blob: cd1e9681a0cf5d0bdfd64d84ae90c34bd10f265b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
Al Virod10577a2011-12-07 13:06:11 -050012#include <linux/export.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080013#include <linux/capability.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080014#include <linux/mnt_namespace.h>
Eric W. Biederman771b1372012-07-26 21:08:32 -070015#include <linux/user_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/namei.h>
17#include <linux/security.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010018#include <linux/idr.h>
Rob Landley57f150a2013-09-11 14:26:10 -070019#include <linux/init.h> /* init_rootfs */
Al Virod10577a2011-12-07 13:06:11 -050020#include <linux/fs_struct.h> /* get_fs_root et.al. */
21#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22#include <linux/uaccess.h>
David Howells0bb80f22013-04-12 01:50:06 +010023#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070024#include <linux/magic.h>
Al Viro0818bf22014-02-28 13:46:44 -050025#include <linux/bootmem.h>
Al Viro9ea459e2014-08-08 13:08:20 -040026#include <linux/task_work.h>
Ram Pai07b20882005-11-07 17:19:07 -050027#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070028#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Al Viro0818bf22014-02-28 13:46:44 -050030static unsigned int m_hash_mask __read_mostly;
31static unsigned int m_hash_shift __read_mostly;
32static unsigned int mp_hash_mask __read_mostly;
33static unsigned int mp_hash_shift __read_mostly;
34
35static __initdata unsigned long mhash_entries;
36static int __init set_mhash_entries(char *str)
37{
38 if (!str)
39 return 0;
40 mhash_entries = simple_strtoul(str, &str, 0);
41 return 1;
42}
43__setup("mhash_entries=", set_mhash_entries);
44
45static __initdata unsigned long mphash_entries;
46static int __init set_mphash_entries(char *str)
47{
48 if (!str)
49 return 0;
50 mphash_entries = simple_strtoul(str, &str, 0);
51 return 1;
52}
53__setup("mphash_entries=", set_mphash_entries);
Eric Dumazet13f14b42008-02-06 01:37:57 -080054
Al Viroc7999c32014-02-27 14:40:10 -050055static u64 event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010056static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010057static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100058static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040059static int mnt_id_start = 0;
60static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Al Viro38129a12014-03-20 21:10:51 -040062static struct hlist_head *mount_hashtable __read_mostly;
Al Viro0818bf22014-02-28 13:46:44 -050063static struct hlist_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080064static struct kmem_cache *mnt_cache __read_mostly;
Al Viro59aa0da2013-09-16 21:34:53 -040065static DECLARE_RWSEM(namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080067/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060068struct kobject *fs_kobj;
69EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080070
Nick Piggin99b7db72010-08-18 04:37:39 +100071/*
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
74 * up the tree.
75 *
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
78 */
Al Viro48a066e2013-09-29 22:06:07 -040079__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +100080
Al Viro38129a12014-03-20 21:10:51 -040081static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082{
Ram Paib58fed82005-11-07 17:16:09 -050083 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Al Viro0818bf22014-02-28 13:46:44 -050085 tmp = tmp + (tmp >> m_hash_shift);
86 return &mount_hashtable[tmp & m_hash_mask];
87}
88
89static inline struct hlist_head *mp_hash(struct dentry *dentry)
90{
91 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 tmp = tmp + (tmp >> mp_hash_shift);
93 return &mountpoint_hashtable[tmp & mp_hash_mask];
Linus Torvalds1da177e2005-04-16 15:20:36 -070094}
95
Nick Piggin99b7db72010-08-18 04:37:39 +100096/*
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
99 */
Al Virob105e272011-11-24 20:38:33 -0500100static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100101{
102 int res;
103
104retry:
105 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +1000106 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -0500107 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -0400108 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500109 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +1000110 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100111 if (res == -EAGAIN)
112 goto retry;
113
114 return res;
115}
116
Al Virob105e272011-11-24 20:38:33 -0500117static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100118{
Al Viro15169fe2011-11-25 00:50:41 -0500119 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000120 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -0400121 ida_remove(&mnt_id_ida, id);
122 if (mnt_id_start > id)
123 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000124 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100125}
126
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100127/*
128 * Allocate a new peer group ID
129 *
130 * mnt_group_ida is protected by namespace_sem
131 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500132static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100133{
Al Virof21f6222009-06-24 03:12:00 -0400134 int res;
135
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100136 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
137 return -ENOMEM;
138
Al Virof21f6222009-06-24 03:12:00 -0400139 res = ida_get_new_above(&mnt_group_ida,
140 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500141 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400142 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500143 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400144
145 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100146}
147
148/*
149 * Release a peer group ID
150 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500151void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100152{
Al Viro15169fe2011-11-25 00:50:41 -0500153 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400154 ida_remove(&mnt_group_ida, id);
155 if (mnt_group_start > id)
156 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500157 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100158}
159
Nick Pigginb3e19d92011-01-07 17:50:11 +1100160/*
161 * vfsmount lock must be held for read
162 */
Al Viro83adc752011-11-24 22:37:54 -0500163static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100164{
165#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500166 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100167#else
168 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500169 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100170 preempt_enable();
171#endif
172}
173
Nick Pigginb3e19d92011-01-07 17:50:11 +1100174/*
175 * vfsmount lock must be held for write
176 */
Al Viro83adc752011-11-24 22:37:54 -0500177unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100178{
179#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500180 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100181 int cpu;
182
183 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500184 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100185 }
186
187 return count;
188#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500189 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100190#endif
191}
192
Al Virob105e272011-11-24 20:38:33 -0500193static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194{
Al Viroc63181e2011-11-25 02:35:16 -0500195 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
196 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100197 int err;
198
Al Viroc63181e2011-11-25 02:35:16 -0500199 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800200 if (err)
201 goto out_free_cache;
202
203 if (name) {
Al Viroc63181e2011-11-25 02:35:16 -0500204 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
205 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800206 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100207 }
208
Nick Pigginb3e19d92011-01-07 17:50:11 +1100209#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500210 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
211 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100212 goto out_free_devname;
213
Al Viroc63181e2011-11-25 02:35:16 -0500214 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100215#else
Al Viroc63181e2011-11-25 02:35:16 -0500216 mnt->mnt_count = 1;
217 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100218#endif
219
Al Viro38129a12014-03-20 21:10:51 -0400220 INIT_HLIST_NODE(&mnt->mnt_hash);
Al Viroc63181e2011-11-25 02:35:16 -0500221 INIT_LIST_HEAD(&mnt->mnt_child);
222 INIT_LIST_HEAD(&mnt->mnt_mounts);
223 INIT_LIST_HEAD(&mnt->mnt_list);
224 INIT_LIST_HEAD(&mnt->mnt_expire);
225 INIT_LIST_HEAD(&mnt->mnt_share);
226 INIT_LIST_HEAD(&mnt->mnt_slave_list);
227 INIT_LIST_HEAD(&mnt->mnt_slave);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700228 INIT_HLIST_NODE(&mnt->mnt_mp_list);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500229#ifdef CONFIG_FSNOTIFY
230 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
231#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 }
Al Viroc63181e2011-11-25 02:35:16 -0500233 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800234
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000235#ifdef CONFIG_SMP
236out_free_devname:
Al Viroc63181e2011-11-25 02:35:16 -0500237 kfree(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000238#endif
Li Zefan88b38782008-07-21 18:06:36 +0800239out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500240 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800241out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500242 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800243 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
245
Dave Hansen83660252008-02-15 14:37:30 -0800246/*
247 * Most r/o checks on a fs are for operations that take
248 * discrete amounts of time, like a write() or unlink().
249 * We must keep track of when those operations start
250 * (for permission checks) and when they end, so that
251 * we can determine when writes are able to occur to
252 * a filesystem.
253 */
Dave Hansen3d733632008-02-15 14:37:59 -0800254/*
255 * __mnt_is_readonly: check whether a mount is read-only
256 * @mnt: the mount to check for its write status
257 *
258 * This shouldn't be used directly ouside of the VFS.
259 * It does not guarantee that the filesystem will stay
260 * r/w, just that it is right *now*. This can not and
261 * should not be used in place of IS_RDONLY(inode).
262 * mnt_want/drop_write() will _keep_ the filesystem
263 * r/w.
264 */
265int __mnt_is_readonly(struct vfsmount *mnt)
266{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800267 if (mnt->mnt_flags & MNT_READONLY)
268 return 1;
269 if (mnt->mnt_sb->s_flags & MS_RDONLY)
270 return 1;
271 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800272}
273EXPORT_SYMBOL_GPL(__mnt_is_readonly);
274
Al Viro83adc752011-11-24 22:37:54 -0500275static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800276{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000277#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500278 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000279#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500280 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000281#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800282}
Dave Hansen3d733632008-02-15 14:37:59 -0800283
Al Viro83adc752011-11-24 22:37:54 -0500284static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800285{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000286#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500287 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000288#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500289 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000290#endif
291}
292
Al Viro83adc752011-11-24 22:37:54 -0500293static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000294{
295#ifdef CONFIG_SMP
296 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800297 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800298
299 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500300 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800301 }
Dave Hansen3d733632008-02-15 14:37:59 -0800302
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000303 return count;
304#else
305 return mnt->mnt_writers;
306#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800307}
308
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100309static int mnt_is_readonly(struct vfsmount *mnt)
310{
311 if (mnt->mnt_sb->s_readonly_remount)
312 return 1;
313 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
314 smp_rmb();
315 return __mnt_is_readonly(mnt);
316}
317
Dave Hansen3d733632008-02-15 14:37:59 -0800318/*
Jan Karaeb04c282012-06-12 16:20:35 +0200319 * Most r/o & frozen checks on a fs are for operations that take discrete
320 * amounts of time, like a write() or unlink(). We must keep track of when
321 * those operations start (for permission checks) and when they end, so that we
322 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800323 */
Dave Hansen83660252008-02-15 14:37:30 -0800324/**
Jan Karaeb04c282012-06-12 16:20:35 +0200325 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500326 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800327 *
Jan Karaeb04c282012-06-12 16:20:35 +0200328 * This tells the low-level filesystem that a write is about to be performed to
329 * it, and makes sure that writes are allowed (mnt it read-write) before
330 * returning success. This operation does not protect against filesystem being
331 * frozen. When the write operation is finished, __mnt_drop_write() must be
332 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800333 */
Jan Karaeb04c282012-06-12 16:20:35 +0200334int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800335{
Al Viro83adc752011-11-24 22:37:54 -0500336 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800337 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800338
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000339 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100340 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000341 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100342 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000343 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
344 * incremented count after it has set MNT_WRITE_HOLD.
345 */
346 smp_mb();
Miao Xie1e755292012-11-16 17:23:50 +0800347 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000348 cpu_relax();
349 /*
350 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
351 * be set to match its requirements. So we must not load that until
352 * MNT_WRITE_HOLD is cleared.
353 */
354 smp_rmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100355 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100356 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800357 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800358 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000359 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200360
361 return ret;
362}
363
364/**
365 * mnt_want_write - get write access to a mount
366 * @m: the mount on which to take a write
367 *
368 * This tells the low-level filesystem that a write is about to be performed to
369 * it, and makes sure that writes are allowed (mount is read-write, filesystem
370 * is not frozen) before returning success. When the write operation is
371 * finished, mnt_drop_write() must be called. This is effectively a refcount.
372 */
373int mnt_want_write(struct vfsmount *m)
374{
375 int ret;
376
377 sb_start_write(m->mnt_sb);
378 ret = __mnt_want_write(m);
379 if (ret)
380 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800381 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800382}
383EXPORT_SYMBOL_GPL(mnt_want_write);
384
385/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000386 * mnt_clone_write - get write access to a mount
387 * @mnt: the mount on which to take a write
388 *
389 * This is effectively like mnt_want_write, except
390 * it must only be used to take an extra write reference
391 * on a mountpoint that we already know has a write reference
392 * on it. This allows some optimisation.
393 *
394 * After finished, mnt_drop_write must be called as usual to
395 * drop the reference.
396 */
397int mnt_clone_write(struct vfsmount *mnt)
398{
399 /* superblock may be r/o */
400 if (__mnt_is_readonly(mnt))
401 return -EROFS;
402 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500403 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000404 preempt_enable();
405 return 0;
406}
407EXPORT_SYMBOL_GPL(mnt_clone_write);
408
409/**
Jan Karaeb04c282012-06-12 16:20:35 +0200410 * __mnt_want_write_file - get write access to a file's mount
411 * @file: the file who's mount on which to take a write
412 *
413 * This is like __mnt_want_write, but it takes a file and can
414 * do some optimisations if the file is open for write already
415 */
416int __mnt_want_write_file(struct file *file)
417{
Al Viro83f936c2014-03-14 12:02:47 -0400418 if (!(file->f_mode & FMODE_WRITER))
Jan Karaeb04c282012-06-12 16:20:35 +0200419 return __mnt_want_write(file->f_path.mnt);
420 else
421 return mnt_clone_write(file->f_path.mnt);
422}
423
424/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000425 * mnt_want_write_file - get write access to a file's mount
426 * @file: the file who's mount on which to take a write
427 *
428 * This is like mnt_want_write, but it takes a file and can
429 * do some optimisations if the file is open for write already
430 */
431int mnt_want_write_file(struct file *file)
432{
Jan Karaeb04c282012-06-12 16:20:35 +0200433 int ret;
434
435 sb_start_write(file->f_path.mnt->mnt_sb);
436 ret = __mnt_want_write_file(file);
437 if (ret)
438 sb_end_write(file->f_path.mnt->mnt_sb);
439 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000440}
441EXPORT_SYMBOL_GPL(mnt_want_write_file);
442
443/**
Jan Karaeb04c282012-06-12 16:20:35 +0200444 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800445 * @mnt: the mount on which to give up write access
446 *
447 * Tells the low-level filesystem that we are done
448 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200449 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800450 */
Jan Karaeb04c282012-06-12 16:20:35 +0200451void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800452{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000453 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500454 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000455 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800456}
Jan Karaeb04c282012-06-12 16:20:35 +0200457
458/**
459 * mnt_drop_write - give up write access to a mount
460 * @mnt: the mount on which to give up write access
461 *
462 * Tells the low-level filesystem that we are done performing writes to it and
463 * also allows filesystem to be frozen again. Must be matched with
464 * mnt_want_write() call above.
465 */
466void mnt_drop_write(struct vfsmount *mnt)
467{
468 __mnt_drop_write(mnt);
469 sb_end_write(mnt->mnt_sb);
470}
Dave Hansen83660252008-02-15 14:37:30 -0800471EXPORT_SYMBOL_GPL(mnt_drop_write);
472
Jan Karaeb04c282012-06-12 16:20:35 +0200473void __mnt_drop_write_file(struct file *file)
474{
475 __mnt_drop_write(file->f_path.mnt);
476}
477
Al Viro2a79f172011-12-09 08:06:57 -0500478void mnt_drop_write_file(struct file *file)
479{
480 mnt_drop_write(file->f_path.mnt);
481}
482EXPORT_SYMBOL(mnt_drop_write_file);
483
Al Viro83adc752011-11-24 22:37:54 -0500484static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800485{
Dave Hansen3d733632008-02-15 14:37:59 -0800486 int ret = 0;
487
Al Viro719ea2f2013-09-29 11:24:49 -0400488 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500489 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000490 /*
491 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
492 * should be visible before we do.
493 */
494 smp_mb();
495
496 /*
497 * With writers on hold, if this value is zero, then there are
498 * definitely no active writers (although held writers may subsequently
499 * increment the count, they'll have to wait, and decrement it after
500 * seeing MNT_READONLY).
501 *
502 * It is OK to have counter incremented on one CPU and decremented on
503 * another: the sum will add up correctly. The danger would be when we
504 * sum up each counter, if we read a counter before it is incremented,
505 * but then read another CPU's count which it has been subsequently
506 * decremented from -- we would see more decrements than we should.
507 * MNT_WRITE_HOLD protects against this scenario, because
508 * mnt_want_write first increments count, then smp_mb, then spins on
509 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
510 * we're counting up here.
511 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100512 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000513 ret = -EBUSY;
514 else
Al Viro83adc752011-11-24 22:37:54 -0500515 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000516 /*
517 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
518 * that become unheld will see MNT_READONLY.
519 */
520 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500521 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Al Viro719ea2f2013-09-29 11:24:49 -0400522 unlock_mount_hash();
Dave Hansen3d733632008-02-15 14:37:59 -0800523 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800524}
Dave Hansen83660252008-02-15 14:37:30 -0800525
Al Viro83adc752011-11-24 22:37:54 -0500526static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800527{
Al Viro719ea2f2013-09-29 11:24:49 -0400528 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500529 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Al Viro719ea2f2013-09-29 11:24:49 -0400530 unlock_mount_hash();
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800531}
532
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100533int sb_prepare_remount_readonly(struct super_block *sb)
534{
535 struct mount *mnt;
536 int err = 0;
537
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100538 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
539 if (atomic_long_read(&sb->s_remove_count))
540 return -EBUSY;
541
Al Viro719ea2f2013-09-29 11:24:49 -0400542 lock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100543 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
544 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
545 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
546 smp_mb();
547 if (mnt_get_writers(mnt) > 0) {
548 err = -EBUSY;
549 break;
550 }
551 }
552 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100553 if (!err && atomic_long_read(&sb->s_remove_count))
554 err = -EBUSY;
555
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100556 if (!err) {
557 sb->s_readonly_remount = 1;
558 smp_wmb();
559 }
560 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
561 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
562 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
563 }
Al Viro719ea2f2013-09-29 11:24:49 -0400564 unlock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100565
566 return err;
567}
568
Al Virob105e272011-11-24 20:38:33 -0500569static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570{
Al Viro52ba1622011-11-25 02:25:17 -0500571 kfree(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000572#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500573 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000574#endif
Al Virob105e272011-11-24 20:38:33 -0500575 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576}
577
David Howells8ffcb322014-01-24 12:17:54 +0000578static void delayed_free_vfsmnt(struct rcu_head *head)
579{
580 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
581}
582
Al Viro48a066e2013-09-29 22:06:07 -0400583/* call under rcu_read_lock */
584bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
585{
586 struct mount *mnt;
587 if (read_seqretry(&mount_lock, seq))
588 return false;
589 if (bastard == NULL)
590 return true;
591 mnt = real_mount(bastard);
592 mnt_add_count(mnt, 1);
593 if (likely(!read_seqretry(&mount_lock, seq)))
594 return true;
595 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
596 mnt_add_count(mnt, -1);
597 return false;
598 }
599 rcu_read_unlock();
600 mntput(bastard);
601 rcu_read_lock();
602 return false;
603}
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605/*
Al Viro474279d2013-10-01 16:11:26 -0400606 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400607 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 */
Al Viro474279d2013-10-01 16:11:26 -0400609struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610{
Al Viro38129a12014-03-20 21:10:51 -0400611 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400612 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Al Viro38129a12014-03-20 21:10:51 -0400614 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400615 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
616 return p;
617 return NULL;
618}
619
620/*
621 * find the last mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400622 * mount_lock must be held.
Al Viro474279d2013-10-01 16:11:26 -0400623 */
624struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
625{
Al Viro38129a12014-03-20 21:10:51 -0400626 struct mount *p, *res;
627 res = p = __lookup_mnt(mnt, dentry);
628 if (!p)
629 goto out;
630 hlist_for_each_entry_continue(p, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -0400631 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
632 break;
633 res = p;
634 }
Al Viro38129a12014-03-20 21:10:51 -0400635out:
Al Viro1d6a32a2014-03-20 20:34:43 -0400636 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
638
Ram Paia05964f2005-11-07 17:20:17 -0500639/*
David Howellsf015f1262012-06-25 12:55:28 +0100640 * lookup_mnt - Return the first child mount mounted at path
641 *
642 * "First" means first mounted chronologically. If you create the
643 * following mounts:
644 *
645 * mount /dev/sda1 /mnt
646 * mount /dev/sda2 /mnt
647 * mount /dev/sda3 /mnt
648 *
649 * Then lookup_mnt() on the base /mnt dentry in the root mount will
650 * return successively the root dentry and vfsmount of /dev/sda1, then
651 * /dev/sda2, then /dev/sda3, then NULL.
652 *
653 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500654 */
Al Viro1c755af2009-04-18 14:06:57 -0400655struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500656{
Al Viroc7105362011-11-24 18:22:03 -0500657 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400658 struct vfsmount *m;
659 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000660
Al Viro48a066e2013-09-29 22:06:07 -0400661 rcu_read_lock();
662 do {
663 seq = read_seqbegin(&mount_lock);
664 child_mnt = __lookup_mnt(path->mnt, path->dentry);
665 m = child_mnt ? &child_mnt->mnt : NULL;
666 } while (!legitimize_mnt(m, seq));
667 rcu_read_unlock();
668 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500669}
670
Eric W. Biederman7af13642013-10-04 19:15:13 -0700671/*
672 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
673 * current mount namespace.
674 *
675 * The common case is dentries are not mountpoints at all and that
676 * test is handled inline. For the slow case when we are actually
677 * dealing with a mountpoint of some kind, walk through all of the
678 * mounts in the current mount namespace and test to see if the dentry
679 * is a mountpoint.
680 *
681 * The mount_hashtable is not usable in the context because we
682 * need to identify all mounts that may be in the current mount
683 * namespace not just a mount that happens to have some specified
684 * parent mount.
685 */
686bool __is_local_mountpoint(struct dentry *dentry)
687{
688 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
689 struct mount *mnt;
690 bool is_covered = false;
691
692 if (!d_mountpoint(dentry))
693 goto out;
694
695 down_read(&namespace_sem);
696 list_for_each_entry(mnt, &ns->list, mnt_list) {
697 is_covered = (mnt->mnt_mountpoint == dentry);
698 if (is_covered)
699 break;
700 }
701 up_read(&namespace_sem);
702out:
703 return is_covered;
704}
705
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800706static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
Al Viro84d17192013-03-15 10:53:28 -0400707{
Al Viro0818bf22014-02-28 13:46:44 -0500708 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d17192013-03-15 10:53:28 -0400709 struct mountpoint *mp;
710
Al Viro0818bf22014-02-28 13:46:44 -0500711 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d17192013-03-15 10:53:28 -0400712 if (mp->m_dentry == dentry) {
713 /* might be worth a WARN_ON() */
714 if (d_unlinked(dentry))
715 return ERR_PTR(-ENOENT);
716 mp->m_count++;
717 return mp;
718 }
719 }
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800720 return NULL;
721}
722
723static struct mountpoint *new_mountpoint(struct dentry *dentry)
724{
725 struct hlist_head *chain = mp_hash(dentry);
726 struct mountpoint *mp;
727 int ret;
Al Viro84d17192013-03-15 10:53:28 -0400728
729 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
730 if (!mp)
731 return ERR_PTR(-ENOMEM);
732
Miklos Szeredieed81002013-09-05 14:39:11 +0200733 ret = d_set_mounted(dentry);
734 if (ret) {
Al Viro84d17192013-03-15 10:53:28 -0400735 kfree(mp);
Miklos Szeredieed81002013-09-05 14:39:11 +0200736 return ERR_PTR(ret);
Al Viro84d17192013-03-15 10:53:28 -0400737 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200738
Al Viro84d17192013-03-15 10:53:28 -0400739 mp->m_dentry = dentry;
740 mp->m_count = 1;
Al Viro0818bf22014-02-28 13:46:44 -0500741 hlist_add_head(&mp->m_hash, chain);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700742 INIT_HLIST_HEAD(&mp->m_list);
Al Viro84d17192013-03-15 10:53:28 -0400743 return mp;
744}
745
746static void put_mountpoint(struct mountpoint *mp)
747{
748 if (!--mp->m_count) {
749 struct dentry *dentry = mp->m_dentry;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700750 BUG_ON(!hlist_empty(&mp->m_list));
Al Viro84d17192013-03-15 10:53:28 -0400751 spin_lock(&dentry->d_lock);
752 dentry->d_flags &= ~DCACHE_MOUNTED;
753 spin_unlock(&dentry->d_lock);
Al Viro0818bf22014-02-28 13:46:44 -0500754 hlist_del(&mp->m_hash);
Al Viro84d17192013-03-15 10:53:28 -0400755 kfree(mp);
756 }
757}
758
Al Viro143c8c92011-11-25 00:46:35 -0500759static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800761 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
Nick Piggin99b7db72010-08-18 04:37:39 +1000764/*
765 * vfsmount lock must be held for write
766 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800767static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500768{
769 if (ns) {
770 ns->event = ++event;
771 wake_up_interruptible(&ns->poll);
772 }
773}
774
Nick Piggin99b7db72010-08-18 04:37:39 +1000775/*
776 * vfsmount lock must be held for write
777 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800778static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500779{
780 if (ns && ns->event != event) {
781 ns->event = event;
782 wake_up_interruptible(&ns->poll);
783 }
784}
785
Nick Piggin99b7db72010-08-18 04:37:39 +1000786/*
787 * vfsmount lock must be held for write
788 */
Al Viro419148d2011-11-24 19:41:16 -0500789static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
Al Viroa73324d2011-11-24 22:25:07 -0500791 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500792 old_path->mnt = &mnt->mnt_parent->mnt;
793 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500794 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500795 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400796 hlist_del_init_rcu(&mnt->mnt_hash);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700797 hlist_del_init(&mnt->mnt_mp_list);
Al Viro84d17192013-03-15 10:53:28 -0400798 put_mountpoint(mnt->mnt_mp);
799 mnt->mnt_mp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800}
801
Nick Piggin99b7db72010-08-18 04:37:39 +1000802/*
803 * vfsmount lock must be held for write
804 */
Al Viro84d17192013-03-15 10:53:28 -0400805void mnt_set_mountpoint(struct mount *mnt,
806 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500807 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500808{
Al Viro84d17192013-03-15 10:53:28 -0400809 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500810 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro84d17192013-03-15 10:53:28 -0400811 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500812 child_mnt->mnt_parent = mnt;
Al Viro84d17192013-03-15 10:53:28 -0400813 child_mnt->mnt_mp = mp;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700814 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500815}
816
Nick Piggin99b7db72010-08-18 04:37:39 +1000817/*
818 * vfsmount lock must be held for write
819 */
Al Viro84d17192013-03-15 10:53:28 -0400820static void attach_mnt(struct mount *mnt,
821 struct mount *parent,
822 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
Al Viro84d17192013-03-15 10:53:28 -0400824 mnt_set_mountpoint(parent, mp, mnt);
Al Viro38129a12014-03-20 21:10:51 -0400825 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
Al Viro84d17192013-03-15 10:53:28 -0400826 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500827}
828
Al Viro12a5b522014-08-10 03:44:55 -0400829static void attach_shadowed(struct mount *mnt,
830 struct mount *parent,
831 struct mount *shadows)
832{
833 if (shadows) {
Linus Torvaldsf6f99332014-08-11 11:44:11 -0700834 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
Al Viro12a5b522014-08-10 03:44:55 -0400835 list_add(&mnt->mnt_child, &shadows->mnt_child);
836 } else {
837 hlist_add_head_rcu(&mnt->mnt_hash,
838 m_hash(&parent->mnt, mnt->mnt_mountpoint));
839 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
840 }
841}
842
Ram Paib90fa9a2005-11-07 17:19:50 -0500843/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000844 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500845 */
Al Viro1d6a32a2014-03-20 20:34:43 -0400846static void commit_tree(struct mount *mnt, struct mount *shadows)
Ram Paib90fa9a2005-11-07 17:19:50 -0500847{
Al Viro0714a532011-11-24 22:19:58 -0500848 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500849 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500850 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500851 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500852
Al Viro0714a532011-11-24 22:19:58 -0500853 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500854
Al Viro1a4eeaf2011-11-25 02:19:55 -0500855 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400856 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500857 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500858
Ram Paib90fa9a2005-11-07 17:19:50 -0500859 list_splice(&head, n->list.prev);
860
Al Viro12a5b522014-08-10 03:44:55 -0400861 attach_shadowed(mnt, parent, shadows);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800862 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863}
864
Al Viro909b0a82011-11-25 03:06:56 -0500865static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
Al Viro6b41d532011-11-24 23:24:33 -0500867 struct list_head *next = p->mnt_mounts.next;
868 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500870 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500872 next = p->mnt_child.next;
873 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 break;
Al Viro0714a532011-11-24 22:19:58 -0500875 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 }
877 }
Al Viro6b41d532011-11-24 23:24:33 -0500878 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879}
880
Al Viro315fc832011-11-24 18:57:30 -0500881static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500882{
Al Viro6b41d532011-11-24 23:24:33 -0500883 struct list_head *prev = p->mnt_mounts.prev;
884 while (prev != &p->mnt_mounts) {
885 p = list_entry(prev, struct mount, mnt_child);
886 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500887 }
888 return p;
889}
890
Al Viro9d412a42011-03-17 22:08:28 -0400891struct vfsmount *
892vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
893{
Al Virob105e272011-11-24 20:38:33 -0500894 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400895 struct dentry *root;
896
897 if (!type)
898 return ERR_PTR(-ENODEV);
899
900 mnt = alloc_vfsmnt(name);
901 if (!mnt)
902 return ERR_PTR(-ENOMEM);
903
904 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500905 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400906
907 root = mount_fs(type, flags, name, data);
908 if (IS_ERR(root)) {
David Howells8ffcb322014-01-24 12:17:54 +0000909 mnt_free_id(mnt);
Al Viro9d412a42011-03-17 22:08:28 -0400910 free_vfsmnt(mnt);
911 return ERR_CAST(root);
912 }
913
Al Virob105e272011-11-24 20:38:33 -0500914 mnt->mnt.mnt_root = root;
915 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500916 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500917 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400918 lock_mount_hash();
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100919 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400920 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -0500921 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400922}
923EXPORT_SYMBOL_GPL(vfs_kern_mount);
924
Al Viro87129cc2011-11-24 21:24:27 -0500925static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500926 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927{
Al Viro87129cc2011-11-24 21:24:27 -0500928 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100929 struct mount *mnt;
930 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
David Howellsbe34d1a2012-06-25 12:55:18 +0100932 mnt = alloc_vfsmnt(old->mnt_devname);
933 if (!mnt)
934 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100935
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700936 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +0100937 mnt->mnt_group_id = 0; /* not a peer of original */
938 else
939 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100940
David Howellsbe34d1a2012-06-25 12:55:18 +0100941 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
942 err = mnt_alloc_group_id(mnt);
943 if (err)
944 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100946
Al Virof2ebb3a2014-02-27 09:35:45 -0500947 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700948 /* Don't allow unprivileged users to change mount flags */
Eric W. Biederman9566d672014-07-28 17:26:07 -0700949 if (flag & CL_UNPRIVILEGED) {
950 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
951
952 if (mnt->mnt.mnt_flags & MNT_READONLY)
953 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
954
955 if (mnt->mnt.mnt_flags & MNT_NODEV)
956 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
957
958 if (mnt->mnt.mnt_flags & MNT_NOSUID)
959 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
960
961 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
962 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
963 }
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700964
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -0700965 /* Don't allow unprivileged users to reveal what is under a mount */
Eric W. Biederman381cacb2014-10-07 17:11:46 -0700966 if ((flag & CL_UNPRIVILEGED) &&
967 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -0700968 mnt->mnt.mnt_flags |= MNT_LOCKED;
969
David Howellsbe34d1a2012-06-25 12:55:18 +0100970 atomic_inc(&sb->s_active);
971 mnt->mnt.mnt_sb = sb;
972 mnt->mnt.mnt_root = dget(root);
973 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
974 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400975 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +0100976 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400977 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +0100978
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700979 if ((flag & CL_SLAVE) ||
980 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +0100981 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
982 mnt->mnt_master = old;
983 CLEAR_MNT_SHARED(mnt);
984 } else if (!(flag & CL_PRIVATE)) {
985 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
986 list_add(&mnt->mnt_share, &old->mnt_share);
987 if (IS_MNT_SLAVE(old))
988 list_add(&mnt->mnt_slave, &old->mnt_slave);
989 mnt->mnt_master = old->mnt_master;
990 }
991 if (flag & CL_MAKE_SHARED)
992 set_mnt_shared(mnt);
993
994 /* stick the duplicate mount on the same expiry list
995 * as the original if that was on one */
996 if (flag & CL_EXPIRE) {
997 if (!list_empty(&old->mnt_expire))
998 list_add(&mnt->mnt_expire, &old->mnt_expire);
999 }
1000
Al Virocb338d02011-11-24 20:55:08 -05001001 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001002
1003 out_free:
David Howells8ffcb322014-01-24 12:17:54 +00001004 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001005 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001006 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007}
1008
Al Viro9ea459e2014-08-08 13:08:20 -04001009static void cleanup_mnt(struct mount *mnt)
1010{
1011 /*
1012 * This probably indicates that somebody messed
1013 * up a mnt_want/drop_write() pair. If this
1014 * happens, the filesystem was probably unable
1015 * to make r/w->r/o transitions.
1016 */
1017 /*
1018 * The locking used to deal with mnt_count decrement provides barriers,
1019 * so mnt_get_writers() below is safe.
1020 */
1021 WARN_ON(mnt_get_writers(mnt));
1022 if (unlikely(mnt->mnt_pins.first))
1023 mnt_pin_kill(mnt);
1024 fsnotify_vfsmount_delete(&mnt->mnt);
1025 dput(mnt->mnt.mnt_root);
1026 deactivate_super(mnt->mnt.mnt_sb);
1027 mnt_free_id(mnt);
1028 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1029}
1030
1031static void __cleanup_mnt(struct rcu_head *head)
1032{
1033 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1034}
1035
1036static LLIST_HEAD(delayed_mntput_list);
1037static void delayed_mntput(struct work_struct *unused)
1038{
1039 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1040 struct llist_node *next;
1041
1042 for (; node; node = next) {
1043 next = llist_next(node);
1044 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1045 }
1046}
1047static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1048
Al Viro900148d2011-11-25 00:33:11 -05001049static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001050{
Al Viro48a066e2013-09-29 22:06:07 -04001051 rcu_read_lock();
1052 mnt_add_count(mnt, -1);
1053 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1054 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -05001055 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001056 }
Al Viro719ea2f2013-09-29 11:24:49 -04001057 lock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001058 if (mnt_get_count(mnt)) {
Al Viro48a066e2013-09-29 22:06:07 -04001059 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -04001060 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001061 return;
1062 }
Al Viro48a066e2013-09-29 22:06:07 -04001063 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1064 rcu_read_unlock();
1065 unlock_mount_hash();
1066 return;
1067 }
1068 mnt->mnt.mnt_flags |= MNT_DOOMED;
1069 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301070
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001071 list_del(&mnt->mnt_instance);
Al Viro719ea2f2013-09-29 11:24:49 -04001072 unlock_mount_hash();
Al Viro649a7952013-09-28 12:41:25 -04001073
Al Viro9ea459e2014-08-08 13:08:20 -04001074 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1075 struct task_struct *task = current;
1076 if (likely(!(task->flags & PF_KTHREAD))) {
1077 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1078 if (!task_work_add(task, &mnt->mnt_rcu, true))
1079 return;
1080 }
1081 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1082 schedule_delayed_work(&delayed_mntput_work, 1);
1083 return;
1084 }
1085 cleanup_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001086}
Nick Pigginb3e19d92011-01-07 17:50:11 +11001087
1088void mntput(struct vfsmount *mnt)
1089{
1090 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -05001091 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001092 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -05001093 if (unlikely(m->mnt_expiry_mark))
1094 m->mnt_expiry_mark = 0;
1095 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001096 }
1097}
1098EXPORT_SYMBOL(mntput);
1099
1100struct vfsmount *mntget(struct vfsmount *mnt)
1101{
1102 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001103 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001104 return mnt;
1105}
1106EXPORT_SYMBOL(mntget);
1107
Al Viro3064c352014-08-07 09:12:31 -04001108struct vfsmount *mnt_clone_internal(struct path *path)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001109{
Al Viro3064c352014-08-07 09:12:31 -04001110 struct mount *p;
1111 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1112 if (IS_ERR(p))
1113 return ERR_CAST(p);
1114 p->mnt.mnt_flags |= MNT_INTERNAL;
1115 return &p->mnt;
Al Viro7b7b1ac2005-11-07 17:13:39 -05001116}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001118static inline void mangle(struct seq_file *m, const char *s)
1119{
1120 seq_escape(m, s, " \t\n\\");
1121}
1122
1123/*
1124 * Simple .show_options callback for filesystems which don't want to
1125 * implement more complex mount option showing.
1126 *
1127 * See also save_mount_options().
1128 */
Al Viro34c80b12011-12-08 21:32:45 -05001129int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001130{
Al Viro2a32ceb2009-05-08 16:05:57 -04001131 const char *options;
1132
1133 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -05001134 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001135
1136 if (options != NULL && options[0]) {
1137 seq_putc(m, ',');
1138 mangle(m, options);
1139 }
Al Viro2a32ceb2009-05-08 16:05:57 -04001140 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001141
1142 return 0;
1143}
1144EXPORT_SYMBOL(generic_show_options);
1145
1146/*
1147 * If filesystem uses generic_show_options(), this function should be
1148 * called from the fill_super() callback.
1149 *
1150 * The .remount_fs callback usually needs to be handled in a special
1151 * way, to make sure, that previous options are not overwritten if the
1152 * remount fails.
1153 *
1154 * Also note, that if the filesystem's .remount_fs function doesn't
1155 * reset all options to their default value, but changes only newly
1156 * given options, then the displayed options will not reflect reality
1157 * any more.
1158 */
1159void save_mount_options(struct super_block *sb, char *options)
1160{
Al Viro2a32ceb2009-05-08 16:05:57 -04001161 BUG_ON(sb->s_options);
1162 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001163}
1164EXPORT_SYMBOL(save_mount_options);
1165
Al Viro2a32ceb2009-05-08 16:05:57 -04001166void replace_mount_options(struct super_block *sb, char *options)
1167{
1168 char *old = sb->s_options;
1169 rcu_assign_pointer(sb->s_options, options);
1170 if (old) {
1171 synchronize_rcu();
1172 kfree(old);
1173 }
1174}
1175EXPORT_SYMBOL(replace_mount_options);
1176
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001177#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001178/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179static void *m_start(struct seq_file *m, loff_t *pos)
1180{
Al Viro6ce6e242012-06-09 01:16:59 -04001181 struct proc_mounts *p = proc_mounts(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
Ram Pai390c6842005-11-07 17:17:51 -05001183 down_read(&namespace_sem);
Al Viroc7999c32014-02-27 14:40:10 -05001184 if (p->cached_event == p->ns->event) {
1185 void *v = p->cached_mount;
1186 if (*pos == p->cached_index)
1187 return v;
1188 if (*pos == p->cached_index + 1) {
1189 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1190 return p->cached_mount = v;
1191 }
1192 }
1193
1194 p->cached_event = p->ns->event;
1195 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1196 p->cached_index = *pos;
1197 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198}
1199
1200static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1201{
Al Viro6ce6e242012-06-09 01:16:59 -04001202 struct proc_mounts *p = proc_mounts(m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001203
Al Viroc7999c32014-02-27 14:40:10 -05001204 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1205 p->cached_index = *pos;
1206 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207}
1208
1209static void m_stop(struct seq_file *m, void *v)
1210{
Ram Pai390c6842005-11-07 17:17:51 -05001211 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212}
1213
Al Viro0226f492011-12-06 12:21:54 -05001214static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001215{
Al Viro6ce6e242012-06-09 01:16:59 -04001216 struct proc_mounts *p = proc_mounts(m);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001217 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001218 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219}
1220
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001221const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 .start = m_start,
1223 .next = m_next,
1224 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001225 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001226};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001227#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001228
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229/**
1230 * may_umount_tree - check if a mount tree is busy
1231 * @mnt: root of mount tree
1232 *
1233 * This is called to check if a tree of mounts has any
1234 * open files, pwds, chroots or sub mounts that are
1235 * busy.
1236 */
Al Viro909b0a82011-11-25 03:06:56 -05001237int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238{
Al Viro909b0a82011-11-25 03:06:56 -05001239 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001240 int actual_refs = 0;
1241 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001242 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001243 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Nick Pigginb3e19d92011-01-07 17:50:11 +11001245 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001246 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001247 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001248 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 }
Al Viro719ea2f2013-09-29 11:24:49 -04001251 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
1253 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001254 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
Ian Kente3474a82006-03-27 01:14:51 -08001256 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
1259EXPORT_SYMBOL(may_umount_tree);
1260
1261/**
1262 * may_umount - check if a mount point is busy
1263 * @mnt: root of mount
1264 *
1265 * This is called to check if a mount point has any
1266 * open files, pwds, chroots or sub mounts. If the
1267 * mount has sub mounts this will return busy
1268 * regardless of whether the sub mounts are busy.
1269 *
1270 * Doesn't take quota and stuff into account. IOW, in some cases it will
1271 * give false negatives. The main reason why it's here is that we need
1272 * a non-destructive way to look for easily umountable filesystems.
1273 */
1274int may_umount(struct vfsmount *mnt)
1275{
Ian Kente3474a82006-03-27 01:14:51 -08001276 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001277 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001278 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001279 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001280 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001281 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001282 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001283 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284}
1285
1286EXPORT_SYMBOL(may_umount);
1287
Al Viro38129a12014-03-20 21:10:51 -04001288static HLIST_HEAD(unmounted); /* protected by namespace_sem */
Al Viroe3197d82013-03-16 14:35:16 -04001289
Al Viro97216be2013-03-16 15:12:40 -04001290static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291{
Al Virod5e50f72011-11-24 20:58:57 -05001292 struct mount *mnt;
Al Viro38129a12014-03-20 21:10:51 -04001293 struct hlist_head head = unmounted;
Al Viro97216be2013-03-16 15:12:40 -04001294
Al Viro38129a12014-03-20 21:10:51 -04001295 if (likely(hlist_empty(&head))) {
Al Viro97216be2013-03-16 15:12:40 -04001296 up_write(&namespace_sem);
1297 return;
1298 }
1299
Al Viro38129a12014-03-20 21:10:51 -04001300 head.first->pprev = &head.first;
1301 INIT_HLIST_HEAD(&unmounted);
1302
Al Viro81b6b062014-08-30 18:32:05 -04001303 /* undo decrements we'd done in umount_tree() */
1304 hlist_for_each_entry(mnt, &head, mnt_hash)
1305 if (mnt->mnt_ex_mountpoint.mnt)
1306 mntget(mnt->mnt_ex_mountpoint.mnt);
1307
Al Viro97216be2013-03-16 15:12:40 -04001308 up_write(&namespace_sem);
1309
Al Viro48a066e2013-09-29 22:06:07 -04001310 synchronize_rcu();
1311
Al Viro38129a12014-03-20 21:10:51 -04001312 while (!hlist_empty(&head)) {
1313 mnt = hlist_entry(head.first, struct mount, mnt_hash);
1314 hlist_del_init(&mnt->mnt_hash);
Al Viroaba809cf2013-09-28 23:10:55 -04001315 if (mnt->mnt_ex_mountpoint.mnt)
1316 path_put(&mnt->mnt_ex_mountpoint);
Al Virod5e50f72011-11-24 20:58:57 -05001317 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001318 }
1319}
1320
Al Viro97216be2013-03-16 15:12:40 -04001321static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001322{
Al Viro97216be2013-03-16 15:12:40 -04001323 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001324}
1325
Nick Piggin99b7db72010-08-18 04:37:39 +10001326/*
Al Viro48a066e2013-09-29 22:06:07 -04001327 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001328 * namespace_sem must be held for write
Al Viro48a066e2013-09-29 22:06:07 -04001329 * how = 0 => just this tree, don't propagate
1330 * how = 1 => propagate; we know that nobody else has reference to any victims
1331 * how = 2 => lazy umount
Nick Piggin99b7db72010-08-18 04:37:39 +10001332 */
Al Viro48a066e2013-09-29 22:06:07 -04001333void umount_tree(struct mount *mnt, int how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001334{
Al Viro38129a12014-03-20 21:10:51 -04001335 HLIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001336 struct mount *p;
Al Viro38129a12014-03-20 21:10:51 -04001337 struct mount *last = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001338
Al Viro38129a12014-03-20 21:10:51 -04001339 for (p = mnt; p; p = next_mnt(p, mnt)) {
1340 hlist_del_init_rcu(&p->mnt_hash);
1341 hlist_add_head(&p->mnt_hash, &tmp_list);
1342 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001343
Al Viro88b368f2014-08-18 15:09:26 -04001344 hlist_for_each_entry(p, &tmp_list, mnt_hash)
1345 list_del_init(&p->mnt_child);
1346
Al Viro48a066e2013-09-29 22:06:07 -04001347 if (how)
Al Viro7b8a53f2011-01-15 20:08:44 -05001348 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001349
Al Viro38129a12014-03-20 21:10:51 -04001350 hlist_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001351 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001352 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001353 __touch_mnt_namespace(p->mnt_ns);
1354 p->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04001355 if (how < 2)
1356 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro676da582011-11-24 21:47:05 -05001357 if (mnt_has_parent(p)) {
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -07001358 hlist_del_init(&p->mnt_mp_list);
Al Viro84d17192013-03-15 10:53:28 -04001359 put_mountpoint(p->mnt_mp);
Al Viro81b6b062014-08-30 18:32:05 -04001360 mnt_add_count(p->mnt_parent, -1);
Al Viroaba809cf2013-09-28 23:10:55 -04001361 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1362 p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint;
1363 p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt;
1364 p->mnt_mountpoint = p->mnt.mnt_root;
1365 p->mnt_parent = p;
Al Viro84d17192013-03-15 10:53:28 -04001366 p->mnt_mp = NULL;
Al Viro7c4b93d2008-03-21 23:59:49 -04001367 }
Al Viro0f0afb12011-11-24 20:43:10 -05001368 change_mnt_propagation(p, MS_PRIVATE);
Al Viro38129a12014-03-20 21:10:51 -04001369 last = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 }
Al Viro38129a12014-03-20 21:10:51 -04001371 if (last) {
1372 last->mnt_hash.next = unmounted.first;
Eric W. Biedermanc297abf2014-12-18 10:57:19 -06001373 if (unmounted.first)
1374 unmounted.first->pprev = &last->mnt_hash.next;
Al Viro38129a12014-03-20 21:10:51 -04001375 unmounted.first = tmp_list.first;
1376 unmounted.first->pprev = &unmounted.first;
1377 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378}
1379
Al Virob54b9be2013-03-16 14:39:34 -04001380static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001381
Al Viro1ab59732011-11-24 21:35:16 -05001382static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383{
Al Viro1ab59732011-11-24 21:35:16 -05001384 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 int retval;
1386
Al Viro1ab59732011-11-24 21:35:16 -05001387 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 if (retval)
1389 return retval;
1390
1391 /*
1392 * Allow userspace to request a mountpoint be expired rather than
1393 * unmounting unconditionally. Unmount only happens if:
1394 * (1) the mark is already set (the mark is cleared by mntput())
1395 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1396 */
1397 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001398 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 flags & (MNT_FORCE | MNT_DETACH))
1400 return -EINVAL;
1401
Nick Pigginb3e19d92011-01-07 17:50:11 +11001402 /*
1403 * probably don't strictly need the lock here if we examined
1404 * all race cases, but it's a slowpath.
1405 */
Al Viro719ea2f2013-09-29 11:24:49 -04001406 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001407 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001408 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001410 }
Al Viro719ea2f2013-09-29 11:24:49 -04001411 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
Al Viro863d6842011-11-25 00:57:42 -05001413 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 return -EAGAIN;
1415 }
1416
1417 /*
1418 * If we may have to abort operations to get out of this
1419 * mount, and they will themselves hold resources we must
1420 * allow the fs to do things. In the Unix tradition of
1421 * 'Gee thats tricky lets do it in userspace' the umount_begin
1422 * might fail to complete on the first run through as other tasks
1423 * must return, and the like. Thats for the mount program to worry
1424 * about for the moment.
1425 */
1426
Al Viro42faad92008-04-24 07:21:56 -04001427 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001428 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001429 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431 /*
1432 * No sense to grab the lock for this test, but test itself looks
1433 * somewhat bogus. Suggestions for better replacement?
1434 * Ho-hum... In principle, we might treat that as umount + switch
1435 * to rootfs. GC would eventually take care of the old vfsmount.
1436 * Actually it makes sense, especially if rootfs would contain a
1437 * /reboot - static binary that would close all descriptors and
1438 * call reboot(9). Then init(8) could umount root and exec /reboot.
1439 */
Al Viro1ab59732011-11-24 21:35:16 -05001440 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 /*
1442 * Special case for "unmounting" root ...
1443 * we just try to remount it readonly.
1444 */
Andy Lutomirskia1480dc2014-10-08 12:32:47 -07001445 if (!capable(CAP_SYS_ADMIN))
1446 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001448 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 up_write(&sb->s_umount);
1451 return retval;
1452 }
1453
Al Viro97216be2013-03-16 15:12:40 -04001454 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001455 lock_mount_hash();
Al Viro5addc5d2005-11-07 17:15:49 -05001456 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
Al Viro48a066e2013-09-29 22:06:07 -04001458 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001459 if (!list_empty(&mnt->mnt_list))
Al Viro48a066e2013-09-29 22:06:07 -04001460 umount_tree(mnt, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001462 } else {
1463 shrink_submounts(mnt);
1464 retval = -EBUSY;
1465 if (!propagate_mount_busy(mnt, 2)) {
1466 if (!list_empty(&mnt->mnt_list))
1467 umount_tree(mnt, 1);
1468 retval = 0;
1469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 }
Al Viro719ea2f2013-09-29 11:24:49 -04001471 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001472 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 return retval;
1474}
1475
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001476/*
1477 * __detach_mounts - lazily unmount all mounts on the specified dentry
1478 *
1479 * During unlink, rmdir, and d_drop it is possible to loose the path
1480 * to an existing mountpoint, and wind up leaking the mount.
1481 * detach_mounts allows lazily unmounting those mounts instead of
1482 * leaking them.
1483 *
1484 * The caller may hold dentry->d_inode->i_mutex.
1485 */
1486void __detach_mounts(struct dentry *dentry)
1487{
1488 struct mountpoint *mp;
1489 struct mount *mnt;
1490
1491 namespace_lock();
1492 mp = lookup_mountpoint(dentry);
1493 if (!mp)
1494 goto out_unlock;
1495
1496 lock_mount_hash();
1497 while (!hlist_empty(&mp->m_list)) {
1498 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1499 umount_tree(mnt, 2);
1500 }
1501 unlock_mount_hash();
1502 put_mountpoint(mp);
1503out_unlock:
1504 namespace_unlock();
1505}
1506
Al Viro9b40bc92013-02-22 22:45:42 -05001507/*
1508 * Is the caller allowed to modify his namespace?
1509 */
1510static inline bool may_mount(void)
1511{
1512 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1513}
1514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515/*
1516 * Now umount can handle mount points as well as block devices.
1517 * This is important for filesystems which use unnamed block devices.
1518 *
1519 * We now support a flag for forced unmount like the other 'big iron'
1520 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1521 */
1522
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001523SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
Al Viro2d8f3032008-07-22 09:59:21 -04001525 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001526 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001528 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001530 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1531 return -EINVAL;
1532
Al Viro9b40bc92013-02-22 22:45:42 -05001533 if (!may_mount())
1534 return -EPERM;
1535
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001536 if (!(flags & UMOUNT_NOFOLLOW))
1537 lookup_flags |= LOOKUP_FOLLOW;
1538
Al Viro197df042013-09-08 14:03:27 -04001539 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 if (retval)
1541 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001542 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001544 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001546 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 goto dput_and_out;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001548 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1549 goto dput_and_out;
Eric W. Biedermanb2f5d4d2014-10-04 14:44:03 -07001550 retval = -EPERM;
1551 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1552 goto dput_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
Al Viro900148d2011-11-25 00:33:11 -05001554 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001556 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001557 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001558 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559out:
1560 return retval;
1561}
1562
1563#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1564
1565/*
Ram Paib58fed82005-11-07 17:16:09 -05001566 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001568SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569{
Ram Paib58fed82005-11-07 17:16:09 -05001570 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571}
1572
1573#endif
1574
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001575static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001576{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001577 /* Is this a proxy for a mount namespace? */
Al Viroe149ed22014-11-01 10:57:28 -04001578 return dentry->d_op == &ns_dentry_operations &&
1579 dentry->d_fsdata == &mntns_operations;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001580}
1581
Al Viro58be28252014-11-01 00:00:23 -04001582struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1583{
1584 return container_of(ns, struct mnt_namespace, ns);
1585}
1586
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001587static bool mnt_ns_loop(struct dentry *dentry)
1588{
1589 /* Could bind mounting the mount namespace inode cause a
1590 * mount namespace loop?
1591 */
1592 struct mnt_namespace *mnt_ns;
1593 if (!is_mnt_ns_file(dentry))
1594 return false;
1595
Al Virof77c8012014-11-01 03:13:17 -04001596 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
Eric W. Biederman8823c072010-03-07 18:49:36 -08001597 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1598}
1599
Al Viro87129cc2011-11-24 21:24:27 -05001600struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001601 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602{
Al Viro84d17192013-03-15 10:53:28 -04001603 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001605 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1606 return ERR_PTR(-EINVAL);
1607
1608 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001609 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001610
Ram Pai36341f62005-11-07 17:17:22 -05001611 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001612 if (IS_ERR(q))
1613 return q;
1614
Al Viroa73324d2011-11-24 22:25:07 -05001615 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
1617 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001618 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001619 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001620 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 continue;
1622
Al Viro909b0a82011-11-25 03:06:56 -05001623 for (s = r; s; s = next_mnt(s, r)) {
Al Viro12a5b522014-08-10 03:44:55 -04001624 struct mount *t = NULL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001625 if (!(flag & CL_COPY_UNBINDABLE) &&
1626 IS_MNT_UNBINDABLE(s)) {
1627 s = skip_mnt_tree(s);
1628 continue;
1629 }
1630 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1631 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001632 s = skip_mnt_tree(s);
1633 continue;
1634 }
Al Viro0714a532011-11-24 22:19:58 -05001635 while (p != s->mnt_parent) {
1636 p = p->mnt_parent;
1637 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
Al Viro87129cc2011-11-24 21:24:27 -05001639 p = s;
Al Viro84d17192013-03-15 10:53:28 -04001640 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001641 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001642 if (IS_ERR(q))
1643 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04001644 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05001645 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro12a5b522014-08-10 03:44:55 -04001646 mnt_set_mountpoint(parent, p->mnt_mp, q);
1647 if (!list_empty(&parent->mnt_mounts)) {
1648 t = list_last_entry(&parent->mnt_mounts,
1649 struct mount, mnt_child);
1650 if (t->mnt_mp != p->mnt_mp)
1651 t = NULL;
1652 }
1653 attach_shadowed(q, parent, t);
Al Viro719ea2f2013-09-29 11:24:49 -04001654 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 }
1656 }
1657 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001658out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04001660 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001661 umount_tree(res, 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001662 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001664 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665}
1666
David Howellsbe34d1a2012-06-25 12:55:18 +01001667/* Caller should check returned pointer for errors */
1668
Al Viro589ff872009-04-18 03:28:19 -04001669struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001670{
Al Virocb338d02011-11-24 20:55:08 -05001671 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001672 namespace_lock();
Al Viro87129cc2011-11-24 21:24:27 -05001673 tree = copy_tree(real_mount(path->mnt), path->dentry,
1674 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001675 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001676 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001677 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001678 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001679}
1680
1681void drop_collected_mounts(struct vfsmount *mnt)
1682{
Al Viro97216be2013-03-16 15:12:40 -04001683 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001684 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001685 umount_tree(real_mount(mnt), 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001686 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04001687 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001688}
1689
Miklos Szeredic771d682014-10-24 00:14:36 +02001690/**
1691 * clone_private_mount - create a private clone of a path
1692 *
1693 * This creates a new vfsmount, which will be the clone of @path. The new will
1694 * not be attached anywhere in the namespace and will be private (i.e. changes
1695 * to the originating mount won't be propagated into this).
1696 *
1697 * Release with mntput().
1698 */
1699struct vfsmount *clone_private_mount(struct path *path)
1700{
1701 struct mount *old_mnt = real_mount(path->mnt);
1702 struct mount *new_mnt;
1703
1704 if (IS_MNT_UNBINDABLE(old_mnt))
1705 return ERR_PTR(-EINVAL);
1706
1707 down_read(&namespace_sem);
1708 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1709 up_read(&namespace_sem);
1710 if (IS_ERR(new_mnt))
1711 return ERR_CAST(new_mnt);
1712
1713 return &new_mnt->mnt;
1714}
1715EXPORT_SYMBOL_GPL(clone_private_mount);
1716
Al Viro1f707132010-01-30 22:51:25 -05001717int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1718 struct vfsmount *root)
1719{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001720 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001721 int res = f(root, arg);
1722 if (res)
1723 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001724 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1725 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001726 if (res)
1727 return res;
1728 }
1729 return 0;
1730}
1731
Al Viro4b8b21f2011-11-24 19:54:23 -05001732static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001733{
Al Viro315fc832011-11-24 18:57:30 -05001734 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001735
Al Viro909b0a82011-11-25 03:06:56 -05001736 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001737 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001738 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001739 }
1740}
1741
Al Viro4b8b21f2011-11-24 19:54:23 -05001742static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001743{
Al Viro315fc832011-11-24 18:57:30 -05001744 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001745
Al Viro909b0a82011-11-25 03:06:56 -05001746 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001747 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001748 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001749 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001750 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001751 return err;
1752 }
1753 }
1754 }
1755
1756 return 0;
1757}
1758
Ram Paib90fa9a2005-11-07 17:19:50 -05001759/*
1760 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001761 * @nd : place the mount tree @source_mnt is attached
1762 * @parent_nd : if non-null, detach the source_mnt from its parent and
1763 * store the parent mount and mountpoint dentry.
1764 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001765 *
1766 * NOTE: in the table below explains the semantics when a source mount
1767 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001768 * ---------------------------------------------------------------------------
1769 * | BIND MOUNT OPERATION |
1770 * |**************************************************************************
1771 * | source-->| shared | private | slave | unbindable |
1772 * | dest | | | | |
1773 * | | | | | | |
1774 * | v | | | | |
1775 * |**************************************************************************
1776 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1777 * | | | | | |
1778 * |non-shared| shared (+) | private | slave (*) | invalid |
1779 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001780 * A bind operation clones the source mount and mounts the clone on the
1781 * destination mount.
1782 *
1783 * (++) the cloned mount is propagated to all the mounts in the propagation
1784 * tree of the destination mount and the cloned mount is added to
1785 * the peer group of the source mount.
1786 * (+) the cloned mount is created under the destination mount and is marked
1787 * as shared. The cloned mount is added to the peer group of the source
1788 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001789 * (+++) the mount is propagated to all the mounts in the propagation tree
1790 * of the destination mount and the cloned mount is made slave
1791 * of the same master as that of the source mount. The cloned mount
1792 * is marked as 'shared and slave'.
1793 * (*) the cloned mount is made a slave of the same master as that of the
1794 * source mount.
1795 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001796 * ---------------------------------------------------------------------------
1797 * | MOVE MOUNT OPERATION |
1798 * |**************************************************************************
1799 * | source-->| shared | private | slave | unbindable |
1800 * | dest | | | | |
1801 * | | | | | | |
1802 * | v | | | | |
1803 * |**************************************************************************
1804 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1805 * | | | | | |
1806 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1807 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001808 *
1809 * (+) the mount is moved to the destination. And is then propagated to
1810 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001811 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001812 * (+++) the mount is moved to the destination and is then propagated to
1813 * all the mounts belonging to the destination mount's propagation tree.
1814 * the mount is marked as 'shared and slave'.
1815 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001816 *
1817 * if the source mount is a tree, the operations explained above is
1818 * applied to each mount in the tree.
1819 * Must be called without spinlocks held, since this function can sleep
1820 * in allocations.
1821 */
Al Viro0fb54e52011-11-24 19:59:16 -05001822static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d17192013-03-15 10:53:28 -04001823 struct mount *dest_mnt,
1824 struct mountpoint *dest_mp,
1825 struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001826{
Al Viro38129a12014-03-20 21:10:51 -04001827 HLIST_HEAD(tree_list);
Al Viro315fc832011-11-24 18:57:30 -05001828 struct mount *child, *p;
Al Viro38129a12014-03-20 21:10:51 -04001829 struct hlist_node *n;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001830 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001831
Al Virofc7be132011-11-25 01:05:37 -05001832 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001833 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001834 if (err)
1835 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04001836 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Al Virof2ebb3a2014-02-27 09:35:45 -05001837 lock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001838 if (err)
1839 goto out_cleanup_ids;
Al Viro909b0a82011-11-25 03:06:56 -05001840 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001841 set_mnt_shared(p);
Al Viro0b1b9012014-03-21 10:14:08 -04001842 } else {
1843 lock_mount_hash();
Ram Paib90fa9a2005-11-07 17:19:50 -05001844 }
Al Viro1a390682008-03-21 20:48:19 -04001845 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001846 detach_mnt(source_mnt, parent_path);
Al Viro84d17192013-03-15 10:53:28 -04001847 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05001848 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001849 } else {
Al Viro84d17192013-03-15 10:53:28 -04001850 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Al Viro1d6a32a2014-03-20 20:34:43 -04001851 commit_tree(source_mnt, NULL);
Ram Pai21444402005-11-07 17:20:03 -05001852 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001853
Al Viro38129a12014-03-20 21:10:51 -04001854 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04001855 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04001856 hlist_del_init(&child->mnt_hash);
Al Viro1d6a32a2014-03-20 20:34:43 -04001857 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1858 child->mnt_mountpoint);
1859 commit_tree(child, q);
Ram Paib90fa9a2005-11-07 17:19:50 -05001860 }
Al Viro719ea2f2013-09-29 11:24:49 -04001861 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10001862
Ram Paib90fa9a2005-11-07 17:19:50 -05001863 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001864
1865 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05001866 while (!hlist_empty(&tree_list)) {
1867 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1868 umount_tree(child, 0);
1869 }
1870 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001871 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001872 out:
1873 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001874}
1875
Al Viro84d17192013-03-15 10:53:28 -04001876static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04001877{
1878 struct vfsmount *mnt;
Al Viro84d17192013-03-15 10:53:28 -04001879 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04001880retry:
Al Viro84d17192013-03-15 10:53:28 -04001881 mutex_lock(&dentry->d_inode->i_mutex);
1882 if (unlikely(cant_mount(dentry))) {
1883 mutex_unlock(&dentry->d_inode->i_mutex);
1884 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04001885 }
Al Viro97216be2013-03-16 15:12:40 -04001886 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04001887 mnt = lookup_mnt(path);
Al Viro84d17192013-03-15 10:53:28 -04001888 if (likely(!mnt)) {
Eric W. Biedermane2dfa932014-02-24 17:32:34 -08001889 struct mountpoint *mp = lookup_mountpoint(dentry);
1890 if (!mp)
1891 mp = new_mountpoint(dentry);
Al Viro84d17192013-03-15 10:53:28 -04001892 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04001893 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001894 mutex_unlock(&dentry->d_inode->i_mutex);
1895 return mp;
1896 }
1897 return mp;
1898 }
Al Viro97216be2013-03-16 15:12:40 -04001899 namespace_unlock();
Al Virob12cea92011-03-18 08:55:38 -04001900 mutex_unlock(&path->dentry->d_inode->i_mutex);
1901 path_put(path);
1902 path->mnt = mnt;
Al Viro84d17192013-03-15 10:53:28 -04001903 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04001904 goto retry;
1905}
1906
Al Viro84d17192013-03-15 10:53:28 -04001907static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04001908{
Al Viro84d17192013-03-15 10:53:28 -04001909 struct dentry *dentry = where->m_dentry;
1910 put_mountpoint(where);
Al Viro328e6d92013-03-16 14:49:45 -04001911 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001912 mutex_unlock(&dentry->d_inode->i_mutex);
Al Virob12cea92011-03-18 08:55:38 -04001913}
1914
Al Viro84d17192013-03-15 10:53:28 -04001915static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916{
Al Viro95bc5f22011-11-25 00:30:56 -05001917 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 return -EINVAL;
1919
Al Viro84d17192013-03-15 10:53:28 -04001920 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001921 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 return -ENOTDIR;
1923
Al Viro84d17192013-03-15 10:53:28 -04001924 return attach_recursive_mnt(mnt, p, mp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925}
1926
1927/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001928 * Sanity check the flags to change_mnt_propagation.
1929 */
1930
1931static int flags_to_propagation_type(int flags)
1932{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001933 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001934
1935 /* Fail if any non-propagation flags are set */
1936 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1937 return 0;
1938 /* Only one propagation flag should be set */
1939 if (!is_power_of_2(type))
1940 return 0;
1941 return type;
1942}
1943
1944/*
Ram Pai07b20882005-11-07 17:19:07 -05001945 * recursively change the type of the mountpoint.
1946 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001947static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001948{
Al Viro315fc832011-11-24 18:57:30 -05001949 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001950 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001951 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001952 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001953 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001954
Al Viro2d92ab32008-08-02 00:51:11 -04001955 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001956 return -EINVAL;
1957
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001958 type = flags_to_propagation_type(flag);
1959 if (!type)
1960 return -EINVAL;
1961
Al Viro97216be2013-03-16 15:12:40 -04001962 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001963 if (type == MS_SHARED) {
1964 err = invent_group_ids(mnt, recurse);
1965 if (err)
1966 goto out_unlock;
1967 }
1968
Al Viro719ea2f2013-09-29 11:24:49 -04001969 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001970 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001971 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04001972 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001973
1974 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04001975 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001976 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001977}
1978
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001979static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1980{
1981 struct mount *child;
1982 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1983 if (!is_subdir(child->mnt_mountpoint, dentry))
1984 continue;
1985
1986 if (child->mnt.mnt_flags & MNT_LOCKED)
1987 return true;
1988 }
1989 return false;
1990}
1991
Ram Pai07b20882005-11-07 17:19:07 -05001992/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 * do loopback mount.
1994 */
Al Viro808d4e32012-10-11 11:42:01 -04001995static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001996 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997{
Al Viro2d92ab32008-08-02 00:51:11 -04001998 struct path old_path;
Al Viro84d17192013-03-15 10:53:28 -04001999 struct mount *mnt = NULL, *old, *parent;
2000 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002001 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 if (!old_name || !*old_name)
2003 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04002004 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 if (err)
2006 return err;
2007
Eric W. Biederman8823c072010-03-07 18:49:36 -08002008 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002009 if (mnt_ns_loop(old_path.dentry))
Eric W. Biederman8823c072010-03-07 18:49:36 -08002010 goto out;
2011
Al Viro84d17192013-03-15 10:53:28 -04002012 mp = lock_mount(path);
2013 err = PTR_ERR(mp);
2014 if (IS_ERR(mp))
Jan Blunck4ac91372008-02-14 19:34:32 -08002015 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05002016
Al Viro87129cc2011-11-24 21:24:27 -05002017 old = real_mount(old_path.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002018 parent = real_mount(path->mnt);
Al Viro87129cc2011-11-24 21:24:27 -05002019
Al Virob12cea92011-03-18 08:55:38 -04002020 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002021 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04002022 goto out2;
2023
Al Viroe149ed22014-11-01 10:57:28 -04002024 if (!check_mnt(parent))
2025 goto out2;
2026
2027 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
Al Virob12cea92011-03-18 08:55:38 -04002028 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002030 if (!recurse && has_locked_children(old, old_path.dentry))
2031 goto out2;
2032
Al Viroccd48bc2005-11-07 17:15:04 -05002033 if (recurse)
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002034 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
Al Viroccd48bc2005-11-07 17:15:04 -05002035 else
Al Viro87129cc2011-11-24 21:24:27 -05002036 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05002037
David Howellsbe34d1a2012-06-25 12:55:18 +01002038 if (IS_ERR(mnt)) {
2039 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04002040 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01002041 }
Al Viroccd48bc2005-11-07 17:15:04 -05002042
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002043 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2044
Al Viro84d17192013-03-15 10:53:28 -04002045 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002046 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002047 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04002048 umount_tree(mnt, 0);
Al Viro719ea2f2013-09-29 11:24:49 -04002049 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05002050 }
Al Virob12cea92011-03-18 08:55:38 -04002051out2:
Al Viro84d17192013-03-15 10:53:28 -04002052 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002053out:
Al Viro2d92ab32008-08-02 00:51:11 -04002054 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 return err;
2056}
2057
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002058static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2059{
2060 int error = 0;
2061 int readonly_request = 0;
2062
2063 if (ms_flags & MS_RDONLY)
2064 readonly_request = 1;
2065 if (readonly_request == __mnt_is_readonly(mnt))
2066 return 0;
2067
2068 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05002069 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002070 else
Al Viro83adc752011-11-24 22:37:54 -05002071 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002072 return error;
2073}
2074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075/*
2076 * change filesystem flags. dir should be a physical root of filesystem.
2077 * If you've mounted a non-root directory somewhere and want to do remount
2078 * on it - tough luck.
2079 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002080static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 void *data)
2082{
2083 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04002084 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05002085 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086
Al Viro143c8c92011-11-25 00:46:35 -05002087 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 return -EINVAL;
2089
Al Viro2d92ab32008-08-02 00:51:11 -04002090 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 return -EINVAL;
2092
Eric W. Biederman07b64552014-07-28 17:10:56 -07002093 /* Don't allow changing of locked mnt flags.
2094 *
2095 * No locks need to be held here while testing the various
2096 * MNT_LOCK flags because those flags can never be cleared
2097 * once they are set.
2098 */
2099 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2100 !(mnt_flags & MNT_READONLY)) {
2101 return -EPERM;
2102 }
Eric W. Biederman9566d672014-07-28 17:26:07 -07002103 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2104 !(mnt_flags & MNT_NODEV)) {
Eric W. Biederman3e186642014-08-13 01:33:38 -07002105 /* Was the nodev implicitly added in mount? */
2106 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2107 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2108 mnt_flags |= MNT_NODEV;
2109 } else {
2110 return -EPERM;
2111 }
Eric W. Biederman9566d672014-07-28 17:26:07 -07002112 }
2113 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2114 !(mnt_flags & MNT_NOSUID)) {
2115 return -EPERM;
2116 }
2117 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2118 !(mnt_flags & MNT_NOEXEC)) {
2119 return -EPERM;
2120 }
2121 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2122 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2123 return -EPERM;
2124 }
2125
Eric Parisff36fe22011-03-03 16:09:14 -05002126 err = security_sb_remount(sb, data);
2127 if (err)
2128 return err;
2129
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002131 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002132 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05002133 else if (!capable(CAP_SYS_ADMIN))
2134 err = -EPERM;
Al Viro4aa98cf2009-05-08 13:36:58 -04002135 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002136 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05002137 if (!err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002138 lock_mount_hash();
Eric W. Biedermana6138db2014-07-28 16:26:53 -07002139 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
Al Viro143c8c92011-11-25 00:46:35 -05002140 mnt->mnt.mnt_flags = mnt_flags;
Al Viro143c8c92011-11-25 00:46:35 -05002141 touch_mnt_namespace(mnt->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04002142 unlock_mount_hash();
Dan Williams0e55a7c2008-09-26 19:01:20 -07002143 }
Al Viro6339dab2013-09-16 22:41:01 -04002144 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 return err;
2146}
2147
Al Virocbbe3622011-11-24 20:01:19 -05002148static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05002149{
Al Viro315fc832011-11-24 18:57:30 -05002150 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05002151 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002152 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05002153 return 1;
2154 }
2155 return 0;
2156}
2157
Al Viro808d4e32012-10-11 11:42:01 -04002158static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159{
Al Viro2d92ab32008-08-02 00:51:11 -04002160 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05002161 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05002162 struct mount *old;
Al Viro84d17192013-03-15 10:53:28 -04002163 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002164 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 if (!old_name || !*old_name)
2166 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002167 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 if (err)
2169 return err;
2170
Al Viro84d17192013-03-15 10:53:28 -04002171 mp = lock_mount(path);
2172 err = PTR_ERR(mp);
2173 if (IS_ERR(mp))
David Howellscc53ce52011-01-14 18:45:26 +00002174 goto out;
2175
Al Viro143c8c92011-11-25 00:46:35 -05002176 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05002177 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05002178
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05002180 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 goto out1;
2182
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002183 if (old->mnt.mnt_flags & MNT_LOCKED)
2184 goto out1;
2185
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04002187 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05002188 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
Al Viro676da582011-11-24 21:47:05 -05002190 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05002191 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
Al Viro2d92ab32008-08-02 00:51:11 -04002193 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
2194 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05002195 goto out1;
2196 /*
2197 * Don't move a mount residing in a shared parent.
2198 */
Al Virofc7be132011-11-25 01:05:37 -05002199 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002200 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05002201 /*
2202 * Don't move a mount tree containing unbindable mounts to a destination
2203 * mount which is shared.
2204 */
Al Virofc7be132011-11-25 01:05:37 -05002205 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05002206 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05002208 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05002209 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05002210 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211
Al Viro84d17192013-03-15 10:53:28 -04002212 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08002213 if (err)
Ram Pai21444402005-11-07 17:20:03 -05002214 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215
2216 /* if the mount is moved, it should no longer be expire
2217 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05002218 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219out1:
Al Viro84d17192013-03-15 10:53:28 -04002220 unlock_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04002223 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04002224 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 return err;
2226}
2227
Al Viro9d412a42011-03-17 22:08:28 -04002228static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2229{
2230 int err;
2231 const char *subtype = strchr(fstype, '.');
2232 if (subtype) {
2233 subtype++;
2234 err = -EINVAL;
2235 if (!subtype[0])
2236 goto err;
2237 } else
2238 subtype = "";
2239
2240 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2241 err = -ENOMEM;
2242 if (!mnt->mnt_sb->s_subtype)
2243 goto err;
2244 return mnt;
2245
2246 err:
2247 mntput(mnt);
2248 return ERR_PTR(err);
2249}
2250
Al Viro9d412a42011-03-17 22:08:28 -04002251/*
2252 * add a mount into a namespace's mount tree
2253 */
Al Viro95bc5f22011-11-25 00:30:56 -05002254static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04002255{
Al Viro84d17192013-03-15 10:53:28 -04002256 struct mountpoint *mp;
2257 struct mount *parent;
Al Viro9d412a42011-03-17 22:08:28 -04002258 int err;
2259
Al Virof2ebb3a2014-02-27 09:35:45 -05002260 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04002261
Al Viro84d17192013-03-15 10:53:28 -04002262 mp = lock_mount(path);
2263 if (IS_ERR(mp))
2264 return PTR_ERR(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002265
Al Viro84d17192013-03-15 10:53:28 -04002266 parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04002267 err = -EINVAL;
Al Viro84d17192013-03-15 10:53:28 -04002268 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04002269 /* that's acceptable only for automounts done in private ns */
2270 if (!(mnt_flags & MNT_SHRINKABLE))
2271 goto unlock;
2272 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d17192013-03-15 10:53:28 -04002273 if (!parent->mnt_ns)
Al Viro156cacb2012-09-21 08:19:02 -04002274 goto unlock;
2275 }
Al Viro9d412a42011-03-17 22:08:28 -04002276
2277 /* Refuse the same filesystem on the same mount point */
2278 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05002279 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04002280 path->mnt->mnt_root == path->dentry)
2281 goto unlock;
2282
2283 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05002284 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04002285 goto unlock;
2286
Al Viro95bc5f22011-11-25 00:30:56 -05002287 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro84d17192013-03-15 10:53:28 -04002288 err = graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04002289
2290unlock:
Al Viro84d17192013-03-15 10:53:28 -04002291 unlock_mount(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002292 return err;
2293}
Al Virob1e75df2011-01-17 01:47:59 -05002294
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295/*
2296 * create a new mount for userspace and request it to be added into the
2297 * namespace's tree
2298 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002299static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04002300 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002302 struct file_system_type *type;
Al Viro9b40bc92013-02-22 22:45:42 -05002303 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002305 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002307 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 return -EINVAL;
2309
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002310 type = get_fs_type(fstype);
2311 if (!type)
2312 return -ENODEV;
2313
2314 if (user_ns != &init_user_ns) {
2315 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2316 put_filesystem(type);
2317 return -EPERM;
2318 }
2319 /* Only in special cases allow devices from mounts
2320 * created outside the initial user namespace.
2321 */
2322 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2323 flags |= MS_NODEV;
Eric W. Biederman9566d672014-07-28 17:26:07 -07002324 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002325 }
2326 }
2327
2328 mnt = vfs_kern_mount(type, flags, name, data);
2329 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2330 !mnt->mnt_sb->s_subtype)
2331 mnt = fs_set_subtype(mnt, fstype);
2332
2333 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 if (IS_ERR(mnt))
2335 return PTR_ERR(mnt);
2336
Al Viro95bc5f22011-11-25 00:30:56 -05002337 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002338 if (err)
2339 mntput(mnt);
2340 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341}
2342
Al Viro19a167a2011-01-17 01:35:23 -05002343int finish_automount(struct vfsmount *m, struct path *path)
2344{
Al Viro6776db3d2011-11-25 00:22:05 -05002345 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002346 int err;
2347 /* The new mount record should have at least 2 refs to prevent it being
2348 * expired before we get a chance to add it
2349 */
Al Viro6776db3d2011-11-25 00:22:05 -05002350 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002351
2352 if (m->mnt_sb == path->mnt->mnt_sb &&
2353 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002354 err = -ELOOP;
2355 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002356 }
2357
Al Viro95bc5f22011-11-25 00:30:56 -05002358 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002359 if (!err)
2360 return 0;
2361fail:
2362 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002363 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002364 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05002365 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04002366 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002367 }
Al Virob1e75df2011-01-17 01:47:59 -05002368 mntput(m);
2369 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002370 return err;
2371}
2372
David Howellsea5b7782011-01-14 19:10:03 +00002373/**
2374 * mnt_set_expiry - Put a mount on an expiration list
2375 * @mnt: The mount to list.
2376 * @expiry_list: The list to add the mount to.
2377 */
2378void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2379{
Al Viro97216be2013-03-16 15:12:40 -04002380 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00002381
Al Viro6776db3d2011-11-25 00:22:05 -05002382 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002383
Al Viro97216be2013-03-16 15:12:40 -04002384 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002385}
2386EXPORT_SYMBOL(mnt_set_expiry);
2387
2388/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 * process a list of expirable mountpoints with the intent of discarding any
2390 * mountpoints that aren't in use and haven't been touched since last we came
2391 * here
2392 */
2393void mark_mounts_for_expiry(struct list_head *mounts)
2394{
Al Viro761d5c32011-11-24 21:07:43 -05002395 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 LIST_HEAD(graveyard);
2397
2398 if (list_empty(mounts))
2399 return;
2400
Al Viro97216be2013-03-16 15:12:40 -04002401 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002402 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403
2404 /* extract from the expiration list every vfsmount that matches the
2405 * following criteria:
2406 * - only referenced by its parent vfsmount
2407 * - still marked for expiry (marked on the last call here; marks are
2408 * cleared by mntput())
2409 */
Al Viro6776db3d2011-11-25 00:22:05 -05002410 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002411 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002412 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002414 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002416 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002417 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002418 touch_mnt_namespace(mnt->mnt_ns);
Al Viro328e6d92013-03-16 14:49:45 -04002419 umount_tree(mnt, 1);
Al Virobcc5c7d2008-03-22 00:21:53 -04002420 }
Al Viro719ea2f2013-09-29 11:24:49 -04002421 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002422 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423}
2424
2425EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2426
2427/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002428 * Ripoff of 'select_parent()'
2429 *
2430 * search the list of submounts for a given mountpoint, and move any
2431 * shrinkable submounts to the 'graveyard' list.
2432 */
Al Viro692afc32011-11-24 21:15:14 -05002433static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002434{
Al Viro692afc32011-11-24 21:15:14 -05002435 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002436 struct list_head *next;
2437 int found = 0;
2438
2439repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002440 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002441resume:
Al Viro6b41d532011-11-24 23:24:33 -05002442 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002443 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002444 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002445
2446 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002447 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002448 continue;
2449 /*
2450 * Descend a level if the d_mounts list is non-empty.
2451 */
Al Viro6b41d532011-11-24 23:24:33 -05002452 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002453 this_parent = mnt;
2454 goto repeat;
2455 }
2456
Al Viro1ab59732011-11-24 21:35:16 -05002457 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002458 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002459 found++;
2460 }
2461 }
2462 /*
2463 * All done at this level ... ascend and resume the search
2464 */
2465 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002466 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002467 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002468 goto resume;
2469 }
2470 return found;
2471}
2472
2473/*
2474 * process a list of expirable mountpoints with the intent of discarding any
2475 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002476 *
Al Viro48a066e2013-09-29 22:06:07 -04002477 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002478 */
Al Virob54b9be2013-03-16 14:39:34 -04002479static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002480{
2481 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002482 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002483
Trond Myklebust5528f9112006-06-09 09:34:17 -04002484 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002485 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002486 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002487 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002488 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002489 touch_mnt_namespace(m->mnt_ns);
Al Viro328e6d92013-03-16 14:49:45 -04002490 umount_tree(m, 1);
Al Virobcc5c7d2008-03-22 00:21:53 -04002491 }
2492 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002493}
2494
Trond Myklebust5528f9112006-06-09 09:34:17 -04002495/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 * Some copy_from_user() implementations do not return the exact number of
2497 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2498 * Note that this function differs from copy_from_user() in that it will oops
2499 * on bad values of `to', rather than returning a short copy.
2500 */
Ram Paib58fed82005-11-07 17:16:09 -05002501static long exact_copy_from_user(void *to, const void __user * from,
2502 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503{
2504 char *t = to;
2505 const char __user *f = from;
2506 char c;
2507
2508 if (!access_ok(VERIFY_READ, from, n))
2509 return n;
2510
2511 while (n) {
2512 if (__get_user(c, f)) {
2513 memset(t, 0, n);
2514 break;
2515 }
2516 *t++ = c;
2517 f++;
2518 n--;
2519 }
2520 return n;
2521}
2522
Ram Paib58fed82005-11-07 17:16:09 -05002523int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524{
2525 int i;
2526 unsigned long page;
2527 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002528
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 *where = 0;
2530 if (!data)
2531 return 0;
2532
2533 if (!(page = __get_free_page(GFP_KERNEL)))
2534 return -ENOMEM;
2535
2536 /* We only care that *some* data at the address the user
2537 * gave us is valid. Just in case, we'll zero
2538 * the remainder of the page.
2539 */
2540 /* copy_from_user cannot cross TASK_SIZE ! */
2541 size = TASK_SIZE - (unsigned long)data;
2542 if (size > PAGE_SIZE)
2543 size = PAGE_SIZE;
2544
2545 i = size - exact_copy_from_user((void *)page, data, size);
2546 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002547 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 return -EFAULT;
2549 }
2550 if (i != PAGE_SIZE)
2551 memset((char *)page + i, 0, PAGE_SIZE - i);
2552 *where = page;
2553 return 0;
2554}
2555
Tim Gardnerb8850d12014-08-28 11:26:03 -06002556char *copy_mount_string(const void __user *data)
Vegard Nossumeca6f532009-09-18 13:05:45 -07002557{
Tim Gardnerb8850d12014-08-28 11:26:03 -06002558 return data ? strndup_user(data, PAGE_SIZE) : NULL;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002559}
2560
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561/*
2562 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2563 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2564 *
2565 * data is a (void *) that can point to any structure up to
2566 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2567 * information (or be NULL).
2568 *
2569 * Pre-0.97 versions of mount() didn't have a flags word.
2570 * When the flags word was introduced its top half was required
2571 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2572 * Therefore, if this magic number is present, it carries no information
2573 * and must be discarded.
2574 */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002575long do_mount(const char *dev_name, const char __user *dir_name,
Al Viro808d4e32012-10-11 11:42:01 -04002576 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577{
Al Viro2d92ab32008-08-02 00:51:11 -04002578 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 int retval = 0;
2580 int mnt_flags = 0;
2581
2582 /* Discard magic */
2583 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2584 flags &= ~MS_MGC_MSK;
2585
2586 /* Basic sanity checks */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 if (data_page)
2588 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2589
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002590 /* ... and get the mountpoint */
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002591 retval = user_path(dir_name, &path);
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002592 if (retval)
2593 return retval;
2594
2595 retval = security_sb_mount(dev_name, &path,
2596 type_page, flags, data_page);
Al Viro0d5cadb2013-05-04 14:40:51 -04002597 if (!retval && !may_mount())
2598 retval = -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002599 if (retval)
2600 goto dput_out;
2601
Andi Kleen613cbe32009-04-19 18:40:43 +02002602 /* Default to relatime unless overriden */
2603 if (!(flags & MS_NOATIME))
2604 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002605
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 /* Separate the per-mountpoint flags */
2607 if (flags & MS_NOSUID)
2608 mnt_flags |= MNT_NOSUID;
2609 if (flags & MS_NODEV)
2610 mnt_flags |= MNT_NODEV;
2611 if (flags & MS_NOEXEC)
2612 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002613 if (flags & MS_NOATIME)
2614 mnt_flags |= MNT_NOATIME;
2615 if (flags & MS_NODIRATIME)
2616 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002617 if (flags & MS_STRICTATIME)
2618 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002619 if (flags & MS_RDONLY)
2620 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002621
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07002622 /* The default atime for remount is preservation */
2623 if ((flags & MS_REMOUNT) &&
2624 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2625 MS_STRICTATIME)) == 0)) {
2626 mnt_flags &= ~MNT_ATIME_MASK;
2627 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2628 }
2629
Al Viro7a4dec52010-08-09 12:05:43 -04002630 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002631 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2632 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002635 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 data_page);
2637 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002638 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002639 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002640 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002642 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 else
Al Viro2d92ab32008-08-02 00:51:11 -04002644 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 dev_name, data_page);
2646dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002647 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 return retval;
2649}
2650
Eric W. Biederman771b1372012-07-26 21:08:32 -07002651static void free_mnt_ns(struct mnt_namespace *ns)
2652{
Al Viro6344c432014-11-01 00:45:45 -04002653 ns_free_inum(&ns->ns);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002654 put_user_ns(ns->user_ns);
2655 kfree(ns);
2656}
2657
Eric W. Biederman8823c072010-03-07 18:49:36 -08002658/*
2659 * Assign a sequence number so we can detect when we attempt to bind
2660 * mount a reference to an older mount namespace into the current
2661 * mount namespace, preventing reference counting loops. A 64bit
2662 * number incrementing at 10Ghz will take 12,427 years to wrap which
2663 * is effectively never, so we can ignore the possibility.
2664 */
2665static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2666
Eric W. Biederman771b1372012-07-26 21:08:32 -07002667static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002668{
2669 struct mnt_namespace *new_ns;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002670 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002671
2672 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2673 if (!new_ns)
2674 return ERR_PTR(-ENOMEM);
Al Viro6344c432014-11-01 00:45:45 -04002675 ret = ns_alloc_inum(&new_ns->ns);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002676 if (ret) {
2677 kfree(new_ns);
2678 return ERR_PTR(ret);
2679 }
Al Viro33c42942014-11-01 02:32:53 -04002680 new_ns->ns.ops = &mntns_operations;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002681 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002682 atomic_set(&new_ns->count, 1);
2683 new_ns->root = NULL;
2684 INIT_LIST_HEAD(&new_ns->list);
2685 init_waitqueue_head(&new_ns->poll);
2686 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002687 new_ns->user_ns = get_user_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002688 return new_ns;
2689}
2690
Al Viro9559f682013-09-28 20:47:57 -04002691struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2692 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002694 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002695 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002696 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04002697 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05002698 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002699 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700
Al Viro9559f682013-09-28 20:47:57 -04002701 BUG_ON(!ns);
2702
2703 if (likely(!(flags & CLONE_NEWNS))) {
2704 get_mnt_ns(ns);
2705 return ns;
2706 }
2707
2708 old = ns->root;
2709
Eric W. Biederman771b1372012-07-26 21:08:32 -07002710 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002711 if (IS_ERR(new_ns))
2712 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713
Al Viro97216be2013-03-16 15:12:40 -04002714 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002716 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04002717 if (user_ns != ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002718 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002719 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002720 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04002721 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002722 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002723 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 }
Al Virobe08d6d2011-12-06 13:32:36 -05002725 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002726 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727
2728 /*
2729 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2730 * as belonging to new namespace. We have already acquired a private
2731 * fs_struct, so tsk->fs->lock is not needed.
2732 */
Al Viro909b0a82011-11-25 03:06:56 -05002733 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002734 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002736 q->mnt_ns = new_ns;
Al Viro9559f682013-09-28 20:47:57 -04002737 if (new_fs) {
2738 if (&p->mnt == new_fs->root.mnt) {
2739 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002740 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 }
Al Viro9559f682013-09-28 20:47:57 -04002742 if (&p->mnt == new_fs->pwd.mnt) {
2743 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002744 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 }
Al Viro909b0a82011-11-25 03:06:56 -05002747 p = next_mnt(p, old);
2748 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002749 if (!q)
2750 break;
2751 while (p->mnt.mnt_root != q->mnt.mnt_root)
2752 p = next_mnt(p, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 }
Al Viro328e6d92013-03-16 14:49:45 -04002754 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002757 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002759 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760
JANAK DESAI741a2952006-02-07 12:59:00 -08002761 return new_ns;
2762}
2763
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002764/**
2765 * create_mnt_ns - creates a private namespace and adds a root filesystem
2766 * @mnt: pointer to the new root filesystem mountpoint
2767 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002768static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002769{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002770 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002771 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002772 struct mount *mnt = real_mount(m);
2773 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002774 new_ns->root = mnt;
Al Virob1983cd2013-05-04 15:18:53 -04002775 list_add(&mnt->mnt_list, &new_ns->list);
Al Viroc1334492011-11-16 16:12:14 -05002776 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002777 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002778 }
2779 return new_ns;
2780}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002781
Al Viroea441d12011-11-16 21:43:59 -05002782struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2783{
2784 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002785 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002786 struct path path;
2787 int err;
2788
2789 ns = create_mnt_ns(mnt);
2790 if (IS_ERR(ns))
2791 return ERR_CAST(ns);
2792
2793 err = vfs_path_lookup(mnt->mnt_root, mnt,
2794 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2795
2796 put_mnt_ns(ns);
2797
2798 if (err)
2799 return ERR_PTR(err);
2800
2801 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002802 s = path.mnt->mnt_sb;
2803 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002804 mntput(path.mnt);
2805 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002806 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002807 /* ... and return the root of (sub)tree on it */
2808 return path.dentry;
2809}
2810EXPORT_SYMBOL(mount_subtree);
2811
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002812SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2813 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002815 int ret;
2816 char *kernel_type;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002817 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819
Tim Gardnerb8850d12014-08-28 11:26:03 -06002820 kernel_type = copy_mount_string(type);
2821 ret = PTR_ERR(kernel_type);
2822 if (IS_ERR(kernel_type))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002823 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824
Tim Gardnerb8850d12014-08-28 11:26:03 -06002825 kernel_dev = copy_mount_string(dev_name);
2826 ret = PTR_ERR(kernel_dev);
2827 if (IS_ERR(kernel_dev))
Vegard Nossumeca6f532009-09-18 13:05:45 -07002828 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829
Vegard Nossumeca6f532009-09-18 13:05:45 -07002830 ret = copy_mount_options(data, &data_page);
2831 if (ret < 0)
2832 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833
Seunghun Lee5e6123f2014-09-14 22:15:10 +09002834 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
Vegard Nossumeca6f532009-09-18 13:05:45 -07002835 (void *) data_page);
2836
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002838out_data:
2839 kfree(kernel_dev);
2840out_dev:
Vegard Nossumeca6f532009-09-18 13:05:45 -07002841 kfree(kernel_type);
2842out_type:
2843 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844}
2845
2846/*
Al Viroafac7cb2011-11-23 19:34:49 -05002847 * Return true if path is reachable from root
2848 *
Al Viro48a066e2013-09-29 22:06:07 -04002849 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05002850 */
Al Viro643822b2011-11-24 22:00:28 -05002851bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002852 const struct path *root)
2853{
Al Viro643822b2011-11-24 22:00:28 -05002854 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002855 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002856 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002857 }
Al Viro643822b2011-11-24 22:00:28 -05002858 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002859}
2860
2861int path_is_under(struct path *path1, struct path *path2)
2862{
2863 int res;
Al Viro48a066e2013-09-29 22:06:07 -04002864 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002865 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04002866 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002867 return res;
2868}
2869EXPORT_SYMBOL(path_is_under);
2870
2871/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 * pivot_root Semantics:
2873 * Moves the root file system of the current process to the directory put_old,
2874 * makes new_root as the new root file system of the current process, and sets
2875 * root/cwd of all processes which had them on the current root to new_root.
2876 *
2877 * Restrictions:
2878 * The new_root and put_old must be directories, and must not be on the
2879 * same file system as the current process root. The put_old must be
2880 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2881 * pointed to by put_old must yield the same directory as new_root. No other
2882 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2883 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002884 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2885 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2886 * in this situation.
2887 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 * Notes:
2889 * - we don't move root/cwd if they are not at the root (reason: if something
2890 * cared enough to change them, it's probably wrong to force them elsewhere)
2891 * - it's okay to pick a root that isn't the root of a file system, e.g.
2892 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2893 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2894 * first.
2895 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002896SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2897 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898{
Al Viro2d8f3032008-07-22 09:59:21 -04002899 struct path new, old, parent_path, root_parent, root;
Al Viro84d17192013-03-15 10:53:28 -04002900 struct mount *new_mnt, *root_mnt, *old_mnt;
2901 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 int error;
2903
Al Viro9b40bc92013-02-22 22:45:42 -05002904 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 return -EPERM;
2906
Al Viro2d8f3032008-07-22 09:59:21 -04002907 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 if (error)
2909 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910
Al Viro2d8f3032008-07-22 09:59:21 -04002911 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 if (error)
2913 goto out1;
2914
Al Viro2d8f3032008-07-22 09:59:21 -04002915 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002916 if (error)
2917 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002919 get_fs_root(current->fs, &root);
Al Viro84d17192013-03-15 10:53:28 -04002920 old_mp = lock_mount(&old);
2921 error = PTR_ERR(old_mp);
2922 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04002923 goto out3;
2924
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002926 new_mnt = real_mount(new.mnt);
2927 root_mnt = real_mount(root.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002928 old_mnt = real_mount(old.mnt);
2929 if (IS_MNT_SHARED(old_mnt) ||
Al Virofc7be132011-11-25 01:05:37 -05002930 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2931 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002932 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002933 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002934 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002935 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2936 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002938 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002939 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 error = -EBUSY;
Al Viro84d17192013-03-15 10:53:28 -04002941 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04002942 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002944 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002945 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002946 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002947 goto out4; /* not attached */
Al Viro84d17192013-03-15 10:53:28 -04002948 root_mp = root_mnt->mnt_mp;
Al Viro2d8f3032008-07-22 09:59:21 -04002949 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002950 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002951 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002952 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002953 /* make sure we can reach put_old from new_root */
Al Viro84d17192013-03-15 10:53:28 -04002954 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002955 goto out4;
Eric W. Biederman0d082602014-10-08 10:42:27 -07002956 /* make certain new is below the root */
2957 if (!is_path_reachable(new_mnt, new.dentry, &root))
2958 goto out4;
Al Viro84d17192013-03-15 10:53:28 -04002959 root_mp->m_count++; /* pin it so it won't go away */
Al Viro719ea2f2013-09-29 11:24:49 -04002960 lock_mount_hash();
Al Viro419148d2011-11-24 19:41:16 -05002961 detach_mnt(new_mnt, &parent_path);
2962 detach_mnt(root_mnt, &root_parent);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002963 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2964 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2965 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2966 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002967 /* mount old root on put_old */
Al Viro84d17192013-03-15 10:53:28 -04002968 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08002969 /* mount new_root on / */
Al Viro84d17192013-03-15 10:53:28 -04002970 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002971 touch_mnt_namespace(current->nsproxy->mnt_ns);
Eric W. Biederman4fed6552014-10-08 10:42:57 -07002972 /* A moved mount should not expire automatically */
2973 list_del_init(&new_mnt->mnt_expire);
Al Viro719ea2f2013-09-29 11:24:49 -04002974 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04002975 chroot_fs_refs(&root, &new);
Al Viro84d17192013-03-15 10:53:28 -04002976 put_mountpoint(root_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002978out4:
Al Viro84d17192013-03-15 10:53:28 -04002979 unlock_mount(old_mp);
Al Virob12cea92011-03-18 08:55:38 -04002980 if (!error) {
2981 path_put(&root_parent);
2982 path_put(&parent_path);
2983 }
2984out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002985 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002986out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002987 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002989 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992}
2993
2994static void __init init_mount_tree(void)
2995{
2996 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002997 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002998 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002999 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003001 type = get_fs_type("rootfs");
3002 if (!type)
3003 panic("Can't find rootfs type");
3004 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3005 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 if (IS_ERR(mnt))
3007 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11003008
Trond Myklebust3b22edc2009-06-23 17:29:49 -04003009 ns = create_mnt_ns(mnt);
3010 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003013 init_task.nsproxy->mnt_ns = ns;
3014 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015
Al Virobe08d6d2011-12-06 13:32:36 -05003016 root.mnt = mnt;
3017 root.dentry = mnt->mnt_root;
Eric W. Biedermanda362b02014-10-07 12:19:53 -07003018 mnt->mnt_flags |= MNT_LOCKED;
Jan Blunckac748a02008-02-14 19:34:39 -08003019
3020 set_fs_pwd(current->fs, &root);
3021 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022}
3023
Denis Cheng74bf17c2007-10-16 23:26:30 -07003024void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025{
Eric Dumazet13f14b42008-02-06 01:37:57 -08003026 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003027 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028
Al Viro7d6fec42011-11-23 12:14:10 -05003029 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09003030 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031
Al Viro0818bf22014-02-28 13:46:44 -05003032 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04003033 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05003034 mhash_entries, 19,
3035 0,
3036 &m_hash_shift, &m_hash_mask, 0, 0);
3037 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3038 sizeof(struct hlist_head),
3039 mphash_entries, 19,
3040 0,
3041 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042
Al Viro84d17192013-03-15 10:53:28 -04003043 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 panic("Failed to allocate mount hash table\n");
3045
Al Viro0818bf22014-02-28 13:46:44 -05003046 for (u = 0; u <= m_hash_mask; u++)
Al Viro38129a12014-03-20 21:10:51 -04003047 INIT_HLIST_HEAD(&mount_hashtable[u]);
Al Viro0818bf22014-02-28 13:46:44 -05003048 for (u = 0; u <= mp_hash_mask; u++)
3049 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050
Tejun Heo4b93dc92013-11-28 14:54:43 -05003051 kernfs_init();
3052
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003053 err = sysfs_init();
3054 if (err)
3055 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003056 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06003057 fs_kobj = kobject_create_and_add("fs", NULL);
3058 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003059 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 init_rootfs();
3061 init_mount_tree();
3062}
3063
Trond Myklebust616511d2009-06-22 15:09:13 -04003064void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065{
Al Virod498b252010-02-05 02:21:06 -05003066 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04003067 return;
Al Viro7b00ed62013-09-16 21:19:20 -04003068 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003069 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070}
Al Viro9d412a42011-03-17 22:08:28 -04003071
3072struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3073{
Tim Chen423e0ab2011-07-19 09:32:38 -07003074 struct vfsmount *mnt;
3075 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3076 if (!IS_ERR(mnt)) {
3077 /*
3078 * it is a longterm mount, don't release mnt until
3079 * we unmount before file sys is unregistered
3080 */
Al Virof7a99c52012-06-09 00:59:08 -04003081 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07003082 }
3083 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04003084}
3085EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07003086
3087void kern_unmount(struct vfsmount *mnt)
3088{
3089 /* release long term mount so mount point can be released */
3090 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04003091 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04003092 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07003093 mntput(mnt);
3094 }
3095}
3096EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05003097
3098bool our_mnt(struct vfsmount *mnt)
3099{
Al Viro143c8c92011-11-25 00:46:35 -05003100 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05003101}
Eric W. Biederman8823c072010-03-07 18:49:36 -08003102
Eric W. Biederman31515272013-03-15 01:45:51 -07003103bool current_chrooted(void)
3104{
3105 /* Does the current process have a non-standard root */
3106 struct path ns_root;
3107 struct path fs_root;
3108 bool chrooted;
3109
3110 /* Find the namespace root */
3111 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3112 ns_root.dentry = ns_root.mnt->mnt_root;
3113 path_get(&ns_root);
3114 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3115 ;
3116
3117 get_fs_root(current->fs, &fs_root);
3118
3119 chrooted = !path_equal(&fs_root, &ns_root);
3120
3121 path_put(&fs_root);
3122 path_put(&ns_root);
3123
3124 return chrooted;
3125}
3126
Eric W. Biedermane51db732013-03-30 19:57:41 -07003127bool fs_fully_visible(struct file_system_type *type)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003128{
3129 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3130 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07003131 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003132
Eric W. Biedermane51db732013-03-30 19:57:41 -07003133 if (unlikely(!ns))
3134 return false;
3135
Al Viro44bb4382013-09-16 21:37:36 -04003136 down_read(&namespace_sem);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003137 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07003138 struct mount *child;
3139 if (mnt->mnt.mnt_sb->s_type != type)
3140 continue;
3141
3142 /* This mount is not fully visible if there are any child mounts
3143 * that cover anything except for empty directories.
3144 */
3145 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3146 struct inode *inode = child->mnt_mountpoint->d_inode;
3147 if (!S_ISDIR(inode->i_mode))
3148 goto next;
Eric W. Biederman41301ae2013-11-14 21:22:25 -08003149 if (inode->i_nlink > 2)
Eric W. Biedermane51db732013-03-30 19:57:41 -07003150 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003151 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07003152 visible = true;
3153 goto found;
3154 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003155 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07003156found:
Al Viro44bb4382013-09-16 21:37:36 -04003157 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003158 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003159}
3160
Al Viro64964522014-11-01 00:37:32 -04003161static struct ns_common *mntns_get(struct task_struct *task)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003162{
Al Viro58be28252014-11-01 00:00:23 -04003163 struct ns_common *ns = NULL;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003164 struct nsproxy *nsproxy;
3165
Eric W. Biederman728dba32014-02-03 19:13:49 -08003166 task_lock(task);
3167 nsproxy = task->nsproxy;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003168 if (nsproxy) {
Al Viro58be28252014-11-01 00:00:23 -04003169 ns = &nsproxy->mnt_ns->ns;
3170 get_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003171 }
Eric W. Biederman728dba32014-02-03 19:13:49 -08003172 task_unlock(task);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003173
3174 return ns;
3175}
3176
Al Viro64964522014-11-01 00:37:32 -04003177static void mntns_put(struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003178{
Al Viro58be28252014-11-01 00:00:23 -04003179 put_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08003180}
3181
Al Viro64964522014-11-01 00:37:32 -04003182static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08003183{
3184 struct fs_struct *fs = current->fs;
Al Viro58be28252014-11-01 00:00:23 -04003185 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
Eric W. Biederman8823c072010-03-07 18:49:36 -08003186 struct path root;
3187
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003188 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -07003189 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3190 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08003191 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003192
3193 if (fs->users != 1)
3194 return -EINVAL;
3195
3196 get_mnt_ns(mnt_ns);
3197 put_mnt_ns(nsproxy->mnt_ns);
3198 nsproxy->mnt_ns = mnt_ns;
3199
3200 /* Find the root */
3201 root.mnt = &mnt_ns->root->mnt;
3202 root.dentry = mnt_ns->root->mnt.mnt_root;
3203 path_get(&root);
3204 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3205 ;
3206
3207 /* Update the pwd and root */
3208 set_fs_pwd(fs, &root);
3209 set_fs_root(fs, &root);
3210
3211 path_put(&root);
3212 return 0;
3213}
3214
3215const struct proc_ns_operations mntns_operations = {
3216 .name = "mnt",
3217 .type = CLONE_NEWNS,
3218 .get = mntns_get,
3219 .put = mntns_put,
3220 .install = mntns_install,
3221};