blob: 77ffdb82f63fdcbb32e114ea2cc494ef87fc0115 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
Al Virod10577a2011-12-07 13:06:11 -050012#include <linux/export.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080013#include <linux/capability.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080014#include <linux/mnt_namespace.h>
Eric W. Biederman771b1372012-07-26 21:08:32 -070015#include <linux/user_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/namei.h>
17#include <linux/security.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010018#include <linux/idr.h>
Rob Landley57f150a2013-09-11 14:26:10 -070019#include <linux/init.h> /* init_rootfs */
Al Virod10577a2011-12-07 13:06:11 -050020#include <linux/fs_struct.h> /* get_fs_root et.al. */
21#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22#include <linux/uaccess.h>
David Howells0bb80f22013-04-12 01:50:06 +010023#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070024#include <linux/magic.h>
Al Viro0818bf22014-02-28 13:46:44 -050025#include <linux/bootmem.h>
Al Viro9ea459e2014-08-08 13:08:20 -040026#include <linux/task_work.h>
Ram Pai07b20882005-11-07 17:19:07 -050027#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070028#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Al Viro0818bf22014-02-28 13:46:44 -050030static unsigned int m_hash_mask __read_mostly;
31static unsigned int m_hash_shift __read_mostly;
32static unsigned int mp_hash_mask __read_mostly;
33static unsigned int mp_hash_shift __read_mostly;
34
35static __initdata unsigned long mhash_entries;
36static int __init set_mhash_entries(char *str)
37{
38 if (!str)
39 return 0;
40 mhash_entries = simple_strtoul(str, &str, 0);
41 return 1;
42}
43__setup("mhash_entries=", set_mhash_entries);
44
45static __initdata unsigned long mphash_entries;
46static int __init set_mphash_entries(char *str)
47{
48 if (!str)
49 return 0;
50 mphash_entries = simple_strtoul(str, &str, 0);
51 return 1;
52}
53__setup("mphash_entries=", set_mphash_entries);
Eric Dumazet13f14b42008-02-06 01:37:57 -080054
Al Viroc7999c32014-02-27 14:40:10 -050055static u64 event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010056static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010057static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100058static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040059static int mnt_id_start = 0;
60static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Al Viro38129a12014-03-20 21:10:51 -040062static struct hlist_head *mount_hashtable __read_mostly;
Al Viro0818bf22014-02-28 13:46:44 -050063static struct hlist_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080064static struct kmem_cache *mnt_cache __read_mostly;
Al Viro59aa0da2013-09-16 21:34:53 -040065static DECLARE_RWSEM(namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080067/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060068struct kobject *fs_kobj;
69EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080070
Nick Piggin99b7db72010-08-18 04:37:39 +100071/*
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
74 * up the tree.
75 *
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
78 */
Al Viro48a066e2013-09-29 22:06:07 -040079__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +100080
Al Viro38129a12014-03-20 21:10:51 -040081static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082{
Ram Paib58fed82005-11-07 17:16:09 -050083 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Al Viro0818bf22014-02-28 13:46:44 -050085 tmp = tmp + (tmp >> m_hash_shift);
86 return &mount_hashtable[tmp & m_hash_mask];
87}
88
89static inline struct hlist_head *mp_hash(struct dentry *dentry)
90{
91 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 tmp = tmp + (tmp >> mp_hash_shift);
93 return &mountpoint_hashtable[tmp & mp_hash_mask];
Linus Torvalds1da177e2005-04-16 15:20:36 -070094}
95
Nick Piggin99b7db72010-08-18 04:37:39 +100096/*
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
99 */
Al Virob105e272011-11-24 20:38:33 -0500100static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100101{
102 int res;
103
104retry:
105 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +1000106 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -0500107 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -0400108 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500109 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +1000110 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100111 if (res == -EAGAIN)
112 goto retry;
113
114 return res;
115}
116
Al Virob105e272011-11-24 20:38:33 -0500117static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100118{
Al Viro15169fe2011-11-25 00:50:41 -0500119 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000120 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -0400121 ida_remove(&mnt_id_ida, id);
122 if (mnt_id_start > id)
123 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000124 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100125}
126
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100127/*
128 * Allocate a new peer group ID
129 *
130 * mnt_group_ida is protected by namespace_sem
131 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500132static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100133{
Al Virof21f6222009-06-24 03:12:00 -0400134 int res;
135
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100136 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
137 return -ENOMEM;
138
Al Virof21f6222009-06-24 03:12:00 -0400139 res = ida_get_new_above(&mnt_group_ida,
140 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500141 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400142 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500143 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400144
145 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100146}
147
148/*
149 * Release a peer group ID
150 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500151void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100152{
Al Viro15169fe2011-11-25 00:50:41 -0500153 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400154 ida_remove(&mnt_group_ida, id);
155 if (mnt_group_start > id)
156 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500157 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100158}
159
Nick Pigginb3e19d92011-01-07 17:50:11 +1100160/*
161 * vfsmount lock must be held for read
162 */
Al Viro83adc752011-11-24 22:37:54 -0500163static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100164{
165#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500166 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100167#else
168 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500169 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100170 preempt_enable();
171#endif
172}
173
Nick Pigginb3e19d92011-01-07 17:50:11 +1100174/*
175 * vfsmount lock must be held for write
176 */
Al Viro83adc752011-11-24 22:37:54 -0500177unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100178{
179#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500180 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100181 int cpu;
182
183 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500184 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100185 }
186
187 return count;
188#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500189 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100190#endif
191}
192
Al Virob105e272011-11-24 20:38:33 -0500193static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194{
Al Viroc63181e2011-11-25 02:35:16 -0500195 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
196 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100197 int err;
198
Al Viroc63181e2011-11-25 02:35:16 -0500199 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800200 if (err)
201 goto out_free_cache;
202
203 if (name) {
Al Viroc63181e2011-11-25 02:35:16 -0500204 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
205 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800206 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100207 }
208
Nick Pigginb3e19d92011-01-07 17:50:11 +1100209#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500210 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
211 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100212 goto out_free_devname;
213
Al Viroc63181e2011-11-25 02:35:16 -0500214 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100215#else
Al Viroc63181e2011-11-25 02:35:16 -0500216 mnt->mnt_count = 1;
217 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100218#endif
219
Al Viro38129a12014-03-20 21:10:51 -0400220 INIT_HLIST_NODE(&mnt->mnt_hash);
Al Viroc63181e2011-11-25 02:35:16 -0500221 INIT_LIST_HEAD(&mnt->mnt_child);
222 INIT_LIST_HEAD(&mnt->mnt_mounts);
223 INIT_LIST_HEAD(&mnt->mnt_list);
224 INIT_LIST_HEAD(&mnt->mnt_expire);
225 INIT_LIST_HEAD(&mnt->mnt_share);
226 INIT_LIST_HEAD(&mnt->mnt_slave_list);
227 INIT_LIST_HEAD(&mnt->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500228#ifdef CONFIG_FSNOTIFY
229 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
230#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 }
Al Viroc63181e2011-11-25 02:35:16 -0500232 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800233
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000234#ifdef CONFIG_SMP
235out_free_devname:
Al Viroc63181e2011-11-25 02:35:16 -0500236 kfree(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000237#endif
Li Zefan88b38782008-07-21 18:06:36 +0800238out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500239 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800240out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500241 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800242 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
244
Dave Hansen83660252008-02-15 14:37:30 -0800245/*
246 * Most r/o checks on a fs are for operations that take
247 * discrete amounts of time, like a write() or unlink().
248 * We must keep track of when those operations start
249 * (for permission checks) and when they end, so that
250 * we can determine when writes are able to occur to
251 * a filesystem.
252 */
Dave Hansen3d733632008-02-15 14:37:59 -0800253/*
254 * __mnt_is_readonly: check whether a mount is read-only
255 * @mnt: the mount to check for its write status
256 *
257 * This shouldn't be used directly ouside of the VFS.
258 * It does not guarantee that the filesystem will stay
259 * r/w, just that it is right *now*. This can not and
260 * should not be used in place of IS_RDONLY(inode).
261 * mnt_want/drop_write() will _keep_ the filesystem
262 * r/w.
263 */
264int __mnt_is_readonly(struct vfsmount *mnt)
265{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800266 if (mnt->mnt_flags & MNT_READONLY)
267 return 1;
268 if (mnt->mnt_sb->s_flags & MS_RDONLY)
269 return 1;
270 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800271}
272EXPORT_SYMBOL_GPL(__mnt_is_readonly);
273
Al Viro83adc752011-11-24 22:37:54 -0500274static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800275{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000276#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500277 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000278#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500279 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000280#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800281}
Dave Hansen3d733632008-02-15 14:37:59 -0800282
Al Viro83adc752011-11-24 22:37:54 -0500283static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800284{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000285#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500286 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000287#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500288 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000289#endif
290}
291
Al Viro83adc752011-11-24 22:37:54 -0500292static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000293{
294#ifdef CONFIG_SMP
295 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800296 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800297
298 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500299 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800300 }
Dave Hansen3d733632008-02-15 14:37:59 -0800301
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000302 return count;
303#else
304 return mnt->mnt_writers;
305#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800306}
307
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100308static int mnt_is_readonly(struct vfsmount *mnt)
309{
310 if (mnt->mnt_sb->s_readonly_remount)
311 return 1;
312 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
313 smp_rmb();
314 return __mnt_is_readonly(mnt);
315}
316
Dave Hansen3d733632008-02-15 14:37:59 -0800317/*
Jan Karaeb04c282012-06-12 16:20:35 +0200318 * Most r/o & frozen checks on a fs are for operations that take discrete
319 * amounts of time, like a write() or unlink(). We must keep track of when
320 * those operations start (for permission checks) and when they end, so that we
321 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800322 */
Dave Hansen83660252008-02-15 14:37:30 -0800323/**
Jan Karaeb04c282012-06-12 16:20:35 +0200324 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500325 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800326 *
Jan Karaeb04c282012-06-12 16:20:35 +0200327 * This tells the low-level filesystem that a write is about to be performed to
328 * it, and makes sure that writes are allowed (mnt it read-write) before
329 * returning success. This operation does not protect against filesystem being
330 * frozen. When the write operation is finished, __mnt_drop_write() must be
331 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800332 */
Jan Karaeb04c282012-06-12 16:20:35 +0200333int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800334{
Al Viro83adc752011-11-24 22:37:54 -0500335 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800336 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800337
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000338 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100339 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000340 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100341 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000342 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
343 * incremented count after it has set MNT_WRITE_HOLD.
344 */
345 smp_mb();
Miao Xie1e755292012-11-16 17:23:50 +0800346 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000347 cpu_relax();
348 /*
349 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
350 * be set to match its requirements. So we must not load that until
351 * MNT_WRITE_HOLD is cleared.
352 */
353 smp_rmb();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100354 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100355 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800356 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800357 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000358 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200359
360 return ret;
361}
362
363/**
364 * mnt_want_write - get write access to a mount
365 * @m: the mount on which to take a write
366 *
367 * This tells the low-level filesystem that a write is about to be performed to
368 * it, and makes sure that writes are allowed (mount is read-write, filesystem
369 * is not frozen) before returning success. When the write operation is
370 * finished, mnt_drop_write() must be called. This is effectively a refcount.
371 */
372int mnt_want_write(struct vfsmount *m)
373{
374 int ret;
375
376 sb_start_write(m->mnt_sb);
377 ret = __mnt_want_write(m);
378 if (ret)
379 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800380 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800381}
382EXPORT_SYMBOL_GPL(mnt_want_write);
383
384/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000385 * mnt_clone_write - get write access to a mount
386 * @mnt: the mount on which to take a write
387 *
388 * This is effectively like mnt_want_write, except
389 * it must only be used to take an extra write reference
390 * on a mountpoint that we already know has a write reference
391 * on it. This allows some optimisation.
392 *
393 * After finished, mnt_drop_write must be called as usual to
394 * drop the reference.
395 */
396int mnt_clone_write(struct vfsmount *mnt)
397{
398 /* superblock may be r/o */
399 if (__mnt_is_readonly(mnt))
400 return -EROFS;
401 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500402 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000403 preempt_enable();
404 return 0;
405}
406EXPORT_SYMBOL_GPL(mnt_clone_write);
407
408/**
Jan Karaeb04c282012-06-12 16:20:35 +0200409 * __mnt_want_write_file - get write access to a file's mount
410 * @file: the file who's mount on which to take a write
411 *
412 * This is like __mnt_want_write, but it takes a file and can
413 * do some optimisations if the file is open for write already
414 */
415int __mnt_want_write_file(struct file *file)
416{
Al Viro83f936c2014-03-14 12:02:47 -0400417 if (!(file->f_mode & FMODE_WRITER))
Jan Karaeb04c282012-06-12 16:20:35 +0200418 return __mnt_want_write(file->f_path.mnt);
419 else
420 return mnt_clone_write(file->f_path.mnt);
421}
422
423/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000424 * mnt_want_write_file - get write access to a file's mount
425 * @file: the file who's mount on which to take a write
426 *
427 * This is like mnt_want_write, but it takes a file and can
428 * do some optimisations if the file is open for write already
429 */
430int mnt_want_write_file(struct file *file)
431{
Jan Karaeb04c282012-06-12 16:20:35 +0200432 int ret;
433
434 sb_start_write(file->f_path.mnt->mnt_sb);
435 ret = __mnt_want_write_file(file);
436 if (ret)
437 sb_end_write(file->f_path.mnt->mnt_sb);
438 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000439}
440EXPORT_SYMBOL_GPL(mnt_want_write_file);
441
442/**
Jan Karaeb04c282012-06-12 16:20:35 +0200443 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800444 * @mnt: the mount on which to give up write access
445 *
446 * Tells the low-level filesystem that we are done
447 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200448 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800449 */
Jan Karaeb04c282012-06-12 16:20:35 +0200450void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800451{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000452 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500453 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000454 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800455}
Jan Karaeb04c282012-06-12 16:20:35 +0200456
457/**
458 * mnt_drop_write - give up write access to a mount
459 * @mnt: the mount on which to give up write access
460 *
461 * Tells the low-level filesystem that we are done performing writes to it and
462 * also allows filesystem to be frozen again. Must be matched with
463 * mnt_want_write() call above.
464 */
465void mnt_drop_write(struct vfsmount *mnt)
466{
467 __mnt_drop_write(mnt);
468 sb_end_write(mnt->mnt_sb);
469}
Dave Hansen83660252008-02-15 14:37:30 -0800470EXPORT_SYMBOL_GPL(mnt_drop_write);
471
Jan Karaeb04c282012-06-12 16:20:35 +0200472void __mnt_drop_write_file(struct file *file)
473{
474 __mnt_drop_write(file->f_path.mnt);
475}
476
Al Viro2a79f172011-12-09 08:06:57 -0500477void mnt_drop_write_file(struct file *file)
478{
479 mnt_drop_write(file->f_path.mnt);
480}
481EXPORT_SYMBOL(mnt_drop_write_file);
482
Al Viro83adc752011-11-24 22:37:54 -0500483static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800484{
Dave Hansen3d733632008-02-15 14:37:59 -0800485 int ret = 0;
486
Al Viro719ea2f2013-09-29 11:24:49 -0400487 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500488 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000489 /*
490 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
491 * should be visible before we do.
492 */
493 smp_mb();
494
495 /*
496 * With writers on hold, if this value is zero, then there are
497 * definitely no active writers (although held writers may subsequently
498 * increment the count, they'll have to wait, and decrement it after
499 * seeing MNT_READONLY).
500 *
501 * It is OK to have counter incremented on one CPU and decremented on
502 * another: the sum will add up correctly. The danger would be when we
503 * sum up each counter, if we read a counter before it is incremented,
504 * but then read another CPU's count which it has been subsequently
505 * decremented from -- we would see more decrements than we should.
506 * MNT_WRITE_HOLD protects against this scenario, because
507 * mnt_want_write first increments count, then smp_mb, then spins on
508 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
509 * we're counting up here.
510 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100511 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000512 ret = -EBUSY;
513 else
Al Viro83adc752011-11-24 22:37:54 -0500514 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000515 /*
516 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
517 * that become unheld will see MNT_READONLY.
518 */
519 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500520 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Al Viro719ea2f2013-09-29 11:24:49 -0400521 unlock_mount_hash();
Dave Hansen3d733632008-02-15 14:37:59 -0800522 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800523}
Dave Hansen83660252008-02-15 14:37:30 -0800524
Al Viro83adc752011-11-24 22:37:54 -0500525static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800526{
Al Viro719ea2f2013-09-29 11:24:49 -0400527 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500528 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Al Viro719ea2f2013-09-29 11:24:49 -0400529 unlock_mount_hash();
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800530}
531
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100532int sb_prepare_remount_readonly(struct super_block *sb)
533{
534 struct mount *mnt;
535 int err = 0;
536
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100537 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
538 if (atomic_long_read(&sb->s_remove_count))
539 return -EBUSY;
540
Al Viro719ea2f2013-09-29 11:24:49 -0400541 lock_mount_hash();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100542 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
543 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
544 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
545 smp_mb();
546 if (mnt_get_writers(mnt) > 0) {
547 err = -EBUSY;
548 break;
549 }
550 }
551 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100552 if (!err && atomic_long_read(&sb->s_remove_count))
553 err = -EBUSY;
554
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100555 if (!err) {
556 sb->s_readonly_remount = 1;
557 smp_wmb();
558 }
559 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
560 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
561 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
562 }
Al Viro719ea2f2013-09-29 11:24:49 -0400563 unlock_mount_hash();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100564
565 return err;
566}
567
Al Virob105e272011-11-24 20:38:33 -0500568static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
Al Viro52ba1622011-11-25 02:25:17 -0500570 kfree(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000571#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500572 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000573#endif
Al Virob105e272011-11-24 20:38:33 -0500574 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575}
576
David Howells8ffcb322014-01-24 12:17:54 +0000577static void delayed_free_vfsmnt(struct rcu_head *head)
578{
579 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
580}
581
Al Viro48a066e2013-09-29 22:06:07 -0400582/* call under rcu_read_lock */
583bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
584{
585 struct mount *mnt;
586 if (read_seqretry(&mount_lock, seq))
587 return false;
588 if (bastard == NULL)
589 return true;
590 mnt = real_mount(bastard);
591 mnt_add_count(mnt, 1);
592 if (likely(!read_seqretry(&mount_lock, seq)))
593 return true;
594 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
595 mnt_add_count(mnt, -1);
596 return false;
597 }
598 rcu_read_unlock();
599 mntput(bastard);
600 rcu_read_lock();
601 return false;
602}
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604/*
Al Viro474279d2013-10-01 16:11:26 -0400605 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400606 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
Al Viro474279d2013-10-01 16:11:26 -0400608struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Al Viro38129a12014-03-20 21:10:51 -0400610 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400611 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Al Viro38129a12014-03-20 21:10:51 -0400613 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400614 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
615 return p;
616 return NULL;
617}
618
619/*
620 * find the last mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400621 * mount_lock must be held.
Al Viro474279d2013-10-01 16:11:26 -0400622 */
623struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
624{
Al Viro38129a12014-03-20 21:10:51 -0400625 struct mount *p, *res;
626 res = p = __lookup_mnt(mnt, dentry);
627 if (!p)
628 goto out;
629 hlist_for_each_entry_continue(p, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -0400630 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
631 break;
632 res = p;
633 }
Al Viro38129a12014-03-20 21:10:51 -0400634out:
Al Viro1d6a32a2014-03-20 20:34:43 -0400635 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636}
637
Ram Paia05964f2005-11-07 17:20:17 -0500638/*
David Howellsf015f1262012-06-25 12:55:28 +0100639 * lookup_mnt - Return the first child mount mounted at path
640 *
641 * "First" means first mounted chronologically. If you create the
642 * following mounts:
643 *
644 * mount /dev/sda1 /mnt
645 * mount /dev/sda2 /mnt
646 * mount /dev/sda3 /mnt
647 *
648 * Then lookup_mnt() on the base /mnt dentry in the root mount will
649 * return successively the root dentry and vfsmount of /dev/sda1, then
650 * /dev/sda2, then /dev/sda3, then NULL.
651 *
652 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500653 */
Al Viro1c755af2009-04-18 14:06:57 -0400654struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500655{
Al Viroc7105362011-11-24 18:22:03 -0500656 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400657 struct vfsmount *m;
658 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000659
Al Viro48a066e2013-09-29 22:06:07 -0400660 rcu_read_lock();
661 do {
662 seq = read_seqbegin(&mount_lock);
663 child_mnt = __lookup_mnt(path->mnt, path->dentry);
664 m = child_mnt ? &child_mnt->mnt : NULL;
665 } while (!legitimize_mnt(m, seq));
666 rcu_read_unlock();
667 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500668}
669
Eric W. Biederman7af13642013-10-04 19:15:13 -0700670/*
671 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
672 * current mount namespace.
673 *
674 * The common case is dentries are not mountpoints at all and that
675 * test is handled inline. For the slow case when we are actually
676 * dealing with a mountpoint of some kind, walk through all of the
677 * mounts in the current mount namespace and test to see if the dentry
678 * is a mountpoint.
679 *
680 * The mount_hashtable is not usable in the context because we
681 * need to identify all mounts that may be in the current mount
682 * namespace not just a mount that happens to have some specified
683 * parent mount.
684 */
685bool __is_local_mountpoint(struct dentry *dentry)
686{
687 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
688 struct mount *mnt;
689 bool is_covered = false;
690
691 if (!d_mountpoint(dentry))
692 goto out;
693
694 down_read(&namespace_sem);
695 list_for_each_entry(mnt, &ns->list, mnt_list) {
696 is_covered = (mnt->mnt_mountpoint == dentry);
697 if (is_covered)
698 break;
699 }
700 up_read(&namespace_sem);
701out:
702 return is_covered;
703}
704
Al Viro84d17192013-03-15 10:53:28 -0400705static struct mountpoint *new_mountpoint(struct dentry *dentry)
706{
Al Viro0818bf22014-02-28 13:46:44 -0500707 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d17192013-03-15 10:53:28 -0400708 struct mountpoint *mp;
Miklos Szeredieed81002013-09-05 14:39:11 +0200709 int ret;
Al Viro84d17192013-03-15 10:53:28 -0400710
Al Viro0818bf22014-02-28 13:46:44 -0500711 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d17192013-03-15 10:53:28 -0400712 if (mp->m_dentry == dentry) {
713 /* might be worth a WARN_ON() */
714 if (d_unlinked(dentry))
715 return ERR_PTR(-ENOENT);
716 mp->m_count++;
717 return mp;
718 }
719 }
720
721 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
722 if (!mp)
723 return ERR_PTR(-ENOMEM);
724
Miklos Szeredieed81002013-09-05 14:39:11 +0200725 ret = d_set_mounted(dentry);
726 if (ret) {
Al Viro84d17192013-03-15 10:53:28 -0400727 kfree(mp);
Miklos Szeredieed81002013-09-05 14:39:11 +0200728 return ERR_PTR(ret);
Al Viro84d17192013-03-15 10:53:28 -0400729 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200730
Al Viro84d17192013-03-15 10:53:28 -0400731 mp->m_dentry = dentry;
732 mp->m_count = 1;
Al Viro0818bf22014-02-28 13:46:44 -0500733 hlist_add_head(&mp->m_hash, chain);
Al Viro84d17192013-03-15 10:53:28 -0400734 return mp;
735}
736
737static void put_mountpoint(struct mountpoint *mp)
738{
739 if (!--mp->m_count) {
740 struct dentry *dentry = mp->m_dentry;
741 spin_lock(&dentry->d_lock);
742 dentry->d_flags &= ~DCACHE_MOUNTED;
743 spin_unlock(&dentry->d_lock);
Al Viro0818bf22014-02-28 13:46:44 -0500744 hlist_del(&mp->m_hash);
Al Viro84d17192013-03-15 10:53:28 -0400745 kfree(mp);
746 }
747}
748
Al Viro143c8c92011-11-25 00:46:35 -0500749static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800751 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752}
753
Nick Piggin99b7db72010-08-18 04:37:39 +1000754/*
755 * vfsmount lock must be held for write
756 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800757static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500758{
759 if (ns) {
760 ns->event = ++event;
761 wake_up_interruptible(&ns->poll);
762 }
763}
764
Nick Piggin99b7db72010-08-18 04:37:39 +1000765/*
766 * vfsmount lock must be held for write
767 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800768static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500769{
770 if (ns && ns->event != event) {
771 ns->event = event;
772 wake_up_interruptible(&ns->poll);
773 }
774}
775
Nick Piggin99b7db72010-08-18 04:37:39 +1000776/*
777 * vfsmount lock must be held for write
778 */
Al Viro419148d2011-11-24 19:41:16 -0500779static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780{
Al Viroa73324d2011-11-24 22:25:07 -0500781 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500782 old_path->mnt = &mnt->mnt_parent->mnt;
783 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500784 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500785 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400786 hlist_del_init_rcu(&mnt->mnt_hash);
Al Viro84d17192013-03-15 10:53:28 -0400787 put_mountpoint(mnt->mnt_mp);
788 mnt->mnt_mp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789}
790
Nick Piggin99b7db72010-08-18 04:37:39 +1000791/*
792 * vfsmount lock must be held for write
793 */
Al Viro84d17192013-03-15 10:53:28 -0400794void mnt_set_mountpoint(struct mount *mnt,
795 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500796 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500797{
Al Viro84d17192013-03-15 10:53:28 -0400798 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500799 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro84d17192013-03-15 10:53:28 -0400800 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500801 child_mnt->mnt_parent = mnt;
Al Viro84d17192013-03-15 10:53:28 -0400802 child_mnt->mnt_mp = mp;
Ram Paib90fa9a2005-11-07 17:19:50 -0500803}
804
Nick Piggin99b7db72010-08-18 04:37:39 +1000805/*
806 * vfsmount lock must be held for write
807 */
Al Viro84d17192013-03-15 10:53:28 -0400808static void attach_mnt(struct mount *mnt,
809 struct mount *parent,
810 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
Al Viro84d17192013-03-15 10:53:28 -0400812 mnt_set_mountpoint(parent, mp, mnt);
Al Viro38129a12014-03-20 21:10:51 -0400813 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
Al Viro84d17192013-03-15 10:53:28 -0400814 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500815}
816
Al Viro12a5b522014-08-10 03:44:55 -0400817static void attach_shadowed(struct mount *mnt,
818 struct mount *parent,
819 struct mount *shadows)
820{
821 if (shadows) {
Linus Torvaldsf6f99332014-08-11 11:44:11 -0700822 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
Al Viro12a5b522014-08-10 03:44:55 -0400823 list_add(&mnt->mnt_child, &shadows->mnt_child);
824 } else {
825 hlist_add_head_rcu(&mnt->mnt_hash,
826 m_hash(&parent->mnt, mnt->mnt_mountpoint));
827 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
828 }
829}
830
Ram Paib90fa9a2005-11-07 17:19:50 -0500831/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000832 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500833 */
Al Viro1d6a32a2014-03-20 20:34:43 -0400834static void commit_tree(struct mount *mnt, struct mount *shadows)
Ram Paib90fa9a2005-11-07 17:19:50 -0500835{
Al Viro0714a532011-11-24 22:19:58 -0500836 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500837 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500838 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500839 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500840
Al Viro0714a532011-11-24 22:19:58 -0500841 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500842
Al Viro1a4eeaf2011-11-25 02:19:55 -0500843 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400844 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500845 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500846
Ram Paib90fa9a2005-11-07 17:19:50 -0500847 list_splice(&head, n->list.prev);
848
Al Viro12a5b522014-08-10 03:44:55 -0400849 attach_shadowed(mnt, parent, shadows);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800850 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851}
852
Al Viro909b0a82011-11-25 03:06:56 -0500853static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854{
Al Viro6b41d532011-11-24 23:24:33 -0500855 struct list_head *next = p->mnt_mounts.next;
856 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500858 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500860 next = p->mnt_child.next;
861 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 break;
Al Viro0714a532011-11-24 22:19:58 -0500863 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 }
865 }
Al Viro6b41d532011-11-24 23:24:33 -0500866 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867}
868
Al Viro315fc832011-11-24 18:57:30 -0500869static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500870{
Al Viro6b41d532011-11-24 23:24:33 -0500871 struct list_head *prev = p->mnt_mounts.prev;
872 while (prev != &p->mnt_mounts) {
873 p = list_entry(prev, struct mount, mnt_child);
874 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500875 }
876 return p;
877}
878
Al Viro9d412a42011-03-17 22:08:28 -0400879struct vfsmount *
880vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
881{
Al Virob105e272011-11-24 20:38:33 -0500882 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400883 struct dentry *root;
884
885 if (!type)
886 return ERR_PTR(-ENODEV);
887
888 mnt = alloc_vfsmnt(name);
889 if (!mnt)
890 return ERR_PTR(-ENOMEM);
891
892 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500893 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400894
895 root = mount_fs(type, flags, name, data);
896 if (IS_ERR(root)) {
David Howells8ffcb322014-01-24 12:17:54 +0000897 mnt_free_id(mnt);
Al Viro9d412a42011-03-17 22:08:28 -0400898 free_vfsmnt(mnt);
899 return ERR_CAST(root);
900 }
901
Al Virob105e272011-11-24 20:38:33 -0500902 mnt->mnt.mnt_root = root;
903 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500904 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500905 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400906 lock_mount_hash();
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100907 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400908 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -0500909 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400910}
911EXPORT_SYMBOL_GPL(vfs_kern_mount);
912
Al Viro87129cc2011-11-24 21:24:27 -0500913static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500914 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915{
Al Viro87129cc2011-11-24 21:24:27 -0500916 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100917 struct mount *mnt;
918 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
David Howellsbe34d1a2012-06-25 12:55:18 +0100920 mnt = alloc_vfsmnt(old->mnt_devname);
921 if (!mnt)
922 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100923
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700924 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +0100925 mnt->mnt_group_id = 0; /* not a peer of original */
926 else
927 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100928
David Howellsbe34d1a2012-06-25 12:55:18 +0100929 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
930 err = mnt_alloc_group_id(mnt);
931 if (err)
932 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100934
Al Virof2ebb3a2014-02-27 09:35:45 -0500935 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700936 /* Don't allow unprivileged users to change mount flags */
Eric W. Biederman9566d672014-07-28 17:26:07 -0700937 if (flag & CL_UNPRIVILEGED) {
938 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
939
940 if (mnt->mnt.mnt_flags & MNT_READONLY)
941 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
942
943 if (mnt->mnt.mnt_flags & MNT_NODEV)
944 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
945
946 if (mnt->mnt.mnt_flags & MNT_NOSUID)
947 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
948
949 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
950 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
951 }
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700952
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -0700953 /* Don't allow unprivileged users to reveal what is under a mount */
954 if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
955 mnt->mnt.mnt_flags |= MNT_LOCKED;
956
David Howellsbe34d1a2012-06-25 12:55:18 +0100957 atomic_inc(&sb->s_active);
958 mnt->mnt.mnt_sb = sb;
959 mnt->mnt.mnt_root = dget(root);
960 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
961 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400962 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +0100963 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400964 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +0100965
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700966 if ((flag & CL_SLAVE) ||
967 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +0100968 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
969 mnt->mnt_master = old;
970 CLEAR_MNT_SHARED(mnt);
971 } else if (!(flag & CL_PRIVATE)) {
972 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
973 list_add(&mnt->mnt_share, &old->mnt_share);
974 if (IS_MNT_SLAVE(old))
975 list_add(&mnt->mnt_slave, &old->mnt_slave);
976 mnt->mnt_master = old->mnt_master;
977 }
978 if (flag & CL_MAKE_SHARED)
979 set_mnt_shared(mnt);
980
981 /* stick the duplicate mount on the same expiry list
982 * as the original if that was on one */
983 if (flag & CL_EXPIRE) {
984 if (!list_empty(&old->mnt_expire))
985 list_add(&mnt->mnt_expire, &old->mnt_expire);
986 }
987
Al Virocb338d02011-11-24 20:55:08 -0500988 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100989
990 out_free:
David Howells8ffcb322014-01-24 12:17:54 +0000991 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100992 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +0100993 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994}
995
Al Viro9ea459e2014-08-08 13:08:20 -0400996static void cleanup_mnt(struct mount *mnt)
997{
998 /*
999 * This probably indicates that somebody messed
1000 * up a mnt_want/drop_write() pair. If this
1001 * happens, the filesystem was probably unable
1002 * to make r/w->r/o transitions.
1003 */
1004 /*
1005 * The locking used to deal with mnt_count decrement provides barriers,
1006 * so mnt_get_writers() below is safe.
1007 */
1008 WARN_ON(mnt_get_writers(mnt));
1009 if (unlikely(mnt->mnt_pins.first))
1010 mnt_pin_kill(mnt);
1011 fsnotify_vfsmount_delete(&mnt->mnt);
1012 dput(mnt->mnt.mnt_root);
1013 deactivate_super(mnt->mnt.mnt_sb);
1014 mnt_free_id(mnt);
1015 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1016}
1017
1018static void __cleanup_mnt(struct rcu_head *head)
1019{
1020 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1021}
1022
1023static LLIST_HEAD(delayed_mntput_list);
1024static void delayed_mntput(struct work_struct *unused)
1025{
1026 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1027 struct llist_node *next;
1028
1029 for (; node; node = next) {
1030 next = llist_next(node);
1031 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1032 }
1033}
1034static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1035
Al Viro900148d2011-11-25 00:33:11 -05001036static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001037{
Al Viro48a066e2013-09-29 22:06:07 -04001038 rcu_read_lock();
1039 mnt_add_count(mnt, -1);
1040 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1041 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -05001042 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001043 }
Al Viro719ea2f2013-09-29 11:24:49 -04001044 lock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001045 if (mnt_get_count(mnt)) {
Al Viro48a066e2013-09-29 22:06:07 -04001046 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -04001047 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001048 return;
1049 }
Al Viro48a066e2013-09-29 22:06:07 -04001050 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1051 rcu_read_unlock();
1052 unlock_mount_hash();
1053 return;
1054 }
1055 mnt->mnt.mnt_flags |= MNT_DOOMED;
1056 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301057
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001058 list_del(&mnt->mnt_instance);
Al Viro719ea2f2013-09-29 11:24:49 -04001059 unlock_mount_hash();
Al Viro649a7952013-09-28 12:41:25 -04001060
Al Viro9ea459e2014-08-08 13:08:20 -04001061 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1062 struct task_struct *task = current;
1063 if (likely(!(task->flags & PF_KTHREAD))) {
1064 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1065 if (!task_work_add(task, &mnt->mnt_rcu, true))
1066 return;
1067 }
1068 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1069 schedule_delayed_work(&delayed_mntput_work, 1);
1070 return;
1071 }
1072 cleanup_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001073}
Nick Pigginb3e19d92011-01-07 17:50:11 +11001074
1075void mntput(struct vfsmount *mnt)
1076{
1077 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -05001078 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001079 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -05001080 if (unlikely(m->mnt_expiry_mark))
1081 m->mnt_expiry_mark = 0;
1082 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001083 }
1084}
1085EXPORT_SYMBOL(mntput);
1086
1087struct vfsmount *mntget(struct vfsmount *mnt)
1088{
1089 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001090 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001091 return mnt;
1092}
1093EXPORT_SYMBOL(mntget);
1094
Al Viro3064c352014-08-07 09:12:31 -04001095struct vfsmount *mnt_clone_internal(struct path *path)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001096{
Al Viro3064c352014-08-07 09:12:31 -04001097 struct mount *p;
1098 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1099 if (IS_ERR(p))
1100 return ERR_CAST(p);
1101 p->mnt.mnt_flags |= MNT_INTERNAL;
1102 return &p->mnt;
Al Viro7b7b1ac2005-11-07 17:13:39 -05001103}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001105static inline void mangle(struct seq_file *m, const char *s)
1106{
1107 seq_escape(m, s, " \t\n\\");
1108}
1109
1110/*
1111 * Simple .show_options callback for filesystems which don't want to
1112 * implement more complex mount option showing.
1113 *
1114 * See also save_mount_options().
1115 */
Al Viro34c80b12011-12-08 21:32:45 -05001116int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001117{
Al Viro2a32ceb2009-05-08 16:05:57 -04001118 const char *options;
1119
1120 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -05001121 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001122
1123 if (options != NULL && options[0]) {
1124 seq_putc(m, ',');
1125 mangle(m, options);
1126 }
Al Viro2a32ceb2009-05-08 16:05:57 -04001127 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001128
1129 return 0;
1130}
1131EXPORT_SYMBOL(generic_show_options);
1132
1133/*
1134 * If filesystem uses generic_show_options(), this function should be
1135 * called from the fill_super() callback.
1136 *
1137 * The .remount_fs callback usually needs to be handled in a special
1138 * way, to make sure, that previous options are not overwritten if the
1139 * remount fails.
1140 *
1141 * Also note, that if the filesystem's .remount_fs function doesn't
1142 * reset all options to their default value, but changes only newly
1143 * given options, then the displayed options will not reflect reality
1144 * any more.
1145 */
1146void save_mount_options(struct super_block *sb, char *options)
1147{
Al Viro2a32ceb2009-05-08 16:05:57 -04001148 BUG_ON(sb->s_options);
1149 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001150}
1151EXPORT_SYMBOL(save_mount_options);
1152
Al Viro2a32ceb2009-05-08 16:05:57 -04001153void replace_mount_options(struct super_block *sb, char *options)
1154{
1155 char *old = sb->s_options;
1156 rcu_assign_pointer(sb->s_options, options);
1157 if (old) {
1158 synchronize_rcu();
1159 kfree(old);
1160 }
1161}
1162EXPORT_SYMBOL(replace_mount_options);
1163
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001164#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001165/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166static void *m_start(struct seq_file *m, loff_t *pos)
1167{
Al Viro6ce6e242012-06-09 01:16:59 -04001168 struct proc_mounts *p = proc_mounts(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
Ram Pai390c6842005-11-07 17:17:51 -05001170 down_read(&namespace_sem);
Al Viroc7999c32014-02-27 14:40:10 -05001171 if (p->cached_event == p->ns->event) {
1172 void *v = p->cached_mount;
1173 if (*pos == p->cached_index)
1174 return v;
1175 if (*pos == p->cached_index + 1) {
1176 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1177 return p->cached_mount = v;
1178 }
1179 }
1180
1181 p->cached_event = p->ns->event;
1182 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1183 p->cached_index = *pos;
1184 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185}
1186
1187static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1188{
Al Viro6ce6e242012-06-09 01:16:59 -04001189 struct proc_mounts *p = proc_mounts(m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001190
Al Viroc7999c32014-02-27 14:40:10 -05001191 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1192 p->cached_index = *pos;
1193 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194}
1195
1196static void m_stop(struct seq_file *m, void *v)
1197{
Ram Pai390c6842005-11-07 17:17:51 -05001198 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199}
1200
Al Viro0226f492011-12-06 12:21:54 -05001201static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001202{
Al Viro6ce6e242012-06-09 01:16:59 -04001203 struct proc_mounts *p = proc_mounts(m);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001204 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001205 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206}
1207
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001208const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 .start = m_start,
1210 .next = m_next,
1211 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001212 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001213};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001214#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216/**
1217 * may_umount_tree - check if a mount tree is busy
1218 * @mnt: root of mount tree
1219 *
1220 * This is called to check if a tree of mounts has any
1221 * open files, pwds, chroots or sub mounts that are
1222 * busy.
1223 */
Al Viro909b0a82011-11-25 03:06:56 -05001224int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225{
Al Viro909b0a82011-11-25 03:06:56 -05001226 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001227 int actual_refs = 0;
1228 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001229 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001230 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
Nick Pigginb3e19d92011-01-07 17:50:11 +11001232 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001233 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001234 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001235 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 }
Al Viro719ea2f2013-09-29 11:24:49 -04001238 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
1240 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001241 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
Ian Kente3474a82006-03-27 01:14:51 -08001243 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244}
1245
1246EXPORT_SYMBOL(may_umount_tree);
1247
1248/**
1249 * may_umount - check if a mount point is busy
1250 * @mnt: root of mount
1251 *
1252 * This is called to check if a mount point has any
1253 * open files, pwds, chroots or sub mounts. If the
1254 * mount has sub mounts this will return busy
1255 * regardless of whether the sub mounts are busy.
1256 *
1257 * Doesn't take quota and stuff into account. IOW, in some cases it will
1258 * give false negatives. The main reason why it's here is that we need
1259 * a non-destructive way to look for easily umountable filesystems.
1260 */
1261int may_umount(struct vfsmount *mnt)
1262{
Ian Kente3474a82006-03-27 01:14:51 -08001263 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001264 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001265 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001266 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001267 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001268 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001269 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001270 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271}
1272
1273EXPORT_SYMBOL(may_umount);
1274
Al Viro38129a12014-03-20 21:10:51 -04001275static HLIST_HEAD(unmounted); /* protected by namespace_sem */
Al Viroe3197d82013-03-16 14:35:16 -04001276
Al Viro97216be2013-03-16 15:12:40 -04001277static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278{
Al Virod5e50f72011-11-24 20:58:57 -05001279 struct mount *mnt;
Al Viro38129a12014-03-20 21:10:51 -04001280 struct hlist_head head = unmounted;
Al Viro97216be2013-03-16 15:12:40 -04001281
Al Viro38129a12014-03-20 21:10:51 -04001282 if (likely(hlist_empty(&head))) {
Al Viro97216be2013-03-16 15:12:40 -04001283 up_write(&namespace_sem);
1284 return;
1285 }
1286
Al Viro38129a12014-03-20 21:10:51 -04001287 head.first->pprev = &head.first;
1288 INIT_HLIST_HEAD(&unmounted);
1289
Al Viro81b6b062014-08-30 18:32:05 -04001290 /* undo decrements we'd done in umount_tree() */
1291 hlist_for_each_entry(mnt, &head, mnt_hash)
1292 if (mnt->mnt_ex_mountpoint.mnt)
1293 mntget(mnt->mnt_ex_mountpoint.mnt);
1294
Al Viro97216be2013-03-16 15:12:40 -04001295 up_write(&namespace_sem);
1296
Al Viro48a066e2013-09-29 22:06:07 -04001297 synchronize_rcu();
1298
Al Viro38129a12014-03-20 21:10:51 -04001299 while (!hlist_empty(&head)) {
1300 mnt = hlist_entry(head.first, struct mount, mnt_hash);
1301 hlist_del_init(&mnt->mnt_hash);
Al Viroaba809cf2013-09-28 23:10:55 -04001302 if (mnt->mnt_ex_mountpoint.mnt)
1303 path_put(&mnt->mnt_ex_mountpoint);
Al Virod5e50f72011-11-24 20:58:57 -05001304 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001305 }
1306}
1307
Al Viro97216be2013-03-16 15:12:40 -04001308static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001309{
Al Viro97216be2013-03-16 15:12:40 -04001310 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001311}
1312
Nick Piggin99b7db72010-08-18 04:37:39 +10001313/*
Al Viro48a066e2013-09-29 22:06:07 -04001314 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001315 * namespace_sem must be held for write
Al Viro48a066e2013-09-29 22:06:07 -04001316 * how = 0 => just this tree, don't propagate
1317 * how = 1 => propagate; we know that nobody else has reference to any victims
1318 * how = 2 => lazy umount
Nick Piggin99b7db72010-08-18 04:37:39 +10001319 */
Al Viro48a066e2013-09-29 22:06:07 -04001320void umount_tree(struct mount *mnt, int how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001321{
Al Viro38129a12014-03-20 21:10:51 -04001322 HLIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001323 struct mount *p;
Al Viro38129a12014-03-20 21:10:51 -04001324 struct mount *last = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001325
Al Viro38129a12014-03-20 21:10:51 -04001326 for (p = mnt; p; p = next_mnt(p, mnt)) {
1327 hlist_del_init_rcu(&p->mnt_hash);
1328 hlist_add_head(&p->mnt_hash, &tmp_list);
1329 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001330
Al Viro88b368f2014-08-18 15:09:26 -04001331 hlist_for_each_entry(p, &tmp_list, mnt_hash)
1332 list_del_init(&p->mnt_child);
1333
Al Viro48a066e2013-09-29 22:06:07 -04001334 if (how)
Al Viro7b8a53f2011-01-15 20:08:44 -05001335 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001336
Al Viro38129a12014-03-20 21:10:51 -04001337 hlist_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001338 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001339 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001340 __touch_mnt_namespace(p->mnt_ns);
1341 p->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04001342 if (how < 2)
1343 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro676da582011-11-24 21:47:05 -05001344 if (mnt_has_parent(p)) {
Al Viro84d17192013-03-15 10:53:28 -04001345 put_mountpoint(p->mnt_mp);
Al Viro81b6b062014-08-30 18:32:05 -04001346 mnt_add_count(p->mnt_parent, -1);
Al Viroaba809cf2013-09-28 23:10:55 -04001347 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1348 p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint;
1349 p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt;
1350 p->mnt_mountpoint = p->mnt.mnt_root;
1351 p->mnt_parent = p;
Al Viro84d17192013-03-15 10:53:28 -04001352 p->mnt_mp = NULL;
Al Viro7c4b93d2008-03-21 23:59:49 -04001353 }
Al Viro0f0afb12011-11-24 20:43:10 -05001354 change_mnt_propagation(p, MS_PRIVATE);
Al Viro38129a12014-03-20 21:10:51 -04001355 last = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
Al Viro38129a12014-03-20 21:10:51 -04001357 if (last) {
1358 last->mnt_hash.next = unmounted.first;
1359 unmounted.first = tmp_list.first;
1360 unmounted.first->pprev = &unmounted.first;
1361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362}
1363
Al Virob54b9be2013-03-16 14:39:34 -04001364static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001365
Al Viro1ab59732011-11-24 21:35:16 -05001366static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367{
Al Viro1ab59732011-11-24 21:35:16 -05001368 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 int retval;
1370
Al Viro1ab59732011-11-24 21:35:16 -05001371 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 if (retval)
1373 return retval;
1374
1375 /*
1376 * Allow userspace to request a mountpoint be expired rather than
1377 * unmounting unconditionally. Unmount only happens if:
1378 * (1) the mark is already set (the mark is cleared by mntput())
1379 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1380 */
1381 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001382 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 flags & (MNT_FORCE | MNT_DETACH))
1384 return -EINVAL;
1385
Nick Pigginb3e19d92011-01-07 17:50:11 +11001386 /*
1387 * probably don't strictly need the lock here if we examined
1388 * all race cases, but it's a slowpath.
1389 */
Al Viro719ea2f2013-09-29 11:24:49 -04001390 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001391 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001392 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001394 }
Al Viro719ea2f2013-09-29 11:24:49 -04001395 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
Al Viro863d6842011-11-25 00:57:42 -05001397 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 return -EAGAIN;
1399 }
1400
1401 /*
1402 * If we may have to abort operations to get out of this
1403 * mount, and they will themselves hold resources we must
1404 * allow the fs to do things. In the Unix tradition of
1405 * 'Gee thats tricky lets do it in userspace' the umount_begin
1406 * might fail to complete on the first run through as other tasks
1407 * must return, and the like. Thats for the mount program to worry
1408 * about for the moment.
1409 */
1410
Al Viro42faad92008-04-24 07:21:56 -04001411 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001412 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001413 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
1415 /*
1416 * No sense to grab the lock for this test, but test itself looks
1417 * somewhat bogus. Suggestions for better replacement?
1418 * Ho-hum... In principle, we might treat that as umount + switch
1419 * to rootfs. GC would eventually take care of the old vfsmount.
1420 * Actually it makes sense, especially if rootfs would contain a
1421 * /reboot - static binary that would close all descriptors and
1422 * call reboot(9). Then init(8) could umount root and exec /reboot.
1423 */
Al Viro1ab59732011-11-24 21:35:16 -05001424 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 /*
1426 * Special case for "unmounting" root ...
1427 * we just try to remount it readonly.
1428 */
1429 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001430 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 up_write(&sb->s_umount);
1433 return retval;
1434 }
1435
Al Viro97216be2013-03-16 15:12:40 -04001436 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001437 lock_mount_hash();
Al Viro5addc5d2005-11-07 17:15:49 -05001438 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Al Viro48a066e2013-09-29 22:06:07 -04001440 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001441 if (!list_empty(&mnt->mnt_list))
Al Viro48a066e2013-09-29 22:06:07 -04001442 umount_tree(mnt, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001444 } else {
1445 shrink_submounts(mnt);
1446 retval = -EBUSY;
1447 if (!propagate_mount_busy(mnt, 2)) {
1448 if (!list_empty(&mnt->mnt_list))
1449 umount_tree(mnt, 1);
1450 retval = 0;
1451 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 }
Al Viro719ea2f2013-09-29 11:24:49 -04001453 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001454 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 return retval;
1456}
1457
Al Viro9b40bc92013-02-22 22:45:42 -05001458/*
1459 * Is the caller allowed to modify his namespace?
1460 */
1461static inline bool may_mount(void)
1462{
1463 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1464}
1465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466/*
1467 * Now umount can handle mount points as well as block devices.
1468 * This is important for filesystems which use unnamed block devices.
1469 *
1470 * We now support a flag for forced unmount like the other 'big iron'
1471 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1472 */
1473
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001474SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475{
Al Viro2d8f3032008-07-22 09:59:21 -04001476 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001477 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001479 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001481 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1482 return -EINVAL;
1483
Al Viro9b40bc92013-02-22 22:45:42 -05001484 if (!may_mount())
1485 return -EPERM;
1486
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001487 if (!(flags & UMOUNT_NOFOLLOW))
1488 lookup_flags |= LOOKUP_FOLLOW;
1489
Al Viro197df042013-09-08 14:03:27 -04001490 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 if (retval)
1492 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001493 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001495 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001497 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 goto dput_and_out;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001499 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1500 goto dput_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
Al Viro900148d2011-11-25 00:33:11 -05001502 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001504 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001505 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001506 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507out:
1508 return retval;
1509}
1510
1511#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1512
1513/*
Ram Paib58fed82005-11-07 17:16:09 -05001514 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001516SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517{
Ram Paib58fed82005-11-07 17:16:09 -05001518 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519}
1520
1521#endif
1522
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001523static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001524{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001525 /* Is this a proxy for a mount namespace? */
1526 struct inode *inode = dentry->d_inode;
David Howells0bb80f22013-04-12 01:50:06 +01001527 struct proc_ns *ei;
Eric W. Biederman8823c072010-03-07 18:49:36 -08001528
1529 if (!proc_ns_inode(inode))
1530 return false;
1531
David Howells0bb80f22013-04-12 01:50:06 +01001532 ei = get_proc_ns(inode);
Eric W. Biederman8823c072010-03-07 18:49:36 -08001533 if (ei->ns_ops != &mntns_operations)
1534 return false;
1535
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001536 return true;
1537}
1538
1539static bool mnt_ns_loop(struct dentry *dentry)
1540{
1541 /* Could bind mounting the mount namespace inode cause a
1542 * mount namespace loop?
1543 */
1544 struct mnt_namespace *mnt_ns;
1545 if (!is_mnt_ns_file(dentry))
1546 return false;
1547
1548 mnt_ns = get_proc_ns(dentry->d_inode)->ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08001549 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1550}
1551
Al Viro87129cc2011-11-24 21:24:27 -05001552struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001553 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554{
Al Viro84d17192013-03-15 10:53:28 -04001555 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001557 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1558 return ERR_PTR(-EINVAL);
1559
1560 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001561 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001562
Ram Pai36341f62005-11-07 17:17:22 -05001563 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001564 if (IS_ERR(q))
1565 return q;
1566
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001567 q->mnt.mnt_flags &= ~MNT_LOCKED;
Al Viroa73324d2011-11-24 22:25:07 -05001568 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
1570 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001571 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001572 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001573 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 continue;
1575
Al Viro909b0a82011-11-25 03:06:56 -05001576 for (s = r; s; s = next_mnt(s, r)) {
Al Viro12a5b522014-08-10 03:44:55 -04001577 struct mount *t = NULL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001578 if (!(flag & CL_COPY_UNBINDABLE) &&
1579 IS_MNT_UNBINDABLE(s)) {
1580 s = skip_mnt_tree(s);
1581 continue;
1582 }
1583 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1584 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001585 s = skip_mnt_tree(s);
1586 continue;
1587 }
Al Viro0714a532011-11-24 22:19:58 -05001588 while (p != s->mnt_parent) {
1589 p = p->mnt_parent;
1590 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 }
Al Viro87129cc2011-11-24 21:24:27 -05001592 p = s;
Al Viro84d17192013-03-15 10:53:28 -04001593 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001594 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001595 if (IS_ERR(q))
1596 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04001597 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05001598 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro12a5b522014-08-10 03:44:55 -04001599 mnt_set_mountpoint(parent, p->mnt_mp, q);
1600 if (!list_empty(&parent->mnt_mounts)) {
1601 t = list_last_entry(&parent->mnt_mounts,
1602 struct mount, mnt_child);
1603 if (t->mnt_mp != p->mnt_mp)
1604 t = NULL;
1605 }
1606 attach_shadowed(q, parent, t);
Al Viro719ea2f2013-09-29 11:24:49 -04001607 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 }
1609 }
1610 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001611out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04001613 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001614 umount_tree(res, 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001615 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001617 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618}
1619
David Howellsbe34d1a2012-06-25 12:55:18 +01001620/* Caller should check returned pointer for errors */
1621
Al Viro589ff872009-04-18 03:28:19 -04001622struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001623{
Al Virocb338d02011-11-24 20:55:08 -05001624 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001625 namespace_lock();
Al Viro87129cc2011-11-24 21:24:27 -05001626 tree = copy_tree(real_mount(path->mnt), path->dentry,
1627 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001628 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001629 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001630 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001631 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001632}
1633
1634void drop_collected_mounts(struct vfsmount *mnt)
1635{
Al Viro97216be2013-03-16 15:12:40 -04001636 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001637 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001638 umount_tree(real_mount(mnt), 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001639 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04001640 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001641}
1642
Al Viro1f707132010-01-30 22:51:25 -05001643int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1644 struct vfsmount *root)
1645{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001646 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001647 int res = f(root, arg);
1648 if (res)
1649 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001650 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1651 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001652 if (res)
1653 return res;
1654 }
1655 return 0;
1656}
1657
Al Viro4b8b21f2011-11-24 19:54:23 -05001658static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001659{
Al Viro315fc832011-11-24 18:57:30 -05001660 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001661
Al Viro909b0a82011-11-25 03:06:56 -05001662 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001663 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001664 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001665 }
1666}
1667
Al Viro4b8b21f2011-11-24 19:54:23 -05001668static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001669{
Al Viro315fc832011-11-24 18:57:30 -05001670 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001671
Al Viro909b0a82011-11-25 03:06:56 -05001672 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001673 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001674 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001675 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001676 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001677 return err;
1678 }
1679 }
1680 }
1681
1682 return 0;
1683}
1684
Ram Paib90fa9a2005-11-07 17:19:50 -05001685/*
1686 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001687 * @nd : place the mount tree @source_mnt is attached
1688 * @parent_nd : if non-null, detach the source_mnt from its parent and
1689 * store the parent mount and mountpoint dentry.
1690 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001691 *
1692 * NOTE: in the table below explains the semantics when a source mount
1693 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001694 * ---------------------------------------------------------------------------
1695 * | BIND MOUNT OPERATION |
1696 * |**************************************************************************
1697 * | source-->| shared | private | slave | unbindable |
1698 * | dest | | | | |
1699 * | | | | | | |
1700 * | v | | | | |
1701 * |**************************************************************************
1702 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1703 * | | | | | |
1704 * |non-shared| shared (+) | private | slave (*) | invalid |
1705 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001706 * A bind operation clones the source mount and mounts the clone on the
1707 * destination mount.
1708 *
1709 * (++) the cloned mount is propagated to all the mounts in the propagation
1710 * tree of the destination mount and the cloned mount is added to
1711 * the peer group of the source mount.
1712 * (+) the cloned mount is created under the destination mount and is marked
1713 * as shared. The cloned mount is added to the peer group of the source
1714 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001715 * (+++) the mount is propagated to all the mounts in the propagation tree
1716 * of the destination mount and the cloned mount is made slave
1717 * of the same master as that of the source mount. The cloned mount
1718 * is marked as 'shared and slave'.
1719 * (*) the cloned mount is made a slave of the same master as that of the
1720 * source mount.
1721 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001722 * ---------------------------------------------------------------------------
1723 * | MOVE MOUNT OPERATION |
1724 * |**************************************************************************
1725 * | source-->| shared | private | slave | unbindable |
1726 * | dest | | | | |
1727 * | | | | | | |
1728 * | v | | | | |
1729 * |**************************************************************************
1730 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1731 * | | | | | |
1732 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1733 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001734 *
1735 * (+) the mount is moved to the destination. And is then propagated to
1736 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001737 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001738 * (+++) the mount is moved to the destination and is then propagated to
1739 * all the mounts belonging to the destination mount's propagation tree.
1740 * the mount is marked as 'shared and slave'.
1741 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001742 *
1743 * if the source mount is a tree, the operations explained above is
1744 * applied to each mount in the tree.
1745 * Must be called without spinlocks held, since this function can sleep
1746 * in allocations.
1747 */
Al Viro0fb54e52011-11-24 19:59:16 -05001748static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d17192013-03-15 10:53:28 -04001749 struct mount *dest_mnt,
1750 struct mountpoint *dest_mp,
1751 struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001752{
Al Viro38129a12014-03-20 21:10:51 -04001753 HLIST_HEAD(tree_list);
Al Viro315fc832011-11-24 18:57:30 -05001754 struct mount *child, *p;
Al Viro38129a12014-03-20 21:10:51 -04001755 struct hlist_node *n;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001756 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001757
Al Virofc7be132011-11-25 01:05:37 -05001758 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001759 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001760 if (err)
1761 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04001762 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Al Virof2ebb3a2014-02-27 09:35:45 -05001763 lock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001764 if (err)
1765 goto out_cleanup_ids;
Al Viro909b0a82011-11-25 03:06:56 -05001766 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001767 set_mnt_shared(p);
Al Viro0b1b9012014-03-21 10:14:08 -04001768 } else {
1769 lock_mount_hash();
Ram Paib90fa9a2005-11-07 17:19:50 -05001770 }
Al Viro1a390682008-03-21 20:48:19 -04001771 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001772 detach_mnt(source_mnt, parent_path);
Al Viro84d17192013-03-15 10:53:28 -04001773 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05001774 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001775 } else {
Al Viro84d17192013-03-15 10:53:28 -04001776 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Al Viro1d6a32a2014-03-20 20:34:43 -04001777 commit_tree(source_mnt, NULL);
Ram Pai21444402005-11-07 17:20:03 -05001778 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001779
Al Viro38129a12014-03-20 21:10:51 -04001780 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04001781 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04001782 hlist_del_init(&child->mnt_hash);
Al Viro1d6a32a2014-03-20 20:34:43 -04001783 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1784 child->mnt_mountpoint);
1785 commit_tree(child, q);
Ram Paib90fa9a2005-11-07 17:19:50 -05001786 }
Al Viro719ea2f2013-09-29 11:24:49 -04001787 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10001788
Ram Paib90fa9a2005-11-07 17:19:50 -05001789 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001790
1791 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05001792 while (!hlist_empty(&tree_list)) {
1793 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1794 umount_tree(child, 0);
1795 }
1796 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001797 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001798 out:
1799 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001800}
1801
Al Viro84d17192013-03-15 10:53:28 -04001802static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04001803{
1804 struct vfsmount *mnt;
Al Viro84d17192013-03-15 10:53:28 -04001805 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04001806retry:
Al Viro84d17192013-03-15 10:53:28 -04001807 mutex_lock(&dentry->d_inode->i_mutex);
1808 if (unlikely(cant_mount(dentry))) {
1809 mutex_unlock(&dentry->d_inode->i_mutex);
1810 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04001811 }
Al Viro97216be2013-03-16 15:12:40 -04001812 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04001813 mnt = lookup_mnt(path);
Al Viro84d17192013-03-15 10:53:28 -04001814 if (likely(!mnt)) {
1815 struct mountpoint *mp = new_mountpoint(dentry);
1816 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04001817 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001818 mutex_unlock(&dentry->d_inode->i_mutex);
1819 return mp;
1820 }
1821 return mp;
1822 }
Al Viro97216be2013-03-16 15:12:40 -04001823 namespace_unlock();
Al Virob12cea92011-03-18 08:55:38 -04001824 mutex_unlock(&path->dentry->d_inode->i_mutex);
1825 path_put(path);
1826 path->mnt = mnt;
Al Viro84d17192013-03-15 10:53:28 -04001827 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04001828 goto retry;
1829}
1830
Al Viro84d17192013-03-15 10:53:28 -04001831static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04001832{
Al Viro84d17192013-03-15 10:53:28 -04001833 struct dentry *dentry = where->m_dentry;
1834 put_mountpoint(where);
Al Viro328e6d92013-03-16 14:49:45 -04001835 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001836 mutex_unlock(&dentry->d_inode->i_mutex);
Al Virob12cea92011-03-18 08:55:38 -04001837}
1838
Al Viro84d17192013-03-15 10:53:28 -04001839static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840{
Al Viro95bc5f22011-11-25 00:30:56 -05001841 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 return -EINVAL;
1843
Al Viro84d17192013-03-15 10:53:28 -04001844 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001845 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 return -ENOTDIR;
1847
Al Viro84d17192013-03-15 10:53:28 -04001848 return attach_recursive_mnt(mnt, p, mp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849}
1850
1851/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001852 * Sanity check the flags to change_mnt_propagation.
1853 */
1854
1855static int flags_to_propagation_type(int flags)
1856{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001857 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001858
1859 /* Fail if any non-propagation flags are set */
1860 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1861 return 0;
1862 /* Only one propagation flag should be set */
1863 if (!is_power_of_2(type))
1864 return 0;
1865 return type;
1866}
1867
1868/*
Ram Pai07b20882005-11-07 17:19:07 -05001869 * recursively change the type of the mountpoint.
1870 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001871static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001872{
Al Viro315fc832011-11-24 18:57:30 -05001873 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001874 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001875 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001876 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001877 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001878
Al Viro2d92ab32008-08-02 00:51:11 -04001879 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001880 return -EINVAL;
1881
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001882 type = flags_to_propagation_type(flag);
1883 if (!type)
1884 return -EINVAL;
1885
Al Viro97216be2013-03-16 15:12:40 -04001886 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001887 if (type == MS_SHARED) {
1888 err = invent_group_ids(mnt, recurse);
1889 if (err)
1890 goto out_unlock;
1891 }
1892
Al Viro719ea2f2013-09-29 11:24:49 -04001893 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001894 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001895 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04001896 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001897
1898 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04001899 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001900 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001901}
1902
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001903static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1904{
1905 struct mount *child;
1906 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1907 if (!is_subdir(child->mnt_mountpoint, dentry))
1908 continue;
1909
1910 if (child->mnt.mnt_flags & MNT_LOCKED)
1911 return true;
1912 }
1913 return false;
1914}
1915
Ram Pai07b20882005-11-07 17:19:07 -05001916/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 * do loopback mount.
1918 */
Al Viro808d4e32012-10-11 11:42:01 -04001919static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001920 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921{
Al Viro2d92ab32008-08-02 00:51:11 -04001922 struct path old_path;
Al Viro84d17192013-03-15 10:53:28 -04001923 struct mount *mnt = NULL, *old, *parent;
1924 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05001925 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 if (!old_name || !*old_name)
1927 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001928 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 if (err)
1930 return err;
1931
Eric W. Biederman8823c072010-03-07 18:49:36 -08001932 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001933 if (mnt_ns_loop(old_path.dentry))
Eric W. Biederman8823c072010-03-07 18:49:36 -08001934 goto out;
1935
Al Viro84d17192013-03-15 10:53:28 -04001936 mp = lock_mount(path);
1937 err = PTR_ERR(mp);
1938 if (IS_ERR(mp))
Jan Blunck4ac91372008-02-14 19:34:32 -08001939 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001940
Al Viro87129cc2011-11-24 21:24:27 -05001941 old = real_mount(old_path.mnt);
Al Viro84d17192013-03-15 10:53:28 -04001942 parent = real_mount(path->mnt);
Al Viro87129cc2011-11-24 21:24:27 -05001943
Al Virob12cea92011-03-18 08:55:38 -04001944 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001945 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04001946 goto out2;
1947
Al Viro84d17192013-03-15 10:53:28 -04001948 if (!check_mnt(parent) || !check_mnt(old))
Al Virob12cea92011-03-18 08:55:38 -04001949 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001951 if (!recurse && has_locked_children(old, old_path.dentry))
1952 goto out2;
1953
Al Viroccd48bc2005-11-07 17:15:04 -05001954 if (recurse)
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001955 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
Al Viroccd48bc2005-11-07 17:15:04 -05001956 else
Al Viro87129cc2011-11-24 21:24:27 -05001957 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001958
David Howellsbe34d1a2012-06-25 12:55:18 +01001959 if (IS_ERR(mnt)) {
1960 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04001961 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01001962 }
Al Viroccd48bc2005-11-07 17:15:04 -05001963
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001964 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
1965
Al Viro84d17192013-03-15 10:53:28 -04001966 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05001967 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04001968 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001969 umount_tree(mnt, 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001970 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001971 }
Al Virob12cea92011-03-18 08:55:38 -04001972out2:
Al Viro84d17192013-03-15 10:53:28 -04001973 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05001974out:
Al Viro2d92ab32008-08-02 00:51:11 -04001975 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 return err;
1977}
1978
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001979static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1980{
1981 int error = 0;
1982 int readonly_request = 0;
1983
1984 if (ms_flags & MS_RDONLY)
1985 readonly_request = 1;
1986 if (readonly_request == __mnt_is_readonly(mnt))
1987 return 0;
1988
1989 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05001990 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001991 else
Al Viro83adc752011-11-24 22:37:54 -05001992 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001993 return error;
1994}
1995
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996/*
1997 * change filesystem flags. dir should be a physical root of filesystem.
1998 * If you've mounted a non-root directory somewhere and want to do remount
1999 * on it - tough luck.
2000 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002001static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 void *data)
2003{
2004 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04002005 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05002006 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007
Al Viro143c8c92011-11-25 00:46:35 -05002008 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 return -EINVAL;
2010
Al Viro2d92ab32008-08-02 00:51:11 -04002011 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 return -EINVAL;
2013
Eric W. Biederman07b64552014-07-28 17:10:56 -07002014 /* Don't allow changing of locked mnt flags.
2015 *
2016 * No locks need to be held here while testing the various
2017 * MNT_LOCK flags because those flags can never be cleared
2018 * once they are set.
2019 */
2020 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2021 !(mnt_flags & MNT_READONLY)) {
2022 return -EPERM;
2023 }
Eric W. Biederman9566d672014-07-28 17:26:07 -07002024 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2025 !(mnt_flags & MNT_NODEV)) {
2026 return -EPERM;
2027 }
2028 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2029 !(mnt_flags & MNT_NOSUID)) {
2030 return -EPERM;
2031 }
2032 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2033 !(mnt_flags & MNT_NOEXEC)) {
2034 return -EPERM;
2035 }
2036 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2037 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2038 return -EPERM;
2039 }
2040
Eric Parisff36fe22011-03-03 16:09:14 -05002041 err = security_sb_remount(sb, data);
2042 if (err)
2043 return err;
2044
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002046 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002047 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05002048 else if (!capable(CAP_SYS_ADMIN))
2049 err = -EPERM;
Al Viro4aa98cf2009-05-08 13:36:58 -04002050 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002051 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05002052 if (!err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002053 lock_mount_hash();
Eric W. Biedermana6138db2014-07-28 16:26:53 -07002054 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
Al Viro143c8c92011-11-25 00:46:35 -05002055 mnt->mnt.mnt_flags = mnt_flags;
Al Viro143c8c92011-11-25 00:46:35 -05002056 touch_mnt_namespace(mnt->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04002057 unlock_mount_hash();
Dan Williams0e55a7c2008-09-26 19:01:20 -07002058 }
Al Viro6339dab2013-09-16 22:41:01 -04002059 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 return err;
2061}
2062
Al Virocbbe3622011-11-24 20:01:19 -05002063static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05002064{
Al Viro315fc832011-11-24 18:57:30 -05002065 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05002066 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002067 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05002068 return 1;
2069 }
2070 return 0;
2071}
2072
Al Viro808d4e32012-10-11 11:42:01 -04002073static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074{
Al Viro2d92ab32008-08-02 00:51:11 -04002075 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05002076 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05002077 struct mount *old;
Al Viro84d17192013-03-15 10:53:28 -04002078 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002079 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 if (!old_name || !*old_name)
2081 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002082 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 if (err)
2084 return err;
2085
Al Viro84d17192013-03-15 10:53:28 -04002086 mp = lock_mount(path);
2087 err = PTR_ERR(mp);
2088 if (IS_ERR(mp))
David Howellscc53ce52011-01-14 18:45:26 +00002089 goto out;
2090
Al Viro143c8c92011-11-25 00:46:35 -05002091 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002092 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05002093
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002095 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 goto out1;
2097
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002098 if (old->mnt.mnt_flags & MNT_LOCKED)
2099 goto out1;
2100
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002102 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05002103 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Al Viro676da582011-11-24 21:47:05 -05002105 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05002106 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Al Viro2d92ab32008-08-02 00:51:11 -04002108 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
2109 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05002110 goto out1;
2111 /*
2112 * Don't move a mount residing in a shared parent.
2113 */
Al Virofc7be132011-11-25 01:05:37 -05002114 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002115 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05002116 /*
2117 * Don't move a mount tree containing unbindable mounts to a destination
2118 * mount which is shared.
2119 */
Al Virofc7be132011-11-25 01:05:37 -05002120 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05002121 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05002123 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05002124 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05002125 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
Al Viro84d17192013-03-15 10:53:28 -04002127 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08002128 if (err)
Ram Pai21444402005-11-07 17:20:03 -05002129 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130
2131 /* if the mount is moved, it should no longer be expire
2132 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05002133 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134out1:
Al Viro84d17192013-03-15 10:53:28 -04002135 unlock_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04002138 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04002139 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 return err;
2141}
2142
Al Viro9d412a42011-03-17 22:08:28 -04002143static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2144{
2145 int err;
2146 const char *subtype = strchr(fstype, '.');
2147 if (subtype) {
2148 subtype++;
2149 err = -EINVAL;
2150 if (!subtype[0])
2151 goto err;
2152 } else
2153 subtype = "";
2154
2155 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2156 err = -ENOMEM;
2157 if (!mnt->mnt_sb->s_subtype)
2158 goto err;
2159 return mnt;
2160
2161 err:
2162 mntput(mnt);
2163 return ERR_PTR(err);
2164}
2165
Al Viro9d412a42011-03-17 22:08:28 -04002166/*
2167 * add a mount into a namespace's mount tree
2168 */
Al Viro95bc5f22011-11-25 00:30:56 -05002169static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04002170{
Al Viro84d17192013-03-15 10:53:28 -04002171 struct mountpoint *mp;
2172 struct mount *parent;
Al Viro9d412a42011-03-17 22:08:28 -04002173 int err;
2174
Al Virof2ebb3a2014-02-27 09:35:45 -05002175 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04002176
Al Viro84d17192013-03-15 10:53:28 -04002177 mp = lock_mount(path);
2178 if (IS_ERR(mp))
2179 return PTR_ERR(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002180
Al Viro84d17192013-03-15 10:53:28 -04002181 parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04002182 err = -EINVAL;
Al Viro84d17192013-03-15 10:53:28 -04002183 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04002184 /* that's acceptable only for automounts done in private ns */
2185 if (!(mnt_flags & MNT_SHRINKABLE))
2186 goto unlock;
2187 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d17192013-03-15 10:53:28 -04002188 if (!parent->mnt_ns)
Al Viro156cacb2012-09-21 08:19:02 -04002189 goto unlock;
2190 }
Al Viro9d412a42011-03-17 22:08:28 -04002191
2192 /* Refuse the same filesystem on the same mount point */
2193 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05002194 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04002195 path->mnt->mnt_root == path->dentry)
2196 goto unlock;
2197
2198 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05002199 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04002200 goto unlock;
2201
Al Viro95bc5f22011-11-25 00:30:56 -05002202 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro84d17192013-03-15 10:53:28 -04002203 err = graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04002204
2205unlock:
Al Viro84d17192013-03-15 10:53:28 -04002206 unlock_mount(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002207 return err;
2208}
Al Virob1e75df2011-01-17 01:47:59 -05002209
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210/*
2211 * create a new mount for userspace and request it to be added into the
2212 * namespace's tree
2213 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002214static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04002215 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002217 struct file_system_type *type;
Al Viro9b40bc92013-02-22 22:45:42 -05002218 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002220 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002222 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 return -EINVAL;
2224
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002225 type = get_fs_type(fstype);
2226 if (!type)
2227 return -ENODEV;
2228
2229 if (user_ns != &init_user_ns) {
2230 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2231 put_filesystem(type);
2232 return -EPERM;
2233 }
2234 /* Only in special cases allow devices from mounts
2235 * created outside the initial user namespace.
2236 */
2237 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2238 flags |= MS_NODEV;
Eric W. Biederman9566d672014-07-28 17:26:07 -07002239 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002240 }
2241 }
2242
2243 mnt = vfs_kern_mount(type, flags, name, data);
2244 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2245 !mnt->mnt_sb->s_subtype)
2246 mnt = fs_set_subtype(mnt, fstype);
2247
2248 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 if (IS_ERR(mnt))
2250 return PTR_ERR(mnt);
2251
Al Viro95bc5f22011-11-25 00:30:56 -05002252 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002253 if (err)
2254 mntput(mnt);
2255 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256}
2257
Al Viro19a167a2011-01-17 01:35:23 -05002258int finish_automount(struct vfsmount *m, struct path *path)
2259{
Al Viro6776db3d2011-11-25 00:22:05 -05002260 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002261 int err;
2262 /* The new mount record should have at least 2 refs to prevent it being
2263 * expired before we get a chance to add it
2264 */
Al Viro6776db3d2011-11-25 00:22:05 -05002265 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002266
2267 if (m->mnt_sb == path->mnt->mnt_sb &&
2268 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002269 err = -ELOOP;
2270 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002271 }
2272
Al Viro95bc5f22011-11-25 00:30:56 -05002273 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002274 if (!err)
2275 return 0;
2276fail:
2277 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002278 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002279 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05002280 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04002281 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002282 }
Al Virob1e75df2011-01-17 01:47:59 -05002283 mntput(m);
2284 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002285 return err;
2286}
2287
David Howellsea5b7782011-01-14 19:10:03 +00002288/**
2289 * mnt_set_expiry - Put a mount on an expiration list
2290 * @mnt: The mount to list.
2291 * @expiry_list: The list to add the mount to.
2292 */
2293void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2294{
Al Viro97216be2013-03-16 15:12:40 -04002295 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00002296
Al Viro6776db3d2011-11-25 00:22:05 -05002297 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002298
Al Viro97216be2013-03-16 15:12:40 -04002299 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002300}
2301EXPORT_SYMBOL(mnt_set_expiry);
2302
2303/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 * process a list of expirable mountpoints with the intent of discarding any
2305 * mountpoints that aren't in use and haven't been touched since last we came
2306 * here
2307 */
2308void mark_mounts_for_expiry(struct list_head *mounts)
2309{
Al Viro761d5c32011-11-24 21:07:43 -05002310 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 LIST_HEAD(graveyard);
2312
2313 if (list_empty(mounts))
2314 return;
2315
Al Viro97216be2013-03-16 15:12:40 -04002316 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002317 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318
2319 /* extract from the expiration list every vfsmount that matches the
2320 * following criteria:
2321 * - only referenced by its parent vfsmount
2322 * - still marked for expiry (marked on the last call here; marks are
2323 * cleared by mntput())
2324 */
Al Viro6776db3d2011-11-25 00:22:05 -05002325 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002326 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002327 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002329 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002331 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002332 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002333 touch_mnt_namespace(mnt->mnt_ns);
Al Viro328e6d92013-03-16 14:49:45 -04002334 umount_tree(mnt, 1);
Al Virobcc5c7d2008-03-22 00:21:53 -04002335 }
Al Viro719ea2f2013-09-29 11:24:49 -04002336 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002337 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338}
2339
2340EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2341
2342/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002343 * Ripoff of 'select_parent()'
2344 *
2345 * search the list of submounts for a given mountpoint, and move any
2346 * shrinkable submounts to the 'graveyard' list.
2347 */
Al Viro692afc32011-11-24 21:15:14 -05002348static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002349{
Al Viro692afc32011-11-24 21:15:14 -05002350 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002351 struct list_head *next;
2352 int found = 0;
2353
2354repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002355 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002356resume:
Al Viro6b41d532011-11-24 23:24:33 -05002357 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002358 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002359 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002360
2361 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002362 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002363 continue;
2364 /*
2365 * Descend a level if the d_mounts list is non-empty.
2366 */
Al Viro6b41d532011-11-24 23:24:33 -05002367 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002368 this_parent = mnt;
2369 goto repeat;
2370 }
2371
Al Viro1ab59732011-11-24 21:35:16 -05002372 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002373 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002374 found++;
2375 }
2376 }
2377 /*
2378 * All done at this level ... ascend and resume the search
2379 */
2380 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002381 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002382 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002383 goto resume;
2384 }
2385 return found;
2386}
2387
2388/*
2389 * process a list of expirable mountpoints with the intent of discarding any
2390 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002391 *
Al Viro48a066e2013-09-29 22:06:07 -04002392 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002393 */
Al Virob54b9be2013-03-16 14:39:34 -04002394static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002395{
2396 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002397 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002398
Trond Myklebust5528f9112006-06-09 09:34:17 -04002399 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002400 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002401 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002402 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002403 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002404 touch_mnt_namespace(m->mnt_ns);
Al Viro328e6d92013-03-16 14:49:45 -04002405 umount_tree(m, 1);
Al Virobcc5c7d2008-03-22 00:21:53 -04002406 }
2407 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002408}
2409
Trond Myklebust5528f9112006-06-09 09:34:17 -04002410/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 * Some copy_from_user() implementations do not return the exact number of
2412 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2413 * Note that this function differs from copy_from_user() in that it will oops
2414 * on bad values of `to', rather than returning a short copy.
2415 */
Ram Paib58fed82005-11-07 17:16:09 -05002416static long exact_copy_from_user(void *to, const void __user * from,
2417 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418{
2419 char *t = to;
2420 const char __user *f = from;
2421 char c;
2422
2423 if (!access_ok(VERIFY_READ, from, n))
2424 return n;
2425
2426 while (n) {
2427 if (__get_user(c, f)) {
2428 memset(t, 0, n);
2429 break;
2430 }
2431 *t++ = c;
2432 f++;
2433 n--;
2434 }
2435 return n;
2436}
2437
Ram Paib58fed82005-11-07 17:16:09 -05002438int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439{
2440 int i;
2441 unsigned long page;
2442 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002443
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 *where = 0;
2445 if (!data)
2446 return 0;
2447
2448 if (!(page = __get_free_page(GFP_KERNEL)))
2449 return -ENOMEM;
2450
2451 /* We only care that *some* data at the address the user
2452 * gave us is valid. Just in case, we'll zero
2453 * the remainder of the page.
2454 */
2455 /* copy_from_user cannot cross TASK_SIZE ! */
2456 size = TASK_SIZE - (unsigned long)data;
2457 if (size > PAGE_SIZE)
2458 size = PAGE_SIZE;
2459
2460 i = size - exact_copy_from_user((void *)page, data, size);
2461 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002462 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 return -EFAULT;
2464 }
2465 if (i != PAGE_SIZE)
2466 memset((char *)page + i, 0, PAGE_SIZE - i);
2467 *where = page;
2468 return 0;
2469}
2470
Vegard Nossumeca6f532009-09-18 13:05:45 -07002471int copy_mount_string(const void __user *data, char **where)
2472{
2473 char *tmp;
2474
2475 if (!data) {
2476 *where = NULL;
2477 return 0;
2478 }
2479
2480 tmp = strndup_user(data, PAGE_SIZE);
2481 if (IS_ERR(tmp))
2482 return PTR_ERR(tmp);
2483
2484 *where = tmp;
2485 return 0;
2486}
2487
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488/*
2489 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2490 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2491 *
2492 * data is a (void *) that can point to any structure up to
2493 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2494 * information (or be NULL).
2495 *
2496 * Pre-0.97 versions of mount() didn't have a flags word.
2497 * When the flags word was introduced its top half was required
2498 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2499 * Therefore, if this magic number is present, it carries no information
2500 * and must be discarded.
2501 */
Al Viro808d4e32012-10-11 11:42:01 -04002502long do_mount(const char *dev_name, const char *dir_name,
2503 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504{
Al Viro2d92ab32008-08-02 00:51:11 -04002505 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 int retval = 0;
2507 int mnt_flags = 0;
2508
2509 /* Discard magic */
2510 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2511 flags &= ~MS_MGC_MSK;
2512
2513 /* Basic sanity checks */
2514
2515 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2516 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517
2518 if (data_page)
2519 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2520
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002521 /* ... and get the mountpoint */
2522 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2523 if (retval)
2524 return retval;
2525
2526 retval = security_sb_mount(dev_name, &path,
2527 type_page, flags, data_page);
Al Viro0d5cadb2013-05-04 14:40:51 -04002528 if (!retval && !may_mount())
2529 retval = -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002530 if (retval)
2531 goto dput_out;
2532
Andi Kleen613cbe32009-04-19 18:40:43 +02002533 /* Default to relatime unless overriden */
2534 if (!(flags & MS_NOATIME))
2535 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002536
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 /* Separate the per-mountpoint flags */
2538 if (flags & MS_NOSUID)
2539 mnt_flags |= MNT_NOSUID;
2540 if (flags & MS_NODEV)
2541 mnt_flags |= MNT_NODEV;
2542 if (flags & MS_NOEXEC)
2543 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002544 if (flags & MS_NOATIME)
2545 mnt_flags |= MNT_NOATIME;
2546 if (flags & MS_NODIRATIME)
2547 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002548 if (flags & MS_STRICTATIME)
2549 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002550 if (flags & MS_RDONLY)
2551 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002552
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07002553 /* The default atime for remount is preservation */
2554 if ((flags & MS_REMOUNT) &&
2555 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2556 MS_STRICTATIME)) == 0)) {
2557 mnt_flags &= ~MNT_ATIME_MASK;
2558 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2559 }
2560
Al Viro7a4dec52010-08-09 12:05:43 -04002561 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002562 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2563 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002566 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 data_page);
2568 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002569 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002570 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002571 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002573 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 else
Al Viro2d92ab32008-08-02 00:51:11 -04002575 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 dev_name, data_page);
2577dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002578 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 return retval;
2580}
2581
Eric W. Biederman771b1372012-07-26 21:08:32 -07002582static void free_mnt_ns(struct mnt_namespace *ns)
2583{
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002584 proc_free_inum(ns->proc_inum);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002585 put_user_ns(ns->user_ns);
2586 kfree(ns);
2587}
2588
Eric W. Biederman8823c072010-03-07 18:49:36 -08002589/*
2590 * Assign a sequence number so we can detect when we attempt to bind
2591 * mount a reference to an older mount namespace into the current
2592 * mount namespace, preventing reference counting loops. A 64bit
2593 * number incrementing at 10Ghz will take 12,427 years to wrap which
2594 * is effectively never, so we can ignore the possibility.
2595 */
2596static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2597
Eric W. Biederman771b1372012-07-26 21:08:32 -07002598static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002599{
2600 struct mnt_namespace *new_ns;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002601 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002602
2603 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2604 if (!new_ns)
2605 return ERR_PTR(-ENOMEM);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002606 ret = proc_alloc_inum(&new_ns->proc_inum);
2607 if (ret) {
2608 kfree(new_ns);
2609 return ERR_PTR(ret);
2610 }
Eric W. Biederman8823c072010-03-07 18:49:36 -08002611 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002612 atomic_set(&new_ns->count, 1);
2613 new_ns->root = NULL;
2614 INIT_LIST_HEAD(&new_ns->list);
2615 init_waitqueue_head(&new_ns->poll);
2616 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002617 new_ns->user_ns = get_user_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002618 return new_ns;
2619}
2620
Al Viro9559f682013-09-28 20:47:57 -04002621struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2622 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002624 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002625 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002626 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04002627 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05002628 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002629 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630
Al Viro9559f682013-09-28 20:47:57 -04002631 BUG_ON(!ns);
2632
2633 if (likely(!(flags & CLONE_NEWNS))) {
2634 get_mnt_ns(ns);
2635 return ns;
2636 }
2637
2638 old = ns->root;
2639
Eric W. Biederman771b1372012-07-26 21:08:32 -07002640 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002641 if (IS_ERR(new_ns))
2642 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
Al Viro97216be2013-03-16 15:12:40 -04002644 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002646 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04002647 if (user_ns != ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002648 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002649 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002650 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04002651 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002652 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002653 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 }
Al Virobe08d6d2011-12-06 13:32:36 -05002655 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002656 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657
2658 /*
2659 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2660 * as belonging to new namespace. We have already acquired a private
2661 * fs_struct, so tsk->fs->lock is not needed.
2662 */
Al Viro909b0a82011-11-25 03:06:56 -05002663 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002664 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002666 q->mnt_ns = new_ns;
Al Viro9559f682013-09-28 20:47:57 -04002667 if (new_fs) {
2668 if (&p->mnt == new_fs->root.mnt) {
2669 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002670 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 }
Al Viro9559f682013-09-28 20:47:57 -04002672 if (&p->mnt == new_fs->pwd.mnt) {
2673 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002674 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 }
Al Viro909b0a82011-11-25 03:06:56 -05002677 p = next_mnt(p, old);
2678 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002679 if (!q)
2680 break;
2681 while (p->mnt.mnt_root != q->mnt.mnt_root)
2682 p = next_mnt(p, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 }
Al Viro328e6d92013-03-16 14:49:45 -04002684 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002687 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002689 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
JANAK DESAI741a2952006-02-07 12:59:00 -08002691 return new_ns;
2692}
2693
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002694/**
2695 * create_mnt_ns - creates a private namespace and adds a root filesystem
2696 * @mnt: pointer to the new root filesystem mountpoint
2697 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002698static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002699{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002700 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002701 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002702 struct mount *mnt = real_mount(m);
2703 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002704 new_ns->root = mnt;
Al Virob1983cd2013-05-04 15:18:53 -04002705 list_add(&mnt->mnt_list, &new_ns->list);
Al Viroc1334492011-11-16 16:12:14 -05002706 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002707 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002708 }
2709 return new_ns;
2710}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002711
Al Viroea441d12011-11-16 21:43:59 -05002712struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2713{
2714 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002715 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002716 struct path path;
2717 int err;
2718
2719 ns = create_mnt_ns(mnt);
2720 if (IS_ERR(ns))
2721 return ERR_CAST(ns);
2722
2723 err = vfs_path_lookup(mnt->mnt_root, mnt,
2724 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2725
2726 put_mnt_ns(ns);
2727
2728 if (err)
2729 return ERR_PTR(err);
2730
2731 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002732 s = path.mnt->mnt_sb;
2733 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002734 mntput(path.mnt);
2735 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002736 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002737 /* ... and return the root of (sub)tree on it */
2738 return path.dentry;
2739}
2740EXPORT_SYMBOL(mount_subtree);
2741
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002742SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2743 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002745 int ret;
2746 char *kernel_type;
Jeff Layton91a27b22012-10-10 15:25:28 -04002747 struct filename *kernel_dir;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002748 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750
Vegard Nossumeca6f532009-09-18 13:05:45 -07002751 ret = copy_mount_string(type, &kernel_type);
2752 if (ret < 0)
2753 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754
Vegard Nossumeca6f532009-09-18 13:05:45 -07002755 kernel_dir = getname(dir_name);
2756 if (IS_ERR(kernel_dir)) {
2757 ret = PTR_ERR(kernel_dir);
2758 goto out_dir;
2759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760
Vegard Nossumeca6f532009-09-18 13:05:45 -07002761 ret = copy_mount_string(dev_name, &kernel_dev);
2762 if (ret < 0)
2763 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764
Vegard Nossumeca6f532009-09-18 13:05:45 -07002765 ret = copy_mount_options(data, &data_page);
2766 if (ret < 0)
2767 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768
Jeff Layton91a27b22012-10-10 15:25:28 -04002769 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
Vegard Nossumeca6f532009-09-18 13:05:45 -07002770 (void *) data_page);
2771
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002773out_data:
2774 kfree(kernel_dev);
2775out_dev:
2776 putname(kernel_dir);
2777out_dir:
2778 kfree(kernel_type);
2779out_type:
2780 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781}
2782
2783/*
Al Viroafac7cb2011-11-23 19:34:49 -05002784 * Return true if path is reachable from root
2785 *
Al Viro48a066e2013-09-29 22:06:07 -04002786 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05002787 */
Al Viro643822b2011-11-24 22:00:28 -05002788bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002789 const struct path *root)
2790{
Al Viro643822b2011-11-24 22:00:28 -05002791 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002792 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002793 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002794 }
Al Viro643822b2011-11-24 22:00:28 -05002795 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002796}
2797
2798int path_is_under(struct path *path1, struct path *path2)
2799{
2800 int res;
Al Viro48a066e2013-09-29 22:06:07 -04002801 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002802 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04002803 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002804 return res;
2805}
2806EXPORT_SYMBOL(path_is_under);
2807
2808/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 * pivot_root Semantics:
2810 * Moves the root file system of the current process to the directory put_old,
2811 * makes new_root as the new root file system of the current process, and sets
2812 * root/cwd of all processes which had them on the current root to new_root.
2813 *
2814 * Restrictions:
2815 * The new_root and put_old must be directories, and must not be on the
2816 * same file system as the current process root. The put_old must be
2817 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2818 * pointed to by put_old must yield the same directory as new_root. No other
2819 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2820 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002821 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2822 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2823 * in this situation.
2824 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 * Notes:
2826 * - we don't move root/cwd if they are not at the root (reason: if something
2827 * cared enough to change them, it's probably wrong to force them elsewhere)
2828 * - it's okay to pick a root that isn't the root of a file system, e.g.
2829 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2830 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2831 * first.
2832 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002833SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2834 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835{
Al Viro2d8f3032008-07-22 09:59:21 -04002836 struct path new, old, parent_path, root_parent, root;
Al Viro84d17192013-03-15 10:53:28 -04002837 struct mount *new_mnt, *root_mnt, *old_mnt;
2838 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 int error;
2840
Al Viro9b40bc92013-02-22 22:45:42 -05002841 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 return -EPERM;
2843
Al Viro2d8f3032008-07-22 09:59:21 -04002844 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 if (error)
2846 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847
Al Viro2d8f3032008-07-22 09:59:21 -04002848 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 if (error)
2850 goto out1;
2851
Al Viro2d8f3032008-07-22 09:59:21 -04002852 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002853 if (error)
2854 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002856 get_fs_root(current->fs, &root);
Al Viro84d17192013-03-15 10:53:28 -04002857 old_mp = lock_mount(&old);
2858 error = PTR_ERR(old_mp);
2859 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04002860 goto out3;
2861
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002863 new_mnt = real_mount(new.mnt);
2864 root_mnt = real_mount(root.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002865 old_mnt = real_mount(old.mnt);
2866 if (IS_MNT_SHARED(old_mnt) ||
Al Virofc7be132011-11-25 01:05:37 -05002867 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2868 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002869 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002870 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002871 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002872 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2873 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002875 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002876 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 error = -EBUSY;
Al Viro84d17192013-03-15 10:53:28 -04002878 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04002879 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002881 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002882 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002883 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002884 goto out4; /* not attached */
Al Viro84d17192013-03-15 10:53:28 -04002885 root_mp = root_mnt->mnt_mp;
Al Viro2d8f3032008-07-22 09:59:21 -04002886 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002887 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002888 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002889 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002890 /* make sure we can reach put_old from new_root */
Al Viro84d17192013-03-15 10:53:28 -04002891 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002892 goto out4;
Al Viro84d17192013-03-15 10:53:28 -04002893 root_mp->m_count++; /* pin it so it won't go away */
Al Viro719ea2f2013-09-29 11:24:49 -04002894 lock_mount_hash();
Al Viro419148d2011-11-24 19:41:16 -05002895 detach_mnt(new_mnt, &parent_path);
2896 detach_mnt(root_mnt, &root_parent);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002897 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2898 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2899 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2900 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002901 /* mount old root on put_old */
Al Viro84d17192013-03-15 10:53:28 -04002902 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08002903 /* mount new_root on / */
Al Viro84d17192013-03-15 10:53:28 -04002904 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002905 touch_mnt_namespace(current->nsproxy->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04002906 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04002907 chroot_fs_refs(&root, &new);
Al Viro84d17192013-03-15 10:53:28 -04002908 put_mountpoint(root_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002910out4:
Al Viro84d17192013-03-15 10:53:28 -04002911 unlock_mount(old_mp);
Al Virob12cea92011-03-18 08:55:38 -04002912 if (!error) {
2913 path_put(&root_parent);
2914 path_put(&parent_path);
2915 }
2916out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002917 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002918out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002919 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002921 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924}
2925
2926static void __init init_mount_tree(void)
2927{
2928 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002929 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002930 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002931 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002933 type = get_fs_type("rootfs");
2934 if (!type)
2935 panic("Can't find rootfs type");
2936 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2937 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 if (IS_ERR(mnt))
2939 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002940
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002941 ns = create_mnt_ns(mnt);
2942 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002945 init_task.nsproxy->mnt_ns = ns;
2946 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947
Al Virobe08d6d2011-12-06 13:32:36 -05002948 root.mnt = mnt;
2949 root.dentry = mnt->mnt_root;
Jan Blunckac748a02008-02-14 19:34:39 -08002950
2951 set_fs_pwd(current->fs, &root);
2952 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953}
2954
Denis Cheng74bf17c2007-10-16 23:26:30 -07002955void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002957 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002958 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959
Al Viro7d6fec42011-11-23 12:14:10 -05002960 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002961 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962
Al Viro0818bf22014-02-28 13:46:44 -05002963 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04002964 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05002965 mhash_entries, 19,
2966 0,
2967 &m_hash_shift, &m_hash_mask, 0, 0);
2968 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
2969 sizeof(struct hlist_head),
2970 mphash_entries, 19,
2971 0,
2972 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973
Al Viro84d17192013-03-15 10:53:28 -04002974 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 panic("Failed to allocate mount hash table\n");
2976
Al Viro0818bf22014-02-28 13:46:44 -05002977 for (u = 0; u <= m_hash_mask; u++)
Al Viro38129a12014-03-20 21:10:51 -04002978 INIT_HLIST_HEAD(&mount_hashtable[u]);
Al Viro0818bf22014-02-28 13:46:44 -05002979 for (u = 0; u <= mp_hash_mask; u++)
2980 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981
Tejun Heo4b93dc92013-11-28 14:54:43 -05002982 kernfs_init();
2983
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002984 err = sysfs_init();
2985 if (err)
2986 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002987 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002988 fs_kobj = kobject_create_and_add("fs", NULL);
2989 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002990 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 init_rootfs();
2992 init_mount_tree();
2993}
2994
Trond Myklebust616511d2009-06-22 15:09:13 -04002995void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996{
Al Virod498b252010-02-05 02:21:06 -05002997 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002998 return;
Al Viro7b00ed62013-09-16 21:19:20 -04002999 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003000 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001}
Al Viro9d412a42011-03-17 22:08:28 -04003002
3003struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3004{
Tim Chen423e0ab2011-07-19 09:32:38 -07003005 struct vfsmount *mnt;
3006 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3007 if (!IS_ERR(mnt)) {
3008 /*
3009 * it is a longterm mount, don't release mnt until
3010 * we unmount before file sys is unregistered
3011 */
Al Virof7a99c52012-06-09 00:59:08 -04003012 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07003013 }
3014 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04003015}
3016EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07003017
3018void kern_unmount(struct vfsmount *mnt)
3019{
3020 /* release long term mount so mount point can be released */
3021 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04003022 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04003023 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07003024 mntput(mnt);
3025 }
3026}
3027EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05003028
3029bool our_mnt(struct vfsmount *mnt)
3030{
Al Viro143c8c92011-11-25 00:46:35 -05003031 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05003032}
Eric W. Biederman8823c072010-03-07 18:49:36 -08003033
Eric W. Biederman31515272013-03-15 01:45:51 -07003034bool current_chrooted(void)
3035{
3036 /* Does the current process have a non-standard root */
3037 struct path ns_root;
3038 struct path fs_root;
3039 bool chrooted;
3040
3041 /* Find the namespace root */
3042 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3043 ns_root.dentry = ns_root.mnt->mnt_root;
3044 path_get(&ns_root);
3045 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3046 ;
3047
3048 get_fs_root(current->fs, &fs_root);
3049
3050 chrooted = !path_equal(&fs_root, &ns_root);
3051
3052 path_put(&fs_root);
3053 path_put(&ns_root);
3054
3055 return chrooted;
3056}
3057
Eric W. Biedermane51db732013-03-30 19:57:41 -07003058bool fs_fully_visible(struct file_system_type *type)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003059{
3060 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3061 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07003062 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003063
Eric W. Biedermane51db732013-03-30 19:57:41 -07003064 if (unlikely(!ns))
3065 return false;
3066
Al Viro44bb4382013-09-16 21:37:36 -04003067 down_read(&namespace_sem);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003068 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07003069 struct mount *child;
3070 if (mnt->mnt.mnt_sb->s_type != type)
3071 continue;
3072
3073 /* This mount is not fully visible if there are any child mounts
3074 * that cover anything except for empty directories.
3075 */
3076 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3077 struct inode *inode = child->mnt_mountpoint->d_inode;
3078 if (!S_ISDIR(inode->i_mode))
3079 goto next;
Eric W. Biederman41301ae2013-11-14 21:22:25 -08003080 if (inode->i_nlink > 2)
Eric W. Biedermane51db732013-03-30 19:57:41 -07003081 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003082 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07003083 visible = true;
3084 goto found;
3085 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003086 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07003087found:
Al Viro44bb4382013-09-16 21:37:36 -04003088 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003089 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003090}
3091
Eric W. Biederman8823c072010-03-07 18:49:36 -08003092static void *mntns_get(struct task_struct *task)
3093{
3094 struct mnt_namespace *ns = NULL;
3095 struct nsproxy *nsproxy;
3096
Eric W. Biederman728dba32014-02-03 19:13:49 -08003097 task_lock(task);
3098 nsproxy = task->nsproxy;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003099 if (nsproxy) {
3100 ns = nsproxy->mnt_ns;
3101 get_mnt_ns(ns);
3102 }
Eric W. Biederman728dba32014-02-03 19:13:49 -08003103 task_unlock(task);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003104
3105 return ns;
3106}
3107
3108static void mntns_put(void *ns)
3109{
3110 put_mnt_ns(ns);
3111}
3112
3113static int mntns_install(struct nsproxy *nsproxy, void *ns)
3114{
3115 struct fs_struct *fs = current->fs;
3116 struct mnt_namespace *mnt_ns = ns;
3117 struct path root;
3118
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003119 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -07003120 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3121 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08003122 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003123
3124 if (fs->users != 1)
3125 return -EINVAL;
3126
3127 get_mnt_ns(mnt_ns);
3128 put_mnt_ns(nsproxy->mnt_ns);
3129 nsproxy->mnt_ns = mnt_ns;
3130
3131 /* Find the root */
3132 root.mnt = &mnt_ns->root->mnt;
3133 root.dentry = mnt_ns->root->mnt.mnt_root;
3134 path_get(&root);
3135 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3136 ;
3137
3138 /* Update the pwd and root */
3139 set_fs_pwd(fs, &root);
3140 set_fs_root(fs, &root);
3141
3142 path_put(&root);
3143 return 0;
3144}
3145
Eric W. Biederman98f842e2011-06-15 10:21:48 -07003146static unsigned int mntns_inum(void *ns)
3147{
3148 struct mnt_namespace *mnt_ns = ns;
3149 return mnt_ns->proc_inum;
3150}
3151
Eric W. Biederman8823c072010-03-07 18:49:36 -08003152const struct proc_ns_operations mntns_operations = {
3153 .name = "mnt",
3154 .type = CLONE_NEWNS,
3155 .get = mntns_get,
3156 .put = mntns_put,
3157 .install = mntns_install,
Eric W. Biederman98f842e2011-06-15 10:21:48 -07003158 .inum = mntns_inum,
Eric W. Biederman8823c072010-03-07 18:49:36 -08003159};