blob: a45ba4f267fe6e834909f122dd166cdd608dcce5 [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>
David Howells0bb80f22013-04-12 01:50:06 +010024#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070025#include <linux/magic.h>
Ram Pai07b20882005-11-07 17:19:07 -050026#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070027#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
Eric Dumazet13f14b42008-02-06 01:37:57 -080029#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
30#define HASH_SIZE (1UL << HASH_SHIFT)
31
Al Viro5addc5d2005-11-07 17:15:49 -050032static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010033static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010034static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100035static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040036static int mnt_id_start = 0;
37static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Eric Dumazetfa3536c2006-03-26 01:37:24 -080039static struct list_head *mount_hashtable __read_mostly;
Al Viro84d17192013-03-15 10:53:28 -040040static struct list_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080041static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050042static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080044/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060045struct kobject *fs_kobj;
46EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080047
Nick Piggin99b7db72010-08-18 04:37:39 +100048/*
49 * vfsmount lock may be taken for read to prevent changes to the
50 * vfsmount hash, ie. during mountpoint lookups or walking back
51 * up the tree.
52 *
53 * It should be taken for write in all cases where the vfsmount
54 * tree or hash is modified or when a vfsmount structure is modified.
55 */
56DEFINE_BRLOCK(vfsmount_lock);
57
Linus Torvalds1da177e2005-04-16 15:20:36 -070058static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
59{
Ram Paib58fed82005-11-07 17:16:09 -050060 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
61 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080062 tmp = tmp + (tmp >> HASH_SHIFT);
63 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064}
65
Dave Hansen3d733632008-02-15 14:37:59 -080066#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
67
Nick Piggin99b7db72010-08-18 04:37:39 +100068/*
69 * allocation is serialized by namespace_sem, but we need the spinlock to
70 * serialize with freeing.
71 */
Al Virob105e272011-11-24 20:38:33 -050072static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010073{
74 int res;
75
76retry:
77 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +100078 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -050079 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -040080 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -050081 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +100082 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010083 if (res == -EAGAIN)
84 goto retry;
85
86 return res;
87}
88
Al Virob105e272011-11-24 20:38:33 -050089static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010090{
Al Viro15169fe2011-11-25 00:50:41 -050091 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +100092 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040093 ida_remove(&mnt_id_ida, id);
94 if (mnt_id_start > id)
95 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +100096 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010097}
98
Miklos Szeredi719f5d72008-03-27 13:06:23 +010099/*
100 * Allocate a new peer group ID
101 *
102 * mnt_group_ida is protected by namespace_sem
103 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500104static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100105{
Al Virof21f6222009-06-24 03:12:00 -0400106 int res;
107
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100108 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
109 return -ENOMEM;
110
Al Virof21f6222009-06-24 03:12:00 -0400111 res = ida_get_new_above(&mnt_group_ida,
112 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500113 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400114 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500115 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400116
117 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100118}
119
120/*
121 * Release a peer group ID
122 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500123void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100124{
Al Viro15169fe2011-11-25 00:50:41 -0500125 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400126 ida_remove(&mnt_group_ida, id);
127 if (mnt_group_start > id)
128 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500129 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100130}
131
Nick Pigginb3e19d92011-01-07 17:50:11 +1100132/*
133 * vfsmount lock must be held for read
134 */
Al Viro83adc752011-11-24 22:37:54 -0500135static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100136{
137#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500138 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100139#else
140 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500141 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100142 preempt_enable();
143#endif
144}
145
Nick Pigginb3e19d92011-01-07 17:50:11 +1100146/*
147 * vfsmount lock must be held for write
148 */
Al Viro83adc752011-11-24 22:37:54 -0500149unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100150{
151#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500152 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100153 int cpu;
154
155 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500156 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100157 }
158
159 return count;
160#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500161 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100162#endif
163}
164
Al Virob105e272011-11-24 20:38:33 -0500165static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166{
Al Viroc63181e2011-11-25 02:35:16 -0500167 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
168 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100169 int err;
170
Al Viroc63181e2011-11-25 02:35:16 -0500171 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800172 if (err)
173 goto out_free_cache;
174
175 if (name) {
Al Viroc63181e2011-11-25 02:35:16 -0500176 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
177 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800178 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100179 }
180
Nick Pigginb3e19d92011-01-07 17:50:11 +1100181#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500182 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
183 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100184 goto out_free_devname;
185
Al Viroc63181e2011-11-25 02:35:16 -0500186 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100187#else
Al Viroc63181e2011-11-25 02:35:16 -0500188 mnt->mnt_count = 1;
189 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100190#endif
191
Al Viroc63181e2011-11-25 02:35:16 -0500192 INIT_LIST_HEAD(&mnt->mnt_hash);
193 INIT_LIST_HEAD(&mnt->mnt_child);
194 INIT_LIST_HEAD(&mnt->mnt_mounts);
195 INIT_LIST_HEAD(&mnt->mnt_list);
196 INIT_LIST_HEAD(&mnt->mnt_expire);
197 INIT_LIST_HEAD(&mnt->mnt_share);
198 INIT_LIST_HEAD(&mnt->mnt_slave_list);
199 INIT_LIST_HEAD(&mnt->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500200#ifdef CONFIG_FSNOTIFY
201 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
202#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 }
Al Viroc63181e2011-11-25 02:35:16 -0500204 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800205
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000206#ifdef CONFIG_SMP
207out_free_devname:
Al Viroc63181e2011-11-25 02:35:16 -0500208 kfree(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000209#endif
Li Zefan88b38782008-07-21 18:06:36 +0800210out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500211 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800212out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500213 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800214 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215}
216
Dave Hansen83660252008-02-15 14:37:30 -0800217/*
218 * Most r/o checks on a fs are for operations that take
219 * discrete amounts of time, like a write() or unlink().
220 * We must keep track of when those operations start
221 * (for permission checks) and when they end, so that
222 * we can determine when writes are able to occur to
223 * a filesystem.
224 */
Dave Hansen3d733632008-02-15 14:37:59 -0800225/*
226 * __mnt_is_readonly: check whether a mount is read-only
227 * @mnt: the mount to check for its write status
228 *
229 * This shouldn't be used directly ouside of the VFS.
230 * It does not guarantee that the filesystem will stay
231 * r/w, just that it is right *now*. This can not and
232 * should not be used in place of IS_RDONLY(inode).
233 * mnt_want/drop_write() will _keep_ the filesystem
234 * r/w.
235 */
236int __mnt_is_readonly(struct vfsmount *mnt)
237{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800238 if (mnt->mnt_flags & MNT_READONLY)
239 return 1;
240 if (mnt->mnt_sb->s_flags & MS_RDONLY)
241 return 1;
242 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800243}
244EXPORT_SYMBOL_GPL(__mnt_is_readonly);
245
Al Viro83adc752011-11-24 22:37:54 -0500246static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800247{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000248#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500249 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000250#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500251 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000252#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800253}
Dave Hansen3d733632008-02-15 14:37:59 -0800254
Al Viro83adc752011-11-24 22:37:54 -0500255static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800256{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000257#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500258 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000259#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500260 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000261#endif
262}
263
Al Viro83adc752011-11-24 22:37:54 -0500264static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000265{
266#ifdef CONFIG_SMP
267 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800268 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800269
270 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500271 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800272 }
Dave Hansen3d733632008-02-15 14:37:59 -0800273
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000274 return count;
275#else
276 return mnt->mnt_writers;
277#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800278}
279
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100280static int mnt_is_readonly(struct vfsmount *mnt)
281{
282 if (mnt->mnt_sb->s_readonly_remount)
283 return 1;
284 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
285 smp_rmb();
286 return __mnt_is_readonly(mnt);
287}
288
Dave Hansen3d733632008-02-15 14:37:59 -0800289/*
Jan Karaeb04c282012-06-12 16:20:35 +0200290 * Most r/o & frozen checks on a fs are for operations that take discrete
291 * amounts of time, like a write() or unlink(). We must keep track of when
292 * those operations start (for permission checks) and when they end, so that we
293 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800294 */
Dave Hansen83660252008-02-15 14:37:30 -0800295/**
Jan Karaeb04c282012-06-12 16:20:35 +0200296 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500297 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800298 *
Jan Karaeb04c282012-06-12 16:20:35 +0200299 * This tells the low-level filesystem that a write is about to be performed to
300 * it, and makes sure that writes are allowed (mnt it read-write) before
301 * returning success. This operation does not protect against filesystem being
302 * frozen. When the write operation is finished, __mnt_drop_write() must be
303 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800304 */
Jan Karaeb04c282012-06-12 16:20:35 +0200305int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800306{
Al Viro83adc752011-11-24 22:37:54 -0500307 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800308 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800309
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000310 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100311 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000312 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100313 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000314 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
315 * incremented count after it has set MNT_WRITE_HOLD.
316 */
317 smp_mb();
Miao Xie1e755292012-11-16 17:23:50 +0800318 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000319 cpu_relax();
320 /*
321 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
322 * be set to match its requirements. So we must not load that until
323 * MNT_WRITE_HOLD is cleared.
324 */
325 smp_rmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100326 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100327 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800328 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800329 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000330 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200331
332 return ret;
333}
334
335/**
336 * mnt_want_write - get write access to a mount
337 * @m: the mount on which to take a write
338 *
339 * This tells the low-level filesystem that a write is about to be performed to
340 * it, and makes sure that writes are allowed (mount is read-write, filesystem
341 * is not frozen) before returning success. When the write operation is
342 * finished, mnt_drop_write() must be called. This is effectively a refcount.
343 */
344int mnt_want_write(struct vfsmount *m)
345{
346 int ret;
347
348 sb_start_write(m->mnt_sb);
349 ret = __mnt_want_write(m);
350 if (ret)
351 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800352 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800353}
354EXPORT_SYMBOL_GPL(mnt_want_write);
355
356/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000357 * mnt_clone_write - get write access to a mount
358 * @mnt: the mount on which to take a write
359 *
360 * This is effectively like mnt_want_write, except
361 * it must only be used to take an extra write reference
362 * on a mountpoint that we already know has a write reference
363 * on it. This allows some optimisation.
364 *
365 * After finished, mnt_drop_write must be called as usual to
366 * drop the reference.
367 */
368int mnt_clone_write(struct vfsmount *mnt)
369{
370 /* superblock may be r/o */
371 if (__mnt_is_readonly(mnt))
372 return -EROFS;
373 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500374 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000375 preempt_enable();
376 return 0;
377}
378EXPORT_SYMBOL_GPL(mnt_clone_write);
379
380/**
Jan Karaeb04c282012-06-12 16:20:35 +0200381 * __mnt_want_write_file - get write access to a file's mount
382 * @file: the file who's mount on which to take a write
383 *
384 * This is like __mnt_want_write, but it takes a file and can
385 * do some optimisations if the file is open for write already
386 */
387int __mnt_want_write_file(struct file *file)
388{
Al Viro496ad9a2013-01-23 17:07:38 -0500389 struct inode *inode = file_inode(file);
Jan Karaeb04c282012-06-12 16:20:35 +0200390
391 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
392 return __mnt_want_write(file->f_path.mnt);
393 else
394 return mnt_clone_write(file->f_path.mnt);
395}
396
397/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000398 * mnt_want_write_file - get write access to a file's mount
399 * @file: the file who's mount on which to take a write
400 *
401 * This is like mnt_want_write, but it takes a file and can
402 * do some optimisations if the file is open for write already
403 */
404int mnt_want_write_file(struct file *file)
405{
Jan Karaeb04c282012-06-12 16:20:35 +0200406 int ret;
407
408 sb_start_write(file->f_path.mnt->mnt_sb);
409 ret = __mnt_want_write_file(file);
410 if (ret)
411 sb_end_write(file->f_path.mnt->mnt_sb);
412 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000413}
414EXPORT_SYMBOL_GPL(mnt_want_write_file);
415
416/**
Jan Karaeb04c282012-06-12 16:20:35 +0200417 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800418 * @mnt: the mount on which to give up write access
419 *
420 * Tells the low-level filesystem that we are done
421 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200422 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800423 */
Jan Karaeb04c282012-06-12 16:20:35 +0200424void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800425{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000426 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500427 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000428 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800429}
Jan Karaeb04c282012-06-12 16:20:35 +0200430
431/**
432 * mnt_drop_write - give up write access to a mount
433 * @mnt: the mount on which to give up write access
434 *
435 * Tells the low-level filesystem that we are done performing writes to it and
436 * also allows filesystem to be frozen again. Must be matched with
437 * mnt_want_write() call above.
438 */
439void mnt_drop_write(struct vfsmount *mnt)
440{
441 __mnt_drop_write(mnt);
442 sb_end_write(mnt->mnt_sb);
443}
Dave Hansen83660252008-02-15 14:37:30 -0800444EXPORT_SYMBOL_GPL(mnt_drop_write);
445
Jan Karaeb04c282012-06-12 16:20:35 +0200446void __mnt_drop_write_file(struct file *file)
447{
448 __mnt_drop_write(file->f_path.mnt);
449}
450
Al Viro2a79f172011-12-09 08:06:57 -0500451void mnt_drop_write_file(struct file *file)
452{
453 mnt_drop_write(file->f_path.mnt);
454}
455EXPORT_SYMBOL(mnt_drop_write_file);
456
Al Viro83adc752011-11-24 22:37:54 -0500457static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800458{
Dave Hansen3d733632008-02-15 14:37:59 -0800459 int ret = 0;
460
Andi Kleen962830d2012-05-08 13:32:02 +0930461 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500462 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000463 /*
464 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
465 * should be visible before we do.
466 */
467 smp_mb();
468
469 /*
470 * With writers on hold, if this value is zero, then there are
471 * definitely no active writers (although held writers may subsequently
472 * increment the count, they'll have to wait, and decrement it after
473 * seeing MNT_READONLY).
474 *
475 * It is OK to have counter incremented on one CPU and decremented on
476 * another: the sum will add up correctly. The danger would be when we
477 * sum up each counter, if we read a counter before it is incremented,
478 * but then read another CPU's count which it has been subsequently
479 * decremented from -- we would see more decrements than we should.
480 * MNT_WRITE_HOLD protects against this scenario, because
481 * mnt_want_write first increments count, then smp_mb, then spins on
482 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
483 * we're counting up here.
484 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100485 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000486 ret = -EBUSY;
487 else
Al Viro83adc752011-11-24 22:37:54 -0500488 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000489 /*
490 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
491 * that become unheld will see MNT_READONLY.
492 */
493 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500494 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Andi Kleen962830d2012-05-08 13:32:02 +0930495 br_write_unlock(&vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800496 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800497}
Dave Hansen83660252008-02-15 14:37:30 -0800498
Al Viro83adc752011-11-24 22:37:54 -0500499static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800500{
Andi Kleen962830d2012-05-08 13:32:02 +0930501 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500502 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Andi Kleen962830d2012-05-08 13:32:02 +0930503 br_write_unlock(&vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800504}
505
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100506int sb_prepare_remount_readonly(struct super_block *sb)
507{
508 struct mount *mnt;
509 int err = 0;
510
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100511 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
512 if (atomic_long_read(&sb->s_remove_count))
513 return -EBUSY;
514
Andi Kleen962830d2012-05-08 13:32:02 +0930515 br_write_lock(&vfsmount_lock);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100516 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
517 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
518 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
519 smp_mb();
520 if (mnt_get_writers(mnt) > 0) {
521 err = -EBUSY;
522 break;
523 }
524 }
525 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100526 if (!err && atomic_long_read(&sb->s_remove_count))
527 err = -EBUSY;
528
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100529 if (!err) {
530 sb->s_readonly_remount = 1;
531 smp_wmb();
532 }
533 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
534 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
535 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
536 }
Andi Kleen962830d2012-05-08 13:32:02 +0930537 br_write_unlock(&vfsmount_lock);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100538
539 return err;
540}
541
Al Virob105e272011-11-24 20:38:33 -0500542static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543{
Al Viro52ba1622011-11-25 02:25:17 -0500544 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100545 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000546#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500547 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000548#endif
Al Virob105e272011-11-24 20:38:33 -0500549 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550}
551
552/*
Ram Paia05964f2005-11-07 17:20:17 -0500553 * find the first or last mount at @dentry on vfsmount @mnt depending on
554 * @dir. If @dir is set return the first mount else return the last mount.
Nick Piggin99b7db72010-08-18 04:37:39 +1000555 * vfsmount_lock must be held for read or write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 */
Al Viroc7105362011-11-24 18:22:03 -0500557struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
Ram Paia05964f2005-11-07 17:20:17 -0500558 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559{
Ram Paib58fed82005-11-07 17:16:09 -0500560 struct list_head *head = mount_hashtable + hash(mnt, dentry);
561 struct list_head *tmp = head;
Al Viroc7105362011-11-24 18:22:03 -0500562 struct mount *p, *found = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500565 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 p = NULL;
567 if (tmp == head)
568 break;
Al Viro1b8e5562011-11-24 21:01:32 -0500569 p = list_entry(tmp, struct mount, mnt_hash);
Al Viroa73324d2011-11-24 22:25:07 -0500570 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500571 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 break;
573 }
574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 return found;
576}
577
Ram Paia05964f2005-11-07 17:20:17 -0500578/*
David Howellsf015f1262012-06-25 12:55:28 +0100579 * lookup_mnt - Return the first child mount mounted at path
580 *
581 * "First" means first mounted chronologically. If you create the
582 * following mounts:
583 *
584 * mount /dev/sda1 /mnt
585 * mount /dev/sda2 /mnt
586 * mount /dev/sda3 /mnt
587 *
588 * Then lookup_mnt() on the base /mnt dentry in the root mount will
589 * return successively the root dentry and vfsmount of /dev/sda1, then
590 * /dev/sda2, then /dev/sda3, then NULL.
591 *
592 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500593 */
Al Viro1c755af2009-04-18 14:06:57 -0400594struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500595{
Al Viroc7105362011-11-24 18:22:03 -0500596 struct mount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000597
Andi Kleen962830d2012-05-08 13:32:02 +0930598 br_read_lock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500599 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
600 if (child_mnt) {
601 mnt_add_count(child_mnt, 1);
Andi Kleen962830d2012-05-08 13:32:02 +0930602 br_read_unlock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500603 return &child_mnt->mnt;
604 } else {
Andi Kleen962830d2012-05-08 13:32:02 +0930605 br_read_unlock(&vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500606 return NULL;
607 }
Ram Paia05964f2005-11-07 17:20:17 -0500608}
609
Al Viro84d17192013-03-15 10:53:28 -0400610static struct mountpoint *new_mountpoint(struct dentry *dentry)
611{
612 struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry);
613 struct mountpoint *mp;
614
615 list_for_each_entry(mp, chain, m_hash) {
616 if (mp->m_dentry == dentry) {
617 /* might be worth a WARN_ON() */
618 if (d_unlinked(dentry))
619 return ERR_PTR(-ENOENT);
620 mp->m_count++;
621 return mp;
622 }
623 }
624
625 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
626 if (!mp)
627 return ERR_PTR(-ENOMEM);
628
629 spin_lock(&dentry->d_lock);
630 if (d_unlinked(dentry)) {
631 spin_unlock(&dentry->d_lock);
632 kfree(mp);
633 return ERR_PTR(-ENOENT);
634 }
635 dentry->d_flags |= DCACHE_MOUNTED;
636 spin_unlock(&dentry->d_lock);
637 mp->m_dentry = dentry;
638 mp->m_count = 1;
639 list_add(&mp->m_hash, chain);
640 return mp;
641}
642
643static void put_mountpoint(struct mountpoint *mp)
644{
645 if (!--mp->m_count) {
646 struct dentry *dentry = mp->m_dentry;
647 spin_lock(&dentry->d_lock);
648 dentry->d_flags &= ~DCACHE_MOUNTED;
649 spin_unlock(&dentry->d_lock);
650 list_del(&mp->m_hash);
651 kfree(mp);
652 }
653}
654
Al Viro143c8c92011-11-25 00:46:35 -0500655static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800657 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
Nick Piggin99b7db72010-08-18 04:37:39 +1000660/*
661 * vfsmount lock must be held for write
662 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800663static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500664{
665 if (ns) {
666 ns->event = ++event;
667 wake_up_interruptible(&ns->poll);
668 }
669}
670
Nick Piggin99b7db72010-08-18 04:37:39 +1000671/*
672 * vfsmount lock must be held for write
673 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800674static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500675{
676 if (ns && ns->event != event) {
677 ns->event = event;
678 wake_up_interruptible(&ns->poll);
679 }
680}
681
Nick Piggin99b7db72010-08-18 04:37:39 +1000682/*
683 * vfsmount lock must be held for write
684 */
Al Viro419148d2011-11-24 19:41:16 -0500685static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
Al Viroa73324d2011-11-24 22:25:07 -0500687 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500688 old_path->mnt = &mnt->mnt_parent->mnt;
689 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500690 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500691 list_del_init(&mnt->mnt_child);
Al Viro1b8e5562011-11-24 21:01:32 -0500692 list_del_init(&mnt->mnt_hash);
Al Viro84d17192013-03-15 10:53:28 -0400693 put_mountpoint(mnt->mnt_mp);
694 mnt->mnt_mp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695}
696
Nick Piggin99b7db72010-08-18 04:37:39 +1000697/*
698 * vfsmount lock must be held for write
699 */
Al Viro84d17192013-03-15 10:53:28 -0400700void mnt_set_mountpoint(struct mount *mnt,
701 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500702 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500703{
Al Viro84d17192013-03-15 10:53:28 -0400704 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500705 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro84d17192013-03-15 10:53:28 -0400706 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500707 child_mnt->mnt_parent = mnt;
Al Viro84d17192013-03-15 10:53:28 -0400708 child_mnt->mnt_mp = mp;
Ram Paib90fa9a2005-11-07 17:19:50 -0500709}
710
Nick Piggin99b7db72010-08-18 04:37:39 +1000711/*
712 * vfsmount lock must be held for write
713 */
Al Viro84d17192013-03-15 10:53:28 -0400714static void attach_mnt(struct mount *mnt,
715 struct mount *parent,
716 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717{
Al Viro84d17192013-03-15 10:53:28 -0400718 mnt_set_mountpoint(parent, mp, mnt);
Al Viro1b8e5562011-11-24 21:01:32 -0500719 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro84d17192013-03-15 10:53:28 -0400720 hash(&parent->mnt, mp->m_dentry));
721 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500722}
723
724/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000725 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500726 */
Al Viro4b2619a2011-11-24 19:47:15 -0500727static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500728{
Al Viro0714a532011-11-24 22:19:58 -0500729 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500730 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500731 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500732 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500733
Al Viro0714a532011-11-24 22:19:58 -0500734 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500735
Al Viro1a4eeaf2011-11-25 02:19:55 -0500736 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400737 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500738 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500739
Ram Paib90fa9a2005-11-07 17:19:50 -0500740 list_splice(&head, n->list.prev);
741
Al Viro1b8e5562011-11-24 21:01:32 -0500742 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viroa73324d2011-11-24 22:25:07 -0500743 hash(&parent->mnt, mnt->mnt_mountpoint));
Al Viro6b41d532011-11-24 23:24:33 -0500744 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800745 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
747
Al Viro909b0a82011-11-25 03:06:56 -0500748static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
Al Viro6b41d532011-11-24 23:24:33 -0500750 struct list_head *next = p->mnt_mounts.next;
751 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500753 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500755 next = p->mnt_child.next;
756 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 break;
Al Viro0714a532011-11-24 22:19:58 -0500758 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
760 }
Al Viro6b41d532011-11-24 23:24:33 -0500761 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
Al Viro315fc832011-11-24 18:57:30 -0500764static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500765{
Al Viro6b41d532011-11-24 23:24:33 -0500766 struct list_head *prev = p->mnt_mounts.prev;
767 while (prev != &p->mnt_mounts) {
768 p = list_entry(prev, struct mount, mnt_child);
769 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500770 }
771 return p;
772}
773
Al Viro9d412a42011-03-17 22:08:28 -0400774struct vfsmount *
775vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
776{
Al Virob105e272011-11-24 20:38:33 -0500777 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400778 struct dentry *root;
779
780 if (!type)
781 return ERR_PTR(-ENODEV);
782
783 mnt = alloc_vfsmnt(name);
784 if (!mnt)
785 return ERR_PTR(-ENOMEM);
786
787 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500788 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400789
790 root = mount_fs(type, flags, name, data);
791 if (IS_ERR(root)) {
792 free_vfsmnt(mnt);
793 return ERR_CAST(root);
794 }
795
Al Virob105e272011-11-24 20:38:33 -0500796 mnt->mnt.mnt_root = root;
797 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500798 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500799 mnt->mnt_parent = mnt;
Andi Kleen962830d2012-05-08 13:32:02 +0930800 br_write_lock(&vfsmount_lock);
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100801 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Andi Kleen962830d2012-05-08 13:32:02 +0930802 br_write_unlock(&vfsmount_lock);
Al Virob105e272011-11-24 20:38:33 -0500803 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400804}
805EXPORT_SYMBOL_GPL(vfs_kern_mount);
806
Al Viro87129cc2011-11-24 21:24:27 -0500807static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500808 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809{
Al Viro87129cc2011-11-24 21:24:27 -0500810 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100811 struct mount *mnt;
812 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
David Howellsbe34d1a2012-06-25 12:55:18 +0100814 mnt = alloc_vfsmnt(old->mnt_devname);
815 if (!mnt)
816 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100817
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700818 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +0100819 mnt->mnt_group_id = 0; /* not a peer of original */
820 else
821 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100822
David Howellsbe34d1a2012-06-25 12:55:18 +0100823 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
824 err = mnt_alloc_group_id(mnt);
825 if (err)
826 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100828
829 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700830 /* Don't allow unprivileged users to change mount flags */
831 if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
832 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
833
David Howellsbe34d1a2012-06-25 12:55:18 +0100834 atomic_inc(&sb->s_active);
835 mnt->mnt.mnt_sb = sb;
836 mnt->mnt.mnt_root = dget(root);
837 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
838 mnt->mnt_parent = mnt;
839 br_write_lock(&vfsmount_lock);
840 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
841 br_write_unlock(&vfsmount_lock);
842
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700843 if ((flag & CL_SLAVE) ||
844 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +0100845 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
846 mnt->mnt_master = old;
847 CLEAR_MNT_SHARED(mnt);
848 } else if (!(flag & CL_PRIVATE)) {
849 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
850 list_add(&mnt->mnt_share, &old->mnt_share);
851 if (IS_MNT_SLAVE(old))
852 list_add(&mnt->mnt_slave, &old->mnt_slave);
853 mnt->mnt_master = old->mnt_master;
854 }
855 if (flag & CL_MAKE_SHARED)
856 set_mnt_shared(mnt);
857
858 /* stick the duplicate mount on the same expiry list
859 * as the original if that was on one */
860 if (flag & CL_EXPIRE) {
861 if (!list_empty(&old->mnt_expire))
862 list_add(&mnt->mnt_expire, &old->mnt_expire);
863 }
864
Al Virocb338d02011-11-24 20:55:08 -0500865 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100866
867 out_free:
868 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +0100869 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870}
871
Al Viro83adc752011-11-24 22:37:54 -0500872static inline void mntfree(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873{
Al Viro83adc752011-11-24 22:37:54 -0500874 struct vfsmount *m = &mnt->mnt;
875 struct super_block *sb = m->mnt_sb;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100876
Dave Hansen3d733632008-02-15 14:37:59 -0800877 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800878 * This probably indicates that somebody messed
879 * up a mnt_want/drop_write() pair. If this
880 * happens, the filesystem was probably unable
881 * to make r/w->r/o transitions.
882 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000883 /*
Nick Pigginb3e19d92011-01-07 17:50:11 +1100884 * The locking used to deal with mnt_count decrement provides barriers,
885 * so mnt_get_writers() below is safe.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000886 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100887 WARN_ON(mnt_get_writers(mnt));
Al Viro83adc752011-11-24 22:37:54 -0500888 fsnotify_vfsmount_delete(m);
889 dput(m->mnt_root);
890 free_vfsmnt(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 deactivate_super(sb);
892}
893
Al Viro900148d2011-11-25 00:33:11 -0500894static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500895{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100896put_again:
Al Virof03c6592011-01-14 22:30:21 -0500897#ifdef CONFIG_SMP
Andi Kleen962830d2012-05-08 13:32:02 +0930898 br_read_lock(&vfsmount_lock);
Al Virof7a99c52012-06-09 00:59:08 -0400899 if (likely(mnt->mnt_ns)) {
900 /* shouldn't be the last one */
Al Viroaa9c0e02011-11-23 19:04:52 -0500901 mnt_add_count(mnt, -1);
Andi Kleen962830d2012-05-08 13:32:02 +0930902 br_read_unlock(&vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500903 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100904 }
Andi Kleen962830d2012-05-08 13:32:02 +0930905 br_read_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100906
Andi Kleen962830d2012-05-08 13:32:02 +0930907 br_write_lock(&vfsmount_lock);
Al Viroaa9c0e02011-11-23 19:04:52 -0500908 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100909 if (mnt_get_count(mnt)) {
Andi Kleen962830d2012-05-08 13:32:02 +0930910 br_write_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100911 return;
912 }
Nick Pigginb3e19d92011-01-07 17:50:11 +1100913#else
Al Viroaa9c0e02011-11-23 19:04:52 -0500914 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100915 if (likely(mnt_get_count(mnt)))
Nick Piggin99b7db72010-08-18 04:37:39 +1000916 return;
Andi Kleen962830d2012-05-08 13:32:02 +0930917 br_write_lock(&vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500918#endif
Al Viro863d6842011-11-25 00:57:42 -0500919 if (unlikely(mnt->mnt_pinned)) {
920 mnt_add_count(mnt, mnt->mnt_pinned + 1);
921 mnt->mnt_pinned = 0;
Andi Kleen962830d2012-05-08 13:32:02 +0930922 br_write_unlock(&vfsmount_lock);
Al Viro900148d2011-11-25 00:33:11 -0500923 acct_auto_close_mnt(&mnt->mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100924 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500925 }
Andi Kleen962830d2012-05-08 13:32:02 +0930926
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100927 list_del(&mnt->mnt_instance);
Andi Kleen962830d2012-05-08 13:32:02 +0930928 br_write_unlock(&vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100929 mntfree(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500930}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100931
932void mntput(struct vfsmount *mnt)
933{
934 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -0500935 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100936 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -0500937 if (unlikely(m->mnt_expiry_mark))
938 m->mnt_expiry_mark = 0;
939 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100940 }
941}
942EXPORT_SYMBOL(mntput);
943
944struct vfsmount *mntget(struct vfsmount *mnt)
945{
946 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -0500947 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100948 return mnt;
949}
950EXPORT_SYMBOL(mntget);
951
Al Viro7b7b1ac2005-11-07 17:13:39 -0500952void mnt_pin(struct vfsmount *mnt)
953{
Andi Kleen962830d2012-05-08 13:32:02 +0930954 br_write_lock(&vfsmount_lock);
Al Viro863d6842011-11-25 00:57:42 -0500955 real_mount(mnt)->mnt_pinned++;
Andi Kleen962830d2012-05-08 13:32:02 +0930956 br_write_unlock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500957}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500958EXPORT_SYMBOL(mnt_pin);
959
Al Viro863d6842011-11-25 00:57:42 -0500960void mnt_unpin(struct vfsmount *m)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500961{
Al Viro863d6842011-11-25 00:57:42 -0500962 struct mount *mnt = real_mount(m);
Andi Kleen962830d2012-05-08 13:32:02 +0930963 br_write_lock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500964 if (mnt->mnt_pinned) {
Al Viro863d6842011-11-25 00:57:42 -0500965 mnt_add_count(mnt, 1);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500966 mnt->mnt_pinned--;
967 }
Andi Kleen962830d2012-05-08 13:32:02 +0930968 br_write_unlock(&vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500969}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500970EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800972static inline void mangle(struct seq_file *m, const char *s)
973{
974 seq_escape(m, s, " \t\n\\");
975}
976
977/*
978 * Simple .show_options callback for filesystems which don't want to
979 * implement more complex mount option showing.
980 *
981 * See also save_mount_options().
982 */
Al Viro34c80b12011-12-08 21:32:45 -0500983int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800984{
Al Viro2a32ceb2009-05-08 16:05:57 -0400985 const char *options;
986
987 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -0500988 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800989
990 if (options != NULL && options[0]) {
991 seq_putc(m, ',');
992 mangle(m, options);
993 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400994 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800995
996 return 0;
997}
998EXPORT_SYMBOL(generic_show_options);
999
1000/*
1001 * If filesystem uses generic_show_options(), this function should be
1002 * called from the fill_super() callback.
1003 *
1004 * The .remount_fs callback usually needs to be handled in a special
1005 * way, to make sure, that previous options are not overwritten if the
1006 * remount fails.
1007 *
1008 * Also note, that if the filesystem's .remount_fs function doesn't
1009 * reset all options to their default value, but changes only newly
1010 * given options, then the displayed options will not reflect reality
1011 * any more.
1012 */
1013void save_mount_options(struct super_block *sb, char *options)
1014{
Al Viro2a32ceb2009-05-08 16:05:57 -04001015 BUG_ON(sb->s_options);
1016 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001017}
1018EXPORT_SYMBOL(save_mount_options);
1019
Al Viro2a32ceb2009-05-08 16:05:57 -04001020void replace_mount_options(struct super_block *sb, char *options)
1021{
1022 char *old = sb->s_options;
1023 rcu_assign_pointer(sb->s_options, options);
1024 if (old) {
1025 synchronize_rcu();
1026 kfree(old);
1027 }
1028}
1029EXPORT_SYMBOL(replace_mount_options);
1030
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001031#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001032/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033static void *m_start(struct seq_file *m, loff_t *pos)
1034{
Al Viro6ce6e242012-06-09 01:16:59 -04001035 struct proc_mounts *p = proc_mounts(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
Ram Pai390c6842005-11-07 17:17:51 -05001037 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001038 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039}
1040
1041static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1042{
Al Viro6ce6e242012-06-09 01:16:59 -04001043 struct proc_mounts *p = proc_mounts(m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001044
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001045 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046}
1047
1048static void m_stop(struct seq_file *m, void *v)
1049{
Ram Pai390c6842005-11-07 17:17:51 -05001050 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051}
1052
Al Viro0226f492011-12-06 12:21:54 -05001053static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001054{
Al Viro6ce6e242012-06-09 01:16:59 -04001055 struct proc_mounts *p = proc_mounts(m);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001056 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001057 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058}
1059
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001060const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 .start = m_start,
1062 .next = m_next,
1063 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001064 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001065};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001066#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001067
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068/**
1069 * may_umount_tree - check if a mount tree is busy
1070 * @mnt: root of mount tree
1071 *
1072 * This is called to check if a tree of mounts has any
1073 * open files, pwds, chroots or sub mounts that are
1074 * busy.
1075 */
Al Viro909b0a82011-11-25 03:06:56 -05001076int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077{
Al Viro909b0a82011-11-25 03:06:56 -05001078 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001079 int actual_refs = 0;
1080 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001081 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001082 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
Nick Pigginb3e19d92011-01-07 17:50:11 +11001084 /* write lock needed for mnt_get_count */
Andi Kleen962830d2012-05-08 13:32:02 +09301085 br_write_lock(&vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -05001086 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001087 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 }
Andi Kleen962830d2012-05-08 13:32:02 +09301090 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
1092 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001093 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Ian Kente3474a82006-03-27 01:14:51 -08001095 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096}
1097
1098EXPORT_SYMBOL(may_umount_tree);
1099
1100/**
1101 * may_umount - check if a mount point is busy
1102 * @mnt: root of mount
1103 *
1104 * This is called to check if a mount point has any
1105 * open files, pwds, chroots or sub mounts. If the
1106 * mount has sub mounts this will return busy
1107 * regardless of whether the sub mounts are busy.
1108 *
1109 * Doesn't take quota and stuff into account. IOW, in some cases it will
1110 * give false negatives. The main reason why it's here is that we need
1111 * a non-destructive way to look for easily umountable filesystems.
1112 */
1113int may_umount(struct vfsmount *mnt)
1114{
Ian Kente3474a82006-03-27 01:14:51 -08001115 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001116 down_read(&namespace_sem);
Andi Kleen962830d2012-05-08 13:32:02 +09301117 br_write_lock(&vfsmount_lock);
Al Viro1ab59732011-11-24 21:35:16 -05001118 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001119 ret = 0;
Andi Kleen962830d2012-05-08 13:32:02 +09301120 br_write_unlock(&vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001121 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001122 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123}
1124
1125EXPORT_SYMBOL(may_umount);
1126
Al Viroe3197d82013-03-16 14:35:16 -04001127static LIST_HEAD(unmounted); /* protected by namespace_sem */
1128
Al Viro97216be2013-03-16 15:12:40 -04001129static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130{
Al Virod5e50f72011-11-24 20:58:57 -05001131 struct mount *mnt;
Al Viro97216be2013-03-16 15:12:40 -04001132 LIST_HEAD(head);
1133
1134 if (likely(list_empty(&unmounted))) {
1135 up_write(&namespace_sem);
1136 return;
1137 }
1138
1139 list_splice_init(&unmounted, &head);
1140 up_write(&namespace_sem);
1141
1142 while (!list_empty(&head)) {
1143 mnt = list_first_entry(&head, struct mount, mnt_hash);
Al Viro1b8e5562011-11-24 21:01:32 -05001144 list_del_init(&mnt->mnt_hash);
Al Viro676da582011-11-24 21:47:05 -05001145 if (mnt_has_parent(mnt)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001146 struct dentry *dentry;
Al Viro863d6842011-11-25 00:57:42 -05001147 struct mount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001148
Andi Kleen962830d2012-05-08 13:32:02 +09301149 br_write_lock(&vfsmount_lock);
Al Viroa73324d2011-11-24 22:25:07 -05001150 dentry = mnt->mnt_mountpoint;
Al Viro863d6842011-11-25 00:57:42 -05001151 m = mnt->mnt_parent;
Al Viroa73324d2011-11-24 22:25:07 -05001152 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -05001153 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001154 m->mnt_ghosts--;
Andi Kleen962830d2012-05-08 13:32:02 +09301155 br_write_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001156 dput(dentry);
Al Viro863d6842011-11-25 00:57:42 -05001157 mntput(&m->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 }
Al Virod5e50f72011-11-24 20:58:57 -05001159 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001160 }
1161}
1162
Al Viro97216be2013-03-16 15:12:40 -04001163static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001164{
Al Viro97216be2013-03-16 15:12:40 -04001165 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001166}
1167
Nick Piggin99b7db72010-08-18 04:37:39 +10001168/*
1169 * vfsmount lock must be held for write
1170 * namespace_sem must be held for write
1171 */
Al Viro328e6d92013-03-16 14:49:45 -04001172void umount_tree(struct mount *mnt, int propagate)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001173{
Al Viro7b8a53f2011-01-15 20:08:44 -05001174 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001175 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001176
Al Viro909b0a82011-11-25 03:06:56 -05001177 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro1b8e5562011-11-24 21:01:32 -05001178 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001179
Ram Paia05964f2005-11-07 17:20:17 -05001180 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001181 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001182
Al Viro1b8e5562011-11-24 21:01:32 -05001183 list_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001184 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001185 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001186 __touch_mnt_namespace(p->mnt_ns);
1187 p->mnt_ns = NULL;
Al Viro6b41d532011-11-24 23:24:33 -05001188 list_del_init(&p->mnt_child);
Al Viro676da582011-11-24 21:47:05 -05001189 if (mnt_has_parent(p)) {
Al Viro863d6842011-11-25 00:57:42 -05001190 p->mnt_parent->mnt_ghosts++;
Al Viro84d17192013-03-15 10:53:28 -04001191 put_mountpoint(p->mnt_mp);
1192 p->mnt_mp = NULL;
Al Viro7c4b93d2008-03-21 23:59:49 -04001193 }
Al Viro0f0afb12011-11-24 20:43:10 -05001194 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 }
Al Viro328e6d92013-03-16 14:49:45 -04001196 list_splice(&tmp_list, &unmounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197}
1198
Al Virob54b9be2013-03-16 14:39:34 -04001199static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001200
Al Viro1ab59732011-11-24 21:35:16 -05001201static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202{
Al Viro1ab59732011-11-24 21:35:16 -05001203 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 int retval;
1205
Al Viro1ab59732011-11-24 21:35:16 -05001206 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 if (retval)
1208 return retval;
1209
1210 /*
1211 * Allow userspace to request a mountpoint be expired rather than
1212 * unmounting unconditionally. Unmount only happens if:
1213 * (1) the mark is already set (the mark is cleared by mntput())
1214 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1215 */
1216 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001217 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 flags & (MNT_FORCE | MNT_DETACH))
1219 return -EINVAL;
1220
Nick Pigginb3e19d92011-01-07 17:50:11 +11001221 /*
1222 * probably don't strictly need the lock here if we examined
1223 * all race cases, but it's a slowpath.
1224 */
Andi Kleen962830d2012-05-08 13:32:02 +09301225 br_write_lock(&vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -05001226 if (mnt_get_count(mnt) != 2) {
Andi Kleen962830d2012-05-08 13:32:02 +09301227 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001229 }
Andi Kleen962830d2012-05-08 13:32:02 +09301230 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
Al Viro863d6842011-11-25 00:57:42 -05001232 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 return -EAGAIN;
1234 }
1235
1236 /*
1237 * If we may have to abort operations to get out of this
1238 * mount, and they will themselves hold resources we must
1239 * allow the fs to do things. In the Unix tradition of
1240 * 'Gee thats tricky lets do it in userspace' the umount_begin
1241 * might fail to complete on the first run through as other tasks
1242 * must return, and the like. Thats for the mount program to worry
1243 * about for the moment.
1244 */
1245
Al Viro42faad92008-04-24 07:21:56 -04001246 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001247 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001248 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
1250 /*
1251 * No sense to grab the lock for this test, but test itself looks
1252 * somewhat bogus. Suggestions for better replacement?
1253 * Ho-hum... In principle, we might treat that as umount + switch
1254 * to rootfs. GC would eventually take care of the old vfsmount.
1255 * Actually it makes sense, especially if rootfs would contain a
1256 * /reboot - static binary that would close all descriptors and
1257 * call reboot(9). Then init(8) could umount root and exec /reboot.
1258 */
Al Viro1ab59732011-11-24 21:35:16 -05001259 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 /*
1261 * Special case for "unmounting" root ...
1262 * we just try to remount it readonly.
1263 */
1264 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001265 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 up_write(&sb->s_umount);
1268 return retval;
1269 }
1270
Al Viro97216be2013-03-16 15:12:40 -04001271 namespace_lock();
Andi Kleen962830d2012-05-08 13:32:02 +09301272 br_write_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001273 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Al Viroc35038b2008-03-22 00:46:23 -04001275 if (!(flags & MNT_DETACH))
Al Virob54b9be2013-03-16 14:39:34 -04001276 shrink_submounts(mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001277
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001279 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001280 if (!list_empty(&mnt->mnt_list))
Al Viro328e6d92013-03-16 14:49:45 -04001281 umount_tree(mnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 retval = 0;
1283 }
Andi Kleen962830d2012-05-08 13:32:02 +09301284 br_write_unlock(&vfsmount_lock);
Al Viroe3197d82013-03-16 14:35:16 -04001285 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 return retval;
1287}
1288
Al Viro9b40bc92013-02-22 22:45:42 -05001289/*
1290 * Is the caller allowed to modify his namespace?
1291 */
1292static inline bool may_mount(void)
1293{
1294 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1295}
1296
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297/*
1298 * Now umount can handle mount points as well as block devices.
1299 * This is important for filesystems which use unnamed block devices.
1300 *
1301 * We now support a flag for forced unmount like the other 'big iron'
1302 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1303 */
1304
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001305SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306{
Al Viro2d8f3032008-07-22 09:59:21 -04001307 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001308 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001310 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001312 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1313 return -EINVAL;
1314
Al Viro9b40bc92013-02-22 22:45:42 -05001315 if (!may_mount())
1316 return -EPERM;
1317
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001318 if (!(flags & UMOUNT_NOFOLLOW))
1319 lookup_flags |= LOOKUP_FOLLOW;
1320
1321 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 if (retval)
1323 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001324 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001326 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001328 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 goto dput_and_out;
1330
Al Viro900148d2011-11-25 00:33:11 -05001331 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001333 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001334 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001335 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336out:
1337 return retval;
1338}
1339
1340#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1341
1342/*
Ram Paib58fed82005-11-07 17:16:09 -05001343 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001345SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346{
Ram Paib58fed82005-11-07 17:16:09 -05001347 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348}
1349
1350#endif
1351
Eric W. Biederman8823c072010-03-07 18:49:36 -08001352static bool mnt_ns_loop(struct path *path)
1353{
1354 /* Could bind mounting the mount namespace inode cause a
1355 * mount namespace loop?
1356 */
1357 struct inode *inode = path->dentry->d_inode;
David Howells0bb80f22013-04-12 01:50:06 +01001358 struct proc_ns *ei;
Eric W. Biederman8823c072010-03-07 18:49:36 -08001359 struct mnt_namespace *mnt_ns;
1360
1361 if (!proc_ns_inode(inode))
1362 return false;
1363
David Howells0bb80f22013-04-12 01:50:06 +01001364 ei = get_proc_ns(inode);
Eric W. Biederman8823c072010-03-07 18:49:36 -08001365 if (ei->ns_ops != &mntns_operations)
1366 return false;
1367
1368 mnt_ns = ei->ns;
1369 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1370}
1371
Al Viro87129cc2011-11-24 21:24:27 -05001372struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001373 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374{
Al Viro84d17192013-03-15 10:53:28 -04001375 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
Al Virofc7be132011-11-25 01:05:37 -05001377 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
David Howellsbe34d1a2012-06-25 12:55:18 +01001378 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001379
Ram Pai36341f62005-11-07 17:17:22 -05001380 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001381 if (IS_ERR(q))
1382 return q;
1383
Al Viroa73324d2011-11-24 22:25:07 -05001384 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
1386 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001387 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001388 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001389 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 continue;
1391
Al Viro909b0a82011-11-25 03:06:56 -05001392 for (s = r; s; s = next_mnt(s, r)) {
Al Virofc7be132011-11-25 01:05:37 -05001393 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001394 s = skip_mnt_tree(s);
1395 continue;
1396 }
Al Viro0714a532011-11-24 22:19:58 -05001397 while (p != s->mnt_parent) {
1398 p = p->mnt_parent;
1399 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 }
Al Viro87129cc2011-11-24 21:24:27 -05001401 p = s;
Al Viro84d17192013-03-15 10:53:28 -04001402 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001403 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001404 if (IS_ERR(q))
1405 goto out;
Andi Kleen962830d2012-05-08 13:32:02 +09301406 br_write_lock(&vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001407 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro84d17192013-03-15 10:53:28 -04001408 attach_mnt(q, parent, p->mnt_mp);
Andi Kleen962830d2012-05-08 13:32:02 +09301409 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 }
1411 }
1412 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001413out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 if (res) {
Andi Kleen962830d2012-05-08 13:32:02 +09301415 br_write_lock(&vfsmount_lock);
Al Viro328e6d92013-03-16 14:49:45 -04001416 umount_tree(res, 0);
Andi Kleen962830d2012-05-08 13:32:02 +09301417 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001419 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420}
1421
David Howellsbe34d1a2012-06-25 12:55:18 +01001422/* Caller should check returned pointer for errors */
1423
Al Viro589ff872009-04-18 03:28:19 -04001424struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001425{
Al Virocb338d02011-11-24 20:55:08 -05001426 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001427 namespace_lock();
Al Viro87129cc2011-11-24 21:24:27 -05001428 tree = copy_tree(real_mount(path->mnt), path->dentry,
1429 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001430 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001431 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001432 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001433 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001434}
1435
1436void drop_collected_mounts(struct vfsmount *mnt)
1437{
Al Viro97216be2013-03-16 15:12:40 -04001438 namespace_lock();
Andi Kleen962830d2012-05-08 13:32:02 +09301439 br_write_lock(&vfsmount_lock);
Al Viro328e6d92013-03-16 14:49:45 -04001440 umount_tree(real_mount(mnt), 0);
Andi Kleen962830d2012-05-08 13:32:02 +09301441 br_write_unlock(&vfsmount_lock);
Al Viro3ab6abe2013-03-16 14:42:19 -04001442 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001443}
1444
Al Viro1f707132010-01-30 22:51:25 -05001445int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1446 struct vfsmount *root)
1447{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001448 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001449 int res = f(root, arg);
1450 if (res)
1451 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001452 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1453 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001454 if (res)
1455 return res;
1456 }
1457 return 0;
1458}
1459
Al Viro4b8b21f2011-11-24 19:54:23 -05001460static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001461{
Al Viro315fc832011-11-24 18:57:30 -05001462 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001463
Al Viro909b0a82011-11-25 03:06:56 -05001464 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001465 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001466 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001467 }
1468}
1469
Al Viro4b8b21f2011-11-24 19:54:23 -05001470static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001471{
Al Viro315fc832011-11-24 18:57:30 -05001472 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001473
Al Viro909b0a82011-11-25 03:06:56 -05001474 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001475 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001476 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001477 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001478 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001479 return err;
1480 }
1481 }
1482 }
1483
1484 return 0;
1485}
1486
Ram Paib90fa9a2005-11-07 17:19:50 -05001487/*
1488 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001489 * @nd : place the mount tree @source_mnt is attached
1490 * @parent_nd : if non-null, detach the source_mnt from its parent and
1491 * store the parent mount and mountpoint dentry.
1492 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001493 *
1494 * NOTE: in the table below explains the semantics when a source mount
1495 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001496 * ---------------------------------------------------------------------------
1497 * | BIND MOUNT OPERATION |
1498 * |**************************************************************************
1499 * | source-->| shared | private | slave | unbindable |
1500 * | dest | | | | |
1501 * | | | | | | |
1502 * | v | | | | |
1503 * |**************************************************************************
1504 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1505 * | | | | | |
1506 * |non-shared| shared (+) | private | slave (*) | invalid |
1507 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001508 * A bind operation clones the source mount and mounts the clone on the
1509 * destination mount.
1510 *
1511 * (++) the cloned mount is propagated to all the mounts in the propagation
1512 * tree of the destination mount and the cloned mount is added to
1513 * the peer group of the source mount.
1514 * (+) the cloned mount is created under the destination mount and is marked
1515 * as shared. The cloned mount is added to the peer group of the source
1516 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001517 * (+++) the mount is propagated to all the mounts in the propagation tree
1518 * of the destination mount and the cloned mount is made slave
1519 * of the same master as that of the source mount. The cloned mount
1520 * is marked as 'shared and slave'.
1521 * (*) the cloned mount is made a slave of the same master as that of the
1522 * source mount.
1523 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001524 * ---------------------------------------------------------------------------
1525 * | MOVE MOUNT OPERATION |
1526 * |**************************************************************************
1527 * | source-->| shared | private | slave | unbindable |
1528 * | dest | | | | |
1529 * | | | | | | |
1530 * | v | | | | |
1531 * |**************************************************************************
1532 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1533 * | | | | | |
1534 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1535 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001536 *
1537 * (+) the mount is moved to the destination. And is then propagated to
1538 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001539 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001540 * (+++) the mount is moved to the destination and is then propagated to
1541 * all the mounts belonging to the destination mount's propagation tree.
1542 * the mount is marked as 'shared and slave'.
1543 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001544 *
1545 * if the source mount is a tree, the operations explained above is
1546 * applied to each mount in the tree.
1547 * Must be called without spinlocks held, since this function can sleep
1548 * in allocations.
1549 */
Al Viro0fb54e52011-11-24 19:59:16 -05001550static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d17192013-03-15 10:53:28 -04001551 struct mount *dest_mnt,
1552 struct mountpoint *dest_mp,
1553 struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001554{
1555 LIST_HEAD(tree_list);
Al Viro315fc832011-11-24 18:57:30 -05001556 struct mount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001557 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001558
Al Virofc7be132011-11-25 01:05:37 -05001559 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001560 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001561 if (err)
1562 goto out;
1563 }
Al Viro84d17192013-03-15 10:53:28 -04001564 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001565 if (err)
1566 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001567
Andi Kleen962830d2012-05-08 13:32:02 +09301568 br_write_lock(&vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001569
Al Virofc7be132011-11-25 01:05:37 -05001570 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro909b0a82011-11-25 03:06:56 -05001571 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001572 set_mnt_shared(p);
Ram Paib90fa9a2005-11-07 17:19:50 -05001573 }
Al Viro1a390682008-03-21 20:48:19 -04001574 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001575 detach_mnt(source_mnt, parent_path);
Al Viro84d17192013-03-15 10:53:28 -04001576 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05001577 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001578 } else {
Al Viro84d17192013-03-15 10:53:28 -04001579 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Al Viro0fb54e52011-11-24 19:59:16 -05001580 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05001581 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001582
Al Viro1b8e5562011-11-24 21:01:32 -05001583 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1584 list_del_init(&child->mnt_hash);
Al Viro4b2619a2011-11-24 19:47:15 -05001585 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05001586 }
Andi Kleen962830d2012-05-08 13:32:02 +09301587 br_write_unlock(&vfsmount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +10001588
Ram Paib90fa9a2005-11-07 17:19:50 -05001589 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001590
1591 out_cleanup_ids:
Al Virofc7be132011-11-25 01:05:37 -05001592 if (IS_MNT_SHARED(dest_mnt))
Al Viro0fb54e52011-11-24 19:59:16 -05001593 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001594 out:
1595 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001596}
1597
Al Viro84d17192013-03-15 10:53:28 -04001598static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04001599{
1600 struct vfsmount *mnt;
Al Viro84d17192013-03-15 10:53:28 -04001601 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04001602retry:
Al Viro84d17192013-03-15 10:53:28 -04001603 mutex_lock(&dentry->d_inode->i_mutex);
1604 if (unlikely(cant_mount(dentry))) {
1605 mutex_unlock(&dentry->d_inode->i_mutex);
1606 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04001607 }
Al Viro97216be2013-03-16 15:12:40 -04001608 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04001609 mnt = lookup_mnt(path);
Al Viro84d17192013-03-15 10:53:28 -04001610 if (likely(!mnt)) {
1611 struct mountpoint *mp = new_mountpoint(dentry);
1612 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04001613 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001614 mutex_unlock(&dentry->d_inode->i_mutex);
1615 return mp;
1616 }
1617 return mp;
1618 }
Al Viro97216be2013-03-16 15:12:40 -04001619 namespace_unlock();
Al Virob12cea92011-03-18 08:55:38 -04001620 mutex_unlock(&path->dentry->d_inode->i_mutex);
1621 path_put(path);
1622 path->mnt = mnt;
Al Viro84d17192013-03-15 10:53:28 -04001623 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04001624 goto retry;
1625}
1626
Al Viro84d17192013-03-15 10:53:28 -04001627static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04001628{
Al Viro84d17192013-03-15 10:53:28 -04001629 struct dentry *dentry = where->m_dentry;
1630 put_mountpoint(where);
Al Viro328e6d92013-03-16 14:49:45 -04001631 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001632 mutex_unlock(&dentry->d_inode->i_mutex);
Al Virob12cea92011-03-18 08:55:38 -04001633}
1634
Al Viro84d17192013-03-15 10:53:28 -04001635static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636{
Al Viro95bc5f22011-11-25 00:30:56 -05001637 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 return -EINVAL;
1639
Al Viro84d17192013-03-15 10:53:28 -04001640 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001641 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 return -ENOTDIR;
1643
Al Viro84d17192013-03-15 10:53:28 -04001644 return attach_recursive_mnt(mnt, p, mp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645}
1646
1647/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001648 * Sanity check the flags to change_mnt_propagation.
1649 */
1650
1651static int flags_to_propagation_type(int flags)
1652{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001653 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001654
1655 /* Fail if any non-propagation flags are set */
1656 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1657 return 0;
1658 /* Only one propagation flag should be set */
1659 if (!is_power_of_2(type))
1660 return 0;
1661 return type;
1662}
1663
1664/*
Ram Pai07b20882005-11-07 17:19:07 -05001665 * recursively change the type of the mountpoint.
1666 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001667static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001668{
Al Viro315fc832011-11-24 18:57:30 -05001669 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001670 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001671 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001672 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001673 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001674
Al Viro2d92ab32008-08-02 00:51:11 -04001675 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001676 return -EINVAL;
1677
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001678 type = flags_to_propagation_type(flag);
1679 if (!type)
1680 return -EINVAL;
1681
Al Viro97216be2013-03-16 15:12:40 -04001682 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001683 if (type == MS_SHARED) {
1684 err = invent_group_ids(mnt, recurse);
1685 if (err)
1686 goto out_unlock;
1687 }
1688
Andi Kleen962830d2012-05-08 13:32:02 +09301689 br_write_lock(&vfsmount_lock);
Al Viro909b0a82011-11-25 03:06:56 -05001690 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001691 change_mnt_propagation(m, type);
Andi Kleen962830d2012-05-08 13:32:02 +09301692 br_write_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001693
1694 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04001695 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001696 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001697}
1698
1699/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 * do loopback mount.
1701 */
Al Viro808d4e32012-10-11 11:42:01 -04001702static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001703 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704{
Al Viro2d92ab32008-08-02 00:51:11 -04001705 struct path old_path;
Al Viro84d17192013-03-15 10:53:28 -04001706 struct mount *mnt = NULL, *old, *parent;
1707 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05001708 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 if (!old_name || !*old_name)
1710 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001711 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 if (err)
1713 return err;
1714
Eric W. Biederman8823c072010-03-07 18:49:36 -08001715 err = -EINVAL;
1716 if (mnt_ns_loop(&old_path))
1717 goto out;
1718
Al Viro84d17192013-03-15 10:53:28 -04001719 mp = lock_mount(path);
1720 err = PTR_ERR(mp);
1721 if (IS_ERR(mp))
Jan Blunck4ac91372008-02-14 19:34:32 -08001722 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001723
Al Viro87129cc2011-11-24 21:24:27 -05001724 old = real_mount(old_path.mnt);
Al Viro84d17192013-03-15 10:53:28 -04001725 parent = real_mount(path->mnt);
Al Viro87129cc2011-11-24 21:24:27 -05001726
Al Virob12cea92011-03-18 08:55:38 -04001727 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001728 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04001729 goto out2;
1730
Al Viro84d17192013-03-15 10:53:28 -04001731 if (!check_mnt(parent) || !check_mnt(old))
Al Virob12cea92011-03-18 08:55:38 -04001732 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
Al Viroccd48bc2005-11-07 17:15:04 -05001734 if (recurse)
Al Viro87129cc2011-11-24 21:24:27 -05001735 mnt = copy_tree(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001736 else
Al Viro87129cc2011-11-24 21:24:27 -05001737 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001738
David Howellsbe34d1a2012-06-25 12:55:18 +01001739 if (IS_ERR(mnt)) {
1740 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04001741 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01001742 }
Al Viroccd48bc2005-11-07 17:15:04 -05001743
Al Viro84d17192013-03-15 10:53:28 -04001744 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05001745 if (err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301746 br_write_lock(&vfsmount_lock);
Al Viro328e6d92013-03-16 14:49:45 -04001747 umount_tree(mnt, 0);
Andi Kleen962830d2012-05-08 13:32:02 +09301748 br_write_unlock(&vfsmount_lock);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001749 }
Al Virob12cea92011-03-18 08:55:38 -04001750out2:
Al Viro84d17192013-03-15 10:53:28 -04001751 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05001752out:
Al Viro2d92ab32008-08-02 00:51:11 -04001753 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 return err;
1755}
1756
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001757static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1758{
1759 int error = 0;
1760 int readonly_request = 0;
1761
1762 if (ms_flags & MS_RDONLY)
1763 readonly_request = 1;
1764 if (readonly_request == __mnt_is_readonly(mnt))
1765 return 0;
1766
Eric W. Biederman90563b12013-03-22 03:10:15 -07001767 if (mnt->mnt_flags & MNT_LOCK_READONLY)
1768 return -EPERM;
1769
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001770 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05001771 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001772 else
Al Viro83adc752011-11-24 22:37:54 -05001773 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001774 return error;
1775}
1776
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777/*
1778 * change filesystem flags. dir should be a physical root of filesystem.
1779 * If you've mounted a non-root directory somewhere and want to do remount
1780 * on it - tough luck.
1781 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001782static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 void *data)
1784{
1785 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001786 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05001787 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Al Viro143c8c92011-11-25 00:46:35 -05001789 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 return -EINVAL;
1791
Al Viro2d92ab32008-08-02 00:51:11 -04001792 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 return -EINVAL;
1794
Eric Parisff36fe22011-03-03 16:09:14 -05001795 err = security_sb_remount(sb, data);
1796 if (err)
1797 return err;
1798
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001800 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001801 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05001802 else if (!capable(CAP_SYS_ADMIN))
1803 err = -EPERM;
Al Viro4aa98cf2009-05-08 13:36:58 -04001804 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001805 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001806 if (!err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301807 br_write_lock(&vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001808 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1809 mnt->mnt.mnt_flags = mnt_flags;
Andi Kleen962830d2012-05-08 13:32:02 +09301810 br_write_unlock(&vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001813 if (!err) {
Andi Kleen962830d2012-05-08 13:32:02 +09301814 br_write_lock(&vfsmount_lock);
Al Viro143c8c92011-11-25 00:46:35 -05001815 touch_mnt_namespace(mnt->mnt_ns);
Andi Kleen962830d2012-05-08 13:32:02 +09301816 br_write_unlock(&vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 return err;
1819}
1820
Al Virocbbe3622011-11-24 20:01:19 -05001821static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05001822{
Al Viro315fc832011-11-24 18:57:30 -05001823 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001824 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001825 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05001826 return 1;
1827 }
1828 return 0;
1829}
1830
Al Viro808d4e32012-10-11 11:42:01 -04001831static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832{
Al Viro2d92ab32008-08-02 00:51:11 -04001833 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05001834 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05001835 struct mount *old;
Al Viro84d17192013-03-15 10:53:28 -04001836 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05001837 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 if (!old_name || !*old_name)
1839 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001840 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 if (err)
1842 return err;
1843
Al Viro84d17192013-03-15 10:53:28 -04001844 mp = lock_mount(path);
1845 err = PTR_ERR(mp);
1846 if (IS_ERR(mp))
David Howellscc53ce52011-01-14 18:45:26 +00001847 goto out;
1848
Al Viro143c8c92011-11-25 00:46:35 -05001849 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05001850 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05001851
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001853 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 goto out1;
1855
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001857 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001858 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
Al Viro676da582011-11-24 21:47:05 -05001860 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05001861 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
Al Viro2d92ab32008-08-02 00:51:11 -04001863 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1864 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001865 goto out1;
1866 /*
1867 * Don't move a mount residing in a shared parent.
1868 */
Al Virofc7be132011-11-25 01:05:37 -05001869 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001870 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001871 /*
1872 * Don't move a mount tree containing unbindable mounts to a destination
1873 * mount which is shared.
1874 */
Al Virofc7be132011-11-25 01:05:37 -05001875 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05001876 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05001878 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05001879 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05001880 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
Al Viro84d17192013-03-15 10:53:28 -04001882 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001883 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001884 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
1886 /* if the mount is moved, it should no longer be expire
1887 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05001888 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889out1:
Al Viro84d17192013-03-15 10:53:28 -04001890 unlock_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001893 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001894 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 return err;
1896}
1897
Al Viro9d412a42011-03-17 22:08:28 -04001898static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1899{
1900 int err;
1901 const char *subtype = strchr(fstype, '.');
1902 if (subtype) {
1903 subtype++;
1904 err = -EINVAL;
1905 if (!subtype[0])
1906 goto err;
1907 } else
1908 subtype = "";
1909
1910 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1911 err = -ENOMEM;
1912 if (!mnt->mnt_sb->s_subtype)
1913 goto err;
1914 return mnt;
1915
1916 err:
1917 mntput(mnt);
1918 return ERR_PTR(err);
1919}
1920
Al Viro9d412a42011-03-17 22:08:28 -04001921/*
1922 * add a mount into a namespace's mount tree
1923 */
Al Viro95bc5f22011-11-25 00:30:56 -05001924static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04001925{
Al Viro84d17192013-03-15 10:53:28 -04001926 struct mountpoint *mp;
1927 struct mount *parent;
Al Viro9d412a42011-03-17 22:08:28 -04001928 int err;
1929
1930 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1931
Al Viro84d17192013-03-15 10:53:28 -04001932 mp = lock_mount(path);
1933 if (IS_ERR(mp))
1934 return PTR_ERR(mp);
Al Viro9d412a42011-03-17 22:08:28 -04001935
Al Viro84d17192013-03-15 10:53:28 -04001936 parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04001937 err = -EINVAL;
Al Viro84d17192013-03-15 10:53:28 -04001938 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04001939 /* that's acceptable only for automounts done in private ns */
1940 if (!(mnt_flags & MNT_SHRINKABLE))
1941 goto unlock;
1942 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d17192013-03-15 10:53:28 -04001943 if (!parent->mnt_ns)
Al Viro156cacb2012-09-21 08:19:02 -04001944 goto unlock;
1945 }
Al Viro9d412a42011-03-17 22:08:28 -04001946
1947 /* Refuse the same filesystem on the same mount point */
1948 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05001949 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04001950 path->mnt->mnt_root == path->dentry)
1951 goto unlock;
1952
1953 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05001954 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04001955 goto unlock;
1956
Al Viro95bc5f22011-11-25 00:30:56 -05001957 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro84d17192013-03-15 10:53:28 -04001958 err = graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04001959
1960unlock:
Al Viro84d17192013-03-15 10:53:28 -04001961 unlock_mount(mp);
Al Viro9d412a42011-03-17 22:08:28 -04001962 return err;
1963}
Al Virob1e75df2011-01-17 01:47:59 -05001964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965/*
1966 * create a new mount for userspace and request it to be added into the
1967 * namespace's tree
1968 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001969static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04001970 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001972 struct file_system_type *type;
Al Viro9b40bc92013-02-22 22:45:42 -05001973 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05001975 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001977 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 return -EINVAL;
1979
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07001980 type = get_fs_type(fstype);
1981 if (!type)
1982 return -ENODEV;
1983
1984 if (user_ns != &init_user_ns) {
1985 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
1986 put_filesystem(type);
1987 return -EPERM;
1988 }
1989 /* Only in special cases allow devices from mounts
1990 * created outside the initial user namespace.
1991 */
1992 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
1993 flags |= MS_NODEV;
1994 mnt_flags |= MNT_NODEV;
1995 }
1996 }
1997
1998 mnt = vfs_kern_mount(type, flags, name, data);
1999 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2000 !mnt->mnt_sb->s_subtype)
2001 mnt = fs_set_subtype(mnt, fstype);
2002
2003 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 if (IS_ERR(mnt))
2005 return PTR_ERR(mnt);
2006
Al Viro95bc5f22011-11-25 00:30:56 -05002007 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002008 if (err)
2009 mntput(mnt);
2010 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011}
2012
Al Viro19a167a2011-01-17 01:35:23 -05002013int finish_automount(struct vfsmount *m, struct path *path)
2014{
Al Viro6776db3d2011-11-25 00:22:05 -05002015 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002016 int err;
2017 /* The new mount record should have at least 2 refs to prevent it being
2018 * expired before we get a chance to add it
2019 */
Al Viro6776db3d2011-11-25 00:22:05 -05002020 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002021
2022 if (m->mnt_sb == path->mnt->mnt_sb &&
2023 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002024 err = -ELOOP;
2025 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002026 }
2027
Al Viro95bc5f22011-11-25 00:30:56 -05002028 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002029 if (!err)
2030 return 0;
2031fail:
2032 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002033 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002034 namespace_lock();
Andi Kleen962830d2012-05-08 13:32:02 +09302035 br_write_lock(&vfsmount_lock);
Al Viro6776db3d2011-11-25 00:22:05 -05002036 list_del_init(&mnt->mnt_expire);
Andi Kleen962830d2012-05-08 13:32:02 +09302037 br_write_unlock(&vfsmount_lock);
Al Viro97216be2013-03-16 15:12:40 -04002038 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002039 }
Al Virob1e75df2011-01-17 01:47:59 -05002040 mntput(m);
2041 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002042 return err;
2043}
2044
David Howellsea5b7782011-01-14 19:10:03 +00002045/**
2046 * mnt_set_expiry - Put a mount on an expiration list
2047 * @mnt: The mount to list.
2048 * @expiry_list: The list to add the mount to.
2049 */
2050void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2051{
Al Viro97216be2013-03-16 15:12:40 -04002052 namespace_lock();
Andi Kleen962830d2012-05-08 13:32:02 +09302053 br_write_lock(&vfsmount_lock);
David Howellsea5b7782011-01-14 19:10:03 +00002054
Al Viro6776db3d2011-11-25 00:22:05 -05002055 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002056
Andi Kleen962830d2012-05-08 13:32:02 +09302057 br_write_unlock(&vfsmount_lock);
Al Viro97216be2013-03-16 15:12:40 -04002058 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002059}
2060EXPORT_SYMBOL(mnt_set_expiry);
2061
2062/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 * process a list of expirable mountpoints with the intent of discarding any
2064 * mountpoints that aren't in use and haven't been touched since last we came
2065 * here
2066 */
2067void mark_mounts_for_expiry(struct list_head *mounts)
2068{
Al Viro761d5c32011-11-24 21:07:43 -05002069 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 LIST_HEAD(graveyard);
2071
2072 if (list_empty(mounts))
2073 return;
2074
Al Viro97216be2013-03-16 15:12:40 -04002075 namespace_lock();
Andi Kleen962830d2012-05-08 13:32:02 +09302076 br_write_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
2078 /* extract from the expiration list every vfsmount that matches the
2079 * following criteria:
2080 * - only referenced by its parent vfsmount
2081 * - still marked for expiry (marked on the last call here; marks are
2082 * cleared by mntput())
2083 */
Al Viro6776db3d2011-11-25 00:22:05 -05002084 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002085 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002086 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002088 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002090 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002091 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002092 touch_mnt_namespace(mnt->mnt_ns);
Al Viro328e6d92013-03-16 14:49:45 -04002093 umount_tree(mnt, 1);
Al Virobcc5c7d2008-03-22 00:21:53 -04002094 }
Andi Kleen962830d2012-05-08 13:32:02 +09302095 br_write_unlock(&vfsmount_lock);
Al Viro3ab6abe2013-03-16 14:42:19 -04002096 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097}
2098
2099EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2100
2101/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002102 * Ripoff of 'select_parent()'
2103 *
2104 * search the list of submounts for a given mountpoint, and move any
2105 * shrinkable submounts to the 'graveyard' list.
2106 */
Al Viro692afc32011-11-24 21:15:14 -05002107static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002108{
Al Viro692afc32011-11-24 21:15:14 -05002109 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002110 struct list_head *next;
2111 int found = 0;
2112
2113repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002114 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002115resume:
Al Viro6b41d532011-11-24 23:24:33 -05002116 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002117 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002118 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002119
2120 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002121 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002122 continue;
2123 /*
2124 * Descend a level if the d_mounts list is non-empty.
2125 */
Al Viro6b41d532011-11-24 23:24:33 -05002126 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002127 this_parent = mnt;
2128 goto repeat;
2129 }
2130
Al Viro1ab59732011-11-24 21:35:16 -05002131 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002132 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002133 found++;
2134 }
2135 }
2136 /*
2137 * All done at this level ... ascend and resume the search
2138 */
2139 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002140 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002141 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002142 goto resume;
2143 }
2144 return found;
2145}
2146
2147/*
2148 * process a list of expirable mountpoints with the intent of discarding any
2149 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002150 *
2151 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002152 */
Al Virob54b9be2013-03-16 14:39:34 -04002153static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002154{
2155 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002156 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002157
Trond Myklebust5528f9112006-06-09 09:34:17 -04002158 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002159 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002160 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002161 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002162 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002163 touch_mnt_namespace(m->mnt_ns);
Al Viro328e6d92013-03-16 14:49:45 -04002164 umount_tree(m, 1);
Al Virobcc5c7d2008-03-22 00:21:53 -04002165 }
2166 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002167}
2168
Trond Myklebust5528f9112006-06-09 09:34:17 -04002169/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 * Some copy_from_user() implementations do not return the exact number of
2171 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2172 * Note that this function differs from copy_from_user() in that it will oops
2173 * on bad values of `to', rather than returning a short copy.
2174 */
Ram Paib58fed82005-11-07 17:16:09 -05002175static long exact_copy_from_user(void *to, const void __user * from,
2176 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177{
2178 char *t = to;
2179 const char __user *f = from;
2180 char c;
2181
2182 if (!access_ok(VERIFY_READ, from, n))
2183 return n;
2184
2185 while (n) {
2186 if (__get_user(c, f)) {
2187 memset(t, 0, n);
2188 break;
2189 }
2190 *t++ = c;
2191 f++;
2192 n--;
2193 }
2194 return n;
2195}
2196
Ram Paib58fed82005-11-07 17:16:09 -05002197int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198{
2199 int i;
2200 unsigned long page;
2201 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002202
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 *where = 0;
2204 if (!data)
2205 return 0;
2206
2207 if (!(page = __get_free_page(GFP_KERNEL)))
2208 return -ENOMEM;
2209
2210 /* We only care that *some* data at the address the user
2211 * gave us is valid. Just in case, we'll zero
2212 * the remainder of the page.
2213 */
2214 /* copy_from_user cannot cross TASK_SIZE ! */
2215 size = TASK_SIZE - (unsigned long)data;
2216 if (size > PAGE_SIZE)
2217 size = PAGE_SIZE;
2218
2219 i = size - exact_copy_from_user((void *)page, data, size);
2220 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002221 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 return -EFAULT;
2223 }
2224 if (i != PAGE_SIZE)
2225 memset((char *)page + i, 0, PAGE_SIZE - i);
2226 *where = page;
2227 return 0;
2228}
2229
Vegard Nossumeca6f532009-09-18 13:05:45 -07002230int copy_mount_string(const void __user *data, char **where)
2231{
2232 char *tmp;
2233
2234 if (!data) {
2235 *where = NULL;
2236 return 0;
2237 }
2238
2239 tmp = strndup_user(data, PAGE_SIZE);
2240 if (IS_ERR(tmp))
2241 return PTR_ERR(tmp);
2242
2243 *where = tmp;
2244 return 0;
2245}
2246
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247/*
2248 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2249 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2250 *
2251 * data is a (void *) that can point to any structure up to
2252 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2253 * information (or be NULL).
2254 *
2255 * Pre-0.97 versions of mount() didn't have a flags word.
2256 * When the flags word was introduced its top half was required
2257 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2258 * Therefore, if this magic number is present, it carries no information
2259 * and must be discarded.
2260 */
Al Viro808d4e32012-10-11 11:42:01 -04002261long do_mount(const char *dev_name, const char *dir_name,
2262 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263{
Al Viro2d92ab32008-08-02 00:51:11 -04002264 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 int retval = 0;
2266 int mnt_flags = 0;
2267
2268 /* Discard magic */
2269 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2270 flags &= ~MS_MGC_MSK;
2271
2272 /* Basic sanity checks */
2273
2274 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2275 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
2277 if (data_page)
2278 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2279
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002280 /* ... and get the mountpoint */
2281 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2282 if (retval)
2283 return retval;
2284
2285 retval = security_sb_mount(dev_name, &path,
2286 type_page, flags, data_page);
Al Viro0d5cadb2013-05-04 14:40:51 -04002287 if (!retval && !may_mount())
2288 retval = -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002289 if (retval)
2290 goto dput_out;
2291
Andi Kleen613cbe32009-04-19 18:40:43 +02002292 /* Default to relatime unless overriden */
2293 if (!(flags & MS_NOATIME))
2294 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002295
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 /* Separate the per-mountpoint flags */
2297 if (flags & MS_NOSUID)
2298 mnt_flags |= MNT_NOSUID;
2299 if (flags & MS_NODEV)
2300 mnt_flags |= MNT_NODEV;
2301 if (flags & MS_NOEXEC)
2302 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002303 if (flags & MS_NOATIME)
2304 mnt_flags |= MNT_NOATIME;
2305 if (flags & MS_NODIRATIME)
2306 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002307 if (flags & MS_STRICTATIME)
2308 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002309 if (flags & MS_RDONLY)
2310 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002311
Al Viro7a4dec52010-08-09 12:05:43 -04002312 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002313 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2314 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002317 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 data_page);
2319 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002320 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002321 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002322 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002324 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 else
Al Viro2d92ab32008-08-02 00:51:11 -04002326 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 dev_name, data_page);
2328dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002329 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 return retval;
2331}
2332
Eric W. Biederman771b1372012-07-26 21:08:32 -07002333static void free_mnt_ns(struct mnt_namespace *ns)
2334{
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002335 proc_free_inum(ns->proc_inum);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002336 put_user_ns(ns->user_ns);
2337 kfree(ns);
2338}
2339
Eric W. Biederman8823c072010-03-07 18:49:36 -08002340/*
2341 * Assign a sequence number so we can detect when we attempt to bind
2342 * mount a reference to an older mount namespace into the current
2343 * mount namespace, preventing reference counting loops. A 64bit
2344 * number incrementing at 10Ghz will take 12,427 years to wrap which
2345 * is effectively never, so we can ignore the possibility.
2346 */
2347static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2348
Eric W. Biederman771b1372012-07-26 21:08:32 -07002349static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002350{
2351 struct mnt_namespace *new_ns;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002352 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002353
2354 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2355 if (!new_ns)
2356 return ERR_PTR(-ENOMEM);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002357 ret = proc_alloc_inum(&new_ns->proc_inum);
2358 if (ret) {
2359 kfree(new_ns);
2360 return ERR_PTR(ret);
2361 }
Eric W. Biederman8823c072010-03-07 18:49:36 -08002362 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002363 atomic_set(&new_ns->count, 1);
2364 new_ns->root = NULL;
2365 INIT_LIST_HEAD(&new_ns->list);
2366 init_waitqueue_head(&new_ns->poll);
2367 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002368 new_ns->user_ns = get_user_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002369 return new_ns;
2370}
2371
JANAK DESAI741a2952006-02-07 12:59:00 -08002372/*
2373 * Allocate a new namespace structure and populate it with contents
2374 * copied from the namespace of the passed in task structure.
2375 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002376static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Eric W. Biederman771b1372012-07-26 21:08:32 -07002377 struct user_namespace *user_ns, struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002379 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002380 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002381 struct mount *p, *q;
Al Virobe08d6d2011-12-06 13:32:36 -05002382 struct mount *old = mnt_ns->root;
Al Virocb338d02011-11-24 20:55:08 -05002383 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002384 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
Eric W. Biederman771b1372012-07-26 21:08:32 -07002386 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002387 if (IS_ERR(new_ns))
2388 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389
Al Viro97216be2013-03-16 15:12:40 -04002390 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 /* First pass: copy the tree topology */
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002392 copy_flags = CL_COPY_ALL | CL_EXPIRE;
2393 if (user_ns != mnt_ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002394 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002395 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002396 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04002397 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002398 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002399 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 }
Al Virobe08d6d2011-12-06 13:32:36 -05002401 new_ns->root = new;
Andi Kleen962830d2012-05-08 13:32:02 +09302402 br_write_lock(&vfsmount_lock);
Al Viro1a4eeaf2011-11-25 02:19:55 -05002403 list_add_tail(&new_ns->list, &new->mnt_list);
Andi Kleen962830d2012-05-08 13:32:02 +09302404 br_write_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
2406 /*
2407 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2408 * as belonging to new namespace. We have already acquired a private
2409 * fs_struct, so tsk->fs->lock is not needed.
2410 */
Al Viro909b0a82011-11-25 03:06:56 -05002411 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002412 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002414 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 if (fs) {
Al Viro315fc832011-11-24 18:57:30 -05002416 if (&p->mnt == fs->root.mnt) {
2417 fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002418 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 }
Al Viro315fc832011-11-24 18:57:30 -05002420 if (&p->mnt == fs->pwd.mnt) {
2421 fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002422 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 }
Al Viro909b0a82011-11-25 03:06:56 -05002425 p = next_mnt(p, old);
2426 q = next_mnt(q, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 }
Al Viro328e6d92013-03-16 14:49:45 -04002428 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002431 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002433 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434
JANAK DESAI741a2952006-02-07 12:59:00 -08002435 return new_ns;
2436}
2437
Eric W. Biederman213dd262007-07-15 23:41:15 -07002438struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Eric W. Biederman771b1372012-07-26 21:08:32 -07002439 struct user_namespace *user_ns, struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002440{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002441 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002442
Badari Pulavartye3222c42007-05-08 00:25:21 -07002443 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002444 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002445
2446 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002447 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002448
Eric W. Biederman771b1372012-07-26 21:08:32 -07002449 new_ns = dup_mnt_ns(ns, user_ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002450
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002451 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002452 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453}
2454
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002455/**
2456 * create_mnt_ns - creates a private namespace and adds a root filesystem
2457 * @mnt: pointer to the new root filesystem mountpoint
2458 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002459static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002460{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002461 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002462 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002463 struct mount *mnt = real_mount(m);
2464 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002465 new_ns->root = mnt;
Al Virob1983cd2013-05-04 15:18:53 -04002466 list_add(&mnt->mnt_list, &new_ns->list);
Al Viroc1334492011-11-16 16:12:14 -05002467 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002468 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002469 }
2470 return new_ns;
2471}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002472
Al Viroea441d12011-11-16 21:43:59 -05002473struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2474{
2475 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002476 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002477 struct path path;
2478 int err;
2479
2480 ns = create_mnt_ns(mnt);
2481 if (IS_ERR(ns))
2482 return ERR_CAST(ns);
2483
2484 err = vfs_path_lookup(mnt->mnt_root, mnt,
2485 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2486
2487 put_mnt_ns(ns);
2488
2489 if (err)
2490 return ERR_PTR(err);
2491
2492 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002493 s = path.mnt->mnt_sb;
2494 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002495 mntput(path.mnt);
2496 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002497 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002498 /* ... and return the root of (sub)tree on it */
2499 return path.dentry;
2500}
2501EXPORT_SYMBOL(mount_subtree);
2502
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002503SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2504 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002506 int ret;
2507 char *kernel_type;
Jeff Layton91a27b22012-10-10 15:25:28 -04002508 struct filename *kernel_dir;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002509 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
Vegard Nossumeca6f532009-09-18 13:05:45 -07002512 ret = copy_mount_string(type, &kernel_type);
2513 if (ret < 0)
2514 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515
Vegard Nossumeca6f532009-09-18 13:05:45 -07002516 kernel_dir = getname(dir_name);
2517 if (IS_ERR(kernel_dir)) {
2518 ret = PTR_ERR(kernel_dir);
2519 goto out_dir;
2520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
Vegard Nossumeca6f532009-09-18 13:05:45 -07002522 ret = copy_mount_string(dev_name, &kernel_dev);
2523 if (ret < 0)
2524 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
Vegard Nossumeca6f532009-09-18 13:05:45 -07002526 ret = copy_mount_options(data, &data_page);
2527 if (ret < 0)
2528 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529
Jeff Layton91a27b22012-10-10 15:25:28 -04002530 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
Vegard Nossumeca6f532009-09-18 13:05:45 -07002531 (void *) data_page);
2532
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002534out_data:
2535 kfree(kernel_dev);
2536out_dev:
2537 putname(kernel_dir);
2538out_dir:
2539 kfree(kernel_type);
2540out_type:
2541 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542}
2543
2544/*
Al Viroafac7cb2011-11-23 19:34:49 -05002545 * Return true if path is reachable from root
2546 *
2547 * namespace_sem or vfsmount_lock is held
2548 */
Al Viro643822b2011-11-24 22:00:28 -05002549bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002550 const struct path *root)
2551{
Al Viro643822b2011-11-24 22:00:28 -05002552 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002553 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002554 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002555 }
Al Viro643822b2011-11-24 22:00:28 -05002556 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002557}
2558
2559int path_is_under(struct path *path1, struct path *path2)
2560{
2561 int res;
Andi Kleen962830d2012-05-08 13:32:02 +09302562 br_read_lock(&vfsmount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002563 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Andi Kleen962830d2012-05-08 13:32:02 +09302564 br_read_unlock(&vfsmount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002565 return res;
2566}
2567EXPORT_SYMBOL(path_is_under);
2568
2569/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 * pivot_root Semantics:
2571 * Moves the root file system of the current process to the directory put_old,
2572 * makes new_root as the new root file system of the current process, and sets
2573 * root/cwd of all processes which had them on the current root to new_root.
2574 *
2575 * Restrictions:
2576 * The new_root and put_old must be directories, and must not be on the
2577 * same file system as the current process root. The put_old must be
2578 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2579 * pointed to by put_old must yield the same directory as new_root. No other
2580 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2581 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002582 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2583 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2584 * in this situation.
2585 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 * Notes:
2587 * - we don't move root/cwd if they are not at the root (reason: if something
2588 * cared enough to change them, it's probably wrong to force them elsewhere)
2589 * - it's okay to pick a root that isn't the root of a file system, e.g.
2590 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2591 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2592 * first.
2593 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002594SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2595 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596{
Al Viro2d8f3032008-07-22 09:59:21 -04002597 struct path new, old, parent_path, root_parent, root;
Al Viro84d17192013-03-15 10:53:28 -04002598 struct mount *new_mnt, *root_mnt, *old_mnt;
2599 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 int error;
2601
Al Viro9b40bc92013-02-22 22:45:42 -05002602 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 return -EPERM;
2604
Al Viro2d8f3032008-07-22 09:59:21 -04002605 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 if (error)
2607 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608
Al Viro2d8f3032008-07-22 09:59:21 -04002609 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 if (error)
2611 goto out1;
2612
Al Viro2d8f3032008-07-22 09:59:21 -04002613 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002614 if (error)
2615 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002617 get_fs_root(current->fs, &root);
Al Viro84d17192013-03-15 10:53:28 -04002618 old_mp = lock_mount(&old);
2619 error = PTR_ERR(old_mp);
2620 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04002621 goto out3;
2622
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002624 new_mnt = real_mount(new.mnt);
2625 root_mnt = real_mount(root.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002626 old_mnt = real_mount(old.mnt);
2627 if (IS_MNT_SHARED(old_mnt) ||
Al Virofc7be132011-11-25 01:05:37 -05002628 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2629 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002630 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002631 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002632 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002634 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002635 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 error = -EBUSY;
Al Viro84d17192013-03-15 10:53:28 -04002637 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04002638 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002640 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002641 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002642 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002643 goto out4; /* not attached */
Al Viro84d17192013-03-15 10:53:28 -04002644 root_mp = root_mnt->mnt_mp;
Al Viro2d8f3032008-07-22 09:59:21 -04002645 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002646 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002647 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002648 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002649 /* make sure we can reach put_old from new_root */
Al Viro84d17192013-03-15 10:53:28 -04002650 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002651 goto out4;
Al Viro84d17192013-03-15 10:53:28 -04002652 root_mp->m_count++; /* pin it so it won't go away */
Andi Kleen962830d2012-05-08 13:32:02 +09302653 br_write_lock(&vfsmount_lock);
Al Viro419148d2011-11-24 19:41:16 -05002654 detach_mnt(new_mnt, &parent_path);
2655 detach_mnt(root_mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002656 /* mount old root on put_old */
Al Viro84d17192013-03-15 10:53:28 -04002657 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08002658 /* mount new_root on / */
Al Viro84d17192013-03-15 10:53:28 -04002659 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002660 touch_mnt_namespace(current->nsproxy->mnt_ns);
Andi Kleen962830d2012-05-08 13:32:02 +09302661 br_write_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002662 chroot_fs_refs(&root, &new);
Al Viro84d17192013-03-15 10:53:28 -04002663 put_mountpoint(root_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002665out4:
Al Viro84d17192013-03-15 10:53:28 -04002666 unlock_mount(old_mp);
Al Virob12cea92011-03-18 08:55:38 -04002667 if (!error) {
2668 path_put(&root_parent);
2669 path_put(&parent_path);
2670 }
2671out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002672 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002673out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002674 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002676 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679}
2680
2681static void __init init_mount_tree(void)
2682{
2683 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002684 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002685 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002686 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002688 type = get_fs_type("rootfs");
2689 if (!type)
2690 panic("Can't find rootfs type");
2691 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2692 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 if (IS_ERR(mnt))
2694 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002695
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002696 ns = create_mnt_ns(mnt);
2697 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002700 init_task.nsproxy->mnt_ns = ns;
2701 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702
Al Virobe08d6d2011-12-06 13:32:36 -05002703 root.mnt = mnt;
2704 root.dentry = mnt->mnt_root;
Jan Blunckac748a02008-02-14 19:34:39 -08002705
2706 set_fs_pwd(current->fs, &root);
2707 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708}
2709
Denis Cheng74bf17c2007-10-16 23:26:30 -07002710void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002712 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002713 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
Ram Pai390c6842005-11-07 17:17:51 -05002715 init_rwsem(&namespace_sem);
2716
Al Viro7d6fec42011-11-23 12:14:10 -05002717 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002718 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719
Ram Paib58fed82005-11-07 17:16:09 -05002720 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Al Viro84d17192013-03-15 10:53:28 -04002721 mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722
Al Viro84d17192013-03-15 10:53:28 -04002723 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 panic("Failed to allocate mount hash table\n");
2725
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002726 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727
Eric Dumazet13f14b42008-02-06 01:37:57 -08002728 for (u = 0; u < HASH_SIZE; u++)
2729 INIT_LIST_HEAD(&mount_hashtable[u]);
Al Viro84d17192013-03-15 10:53:28 -04002730 for (u = 0; u < HASH_SIZE; u++)
2731 INIT_LIST_HEAD(&mountpoint_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732
Andi Kleen962830d2012-05-08 13:32:02 +09302733 br_lock_init(&vfsmount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +10002734
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002735 err = sysfs_init();
2736 if (err)
2737 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002738 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002739 fs_kobj = kobject_create_and_add("fs", NULL);
2740 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002741 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 init_rootfs();
2743 init_mount_tree();
2744}
2745
Trond Myklebust616511d2009-06-22 15:09:13 -04002746void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747{
Al Virod498b252010-02-05 02:21:06 -05002748 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002749 return;
Al Viro97216be2013-03-16 15:12:40 -04002750 namespace_lock();
Andi Kleen962830d2012-05-08 13:32:02 +09302751 br_write_lock(&vfsmount_lock);
Al Viro328e6d92013-03-16 14:49:45 -04002752 umount_tree(ns->root, 0);
Andi Kleen962830d2012-05-08 13:32:02 +09302753 br_write_unlock(&vfsmount_lock);
Al Viro3ab6abe2013-03-16 14:42:19 -04002754 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002755 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756}
Al Viro9d412a42011-03-17 22:08:28 -04002757
2758struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2759{
Tim Chen423e0ab2011-07-19 09:32:38 -07002760 struct vfsmount *mnt;
2761 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2762 if (!IS_ERR(mnt)) {
2763 /*
2764 * it is a longterm mount, don't release mnt until
2765 * we unmount before file sys is unregistered
2766 */
Al Virof7a99c52012-06-09 00:59:08 -04002767 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07002768 }
2769 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002770}
2771EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002772
2773void kern_unmount(struct vfsmount *mnt)
2774{
2775 /* release long term mount so mount point can be released */
2776 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04002777 br_write_lock(&vfsmount_lock);
2778 real_mount(mnt)->mnt_ns = NULL;
2779 br_write_unlock(&vfsmount_lock);
Tim Chen423e0ab2011-07-19 09:32:38 -07002780 mntput(mnt);
2781 }
2782}
2783EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002784
2785bool our_mnt(struct vfsmount *mnt)
2786{
Al Viro143c8c92011-11-25 00:46:35 -05002787 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05002788}
Eric W. Biederman8823c072010-03-07 18:49:36 -08002789
Eric W. Biederman31515272013-03-15 01:45:51 -07002790bool current_chrooted(void)
2791{
2792 /* Does the current process have a non-standard root */
2793 struct path ns_root;
2794 struct path fs_root;
2795 bool chrooted;
2796
2797 /* Find the namespace root */
2798 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
2799 ns_root.dentry = ns_root.mnt->mnt_root;
2800 path_get(&ns_root);
2801 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2802 ;
2803
2804 get_fs_root(current->fs, &fs_root);
2805
2806 chrooted = !path_equal(&fs_root, &ns_root);
2807
2808 path_put(&fs_root);
2809 path_put(&ns_root);
2810
2811 return chrooted;
2812}
2813
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002814void update_mnt_policy(struct user_namespace *userns)
2815{
2816 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2817 struct mount *mnt;
2818
2819 down_read(&namespace_sem);
2820 list_for_each_entry(mnt, &ns->list, mnt_list) {
2821 switch (mnt->mnt.mnt_sb->s_magic) {
2822 case SYSFS_MAGIC:
2823 userns->may_mount_sysfs = true;
2824 break;
2825 case PROC_SUPER_MAGIC:
2826 userns->may_mount_proc = true;
2827 break;
2828 }
2829 if (userns->may_mount_sysfs && userns->may_mount_proc)
2830 break;
2831 }
2832 up_read(&namespace_sem);
2833}
2834
Eric W. Biederman8823c072010-03-07 18:49:36 -08002835static void *mntns_get(struct task_struct *task)
2836{
2837 struct mnt_namespace *ns = NULL;
2838 struct nsproxy *nsproxy;
2839
2840 rcu_read_lock();
2841 nsproxy = task_nsproxy(task);
2842 if (nsproxy) {
2843 ns = nsproxy->mnt_ns;
2844 get_mnt_ns(ns);
2845 }
2846 rcu_read_unlock();
2847
2848 return ns;
2849}
2850
2851static void mntns_put(void *ns)
2852{
2853 put_mnt_ns(ns);
2854}
2855
2856static int mntns_install(struct nsproxy *nsproxy, void *ns)
2857{
2858 struct fs_struct *fs = current->fs;
2859 struct mnt_namespace *mnt_ns = ns;
2860 struct path root;
2861
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002862 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biederman5e4a0842012-12-14 07:55:36 -08002863 !nsown_capable(CAP_SYS_CHROOT) ||
2864 !nsown_capable(CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08002865 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002866
2867 if (fs->users != 1)
2868 return -EINVAL;
2869
2870 get_mnt_ns(mnt_ns);
2871 put_mnt_ns(nsproxy->mnt_ns);
2872 nsproxy->mnt_ns = mnt_ns;
2873
2874 /* Find the root */
2875 root.mnt = &mnt_ns->root->mnt;
2876 root.dentry = mnt_ns->root->mnt.mnt_root;
2877 path_get(&root);
2878 while(d_mountpoint(root.dentry) && follow_down_one(&root))
2879 ;
2880
2881 /* Update the pwd and root */
2882 set_fs_pwd(fs, &root);
2883 set_fs_root(fs, &root);
2884
2885 path_put(&root);
2886 return 0;
2887}
2888
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002889static unsigned int mntns_inum(void *ns)
2890{
2891 struct mnt_namespace *mnt_ns = ns;
2892 return mnt_ns->proc_inum;
2893}
2894
Eric W. Biederman8823c072010-03-07 18:49:36 -08002895const struct proc_ns_operations mntns_operations = {
2896 .name = "mnt",
2897 .type = CLONE_NEWNS,
2898 .get = mntns_get,
2899 .put = mntns_put,
2900 .install = mntns_install,
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002901 .inum = mntns_inum,
Eric W. Biederman8823c072010-03-07 18:49:36 -08002902};