blob: e6c234b1a6456c1364bcc06b029f3cc4a155d5bd [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
Al Virod10577a2011-12-07 13:06:11 -050012#include <linux/export.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080013#include <linux/capability.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080014#include <linux/mnt_namespace.h>
Eric W. Biederman771b1372012-07-26 21:08:32 -070015#include <linux/user_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/namei.h>
17#include <linux/security.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010018#include <linux/idr.h>
Rob Landley57f150a2013-09-11 14:26:10 -070019#include <linux/init.h> /* init_rootfs */
Al Virod10577a2011-12-07 13:06:11 -050020#include <linux/fs_struct.h> /* get_fs_root et.al. */
21#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22#include <linux/uaccess.h>
David Howells0bb80f22013-04-12 01:50:06 +010023#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070024#include <linux/magic.h>
Al Viro0818bf22014-02-28 13:46:44 -050025#include <linux/bootmem.h>
Al Viro9ea459e2014-08-08 13:08:20 -040026#include <linux/task_work.h>
Ram Pai07b20882005-11-07 17:19:07 -050027#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070028#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Eric W. Biedermand2921682016-09-28 00:27:17 -050030/* Maximum number of mounts in a mount namespace */
31unsigned int sysctl_mount_max __read_mostly = 100000;
32
Al Viro0818bf22014-02-28 13:46:44 -050033static unsigned int m_hash_mask __read_mostly;
34static unsigned int m_hash_shift __read_mostly;
35static unsigned int mp_hash_mask __read_mostly;
36static unsigned int mp_hash_shift __read_mostly;
37
38static __initdata unsigned long mhash_entries;
39static int __init set_mhash_entries(char *str)
40{
41 if (!str)
42 return 0;
43 mhash_entries = simple_strtoul(str, &str, 0);
44 return 1;
45}
46__setup("mhash_entries=", set_mhash_entries);
47
48static __initdata unsigned long mphash_entries;
49static int __init set_mphash_entries(char *str)
50{
51 if (!str)
52 return 0;
53 mphash_entries = simple_strtoul(str, &str, 0);
54 return 1;
55}
56__setup("mphash_entries=", set_mphash_entries);
Eric Dumazet13f14b42008-02-06 01:37:57 -080057
Al Viroc7999c32014-02-27 14:40:10 -050058static u64 event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010059static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010060static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100061static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040062static int mnt_id_start = 0;
63static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Al Viro38129a12014-03-20 21:10:51 -040065static struct hlist_head *mount_hashtable __read_mostly;
Al Viro0818bf22014-02-28 13:46:44 -050066static struct hlist_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080067static struct kmem_cache *mnt_cache __read_mostly;
Al Viro59aa0da2013-09-16 21:34:53 -040068static DECLARE_RWSEM(namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080070/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060071struct kobject *fs_kobj;
72EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080073
Nick Piggin99b7db72010-08-18 04:37:39 +100074/*
75 * vfsmount lock may be taken for read to prevent changes to the
76 * vfsmount hash, ie. during mountpoint lookups or walking back
77 * up the tree.
78 *
79 * It should be taken for write in all cases where the vfsmount
80 * tree or hash is modified or when a vfsmount structure is modified.
81 */
Al Viro48a066e2013-09-29 22:06:07 -040082__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +100083
Al Viro38129a12014-03-20 21:10:51 -040084static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085{
Ram Paib58fed82005-11-07 17:16:09 -050086 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
87 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Al Viro0818bf22014-02-28 13:46:44 -050088 tmp = tmp + (tmp >> m_hash_shift);
89 return &mount_hashtable[tmp & m_hash_mask];
90}
91
92static inline struct hlist_head *mp_hash(struct dentry *dentry)
93{
94 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
95 tmp = tmp + (tmp >> mp_hash_shift);
96 return &mountpoint_hashtable[tmp & mp_hash_mask];
Linus Torvalds1da177e2005-04-16 15:20:36 -070097}
98
Nick Piggin99b7db72010-08-18 04:37:39 +100099/*
100 * allocation is serialized by namespace_sem, but we need the spinlock to
101 * serialize with freeing.
102 */
Al Virob105e272011-11-24 20:38:33 -0500103static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100104{
105 int res;
106
107retry:
108 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +1000109 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -0500110 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -0400111 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500112 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +1000113 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100114 if (res == -EAGAIN)
115 goto retry;
116
117 return res;
118}
119
Al Virob105e272011-11-24 20:38:33 -0500120static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100121{
Al Viro15169fe2011-11-25 00:50:41 -0500122 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000123 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -0400124 ida_remove(&mnt_id_ida, id);
125 if (mnt_id_start > id)
126 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000127 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100128}
129
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100130/*
131 * Allocate a new peer group ID
132 *
133 * mnt_group_ida is protected by namespace_sem
134 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500135static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100136{
Al Virof21f6222009-06-24 03:12:00 -0400137 int res;
138
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100139 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
140 return -ENOMEM;
141
Al Virof21f6222009-06-24 03:12:00 -0400142 res = ida_get_new_above(&mnt_group_ida,
143 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500144 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400145 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500146 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400147
148 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100149}
150
151/*
152 * Release a peer group ID
153 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500154void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100155{
Al Viro15169fe2011-11-25 00:50:41 -0500156 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400157 ida_remove(&mnt_group_ida, id);
158 if (mnt_group_start > id)
159 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500160 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100161}
162
Nick Pigginb3e19d92011-01-07 17:50:11 +1100163/*
164 * vfsmount lock must be held for read
165 */
Al Viro83adc752011-11-24 22:37:54 -0500166static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100167{
168#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500169 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100170#else
171 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500172 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100173 preempt_enable();
174#endif
175}
176
Nick Pigginb3e19d92011-01-07 17:50:11 +1100177/*
178 * vfsmount lock must be held for write
179 */
Al Viro83adc752011-11-24 22:37:54 -0500180unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100181{
182#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500183 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100184 int cpu;
185
186 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500187 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100188 }
189
190 return count;
191#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500192 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100193#endif
194}
195
Al Viro87b95ce2015-01-10 19:01:08 -0500196static void drop_mountpoint(struct fs_pin *p)
197{
198 struct mount *m = container_of(p, struct mount, mnt_umount);
199 dput(m->mnt_ex_mountpoint);
200 pin_remove(p);
201 mntput(&m->mnt);
202}
203
Al Virob105e272011-11-24 20:38:33 -0500204static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205{
Al Viroc63181e2011-11-25 02:35:16 -0500206 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
207 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100208 int err;
209
Al Viroc63181e2011-11-25 02:35:16 -0500210 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800211 if (err)
212 goto out_free_cache;
213
214 if (name) {
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800215 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
Al Viroc63181e2011-11-25 02:35:16 -0500216 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800217 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100218 }
219
Nick Pigginb3e19d92011-01-07 17:50:11 +1100220#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500221 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
222 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100223 goto out_free_devname;
224
Al Viroc63181e2011-11-25 02:35:16 -0500225 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100226#else
Al Viroc63181e2011-11-25 02:35:16 -0500227 mnt->mnt_count = 1;
228 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100229#endif
230
Al Viro38129a12014-03-20 21:10:51 -0400231 INIT_HLIST_NODE(&mnt->mnt_hash);
Al Viroc63181e2011-11-25 02:35:16 -0500232 INIT_LIST_HEAD(&mnt->mnt_child);
233 INIT_LIST_HEAD(&mnt->mnt_mounts);
234 INIT_LIST_HEAD(&mnt->mnt_list);
235 INIT_LIST_HEAD(&mnt->mnt_expire);
236 INIT_LIST_HEAD(&mnt->mnt_share);
237 INIT_LIST_HEAD(&mnt->mnt_slave_list);
238 INIT_LIST_HEAD(&mnt->mnt_slave);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700239 INIT_HLIST_NODE(&mnt->mnt_mp_list);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500240#ifdef CONFIG_FSNOTIFY
241 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
242#endif
Al Viro87b95ce2015-01-10 19:01:08 -0500243 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 }
Al Viroc63181e2011-11-25 02:35:16 -0500245 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800246
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000247#ifdef CONFIG_SMP
248out_free_devname:
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800249 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000250#endif
Li Zefan88b38782008-07-21 18:06:36 +0800251out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500252 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800253out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500254 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800255 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
257
Dave Hansen83660252008-02-15 14:37:30 -0800258/*
259 * Most r/o checks on a fs are for operations that take
260 * discrete amounts of time, like a write() or unlink().
261 * We must keep track of when those operations start
262 * (for permission checks) and when they end, so that
263 * we can determine when writes are able to occur to
264 * a filesystem.
265 */
Dave Hansen3d733632008-02-15 14:37:59 -0800266/*
267 * __mnt_is_readonly: check whether a mount is read-only
268 * @mnt: the mount to check for its write status
269 *
270 * This shouldn't be used directly ouside of the VFS.
271 * It does not guarantee that the filesystem will stay
272 * r/w, just that it is right *now*. This can not and
273 * should not be used in place of IS_RDONLY(inode).
274 * mnt_want/drop_write() will _keep_ the filesystem
275 * r/w.
276 */
277int __mnt_is_readonly(struct vfsmount *mnt)
278{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800279 if (mnt->mnt_flags & MNT_READONLY)
280 return 1;
281 if (mnt->mnt_sb->s_flags & MS_RDONLY)
282 return 1;
283 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800284}
285EXPORT_SYMBOL_GPL(__mnt_is_readonly);
286
Al Viro83adc752011-11-24 22:37:54 -0500287static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800288{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000289#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500290 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000291#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500292 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000293#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800294}
Dave Hansen3d733632008-02-15 14:37:59 -0800295
Al Viro83adc752011-11-24 22:37:54 -0500296static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800297{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000298#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500299 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000300#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500301 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000302#endif
303}
304
Al Viro83adc752011-11-24 22:37:54 -0500305static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000306{
307#ifdef CONFIG_SMP
308 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800309 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800310
311 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500312 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800313 }
Dave Hansen3d733632008-02-15 14:37:59 -0800314
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000315 return count;
316#else
317 return mnt->mnt_writers;
318#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800319}
320
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100321static int mnt_is_readonly(struct vfsmount *mnt)
322{
323 if (mnt->mnt_sb->s_readonly_remount)
324 return 1;
325 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
326 smp_rmb();
327 return __mnt_is_readonly(mnt);
328}
329
Dave Hansen3d733632008-02-15 14:37:59 -0800330/*
Jan Karaeb04c282012-06-12 16:20:35 +0200331 * Most r/o & frozen checks on a fs are for operations that take discrete
332 * amounts of time, like a write() or unlink(). We must keep track of when
333 * those operations start (for permission checks) and when they end, so that we
334 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800335 */
Dave Hansen83660252008-02-15 14:37:30 -0800336/**
Jan Karaeb04c282012-06-12 16:20:35 +0200337 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500338 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800339 *
Jan Karaeb04c282012-06-12 16:20:35 +0200340 * This tells the low-level filesystem that a write is about to be performed to
341 * it, and makes sure that writes are allowed (mnt it read-write) before
342 * returning success. This operation does not protect against filesystem being
343 * frozen. When the write operation is finished, __mnt_drop_write() must be
344 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800345 */
Jan Karaeb04c282012-06-12 16:20:35 +0200346int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800347{
Al Viro83adc752011-11-24 22:37:54 -0500348 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800349 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800350
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000351 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100352 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000353 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100354 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000355 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
356 * incremented count after it has set MNT_WRITE_HOLD.
357 */
358 smp_mb();
Miao Xie1e755292012-11-16 17:23:50 +0800359 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000360 cpu_relax();
361 /*
362 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
363 * be set to match its requirements. So we must not load that until
364 * MNT_WRITE_HOLD is cleared.
365 */
366 smp_rmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100367 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100368 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800369 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800370 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000371 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200372
373 return ret;
374}
375
376/**
377 * mnt_want_write - get write access to a mount
378 * @m: the mount on which to take a write
379 *
380 * This tells the low-level filesystem that a write is about to be performed to
381 * it, and makes sure that writes are allowed (mount is read-write, filesystem
382 * is not frozen) before returning success. When the write operation is
383 * finished, mnt_drop_write() must be called. This is effectively a refcount.
384 */
385int mnt_want_write(struct vfsmount *m)
386{
387 int ret;
388
389 sb_start_write(m->mnt_sb);
390 ret = __mnt_want_write(m);
391 if (ret)
392 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800393 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800394}
395EXPORT_SYMBOL_GPL(mnt_want_write);
396
397/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000398 * mnt_clone_write - get write access to a mount
399 * @mnt: the mount on which to take a write
400 *
401 * This is effectively like mnt_want_write, except
402 * it must only be used to take an extra write reference
403 * on a mountpoint that we already know has a write reference
404 * on it. This allows some optimisation.
405 *
406 * After finished, mnt_drop_write must be called as usual to
407 * drop the reference.
408 */
409int mnt_clone_write(struct vfsmount *mnt)
410{
411 /* superblock may be r/o */
412 if (__mnt_is_readonly(mnt))
413 return -EROFS;
414 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500415 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000416 preempt_enable();
417 return 0;
418}
419EXPORT_SYMBOL_GPL(mnt_clone_write);
420
421/**
Jan Karaeb04c282012-06-12 16:20:35 +0200422 * __mnt_want_write_file - get write access to a file's mount
423 * @file: the file who's mount on which to take a write
424 *
425 * This is like __mnt_want_write, but it takes a file and can
426 * do some optimisations if the file is open for write already
427 */
428int __mnt_want_write_file(struct file *file)
429{
Al Viro83f936c2014-03-14 12:02:47 -0400430 if (!(file->f_mode & FMODE_WRITER))
Jan Karaeb04c282012-06-12 16:20:35 +0200431 return __mnt_want_write(file->f_path.mnt);
432 else
433 return mnt_clone_write(file->f_path.mnt);
434}
435
436/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000437 * mnt_want_write_file - get write access to a file's mount
438 * @file: the file who's mount on which to take a write
439 *
440 * This is like mnt_want_write, but it takes a file and can
441 * do some optimisations if the file is open for write already
442 */
443int mnt_want_write_file(struct file *file)
444{
Jan Karaeb04c282012-06-12 16:20:35 +0200445 int ret;
446
447 sb_start_write(file->f_path.mnt->mnt_sb);
448 ret = __mnt_want_write_file(file);
449 if (ret)
450 sb_end_write(file->f_path.mnt->mnt_sb);
451 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000452}
453EXPORT_SYMBOL_GPL(mnt_want_write_file);
454
455/**
Jan Karaeb04c282012-06-12 16:20:35 +0200456 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800457 * @mnt: the mount on which to give up write access
458 *
459 * Tells the low-level filesystem that we are done
460 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200461 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800462 */
Jan Karaeb04c282012-06-12 16:20:35 +0200463void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800464{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000465 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500466 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000467 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800468}
Jan Karaeb04c282012-06-12 16:20:35 +0200469
470/**
471 * mnt_drop_write - give up write access to a mount
472 * @mnt: the mount on which to give up write access
473 *
474 * Tells the low-level filesystem that we are done performing writes to it and
475 * also allows filesystem to be frozen again. Must be matched with
476 * mnt_want_write() call above.
477 */
478void mnt_drop_write(struct vfsmount *mnt)
479{
480 __mnt_drop_write(mnt);
481 sb_end_write(mnt->mnt_sb);
482}
Dave Hansen83660252008-02-15 14:37:30 -0800483EXPORT_SYMBOL_GPL(mnt_drop_write);
484
Jan Karaeb04c282012-06-12 16:20:35 +0200485void __mnt_drop_write_file(struct file *file)
486{
487 __mnt_drop_write(file->f_path.mnt);
488}
489
Al Viro2a79f172011-12-09 08:06:57 -0500490void mnt_drop_write_file(struct file *file)
491{
492 mnt_drop_write(file->f_path.mnt);
493}
494EXPORT_SYMBOL(mnt_drop_write_file);
495
Al Viro83adc752011-11-24 22:37:54 -0500496static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800497{
Dave Hansen3d733632008-02-15 14:37:59 -0800498 int ret = 0;
499
Al Viro719ea2f2013-09-29 11:24:49 -0400500 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500501 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000502 /*
503 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
504 * should be visible before we do.
505 */
506 smp_mb();
507
508 /*
509 * With writers on hold, if this value is zero, then there are
510 * definitely no active writers (although held writers may subsequently
511 * increment the count, they'll have to wait, and decrement it after
512 * seeing MNT_READONLY).
513 *
514 * It is OK to have counter incremented on one CPU and decremented on
515 * another: the sum will add up correctly. The danger would be when we
516 * sum up each counter, if we read a counter before it is incremented,
517 * but then read another CPU's count which it has been subsequently
518 * decremented from -- we would see more decrements than we should.
519 * MNT_WRITE_HOLD protects against this scenario, because
520 * mnt_want_write first increments count, then smp_mb, then spins on
521 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
522 * we're counting up here.
523 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100524 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000525 ret = -EBUSY;
526 else
Al Viro83adc752011-11-24 22:37:54 -0500527 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000528 /*
529 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
530 * that become unheld will see MNT_READONLY.
531 */
532 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500533 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Al Viro719ea2f2013-09-29 11:24:49 -0400534 unlock_mount_hash();
Dave Hansen3d733632008-02-15 14:37:59 -0800535 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800536}
Dave Hansen83660252008-02-15 14:37:30 -0800537
Al Viro83adc752011-11-24 22:37:54 -0500538static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800539{
Al Viro719ea2f2013-09-29 11:24:49 -0400540 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500541 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Al Viro719ea2f2013-09-29 11:24:49 -0400542 unlock_mount_hash();
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800543}
544
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100545int sb_prepare_remount_readonly(struct super_block *sb)
546{
547 struct mount *mnt;
548 int err = 0;
549
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100550 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
551 if (atomic_long_read(&sb->s_remove_count))
552 return -EBUSY;
553
Al Viro719ea2f2013-09-29 11:24:49 -0400554 lock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100555 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
556 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
557 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
558 smp_mb();
559 if (mnt_get_writers(mnt) > 0) {
560 err = -EBUSY;
561 break;
562 }
563 }
564 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100565 if (!err && atomic_long_read(&sb->s_remove_count))
566 err = -EBUSY;
567
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100568 if (!err) {
569 sb->s_readonly_remount = 1;
570 smp_wmb();
571 }
572 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
573 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
574 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
575 }
Al Viro719ea2f2013-09-29 11:24:49 -0400576 unlock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100577
578 return err;
579}
580
Al Virob105e272011-11-24 20:38:33 -0500581static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800583 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000584#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500585 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000586#endif
Al Virob105e272011-11-24 20:38:33 -0500587 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588}
589
David Howells8ffcb322014-01-24 12:17:54 +0000590static void delayed_free_vfsmnt(struct rcu_head *head)
591{
592 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
593}
594
Al Viro48a066e2013-09-29 22:06:07 -0400595/* call under rcu_read_lock */
Al Viro294d71f2015-05-08 11:43:53 -0400596int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
Al Viro48a066e2013-09-29 22:06:07 -0400597{
598 struct mount *mnt;
599 if (read_seqretry(&mount_lock, seq))
Al Viro294d71f2015-05-08 11:43:53 -0400600 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400601 if (bastard == NULL)
Al Viro294d71f2015-05-08 11:43:53 -0400602 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400603 mnt = real_mount(bastard);
604 mnt_add_count(mnt, 1);
605 if (likely(!read_seqretry(&mount_lock, seq)))
Al Viro294d71f2015-05-08 11:43:53 -0400606 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400607 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
608 mnt_add_count(mnt, -1);
Al Viro294d71f2015-05-08 11:43:53 -0400609 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400610 }
Al Viro294d71f2015-05-08 11:43:53 -0400611 return -1;
612}
613
614/* call under rcu_read_lock */
615bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
616{
617 int res = __legitimize_mnt(bastard, seq);
618 if (likely(!res))
619 return true;
620 if (unlikely(res < 0)) {
621 rcu_read_unlock();
622 mntput(bastard);
623 rcu_read_lock();
624 }
Al Viro48a066e2013-09-29 22:06:07 -0400625 return false;
626}
627
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628/*
Al Viro474279d2013-10-01 16:11:26 -0400629 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400630 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 */
Al Viro474279d2013-10-01 16:11:26 -0400632struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633{
Al Viro38129a12014-03-20 21:10:51 -0400634 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400635 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
Al Viro38129a12014-03-20 21:10:51 -0400637 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400638 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
639 return p;
640 return NULL;
641}
642
643/*
644 * find the last mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400645 * mount_lock must be held.
Al Viro474279d2013-10-01 16:11:26 -0400646 */
647struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
648{
Eric W. Biederman411a9382014-12-22 19:12:07 -0600649 struct mount *p, *res = NULL;
650 p = __lookup_mnt(mnt, dentry);
Al Viro38129a12014-03-20 21:10:51 -0400651 if (!p)
652 goto out;
Eric W. Biederman411a9382014-12-22 19:12:07 -0600653 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
654 res = p;
Al Viro38129a12014-03-20 21:10:51 -0400655 hlist_for_each_entry_continue(p, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -0400656 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
657 break;
Eric W. Biederman411a9382014-12-22 19:12:07 -0600658 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
659 res = p;
Al Viro1d6a32a2014-03-20 20:34:43 -0400660 }
Al Viro38129a12014-03-20 21:10:51 -0400661out:
Al Viro1d6a32a2014-03-20 20:34:43 -0400662 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
Ram Paia05964f2005-11-07 17:20:17 -0500665/*
David Howellsf015f1262012-06-25 12:55:28 +0100666 * lookup_mnt - Return the first child mount mounted at path
667 *
668 * "First" means first mounted chronologically. If you create the
669 * following mounts:
670 *
671 * mount /dev/sda1 /mnt
672 * mount /dev/sda2 /mnt
673 * mount /dev/sda3 /mnt
674 *
675 * Then lookup_mnt() on the base /mnt dentry in the root mount will
676 * return successively the root dentry and vfsmount of /dev/sda1, then
677 * /dev/sda2, then /dev/sda3, then NULL.
678 *
679 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500680 */
Al Viro1c755af2009-04-18 14:06:57 -0400681struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500682{
Al Viroc7105362011-11-24 18:22:03 -0500683 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400684 struct vfsmount *m;
685 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000686
Al Viro48a066e2013-09-29 22:06:07 -0400687 rcu_read_lock();
688 do {
689 seq = read_seqbegin(&mount_lock);
690 child_mnt = __lookup_mnt(path->mnt, path->dentry);
691 m = child_mnt ? &child_mnt->mnt : NULL;
692 } while (!legitimize_mnt(m, seq));
693 rcu_read_unlock();
694 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500695}
696
Eric W. Biederman7af13642013-10-04 19:15:13 -0700697/*
698 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
699 * current mount namespace.
700 *
701 * The common case is dentries are not mountpoints at all and that
702 * test is handled inline. For the slow case when we are actually
703 * dealing with a mountpoint of some kind, walk through all of the
704 * mounts in the current mount namespace and test to see if the dentry
705 * is a mountpoint.
706 *
707 * The mount_hashtable is not usable in the context because we
708 * need to identify all mounts that may be in the current mount
709 * namespace not just a mount that happens to have some specified
710 * parent mount.
711 */
712bool __is_local_mountpoint(struct dentry *dentry)
713{
714 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
715 struct mount *mnt;
716 bool is_covered = false;
717
718 if (!d_mountpoint(dentry))
719 goto out;
720
721 down_read(&namespace_sem);
722 list_for_each_entry(mnt, &ns->list, mnt_list) {
723 is_covered = (mnt->mnt_mountpoint == dentry);
724 if (is_covered)
725 break;
726 }
727 up_read(&namespace_sem);
728out:
729 return is_covered;
730}
731
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800732static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
Al Viro84d17192013-03-15 10:53:28 -0400733{
Al Viro0818bf22014-02-28 13:46:44 -0500734 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d17192013-03-15 10:53:28 -0400735 struct mountpoint *mp;
736
Al Viro0818bf22014-02-28 13:46:44 -0500737 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d17192013-03-15 10:53:28 -0400738 if (mp->m_dentry == dentry) {
739 /* might be worth a WARN_ON() */
740 if (d_unlinked(dentry))
741 return ERR_PTR(-ENOENT);
742 mp->m_count++;
743 return mp;
744 }
745 }
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800746 return NULL;
747}
748
749static struct mountpoint *new_mountpoint(struct dentry *dentry)
750{
751 struct hlist_head *chain = mp_hash(dentry);
752 struct mountpoint *mp;
753 int ret;
Al Viro84d17192013-03-15 10:53:28 -0400754
755 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
756 if (!mp)
757 return ERR_PTR(-ENOMEM);
758
Miklos Szeredieed81002013-09-05 14:39:11 +0200759 ret = d_set_mounted(dentry);
760 if (ret) {
Al Viro84d17192013-03-15 10:53:28 -0400761 kfree(mp);
Miklos Szeredieed81002013-09-05 14:39:11 +0200762 return ERR_PTR(ret);
Al Viro84d17192013-03-15 10:53:28 -0400763 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200764
Al Viro84d17192013-03-15 10:53:28 -0400765 mp->m_dentry = dentry;
766 mp->m_count = 1;
Al Viro0818bf22014-02-28 13:46:44 -0500767 hlist_add_head(&mp->m_hash, chain);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700768 INIT_HLIST_HEAD(&mp->m_list);
Al Viro84d17192013-03-15 10:53:28 -0400769 return mp;
770}
771
772static void put_mountpoint(struct mountpoint *mp)
773{
774 if (!--mp->m_count) {
775 struct dentry *dentry = mp->m_dentry;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700776 BUG_ON(!hlist_empty(&mp->m_list));
Al Viro84d17192013-03-15 10:53:28 -0400777 spin_lock(&dentry->d_lock);
778 dentry->d_flags &= ~DCACHE_MOUNTED;
779 spin_unlock(&dentry->d_lock);
Al Viro0818bf22014-02-28 13:46:44 -0500780 hlist_del(&mp->m_hash);
Al Viro84d17192013-03-15 10:53:28 -0400781 kfree(mp);
782 }
783}
784
Al Viro143c8c92011-11-25 00:46:35 -0500785static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800787 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788}
789
Nick Piggin99b7db72010-08-18 04:37:39 +1000790/*
791 * vfsmount lock must be held for write
792 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800793static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500794{
795 if (ns) {
796 ns->event = ++event;
797 wake_up_interruptible(&ns->poll);
798 }
799}
800
Nick Piggin99b7db72010-08-18 04:37:39 +1000801/*
802 * vfsmount lock must be held for write
803 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800804static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500805{
806 if (ns && ns->event != event) {
807 ns->event = event;
808 wake_up_interruptible(&ns->poll);
809 }
810}
811
Nick Piggin99b7db72010-08-18 04:37:39 +1000812/*
813 * vfsmount lock must be held for write
814 */
Eric W. Biederman7bdb11d2014-12-29 13:03:41 -0600815static void unhash_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816{
Al Viro0714a532011-11-24 22:19:58 -0500817 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500818 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500819 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400820 hlist_del_init_rcu(&mnt->mnt_hash);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700821 hlist_del_init(&mnt->mnt_mp_list);
Al Viro84d17192013-03-15 10:53:28 -0400822 put_mountpoint(mnt->mnt_mp);
823 mnt->mnt_mp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824}
825
Nick Piggin99b7db72010-08-18 04:37:39 +1000826/*
827 * vfsmount lock must be held for write
828 */
Eric W. Biederman7bdb11d2014-12-29 13:03:41 -0600829static void detach_mnt(struct mount *mnt, struct path *old_path)
830{
831 old_path->dentry = mnt->mnt_mountpoint;
832 old_path->mnt = &mnt->mnt_parent->mnt;
833 unhash_mnt(mnt);
834}
835
836/*
837 * vfsmount lock must be held for write
838 */
Eric W. Biederman6a46c572015-01-15 22:58:33 -0600839static void umount_mnt(struct mount *mnt)
840{
841 /* old mountpoint will be dropped when we can do that */
842 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
843 unhash_mnt(mnt);
844}
845
846/*
847 * vfsmount lock must be held for write
848 */
Al Viro84d17192013-03-15 10:53:28 -0400849void mnt_set_mountpoint(struct mount *mnt,
850 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500851 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500852{
Al Viro84d17192013-03-15 10:53:28 -0400853 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500854 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro84d17192013-03-15 10:53:28 -0400855 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500856 child_mnt->mnt_parent = mnt;
Al Viro84d17192013-03-15 10:53:28 -0400857 child_mnt->mnt_mp = mp;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700858 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500859}
860
Nick Piggin99b7db72010-08-18 04:37:39 +1000861/*
862 * vfsmount lock must be held for write
863 */
Al Viro84d17192013-03-15 10:53:28 -0400864static void attach_mnt(struct mount *mnt,
865 struct mount *parent,
866 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867{
Al Viro84d17192013-03-15 10:53:28 -0400868 mnt_set_mountpoint(parent, mp, mnt);
Al Viro38129a12014-03-20 21:10:51 -0400869 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
Al Viro84d17192013-03-15 10:53:28 -0400870 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500871}
872
Al Viro12a5b522014-08-10 03:44:55 -0400873static void attach_shadowed(struct mount *mnt,
874 struct mount *parent,
875 struct mount *shadows)
876{
877 if (shadows) {
Linus Torvaldsf6f99332014-08-11 11:44:11 -0700878 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
Al Viro12a5b522014-08-10 03:44:55 -0400879 list_add(&mnt->mnt_child, &shadows->mnt_child);
880 } else {
881 hlist_add_head_rcu(&mnt->mnt_hash,
882 m_hash(&parent->mnt, mnt->mnt_mountpoint));
883 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
884 }
885}
886
Ram Paib90fa9a2005-11-07 17:19:50 -0500887/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000888 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500889 */
Al Viro1d6a32a2014-03-20 20:34:43 -0400890static void commit_tree(struct mount *mnt, struct mount *shadows)
Ram Paib90fa9a2005-11-07 17:19:50 -0500891{
Al Viro0714a532011-11-24 22:19:58 -0500892 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500893 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500894 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500895 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500896
Al Viro0714a532011-11-24 22:19:58 -0500897 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500898
Al Viro1a4eeaf2011-11-25 02:19:55 -0500899 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400900 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500901 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500902
Ram Paib90fa9a2005-11-07 17:19:50 -0500903 list_splice(&head, n->list.prev);
904
Eric W. Biedermand2921682016-09-28 00:27:17 -0500905 n->mounts += n->pending_mounts;
906 n->pending_mounts = 0;
907
Al Viro12a5b522014-08-10 03:44:55 -0400908 attach_shadowed(mnt, parent, shadows);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800909 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910}
911
Al Viro909b0a82011-11-25 03:06:56 -0500912static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
Al Viro6b41d532011-11-24 23:24:33 -0500914 struct list_head *next = p->mnt_mounts.next;
915 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500917 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500919 next = p->mnt_child.next;
920 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 break;
Al Viro0714a532011-11-24 22:19:58 -0500922 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 }
924 }
Al Viro6b41d532011-11-24 23:24:33 -0500925 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926}
927
Al Viro315fc832011-11-24 18:57:30 -0500928static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500929{
Al Viro6b41d532011-11-24 23:24:33 -0500930 struct list_head *prev = p->mnt_mounts.prev;
931 while (prev != &p->mnt_mounts) {
932 p = list_entry(prev, struct mount, mnt_child);
933 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500934 }
935 return p;
936}
937
Al Viro9d412a42011-03-17 22:08:28 -0400938struct vfsmount *
939vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
940{
Al Virob105e272011-11-24 20:38:33 -0500941 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400942 struct dentry *root;
943
944 if (!type)
945 return ERR_PTR(-ENODEV);
946
947 mnt = alloc_vfsmnt(name);
948 if (!mnt)
949 return ERR_PTR(-ENOMEM);
950
951 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500952 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400953
954 root = mount_fs(type, flags, name, data);
955 if (IS_ERR(root)) {
David Howells8ffcb322014-01-24 12:17:54 +0000956 mnt_free_id(mnt);
Al Viro9d412a42011-03-17 22:08:28 -0400957 free_vfsmnt(mnt);
958 return ERR_CAST(root);
959 }
960
Al Virob105e272011-11-24 20:38:33 -0500961 mnt->mnt.mnt_root = root;
962 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500963 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500964 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400965 lock_mount_hash();
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100966 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400967 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -0500968 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400969}
970EXPORT_SYMBOL_GPL(vfs_kern_mount);
971
Al Viro87129cc2011-11-24 21:24:27 -0500972static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500973 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974{
Al Viro87129cc2011-11-24 21:24:27 -0500975 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100976 struct mount *mnt;
977 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
David Howellsbe34d1a2012-06-25 12:55:18 +0100979 mnt = alloc_vfsmnt(old->mnt_devname);
980 if (!mnt)
981 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100982
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700983 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +0100984 mnt->mnt_group_id = 0; /* not a peer of original */
985 else
986 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100987
David Howellsbe34d1a2012-06-25 12:55:18 +0100988 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
989 err = mnt_alloc_group_id(mnt);
990 if (err)
991 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100993
Al Virof2ebb3a2014-02-27 09:35:45 -0500994 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700995 /* Don't allow unprivileged users to change mount flags */
Eric W. Biederman9566d672014-07-28 17:26:07 -0700996 if (flag & CL_UNPRIVILEGED) {
997 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
998
999 if (mnt->mnt.mnt_flags & MNT_READONLY)
1000 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1001
1002 if (mnt->mnt.mnt_flags & MNT_NODEV)
1003 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1004
1005 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1006 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1007
1008 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1009 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1010 }
Eric W. Biederman132c94e2013-03-22 04:08:05 -07001011
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001012 /* Don't allow unprivileged users to reveal what is under a mount */
Eric W. Biederman381cacb2014-10-07 17:11:46 -07001013 if ((flag & CL_UNPRIVILEGED) &&
1014 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001015 mnt->mnt.mnt_flags |= MNT_LOCKED;
1016
David Howellsbe34d1a2012-06-25 12:55:18 +01001017 atomic_inc(&sb->s_active);
1018 mnt->mnt.mnt_sb = sb;
1019 mnt->mnt.mnt_root = dget(root);
1020 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1021 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -04001022 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001023 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001024 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001025
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001026 if ((flag & CL_SLAVE) ||
1027 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +01001028 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1029 mnt->mnt_master = old;
1030 CLEAR_MNT_SHARED(mnt);
1031 } else if (!(flag & CL_PRIVATE)) {
1032 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1033 list_add(&mnt->mnt_share, &old->mnt_share);
1034 if (IS_MNT_SLAVE(old))
1035 list_add(&mnt->mnt_slave, &old->mnt_slave);
1036 mnt->mnt_master = old->mnt_master;
1037 }
1038 if (flag & CL_MAKE_SHARED)
1039 set_mnt_shared(mnt);
1040
1041 /* stick the duplicate mount on the same expiry list
1042 * as the original if that was on one */
1043 if (flag & CL_EXPIRE) {
1044 if (!list_empty(&old->mnt_expire))
1045 list_add(&mnt->mnt_expire, &old->mnt_expire);
1046 }
1047
Al Virocb338d02011-11-24 20:55:08 -05001048 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001049
1050 out_free:
David Howells8ffcb322014-01-24 12:17:54 +00001051 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001052 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001053 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054}
1055
Al Viro9ea459e2014-08-08 13:08:20 -04001056static void cleanup_mnt(struct mount *mnt)
1057{
1058 /*
1059 * This probably indicates that somebody messed
1060 * up a mnt_want/drop_write() pair. If this
1061 * happens, the filesystem was probably unable
1062 * to make r/w->r/o transitions.
1063 */
1064 /*
1065 * The locking used to deal with mnt_count decrement provides barriers,
1066 * so mnt_get_writers() below is safe.
1067 */
1068 WARN_ON(mnt_get_writers(mnt));
1069 if (unlikely(mnt->mnt_pins.first))
1070 mnt_pin_kill(mnt);
1071 fsnotify_vfsmount_delete(&mnt->mnt);
1072 dput(mnt->mnt.mnt_root);
1073 deactivate_super(mnt->mnt.mnt_sb);
1074 mnt_free_id(mnt);
1075 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1076}
1077
1078static void __cleanup_mnt(struct rcu_head *head)
1079{
1080 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1081}
1082
1083static LLIST_HEAD(delayed_mntput_list);
1084static void delayed_mntput(struct work_struct *unused)
1085{
1086 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1087 struct llist_node *next;
1088
1089 for (; node; node = next) {
1090 next = llist_next(node);
1091 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1092 }
1093}
1094static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1095
Al Viro900148d2011-11-25 00:33:11 -05001096static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001097{
Al Viro48a066e2013-09-29 22:06:07 -04001098 rcu_read_lock();
1099 mnt_add_count(mnt, -1);
1100 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1101 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -05001102 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001103 }
Al Viro719ea2f2013-09-29 11:24:49 -04001104 lock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001105 if (mnt_get_count(mnt)) {
Al Viro48a066e2013-09-29 22:06:07 -04001106 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -04001107 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001108 return;
1109 }
Al Viro48a066e2013-09-29 22:06:07 -04001110 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1111 rcu_read_unlock();
1112 unlock_mount_hash();
1113 return;
1114 }
1115 mnt->mnt.mnt_flags |= MNT_DOOMED;
1116 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301117
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001118 list_del(&mnt->mnt_instance);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001119
1120 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1121 struct mount *p, *tmp;
1122 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1123 umount_mnt(p);
1124 }
1125 }
Al Viro719ea2f2013-09-29 11:24:49 -04001126 unlock_mount_hash();
Al Viro649a7952013-09-28 12:41:25 -04001127
Al Viro9ea459e2014-08-08 13:08:20 -04001128 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1129 struct task_struct *task = current;
1130 if (likely(!(task->flags & PF_KTHREAD))) {
1131 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1132 if (!task_work_add(task, &mnt->mnt_rcu, true))
1133 return;
1134 }
1135 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1136 schedule_delayed_work(&delayed_mntput_work, 1);
1137 return;
1138 }
1139 cleanup_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001140}
Nick Pigginb3e19d92011-01-07 17:50:11 +11001141
1142void mntput(struct vfsmount *mnt)
1143{
1144 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -05001145 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001146 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -05001147 if (unlikely(m->mnt_expiry_mark))
1148 m->mnt_expiry_mark = 0;
1149 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001150 }
1151}
1152EXPORT_SYMBOL(mntput);
1153
1154struct vfsmount *mntget(struct vfsmount *mnt)
1155{
1156 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001157 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001158 return mnt;
1159}
1160EXPORT_SYMBOL(mntget);
1161
Al Viro3064c352014-08-07 09:12:31 -04001162struct vfsmount *mnt_clone_internal(struct path *path)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001163{
Al Viro3064c352014-08-07 09:12:31 -04001164 struct mount *p;
1165 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1166 if (IS_ERR(p))
1167 return ERR_CAST(p);
1168 p->mnt.mnt_flags |= MNT_INTERNAL;
1169 return &p->mnt;
Al Viro7b7b1ac2005-11-07 17:13:39 -05001170}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001172static inline void mangle(struct seq_file *m, const char *s)
1173{
1174 seq_escape(m, s, " \t\n\\");
1175}
1176
1177/*
1178 * Simple .show_options callback for filesystems which don't want to
1179 * implement more complex mount option showing.
1180 *
1181 * See also save_mount_options().
1182 */
Al Viro34c80b12011-12-08 21:32:45 -05001183int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001184{
Al Viro2a32ceb2009-05-08 16:05:57 -04001185 const char *options;
1186
1187 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -05001188 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001189
1190 if (options != NULL && options[0]) {
1191 seq_putc(m, ',');
1192 mangle(m, options);
1193 }
Al Viro2a32ceb2009-05-08 16:05:57 -04001194 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001195
1196 return 0;
1197}
1198EXPORT_SYMBOL(generic_show_options);
1199
1200/*
1201 * If filesystem uses generic_show_options(), this function should be
1202 * called from the fill_super() callback.
1203 *
1204 * The .remount_fs callback usually needs to be handled in a special
1205 * way, to make sure, that previous options are not overwritten if the
1206 * remount fails.
1207 *
1208 * Also note, that if the filesystem's .remount_fs function doesn't
1209 * reset all options to their default value, but changes only newly
1210 * given options, then the displayed options will not reflect reality
1211 * any more.
1212 */
1213void save_mount_options(struct super_block *sb, char *options)
1214{
Al Viro2a32ceb2009-05-08 16:05:57 -04001215 BUG_ON(sb->s_options);
1216 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001217}
1218EXPORT_SYMBOL(save_mount_options);
1219
Al Viro2a32ceb2009-05-08 16:05:57 -04001220void replace_mount_options(struct super_block *sb, char *options)
1221{
1222 char *old = sb->s_options;
1223 rcu_assign_pointer(sb->s_options, options);
1224 if (old) {
1225 synchronize_rcu();
1226 kfree(old);
1227 }
1228}
1229EXPORT_SYMBOL(replace_mount_options);
1230
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001231#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001232/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233static void *m_start(struct seq_file *m, loff_t *pos)
1234{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001235 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
Ram Pai390c6842005-11-07 17:17:51 -05001237 down_read(&namespace_sem);
Al Viroc7999c32014-02-27 14:40:10 -05001238 if (p->cached_event == p->ns->event) {
1239 void *v = p->cached_mount;
1240 if (*pos == p->cached_index)
1241 return v;
1242 if (*pos == p->cached_index + 1) {
1243 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1244 return p->cached_mount = v;
1245 }
1246 }
1247
1248 p->cached_event = p->ns->event;
1249 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1250 p->cached_index = *pos;
1251 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252}
1253
1254static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1255{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001256 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001257
Al Viroc7999c32014-02-27 14:40:10 -05001258 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1259 p->cached_index = *pos;
1260 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261}
1262
1263static void m_stop(struct seq_file *m, void *v)
1264{
Ram Pai390c6842005-11-07 17:17:51 -05001265 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266}
1267
Al Viro0226f492011-12-06 12:21:54 -05001268static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001269{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001270 struct proc_mounts *p = m->private;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001271 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001272 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273}
1274
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001275const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 .start = m_start,
1277 .next = m_next,
1278 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001279 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001280};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001281#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283/**
1284 * may_umount_tree - check if a mount tree is busy
1285 * @mnt: root of mount tree
1286 *
1287 * This is called to check if a tree of mounts has any
1288 * open files, pwds, chroots or sub mounts that are
1289 * busy.
1290 */
Al Viro909b0a82011-11-25 03:06:56 -05001291int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292{
Al Viro909b0a82011-11-25 03:06:56 -05001293 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001294 int actual_refs = 0;
1295 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001296 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001297 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Nick Pigginb3e19d92011-01-07 17:50:11 +11001299 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001300 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001301 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001302 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 }
Al Viro719ea2f2013-09-29 11:24:49 -04001305 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
1307 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001308 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
Ian Kente3474a82006-03-27 01:14:51 -08001310 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311}
1312
1313EXPORT_SYMBOL(may_umount_tree);
1314
1315/**
1316 * may_umount - check if a mount point is busy
1317 * @mnt: root of mount
1318 *
1319 * This is called to check if a mount point has any
1320 * open files, pwds, chroots or sub mounts. If the
1321 * mount has sub mounts this will return busy
1322 * regardless of whether the sub mounts are busy.
1323 *
1324 * Doesn't take quota and stuff into account. IOW, in some cases it will
1325 * give false negatives. The main reason why it's here is that we need
1326 * a non-destructive way to look for easily umountable filesystems.
1327 */
1328int may_umount(struct vfsmount *mnt)
1329{
Ian Kente3474a82006-03-27 01:14:51 -08001330 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001331 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001332 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001333 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001334 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001335 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001336 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001337 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338}
1339
1340EXPORT_SYMBOL(may_umount);
1341
Al Viro38129a12014-03-20 21:10:51 -04001342static HLIST_HEAD(unmounted); /* protected by namespace_sem */
Al Viroe3197d82013-03-16 14:35:16 -04001343
Al Viro97216be2013-03-16 15:12:40 -04001344static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001346 struct hlist_head head;
Al Viro97216be2013-03-16 15:12:40 -04001347
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001348 hlist_move_list(&unmounted, &head);
Al Viro97216be2013-03-16 15:12:40 -04001349
Al Viro97216be2013-03-16 15:12:40 -04001350 up_write(&namespace_sem);
1351
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001352 if (likely(hlist_empty(&head)))
1353 return;
1354
Al Viro48a066e2013-09-29 22:06:07 -04001355 synchronize_rcu();
1356
Al Viro87b95ce2015-01-10 19:01:08 -05001357 group_pin_kill(&head);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001358}
1359
Al Viro97216be2013-03-16 15:12:40 -04001360static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001361{
Al Viro97216be2013-03-16 15:12:40 -04001362 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001363}
1364
Eric W. Biedermane819f152014-12-24 07:20:01 -06001365enum umount_tree_flags {
1366 UMOUNT_SYNC = 1,
1367 UMOUNT_PROPAGATE = 2,
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001368 UMOUNT_CONNECTED = 4,
Eric W. Biedermane819f152014-12-24 07:20:01 -06001369};
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001370
1371static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1372{
1373 /* Leaving mounts connected is only valid for lazy umounts */
1374 if (how & UMOUNT_SYNC)
1375 return true;
1376
1377 /* A mount without a parent has nothing to be connected to */
1378 if (!mnt_has_parent(mnt))
1379 return true;
1380
1381 /* Because the reference counting rules change when mounts are
1382 * unmounted and connected, umounted mounts may not be
1383 * connected to mounted mounts.
1384 */
1385 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1386 return true;
1387
1388 /* Has it been requested that the mount remain connected? */
1389 if (how & UMOUNT_CONNECTED)
1390 return false;
1391
1392 /* Is the mount locked such that it needs to remain connected? */
1393 if (IS_MNT_LOCKED(mnt))
1394 return false;
1395
1396 /* By default disconnect the mount */
1397 return true;
1398}
1399
Nick Piggin99b7db72010-08-18 04:37:39 +10001400/*
Al Viro48a066e2013-09-29 22:06:07 -04001401 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001402 * namespace_sem must be held for write
1403 */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001404static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001405{
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001406 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001407 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001408
Eric W. Biederman5d884572015-01-03 05:39:35 -06001409 if (how & UMOUNT_PROPAGATE)
1410 propagate_mount_unlock(mnt);
1411
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001412 /* Gather the mounts to umount */
Eric W. Biederman590ce4b2014-12-22 18:30:08 -06001413 for (p = mnt; p; p = next_mnt(p, mnt)) {
1414 p->mnt.mnt_flags |= MNT_UMOUNT;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001415 list_move(&p->mnt_list, &tmp_list);
Eric W. Biederman590ce4b2014-12-22 18:30:08 -06001416 }
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001417
Eric W. Biederman411a9382014-12-22 19:12:07 -06001418 /* Hide the mounts from mnt_mounts */
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001419 list_for_each_entry(p, &tmp_list, mnt_list) {
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001420 list_del_init(&p->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -04001421 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001422
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001423 /* Add propogated mounts to the tmp_list */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001424 if (how & UMOUNT_PROPAGATE)
Al Viro7b8a53f2011-01-15 20:08:44 -05001425 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001426
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001427 while (!list_empty(&tmp_list)) {
Eric W. Biedermand2921682016-09-28 00:27:17 -05001428 struct mnt_namespace *ns;
Eric W. Biedermance07d892014-12-23 21:37:03 -06001429 bool disconnect;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001430 p = list_first_entry(&tmp_list, struct mount, mnt_list);
Al Viro6776db3d2011-11-25 00:22:05 -05001431 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001432 list_del_init(&p->mnt_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05001433 ns = p->mnt_ns;
1434 if (ns) {
1435 ns->mounts--;
1436 __touch_mnt_namespace(ns);
1437 }
Al Viro143c8c92011-11-25 00:46:35 -05001438 p->mnt_ns = NULL;
Eric W. Biedermane819f152014-12-24 07:20:01 -06001439 if (how & UMOUNT_SYNC)
Al Viro48a066e2013-09-29 22:06:07 -04001440 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro87b95ce2015-01-10 19:01:08 -05001441
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001442 disconnect = disconnect_mount(p, how);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001443
1444 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1445 disconnect ? &unmounted : NULL);
Al Viro676da582011-11-24 21:47:05 -05001446 if (mnt_has_parent(p)) {
Al Viro81b6b062014-08-30 18:32:05 -04001447 mnt_add_count(p->mnt_parent, -1);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001448 if (!disconnect) {
1449 /* Don't forget about p */
1450 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1451 } else {
1452 umount_mnt(p);
1453 }
Al Viro7c4b93d2008-03-21 23:59:49 -04001454 }
Al Viro0f0afb12011-11-24 20:43:10 -05001455 change_mnt_propagation(p, MS_PRIVATE);
Al Viro38129a12014-03-20 21:10:51 -04001456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457}
1458
Al Virob54b9be2013-03-16 14:39:34 -04001459static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001460
Al Viro1ab59732011-11-24 21:35:16 -05001461static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462{
Al Viro1ab59732011-11-24 21:35:16 -05001463 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 int retval;
1465
Al Viro1ab59732011-11-24 21:35:16 -05001466 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 if (retval)
1468 return retval;
1469
1470 /*
1471 * Allow userspace to request a mountpoint be expired rather than
1472 * unmounting unconditionally. Unmount only happens if:
1473 * (1) the mark is already set (the mark is cleared by mntput())
1474 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1475 */
1476 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001477 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 flags & (MNT_FORCE | MNT_DETACH))
1479 return -EINVAL;
1480
Nick Pigginb3e19d92011-01-07 17:50:11 +11001481 /*
1482 * probably don't strictly need the lock here if we examined
1483 * all race cases, but it's a slowpath.
1484 */
Al Viro719ea2f2013-09-29 11:24:49 -04001485 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001486 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001487 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001489 }
Al Viro719ea2f2013-09-29 11:24:49 -04001490 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491
Al Viro863d6842011-11-25 00:57:42 -05001492 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return -EAGAIN;
1494 }
1495
1496 /*
1497 * If we may have to abort operations to get out of this
1498 * mount, and they will themselves hold resources we must
1499 * allow the fs to do things. In the Unix tradition of
1500 * 'Gee thats tricky lets do it in userspace' the umount_begin
1501 * might fail to complete on the first run through as other tasks
1502 * must return, and the like. Thats for the mount program to worry
1503 * about for the moment.
1504 */
1505
Al Viro42faad92008-04-24 07:21:56 -04001506 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001507 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
1510 /*
1511 * No sense to grab the lock for this test, but test itself looks
1512 * somewhat bogus. Suggestions for better replacement?
1513 * Ho-hum... In principle, we might treat that as umount + switch
1514 * to rootfs. GC would eventually take care of the old vfsmount.
1515 * Actually it makes sense, especially if rootfs would contain a
1516 * /reboot - static binary that would close all descriptors and
1517 * call reboot(9). Then init(8) could umount root and exec /reboot.
1518 */
Al Viro1ab59732011-11-24 21:35:16 -05001519 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 /*
1521 * Special case for "unmounting" root ...
1522 * we just try to remount it readonly.
1523 */
Andy Lutomirskia1480dc2014-10-08 12:32:47 -07001524 if (!capable(CAP_SYS_ADMIN))
1525 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001527 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 up_write(&sb->s_umount);
1530 return retval;
1531 }
1532
Al Viro97216be2013-03-16 15:12:40 -04001533 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001534 lock_mount_hash();
Al Viro5addc5d2005-11-07 17:15:49 -05001535 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536
Al Viro48a066e2013-09-29 22:06:07 -04001537 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001538 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001539 umount_tree(mnt, UMOUNT_PROPAGATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001541 } else {
1542 shrink_submounts(mnt);
1543 retval = -EBUSY;
1544 if (!propagate_mount_busy(mnt, 2)) {
1545 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001546 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Viro48a066e2013-09-29 22:06:07 -04001547 retval = 0;
1548 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 }
Al Viro719ea2f2013-09-29 11:24:49 -04001550 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001551 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 return retval;
1553}
1554
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001555/*
1556 * __detach_mounts - lazily unmount all mounts on the specified dentry
1557 *
1558 * During unlink, rmdir, and d_drop it is possible to loose the path
1559 * to an existing mountpoint, and wind up leaking the mount.
1560 * detach_mounts allows lazily unmounting those mounts instead of
1561 * leaking them.
1562 *
1563 * The caller may hold dentry->d_inode->i_mutex.
1564 */
1565void __detach_mounts(struct dentry *dentry)
1566{
1567 struct mountpoint *mp;
1568 struct mount *mnt;
1569
1570 namespace_lock();
1571 mp = lookup_mountpoint(dentry);
Eric W. Biedermanf53e5792015-01-19 11:48:45 -06001572 if (IS_ERR_OR_NULL(mp))
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001573 goto out_unlock;
1574
1575 lock_mount_hash();
Andrey Ulanove06b9332016-04-15 14:24:41 -07001576 event++;
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001577 while (!hlist_empty(&mp->m_list)) {
1578 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001579 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
Eric W. Biedermanfe78fcc2015-07-17 14:54:27 -05001580 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1581 umount_mnt(mnt);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001582 }
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001583 else umount_tree(mnt, UMOUNT_CONNECTED);
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001584 }
1585 unlock_mount_hash();
1586 put_mountpoint(mp);
1587out_unlock:
1588 namespace_unlock();
1589}
1590
Al Viro9b40bc92013-02-22 22:45:42 -05001591/*
1592 * Is the caller allowed to modify his namespace?
1593 */
1594static inline bool may_mount(void)
1595{
1596 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1597}
1598
Jeff Layton9e8925b2015-11-16 09:49:34 -05001599static inline bool may_mandlock(void)
1600{
1601#ifndef CONFIG_MANDATORY_FILE_LOCKING
1602 return false;
1603#endif
Eric W. Biederman95ace752015-11-11 17:22:33 -06001604 return capable(CAP_SYS_ADMIN);
Jeff Layton9e8925b2015-11-16 09:49:34 -05001605}
1606
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607/*
1608 * Now umount can handle mount points as well as block devices.
1609 * This is important for filesystems which use unnamed block devices.
1610 *
1611 * We now support a flag for forced unmount like the other 'big iron'
1612 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1613 */
1614
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001615SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616{
Al Viro2d8f3032008-07-22 09:59:21 -04001617 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001618 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001620 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001622 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1623 return -EINVAL;
1624
Al Viro9b40bc92013-02-22 22:45:42 -05001625 if (!may_mount())
1626 return -EPERM;
1627
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001628 if (!(flags & UMOUNT_NOFOLLOW))
1629 lookup_flags |= LOOKUP_FOLLOW;
1630
Al Viro197df042013-09-08 14:03:27 -04001631 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 if (retval)
1633 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001634 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001636 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001638 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 goto dput_and_out;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001640 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1641 goto dput_and_out;
Eric W. Biedermanb2f5d4d2014-10-04 14:44:03 -07001642 retval = -EPERM;
1643 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1644 goto dput_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
Al Viro900148d2011-11-25 00:33:11 -05001646 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001648 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001649 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001650 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651out:
1652 return retval;
1653}
1654
1655#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1656
1657/*
Ram Paib58fed82005-11-07 17:16:09 -05001658 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001660SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661{
Ram Paib58fed82005-11-07 17:16:09 -05001662 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663}
1664
1665#endif
1666
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001667static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001668{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001669 /* Is this a proxy for a mount namespace? */
Al Viroe149ed22014-11-01 10:57:28 -04001670 return dentry->d_op == &ns_dentry_operations &&
1671 dentry->d_fsdata == &mntns_operations;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001672}
1673
Al Viro58be28252014-11-01 00:00:23 -04001674struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1675{
1676 return container_of(ns, struct mnt_namespace, ns);
1677}
1678
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001679static bool mnt_ns_loop(struct dentry *dentry)
1680{
1681 /* Could bind mounting the mount namespace inode cause a
1682 * mount namespace loop?
1683 */
1684 struct mnt_namespace *mnt_ns;
1685 if (!is_mnt_ns_file(dentry))
1686 return false;
1687
Al Virof77c8012014-11-01 03:13:17 -04001688 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
Eric W. Biederman8823c072010-03-07 18:49:36 -08001689 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1690}
1691
Al Viro87129cc2011-11-24 21:24:27 -05001692struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001693 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694{
Al Viro84d17192013-03-15 10:53:28 -04001695 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001697 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1698 return ERR_PTR(-EINVAL);
1699
1700 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001701 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001702
Ram Pai36341f62005-11-07 17:17:22 -05001703 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001704 if (IS_ERR(q))
1705 return q;
1706
Al Viroa73324d2011-11-24 22:25:07 -05001707 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
1709 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001710 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001711 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001712 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 continue;
1714
Al Viro909b0a82011-11-25 03:06:56 -05001715 for (s = r; s; s = next_mnt(s, r)) {
Al Viro12a5b522014-08-10 03:44:55 -04001716 struct mount *t = NULL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001717 if (!(flag & CL_COPY_UNBINDABLE) &&
1718 IS_MNT_UNBINDABLE(s)) {
1719 s = skip_mnt_tree(s);
1720 continue;
1721 }
1722 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1723 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001724 s = skip_mnt_tree(s);
1725 continue;
1726 }
Al Viro0714a532011-11-24 22:19:58 -05001727 while (p != s->mnt_parent) {
1728 p = p->mnt_parent;
1729 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 }
Al Viro87129cc2011-11-24 21:24:27 -05001731 p = s;
Al Viro84d17192013-03-15 10:53:28 -04001732 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001733 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001734 if (IS_ERR(q))
1735 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04001736 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05001737 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro12a5b522014-08-10 03:44:55 -04001738 mnt_set_mountpoint(parent, p->mnt_mp, q);
1739 if (!list_empty(&parent->mnt_mounts)) {
1740 t = list_last_entry(&parent->mnt_mounts,
1741 struct mount, mnt_child);
1742 if (t->mnt_mp != p->mnt_mp)
1743 t = NULL;
1744 }
1745 attach_shadowed(q, parent, t);
Al Viro719ea2f2013-09-29 11:24:49 -04001746 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 }
1748 }
1749 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001750out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04001752 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001753 umount_tree(res, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001754 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001756 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757}
1758
David Howellsbe34d1a2012-06-25 12:55:18 +01001759/* Caller should check returned pointer for errors */
1760
Al Viro589ff872009-04-18 03:28:19 -04001761struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001762{
Al Virocb338d02011-11-24 20:55:08 -05001763 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001764 namespace_lock();
Eric W. Biedermancd4a4012015-01-07 14:28:26 -06001765 if (!check_mnt(real_mount(path->mnt)))
1766 tree = ERR_PTR(-EINVAL);
1767 else
1768 tree = copy_tree(real_mount(path->mnt), path->dentry,
1769 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001770 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001771 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001772 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001773 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001774}
1775
1776void drop_collected_mounts(struct vfsmount *mnt)
1777{
Al Viro97216be2013-03-16 15:12:40 -04001778 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001779 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001780 umount_tree(real_mount(mnt), UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001781 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04001782 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001783}
1784
Miklos Szeredic771d682014-10-24 00:14:36 +02001785/**
1786 * clone_private_mount - create a private clone of a path
1787 *
1788 * This creates a new vfsmount, which will be the clone of @path. The new will
1789 * not be attached anywhere in the namespace and will be private (i.e. changes
1790 * to the originating mount won't be propagated into this).
1791 *
1792 * Release with mntput().
1793 */
1794struct vfsmount *clone_private_mount(struct path *path)
1795{
1796 struct mount *old_mnt = real_mount(path->mnt);
1797 struct mount *new_mnt;
1798
1799 if (IS_MNT_UNBINDABLE(old_mnt))
1800 return ERR_PTR(-EINVAL);
1801
1802 down_read(&namespace_sem);
1803 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1804 up_read(&namespace_sem);
1805 if (IS_ERR(new_mnt))
1806 return ERR_CAST(new_mnt);
1807
1808 return &new_mnt->mnt;
1809}
1810EXPORT_SYMBOL_GPL(clone_private_mount);
1811
Al Viro1f707132010-01-30 22:51:25 -05001812int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1813 struct vfsmount *root)
1814{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001815 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001816 int res = f(root, arg);
1817 if (res)
1818 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001819 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1820 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001821 if (res)
1822 return res;
1823 }
1824 return 0;
1825}
1826
Al Viro4b8b21f2011-11-24 19:54:23 -05001827static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001828{
Al Viro315fc832011-11-24 18:57:30 -05001829 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001830
Al Viro909b0a82011-11-25 03:06:56 -05001831 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001832 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001833 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001834 }
1835}
1836
Al Viro4b8b21f2011-11-24 19:54:23 -05001837static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001838{
Al Viro315fc832011-11-24 18:57:30 -05001839 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001840
Al Viro909b0a82011-11-25 03:06:56 -05001841 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001842 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001843 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001844 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001845 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001846 return err;
1847 }
1848 }
1849 }
1850
1851 return 0;
1852}
1853
Eric W. Biedermand2921682016-09-28 00:27:17 -05001854int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1855{
1856 unsigned int max = READ_ONCE(sysctl_mount_max);
1857 unsigned int mounts = 0, old, pending, sum;
1858 struct mount *p;
1859
1860 for (p = mnt; p; p = next_mnt(p, mnt))
1861 mounts++;
1862
1863 old = ns->mounts;
1864 pending = ns->pending_mounts;
1865 sum = old + pending;
1866 if ((old > sum) ||
1867 (pending > sum) ||
1868 (max < sum) ||
1869 (mounts > (max - sum)))
1870 return -ENOSPC;
1871
1872 ns->pending_mounts = pending + mounts;
1873 return 0;
1874}
1875
Ram Paib90fa9a2005-11-07 17:19:50 -05001876/*
1877 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001878 * @nd : place the mount tree @source_mnt is attached
1879 * @parent_nd : if non-null, detach the source_mnt from its parent and
1880 * store the parent mount and mountpoint dentry.
1881 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001882 *
1883 * NOTE: in the table below explains the semantics when a source mount
1884 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001885 * ---------------------------------------------------------------------------
1886 * | BIND MOUNT OPERATION |
1887 * |**************************************************************************
1888 * | source-->| shared | private | slave | unbindable |
1889 * | dest | | | | |
1890 * | | | | | | |
1891 * | v | | | | |
1892 * |**************************************************************************
1893 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1894 * | | | | | |
1895 * |non-shared| shared (+) | private | slave (*) | invalid |
1896 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001897 * A bind operation clones the source mount and mounts the clone on the
1898 * destination mount.
1899 *
1900 * (++) the cloned mount is propagated to all the mounts in the propagation
1901 * tree of the destination mount and the cloned mount is added to
1902 * the peer group of the source mount.
1903 * (+) the cloned mount is created under the destination mount and is marked
1904 * as shared. The cloned mount is added to the peer group of the source
1905 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001906 * (+++) the mount is propagated to all the mounts in the propagation tree
1907 * of the destination mount and the cloned mount is made slave
1908 * of the same master as that of the source mount. The cloned mount
1909 * is marked as 'shared and slave'.
1910 * (*) the cloned mount is made a slave of the same master as that of the
1911 * source mount.
1912 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001913 * ---------------------------------------------------------------------------
1914 * | MOVE MOUNT OPERATION |
1915 * |**************************************************************************
1916 * | source-->| shared | private | slave | unbindable |
1917 * | dest | | | | |
1918 * | | | | | | |
1919 * | v | | | | |
1920 * |**************************************************************************
1921 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1922 * | | | | | |
1923 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1924 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001925 *
1926 * (+) the mount is moved to the destination. And is then propagated to
1927 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001928 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001929 * (+++) the mount is moved to the destination and is then propagated to
1930 * all the mounts belonging to the destination mount's propagation tree.
1931 * the mount is marked as 'shared and slave'.
1932 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001933 *
1934 * if the source mount is a tree, the operations explained above is
1935 * applied to each mount in the tree.
1936 * Must be called without spinlocks held, since this function can sleep
1937 * in allocations.
1938 */
Al Viro0fb54e52011-11-24 19:59:16 -05001939static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d17192013-03-15 10:53:28 -04001940 struct mount *dest_mnt,
1941 struct mountpoint *dest_mp,
1942 struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001943{
Al Viro38129a12014-03-20 21:10:51 -04001944 HLIST_HEAD(tree_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05001945 struct mnt_namespace *ns = dest_mnt->mnt_ns;
Al Viro315fc832011-11-24 18:57:30 -05001946 struct mount *child, *p;
Al Viro38129a12014-03-20 21:10:51 -04001947 struct hlist_node *n;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001948 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001949
Eric W. Biedermand2921682016-09-28 00:27:17 -05001950 /* Is there space to add these mounts to the mount namespace? */
1951 if (!parent_path) {
1952 err = count_mounts(ns, source_mnt);
1953 if (err)
1954 goto out;
1955 }
1956
Al Virofc7be132011-11-25 01:05:37 -05001957 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001958 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001959 if (err)
1960 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04001961 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Al Virof2ebb3a2014-02-27 09:35:45 -05001962 lock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001963 if (err)
1964 goto out_cleanup_ids;
Al Viro909b0a82011-11-25 03:06:56 -05001965 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001966 set_mnt_shared(p);
Al Viro0b1b9012014-03-21 10:14:08 -04001967 } else {
1968 lock_mount_hash();
Ram Paib90fa9a2005-11-07 17:19:50 -05001969 }
Al Viro1a390682008-03-21 20:48:19 -04001970 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001971 detach_mnt(source_mnt, parent_path);
Al Viro84d17192013-03-15 10:53:28 -04001972 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05001973 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001974 } else {
Al Viro84d17192013-03-15 10:53:28 -04001975 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Al Viro1d6a32a2014-03-20 20:34:43 -04001976 commit_tree(source_mnt, NULL);
Ram Pai21444402005-11-07 17:20:03 -05001977 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001978
Al Viro38129a12014-03-20 21:10:51 -04001979 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04001980 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04001981 hlist_del_init(&child->mnt_hash);
Al Viro1d6a32a2014-03-20 20:34:43 -04001982 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1983 child->mnt_mountpoint);
1984 commit_tree(child, q);
Ram Paib90fa9a2005-11-07 17:19:50 -05001985 }
Al Viro719ea2f2013-09-29 11:24:49 -04001986 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10001987
Ram Paib90fa9a2005-11-07 17:19:50 -05001988 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001989
1990 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05001991 while (!hlist_empty(&tree_list)) {
1992 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
Eric W. Biedermand2921682016-09-28 00:27:17 -05001993 child->mnt_parent->mnt_ns->pending_mounts = 0;
Eric W. Biedermane819f152014-12-24 07:20:01 -06001994 umount_tree(child, UMOUNT_SYNC);
Al Virof2ebb3a2014-02-27 09:35:45 -05001995 }
1996 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001997 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001998 out:
Eric W. Biedermand2921682016-09-28 00:27:17 -05001999 ns->pending_mounts = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002000 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05002001}
2002
Al Viro84d17192013-03-15 10:53:28 -04002003static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04002004{
2005 struct vfsmount *mnt;
Al Viro84d17192013-03-15 10:53:28 -04002006 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04002007retry:
Al Viro59551022016-01-22 15:40:57 -05002008 inode_lock(dentry->d_inode);
Al Viro84d17192013-03-15 10:53:28 -04002009 if (unlikely(cant_mount(dentry))) {
Al Viro59551022016-01-22 15:40:57 -05002010 inode_unlock(dentry->d_inode);
Al Viro84d17192013-03-15 10:53:28 -04002011 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04002012 }
Al Viro97216be2013-03-16 15:12:40 -04002013 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04002014 mnt = lookup_mnt(path);
Al Viro84d17192013-03-15 10:53:28 -04002015 if (likely(!mnt)) {
Eric W. Biedermane2dfa932014-02-24 17:32:34 -08002016 struct mountpoint *mp = lookup_mountpoint(dentry);
2017 if (!mp)
2018 mp = new_mountpoint(dentry);
Al Viro84d17192013-03-15 10:53:28 -04002019 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04002020 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002021 inode_unlock(dentry->d_inode);
Al Viro84d17192013-03-15 10:53:28 -04002022 return mp;
2023 }
2024 return mp;
2025 }
Al Viro97216be2013-03-16 15:12:40 -04002026 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002027 inode_unlock(path->dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002028 path_put(path);
2029 path->mnt = mnt;
Al Viro84d17192013-03-15 10:53:28 -04002030 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04002031 goto retry;
2032}
2033
Al Viro84d17192013-03-15 10:53:28 -04002034static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04002035{
Al Viro84d17192013-03-15 10:53:28 -04002036 struct dentry *dentry = where->m_dentry;
2037 put_mountpoint(where);
Al Viro328e6d92013-03-16 14:49:45 -04002038 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002039 inode_unlock(dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002040}
2041
Al Viro84d17192013-03-15 10:53:28 -04002042static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043{
Al Viro95bc5f22011-11-25 00:30:56 -05002044 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 return -EINVAL;
2046
David Howellse36cb0b2015-01-29 12:02:35 +00002047 if (d_is_dir(mp->m_dentry) !=
2048 d_is_dir(mnt->mnt.mnt_root))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 return -ENOTDIR;
2050
Al Viro84d17192013-03-15 10:53:28 -04002051 return attach_recursive_mnt(mnt, p, mp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052}
2053
2054/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002055 * Sanity check the flags to change_mnt_propagation.
2056 */
2057
2058static int flags_to_propagation_type(int flags)
2059{
Roman Borisov7c6e9842011-05-25 16:26:48 -07002060 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002061
2062 /* Fail if any non-propagation flags are set */
2063 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2064 return 0;
2065 /* Only one propagation flag should be set */
2066 if (!is_power_of_2(type))
2067 return 0;
2068 return type;
2069}
2070
2071/*
Ram Pai07b20882005-11-07 17:19:07 -05002072 * recursively change the type of the mountpoint.
2073 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002074static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05002075{
Al Viro315fc832011-11-24 18:57:30 -05002076 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05002077 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05002078 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002079 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002080 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05002081
Al Viro2d92ab32008-08-02 00:51:11 -04002082 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05002083 return -EINVAL;
2084
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002085 type = flags_to_propagation_type(flag);
2086 if (!type)
2087 return -EINVAL;
2088
Al Viro97216be2013-03-16 15:12:40 -04002089 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002090 if (type == MS_SHARED) {
2091 err = invent_group_ids(mnt, recurse);
2092 if (err)
2093 goto out_unlock;
2094 }
2095
Al Viro719ea2f2013-09-29 11:24:49 -04002096 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05002097 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05002098 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04002099 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002100
2101 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04002102 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002103 return err;
Ram Pai07b20882005-11-07 17:19:07 -05002104}
2105
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002106static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2107{
2108 struct mount *child;
2109 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2110 if (!is_subdir(child->mnt_mountpoint, dentry))
2111 continue;
2112
2113 if (child->mnt.mnt_flags & MNT_LOCKED)
2114 return true;
2115 }
2116 return false;
2117}
2118
Ram Pai07b20882005-11-07 17:19:07 -05002119/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 * do loopback mount.
2121 */
Al Viro808d4e32012-10-11 11:42:01 -04002122static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08002123 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124{
Al Viro2d92ab32008-08-02 00:51:11 -04002125 struct path old_path;
Al Viro84d17192013-03-15 10:53:28 -04002126 struct mount *mnt = NULL, *old, *parent;
2127 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002128 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 if (!old_name || !*old_name)
2130 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04002131 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 if (err)
2133 return err;
2134
Eric W. Biederman8823c072010-03-07 18:49:36 -08002135 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002136 if (mnt_ns_loop(old_path.dentry))
Eric W. Biederman8823c072010-03-07 18:49:36 -08002137 goto out;
2138
Al Viro84d17192013-03-15 10:53:28 -04002139 mp = lock_mount(path);
2140 err = PTR_ERR(mp);
2141 if (IS_ERR(mp))
Jan Blunck4ac91372008-02-14 19:34:32 -08002142 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05002143
Al Viro87129cc2011-11-24 21:24:27 -05002144 old = real_mount(old_path.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002145 parent = real_mount(path->mnt);
Al Viro87129cc2011-11-24 21:24:27 -05002146
Al Virob12cea92011-03-18 08:55:38 -04002147 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002148 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04002149 goto out2;
2150
Al Viroe149ed22014-11-01 10:57:28 -04002151 if (!check_mnt(parent))
2152 goto out2;
2153
2154 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
Al Virob12cea92011-03-18 08:55:38 -04002155 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002157 if (!recurse && has_locked_children(old, old_path.dentry))
2158 goto out2;
2159
Al Viroccd48bc2005-11-07 17:15:04 -05002160 if (recurse)
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002161 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
Al Viroccd48bc2005-11-07 17:15:04 -05002162 else
Al Viro87129cc2011-11-24 21:24:27 -05002163 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05002164
David Howellsbe34d1a2012-06-25 12:55:18 +01002165 if (IS_ERR(mnt)) {
2166 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04002167 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01002168 }
Al Viroccd48bc2005-11-07 17:15:04 -05002169
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002170 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2171
Al Viro84d17192013-03-15 10:53:28 -04002172 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002173 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002174 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06002175 umount_tree(mnt, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04002176 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05002177 }
Al Virob12cea92011-03-18 08:55:38 -04002178out2:
Al Viro84d17192013-03-15 10:53:28 -04002179 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002180out:
Al Viro2d92ab32008-08-02 00:51:11 -04002181 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 return err;
2183}
2184
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002185static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2186{
2187 int error = 0;
2188 int readonly_request = 0;
2189
2190 if (ms_flags & MS_RDONLY)
2191 readonly_request = 1;
2192 if (readonly_request == __mnt_is_readonly(mnt))
2193 return 0;
2194
2195 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05002196 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002197 else
Al Viro83adc752011-11-24 22:37:54 -05002198 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002199 return error;
2200}
2201
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202/*
2203 * change filesystem flags. dir should be a physical root of filesystem.
2204 * If you've mounted a non-root directory somewhere and want to do remount
2205 * on it - tough luck.
2206 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002207static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 void *data)
2209{
2210 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04002211 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05002212 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213
Al Viro143c8c92011-11-25 00:46:35 -05002214 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 return -EINVAL;
2216
Al Viro2d92ab32008-08-02 00:51:11 -04002217 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 return -EINVAL;
2219
Eric W. Biederman07b64552014-07-28 17:10:56 -07002220 /* Don't allow changing of locked mnt flags.
2221 *
2222 * No locks need to be held here while testing the various
2223 * MNT_LOCK flags because those flags can never be cleared
2224 * once they are set.
2225 */
2226 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2227 !(mnt_flags & MNT_READONLY)) {
2228 return -EPERM;
2229 }
Eric W. Biederman9566d672014-07-28 17:26:07 -07002230 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2231 !(mnt_flags & MNT_NODEV)) {
Eric W. Biederman67690f92016-05-18 13:50:06 -05002232 return -EPERM;
Eric W. Biederman9566d672014-07-28 17:26:07 -07002233 }
2234 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2235 !(mnt_flags & MNT_NOSUID)) {
2236 return -EPERM;
2237 }
2238 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2239 !(mnt_flags & MNT_NOEXEC)) {
2240 return -EPERM;
2241 }
2242 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2243 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2244 return -EPERM;
2245 }
2246
Eric Parisff36fe22011-03-03 16:09:14 -05002247 err = security_sb_remount(sb, data);
2248 if (err)
2249 return err;
2250
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002252 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002253 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05002254 else if (!capable(CAP_SYS_ADMIN))
2255 err = -EPERM;
Al Viro4aa98cf2009-05-08 13:36:58 -04002256 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002257 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05002258 if (!err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002259 lock_mount_hash();
Eric W. Biedermana6138db2014-07-28 16:26:53 -07002260 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
Al Viro143c8c92011-11-25 00:46:35 -05002261 mnt->mnt.mnt_flags = mnt_flags;
Al Viro143c8c92011-11-25 00:46:35 -05002262 touch_mnt_namespace(mnt->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04002263 unlock_mount_hash();
Dan Williams0e55a7c2008-09-26 19:01:20 -07002264 }
Al Viro6339dab2013-09-16 22:41:01 -04002265 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 return err;
2267}
2268
Al Virocbbe3622011-11-24 20:01:19 -05002269static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05002270{
Al Viro315fc832011-11-24 18:57:30 -05002271 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05002272 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002273 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05002274 return 1;
2275 }
2276 return 0;
2277}
2278
Al Viro808d4e32012-10-11 11:42:01 -04002279static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280{
Al Viro2d92ab32008-08-02 00:51:11 -04002281 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05002282 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05002283 struct mount *old;
Al Viro84d17192013-03-15 10:53:28 -04002284 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002285 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 if (!old_name || !*old_name)
2287 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002288 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 if (err)
2290 return err;
2291
Al Viro84d17192013-03-15 10:53:28 -04002292 mp = lock_mount(path);
2293 err = PTR_ERR(mp);
2294 if (IS_ERR(mp))
David Howellscc53ce52011-01-14 18:45:26 +00002295 goto out;
2296
Al Viro143c8c92011-11-25 00:46:35 -05002297 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002298 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05002299
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002301 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 goto out1;
2303
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002304 if (old->mnt.mnt_flags & MNT_LOCKED)
2305 goto out1;
2306
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002308 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05002309 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
Al Viro676da582011-11-24 21:47:05 -05002311 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05002312 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
David Howellse36cb0b2015-01-29 12:02:35 +00002314 if (d_is_dir(path->dentry) !=
2315 d_is_dir(old_path.dentry))
Ram Pai21444402005-11-07 17:20:03 -05002316 goto out1;
2317 /*
2318 * Don't move a mount residing in a shared parent.
2319 */
Al Virofc7be132011-11-25 01:05:37 -05002320 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002321 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05002322 /*
2323 * Don't move a mount tree containing unbindable mounts to a destination
2324 * mount which is shared.
2325 */
Al Virofc7be132011-11-25 01:05:37 -05002326 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05002327 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05002329 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05002330 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05002331 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332
Al Viro84d17192013-03-15 10:53:28 -04002333 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08002334 if (err)
Ram Pai21444402005-11-07 17:20:03 -05002335 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
2337 /* if the mount is moved, it should no longer be expire
2338 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05002339 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340out1:
Al Viro84d17192013-03-15 10:53:28 -04002341 unlock_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04002344 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04002345 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 return err;
2347}
2348
Al Viro9d412a42011-03-17 22:08:28 -04002349static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2350{
2351 int err;
2352 const char *subtype = strchr(fstype, '.');
2353 if (subtype) {
2354 subtype++;
2355 err = -EINVAL;
2356 if (!subtype[0])
2357 goto err;
2358 } else
2359 subtype = "";
2360
2361 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2362 err = -ENOMEM;
2363 if (!mnt->mnt_sb->s_subtype)
2364 goto err;
2365 return mnt;
2366
2367 err:
2368 mntput(mnt);
2369 return ERR_PTR(err);
2370}
2371
Al Viro9d412a42011-03-17 22:08:28 -04002372/*
2373 * add a mount into a namespace's mount tree
2374 */
Al Viro95bc5f22011-11-25 00:30:56 -05002375static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04002376{
Al Viro84d17192013-03-15 10:53:28 -04002377 struct mountpoint *mp;
2378 struct mount *parent;
Al Viro9d412a42011-03-17 22:08:28 -04002379 int err;
2380
Al Virof2ebb3a2014-02-27 09:35:45 -05002381 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04002382
Al Viro84d17192013-03-15 10:53:28 -04002383 mp = lock_mount(path);
2384 if (IS_ERR(mp))
2385 return PTR_ERR(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002386
Al Viro84d17192013-03-15 10:53:28 -04002387 parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04002388 err = -EINVAL;
Al Viro84d17192013-03-15 10:53:28 -04002389 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04002390 /* that's acceptable only for automounts done in private ns */
2391 if (!(mnt_flags & MNT_SHRINKABLE))
2392 goto unlock;
2393 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d17192013-03-15 10:53:28 -04002394 if (!parent->mnt_ns)
Al Viro156cacb2012-09-21 08:19:02 -04002395 goto unlock;
2396 }
Al Viro9d412a42011-03-17 22:08:28 -04002397
2398 /* Refuse the same filesystem on the same mount point */
2399 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05002400 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04002401 path->mnt->mnt_root == path->dentry)
2402 goto unlock;
2403
2404 err = -EINVAL;
David Howellse36cb0b2015-01-29 12:02:35 +00002405 if (d_is_symlink(newmnt->mnt.mnt_root))
Al Viro9d412a42011-03-17 22:08:28 -04002406 goto unlock;
2407
Al Viro95bc5f22011-11-25 00:30:56 -05002408 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro84d17192013-03-15 10:53:28 -04002409 err = graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04002410
2411unlock:
Al Viro84d17192013-03-15 10:53:28 -04002412 unlock_mount(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002413 return err;
2414}
Al Virob1e75df2011-01-17 01:47:59 -05002415
Eric W. Biederman8654df42016-06-09 16:06:06 -05002416static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05002417
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418/*
2419 * create a new mount for userspace and request it to be added into the
2420 * namespace's tree
2421 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002422static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04002423 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002425 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002427 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002429 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 return -EINVAL;
2431
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002432 type = get_fs_type(fstype);
2433 if (!type)
2434 return -ENODEV;
2435
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002436 mnt = vfs_kern_mount(type, flags, name, data);
2437 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2438 !mnt->mnt_sb->s_subtype)
2439 mnt = fs_set_subtype(mnt, fstype);
2440
2441 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 if (IS_ERR(mnt))
2443 return PTR_ERR(mnt);
2444
Eric W. Biederman8654df42016-06-09 16:06:06 -05002445 if (mount_too_revealing(mnt, &mnt_flags)) {
2446 mntput(mnt);
2447 return -EPERM;
2448 }
2449
Al Viro95bc5f22011-11-25 00:30:56 -05002450 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002451 if (err)
2452 mntput(mnt);
2453 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454}
2455
Al Viro19a167a2011-01-17 01:35:23 -05002456int finish_automount(struct vfsmount *m, struct path *path)
2457{
Al Viro6776db3d2011-11-25 00:22:05 -05002458 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002459 int err;
2460 /* The new mount record should have at least 2 refs to prevent it being
2461 * expired before we get a chance to add it
2462 */
Al Viro6776db3d2011-11-25 00:22:05 -05002463 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002464
2465 if (m->mnt_sb == path->mnt->mnt_sb &&
2466 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002467 err = -ELOOP;
2468 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002469 }
2470
Al Viro95bc5f22011-11-25 00:30:56 -05002471 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002472 if (!err)
2473 return 0;
2474fail:
2475 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002476 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002477 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05002478 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04002479 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002480 }
Al Virob1e75df2011-01-17 01:47:59 -05002481 mntput(m);
2482 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002483 return err;
2484}
2485
David Howellsea5b7782011-01-14 19:10:03 +00002486/**
2487 * mnt_set_expiry - Put a mount on an expiration list
2488 * @mnt: The mount to list.
2489 * @expiry_list: The list to add the mount to.
2490 */
2491void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2492{
Al Viro97216be2013-03-16 15:12:40 -04002493 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00002494
Al Viro6776db3d2011-11-25 00:22:05 -05002495 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002496
Al Viro97216be2013-03-16 15:12:40 -04002497 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002498}
2499EXPORT_SYMBOL(mnt_set_expiry);
2500
2501/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 * process a list of expirable mountpoints with the intent of discarding any
2503 * mountpoints that aren't in use and haven't been touched since last we came
2504 * here
2505 */
2506void mark_mounts_for_expiry(struct list_head *mounts)
2507{
Al Viro761d5c32011-11-24 21:07:43 -05002508 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 LIST_HEAD(graveyard);
2510
2511 if (list_empty(mounts))
2512 return;
2513
Al Viro97216be2013-03-16 15:12:40 -04002514 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002515 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
2517 /* extract from the expiration list every vfsmount that matches the
2518 * following criteria:
2519 * - only referenced by its parent vfsmount
2520 * - still marked for expiry (marked on the last call here; marks are
2521 * cleared by mntput())
2522 */
Al Viro6776db3d2011-11-25 00:22:05 -05002523 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002524 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002525 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002527 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002529 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002530 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002531 touch_mnt_namespace(mnt->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06002532 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04002533 }
Al Viro719ea2f2013-09-29 11:24:49 -04002534 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002535 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536}
2537
2538EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2539
2540/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002541 * Ripoff of 'select_parent()'
2542 *
2543 * search the list of submounts for a given mountpoint, and move any
2544 * shrinkable submounts to the 'graveyard' list.
2545 */
Al Viro692afc32011-11-24 21:15:14 -05002546static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002547{
Al Viro692afc32011-11-24 21:15:14 -05002548 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002549 struct list_head *next;
2550 int found = 0;
2551
2552repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002553 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002554resume:
Al Viro6b41d532011-11-24 23:24:33 -05002555 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002556 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002557 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002558
2559 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002560 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002561 continue;
2562 /*
2563 * Descend a level if the d_mounts list is non-empty.
2564 */
Al Viro6b41d532011-11-24 23:24:33 -05002565 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002566 this_parent = mnt;
2567 goto repeat;
2568 }
2569
Al Viro1ab59732011-11-24 21:35:16 -05002570 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002571 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002572 found++;
2573 }
2574 }
2575 /*
2576 * All done at this level ... ascend and resume the search
2577 */
2578 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002579 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002580 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002581 goto resume;
2582 }
2583 return found;
2584}
2585
2586/*
2587 * process a list of expirable mountpoints with the intent of discarding any
2588 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002589 *
Al Viro48a066e2013-09-29 22:06:07 -04002590 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002591 */
Al Virob54b9be2013-03-16 14:39:34 -04002592static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002593{
2594 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002595 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002596
Trond Myklebust5528f9112006-06-09 09:34:17 -04002597 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002598 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002599 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002600 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002601 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002602 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06002603 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04002604 }
2605 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002606}
2607
Trond Myklebust5528f9112006-06-09 09:34:17 -04002608/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 * Some copy_from_user() implementations do not return the exact number of
2610 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2611 * Note that this function differs from copy_from_user() in that it will oops
2612 * on bad values of `to', rather than returning a short copy.
2613 */
Ram Paib58fed82005-11-07 17:16:09 -05002614static long exact_copy_from_user(void *to, const void __user * from,
2615 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616{
2617 char *t = to;
2618 const char __user *f = from;
2619 char c;
2620
2621 if (!access_ok(VERIFY_READ, from, n))
2622 return n;
2623
2624 while (n) {
2625 if (__get_user(c, f)) {
2626 memset(t, 0, n);
2627 break;
2628 }
2629 *t++ = c;
2630 f++;
2631 n--;
2632 }
2633 return n;
2634}
2635
Al Virob40ef862015-12-14 18:44:44 -05002636void *copy_mount_options(const void __user * data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637{
2638 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 unsigned long size;
Al Virob40ef862015-12-14 18:44:44 -05002640 char *copy;
Ram Paib58fed82005-11-07 17:16:09 -05002641
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 if (!data)
Al Virob40ef862015-12-14 18:44:44 -05002643 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644
Al Virob40ef862015-12-14 18:44:44 -05002645 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2646 if (!copy)
2647 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648
2649 /* We only care that *some* data at the address the user
2650 * gave us is valid. Just in case, we'll zero
2651 * the remainder of the page.
2652 */
2653 /* copy_from_user cannot cross TASK_SIZE ! */
2654 size = TASK_SIZE - (unsigned long)data;
2655 if (size > PAGE_SIZE)
2656 size = PAGE_SIZE;
2657
Al Virob40ef862015-12-14 18:44:44 -05002658 i = size - exact_copy_from_user(copy, data, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 if (!i) {
Al Virob40ef862015-12-14 18:44:44 -05002660 kfree(copy);
2661 return ERR_PTR(-EFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 }
2663 if (i != PAGE_SIZE)
Al Virob40ef862015-12-14 18:44:44 -05002664 memset(copy + i, 0, PAGE_SIZE - i);
2665 return copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666}
2667
Tim Gardnerb8850d12014-08-28 11:26:03 -06002668char *copy_mount_string(const void __user *data)
Vegard Nossumeca6f532009-09-18 13:05:45 -07002669{
Tim Gardnerb8850d12014-08-28 11:26:03 -06002670 return data ? strndup_user(data, PAGE_SIZE) : NULL;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002671}
2672
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673/*
2674 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2675 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2676 *
2677 * data is a (void *) that can point to any structure up to
2678 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2679 * information (or be NULL).
2680 *
2681 * Pre-0.97 versions of mount() didn't have a flags word.
2682 * When the flags word was introduced its top half was required
2683 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2684 * Therefore, if this magic number is present, it carries no information
2685 * and must be discarded.
2686 */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002687long do_mount(const char *dev_name, const char __user *dir_name,
Al Viro808d4e32012-10-11 11:42:01 -04002688 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689{
Al Viro2d92ab32008-08-02 00:51:11 -04002690 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 int retval = 0;
2692 int mnt_flags = 0;
2693
2694 /* Discard magic */
2695 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2696 flags &= ~MS_MGC_MSK;
2697
2698 /* Basic sanity checks */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 if (data_page)
2700 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2701
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002702 /* ... and get the mountpoint */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002703 retval = user_path(dir_name, &path);
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002704 if (retval)
2705 return retval;
2706
2707 retval = security_sb_mount(dev_name, &path,
2708 type_page, flags, data_page);
Al Viro0d5cadb2013-05-04 14:40:51 -04002709 if (!retval && !may_mount())
2710 retval = -EPERM;
Jeff Layton9e8925b2015-11-16 09:49:34 -05002711 if (!retval && (flags & MS_MANDLOCK) && !may_mandlock())
2712 retval = -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002713 if (retval)
2714 goto dput_out;
2715
Andi Kleen613cbe32009-04-19 18:40:43 +02002716 /* Default to relatime unless overriden */
2717 if (!(flags & MS_NOATIME))
2718 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002719
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 /* Separate the per-mountpoint flags */
2721 if (flags & MS_NOSUID)
2722 mnt_flags |= MNT_NOSUID;
2723 if (flags & MS_NODEV)
2724 mnt_flags |= MNT_NODEV;
2725 if (flags & MS_NOEXEC)
2726 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002727 if (flags & MS_NOATIME)
2728 mnt_flags |= MNT_NOATIME;
2729 if (flags & MS_NODIRATIME)
2730 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002731 if (flags & MS_STRICTATIME)
2732 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002733 if (flags & MS_RDONLY)
2734 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002735
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07002736 /* The default atime for remount is preservation */
2737 if ((flags & MS_REMOUNT) &&
2738 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2739 MS_STRICTATIME)) == 0)) {
2740 mnt_flags &= ~MNT_ATIME_MASK;
2741 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2742 }
2743
Al Viro7a4dec52010-08-09 12:05:43 -04002744 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002745 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
Miklos Szeredic568d682016-09-16 12:44:20 +02002746 MS_STRICTATIME | MS_NOREMOTELOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002749 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 data_page);
2751 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002752 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002753 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002754 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002756 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 else
Al Viro2d92ab32008-08-02 00:51:11 -04002758 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 dev_name, data_page);
2760dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002761 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 return retval;
2763}
2764
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002765static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2766{
2767 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2768}
2769
2770static void dec_mnt_namespaces(struct ucounts *ucounts)
2771{
2772 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2773}
2774
Eric W. Biederman771b1372012-07-26 21:08:32 -07002775static void free_mnt_ns(struct mnt_namespace *ns)
2776{
Al Viro6344c432014-11-01 00:45:45 -04002777 ns_free_inum(&ns->ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002778 dec_mnt_namespaces(ns->ucounts);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002779 put_user_ns(ns->user_ns);
2780 kfree(ns);
2781}
2782
Eric W. Biederman8823c072010-03-07 18:49:36 -08002783/*
2784 * Assign a sequence number so we can detect when we attempt to bind
2785 * mount a reference to an older mount namespace into the current
2786 * mount namespace, preventing reference counting loops. A 64bit
2787 * number incrementing at 10Ghz will take 12,427 years to wrap which
2788 * is effectively never, so we can ignore the possibility.
2789 */
2790static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2791
Eric W. Biederman771b1372012-07-26 21:08:32 -07002792static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002793{
2794 struct mnt_namespace *new_ns;
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002795 struct ucounts *ucounts;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002796 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002797
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002798 ucounts = inc_mnt_namespaces(user_ns);
2799 if (!ucounts)
Eric W. Biedermandf75e772016-09-22 13:08:36 -05002800 return ERR_PTR(-ENOSPC);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002801
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002802 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002803 if (!new_ns) {
2804 dec_mnt_namespaces(ucounts);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002805 return ERR_PTR(-ENOMEM);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002806 }
Al Viro6344c432014-11-01 00:45:45 -04002807 ret = ns_alloc_inum(&new_ns->ns);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002808 if (ret) {
2809 kfree(new_ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002810 dec_mnt_namespaces(ucounts);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002811 return ERR_PTR(ret);
2812 }
Al Viro33c42942014-11-01 02:32:53 -04002813 new_ns->ns.ops = &mntns_operations;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002814 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002815 atomic_set(&new_ns->count, 1);
2816 new_ns->root = NULL;
2817 INIT_LIST_HEAD(&new_ns->list);
2818 init_waitqueue_head(&new_ns->poll);
2819 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002820 new_ns->user_ns = get_user_ns(user_ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05002821 new_ns->ucounts = ucounts;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002822 new_ns->mounts = 0;
2823 new_ns->pending_mounts = 0;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002824 return new_ns;
2825}
2826
Emese Revfy0766f782016-06-20 20:42:34 +02002827__latent_entropy
Al Viro9559f682013-09-28 20:47:57 -04002828struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2829 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002831 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002832 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002833 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04002834 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05002835 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002836 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837
Al Viro9559f682013-09-28 20:47:57 -04002838 BUG_ON(!ns);
2839
2840 if (likely(!(flags & CLONE_NEWNS))) {
2841 get_mnt_ns(ns);
2842 return ns;
2843 }
2844
2845 old = ns->root;
2846
Eric W. Biederman771b1372012-07-26 21:08:32 -07002847 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002848 if (IS_ERR(new_ns))
2849 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850
Al Viro97216be2013-03-16 15:12:40 -04002851 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002853 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04002854 if (user_ns != ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002855 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002856 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002857 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04002858 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002859 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002860 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 }
Al Virobe08d6d2011-12-06 13:32:36 -05002862 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002863 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864
2865 /*
2866 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2867 * as belonging to new namespace. We have already acquired a private
2868 * fs_struct, so tsk->fs->lock is not needed.
2869 */
Al Viro909b0a82011-11-25 03:06:56 -05002870 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002871 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002873 q->mnt_ns = new_ns;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002874 new_ns->mounts++;
Al Viro9559f682013-09-28 20:47:57 -04002875 if (new_fs) {
2876 if (&p->mnt == new_fs->root.mnt) {
2877 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002878 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 }
Al Viro9559f682013-09-28 20:47:57 -04002880 if (&p->mnt == new_fs->pwd.mnt) {
2881 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002882 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 }
Al Viro909b0a82011-11-25 03:06:56 -05002885 p = next_mnt(p, old);
2886 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002887 if (!q)
2888 break;
2889 while (p->mnt.mnt_root != q->mnt.mnt_root)
2890 p = next_mnt(p, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 }
Al Viro328e6d92013-03-16 14:49:45 -04002892 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002895 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002897 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898
JANAK DESAI741a2952006-02-07 12:59:00 -08002899 return new_ns;
2900}
2901
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002902/**
2903 * create_mnt_ns - creates a private namespace and adds a root filesystem
2904 * @mnt: pointer to the new root filesystem mountpoint
2905 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002906static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002907{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002908 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002909 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002910 struct mount *mnt = real_mount(m);
2911 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002912 new_ns->root = mnt;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002913 new_ns->mounts++;
Al Virob1983cd2013-05-04 15:18:53 -04002914 list_add(&mnt->mnt_list, &new_ns->list);
Al Viroc1334492011-11-16 16:12:14 -05002915 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002916 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002917 }
2918 return new_ns;
2919}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002920
Al Viroea441d12011-11-16 21:43:59 -05002921struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2922{
2923 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002924 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002925 struct path path;
2926 int err;
2927
2928 ns = create_mnt_ns(mnt);
2929 if (IS_ERR(ns))
2930 return ERR_CAST(ns);
2931
2932 err = vfs_path_lookup(mnt->mnt_root, mnt,
2933 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2934
2935 put_mnt_ns(ns);
2936
2937 if (err)
2938 return ERR_PTR(err);
2939
2940 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002941 s = path.mnt->mnt_sb;
2942 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002943 mntput(path.mnt);
2944 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002945 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002946 /* ... and return the root of (sub)tree on it */
2947 return path.dentry;
2948}
2949EXPORT_SYMBOL(mount_subtree);
2950
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002951SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2952 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002954 int ret;
2955 char *kernel_type;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002956 char *kernel_dev;
Al Virob40ef862015-12-14 18:44:44 -05002957 void *options;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958
Tim Gardnerb8850d12014-08-28 11:26:03 -06002959 kernel_type = copy_mount_string(type);
2960 ret = PTR_ERR(kernel_type);
2961 if (IS_ERR(kernel_type))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002962 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963
Tim Gardnerb8850d12014-08-28 11:26:03 -06002964 kernel_dev = copy_mount_string(dev_name);
2965 ret = PTR_ERR(kernel_dev);
2966 if (IS_ERR(kernel_dev))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002967 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968
Al Virob40ef862015-12-14 18:44:44 -05002969 options = copy_mount_options(data);
2970 ret = PTR_ERR(options);
2971 if (IS_ERR(options))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002972 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973
Al Virob40ef862015-12-14 18:44:44 -05002974 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002975
Al Virob40ef862015-12-14 18:44:44 -05002976 kfree(options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002977out_data:
2978 kfree(kernel_dev);
2979out_dev:
Vegard Nossumeca6f532009-09-18 13:05:45 -07002980 kfree(kernel_type);
2981out_type:
2982 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983}
2984
2985/*
Al Viroafac7cb2011-11-23 19:34:49 -05002986 * Return true if path is reachable from root
2987 *
Al Viro48a066e2013-09-29 22:06:07 -04002988 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05002989 */
Al Viro643822b2011-11-24 22:00:28 -05002990bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002991 const struct path *root)
2992{
Al Viro643822b2011-11-24 22:00:28 -05002993 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002994 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002995 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002996 }
Al Viro643822b2011-11-24 22:00:28 -05002997 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002998}
2999
Yaowei Bai25ab4c92015-11-17 14:40:10 +08003000bool path_is_under(struct path *path1, struct path *path2)
Al Viroafac7cb2011-11-23 19:34:49 -05003001{
Yaowei Bai25ab4c92015-11-17 14:40:10 +08003002 bool res;
Al Viro48a066e2013-09-29 22:06:07 -04003003 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05003004 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04003005 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05003006 return res;
3007}
3008EXPORT_SYMBOL(path_is_under);
3009
3010/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 * pivot_root Semantics:
3012 * Moves the root file system of the current process to the directory put_old,
3013 * makes new_root as the new root file system of the current process, and sets
3014 * root/cwd of all processes which had them on the current root to new_root.
3015 *
3016 * Restrictions:
3017 * The new_root and put_old must be directories, and must not be on the
3018 * same file system as the current process root. The put_old must be
3019 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3020 * pointed to by put_old must yield the same directory as new_root. No other
3021 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3022 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08003023 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3024 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3025 * in this situation.
3026 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027 * Notes:
3028 * - we don't move root/cwd if they are not at the root (reason: if something
3029 * cared enough to change them, it's probably wrong to force them elsewhere)
3030 * - it's okay to pick a root that isn't the root of a file system, e.g.
3031 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3032 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3033 * first.
3034 */
Heiko Carstens3480b252009-01-14 14:14:16 +01003035SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3036 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037{
Al Viro2d8f3032008-07-22 09:59:21 -04003038 struct path new, old, parent_path, root_parent, root;
Al Viro84d17192013-03-15 10:53:28 -04003039 struct mount *new_mnt, *root_mnt, *old_mnt;
3040 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 int error;
3042
Al Viro9b40bc92013-02-22 22:45:42 -05003043 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 return -EPERM;
3045
Al Viro2d8f3032008-07-22 09:59:21 -04003046 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047 if (error)
3048 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049
Al Viro2d8f3032008-07-22 09:59:21 -04003050 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 if (error)
3052 goto out1;
3053
Al Viro2d8f3032008-07-22 09:59:21 -04003054 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04003055 if (error)
3056 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02003058 get_fs_root(current->fs, &root);
Al Viro84d17192013-03-15 10:53:28 -04003059 old_mp = lock_mount(&old);
3060 error = PTR_ERR(old_mp);
3061 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04003062 goto out3;
3063
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05003065 new_mnt = real_mount(new.mnt);
3066 root_mnt = real_mount(root.mnt);
Al Viro84d17192013-03-15 10:53:28 -04003067 old_mnt = real_mount(old.mnt);
3068 if (IS_MNT_SHARED(old_mnt) ||
Al Virofc7be132011-11-25 01:05:37 -05003069 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3070 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04003071 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05003072 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003073 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003074 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3075 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003077 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04003078 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 error = -EBUSY;
Al Viro84d17192013-03-15 10:53:28 -04003080 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04003081 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04003083 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04003084 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05003085 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003086 goto out4; /* not attached */
Al Viro84d17192013-03-15 10:53:28 -04003087 root_mp = root_mnt->mnt_mp;
Al Viro2d8f3032008-07-22 09:59:21 -04003088 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04003089 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05003090 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003091 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08003092 /* make sure we can reach put_old from new_root */
Al Viro84d17192013-03-15 10:53:28 -04003093 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04003094 goto out4;
Eric W. Biederman0d082602014-10-08 10:42:27 -07003095 /* make certain new is below the root */
3096 if (!is_path_reachable(new_mnt, new.dentry, &root))
3097 goto out4;
Al Viro84d17192013-03-15 10:53:28 -04003098 root_mp->m_count++; /* pin it so it won't go away */
Al Viro719ea2f2013-09-29 11:24:49 -04003099 lock_mount_hash();
Al Viro419148d2011-11-24 19:41:16 -05003100 detach_mnt(new_mnt, &parent_path);
3101 detach_mnt(root_mnt, &root_parent);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003102 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3103 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3104 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3105 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003106 /* mount old root on put_old */
Al Viro84d17192013-03-15 10:53:28 -04003107 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08003108 /* mount new_root on / */
Al Viro84d17192013-03-15 10:53:28 -04003109 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003110 touch_mnt_namespace(current->nsproxy->mnt_ns);
Eric W. Biederman4fed6552014-10-08 10:42:57 -07003111 /* A moved mount should not expire automatically */
3112 list_del_init(&new_mnt->mnt_expire);
Al Viro719ea2f2013-09-29 11:24:49 -04003113 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04003114 chroot_fs_refs(&root, &new);
Al Viro84d17192013-03-15 10:53:28 -04003115 put_mountpoint(root_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04003117out4:
Al Viro84d17192013-03-15 10:53:28 -04003118 unlock_mount(old_mp);
Al Virob12cea92011-03-18 08:55:38 -04003119 if (!error) {
3120 path_put(&root_parent);
3121 path_put(&parent_path);
3122 }
3123out3:
Al Viro8c3ee422008-03-22 18:00:39 -04003124 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04003125out2:
Al Viro2d8f3032008-07-22 09:59:21 -04003126 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127out1:
Al Viro2d8f3032008-07-22 09:59:21 -04003128 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131}
3132
3133static void __init init_mount_tree(void)
3134{
3135 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003136 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08003137 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003138 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003140 type = get_fs_type("rootfs");
3141 if (!type)
3142 panic("Can't find rootfs type");
3143 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3144 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 if (IS_ERR(mnt))
3146 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11003147
Trond Myklebust3b22edc2009-06-23 17:29:49 -04003148 ns = create_mnt_ns(mnt);
3149 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003152 init_task.nsproxy->mnt_ns = ns;
3153 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154
Al Virobe08d6d2011-12-06 13:32:36 -05003155 root.mnt = mnt;
3156 root.dentry = mnt->mnt_root;
Eric W. Biedermanda362b02014-10-07 12:19:53 -07003157 mnt->mnt_flags |= MNT_LOCKED;
Jan Blunckac748a02008-02-14 19:34:39 -08003158
3159 set_fs_pwd(current->fs, &root);
3160 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161}
3162
Denis Cheng74bf17c2007-10-16 23:26:30 -07003163void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164{
Eric Dumazet13f14b42008-02-06 01:37:57 -08003165 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003166 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167
Al Viro7d6fec42011-11-23 12:14:10 -05003168 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09003169 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170
Al Viro0818bf22014-02-28 13:46:44 -05003171 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04003172 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05003173 mhash_entries, 19,
3174 0,
3175 &m_hash_shift, &m_hash_mask, 0, 0);
3176 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3177 sizeof(struct hlist_head),
3178 mphash_entries, 19,
3179 0,
3180 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181
Al Viro84d17192013-03-15 10:53:28 -04003182 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 panic("Failed to allocate mount hash table\n");
3184
Al Viro0818bf22014-02-28 13:46:44 -05003185 for (u = 0; u <= m_hash_mask; u++)
Al Viro38129a12014-03-20 21:10:51 -04003186 INIT_HLIST_HEAD(&mount_hashtable[u]);
Al Viro0818bf22014-02-28 13:46:44 -05003187 for (u = 0; u <= mp_hash_mask; u++)
3188 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189
Tejun Heo4b93dc92013-11-28 14:54:43 -05003190 kernfs_init();
3191
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003192 err = sysfs_init();
3193 if (err)
3194 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003195 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06003196 fs_kobj = kobject_create_and_add("fs", NULL);
3197 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003198 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199 init_rootfs();
3200 init_mount_tree();
3201}
3202
Trond Myklebust616511d2009-06-22 15:09:13 -04003203void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204{
Al Virod498b252010-02-05 02:21:06 -05003205 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04003206 return;
Al Viro7b00ed62013-09-16 21:19:20 -04003207 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003208 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209}
Al Viro9d412a42011-03-17 22:08:28 -04003210
3211struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3212{
Tim Chen423e0ab2011-07-19 09:32:38 -07003213 struct vfsmount *mnt;
3214 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3215 if (!IS_ERR(mnt)) {
3216 /*
3217 * it is a longterm mount, don't release mnt until
3218 * we unmount before file sys is unregistered
3219 */
Al Virof7a99c52012-06-09 00:59:08 -04003220 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07003221 }
3222 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04003223}
3224EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07003225
3226void kern_unmount(struct vfsmount *mnt)
3227{
3228 /* release long term mount so mount point can be released */
3229 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04003230 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04003231 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07003232 mntput(mnt);
3233 }
3234}
3235EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05003236
3237bool our_mnt(struct vfsmount *mnt)
3238{
Al Viro143c8c92011-11-25 00:46:35 -05003239 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05003240}
Eric W. Biederman8823c072010-03-07 18:49:36 -08003241
Eric W. Biederman31515272013-03-15 01:45:51 -07003242bool current_chrooted(void)
3243{
3244 /* Does the current process have a non-standard root */
3245 struct path ns_root;
3246 struct path fs_root;
3247 bool chrooted;
3248
3249 /* Find the namespace root */
3250 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3251 ns_root.dentry = ns_root.mnt->mnt_root;
3252 path_get(&ns_root);
3253 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3254 ;
3255
3256 get_fs_root(current->fs, &fs_root);
3257
3258 chrooted = !path_equal(&fs_root, &ns_root);
3259
3260 path_put(&fs_root);
3261 path_put(&ns_root);
3262
3263 return chrooted;
3264}
3265
Eric W. Biederman8654df42016-06-09 16:06:06 -05003266static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3267 int *new_mnt_flags)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003268{
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003269 int new_flags = *new_mnt_flags;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003270 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07003271 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003272
Al Viro44bb4382013-09-16 21:37:36 -04003273 down_read(&namespace_sem);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003274 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07003275 struct mount *child;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003276 int mnt_flags;
3277
Eric W. Biederman8654df42016-06-09 16:06:06 -05003278 if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
Eric W. Biedermane51db732013-03-30 19:57:41 -07003279 continue;
3280
Eric W. Biederman7e96c1b2015-05-08 16:36:50 -05003281 /* This mount is not fully visible if it's root directory
3282 * is not the root directory of the filesystem.
3283 */
3284 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3285 continue;
3286
Eric W. Biedermana1935c12016-06-15 06:59:49 -05003287 /* A local view of the mount flags */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003288 mnt_flags = mnt->mnt.mnt_flags;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003289
Eric W. Biederman695e9df2016-06-10 12:21:40 -05003290 /* Don't miss readonly hidden in the superblock flags */
3291 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3292 mnt_flags |= MNT_LOCK_READONLY;
3293
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003294 /* Verify the mount flags are equal to or more permissive
3295 * than the proposed new mount.
3296 */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003297 if ((mnt_flags & MNT_LOCK_READONLY) &&
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003298 !(new_flags & MNT_READONLY))
3299 continue;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003300 if ((mnt_flags & MNT_LOCK_ATIME) &&
3301 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003302 continue;
3303
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003304 /* This mount is not fully visible if there are any
3305 * locked child mounts that cover anything except for
3306 * empty directories.
Eric W. Biedermane51db732013-03-30 19:57:41 -07003307 */
3308 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3309 struct inode *inode = child->mnt_mountpoint->d_inode;
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003310 /* Only worry about locked mounts */
Eric W. Biedermand71ed6c2016-05-27 14:50:05 -05003311 if (!(child->mnt.mnt_flags & MNT_LOCKED))
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003312 continue;
Eric W. Biederman7236c852015-05-13 20:51:09 -05003313 /* Is the directory permanetly empty? */
3314 if (!is_empty_dir_inode(inode))
Eric W. Biedermane51db732013-03-30 19:57:41 -07003315 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003316 }
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003317 /* Preserve the locked attributes */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003318 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003319 MNT_LOCK_ATIME);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003320 visible = true;
3321 goto found;
3322 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003323 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07003324found:
Al Viro44bb4382013-09-16 21:37:36 -04003325 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003326 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003327}
3328
Eric W. Biederman8654df42016-06-09 16:06:06 -05003329static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3330{
Eric W. Biedermana1935c12016-06-15 06:59:49 -05003331 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
Eric W. Biederman8654df42016-06-09 16:06:06 -05003332 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3333 unsigned long s_iflags;
3334
3335 if (ns->user_ns == &init_user_ns)
3336 return false;
3337
3338 /* Can this filesystem be too revealing? */
3339 s_iflags = mnt->mnt_sb->s_iflags;
3340 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3341 return false;
3342
Eric W. Biedermana1935c12016-06-15 06:59:49 -05003343 if ((s_iflags & required_iflags) != required_iflags) {
3344 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3345 required_iflags);
3346 return true;
3347 }
3348
Eric W. Biederman8654df42016-06-09 16:06:06 -05003349 return !mnt_already_visible(ns, mnt, new_mnt_flags);
3350}
3351
Andy Lutomirski380cf5b2016-06-23 16:41:05 -05003352bool mnt_may_suid(struct vfsmount *mnt)
3353{
3354 /*
3355 * Foreign mounts (accessed via fchdir or through /proc
3356 * symlinks) are always treated as if they are nosuid. This
3357 * prevents namespaces from trusting potentially unsafe
3358 * suid/sgid bits, file caps, or security labels that originate
3359 * in other namespaces.
3360 */
3361 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3362 current_in_userns(mnt->mnt_sb->s_user_ns);
3363}
3364
Al Viro64964522014-11-01 00:37:32 -04003365static struct ns_common *mntns_get(struct task_struct *task)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003366{
Al Viro58be28252014-11-01 00:00:23 -04003367 struct ns_common *ns = NULL;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003368 struct nsproxy *nsproxy;
3369
Eric W. Biederman728dba32014-02-03 19:13:49 -08003370 task_lock(task);
3371 nsproxy = task->nsproxy;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003372 if (nsproxy) {
Al Viro58be28252014-11-01 00:00:23 -04003373 ns = &nsproxy->mnt_ns->ns;
3374 get_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003375 }
Eric W. Biederman728dba32014-02-03 19:13:49 -08003376 task_unlock(task);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003377
3378 return ns;
3379}
3380
Al Viro64964522014-11-01 00:37:32 -04003381static void mntns_put(struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003382{
Al Viro58be28252014-11-01 00:00:23 -04003383 put_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003384}
3385
Al Viro64964522014-11-01 00:37:32 -04003386static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003387{
3388 struct fs_struct *fs = current->fs;
Al Viro58be28252014-11-01 00:00:23 -04003389 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003390 struct path root;
3391
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003392 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -07003393 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3394 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08003395 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003396
3397 if (fs->users != 1)
3398 return -EINVAL;
3399
3400 get_mnt_ns(mnt_ns);
3401 put_mnt_ns(nsproxy->mnt_ns);
3402 nsproxy->mnt_ns = mnt_ns;
3403
3404 /* Find the root */
3405 root.mnt = &mnt_ns->root->mnt;
3406 root.dentry = mnt_ns->root->mnt.mnt_root;
3407 path_get(&root);
3408 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3409 ;
3410
3411 /* Update the pwd and root */
3412 set_fs_pwd(fs, &root);
3413 set_fs_root(fs, &root);
3414
3415 path_put(&root);
3416 return 0;
3417}
3418
Andrey Vaginbcac25a2016-09-06 00:47:13 -07003419static struct user_namespace *mntns_owner(struct ns_common *ns)
3420{
3421 return to_mnt_ns(ns)->user_ns;
3422}
3423
Eric W. Biederman8823c072010-03-07 18:49:36 -08003424const struct proc_ns_operations mntns_operations = {
3425 .name = "mnt",
3426 .type = CLONE_NEWNS,
3427 .get = mntns_get,
3428 .put = mntns_put,
3429 .install = mntns_install,
Andrey Vaginbcac25a2016-09-06 00:47:13 -07003430 .owner = mntns_owner,
Eric W. Biederman8823c072010-03-07 18:49:36 -08003431};