the operations explained above is
1893 * applied to each mount in the tree.
1894 * Must be called without spinlocks held, since this function can sleep
1895 * in allocations.
1896 */
Al Viro0fb54e52011-11-24 19:59:16 -05001897static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d17192013-03-15 10:53:28 -04001898 struct mount *dest_mnt,
1899 struct mountpoint *dest_mp,
1900 struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001901{
Al Viro38129a12014-03-20 21:10:51 -04001902 HLIST_HEAD(tree_list);
Al Viro315fc832011-11-24 18:57:30 -05001903 struct mount *child, *p;
Al Viro38129a12014-03-20 21:10:51 -04001904 struct hlist_node *n;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001905 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001906
Al Virofc7be132011-11-25 01:05:37 -05001907 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001908 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001909 if (err)
1910 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04001911 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Al Virof2ebb3a2014-02-27 09:35:45 -05001912 lock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001913 if (err)
1914 goto out_cleanup_ids;
Al Viro909b0a82011-11-25 03:06:56 -05001915 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001916 set_mnt_shared(p);
Al Viro0b1b9012014-03-21 10:14:08 -04001917 } else {
1918 lock_mount_hash();
Ram Paib90fa9a2005-11-07 17:19:50 -05001919 }
Al Viro1a390682008-03-21 20:48:19 -04001920 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001921 detach_mnt(source_mnt, parent_path);
Al Viro84d17192013-03-15 10:53:28 -04001922 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05001923 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001924 } else {
Al Viro84d17192013-03-15 10:53:28 -04001925 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Al Viro1d6a32a2014-03-20 20:34:43 -04001926 commit_tree(source_mnt, NULL);
Ram Pai21444402005-11-07 17:20:03 -05001927 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001928
Al Viro38129a12014-03-20 21:10:51 -04001929 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04001930 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04001931 hlist_del_init(&child->mnt_hash);
Al Viro1d6a32a2014-03-20 20:34:43 -04001932 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1933 child->mnt_mountpoint);
1934 commit_tree(child, q);
Ram Paib90fa9a2005-11-07 17:19:50 -05001935 }
Al Viro719ea2f2013-09-29 11:24:49 -04001936 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10001937
Ram Paib90fa9a2005-11-07 17:19:50 -05001938 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001939
1940 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05001941 while (!hlist_empty(&tree_list)) {
1942 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
Eric W. Biedermane819f152014-12-24 07:20:01 -06001943 umount_tree(child, UMOUNT_SYNC);
Al Virof2ebb3a2014-02-27 09:35:45 -05001944 }
1945 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001946 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001947 out:
1948 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001949}
1950
Al Viro84d17192013-03-15 10:53:28 -04001951static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04001952{
1953 struct vfsmount *mnt;
Al Viro84d17192013-03-15 10:53:28 -04001954 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04001955retry:
Al Viro84d17192013-03-15 10:53:28 -04001956 mutex_lock(&dentry->d_inode->i_mutex);
1957 if (unlikely(cant_mount(dentry))) {
1958 mutex_unlock(&dentry->d_inode->i_mutex);
1959 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04001960 }
Al Viro97216be2013-03-16 15:12:40 -04001961 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04001962 mnt = lookup_mnt(path);
Al Viro84d17192013-03-15 10:53:28 -04001963 if (likely(!mnt)) {
Eric W. Biedermane2dfa932014-02-24 17:32:34 -08001964 struct mountpoint *mp = lookup_mountpoint(dentry);
1965 if (!mp)
1966 mp = new_mountpoint(dentry);
Al Viro84d17192013-03-15 10:53:28 -04001967 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04001968 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001969 mutex_unlock(&dentry->d_inode->i_mutex);
1970 return mp;
1971 }
1972 return mp;
1973 }
Al Viro97216be2013-03-16 15:12:40 -04001974 namespace_unlock();
Al Virob12cea92011-03-18 08:55:38 -04001975 mutex_unlock(&path->dentry->d_inode->i_mutex);
1976 path_put(path);
1977 path->mnt = mnt;
Al Viro84d17192013-03-15 10:53:28 -04001978 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04001979 goto retry;
1980}
1981
Al Viro84d17192013-03-15 10:53:28 -04001982static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04001983{
Al Viro84d17192013-03-15 10:53:28 -04001984 struct dentry *dentry = where->m_dentry;
1985 put_mountpoint(where);
Al Viro328e6d92013-03-16 14:49:45 -04001986 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001987 mutex_unlock(&dentry->d_inode->i_mutex);
Al Virob12cea92011-03-18 08:55:38 -04001988}
1989
Al Viro84d17192013-03-15 10:53:28 -04001990static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991{
Al Viro95bc5f22011-11-25 00:30:56 -05001992 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 return -EINVAL;
1994
David Howellse36cb0b2015-01-29 12:02:35 +00001995 if (d_is_dir(mp->m_dentry) !=
1996 d_is_dir(mnt->mnt.mnt_root))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 return -ENOTDIR;
1998
Al Viro84d17192013-03-15 10:53:28 -04001999 return attach_recursive_mnt(mnt, p, mp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000}
2001
2002/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002003 * Sanity check the flags to change_mnt_propagation.
2004 */
2005
2006static int flags_to_propagation_type(int flags)
2007{
Roman Borisov7c6e9842011-05-25 16:26:48 -07002008 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002009
2010 /* Fail if any non-propagation flags are set */
2011 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2012 return 0;
2013 /* Only one propagation flag should be set */
2014 if (!is_power_of_2(type))
2015 return 0;
2016 return type;
2017}
2018
2019/*
Ram Pai07b20882005-11-07 17:19:07 -05002020 * recursively change the type of the mountpoint.
2021 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002022static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05002023{
Al Viro315fc832011-11-24 18:57:30 -05002024 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05002025 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05002026 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002027 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002028 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05002029
Al Viro2d92ab32008-08-02 00:51:11 -04002030 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05002031 return -EINVAL;
2032
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002033 type = flags_to_propagation_type(flag);
2034 if (!type)
2035 return -EINVAL;
2036
Al Viro97216be2013-03-16 15:12:40 -04002037 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002038 if (type == MS_SHARED) {
2039 err = invent_group_ids(mnt, recurse);
2040 if (err)
2041 goto out_unlock;
2042 }
2043
Al Viro719ea2f2013-09-29 11:24:49 -04002044 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05002045 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05002046 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04002047 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002048
2049 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04002050 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002051 return err;
Ram Pai07b20882005-11-07 17:19:07 -05002052}
2053
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002054static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2055{
2056 struct mount *child;
2057 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2058 if (!is_subdir(child->mnt_mountpoint, dentry))
2059 continue;
2060
2061 if (child->mnt.mnt_flags & MNT_LOCKED)
2062 return true;
2063 }
2064 return false;
2065}
2066
Ram Pai07b20882005-11-07 17:19:07 -05002067/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 * do loopback mount.
2069 */
Al Viro808d4e32012-10-11 11:42:01 -04002070static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08002071 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072{
Al Viro2d92ab32008-08-02 00:51:11 -04002073 struct path old_path;
Al Viro84d17192013-03-15 10:53:28 -04002074 struct mount *mnt = NULL, *old, *parent;
2075 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002076 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 if (!old_name || !*old_name)
2078 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04002079 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 if (err)
2081 return err;
2082
Eric W. Biederman8823c072010-03-07 18:49:36 -08002083 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002084 if (mnt_ns_loop(old_path.dentry))
Eric W. Biederman8823c072010-03-07 18:49:36 -08002085 goto out;
2086
Al Viro84d17192013-03-15 10:53:28 -04002087 mp = lock_mount(path);
2088 err = PTR_ERR(mp);
2089 if (IS_ERR(mp))
Jan Blunck4ac91372008-02-14 19:34:32 -08002090 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05002091
Al Viro87129cc2011-11-24 21:24:27 -05002092 old = real_mount(old_path.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002093 parent = real_mount(path->mnt);
Al Viro87129cc2011-11-24 21:24:27 -05002094
Al Virob12cea92011-03-18 08:55:38 -04002095 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002096 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04002097 goto out2;
2098
Al Viroe149ed22014-11-01 10:57:28 -04002099 if (!check_mnt(parent))
2100 goto out2;
2101
2102 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
Al Virob12cea92011-03-18 08:55:38 -04002103 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002105 if (!recurse && has_locked_children(old, old_path.dentry))
2106 goto out2;
2107
Al Viroccd48bc2005-11-07 17:15:04 -05002108 if (recurse)
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002109 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
Al Viroccd48bc2005-11-07 17:15:04 -05002110 else
Al Viro87129cc2011-11-24 21:24:27 -05002111 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05002112
David Howellsbe34d1a2012-06-25 12:55:18 +01002113 if (IS_ERR(mnt)) {
2114 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04002115 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01002116 }
Al Viroccd48bc2005-11-07 17:15:04 -05002117
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002118 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2119
Al Viro84d17192013-03-15 10:53:28 -04002120 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002121 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002122 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06002123 umount_tree(mnt, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04002124 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05002125 }
Al Virob12cea92011-03-18 08:55:38 -04002126out2:
Al Viro84d17192013-03-15 10:53:28 -04002127 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002128out:
Al Viro2d92ab32008-08-02 00:51:11 -04002129 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 return err;
2131}
2132
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002133static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2134{
2135 int error = 0;
2136 int readonly_request = 0;
2137
2138 if (ms_flags & MS_RDONLY)
2139 readonly_request = 1;
2140 if (readonly_request == __mnt_is_readonly(mnt))
2141 return 0;
2142
2143 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05002144 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002145 else
Al Viro83adc752011-11-24 22:37:54 -05002146 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002147 return error;
2148}
2149
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150/*
2151 * change filesystem flags. dir should be a physical root of filesystem.
2152 * If you've mounted a non-root directory somewhere and want to do remount
2153 * on it - tough luck.
2154 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002155static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 void *data)
2157{
2158 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04002159 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05002160 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161
Al Viro143c8c92011-11-25 00:46:35 -05002162 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 return -EINVAL;
2164
Al Viro2d92ab32008-08-02 00:51:11 -04002165 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 return -EINVAL;
2167
Eric W. Biederman07b64552014-07-28 17:10:56 -07002168 /* Don't allow changing of locked mnt flags.
2169 *
2170 * No locks need to be held here while testing the various
2171 * MNT_LOCK flags because those flags can never be cleared
2172 * once they are set.
2173 */
2174 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2175 !(mnt_flags & MNT_READONLY)) {
2176 return -EPERM;
2177 }
Eric W. Biederman9566d672014-07-28 17:26:07 -07002178 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2179 !(mnt_flags & MNT_NODEV)) {
Eric W. Biederman3e186642014-08-13 01:33:38 -07002180 /* Was the nodev implicitly added in mount? */
2181 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2182 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2183 mnt_flags |= MNT_NODEV;
2184 } else {
2185 return -EPERM;
2186 }
Eric W. Biederman9566d672014-07-28 17:26:07 -07002187 }
2188 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2189 !(mnt_flags & MNT_NOSUID)) {
2190 return -EPERM;
2191 }
2192 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2193 !(mnt_flags & MNT_NOEXEC)) {
2194 return -EPERM;
2195 }
2196 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2197 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2198 return -EPERM;
2199 }
2200
Eric Parisff36fe22011-03-03 16:09:14 -05002201 err = security_sb_remount(sb, data);
2202 if (err)
2203 return err;
2204
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002206 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002207 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05002208 else if (!capable(CAP_SYS_ADMIN))
2209 err = -EPERM;
Al Viro4aa98cf2009-05-08 13:36:58 -04002210 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002211 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05002212 if (!err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002213 lock_mount_hash();
Eric W. Biedermana6138db2014-07-28 16:26:53 -07002214 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
Al Viro143c8c92011-11-25 00:46:35 -05002215 mnt->mnt.mnt_flags = mnt_flags;
Al Viro143c8c92011-11-25 00:46:35 -05002216 touch_mnt_namespace(mnt->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04002217 unlock_mount_hash();
Dan Williams0e55a7c2008-09-26 19:01:20 -07002218 }
Al Viro6339dab2013-09-16 22:41:01 -04002219 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 return err;
2221}
2222
Al Virocbbe3622011-11-24 20:01:19 -05002223static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05002224{
Al Viro315fc832011-11-24 18:57:30 -05002225 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05002226 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002227 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05002228 return 1;
2229 }
2230 return 0;
2231}
2232
Al Viro808d4e32012-10-11 11:42:01 -04002233static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234{
Al Viro2d92ab32008-08-02 00:51:11 -04002235 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05002236 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05002237 struct mount *old;
Al Viro84d17192013-03-15 10:53:28 -04002238 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002239 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 if (!old_name || !*old_name)
2241 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002242 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 if (err)
2244 return err;
2245
Al Viro84d17192013-03-15 10:53:28 -04002246 mp = lock_mount(path);
2247 err = PTR_ERR(mp);
2248 if (IS_ERR(mp))
David Howellscc53ce52011-01-14 18:45:26 +00002249 goto out;
2250
Al Viro143c8c92011-11-25 00:46:35 -05002251 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002252 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05002253
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002255 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 goto out1;
2257
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002258 if (old->mnt.mnt_flags & MNT_LOCKED)
2259 goto out1;
2260
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002262 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05002263 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
Al Viro676da582011-11-24 21:47:05 -05002265 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05002266 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
David Howellse36cb0b2015-01-29 12:02:35 +00002268 if (d_is_dir(path->dentry) !=
2269 d_is_dir(old_path.dentry))
Ram Pai21444402005-11-07 17:20:03 -05002270 goto out1;
2271 /*
2272 * Don't move a mount residing in a shared parent.
2273 */
Al Virofc7be132011-11-25 01:05:37 -05002274 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002275 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05002276 /*
2277 * Don't move a mount tree containing unbindable mounts to a destination
2278 * mount which is shared.
2279 */
Al Virofc7be132011-11-25 01:05:37 -05002280 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05002281 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05002283 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05002284 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05002285 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
Al Viro84d17192013-03-15 10:53:28 -04002287 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08002288 if (err)
Ram Pai21444402005-11-07 17:20:03 -05002289 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
2291 /* if the mount is moved, it should no longer be expire
2292 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05002293 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294out1:
Al Viro84d17192013-03-15 10:53:28 -04002295 unlock_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04002298 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04002299 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 return err;
2301}
2302
Al Viro9d412a42011-03-17 22:08:28 -04002303static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2304{
2305 int err;
2306 const char *subtype = strchr(fstype, '.');
2307 if (subtype) {
2308 subtype++;
2309 err = -EINVAL;
2310 if (!subtype[0])
2311 goto err;
2312 } else
2313 subtype = "";
2314
2315 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2316 err = -ENOMEM;
2317 if (!mnt->mnt_sb->s_subtype)
2318 goto err;
2319 return mnt;
2320
2321 err:
2322 mntput(mnt);
2323 return ERR_PTR(err);
2324}
2325
Al Viro9d412a42011-03-17 22:08:28 -04002326/*
2327 * add a mount into a namespace's mount tree
2328 */
Al Viro95bc5f22011-11-25 00:30:56 -05002329static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04002330{
Al Viro84d17192013-03-15 10:53:28 -04002331 struct mountpoint *mp;
2332 struct mount *parent;
Al Viro9d412a42011-03-17 22:08:28 -04002333 int err;
2334
Al Virof2ebb3a2014-02-27 09:35:45 -05002335 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04002336
Al Viro84d17192013-03-15 10:53:28 -04002337 mp = lock_mount(path);
2338 if (IS_ERR(mp))
2339 return PTR_ERR(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002340
Al Viro84d17192013-03-15 10:53:28 -04002341 parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04002342 err = -EINVAL;
Al Viro84d17192013-03-15 10:53:28 -04002343 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04002344 /* that's acceptable only for automounts done in private ns */
2345 if (!(mnt_flags & MNT_SHRINKABLE))
2346 goto unlock;
2347 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d17192013-03-15 10:53:28 -04002348 if (!parent->mnt_ns)
Al Viro156cacb2012-09-21 08:19:02 -04002349 goto unlock;
2350 }
Al Viro9d412a42011-03-17 22:08:28 -04002351
2352 /* Refuse the same filesystem on the same mount point */
2353 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05002354 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04002355 path->mnt->mnt_root == path->dentry)
2356 goto unlock;
2357
2358 err = -EINVAL;
David Howellse36cb0b2015-01-29 12:02:35 +00002359 if (d_is_symlink(newmnt->mnt.mnt_root))
Al Viro9d412a42011-03-17 22:08:28 -04002360 goto unlock;
2361
Al Viro95bc5f22011-11-25 00:30:56 -05002362 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro84d17192013-03-15 10:53:28 -04002363 err = graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04002364
2365unlock:
Al Viro84d17192013-03-15 10:53:28 -04002366 unlock_mount(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002367 return err;
2368}
Al Virob1e75df2011-01-17 01:47:59 -05002369
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05002370static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05002371
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372/*
2373 * create a new mount for userspace and request it to be added into the
2374 * namespace's tree
2375 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002376static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04002377 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002379 struct file_system_type *type;
Al Viro9b40bc92013-02-22 22:45:42 -05002380 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002382 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002384 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 return -EINVAL;
2386
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002387 type = get_fs_type(fstype);
2388 if (!type)
2389 return -ENODEV;
2390
2391 if (user_ns != &init_user_ns) {
2392 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2393 put_filesystem(type);
2394 return -EPERM;
2395 }
2396 /* Only in special cases allow devices from mounts
2397 * created outside the initial user namespace.
2398 */
2399 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2400 flags |= MS_NODEV;
Eric W. Biederman9566d672014-07-28 17:26:07 -07002401 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002402 }
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05002403 if (type->fs_flags & FS_USERNS_VISIBLE) {
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05002404 if (!fs_fully_visible(type, &mnt_flags))
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05002405 return -EPERM;
2406 }
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002407 }
2408
2409 mnt = vfs_kern_mount(type, flags, name, data);
2410 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2411 !mnt->mnt_sb->s_subtype)
2412 mnt = fs_set_subtype(mnt, fstype);
2413
2414 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 if (IS_ERR(mnt))
2416 return PTR_ERR(mnt);
2417
Al Viro95bc5f22011-11-25 00:30:56 -05002418 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002419 if (err)
2420 mntput(mnt);
2421 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422}
2423
Al Viro19a167a2011-01-17 01:35:23 -05002424int finish_automount(struct vfsmount *m, struct path *path)
2425{
Al Viro6776db3d2011-11-25 00:22:05 -05002426 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002427 int err;
2428 /* The new mount record should have at least 2 refs to prevent it being
2429 * expired before we get a chance to add it
2430 */
Al Viro6776db3d2011-11-25 00:22:05 -05002431 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002432
2433 if (m->mnt_sb == path->mnt->mnt_sb &&
2434 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002435 err = -ELOOP;
2436 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002437 }
2438
Al Viro95bc5f22011-11-25 00:30:56 -05002439 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002440 if (!err)
2441 return 0;
2442fail:
2443 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002444 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002445 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05002446 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04002447 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002448 }
Al Virob1e75df2011-01-17 01:47:59 -05002449 mntput(m);
2450 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002451 return err;
2452}
2453
David Howellsea5b7782011-01-14 19:10:03 +00002454/**
2455 * mnt_set_expiry - Put a mount on an expiration list
2456 * @mnt: The mount to list.
2457 * @expiry_list: The list to add the mount to.
2458 */
2459void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2460{
Al Viro97216be2013-03-16 15:12:40 -04002461 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00002462
Al Viro6776db3d2011-11-25 00:22:05 -05002463 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002464
Al Viro97216be2013-03-16 15:12:40 -04002465 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002466}
2467EXPORT_SYMBOL(mnt_set_expiry);
2468
2469/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 * process a list of expirable mountpoints with the intent of discarding any
2471 * mountpoints that aren't in use and haven't been touched since last we came
2472 * here
2473 */
2474void mark_mounts_for_expiry(struct list_head *mounts)
2475{
Al Viro761d5c32011-11-24 21:07:43 -05002476 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 LIST_HEAD(graveyard);
2478
2479 if (list_empty(mounts))
2480 return;
2481
Al Viro97216be2013-03-16 15:12:40 -04002482 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002483 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
2485 /* extract from the expiration list every vfsmount that matches the
2486 * following criteria:
2487 * - only referenced by its parent vfsmount
2488 * - still marked for expiry (marked on the last call here; marks are
2489 * cleared by mntput())
2490 */
Al Viro6776db3d2011-11-25 00:22:05 -05002491 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002492 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002493 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002495 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002497 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002498 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002499 touch_mnt_namespace(mnt->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06002500 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04002501 }
Al Viro719ea2f2013-09-29 11:24:49 -04002502 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002503 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504}
2505
2506EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2507
2508/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002509 * Ripoff of 'select_parent()'
2510 *
2511 * search the list of submounts for a given mountpoint, and move any
2512 * shrinkable submounts to the 'graveyard' list.
2513 */
Al Viro692afc32011-11-24 21:15:14 -05002514static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002515{
Al Viro692afc32011-11-24 21:15:14 -05002516 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002517 struct list_head *next;
2518 int found = 0;
2519
2520repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002521 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002522resume:
Al Viro6b41d532011-11-24 23:24:33 -05002523 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002524 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002525 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002526
2527 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002528 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002529 continue;
2530 /*
2531 * Descend a level if the d_mounts list is non-empty.
2532 */
Al Viro6b41d532011-11-24 23:24:33 -05002533 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002534 this_parent = mnt;
2535 goto repeat;
2536 }
2537
Al Viro1ab59732011-11-24 21:35:16 -05002538 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002539 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002540 found++;
2541 }
2542 }
2543 /*
2544 * All done at this level ... ascend and resume the search
2545 */
2546 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002547 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002548 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002549 goto resume;
2550 }
2551 return found;
2552}
2553
2554/*
2555 * process a list of expirable mountpoints with the intent of discarding any
2556 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002557 *
Al Viro48a066e2013-09-29 22:06:07 -04002558 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002559 */
Al Virob54b9be2013-03-16 14:39:34 -04002560static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002561{
2562 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002563 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002564
Trond Myklebust5528f9112006-06-09 09:34:17 -04002565 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002566 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002567 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002568 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002569 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002570 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06002571 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04002572 }
2573 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002574}
2575
Trond Myklebust5528f9112006-06-09 09:34:17 -04002576/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 * Some copy_from_user() implementations do not return the exact number of
2578 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2579 * Note that this function differs from copy_from_user() in that it will oops
2580 * on bad values of `to', rather than returning a short copy.
2581 */
Ram Paib58fed82005-11-07 17:16:09 -05002582static long exact_copy_from_user(void *to, const void __user * from,
2583 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584{
2585 char *t = to;
2586 const char __user *f = from;
2587 char c;
2588
2589 if (!access_ok(VERIFY_READ, from, n))
2590 return n;
2591
2592 while (n) {
2593 if (__get_user(c, f)) {
2594 memset(t, 0, n);
2595 break;
2596 }
2597 *t++ = c;
2598 f++;
2599 n--;
2600 }
2601 return n;
2602}
2603
Ram Paib58fed82005-11-07 17:16:09 -05002604int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605{
2606 int i;
2607 unsigned long page;
2608 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002609
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 *where = 0;
2611 if (!data)
2612 return 0;
2613
2614 if (!(page = __get_free_page(GFP_KERNEL)))
2615 return -ENOMEM;
2616
2617 /* We only care that *some* data at the address the user
2618 * gave us is valid. Just in case, we'll zero
2619 * the remainder of the page.
2620 */
2621 /* copy_from_user cannot cross TASK_SIZE ! */
2622 size = TASK_SIZE - (unsigned long)data;
2623 if (size > PAGE_SIZE)
2624 size = PAGE_SIZE;
2625
2626 i = size - exact_copy_from_user((void *)page, data, size);
2627 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002628 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 return -EFAULT;
2630 }
2631 if (i != PAGE_SIZE)
2632 memset((char *)page + i, 0, PAGE_SIZE - i);
2633 *where = page;
2634 return 0;
2635}
2636
Tim Gardnerb8850d12014-08-28 11:26:03 -06002637char *copy_mount_string(const void __user *data)
Vegard Nossumeca6f532009-09-18 13:05:45 -07002638{
Tim Gardnerb8850d12014-08-28 11:26:03 -06002639 return data ? strndup_user(data, PAGE_SIZE) : NULL;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002640}
2641
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642/*
2643 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2644 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2645 *
2646 * data is a (void *) that can point to any structure up to
2647 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2648 * information (or be NULL).
2649 *
2650 * Pre-0.97 versions of mount() didn't have a flags word.
2651 * When the flags word was introduced its top half was required
2652 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2653 * Therefore, if this magic number is present, it carries no information
2654 * and must be discarded.
2655 */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002656long do_mount(const char *dev_name, const char __user *dir_name,
Al Viro808d4e32012-10-11 11:42:01 -04002657 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658{
Al Viro2d92ab32008-08-02 00:51:11 -04002659 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 int retval = 0;
2661 int mnt_flags = 0;
2662
2663 /* Discard magic */
2664 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2665 flags &= ~MS_MGC_MSK;
2666
2667 /* Basic sanity checks */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 if (data_page)
2669 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2670
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002671 /* ... and get the mountpoint */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002672 retval = user_path(dir_name, &path);
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002673 if (retval)
2674 return retval;
2675
2676 retval = security_sb_mount(dev_name, &path,
2677 type_page, flags, data_page);
Al Viro0d5cadb2013-05-04 14:40:51 -04002678 if (!retval && !may_mount())
2679 retval = -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002680 if (retval)
2681 goto dput_out;
2682
Andi Kleen613cbe32009-04-19 18:40:43 +02002683 /* Default to relatime unless overriden */
2684 if (!(flags & MS_NOATIME))
2685 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002686
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 /* Separate the per-mountpoint flags */
2688 if (flags & MS_NOSUID)
2689 mnt_flags |= MNT_NOSUID;
2690 if (flags & MS_NODEV)
2691 mnt_flags |= MNT_NODEV;
2692 if (flags & MS_NOEXEC)
2693 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002694 if (flags & MS_NOATIME)
2695 mnt_flags |= MNT_NOATIME;
2696 if (flags & MS_NODIRATIME)
2697 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002698 if (flags & MS_STRICTATIME)
2699 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002700 if (flags & MS_RDONLY)
2701 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002702
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07002703 /* The default atime for remount is preservation */
2704 if ((flags & MS_REMOUNT) &&
2705 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2706 MS_STRICTATIME)) == 0)) {
2707 mnt_flags &= ~MNT_ATIME_MASK;
2708 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2709 }
2710
Al Viro7a4dec52010-08-09 12:05:43 -04002711 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002712 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2713 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002716 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 data_page);
2718 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002719 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002720 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002721 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002723 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 else
Al Viro2d92ab32008-08-02 00:51:11 -04002725 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 dev_name, data_page);
2727dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002728 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 return retval;
2730}
2731
Eric W. Biederman771b1372012-07-26 21:08:32 -07002732static void free_mnt_ns(struct mnt_namespace *ns)
2733{
Al Viro6344c432014-11-01 00:45:45 -04002734 ns_free_inum(&ns->ns);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002735 put_user_ns(ns->user_ns);
2736 kfree(ns);
2737}
2738
Eric W. Biederman8823c072010-03-07 18:49:36 -08002739/*
2740 * Assign a sequence number so we can detect when we attempt to bind
2741 * mount a reference to an older mount namespace into the current
2742 * mount namespace, preventing reference counting loops. A 64bit
2743 * number incrementing at 10Ghz will take 12,427 years to wrap which
2744 * is effectively never, so we can ignore the possibility.
2745 */
2746static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2747
Eric W. Biederman771b1372012-07-26 21:08:32 -07002748static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002749{
2750 struct mnt_namespace *new_ns;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002751 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002752
2753 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2754 if (!new_ns)
2755 return ERR_PTR(-ENOMEM);
Al Viro6344c432014-11-01 00:45:45 -04002756 ret = ns_alloc_inum(&new_ns->ns);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002757 if (ret) {
2758 kfree(new_ns);
2759 return ERR_PTR(ret);
2760 }
Al Viro33c42942014-11-01 02:32:53 -04002761 new_ns->ns.ops = &mntns_operations;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002762 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002763 atomic_set(&new_ns->count, 1);
2764 new_ns->root = NULL;
2765 INIT_LIST_HEAD(&new_ns->list);
2766 init_waitqueue_head(&new_ns->poll);
2767 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002768 new_ns->user_ns = get_user_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002769 return new_ns;
2770}
2771
Al Viro9559f682013-09-28 20:47:57 -04002772struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2773 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002775 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002776 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002777 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04002778 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05002779 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002780 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781
Al Viro9559f682013-09-28 20:47:57 -04002782 BUG_ON(!ns);
2783
2784 if (likely(!(flags & CLONE_NEWNS))) {
2785 get_mnt_ns(ns);
2786 return ns;
2787 }
2788
2789 old = ns->root;
2790
Eric W. Biederman771b1372012-07-26 21:08:32 -07002791 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002792 if (IS_ERR(new_ns))
2793 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
Al Viro97216be2013-03-16 15:12:40 -04002795 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002797 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04002798 if (user_ns != ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002799 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002800 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002801 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04002802 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002803 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002804 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 }
Al Virobe08d6d2011-12-06 13:32:36 -05002806 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002807 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808
2809 /*
2810 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2811 * as belonging to new namespace. We have already acquired a private
2812 * fs_struct, so tsk->fs->lock is not needed.
2813 */
Al Viro909b0a82011-11-25 03:06:56 -05002814 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002815 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002817 q->mnt_ns = new_ns;
Al Viro9559f682013-09-28 20:47:57 -04002818 if (new_fs) {
2819 if (&p->mnt == new_fs->root.mnt) {
2820 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002821 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 }
Al Viro9559f682013-09-28 20:47:57 -04002823 if (&p->mnt == new_fs->pwd.mnt) {
2824 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002825 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 }
Al Viro909b0a82011-11-25 03:06:56 -05002828 p = next_mnt(p, old);
2829 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002830 if (!q)
2831 break;
2832 while (p->mnt.mnt_root != q->mnt.mnt_root)
2833 p = next_mnt(p, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 }
Al Viro328e6d92013-03-16 14:49:45 -04002835 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002838 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002840 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841
JANAK DESAI741a2952006-02-07 12:59:00 -08002842 return new_ns;
2843}
2844
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002845/**
2846 * create_mnt_ns - creates a private namespace and adds a root filesystem
2847 * @mnt: pointer to the new root filesystem mountpoint
2848 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002849static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002850{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002851 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002852 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002853 struct mount *mnt = real_mount(m);
2854 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002855 new_ns->root = mnt;
Al Virob1983cd2013-05-04 15:18:53 -04002856 list_add(&mnt->mnt_list, &new_ns->list);
Al Viroc1334492011-11-16 16:12:14 -05002857 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002858 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002859 }
2860 return new_ns;
2861}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002862
Al Viroea441d12011-11-16 21:43:59 -05002863struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2864{
2865 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002866 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002867 struct path path;
2868 int err;
2869
2870 ns = create_mnt_ns(mnt);
2871 if (IS_ERR(ns))
2872 return ERR_CAST(ns);
2873
2874 err = vfs_path_lookup(mnt->mnt_root, mnt,
2875 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2876
2877 put_mnt_ns(ns);
2878
2879 if (err)
2880 return ERR_PTR(err);
2881
2882 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002883 s = path.mnt->mnt_sb;
2884 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002885 mntput(path.mnt);
2886 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002887 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002888 /* ... and return the root of (sub)tree on it */
2889 return path.dentry;
2890}
2891EXPORT_SYMBOL(mount_subtree);
2892
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002893SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2894 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002896 int ret;
2897 char *kernel_type;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002898 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900
Tim Gardnerb8850d12014-08-28 11:26:03 -06002901 kernel_type = copy_mount_string(type);
2902 ret = PTR_ERR(kernel_type);
2903 if (IS_ERR(kernel_type))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002904 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905
Tim Gardnerb8850d12014-08-28 11:26:03 -06002906 kernel_dev = copy_mount_string(dev_name);
2907 ret = PTR_ERR(kernel_dev);
2908 if (IS_ERR(kernel_dev))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002909 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910
Vegard Nossumeca6f532009-09-18 13:05:45 -07002911 ret = copy_mount_options(data, &data_page);
2912 if (ret < 0)
2913 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002915 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
Vegard Nossumeca6f532009-09-18 13:05:45 -07002916 (void *) data_page);
2917
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002919out_data:
2920 kfree(kernel_dev);
2921out_dev:
Vegard Nossumeca6f532009-09-18 13:05:45 -07002922 kfree(kernel_type);
2923out_type:
2924 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925}
2926
2927/*
Al Viroafac7cb2011-11-23 19:34:49 -05002928 * Return true if path is reachable from root
2929 *
Al Viro48a066e2013-09-29 22:06:07 -04002930 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05002931 */
Al Viro643822b2011-11-24 22:00:28 -05002932bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002933 const struct path *root)
2934{
Al Viro643822b2011-11-24 22:00:28 -05002935 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002936 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002937 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002938 }
Al Viro643822b2011-11-24 22:00:28 -05002939 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002940}
2941
2942int path_is_under(struct path *path1, struct path *path2)
2943{
2944 int res;
Al Viro48a066e2013-09-29 22:06:07 -04002945 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002946 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04002947 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002948 return res;
2949}
2950EXPORT_SYMBOL(path_is_under);
2951
2952/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 * pivot_root Semantics:
2954 * Moves the root file system of the current process to the directory put_old,
2955 * makes new_root as the new root file system of the current process, and sets
2956 * root/cwd of all processes which had them on the current root to new_root.
2957 *
2958 * Restrictions:
2959 * The new_root and put_old must be directories, and must not be on the
2960 * same file system as the current process root. The put_old must be
2961 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2962 * pointed to by put_old must yield the same directory as new_root. No other
2963 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2964 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002965 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2966 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2967 * in this situation.
2968 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 * Notes:
2970 * - we don't move root/cwd if they are not at the root (reason: if something
2971 * cared enough to change them, it's probably wrong to force them elsewhere)
2972 * - it's okay to pick a root that isn't the root of a file system, e.g.
2973 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2974 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2975 * first.
2976 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002977SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2978 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979{
Al Viro2d8f3032008-07-22 09:59:21 -04002980 struct path new, old, parent_path, root_parent, root;
Al Viro84d17192013-03-15 10:53:28 -04002981 struct mount *new_mnt, *root_mnt, *old_mnt;
2982 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 int error;
2984
Al Viro9b40bc92013-02-22 22:45:42 -05002985 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 return -EPERM;
2987
Al Viro2d8f3032008-07-22 09:59:21 -04002988 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 if (error)
2990 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991
Al Viro2d8f3032008-07-22 09:59:21 -04002992 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 if (error)
2994 goto out1;
2995
Al Viro2d8f3032008-07-22 09:59:21 -04002996 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002997 if (error)
2998 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02003000 get_fs_root(current->fs, &root);
Al Viro84d17192013-03-15 10:53:28 -04003001 old_mp = lock_mount(&old);
3002 error = PTR_ERR(old_mp);
3003 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04003004 goto out3;
3005
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05003007 new_mnt = real_mount(new.mnt);
3008 root_mnt = real_mount(root.mnt);
Al Viro84d17192013-03-15 10:53:28 -04003009 old_mnt = real_mount(old.mnt);
3010 if (IS_MNT_SHARED(old_mnt) ||
Al Virofc7be132011-11-25 01:05:37 -05003011 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3012 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04003013 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05003014 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003015 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003016 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3017 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003019 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04003020 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 error = -EBUSY;
Al Viro84d17192013-03-15 10:53:28 -04003022 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04003023 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04003025 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04003026 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05003027 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003028 goto out4; /* not attached */
Al Viro84d17192013-03-15 10:53:28 -04003029 root_mp = root_mnt->mnt_mp;
Al Viro2d8f3032008-07-22 09:59:21 -04003030 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04003031 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05003032 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003033 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08003034 /* make sure we can reach put_old from new_root */
Al Viro84d17192013-03-15 10:53:28 -04003035 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04003036 goto out4;
Eric W. Biederman0d082602014-10-08 10:42:27 -07003037 /* make certain new is below the root */
3038 if (!is_path_reachable(new_mnt, new.dentry, &root))
3039 goto out4;
Al Viro84d17192013-03-15 10:53:28 -04003040 root_mp->m_count++; /* pin it so it won't go away */
Al Viro719ea2f2013-09-29 11:24:49 -04003041 lock_mount_hash();
Al Viro419148d2011-11-24 19:41:16 -05003042 detach_mnt(new_mnt, &parent_path);
3043 detach_mnt(root_mnt, &root_parent);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003044 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3045 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3046 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3047 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003048 /* mount old root on put_old */
Al Viro84d17192013-03-15 10:53:28 -04003049 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08003050 /* mount new_root on / */
Al Viro84d17192013-03-15 10:53:28 -04003051 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003052 touch_mnt_namespace(current->nsproxy->mnt_ns);
Eric W. Biederman4fed6552014-10-08 10:42:57 -07003053 /* A moved mount should not expire automatically */
3054 list_del_init(&new_mnt->mnt_expire);
Al Viro719ea2f2013-09-29 11:24:49 -04003055 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04003056 chroot_fs_refs(&root, &new);
Al Viro84d17192013-03-15 10:53:28 -04003057 put_mountpoint(root_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04003059out4:
Al Viro84d17192013-03-15 10:53:28 -04003060 unlock_mount(old_mp);
Al Virob12cea92011-03-18 08:55:38 -04003061 if (!error) {
3062 path_put(&root_parent);
3063 path_put(&parent_path);
3064 }
3065out3:
Al Viro8c3ee422008-03-22 18:00:39 -04003066 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04003067out2:
Al Viro2d8f3032008-07-22 09:59:21 -04003068 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069out1:
Al Viro2d8f3032008-07-22 09:59:21 -04003070 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073}
3074
3075static void __init init_mount_tree(void)
3076{
3077 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003078 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08003079 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003080 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003082 type = get_fs_type("rootfs");
3083 if (!type)
3084 panic("Can't find rootfs type");
3085 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3086 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087 if (IS_ERR(mnt))
3088 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11003089
Trond Myklebust3b22edc2009-06-23 17:29:49 -04003090 ns = create_mnt_ns(mnt);
3091 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003094 init_task.nsproxy->mnt_ns = ns;
3095 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096
Al Virobe08d6d2011-12-06 13:32:36 -05003097 root.mnt = mnt;
3098 root.dentry = mnt->mnt_root;
Eric W. Biedermanda362b02014-10-07 12:19:53 -07003099 mnt->mnt_flags |= MNT_LOCKED;
Jan Blunckac748a02008-02-14 19:34:39 -08003100
3101 set_fs_pwd(current->fs, &root);
3102 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103}
3104
Denis Cheng74bf17c2007-10-16 23:26:30 -07003105void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106{
Eric Dumazet13f14b42008-02-06 01:37:57 -08003107 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003108 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109
Al Viro7d6fec42011-11-23 12:14:10 -05003110 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09003111 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112
Al Viro0818bf22014-02-28 13:46:44 -05003113 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04003114 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05003115 mhash_entries, 19,
3116 0,
3117 &m_hash_shift, &m_hash_mask, 0, 0);
3118 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3119 sizeof(struct hlist_head),
3120 mphash_entries, 19,
3121 0,
3122 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123
Al Viro84d17192013-03-15 10:53:28 -04003124 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125 panic("Failed to allocate mount hash table\n");
3126
Al Viro0818bf22014-02-28 13:46:44 -05003127 for (u = 0; u <= m_hash_mask; u++)
Al Viro38129a12014-03-20 21:10:51 -04003128 INIT_HLIST_HEAD(&mount_hashtable[u]);
Al Viro0818bf22014-02-28 13:46:44 -05003129 for (u = 0; u <= mp_hash_mask; u++)
3130 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131
Tejun Heo4b93dc92013-11-28 14:54:43 -05003132 kernfs_init();
3133
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003134 err = sysfs_init();
3135 if (err)
3136 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003137 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06003138 fs_kobj = kobject_create_and_add("fs", NULL);
3139 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003140 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141 init_rootfs();
3142 init_mount_tree();
3143}
3144
Trond Myklebust616511d2009-06-22 15:09:13 -04003145void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146{
Al Virod498b252010-02-05 02:21:06 -05003147 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04003148 return;
Al Viro7b00ed62013-09-16 21:19:20 -04003149 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003150 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151}
Al Viro9d412a42011-03-17 22:08:28 -04003152
3153struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3154{
Tim Chen423e0ab2011-07-19 09:32:38 -07003155 struct vfsmount *mnt;
3156 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3157 if (!IS_ERR(mnt)) {
3158 /*
3159 * it is a longterm mount, don't release mnt until
3160 * we unmount before file sys is unregistered
3161 */
Al Virof7a99c52012-06-09 00:59:08 -04003162 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07003163 }
3164 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04003165}
3166EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07003167
3168void kern_unmount(struct vfsmount *mnt)
3169{
3170 /* release long term mount so mount point can be released */
3171 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04003172 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04003173 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07003174 mntput(mnt);
3175 }
3176}
3177EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05003178
3179bool our_mnt(struct vfsmount *mnt)
3180{
Al Viro143c8c92011-11-25 00:46:35 -05003181 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05003182}
Eric W. Biederman8823c072010-03-07 18:49:36 -08003183
Eric W. Biederman31515272013-03-15 01:45:51 -07003184bool current_chrooted(void)
3185{
3186 /* Does the current process have a non-standard root */
3187 struct path ns_root;
3188 struct path fs_root;
3189 bool chrooted;
3190
3191 /* Find the namespace root */
3192 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3193 ns_root.dentry = ns_root.mnt->mnt_root;
3194 path_get(&ns_root);
3195 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3196 ;
3197
3198 get_fs_root(current->fs, &fs_root);
3199
3200 chrooted = !path_equal(&fs_root, &ns_root);
3201
3202 path_put(&fs_root);
3203 path_put(&ns_root);
3204
3205 return chrooted;
3206}
3207
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003208static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003209{
3210 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003211 int new_flags = *new_mnt_flags;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003212 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07003213 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003214
Eric W. Biedermane51db732013-03-30 19:57:41 -07003215 if (unlikely(!ns))
3216 return false;
3217
Al Viro44bb4382013-09-16 21:37:36 -04003218 down_read(&namespace_sem);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003219 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07003220 struct mount *child;
3221 if (mnt->mnt.mnt_sb->s_type != type)
3222 continue;
3223
Eric W. Biederman7e96c1b2015-05-08 16:36:50 -05003224 /* This mount is not fully visible if it's root directory
3225 * is not the root directory of the filesystem.
3226 */
3227 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3228 continue;
3229
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003230 /* Verify the mount flags are equal to or more permissive
3231 * than the proposed new mount.
3232 */
3233 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
3234 !(new_flags & MNT_READONLY))
3235 continue;
3236 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
3237 !(new_flags & MNT_NODEV))
3238 continue;
3239 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
3240 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3241 continue;
3242
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003243 /* This mount is not fully visible if there are any
3244 * locked child mounts that cover anything except for
3245 * empty directories.
Eric W. Biedermane51db732013-03-30 19:57:41 -07003246 */
3247 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3248 struct inode *inode = child->mnt_mountpoint->d_inode;
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003249 /* Only worry about locked mounts */
3250 if (!(mnt->mnt.mnt_flags & MNT_LOCKED))
3251 continue;
Eric W. Biederman7236c852015-05-13 20:51:09 -05003252 /* Is the directory permanetly empty? */
3253 if (!is_empty_dir_inode(inode))
Eric W. Biedermane51db732013-03-30 19:57:41 -07003254 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003255 }
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003256 /* Preserve the locked attributes */
3257 *new_mnt_flags |= mnt->mnt.mnt_flags & (MNT_LOCK_READONLY | \
3258 MNT_LOCK_NODEV | \
3259 MNT_LOCK_ATIME);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003260 visible = true;
3261 goto found;
3262 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003263 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07003264found:
Al Viro44bb4382013-09-16 21:37:36 -04003265 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003266 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003267}
3268
Al Viro64964522014-11-01 00:37:32 -04003269static struct ns_common *mntns_get(struct task_struct *task)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003270{
Al Viro58be28252014-11-01 00:00:23 -04003271 struct ns_common *ns = NULL;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003272 struct nsproxy *nsproxy;
3273
Eric W. Biederman728dba32014-02-03 19:13:49 -08003274 task_lock(task);
3275 nsproxy = task->nsproxy;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003276 if (nsproxy) {
Al Viro58be28252014-11-01 00:00:23 -04003277 ns = &nsproxy->mnt_ns->ns;
3278 get_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003279 }
Eric W. Biederman728dba32014-02-03 19:13:49 -08003280 task_unlock(task);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003281
3282 return ns;
3283}
3284
Al Viro64964522014-11-01 00:37:32 -04003285static void mntns_put(struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003286{
Al Viro58be28252014-11-01 00:00:23 -04003287 put_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003288}
3289
Al Viro64964522014-11-01 00:37:32 -04003290static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003291{
3292 struct fs_struct *fs = current->fs;
Al Viro58be28252014-11-01 00:00:23 -04003293 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003294 struct path root;
3295
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003296 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -07003297 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3298 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08003299 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003300
3301 if (fs->users != 1)
3302 return -EINVAL;
3303
3304 get_mnt_ns(mnt_ns);
3305 put_mnt_ns(nsproxy->mnt_ns);
3306 nsproxy->mnt_ns = mnt_ns;
3307
3308 /* Find the root */
3309 root.mnt = &mnt_ns->root->mnt;
3310 root.dentry = mnt_ns->root->mnt.mnt_root;
3311 path_get(&root);
3312 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3313 ;
3314
3315 /* Update the pwd and root */
3316 set_fs_pwd(fs, &root);
3317 set_fs_root(fs, &root);
3318
3319 path_put(&root);
3320 return 0;
3321}
3322
3323const struct proc_ns_operations mntns_operations = {
3324 .name = "mnt",
3325 .type = CLONE_NEWNS,
3326 .get = mntns_get,
3327 .put = mntns_put,
3328 .install = mntns_install,
3329};