blob: edac42c6eff22b556c63bd17d386f6998d6f6141 [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>
Al Virod10577a2011-12-07 13:06:11 -050019#include <linux/acct.h> /* acct_auto_close_mnt */
20#include <linux/ramfs.h> /* init_rootfs */
21#include <linux/fs_struct.h> /* get_fs_root et.al. */
22#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23#include <linux/uaccess.h>
Eric W. Biederman8823c072010-03-07 18:49:36 -080024#include <linux/proc_fs.h>
Ram Pai07b20882005-11-07 17:19:07 -050025#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070026#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
Eric Dumazet13f14b42008-02-06 01:37:57 -080028#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
29#define HASH_SIZE (1UL << HASH_SHIFT)
30
Al Viro5addc5d2005-11-07 17:15:49 -050031static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010032static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010033static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100034static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040035static int mnt_id_start = 0;
36static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Eric Dumazetfa3536c2006-03-26 01:37:24 -080038static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080039static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050040static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080042/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060043struct kobject *fs_kobj;
44EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080045
Nick Piggin99b7db72010-08-18 04:37:39 +100046/*
47 * vfsmount lock may be taken for read to prevent changes to the
48 * vfsmount hash, ie. during mountpoint lookups or walking back
49 * up the tree.
50 *
51 * It should be taken for write in all cases where the vfsmount
52 * tree or hash is modified or when a vfsmount structure is modified.
53 */
54DEFINE_BRLOCK(vfsmount_lock);
55
Linus Torvalds1da177e2005-04-16 15:20:36 -070056static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
57{
Ram Paib58fed82005-11-07 17:16:09 -050058 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
59 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080060 tmp = tmp + (tmp >> HASH_SHIFT);
61 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062}
63
Dave Hansen3d733632008-02-15 14:37:59 -080064#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
65
Nick Piggin99b7db72010-08-18 04:37:39 +100066/*
67 * allocation is serialized by namespace_sem, but we need the spinlock to
68 * serialize with freeing.
69 */
Al Virob105e272011-11-24 20:38:33 -050070static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010071{
72 int res;
73
74retry:
75 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +100076 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -050077 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -040078 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -050079 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +100080 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010081 if (res == -EAGAIN)
82 goto retry;
83
84 return res;
85}
86
Al Virob105e272011-11-24 20:38:33 -050087static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010088{
Al Viro15169fe2011-11-25 00:50:41 -050089 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +100090 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040091 ida_remove(&mnt_id_ida, id);
92 if (mnt_id_start > id)
93 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +100094 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010095}
96
Miklos Szeredi719f5d72008-03-27 13:06:23 +010097/*
98 * Allocate a new peer group ID
99 *
100 * mnt_group_ida is protected by namespace_sem
101 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500102static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100103{
Al Virof21f6222009-06-24 03:12:00 -0400104 int res;
105
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100106 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
107 return -ENOMEM;
108
Al Virof21f6222009-06-24 03:12:00 -0400109 res = ida_get_new_above(&mnt_group_ida,
110 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500111 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400112 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500113 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400114
115 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100116}
117
118/*
119 * Release a peer group ID
120 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500121void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100122{
Al Viro15169fe2011-11-25 00:50:41 -0500123 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400124 ida_remove(&mnt_group_ida, id);
125 if (mnt_group_start > id)
126 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500127 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100128}
129
Nick Pigginb3e19d92011-01-07 17:50:11 +1100130/*
131 * vfsmount lock must be held for read
132 */
Al Viro83adc752011-11-24 22:37:54 -0500133static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100134{
135#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500136 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100137#else
138 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500139 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100140 preempt_enable();
141#endif
142}
143
Nick Pigginb3e19d92011-01-07 17:50:11 +1100144/*
145 * vfsmount lock must be held for write
146 */
Al Viro83adc752011-11-24 22:37:54 -0500147unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100148{
149#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500150 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100151 int cpu;
152
153 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500154 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100155 }
156
157 return count;
158#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500159 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100160#endif
161}
162
Al Virob105e272011-11-24 20:38:33 -0500163static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
Al Viroc63181e2011-11-25 02:35:16 -0500165 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
166 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100167 int err;
168
Al Viroc63181e2011-11-25 02:35:16 -0500169 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800170 if (err)
171 goto out_free_cache;
172
173 if (name) {
Al Viroc63181e2011-11-25 02:35:16 -0500174 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
175 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800176 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100177 }
178
Nick Pigginb3e19d92011-01-07 17:50:11 +1100179#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500180 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
181 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100182 goto out_free_devname;
183
Al Viroc63181e2011-11-25 02:35:16 -0500184 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100185#else
Al Viroc63181e2011-11-25 02:35:16 -0500186 mnt->mnt_count = 1;
187 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100188#endif
189
Al Viroc63181e2011-11-25 02:35:16 -0500190 INIT_LIST_HEAD(&mnt->mnt_hash);
191 INIT_LIST_HEAD(&mnt->mnt_child);
192 INIT_LIST_HEAD(&mnt->mnt_mounts);
193 INIT_LIST_HEAD(&mnt->mnt_list);
194 INIT_LIST_HEAD(&mnt->mnt_expire);
195 INIT_LIST_HEAD(&mnt->mnt_share);
196 INIT_LIST_HEAD(&mnt->mnt_slave_list);
197 INIT_LIST_HEAD(&mnt->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500198#ifdef CONFIG_FSNOTIFY
199 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
200#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 }
Al Viroc63181e2011-11-25 02:35:16 -0500202 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800203
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000204#ifdef CONFIG_SMP
205out_free_devname:
Al Viroc63181e2011-11-25 02:35:16 -0500206 kfree(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000207#endif
Li Zefan88b38782008-07-21 18:06:36 +0800208out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500209 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800210out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500211 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800212 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
Dave Hansen83660252008-02-15 14:37:30 -0800215/*
216 * Most r/o checks on a fs are for operations that take
217 * discrete amounts of time, like a write() or unlink().
218 * We must keep track of when those operations start
219 * (for permission checks) and when they end, so that
220 * we can determine when writes are able to occur to
221 * a filesystem.
222 */
Dave Hansen3d733632008-02-15 14:37:59 -0800223/*
224 * __mnt_is_readonly: check whether a mount is read-only
225 * @mnt: the mount to check for its write status
226 *
227 * This shouldn't be used directly ouside of the VFS.
228 * It does not guarantee that the filesystem will stay
229 * r/w, just that it is right *now*. This can not and
230 * should not be used in place of IS_RDONLY(inode).
231 * mnt_want/drop_write() will _keep_ the filesystem
232 * r/w.
233 */
234int __mnt_is_readonly(struct vfsmount *mnt)
235{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800236 if (mnt->mnt_flags & MNT_READONLY)
237 return 1;
238 if (mnt->mnt_sb->s_flags & MS_RDONLY)
239 return 1;
240 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800241}
242EXPORT_SYMBOL_GPL(__mnt_is_readonly);
243
Al Viro83adc752011-11-24 22:37:54 -0500244static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800245{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000246#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500247 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000248#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500249 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000250#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800251}
Dave Hansen3d733632008-02-15 14:37:59 -0800252
Al Viro83adc752011-11-24 22:37:54 -0500253static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800254{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000255#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500256 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000257#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500258 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000259#endif
260}
261
Al Viro83adc752011-11-24 22:37:54 -0500262static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000263{
264#ifdef CONFIG_SMP
265 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800266 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800267
268 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500269 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800270 }
Dave Hansen3d733632008-02-15 14:37:59 -0800271
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000272 return count;
273#else
274 return mnt->mnt_writers;
275#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800276}
277
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100278static int mnt_is_readonly(struct vfsmount *mnt)
279{
280 if (mnt->mnt_sb->s_readonly_remount)
281 return 1;
282 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
283 smp_rmb();
284 return __mnt_is_readonly(mnt);
285}
286
Dave Hansen3d733632008-02-15 14:37:59 -0800287/*
Jan Karaeb04c282012-06-12 16:20:35 +0200288 * Most r/o & frozen checks on a fs are for operations that take discrete
289 * amounts of time, like a write() or unlink(). We must keep track of when
290 * those operations start (for permission checks) and when they end, so that we
291 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800292 */
Dave Hansen83660252008-02-15 14:37:30 -0800293/**
Jan Karaeb04c282012-06-12 16:20:35 +0200294 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500295 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800296 *
Jan Karaeb04c282012-06-12 16:20:35 +0200297 * This tells the low-level filesystem that a write is about to be performed to
298 * it, and makes sure that writes are allowed (mnt it read-write) before
299 * returning success. This operation does not protect against filesystem being
300 * frozen. When the write operation is finished, __mnt_drop_write() must be
301 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800302 */
Jan Karaeb04c282012-06-12 16:20:35 +0200303int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800304{
Al Viro83adc752011-11-24 22:37:54 -0500305 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800306 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800307
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000308 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100309 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000310 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100311 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000312 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
313 * incremented count after it has set MNT_WRITE_HOLD.
314 */
315 smp_mb();
Miao Xie1e755292012-11-16 17:23:50 +0800316 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000317 cpu_relax();
318 /*
319 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
320 * be set to match its requirements. So we must not load that until
321 * MNT_WRITE_HOLD is cleared.
322 */
323 smp_rmb();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100324 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100325 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800326 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800327 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000328 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200329
330 return ret;
331}
332
333/**
334 * mnt_want_write - get write access to a mount
335 * @m: the mount on which to take a write
336 *
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mount is read-write, filesystem
339 * is not frozen) before returning success. When the write operation is
340 * finished, mnt_drop_write() must be called. This is effectively a refcount.
341 */
342int mnt_want_write(struct vfsmount *m)
343{
344 int ret;
345
346 sb_start_write(m->mnt_sb);
347 ret = __mnt_want_write(m);
348 if (ret)
349 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800350 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800351}
352EXPORT_SYMBOL_GPL(mnt_want_write);
353
354/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000355 * mnt_clone_write - get write access to a mount
356 * @mnt: the mount on which to take a write
357 *
358 * This is effectively like mnt_want_write, except
359 * it must only be used to take an extra write reference
360 * on a mountpoint that we already know has a write reference
361 * on it. This allows some optimisation.
362 *
363 * After finished, mnt_drop_write must be called as usual to
364 * drop the reference.
365 */
366int mnt_clone_write(struct vfsmount *mnt)
367{
368 /* superblock may be r/o */
369 if (__mnt_is_readonly(mnt))
370 return -EROFS;
371 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500372 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000373 preempt_enable();
374 return 0;
375}
376EXPORT_SYMBOL_GPL(mnt_clone_write);
377
378/**
Jan Karaeb04c282012-06-12 16:20:35 +0200379 * __mnt_want_write_file - get write access to a file's mount
380 * @file: the file who's mount on which to take a write
381 *
382 * This is like __mnt_want_write, but it takes a file and can
383 * do some optimisations if the file is open for write already
384 */
385int __mnt_want_write_file(struct file *file)
386{
387 struct inode *inode = file->f_dentry->d_inode;
388
389 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
390 return __mnt_want_write(file->f_path.mnt);
391 else
392 return mnt_clone_write(file->f_path.mnt);
393}
394
395/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000396 * mnt_want_write_file - get write access to a file's mount
397 * @file: the file who's mount on which to take a write
398 *
399 * This is like mnt_want_write, but it takes a file and can
400 * do some optimisations if the file is open for write already
401 */
402int mnt_want_write_file(struct file *file)
403{
Jan Karaeb04c282012-06-12 16:20:35 +0200404 int ret;
405
406 sb_start_write(file->f_path.mnt->mnt_sb);
407 ret = __mnt_want_write_file(file);
408 if (ret)
409 sb_end_write(file->f_path.mnt->mnt_sb);
410 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000411}
412EXPORT_SYMBOL_GPL(mnt_want_write_file);
413
414/**
Jan Karaeb04c282012-06-12 16:20:35 +0200415 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800416 * @mnt: the mount on which to give up write access
417 *
418 * Tells the low-level filesystem that we are done
419 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200420 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800421 */
Jan Karaeb04c282012-06-12 16:20:35 +0200422void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800423{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000424 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500425 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000426 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800427}
Jan Karaeb04c282012-06-12 16:20:35 +0200428
429/**
430 * mnt_drop_write - give up write access to a mount
431 * @mnt: the mount on which to give up write access
432 *
433 * Tells the low-level filesystem that we are done performing writes to it and
434 * also allows filesystem to be frozen again. Must be matched with
435 * mnt_want_write() call above.
436 */
437void mnt_drop_write(struct vfsmount *mnt)
438{
439 __mnt_drop_write(mnt);
440 sb_end_write(mnt->mnt_sb);
441}
Dave Hansen83660252008-02-15 14:37:30 -0800442EXPORT_SYMBOL_GPL(mnt_drop_write);
443
Jan Karaeb04c282012-06-12 16:20:35 +0200444void __mnt_drop_write_file(struct file *file)
445{
446 __mnt_drop_write(file->f_path.mnt);
447}
448
Al Viro2a79f172011-12-09 08:06:57 -0500449void mnt_drop_write_file(struct file *file)
450{
451 mnt_drop_write(file->f_path.mnt);
452}
453EXPORT_SYMBOL(mnt_drop_write_file);
454
Al Viro83adc752011-11-24 22:37:54 -0500455static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800456{
Dave Hansen3d733632008-02-15 14:37:59 -0800457 int ret = 0;
458
Andi Kleen962830d2012-05-08 13:32:02 +0930459 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500460 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000461 /*
462 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
463 * should be visible before we do.
464 */
465 smp_mb();
466
467 /*
468 * With writers on hold, if this value is zero, then there are
469 * definitely no active writers (although held writers may subsequently
470 * increment the count, they'll have to wait, and decrement it after
471 * seeing MNT_READONLY).
472 *
473 * It is OK to have counter incremented on one CPU and decremented on
474 * another: the sum will add up correctly. The danger would be when we
475 * sum up each counter, if we read a counter before it is incremented,
476 * but then read another CPU's count which it has been subsequently
477 * decremented from -- we would see more decrements than we should.
478 * MNT_WRITE_HOLD protects against this scenario, because
479 * mnt_want_write first increments count, then smp_mb, then spins on
480 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
481 * we're counting up here.
482 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100483 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000484 ret = -EBUSY;
485 else
Al Viro83adc752011-11-24 22:37:54 -0500486 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000487 /*
488 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
489 * that become unheld will see MNT_READONLY.
490 */
491 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500492 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Andi Kleen962830d2012-05-08 13:32:02 +0930493 br_write_unlock(&vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800494 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800495}
Dave Hansen83660252008-02-15 14:37:30 -0800496
Al Viro83adc752011-11-24 22:37:54 -0500497static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800498{
Andi Kleen962830d2012-05-08 13:32:02 +0930499 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500500 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Andi Kleen962830d2012-05-08 13:32:02 +0930501 br_write_unlock(&vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800502}
503
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100504int sb_prepare_remount_readonly(struct super_block *sb)
505{
506 struct mount *mnt;
507 int err = 0;
508
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100509 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
510 if (atomic_long_read(&sb->s_remove_count))
511 return -EBUSY;
512
Andi Kleen962830d2012-05-08 13:32:02 +0930513 br_write_lock(&vfsmount_lock);
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100514 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
515 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
516 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
517 smp_mb();
518 if (mnt_get_writers(mnt) > 0) {
519 err = -EBUSY;
520 break;
521 }
522 }
523 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100524 if (!err && atomic_long_read(&sb->s_remove_count))
525 err = -EBUSY;
526
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100527 if (!err) {
528 sb->s_readonly_remount = 1;
529 smp_wmb();
530 }
531 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
532 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
533 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
534 }
Andi Kleen962830d2012-05-08 13:32:02 +0930535 br_write_unlock(&vfsmount_lock);
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100536
537 return err;
538}
539
Al Virob105e272011-11-24 20:38:33 -0500540static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541{
Al Viro52ba1622011-11-25 02:25:17 -0500542 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100543 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000544#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500545 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000546#endif
Al Virob105e272011-11-24 20:38:33 -0500547 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548}
549
550/*
Ram Paia05964f2005-11-07 17:20:17 -0500551 * find the first or last mount at @dentry on vfsmount @mnt depending on
552 * @dir. If @dir is set return the first mount else return the last mount.
Nick Piggin99b7db72010-08-18 04:37:39 +1000553 * vfsmount_lock must be held for read or write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 */
Al Viroc7105362011-11-24 18:22:03 -0500555struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
Ram Paia05964f2005-11-07 17:20:17 -0500556 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557{
Ram Paib58fed82005-11-07 17:16:09 -0500558 struct list_head *head = mount_hashtable + hash(mnt, dentry);
559 struct list_head *tmp = head;
Al Viroc7105362011-11-24 18:22:03 -0500560 struct mount *p, *found = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500563 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 p = NULL;
565 if (tmp == head)
566 break;
Al Viro1b8e5562011-11-24 21:01:32 -0500567 p = list_entry(tmp, struct mount, mnt_hash);
Al Viroa73324d2011-11-24 22:25:07 -0500568 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500569 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 break;
571 }
572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 return found;
574}
575
Ram Paia05964f2005-11-07 17:20:17 -0500576/*
David Howellsf015f1262012-06-25 12:55:28 +0100577 * lookup_mnt - Return the first child mount mounted at path
578 *
579 * "First" means first mounted chronologically. If you create the
580 * following mounts:
581 *
582 * mount /dev/sda1 /mnt
583 * mount /dev/sda2 /mnt
584 * mount /dev/sda3 /mnt
585 *
586 * Then lookup_mnt() on the base /mnt dentry in the root mount will
587 * return successively the root dentry and vfsmount of /dev/sda1, then
588 * /dev/sda2, then /dev/sda3, then NULL.
589 *
590 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500591 */
Al Viro1c755af2009-04-18 14:06:57 -0400592struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500593{
Al Viroc7105362011-11-24 18:22:03 -0500594 struct mount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000595
Andi Kleen962830d2012-05-08 13:32:02 +0930596 br_read_lock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500597 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
598 if (child_mnt) {
599 mnt_add_count(child_mnt, 1);
Andi Kleen962830d2012-05-08 13:32:02 +0930600 br_read_unlock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500601 return &child_mnt->mnt;
602 } else {
Andi Kleen962830d2012-05-08 13:32:02 +0930603 br_read_unlock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500604 return NULL;
605 }
Ram Paia05964f2005-11-07 17:20:17 -0500606}
607
Al Viro143c8c92011-11-25 00:46:35 -0500608static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800610 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611}
612
Nick Piggin99b7db72010-08-18 04:37:39 +1000613/*
614 * vfsmount lock must be held for write
615 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800616static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500617{
618 if (ns) {
619 ns->event = ++event;
620 wake_up_interruptible(&ns->poll);
621 }
622}
623
Nick Piggin99b7db72010-08-18 04:37:39 +1000624/*
625 * vfsmount lock must be held for write
626 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800627static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500628{
629 if (ns && ns->event != event) {
630 ns->event = event;
631 wake_up_interruptible(&ns->poll);
632 }
633}
634
Nick Piggin99b7db72010-08-18 04:37:39 +1000635/*
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100636 * Clear dentry's mounted state if it has no remaining mounts.
637 * vfsmount_lock must be held for write.
638 */
Al Viroaa0a4cf2011-11-24 19:31:36 -0500639static void dentry_reset_mounted(struct dentry *dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100640{
641 unsigned u;
642
643 for (u = 0; u < HASH_SIZE; u++) {
Al Virod5e50f72011-11-24 20:58:57 -0500644 struct mount *p;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100645
Al Viro1b8e5562011-11-24 21:01:32 -0500646 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
Al Viroa73324d2011-11-24 22:25:07 -0500647 if (p->mnt_mountpoint == dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100648 return;
649 }
650 }
651 spin_lock(&dentry->d_lock);
652 dentry->d_flags &= ~DCACHE_MOUNTED;
653 spin_unlock(&dentry->d_lock);
654}
655
656/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000657 * vfsmount lock must be held for write
658 */
Al Viro419148d2011-11-24 19:41:16 -0500659static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
Al Viroa73324d2011-11-24 22:25:07 -0500661 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500662 old_path->mnt = &mnt->mnt_parent->mnt;
663 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500664 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500665 list_del_init(&mnt->mnt_child);
Al Viro1b8e5562011-11-24 21:01:32 -0500666 list_del_init(&mnt->mnt_hash);
Al Viroaa0a4cf2011-11-24 19:31:36 -0500667 dentry_reset_mounted(old_path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668}
669
Nick Piggin99b7db72010-08-18 04:37:39 +1000670/*
671 * vfsmount lock must be held for write
672 */
Al Viro14cf1fa2011-11-25 00:01:17 -0500673void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
Al Viro44d964d62011-11-24 21:28:22 -0500674 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500675{
Al Viro3a2393d2011-11-25 03:19:09 -0500676 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viroa73324d2011-11-24 22:25:07 -0500677 child_mnt->mnt_mountpoint = dget(dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500678 child_mnt->mnt_parent = mnt;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100679 spin_lock(&dentry->d_lock);
680 dentry->d_flags |= DCACHE_MOUNTED;
681 spin_unlock(&dentry->d_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500682}
683
Nick Piggin99b7db72010-08-18 04:37:39 +1000684/*
685 * vfsmount lock must be held for write
686 */
Al Viro419148d2011-11-24 19:41:16 -0500687static void attach_mnt(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688{
Al Viro14cf1fa2011-11-25 00:01:17 -0500689 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
Al Viro1b8e5562011-11-24 21:01:32 -0500690 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400691 hash(path->mnt, path->dentry));
Al Viro6b41d532011-11-24 23:24:33 -0500692 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500693}
694
695/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000696 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500697 */
Al Viro4b2619a2011-11-24 19:47:15 -0500698static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500699{
Al Viro0714a532011-11-24 22:19:58 -0500700 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500701 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500702 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500703 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500704
Al Viro0714a532011-11-24 22:19:58 -0500705 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500706
Al Viro1a4eeaf2011-11-25 02:19:55 -0500707 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400708 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500709 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500710
Ram Paib90fa9a2005-11-07 17:19:50 -0500711 list_splice(&head, n->list.prev);
712
Al Viro1b8e5562011-11-24 21:01:32 -0500713 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viroa73324d2011-11-24 22:25:07 -0500714 hash(&parent->mnt, mnt->mnt_mountpoint));
Al Viro6b41d532011-11-24 23:24:33 -0500715 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800716 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
Al Viro909b0a82011-11-25 03:06:56 -0500719static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720{
Al Viro6b41d532011-11-24 23:24:33 -0500721 struct list_head *next = p->mnt_mounts.next;
722 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500724 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500726 next = p->mnt_child.next;
727 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 break;
Al Viro0714a532011-11-24 22:19:58 -0500729 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 }
731 }
Al Viro6b41d532011-11-24 23:24:33 -0500732 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
734
Al Viro315fc832011-11-24 18:57:30 -0500735static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500736{
Al Viro6b41d532011-11-24 23:24:33 -0500737 struct list_head *prev = p->mnt_mounts.prev;
738 while (prev != &p->mnt_mounts) {
739 p = list_entry(prev, struct mount, mnt_child);
740 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500741 }
742 return p;
743}
744
Al Viro9d412a42011-03-17 22:08:28 -0400745struct vfsmount *
746vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
747{
Al Virob105e272011-11-24 20:38:33 -0500748 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400749 struct dentry *root;
750
751 if (!type)
752 return ERR_PTR(-ENODEV);
753
754 mnt = alloc_vfsmnt(name);
755 if (!mnt)
756 return ERR_PTR(-ENOMEM);
757
758 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500759 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400760
761 root = mount_fs(type, flags, name, data);
762 if (IS_ERR(root)) {
763 free_vfsmnt(mnt);
764 return ERR_CAST(root);
765 }
766
Al Virob105e272011-11-24 20:38:33 -0500767 mnt->mnt.mnt_root = root;
768 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500769 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500770 mnt->mnt_parent = mnt;
Andi Kleen962830d2012-05-08 13:32:02 +0930771 br_write_lock(&vfsmount_lock);
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100772 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Andi Kleen962830d2012-05-08 13:32:02 +0930773 br_write_unlock(&vfsmount_lock);
Al Virob105e272011-11-24 20:38:33 -0500774 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400775}
776EXPORT_SYMBOL_GPL(vfs_kern_mount);
777
Al Viro87129cc2011-11-24 21:24:27 -0500778static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500779 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780{
Al Viro87129cc2011-11-24 21:24:27 -0500781 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100782 struct mount *mnt;
783 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
David Howellsbe34d1a2012-06-25 12:55:18 +0100785 mnt = alloc_vfsmnt(old->mnt_devname);
786 if (!mnt)
787 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100788
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700789 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +0100790 mnt->mnt_group_id = 0; /* not a peer of original */
791 else
792 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100793
David Howellsbe34d1a2012-06-25 12:55:18 +0100794 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
795 err = mnt_alloc_group_id(mnt);
796 if (err)
797 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100799
800 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
801 atomic_inc(&sb->s_active);
802 mnt->mnt.mnt_sb = sb;
803 mnt->mnt.mnt_root = dget(root);
804 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
805 mnt->mnt_parent = mnt;
806 br_write_lock(&vfsmount_lock);
807 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
808 br_write_unlock(&vfsmount_lock);
809
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700810 if ((flag & CL_SLAVE) ||
811 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +0100812 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
813 mnt->mnt_master = old;
814 CLEAR_MNT_SHARED(mnt);
815 } else if (!(flag & CL_PRIVATE)) {
816 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
817 list_add(&mnt->mnt_share, &old->mnt_share);
818 if (IS_MNT_SLAVE(old))
819 list_add(&mnt->mnt_slave, &old->mnt_slave);
820 mnt->mnt_master = old->mnt_master;
821 }
822 if (flag & CL_MAKE_SHARED)
823 set_mnt_shared(mnt);
824
825 /* stick the duplicate mount on the same expiry list
826 * as the original if that was on one */
827 if (flag & CL_EXPIRE) {
828 if (!list_empty(&old->mnt_expire))
829 list_add(&mnt->mnt_expire, &old->mnt_expire);
830 }
831
Al Virocb338d02011-11-24 20:55:08 -0500832 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100833
834 out_free:
835 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +0100836 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837}
838
Al Viro83adc752011-11-24 22:37:54 -0500839static inline void mntfree(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840{
Al Viro83adc752011-11-24 22:37:54 -0500841 struct vfsmount *m = &mnt->mnt;
842 struct super_block *sb = m->mnt_sb;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100843
Dave Hansen3d733632008-02-15 14:37:59 -0800844 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800845 * This probably indicates that somebody messed
846 * up a mnt_want/drop_write() pair. If this
847 * happens, the filesystem was probably unable
848 * to make r/w->r/o transitions.
849 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000850 /*
Nick Pigginb3e19d92011-01-07 17:50:11 +1100851 * The locking used to deal with mnt_count decrement provides barriers,
852 * so mnt_get_writers() below is safe.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000853 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100854 WARN_ON(mnt_get_writers(mnt));
Al Viro83adc752011-11-24 22:37:54 -0500855 fsnotify_vfsmount_delete(m);
856 dput(m->mnt_root);
857 free_vfsmnt(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 deactivate_super(sb);
859}
860
Al Viro900148d2011-11-25 00:33:11 -0500861static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500862{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100863put_again:
Al Virof03c6592011-01-14 22:30:21 -0500864#ifdef CONFIG_SMP
Andi Kleen962830d2012-05-08 13:32:02 +0930865 br_read_lock(&vfsmount_lock);
Al Virof7a99c52012-06-09 00:59:08 -0400866 if (likely(mnt->mnt_ns)) {
867 /* shouldn't be the last one */
Al Viroaa9c0e02011-11-23 19:04:52 -0500868 mnt_add_count(mnt, -1);
Andi Kleen962830d2012-05-08 13:32:02 +0930869 br_read_unlock(&vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500870 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100871 }
Andi Kleen962830d2012-05-08 13:32:02 +0930872 br_read_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100873
Andi Kleen962830d2012-05-08 13:32:02 +0930874 br_write_lock(&vfsmount_lock);
Al Viroaa9c0e02011-11-23 19:04:52 -0500875 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100876 if (mnt_get_count(mnt)) {
Andi Kleen962830d2012-05-08 13:32:02 +0930877 br_write_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100878 return;
879 }
Nick Pigginb3e19d92011-01-07 17:50:11 +1100880#else
Al Viroaa9c0e02011-11-23 19:04:52 -0500881 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100882 if (likely(mnt_get_count(mnt)))
Nick Piggin99b7db72010-08-18 04:37:39 +1000883 return;
Andi Kleen962830d2012-05-08 13:32:02 +0930884 br_write_lock(&vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500885#endif
Al Viro863d6842011-11-25 00:57:42 -0500886 if (unlikely(mnt->mnt_pinned)) {
887 mnt_add_count(mnt, mnt->mnt_pinned + 1);
888 mnt->mnt_pinned = 0;
Andi Kleen962830d2012-05-08 13:32:02 +0930889 br_write_unlock(&vfsmount_lock);
Al Viro900148d2011-11-25 00:33:11 -0500890 acct_auto_close_mnt(&mnt->mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100891 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500892 }
Andi Kleen962830d2012-05-08 13:32:02 +0930893
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100894 list_del(&mnt->mnt_instance);
Andi Kleen962830d2012-05-08 13:32:02 +0930895 br_write_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100896 mntfree(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500897}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100898
899void mntput(struct vfsmount *mnt)
900{
901 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -0500902 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100903 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -0500904 if (unlikely(m->mnt_expiry_mark))
905 m->mnt_expiry_mark = 0;
906 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100907 }
908}
909EXPORT_SYMBOL(mntput);
910
911struct vfsmount *mntget(struct vfsmount *mnt)
912{
913 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -0500914 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100915 return mnt;
916}
917EXPORT_SYMBOL(mntget);
918
Al Viro7b7b1ac2005-11-07 17:13:39 -0500919void mnt_pin(struct vfsmount *mnt)
920{
Andi Kleen962830d2012-05-08 13:32:02 +0930921 br_write_lock(&vfsmount_lock);
Al Viro863d6842011-11-25 00:57:42 -0500922 real_mount(mnt)->mnt_pinned++;
Andi Kleen962830d2012-05-08 13:32:02 +0930923 br_write_unlock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500924}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500925EXPORT_SYMBOL(mnt_pin);
926
Al Viro863d6842011-11-25 00:57:42 -0500927void mnt_unpin(struct vfsmount *m)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500928{
Al Viro863d6842011-11-25 00:57:42 -0500929 struct mount *mnt = real_mount(m);
Andi Kleen962830d2012-05-08 13:32:02 +0930930 br_write_lock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500931 if (mnt->mnt_pinned) {
Al Viro863d6842011-11-25 00:57:42 -0500932 mnt_add_count(mnt, 1);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500933 mnt->mnt_pinned--;
934 }
Andi Kleen962830d2012-05-08 13:32:02 +0930935 br_write_unlock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500936}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500937EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800939static inline void mangle(struct seq_file *m, const char *s)
940{
941 seq_escape(m, s, " \t\n\\");
942}
943
944/*
945 * Simple .show_options callback for filesystems which don't want to
946 * implement more complex mount option showing.
947 *
948 * See also save_mount_options().
949 */
Al Viro34c80b12011-12-08 21:32:45 -0500950int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800951{
Al Viro2a32ceb2009-05-08 16:05:57 -0400952 const char *options;
953
954 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -0500955 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800956
957 if (options != NULL && options[0]) {
958 seq_putc(m, ',');
959 mangle(m, options);
960 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400961 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800962
963 return 0;
964}
965EXPORT_SYMBOL(generic_show_options);
966
967/*
968 * If filesystem uses generic_show_options(), this function should be
969 * called from the fill_super() callback.
970 *
971 * The .remount_fs callback usually needs to be handled in a special
972 * way, to make sure, that previous options are not overwritten if the
973 * remount fails.
974 *
975 * Also note, that if the filesystem's .remount_fs function doesn't
976 * reset all options to their default value, but changes only newly
977 * given options, then the displayed options will not reflect reality
978 * any more.
979 */
980void save_mount_options(struct super_block *sb, char *options)
981{
Al Viro2a32ceb2009-05-08 16:05:57 -0400982 BUG_ON(sb->s_options);
983 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800984}
985EXPORT_SYMBOL(save_mount_options);
986
Al Viro2a32ceb2009-05-08 16:05:57 -0400987void replace_mount_options(struct super_block *sb, char *options)
988{
989 char *old = sb->s_options;
990 rcu_assign_pointer(sb->s_options, options);
991 if (old) {
992 synchronize_rcu();
993 kfree(old);
994 }
995}
996EXPORT_SYMBOL(replace_mount_options);
997
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100998#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -0500999/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000static void *m_start(struct seq_file *m, loff_t *pos)
1001{
Al Viro6ce6e242012-06-09 01:16:59 -04001002 struct proc_mounts *p = proc_mounts(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
Ram Pai390c6842005-11-07 17:17:51 -05001004 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001005 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006}
1007
1008static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1009{
Al Viro6ce6e242012-06-09 01:16:59 -04001010 struct proc_mounts *p = proc_mounts(m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001011
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001012 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013}
1014
1015static void m_stop(struct seq_file *m, void *v)
1016{
Ram Pai390c6842005-11-07 17:17:51 -05001017 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018}
1019
Al Viro0226f492011-12-06 12:21:54 -05001020static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001021{
Al Viro6ce6e242012-06-09 01:16:59 -04001022 struct proc_mounts *p = proc_mounts(m);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001023 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001024 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025}
1026
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001027const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 .start = m_start,
1029 .next = m_next,
1030 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001031 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001032};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001033#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001034
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035/**
1036 * may_umount_tree - check if a mount tree is busy
1037 * @mnt: root of mount tree
1038 *
1039 * This is called to check if a tree of mounts has any
1040 * open files, pwds, chroots or sub mounts that are
1041 * busy.
1042 */
Al Viro909b0a82011-11-25 03:06:56 -05001043int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044{
Al Viro909b0a82011-11-25 03:06:56 -05001045 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001046 int actual_refs = 0;
1047 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001048 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001049 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050
Nick Pigginb3e19d92011-01-07 17:50:11 +11001051 /* write lock needed for mnt_get_count */
Andi Kleen962830d2012-05-08 13:32:02 +09301052 br_write_lock(&vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -05001053 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001054 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 }
Andi Kleen962830d2012-05-08 13:32:02 +09301057 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
1059 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001060 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
Ian Kente3474a82006-03-27 01:14:51 -08001062 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063}
1064
1065EXPORT_SYMBOL(may_umount_tree);
1066
1067/**
1068 * may_umount - check if a mount point is busy
1069 * @mnt: root of mount
1070 *
1071 * This is called to check if a mount point has any
1072 * open files, pwds, chroots or sub mounts. If the
1073 * mount has sub mounts this will return busy
1074 * regardless of whether the sub mounts are busy.
1075 *
1076 * Doesn't take quota and stuff into account. IOW, in some cases it will
1077 * give false negatives. The main reason why it's here is that we need
1078 * a non-destructive way to look for easily umountable filesystems.
1079 */
1080int may_umount(struct vfsmount *mnt)
1081{
Ian Kente3474a82006-03-27 01:14:51 -08001082 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001083 down_read(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301084 br_write_lock(&vfsmount_lock);
Al Viro1ab59732011-11-24 21:35:16 -05001085 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001086 ret = 0;
Andi Kleen962830d2012-05-08 13:32:02 +09301087 br_write_unlock(&vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001088 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001089 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090}
1091
1092EXPORT_SYMBOL(may_umount);
1093
Ram Paib90fa9a2005-11-07 17:19:50 -05001094void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095{
Al Virod5e50f72011-11-24 20:58:57 -05001096 struct mount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001097 while (!list_empty(head)) {
Al Viro1b8e5562011-11-24 21:01:32 -05001098 mnt = list_first_entry(head, struct mount, mnt_hash);
1099 list_del_init(&mnt->mnt_hash);
Al Viro676da582011-11-24 21:47:05 -05001100 if (mnt_has_parent(mnt)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001101 struct dentry *dentry;
Al Viro863d6842011-11-25 00:57:42 -05001102 struct mount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001103
Andi Kleen962830d2012-05-08 13:32:02 +09301104 br_write_lock(&vfsmount_lock);
Al Viroa73324d2011-11-24 22:25:07 -05001105 dentry = mnt->mnt_mountpoint;
Al Viro863d6842011-11-25 00:57:42 -05001106 m = mnt->mnt_parent;
Al Viroa73324d2011-11-24 22:25:07 -05001107 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -05001108 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001109 m->mnt_ghosts--;
Andi Kleen962830d2012-05-08 13:32:02 +09301110 br_write_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001111 dput(dentry);
Al Viro863d6842011-11-25 00:57:42 -05001112 mntput(&m->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 }
Al Virod5e50f72011-11-24 20:58:57 -05001114 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001115 }
1116}
1117
Nick Piggin99b7db72010-08-18 04:37:39 +10001118/*
1119 * vfsmount lock must be held for write
1120 * namespace_sem must be held for write
1121 */
Al Viro761d5c32011-11-24 21:07:43 -05001122void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001123{
Al Viro7b8a53f2011-01-15 20:08:44 -05001124 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001125 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001126
Al Viro909b0a82011-11-25 03:06:56 -05001127 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro1b8e5562011-11-24 21:01:32 -05001128 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001129
Ram Paia05964f2005-11-07 17:20:17 -05001130 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001131 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001132
Al Viro1b8e5562011-11-24 21:01:32 -05001133 list_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001134 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001135 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001136 __touch_mnt_namespace(p->mnt_ns);
1137 p->mnt_ns = NULL;
Al Viro6b41d532011-11-24 23:24:33 -05001138 list_del_init(&p->mnt_child);
Al Viro676da582011-11-24 21:47:05 -05001139 if (mnt_has_parent(p)) {
Al Viro863d6842011-11-25 00:57:42 -05001140 p->mnt_parent->mnt_ghosts++;
Al Viroa73324d2011-11-24 22:25:07 -05001141 dentry_reset_mounted(p->mnt_mountpoint);
Al Viro7c4b93d2008-03-21 23:59:49 -04001142 }
Al Viro0f0afb12011-11-24 20:43:10 -05001143 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 }
Al Viro7b8a53f2011-01-15 20:08:44 -05001145 list_splice(&tmp_list, kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146}
1147
Al Viro692afc32011-11-24 21:15:14 -05001148static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
Al Viroc35038b2008-03-22 00:46:23 -04001149
Al Viro1ab59732011-11-24 21:35:16 -05001150static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
Al Viro1ab59732011-11-24 21:35:16 -05001152 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001154 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
Al Viro1ab59732011-11-24 21:35:16 -05001156 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 if (retval)
1158 return retval;
1159
1160 /*
1161 * Allow userspace to request a mountpoint be expired rather than
1162 * unmounting unconditionally. Unmount only happens if:
1163 * (1) the mark is already set (the mark is cleared by mntput())
1164 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1165 */
1166 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001167 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 flags & (MNT_FORCE | MNT_DETACH))
1169 return -EINVAL;
1170
Nick Pigginb3e19d92011-01-07 17:50:11 +11001171 /*
1172 * probably don't strictly need the lock here if we examined
1173 * all race cases, but it's a slowpath.
1174 */
Andi Kleen962830d2012-05-08 13:32:02 +09301175 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -05001176 if (mnt_get_count(mnt) != 2) {
Andi Kleen962830d2012-05-08 13:32:02 +09301177 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001179 }
Andi Kleen962830d2012-05-08 13:32:02 +09301180 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
Al Viro863d6842011-11-25 00:57:42 -05001182 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 return -EAGAIN;
1184 }
1185
1186 /*
1187 * If we may have to abort operations to get out of this
1188 * mount, and they will themselves hold resources we must
1189 * allow the fs to do things. In the Unix tradition of
1190 * 'Gee thats tricky lets do it in userspace' the umount_begin
1191 * might fail to complete on the first run through as other tasks
1192 * must return, and the like. Thats for the mount program to worry
1193 * about for the moment.
1194 */
1195
Al Viro42faad92008-04-24 07:21:56 -04001196 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001197 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001198 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
1200 /*
1201 * No sense to grab the lock for this test, but test itself looks
1202 * somewhat bogus. Suggestions for better replacement?
1203 * Ho-hum... In principle, we might treat that as umount + switch
1204 * to rootfs. GC would eventually take care of the old vfsmount.
1205 * Actually it makes sense, especially if rootfs would contain a
1206 * /reboot - static binary that would close all descriptors and
1207 * call reboot(9). Then init(8) could umount root and exec /reboot.
1208 */
Al Viro1ab59732011-11-24 21:35:16 -05001209 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 /*
1211 * Special case for "unmounting" root ...
1212 * we just try to remount it readonly.
1213 */
1214 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001215 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 up_write(&sb->s_umount);
1218 return retval;
1219 }
1220
Ram Pai390c6842005-11-07 17:17:51 -05001221 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301222 br_write_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001223 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224
Al Viroc35038b2008-03-22 00:46:23 -04001225 if (!(flags & MNT_DETACH))
Al Viro1ab59732011-11-24 21:35:16 -05001226 shrink_submounts(mnt, &umount_list);
Al Viroc35038b2008-03-22 00:46:23 -04001227
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001229 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001230 if (!list_empty(&mnt->mnt_list))
Al Viro1ab59732011-11-24 21:35:16 -05001231 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 retval = 0;
1233 }
Andi Kleen962830d2012-05-08 13:32:02 +09301234 br_write_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001235 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001236 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 return retval;
1238}
1239
Al Viro9b40bc92013-02-22 22:45:42 -05001240/*
1241 * Is the caller allowed to modify his namespace?
1242 */
1243static inline bool may_mount(void)
1244{
1245 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1246}
1247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248/*
1249 * Now umount can handle mount points as well as block devices.
1250 * This is important for filesystems which use unnamed block devices.
1251 *
1252 * We now support a flag for forced unmount like the other 'big iron'
1253 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1254 */
1255
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001256SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257{
Al Viro2d8f3032008-07-22 09:59:21 -04001258 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001259 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001261 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001263 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1264 return -EINVAL;
1265
Al Viro9b40bc92013-02-22 22:45:42 -05001266 if (!may_mount())
1267 return -EPERM;
1268
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001269 if (!(flags & UMOUNT_NOFOLLOW))
1270 lookup_flags |= LOOKUP_FOLLOW;
1271
1272 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 if (retval)
1274 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001275 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001277 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001279 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 goto dput_and_out;
1281
Al Viro900148d2011-11-25 00:33:11 -05001282 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001284 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001285 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001286 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287out:
1288 return retval;
1289}
1290
1291#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1292
1293/*
Ram Paib58fed82005-11-07 17:16:09 -05001294 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001296SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297{
Ram Paib58fed82005-11-07 17:16:09 -05001298 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299}
1300
1301#endif
1302
Al Viro2d92ab32008-08-02 00:51:11 -04001303static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304{
Al Viro9b40bc92013-02-22 22:45:42 -05001305 if (may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 return 0;
1307 return -EPERM;
1308#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001309 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001311 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001312 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 return -EPERM;
1314 }
Al Viro2d92ab32008-08-02 00:51:11 -04001315 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 return -EPERM;
1317 return 0;
1318#endif
1319}
1320
Eric W. Biederman8823c072010-03-07 18:49:36 -08001321static bool mnt_ns_loop(struct path *path)
1322{
1323 /* Could bind mounting the mount namespace inode cause a
1324 * mount namespace loop?
1325 */
1326 struct inode *inode = path->dentry->d_inode;
1327 struct proc_inode *ei;
1328 struct mnt_namespace *mnt_ns;
1329
1330 if (!proc_ns_inode(inode))
1331 return false;
1332
1333 ei = PROC_I(inode);
1334 if (ei->ns_ops != &mntns_operations)
1335 return false;
1336
1337 mnt_ns = ei->ns;
1338 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1339}
1340
Al Viro87129cc2011-11-24 21:24:27 -05001341struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001342 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343{
Al Viroa73324d2011-11-24 22:25:07 -05001344 struct mount *res, *p, *q, *r;
Al Viro1a390682008-03-21 20:48:19 -04001345 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
Al Virofc7be132011-11-25 01:05:37 -05001347 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
David Howellsbe34d1a2012-06-25 12:55:18 +01001348 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001349
Ram Pai36341f62005-11-07 17:17:22 -05001350 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001351 if (IS_ERR(q))
1352 return q;
1353
Al Viroa73324d2011-11-24 22:25:07 -05001354 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
1356 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001357 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001358 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001359 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 continue;
1361
Al Viro909b0a82011-11-25 03:06:56 -05001362 for (s = r; s; s = next_mnt(s, r)) {
Al Virofc7be132011-11-25 01:05:37 -05001363 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001364 s = skip_mnt_tree(s);
1365 continue;
1366 }
Al Viro0714a532011-11-24 22:19:58 -05001367 while (p != s->mnt_parent) {
1368 p = p->mnt_parent;
1369 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 }
Al Viro87129cc2011-11-24 21:24:27 -05001371 p = s;
Al Virocb338d02011-11-24 20:55:08 -05001372 path.mnt = &q->mnt;
Al Viroa73324d2011-11-24 22:25:07 -05001373 path.dentry = p->mnt_mountpoint;
Al Viro87129cc2011-11-24 21:24:27 -05001374 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001375 if (IS_ERR(q))
1376 goto out;
Andi Kleen962830d2012-05-08 13:32:02 +09301377 br_write_lock(&vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001378 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Virocb338d02011-11-24 20:55:08 -05001379 attach_mnt(q, &path);
Andi Kleen962830d2012-05-08 13:32:02 +09301380 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 }
1382 }
1383 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001384out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001386 LIST_HEAD(umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301387 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001388 umount_tree(res, 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301389 br_write_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001390 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001392 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393}
1394
David Howellsbe34d1a2012-06-25 12:55:18 +01001395/* Caller should check returned pointer for errors */
1396
Al Viro589ff872009-04-18 03:28:19 -04001397struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001398{
Al Virocb338d02011-11-24 20:55:08 -05001399 struct mount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001400 down_write(&namespace_sem);
Al Viro87129cc2011-11-24 21:24:27 -05001401 tree = copy_tree(real_mount(path->mnt), path->dentry,
1402 CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001403 up_write(&namespace_sem);
David Howellsbe34d1a2012-06-25 12:55:18 +01001404 if (IS_ERR(tree))
1405 return NULL;
1406 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001407}
1408
1409void drop_collected_mounts(struct vfsmount *mnt)
1410{
1411 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001412 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301413 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001414 umount_tree(real_mount(mnt), 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301415 br_write_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001416 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001417 release_mounts(&umount_list);
1418}
1419
Al Viro1f707132010-01-30 22:51:25 -05001420int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1421 struct vfsmount *root)
1422{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001423 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001424 int res = f(root, arg);
1425 if (res)
1426 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001427 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1428 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001429 if (res)
1430 return res;
1431 }
1432 return 0;
1433}
1434
Al Viro4b8b21f2011-11-24 19:54:23 -05001435static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001436{
Al Viro315fc832011-11-24 18:57:30 -05001437 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001438
Al Viro909b0a82011-11-25 03:06:56 -05001439 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001440 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001441 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001442 }
1443}
1444
Al Viro4b8b21f2011-11-24 19:54:23 -05001445static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001446{
Al Viro315fc832011-11-24 18:57:30 -05001447 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001448
Al Viro909b0a82011-11-25 03:06:56 -05001449 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001450 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001451 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001452 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001453 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001454 return err;
1455 }
1456 }
1457 }
1458
1459 return 0;
1460}
1461
Ram Paib90fa9a2005-11-07 17:19:50 -05001462/*
1463 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001464 * @nd : place the mount tree @source_mnt is attached
1465 * @parent_nd : if non-null, detach the source_mnt from its parent and
1466 * store the parent mount and mountpoint dentry.
1467 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001468 *
1469 * NOTE: in the table below explains the semantics when a source mount
1470 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001471 * ---------------------------------------------------------------------------
1472 * | BIND MOUNT OPERATION |
1473 * |**************************************************************************
1474 * | source-->| shared | private | slave | unbindable |
1475 * | dest | | | | |
1476 * | | | | | | |
1477 * | v | | | | |
1478 * |**************************************************************************
1479 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1480 * | | | | | |
1481 * |non-shared| shared (+) | private | slave (*) | invalid |
1482 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001483 * A bind operation clones the source mount and mounts the clone on the
1484 * destination mount.
1485 *
1486 * (++) the cloned mount is propagated to all the mounts in the propagation
1487 * tree of the destination mount and the cloned mount is added to
1488 * the peer group of the source mount.
1489 * (+) the cloned mount is created under the destination mount and is marked
1490 * as shared. The cloned mount is added to the peer group of the source
1491 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001492 * (+++) the mount is propagated to all the mounts in the propagation tree
1493 * of the destination mount and the cloned mount is made slave
1494 * of the same master as that of the source mount. The cloned mount
1495 * is marked as 'shared and slave'.
1496 * (*) the cloned mount is made a slave of the same master as that of the
1497 * source mount.
1498 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001499 * ---------------------------------------------------------------------------
1500 * | MOVE MOUNT OPERATION |
1501 * |**************************************************************************
1502 * | source-->| shared | private | slave | unbindable |
1503 * | dest | | | | |
1504 * | | | | | | |
1505 * | v | | | | |
1506 * |**************************************************************************
1507 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1508 * | | | | | |
1509 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1510 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001511 *
1512 * (+) the mount is moved to the destination. And is then propagated to
1513 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001514 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001515 * (+++) the mount is moved to the destination and is then propagated to
1516 * all the mounts belonging to the destination mount's propagation tree.
1517 * the mount is marked as 'shared and slave'.
1518 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001519 *
1520 * if the source mount is a tree, the operations explained above is
1521 * applied to each mount in the tree.
1522 * Must be called without spinlocks held, since this function can sleep
1523 * in allocations.
1524 */
Al Viro0fb54e52011-11-24 19:59:16 -05001525static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001526 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001527{
1528 LIST_HEAD(tree_list);
Al Viroa8d56d82011-11-24 23:59:29 -05001529 struct mount *dest_mnt = real_mount(path->mnt);
Al Viro1a390682008-03-21 20:48:19 -04001530 struct dentry *dest_dentry = path->dentry;
Al Viro315fc832011-11-24 18:57:30 -05001531 struct mount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001532 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001533
Al Virofc7be132011-11-25 01:05:37 -05001534 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001535 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001536 if (err)
1537 goto out;
1538 }
Al Viroa8d56d82011-11-24 23:59:29 -05001539 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001540 if (err)
1541 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001542
Andi Kleen962830d2012-05-08 13:32:02 +09301543 br_write_lock(&vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001544
Al Virofc7be132011-11-25 01:05:37 -05001545 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro909b0a82011-11-25 03:06:56 -05001546 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001547 set_mnt_shared(p);
Ram Paib90fa9a2005-11-07 17:19:50 -05001548 }
Al Viro1a390682008-03-21 20:48:19 -04001549 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001550 detach_mnt(source_mnt, parent_path);
1551 attach_mnt(source_mnt, path);
Al Viro143c8c92011-11-25 00:46:35 -05001552 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001553 } else {
Al Viro14cf1fa2011-11-25 00:01:17 -05001554 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
Al Viro0fb54e52011-11-24 19:59:16 -05001555 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05001556 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001557
Al Viro1b8e5562011-11-24 21:01:32 -05001558 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1559 list_del_init(&child->mnt_hash);
Al Viro4b2619a2011-11-24 19:47:15 -05001560 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05001561 }
Andi Kleen962830d2012-05-08 13:32:02 +09301562 br_write_unlock(&vfsmount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +10001563
Ram Paib90fa9a2005-11-07 17:19:50 -05001564 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001565
1566 out_cleanup_ids:
Al Virofc7be132011-11-25 01:05:37 -05001567 if (IS_MNT_SHARED(dest_mnt))
Al Viro0fb54e52011-11-24 19:59:16 -05001568 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001569 out:
1570 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001571}
1572
Al Virob12cea92011-03-18 08:55:38 -04001573static int lock_mount(struct path *path)
1574{
1575 struct vfsmount *mnt;
1576retry:
1577 mutex_lock(&path->dentry->d_inode->i_mutex);
1578 if (unlikely(cant_mount(path->dentry))) {
1579 mutex_unlock(&path->dentry->d_inode->i_mutex);
1580 return -ENOENT;
1581 }
1582 down_write(&namespace_sem);
1583 mnt = lookup_mnt(path);
1584 if (likely(!mnt))
1585 return 0;
1586 up_write(&namespace_sem);
1587 mutex_unlock(&path->dentry->d_inode->i_mutex);
1588 path_put(path);
1589 path->mnt = mnt;
1590 path->dentry = dget(mnt->mnt_root);
1591 goto retry;
1592}
1593
1594static void unlock_mount(struct path *path)
1595{
1596 up_write(&namespace_sem);
1597 mutex_unlock(&path->dentry->d_inode->i_mutex);
1598}
1599
Al Viro95bc5f22011-11-25 00:30:56 -05001600static int graft_tree(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
Al Viro95bc5f22011-11-25 00:30:56 -05001602 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 return -EINVAL;
1604
Al Viro8c3ee422008-03-22 18:00:39 -04001605 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001606 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 return -ENOTDIR;
1608
Al Virob12cea92011-03-18 08:55:38 -04001609 if (d_unlinked(path->dentry))
1610 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Al Viro95bc5f22011-11-25 00:30:56 -05001612 return attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613}
1614
1615/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001616 * Sanity check the flags to change_mnt_propagation.
1617 */
1618
1619static int flags_to_propagation_type(int flags)
1620{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001621 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001622
1623 /* Fail if any non-propagation flags are set */
1624 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1625 return 0;
1626 /* Only one propagation flag should be set */
1627 if (!is_power_of_2(type))
1628 return 0;
1629 return type;
1630}
1631
1632/*
Ram Pai07b20882005-11-07 17:19:07 -05001633 * recursively change the type of the mountpoint.
1634 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001635static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001636{
Al Viro315fc832011-11-24 18:57:30 -05001637 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001638 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001639 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001640 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001641 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001642
Al Viro9b40bc92013-02-22 22:45:42 -05001643 if (!may_mount())
Miklos Szerediee6f9582007-05-08 00:30:40 -07001644 return -EPERM;
1645
Al Viro2d92ab32008-08-02 00:51:11 -04001646 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001647 return -EINVAL;
1648
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001649 type = flags_to_propagation_type(flag);
1650 if (!type)
1651 return -EINVAL;
1652
Ram Pai07b20882005-11-07 17:19:07 -05001653 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001654 if (type == MS_SHARED) {
1655 err = invent_group_ids(mnt, recurse);
1656 if (err)
1657 goto out_unlock;
1658 }
1659
Andi Kleen962830d2012-05-08 13:32:02 +09301660 br_write_lock(&vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -05001661 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001662 change_mnt_propagation(m, type);
Andi Kleen962830d2012-05-08 13:32:02 +09301663 br_write_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001664
1665 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001666 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001667 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001668}
1669
1670/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 * do loopback mount.
1672 */
Al Viro808d4e32012-10-11 11:42:01 -04001673static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001674 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
Al Virob12cea92011-03-18 08:55:38 -04001676 LIST_HEAD(umount_list);
Al Viro2d92ab32008-08-02 00:51:11 -04001677 struct path old_path;
Al Viro87129cc2011-11-24 21:24:27 -05001678 struct mount *mnt = NULL, *old;
Al Viro2d92ab32008-08-02 00:51:11 -04001679 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 if (err)
1681 return err;
1682 if (!old_name || !*old_name)
1683 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001684 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 if (err)
1686 return err;
1687
Eric W. Biederman8823c072010-03-07 18:49:36 -08001688 err = -EINVAL;
1689 if (mnt_ns_loop(&old_path))
1690 goto out;
1691
Al Virob12cea92011-03-18 08:55:38 -04001692 err = lock_mount(path);
1693 if (err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001694 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001695
Al Viro87129cc2011-11-24 21:24:27 -05001696 old = real_mount(old_path.mnt);
1697
Al Virob12cea92011-03-18 08:55:38 -04001698 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001699 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04001700 goto out2;
1701
Al Viro143c8c92011-11-25 00:46:35 -05001702 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
Al Virob12cea92011-03-18 08:55:38 -04001703 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
Al Viroccd48bc2005-11-07 17:15:04 -05001705 if (recurse)
Al Viro87129cc2011-11-24 21:24:27 -05001706 mnt = copy_tree(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001707 else
Al Viro87129cc2011-11-24 21:24:27 -05001708 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001709
David Howellsbe34d1a2012-06-25 12:55:18 +01001710 if (IS_ERR(mnt)) {
1711 err = PTR_ERR(mnt);
1712 goto out;
1713 }
Al Viroccd48bc2005-11-07 17:15:04 -05001714
Al Viro95bc5f22011-11-25 00:30:56 -05001715 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001716 if (err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301717 br_write_lock(&vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001718 umount_tree(mnt, 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09301719 br_write_unlock(&vfsmount_lock);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001720 }
Al Virob12cea92011-03-18 08:55:38 -04001721out2:
1722 unlock_mount(path);
1723 release_mounts(&umount_list);
Al Viroccd48bc2005-11-07 17:15:04 -05001724out:
Al Viro2d92ab32008-08-02 00:51:11 -04001725 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 return err;
1727}
1728
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001729static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1730{
1731 int error = 0;
1732 int readonly_request = 0;
1733
1734 if (ms_flags & MS_RDONLY)
1735 readonly_request = 1;
1736 if (readonly_request == __mnt_is_readonly(mnt))
1737 return 0;
1738
1739 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05001740 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001741 else
Al Viro83adc752011-11-24 22:37:54 -05001742 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001743 return error;
1744}
1745
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746/*
1747 * change filesystem flags. dir should be a physical root of filesystem.
1748 * If you've mounted a non-root directory somewhere and want to do remount
1749 * on it - tough luck.
1750 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001751static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 void *data)
1753{
1754 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001755 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05001756 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
1758 if (!capable(CAP_SYS_ADMIN))
1759 return -EPERM;
1760
Al Viro143c8c92011-11-25 00:46:35 -05001761 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 return -EINVAL;
1763
Al Viro2d92ab32008-08-02 00:51:11 -04001764 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 return -EINVAL;
1766
Eric Parisff36fe22011-03-03 16:09:14 -05001767 err = security_sb_remount(sb, data);
1768 if (err)
1769 return err;
1770
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001772 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001773 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001774 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001775 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001776 if (!err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301777 br_write_lock(&vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001778 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1779 mnt->mnt.mnt_flags = mnt_flags;
Andi Kleen962830d2012-05-08 13:32:02 +09301780 br_write_unlock(&vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001783 if (!err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301784 br_write_lock(&vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001785 touch_mnt_namespace(mnt->mnt_ns);
Andi Kleen962830d2012-05-08 13:32:02 +09301786 br_write_unlock(&vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 return err;
1789}
1790
Al Virocbbe3622011-11-24 20:01:19 -05001791static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05001792{
Al Viro315fc832011-11-24 18:57:30 -05001793 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001794 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001795 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05001796 return 1;
1797 }
1798 return 0;
1799}
1800
Al Viro808d4e32012-10-11 11:42:01 -04001801static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802{
Al Viro2d92ab32008-08-02 00:51:11 -04001803 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05001804 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05001805 struct mount *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 int err = 0;
Al Viro9b40bc92013-02-22 22:45:42 -05001807 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 return -EPERM;
1809 if (!old_name || !*old_name)
1810 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001811 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 if (err)
1813 return err;
1814
Al Virob12cea92011-03-18 08:55:38 -04001815 err = lock_mount(path);
David Howellscc53ce52011-01-14 18:45:26 +00001816 if (err < 0)
1817 goto out;
1818
Al Viro143c8c92011-11-25 00:46:35 -05001819 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05001820 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05001821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001823 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 goto out1;
1825
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001826 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001827 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
1829 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001830 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001831 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
Al Viro676da582011-11-24 21:47:05 -05001833 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05001834 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
Al Viro2d92ab32008-08-02 00:51:11 -04001836 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1837 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001838 goto out1;
1839 /*
1840 * Don't move a mount residing in a shared parent.
1841 */
Al Virofc7be132011-11-25 01:05:37 -05001842 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001843 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001844 /*
1845 * Don't move a mount tree containing unbindable mounts to a destination
1846 * mount which is shared.
1847 */
Al Virofc7be132011-11-25 01:05:37 -05001848 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05001849 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05001851 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05001852 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05001853 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
Al Viro0fb54e52011-11-24 19:59:16 -05001855 err = attach_recursive_mnt(old, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001856 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001857 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859 /* if the mount is moved, it should no longer be expire
1860 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05001861 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862out1:
Al Virob12cea92011-03-18 08:55:38 -04001863 unlock_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001866 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001867 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 return err;
1869}
1870
Al Viro9d412a42011-03-17 22:08:28 -04001871static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1872{
1873 int err;
1874 const char *subtype = strchr(fstype, '.');
1875 if (subtype) {
1876 subtype++;
1877 err = -EINVAL;
1878 if (!subtype[0])
1879 goto err;
1880 } else
1881 subtype = "";
1882
1883 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1884 err = -ENOMEM;
1885 if (!mnt->mnt_sb->s_subtype)
1886 goto err;
1887 return mnt;
1888
1889 err:
1890 mntput(mnt);
1891 return ERR_PTR(err);
1892}
1893
Al Viro9d412a42011-03-17 22:08:28 -04001894/*
1895 * add a mount into a namespace's mount tree
1896 */
Al Viro95bc5f22011-11-25 00:30:56 -05001897static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04001898{
1899 int err;
1900
1901 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1902
Al Virob12cea92011-03-18 08:55:38 -04001903 err = lock_mount(path);
1904 if (err)
1905 return err;
Al Viro9d412a42011-03-17 22:08:28 -04001906
1907 err = -EINVAL;
Al Viro156cacb2012-09-21 08:19:02 -04001908 if (unlikely(!check_mnt(real_mount(path->mnt)))) {
1909 /* that's acceptable only for automounts done in private ns */
1910 if (!(mnt_flags & MNT_SHRINKABLE))
1911 goto unlock;
1912 /* ... and for those we'd better have mountpoint still alive */
1913 if (!real_mount(path->mnt)->mnt_ns)
1914 goto unlock;
1915 }
Al Viro9d412a42011-03-17 22:08:28 -04001916
1917 /* Refuse the same filesystem on the same mount point */
1918 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05001919 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04001920 path->mnt->mnt_root == path->dentry)
1921 goto unlock;
1922
1923 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05001924 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04001925 goto unlock;
1926
Al Viro95bc5f22011-11-25 00:30:56 -05001927 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro9d412a42011-03-17 22:08:28 -04001928 err = graft_tree(newmnt, path);
1929
1930unlock:
Al Virob12cea92011-03-18 08:55:38 -04001931 unlock_mount(path);
Al Viro9d412a42011-03-17 22:08:28 -04001932 return err;
1933}
Al Virob1e75df2011-01-17 01:47:59 -05001934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935/*
1936 * create a new mount for userspace and request it to be added into the
1937 * namespace's tree
1938 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001939static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04001940 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001942 struct file_system_type *type;
Al Viro9b40bc92013-02-22 22:45:42 -05001943 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05001945 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001947 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 return -EINVAL;
1949
Al Viro9b40bc92013-02-22 22:45:42 -05001950 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 return -EPERM;
1952
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001953 type = get_fs_type(fstype);
1954 if (!type)
1955 return -ENODEV;
1956
1957 if (user_ns != &init_user_ns) {
1958 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
1959 put_filesystem(type);
1960 return -EPERM;
1961 }
1962 /* Only in special cases allow devices from mounts
1963 * created outside the initial user namespace.
1964 */
1965 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
1966 flags |= MS_NODEV;
1967 mnt_flags |= MNT_NODEV;
1968 }
1969 }
1970
1971 mnt = vfs_kern_mount(type, flags, name, data);
1972 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1973 !mnt->mnt_sb->s_subtype)
1974 mnt = fs_set_subtype(mnt, fstype);
1975
1976 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 if (IS_ERR(mnt))
1978 return PTR_ERR(mnt);
1979
Al Viro95bc5f22011-11-25 00:30:56 -05001980 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05001981 if (err)
1982 mntput(mnt);
1983 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984}
1985
Al Viro19a167a2011-01-17 01:35:23 -05001986int finish_automount(struct vfsmount *m, struct path *path)
1987{
Al Viro6776db3d2011-11-25 00:22:05 -05001988 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05001989 int err;
1990 /* The new mount record should have at least 2 refs to prevent it being
1991 * expired before we get a chance to add it
1992 */
Al Viro6776db3d2011-11-25 00:22:05 -05001993 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05001994
1995 if (m->mnt_sb == path->mnt->mnt_sb &&
1996 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05001997 err = -ELOOP;
1998 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05001999 }
2000
Al Viro95bc5f22011-11-25 00:30:56 -05002001 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002002 if (!err)
2003 return 0;
2004fail:
2005 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002006 if (!list_empty(&mnt->mnt_expire)) {
Al Virob1e75df2011-01-17 01:47:59 -05002007 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09302008 br_write_lock(&vfsmount_lock);
Al Viro6776db3d2011-11-25 00:22:05 -05002009 list_del_init(&mnt->mnt_expire);
Andi Kleen962830d2012-05-08 13:32:02 +09302010 br_write_unlock(&vfsmount_lock);
Al Virob1e75df2011-01-17 01:47:59 -05002011 up_write(&namespace_sem);
Al Viro19a167a2011-01-17 01:35:23 -05002012 }
Al Virob1e75df2011-01-17 01:47:59 -05002013 mntput(m);
2014 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002015 return err;
2016}
2017
David Howellsea5b7782011-01-14 19:10:03 +00002018/**
2019 * mnt_set_expiry - Put a mount on an expiration list
2020 * @mnt: The mount to list.
2021 * @expiry_list: The list to add the mount to.
2022 */
2023void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2024{
2025 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09302026 br_write_lock(&vfsmount_lock);
David Howellsea5b7782011-01-14 19:10:03 +00002027
Al Viro6776db3d2011-11-25 00:22:05 -05002028 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002029
Andi Kleen962830d2012-05-08 13:32:02 +09302030 br_write_unlock(&vfsmount_lock);
David Howellsea5b7782011-01-14 19:10:03 +00002031 up_write(&namespace_sem);
2032}
2033EXPORT_SYMBOL(mnt_set_expiry);
2034
2035/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 * process a list of expirable mountpoints with the intent of discarding any
2037 * mountpoints that aren't in use and haven't been touched since last we came
2038 * here
2039 */
2040void mark_mounts_for_expiry(struct list_head *mounts)
2041{
Al Viro761d5c32011-11-24 21:07:43 -05002042 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04002044 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
2046 if (list_empty(mounts))
2047 return;
2048
Al Virobcc5c7d2008-03-22 00:21:53 -04002049 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09302050 br_write_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051
2052 /* extract from the expiration list every vfsmount that matches the
2053 * following criteria:
2054 * - only referenced by its parent vfsmount
2055 * - still marked for expiry (marked on the last call here; marks are
2056 * cleared by mntput())
2057 */
Al Viro6776db3d2011-11-25 00:22:05 -05002058 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002059 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002060 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002062 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002064 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002065 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002066 touch_mnt_namespace(mnt->mnt_ns);
Al Virobcc5c7d2008-03-22 00:21:53 -04002067 umount_tree(mnt, 1, &umounts);
2068 }
Andi Kleen962830d2012-05-08 13:32:02 +09302069 br_write_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04002070 up_write(&namespace_sem);
2071
2072 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073}
2074
2075EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2076
2077/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002078 * Ripoff of 'select_parent()'
2079 *
2080 * search the list of submounts for a given mountpoint, and move any
2081 * shrinkable submounts to the 'graveyard' list.
2082 */
Al Viro692afc32011-11-24 21:15:14 -05002083static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002084{
Al Viro692afc32011-11-24 21:15:14 -05002085 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002086 struct list_head *next;
2087 int found = 0;
2088
2089repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002090 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002091resume:
Al Viro6b41d532011-11-24 23:24:33 -05002092 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002093 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002094 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002095
2096 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002097 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002098 continue;
2099 /*
2100 * Descend a level if the d_mounts list is non-empty.
2101 */
Al Viro6b41d532011-11-24 23:24:33 -05002102 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002103 this_parent = mnt;
2104 goto repeat;
2105 }
2106
Al Viro1ab59732011-11-24 21:35:16 -05002107 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002108 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002109 found++;
2110 }
2111 }
2112 /*
2113 * All done at this level ... ascend and resume the search
2114 */
2115 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002116 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002117 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002118 goto resume;
2119 }
2120 return found;
2121}
2122
2123/*
2124 * process a list of expirable mountpoints with the intent of discarding any
2125 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002126 *
2127 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002128 */
Al Viro692afc32011-11-24 21:15:14 -05002129static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002130{
2131 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002132 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002133
Trond Myklebust5528f9112006-06-09 09:34:17 -04002134 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002135 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002136 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002137 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002138 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002139 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08002140 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04002141 }
2142 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002143}
2144
Trond Myklebust5528f9112006-06-09 09:34:17 -04002145/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 * Some copy_from_user() implementations do not return the exact number of
2147 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2148 * Note that this function differs from copy_from_user() in that it will oops
2149 * on bad values of `to', rather than returning a short copy.
2150 */
Ram Paib58fed82005-11-07 17:16:09 -05002151static long exact_copy_from_user(void *to, const void __user * from,
2152 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153{
2154 char *t = to;
2155 const char __user *f = from;
2156 char c;
2157
2158 if (!access_ok(VERIFY_READ, from, n))
2159 return n;
2160
2161 while (n) {
2162 if (__get_user(c, f)) {
2163 memset(t, 0, n);
2164 break;
2165 }
2166 *t++ = c;
2167 f++;
2168 n--;
2169 }
2170 return n;
2171}
2172
Ram Paib58fed82005-11-07 17:16:09 -05002173int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174{
2175 int i;
2176 unsigned long page;
2177 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002178
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 *where = 0;
2180 if (!data)
2181 return 0;
2182
2183 if (!(page = __get_free_page(GFP_KERNEL)))
2184 return -ENOMEM;
2185
2186 /* We only care that *some* data at the address the user
2187 * gave us is valid. Just in case, we'll zero
2188 * the remainder of the page.
2189 */
2190 /* copy_from_user cannot cross TASK_SIZE ! */
2191 size = TASK_SIZE - (unsigned long)data;
2192 if (size > PAGE_SIZE)
2193 size = PAGE_SIZE;
2194
2195 i = size - exact_copy_from_user((void *)page, data, size);
2196 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002197 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 return -EFAULT;
2199 }
2200 if (i != PAGE_SIZE)
2201 memset((char *)page + i, 0, PAGE_SIZE - i);
2202 *where = page;
2203 return 0;
2204}
2205
Vegard Nossumeca6f532009-09-18 13:05:45 -07002206int copy_mount_string(const void __user *data, char **where)
2207{
2208 char *tmp;
2209
2210 if (!data) {
2211 *where = NULL;
2212 return 0;
2213 }
2214
2215 tmp = strndup_user(data, PAGE_SIZE);
2216 if (IS_ERR(tmp))
2217 return PTR_ERR(tmp);
2218
2219 *where = tmp;
2220 return 0;
2221}
2222
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223/*
2224 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2225 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2226 *
2227 * data is a (void *) that can point to any structure up to
2228 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2229 * information (or be NULL).
2230 *
2231 * Pre-0.97 versions of mount() didn't have a flags word.
2232 * When the flags word was introduced its top half was required
2233 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2234 * Therefore, if this magic number is present, it carries no information
2235 * and must be discarded.
2236 */
Al Viro808d4e32012-10-11 11:42:01 -04002237long do_mount(const char *dev_name, const char *dir_name,
2238 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239{
Al Viro2d92ab32008-08-02 00:51:11 -04002240 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 int retval = 0;
2242 int mnt_flags = 0;
2243
2244 /* Discard magic */
2245 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2246 flags &= ~MS_MGC_MSK;
2247
2248 /* Basic sanity checks */
2249
2250 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2251 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
2253 if (data_page)
2254 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2255
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002256 /* ... and get the mountpoint */
2257 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2258 if (retval)
2259 return retval;
2260
2261 retval = security_sb_mount(dev_name, &path,
2262 type_page, flags, data_page);
2263 if (retval)
2264 goto dput_out;
2265
Andi Kleen613cbe32009-04-19 18:40:43 +02002266 /* Default to relatime unless overriden */
2267 if (!(flags & MS_NOATIME))
2268 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002269
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 /* Separate the per-mountpoint flags */
2271 if (flags & MS_NOSUID)
2272 mnt_flags |= MNT_NOSUID;
2273 if (flags & MS_NODEV)
2274 mnt_flags |= MNT_NODEV;
2275 if (flags & MS_NOEXEC)
2276 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002277 if (flags & MS_NOATIME)
2278 mnt_flags |= MNT_NOATIME;
2279 if (flags & MS_NODIRATIME)
2280 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002281 if (flags & MS_STRICTATIME)
2282 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002283 if (flags & MS_RDONLY)
2284 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002285
Al Viro7a4dec52010-08-09 12:05:43 -04002286 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002287 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2288 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002291 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 data_page);
2293 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002294 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002295 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002296 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002298 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 else
Al Viro2d92ab32008-08-02 00:51:11 -04002300 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 dev_name, data_page);
2302dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002303 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 return retval;
2305}
2306
Eric W. Biederman771b1372012-07-26 21:08:32 -07002307static void free_mnt_ns(struct mnt_namespace *ns)
2308{
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002309 proc_free_inum(ns->proc_inum);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002310 put_user_ns(ns->user_ns);
2311 kfree(ns);
2312}
2313
Eric W. Biederman8823c072010-03-07 18:49:36 -08002314/*
2315 * Assign a sequence number so we can detect when we attempt to bind
2316 * mount a reference to an older mount namespace into the current
2317 * mount namespace, preventing reference counting loops. A 64bit
2318 * number incrementing at 10Ghz will take 12,427 years to wrap which
2319 * is effectively never, so we can ignore the possibility.
2320 */
2321static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2322
Eric W. Biederman771b1372012-07-26 21:08:32 -07002323static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002324{
2325 struct mnt_namespace *new_ns;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002326 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002327
2328 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2329 if (!new_ns)
2330 return ERR_PTR(-ENOMEM);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002331 ret = proc_alloc_inum(&new_ns->proc_inum);
2332 if (ret) {
2333 kfree(new_ns);
2334 return ERR_PTR(ret);
2335 }
Eric W. Biederman8823c072010-03-07 18:49:36 -08002336 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002337 atomic_set(&new_ns->count, 1);
2338 new_ns->root = NULL;
2339 INIT_LIST_HEAD(&new_ns->list);
2340 init_waitqueue_head(&new_ns->poll);
2341 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002342 new_ns->user_ns = get_user_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002343 return new_ns;
2344}
2345
JANAK DESAI741a2952006-02-07 12:59:00 -08002346/*
2347 * Allocate a new namespace structure and populate it with contents
2348 * copied from the namespace of the passed in task structure.
2349 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002350static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Eric W. Biederman771b1372012-07-26 21:08:32 -07002351 struct user_namespace *user_ns, struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002353 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002354 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002355 struct mount *p, *q;
Al Virobe08d6d2011-12-06 13:32:36 -05002356 struct mount *old = mnt_ns->root;
Al Virocb338d02011-11-24 20:55:08 -05002357 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002358 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
Eric W. Biederman771b1372012-07-26 21:08:32 -07002360 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002361 if (IS_ERR(new_ns))
2362 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363
Ram Pai390c6842005-11-07 17:17:51 -05002364 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 /* First pass: copy the tree topology */
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002366 copy_flags = CL_COPY_ALL | CL_EXPIRE;
2367 if (user_ns != mnt_ns->user_ns)
2368 copy_flags |= CL_SHARED_TO_SLAVE;
2369 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002370 if (IS_ERR(new)) {
Ram Pai390c6842005-11-07 17:17:51 -05002371 up_write(&namespace_sem);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002372 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002373 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 }
Al Virobe08d6d2011-12-06 13:32:36 -05002375 new_ns->root = new;
Andi Kleen962830d2012-05-08 13:32:02 +09302376 br_write_lock(&vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05002377 list_add_tail(&new_ns->list, &new->mnt_list);
Andi Kleen962830d2012-05-08 13:32:02 +09302378 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
2380 /*
2381 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2382 * as belonging to new namespace. We have already acquired a private
2383 * fs_struct, so tsk->fs->lock is not needed.
2384 */
Al Viro909b0a82011-11-25 03:06:56 -05002385 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002386 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002388 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 if (fs) {
Al Viro315fc832011-11-24 18:57:30 -05002390 if (&p->mnt == fs->root.mnt) {
2391 fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002392 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 }
Al Viro315fc832011-11-24 18:57:30 -05002394 if (&p->mnt == fs->pwd.mnt) {
2395 fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002396 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 }
Al Viro909b0a82011-11-25 03:06:56 -05002399 p = next_mnt(p, old);
2400 q = next_mnt(q, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 }
Ram Pai390c6842005-11-07 17:17:51 -05002402 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002405 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002407 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408
JANAK DESAI741a2952006-02-07 12:59:00 -08002409 return new_ns;
2410}
2411
Eric W. Biederman213dd262007-07-15 23:41:15 -07002412struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Eric W. Biederman771b1372012-07-26 21:08:32 -07002413 struct user_namespace *user_ns, struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002414{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002415 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002416
Badari Pulavartye3222c42007-05-08 00:25:21 -07002417 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002418 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002419
2420 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002421 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002422
Eric W. Biederman771b1372012-07-26 21:08:32 -07002423 new_ns = dup_mnt_ns(ns, user_ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002424
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002425 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002426 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427}
2428
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002429/**
2430 * create_mnt_ns - creates a private namespace and adds a root filesystem
2431 * @mnt: pointer to the new root filesystem mountpoint
2432 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002433static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002434{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002435 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002436 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002437 struct mount *mnt = real_mount(m);
2438 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002439 new_ns->root = mnt;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002440 list_add(&new_ns->list, &mnt->mnt_list);
Al Viroc1334492011-11-16 16:12:14 -05002441 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002442 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002443 }
2444 return new_ns;
2445}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002446
Al Viroea441d12011-11-16 21:43:59 -05002447struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2448{
2449 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002450 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002451 struct path path;
2452 int err;
2453
2454 ns = create_mnt_ns(mnt);
2455 if (IS_ERR(ns))
2456 return ERR_CAST(ns);
2457
2458 err = vfs_path_lookup(mnt->mnt_root, mnt,
2459 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2460
2461 put_mnt_ns(ns);
2462
2463 if (err)
2464 return ERR_PTR(err);
2465
2466 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002467 s = path.mnt->mnt_sb;
2468 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002469 mntput(path.mnt);
2470 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002471 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002472 /* ... and return the root of (sub)tree on it */
2473 return path.dentry;
2474}
2475EXPORT_SYMBOL(mount_subtree);
2476
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002477SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2478 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002480 int ret;
2481 char *kernel_type;
Jeff Layton91a27b22012-10-10 15:25:28 -04002482 struct filename *kernel_dir;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002483 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485
Vegard Nossumeca6f532009-09-18 13:05:45 -07002486 ret = copy_mount_string(type, &kernel_type);
2487 if (ret < 0)
2488 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489
Vegard Nossumeca6f532009-09-18 13:05:45 -07002490 kernel_dir = getname(dir_name);
2491 if (IS_ERR(kernel_dir)) {
2492 ret = PTR_ERR(kernel_dir);
2493 goto out_dir;
2494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495
Vegard Nossumeca6f532009-09-18 13:05:45 -07002496 ret = copy_mount_string(dev_name, &kernel_dev);
2497 if (ret < 0)
2498 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499
Vegard Nossumeca6f532009-09-18 13:05:45 -07002500 ret = copy_mount_options(data, &data_page);
2501 if (ret < 0)
2502 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503
Jeff Layton91a27b22012-10-10 15:25:28 -04002504 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
Vegard Nossumeca6f532009-09-18 13:05:45 -07002505 (void *) data_page);
2506
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002508out_data:
2509 kfree(kernel_dev);
2510out_dev:
2511 putname(kernel_dir);
2512out_dir:
2513 kfree(kernel_type);
2514out_type:
2515 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516}
2517
2518/*
Al Viroafac7cb2011-11-23 19:34:49 -05002519 * Return true if path is reachable from root
2520 *
2521 * namespace_sem or vfsmount_lock is held
2522 */
Al Viro643822b2011-11-24 22:00:28 -05002523bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002524 const struct path *root)
2525{
Al Viro643822b2011-11-24 22:00:28 -05002526 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002527 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002528 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002529 }
Al Viro643822b2011-11-24 22:00:28 -05002530 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002531}
2532
2533int path_is_under(struct path *path1, struct path *path2)
2534{
2535 int res;
Andi Kleen962830d2012-05-08 13:32:02 +09302536 br_read_lock(&vfsmount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002537 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Andi Kleen962830d2012-05-08 13:32:02 +09302538 br_read_unlock(&vfsmount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002539 return res;
2540}
2541EXPORT_SYMBOL(path_is_under);
2542
2543/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 * pivot_root Semantics:
2545 * Moves the root file system of the current process to the directory put_old,
2546 * makes new_root as the new root file system of the current process, and sets
2547 * root/cwd of all processes which had them on the current root to new_root.
2548 *
2549 * Restrictions:
2550 * The new_root and put_old must be directories, and must not be on the
2551 * same file system as the current process root. The put_old must be
2552 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2553 * pointed to by put_old must yield the same directory as new_root. No other
2554 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2555 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002556 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2557 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2558 * in this situation.
2559 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 * Notes:
2561 * - we don't move root/cwd if they are not at the root (reason: if something
2562 * cared enough to change them, it's probably wrong to force them elsewhere)
2563 * - it's okay to pick a root that isn't the root of a file system, e.g.
2564 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2565 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2566 * first.
2567 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002568SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2569 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570{
Al Viro2d8f3032008-07-22 09:59:21 -04002571 struct path new, old, parent_path, root_parent, root;
Al Viro419148d2011-11-24 19:41:16 -05002572 struct mount *new_mnt, *root_mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 int error;
2574
Al Viro9b40bc92013-02-22 22:45:42 -05002575 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 return -EPERM;
2577
Al Viro2d8f3032008-07-22 09:59:21 -04002578 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 if (error)
2580 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581
Al Viro2d8f3032008-07-22 09:59:21 -04002582 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 if (error)
2584 goto out1;
2585
Al Viro2d8f3032008-07-22 09:59:21 -04002586 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002587 if (error)
2588 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002590 get_fs_root(current->fs, &root);
Al Virob12cea92011-03-18 08:55:38 -04002591 error = lock_mount(&old);
2592 if (error)
2593 goto out3;
2594
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002596 new_mnt = real_mount(new.mnt);
2597 root_mnt = real_mount(root.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002598 if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2599 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2600 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002601 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002602 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002603 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002605 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002606 goto out4;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002607 if (d_unlinked(old.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002608 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002610 if (new.mnt == root.mnt ||
2611 old.mnt == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002612 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002614 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002615 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002616 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002617 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002618 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002619 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002620 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002621 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002622 /* make sure we can reach put_old from new_root */
Al Viro643822b2011-11-24 22:00:28 -05002623 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002624 goto out4;
Andi Kleen962830d2012-05-08 13:32:02 +09302625 br_write_lock(&vfsmount_lock);
Al Viro419148d2011-11-24 19:41:16 -05002626 detach_mnt(new_mnt, &parent_path);
2627 detach_mnt(root_mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002628 /* mount old root on put_old */
Al Viro419148d2011-11-24 19:41:16 -05002629 attach_mnt(root_mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002630 /* mount new_root on / */
Al Viro419148d2011-11-24 19:41:16 -05002631 attach_mnt(new_mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002632 touch_mnt_namespace(current->nsproxy->mnt_ns);
Andi Kleen962830d2012-05-08 13:32:02 +09302633 br_write_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002634 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002636out4:
2637 unlock_mount(&old);
2638 if (!error) {
2639 path_put(&root_parent);
2640 path_put(&parent_path);
2641 }
2642out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002643 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002644out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002645 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002647 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650}
2651
2652static void __init init_mount_tree(void)
2653{
2654 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002655 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002656 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002657 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002659 type = get_fs_type("rootfs");
2660 if (!type)
2661 panic("Can't find rootfs type");
2662 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2663 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 if (IS_ERR(mnt))
2665 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002666
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002667 ns = create_mnt_ns(mnt);
2668 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002671 init_task.nsproxy->mnt_ns = ns;
2672 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673
Al Virobe08d6d2011-12-06 13:32:36 -05002674 root.mnt = mnt;
2675 root.dentry = mnt->mnt_root;
Jan Blunckac748a02008-02-14 19:34:39 -08002676
2677 set_fs_pwd(current->fs, &root);
2678 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679}
2680
Denis Cheng74bf17c2007-10-16 23:26:30 -07002681void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002683 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002684 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685
Ram Pai390c6842005-11-07 17:17:51 -05002686 init_rwsem(&namespace_sem);
2687
Al Viro7d6fec42011-11-23 12:14:10 -05002688 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002689 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
Ram Paib58fed82005-11-07 17:16:09 -05002691 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692
2693 if (!mount_hashtable)
2694 panic("Failed to allocate mount hash table\n");
2695
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002696 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697
Eric Dumazet13f14b42008-02-06 01:37:57 -08002698 for (u = 0; u < HASH_SIZE; u++)
2699 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700
Andi Kleen962830d2012-05-08 13:32:02 +09302701 br_lock_init(&vfsmount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +10002702
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002703 err = sysfs_init();
2704 if (err)
2705 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002706 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002707 fs_kobj = kobject_create_and_add("fs", NULL);
2708 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002709 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 init_rootfs();
2711 init_mount_tree();
2712}
2713
Trond Myklebust616511d2009-06-22 15:09:13 -04002714void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002716 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002717
Al Virod498b252010-02-05 02:21:06 -05002718 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002719 return;
Ram Pai390c6842005-11-07 17:17:51 -05002720 down_write(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09302721 br_write_lock(&vfsmount_lock);
Al Virobe08d6d2011-12-06 13:32:36 -05002722 umount_tree(ns->root, 0, &umount_list);
Andi Kleen962830d2012-05-08 13:32:02 +09302723 br_write_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002724 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002725 release_mounts(&umount_list);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002726 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727}
Al Viro9d412a42011-03-17 22:08:28 -04002728
2729struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2730{
Tim Chen423e0ab2011-07-19 09:32:38 -07002731 struct vfsmount *mnt;
2732 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2733 if (!IS_ERR(mnt)) {
2734 /*
2735 * it is a longterm mount, don't release mnt until
2736 * we unmount before file sys is unregistered
2737 */
Al Virof7a99c52012-06-09 00:59:08 -04002738 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07002739 }
2740 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002741}
2742EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002743
2744void kern_unmount(struct vfsmount *mnt)
2745{
2746 /* release long term mount so mount point can be released */
2747 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04002748 br_write_lock(&vfsmount_lock);
2749 real_mount(mnt)->mnt_ns = NULL;
2750 br_write_unlock(&vfsmount_lock);
Tim Chen423e0ab2011-07-19 09:32:38 -07002751 mntput(mnt);
2752 }
2753}
2754EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002755
2756bool our_mnt(struct vfsmount *mnt)
2757{
Al Viro143c8c92011-11-25 00:46:35 -05002758 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05002759}
Eric W. Biederman8823c072010-03-07 18:49:36 -08002760
2761static void *mntns_get(struct task_struct *task)
2762{
2763 struct mnt_namespace *ns = NULL;
2764 struct nsproxy *nsproxy;
2765
2766 rcu_read_lock();
2767 nsproxy = task_nsproxy(task);
2768 if (nsproxy) {
2769 ns = nsproxy->mnt_ns;
2770 get_mnt_ns(ns);
2771 }
2772 rcu_read_unlock();
2773
2774 return ns;
2775}
2776
2777static void mntns_put(void *ns)
2778{
2779 put_mnt_ns(ns);
2780}
2781
2782static int mntns_install(struct nsproxy *nsproxy, void *ns)
2783{
2784 struct fs_struct *fs = current->fs;
2785 struct mnt_namespace *mnt_ns = ns;
2786 struct path root;
2787
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002788 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biederman5e4a0842012-12-14 07:55:36 -08002789 !nsown_capable(CAP_SYS_CHROOT) ||
2790 !nsown_capable(CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08002791 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002792
2793 if (fs->users != 1)
2794 return -EINVAL;
2795
2796 get_mnt_ns(mnt_ns);
2797 put_mnt_ns(nsproxy->mnt_ns);
2798 nsproxy->mnt_ns = mnt_ns;
2799
2800 /* Find the root */
2801 root.mnt = &mnt_ns->root->mnt;
2802 root.dentry = mnt_ns->root->mnt.mnt_root;
2803 path_get(&root);
2804 while(d_mountpoint(root.dentry) && follow_down_one(&root))
2805 ;
2806
2807 /* Update the pwd and root */
2808 set_fs_pwd(fs, &root);
2809 set_fs_root(fs, &root);
2810
2811 path_put(&root);
2812 return 0;
2813}
2814
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002815static unsigned int mntns_inum(void *ns)
2816{
2817 struct mnt_namespace *mnt_ns = ns;
2818 return mnt_ns->proc_inum;
2819}
2820
Eric W. Biederman8823c072010-03-07 18:49:36 -08002821const struct proc_ns_operations mntns_operations = {
2822 .name = "mnt",
2823 .type = CLONE_NEWNS,
2824 .get = mntns_get,
2825 .put = mntns_put,
2826 .install = mntns_install,
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002827 .inum = mntns_inum,
Eric W. Biederman8823c072010-03-07 18:49:36 -08002828};