blob: 182bc41cd88711d593c4d997171c6ad483a87577 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
Al Virod10577a2011-12-07 13:06:11 -050012#include <linux/export.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080013#include <linux/capability.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080014#include <linux/mnt_namespace.h>
Eric W. Biederman771b1372012-07-26 21:08:32 -070015#include <linux/user_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/namei.h>
17#include <linux/security.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010018#include <linux/idr.h>
Al Virod10577a2011-12-07 13:06:11 -050019#include <linux/acct.h> /* acct_auto_close_mnt */
Rob Landley57f150a2013-09-11 14:26:10 -070020#include <linux/init.h> /* init_rootfs */
Al Virod10577a2011-12-07 13:06:11 -050021#include <linux/fs_struct.h> /* get_fs_root et.al. */
22#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23#include <linux/uaccess.h>
David Howells0bb80f22013-04-12 01:50:06 +010024#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070025#include <linux/magic.h>
Al Viro0818bf22014-02-28 13:46:44 -050026#include <linux/bootmem.h>
Ram Pai07b20882005-11-07 17:19:07 -050027#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070028#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Al Viro0818bf22014-02-28 13:46:44 -050030static unsigned int m_hash_mask __read_mostly;
31static unsigned int m_hash_shift __read_mostly;
32static unsigned int mp_hash_mask __read_mostly;
33static unsigned int mp_hash_shift __read_mostly;
34
35static __initdata unsigned long mhash_entries;
36static int __init set_mhash_entries(char *str)
37{
38 if (!str)
39 return 0;
40 mhash_entries = simple_strtoul(str, &str, 0);
41 return 1;
42}
43__setup("mhash_entries=", set_mhash_entries);
44
45static __initdata unsigned long mphash_entries;
46static int __init set_mphash_entries(char *str)
47{
48 if (!str)
49 return 0;
50 mphash_entries = simple_strtoul(str, &str, 0);
51 return 1;
52}
53__setup("mphash_entries=", set_mphash_entries);
Eric Dumazet13f14b42008-02-06 01:37:57 -080054
Al Viroc7999c32014-02-27 14:40:10 -050055static u64 event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010056static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010057static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100058static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040059static int mnt_id_start = 0;
60static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Al Viro38129a12014-03-20 21:10:51 -040062static struct hlist_head *mount_hashtable __read_mostly;
Al Viro0818bf22014-02-28 13:46:44 -050063static struct hlist_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080064static struct kmem_cache *mnt_cache __read_mostly;
Al Viro59aa0da2013-09-16 21:34:53 -040065static DECLARE_RWSEM(namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080067/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060068struct kobject *fs_kobj;
69EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080070
Nick Piggin99b7db72010-08-18 04:37:39 +100071/*
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
74 * up the tree.
75 *
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
78 */
Al Viro48a066e2013-09-29 22:06:07 -040079__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +100080
Al Viro38129a12014-03-20 21:10:51 -040081static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082{
Ram Paib58fed82005-11-07 17:16:09 -050083 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Al Viro0818bf22014-02-28 13:46:44 -050085 tmp = tmp + (tmp >> m_hash_shift);
86 return &mount_hashtable[tmp & m_hash_mask];
87}
88
89static inline struct hlist_head *mp_hash(struct dentry *dentry)
90{
91 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 tmp = tmp + (tmp >> mp_hash_shift);
93 return &mountpoint_hashtable[tmp & mp_hash_mask];
Linus Torvalds1da177e2005-04-16 15:20:36 -070094}
95
Nick Piggin99b7db72010-08-18 04:37:39 +100096/*
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
99 */
Al Virob105e272011-11-24 20:38:33 -0500100static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100101{
102 int res;
103
104retry:
105 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +1000106 spin_lock(&mnt_id_lock);
Al Viro15169fe2011-11-25 00:50:41 -0500107 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
Al Virof21f6222009-06-24 03:12:00 -0400108 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500109 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +1000110 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100111 if (res == -EAGAIN)
112 goto retry;
113
114 return res;
115}
116
Al Virob105e272011-11-24 20:38:33 -0500117static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100118{
Al Viro15169fe2011-11-25 00:50:41 -0500119 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000120 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -0400121 ida_remove(&mnt_id_ida, id);
122 if (mnt_id_start > id)
123 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000124 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100125}
126
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100127/*
128 * Allocate a new peer group ID
129 *
130 * mnt_group_ida is protected by namespace_sem
131 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500132static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100133{
Al Virof21f6222009-06-24 03:12:00 -0400134 int res;
135
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100136 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
137 return -ENOMEM;
138
Al Virof21f6222009-06-24 03:12:00 -0400139 res = ida_get_new_above(&mnt_group_ida,
140 mnt_group_start,
Al Viro15169fe2011-11-25 00:50:41 -0500141 &mnt->mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400142 if (!res)
Al Viro15169fe2011-11-25 00:50:41 -0500143 mnt_group_start = mnt->mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400144
145 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100146}
147
148/*
149 * Release a peer group ID
150 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500151void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100152{
Al Viro15169fe2011-11-25 00:50:41 -0500153 int id = mnt->mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400154 ida_remove(&mnt_group_ida, id);
155 if (mnt_group_start > id)
156 mnt_group_start = id;
Al Viro15169fe2011-11-25 00:50:41 -0500157 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100158}
159
Nick Pigginb3e19d92011-01-07 17:50:11 +1100160/*
161 * vfsmount lock must be held for read
162 */
Al Viro83adc752011-11-24 22:37:54 -0500163static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100164{
165#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500166 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100167#else
168 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500169 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100170 preempt_enable();
171#endif
172}
173
Nick Pigginb3e19d92011-01-07 17:50:11 +1100174/*
175 * vfsmount lock must be held for write
176 */
Al Viro83adc752011-11-24 22:37:54 -0500177unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100178{
179#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500180 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100181 int cpu;
182
183 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500184 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100185 }
186
187 return count;
188#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500189 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100190#endif
191}
192
Al Virob105e272011-11-24 20:38:33 -0500193static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194{
Al Viroc63181e2011-11-25 02:35:16 -0500195 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
196 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100197 int err;
198
Al Viroc63181e2011-11-25 02:35:16 -0500199 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800200 if (err)
201 goto out_free_cache;
202
203 if (name) {
Al Viroc63181e2011-11-25 02:35:16 -0500204 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
205 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800206 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100207 }
208
Nick Pigginb3e19d92011-01-07 17:50:11 +1100209#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500210 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
211 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100212 goto out_free_devname;
213
Al Viroc63181e2011-11-25 02:35:16 -0500214 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100215#else
Al Viroc63181e2011-11-25 02:35:16 -0500216 mnt->mnt_count = 1;
217 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100218#endif
219
Al Viro38129a12014-03-20 21:10:51 -0400220 INIT_HLIST_NODE(&mnt->mnt_hash);
Al Viroc63181e2011-11-25 02:35:16 -0500221 INIT_LIST_HEAD(&mnt->mnt_child);
222 INIT_LIST_HEAD(&mnt->mnt_mounts);
223 INIT_LIST_HEAD(&mnt->mnt_list);
224 INIT_LIST_HEAD(&mnt->mnt_expire);
225 INIT_LIST_HEAD(&mnt->mnt_share);
226 INIT_LIST_HEAD(&mnt->mnt_slave_list);
227 INIT_LIST_HEAD(&mnt->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500228#ifdef CONFIG_FSNOTIFY
229 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
230#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 }
Al Viroc63181e2011-11-25 02:35:16 -0500232 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800233
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000234#ifdef CONFIG_SMP
235out_free_devname:
Al Viroc63181e2011-11-25 02:35:16 -0500236 kfree(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000237#endif
Li Zefan88b38782008-07-21 18:06:36 +0800238out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500239 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800240out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500241 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800242 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
244
Dave Hansen83660252008-02-15 14:37:30 -0800245/*
246 * Most r/o checks on a fs are for operations that take
247 * discrete amounts of time, like a write() or unlink().
248 * We must keep track of when those operations start
249 * (for permission checks) and when they end, so that
250 * we can determine when writes are able to occur to
251 * a filesystem.
252 */
Dave Hansen3d733632008-02-15 14:37:59 -0800253/*
254 * __mnt_is_readonly: check whether a mount is read-only
255 * @mnt: the mount to check for its write status
256 *
257 * This shouldn't be used directly ouside of the VFS.
258 * It does not guarantee that the filesystem will stay
259 * r/w, just that it is right *now*. This can not and
260 * should not be used in place of IS_RDONLY(inode).
261 * mnt_want/drop_write() will _keep_ the filesystem
262 * r/w.
263 */
264int __mnt_is_readonly(struct vfsmount *mnt)
265{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800266 if (mnt->mnt_flags & MNT_READONLY)
267 return 1;
268 if (mnt->mnt_sb->s_flags & MS_RDONLY)
269 return 1;
270 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800271}
272EXPORT_SYMBOL_GPL(__mnt_is_readonly);
273
Al Viro83adc752011-11-24 22:37:54 -0500274static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800275{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000276#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500277 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000278#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500279 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000280#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800281}
Dave Hansen3d733632008-02-15 14:37:59 -0800282
Al Viro83adc752011-11-24 22:37:54 -0500283static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800284{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000285#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500286 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000287#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500288 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000289#endif
290}
291
Al Viro83adc752011-11-24 22:37:54 -0500292static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000293{
294#ifdef CONFIG_SMP
295 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800296 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800297
298 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500299 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800300 }
Dave Hansen3d733632008-02-15 14:37:59 -0800301
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000302 return count;
303#else
304 return mnt->mnt_writers;
305#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800306}
307
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100308static int mnt_is_readonly(struct vfsmount *mnt)
309{
310 if (mnt->mnt_sb->s_readonly_remount)
311 return 1;
312 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
313 smp_rmb();
314 return __mnt_is_readonly(mnt);
315}
316
Dave Hansen3d733632008-02-15 14:37:59 -0800317/*
Jan Karaeb04c282012-06-12 16:20:35 +0200318 * Most r/o & frozen checks on a fs are for operations that take discrete
319 * amounts of time, like a write() or unlink(). We must keep track of when
320 * those operations start (for permission checks) and when they end, so that we
321 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800322 */
Dave Hansen83660252008-02-15 14:37:30 -0800323/**
Jan Karaeb04c282012-06-12 16:20:35 +0200324 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500325 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800326 *
Jan Karaeb04c282012-06-12 16:20:35 +0200327 * This tells the low-level filesystem that a write is about to be performed to
328 * it, and makes sure that writes are allowed (mnt it read-write) before
329 * returning success. This operation does not protect against filesystem being
330 * frozen. When the write operation is finished, __mnt_drop_write() must be
331 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800332 */
Jan Karaeb04c282012-06-12 16:20:35 +0200333int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800334{
Al Viro83adc752011-11-24 22:37:54 -0500335 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800336 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800337
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000338 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100339 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000340 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100341 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000342 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
343 * incremented count after it has set MNT_WRITE_HOLD.
344 */
345 smp_mb();
Miao Xie1e755292012-11-16 17:23:50 +0800346 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000347 cpu_relax();
348 /*
349 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
350 * be set to match its requirements. So we must not load that until
351 * MNT_WRITE_HOLD is cleared.
352 */
353 smp_rmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100354 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100355 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800356 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800357 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000358 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200359
360 return ret;
361}
362
363/**
364 * mnt_want_write - get write access to a mount
365 * @m: the mount on which to take a write
366 *
367 * This tells the low-level filesystem that a write is about to be performed to
368 * it, and makes sure that writes are allowed (mount is read-write, filesystem
369 * is not frozen) before returning success. When the write operation is
370 * finished, mnt_drop_write() must be called. This is effectively a refcount.
371 */
372int mnt_want_write(struct vfsmount *m)
373{
374 int ret;
375
376 sb_start_write(m->mnt_sb);
377 ret = __mnt_want_write(m);
378 if (ret)
379 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800380 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800381}
382EXPORT_SYMBOL_GPL(mnt_want_write);
383
384/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000385 * mnt_clone_write - get write access to a mount
386 * @mnt: the mount on which to take a write
387 *
388 * This is effectively like mnt_want_write, except
389 * it must only be used to take an extra write reference
390 * on a mountpoint that we already know has a write reference
391 * on it. This allows some optimisation.
392 *
393 * After finished, mnt_drop_write must be called as usual to
394 * drop the reference.
395 */
396int mnt_clone_write(struct vfsmount *mnt)
397{
398 /* superblock may be r/o */
399 if (__mnt_is_readonly(mnt))
400 return -EROFS;
401 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500402 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000403 preempt_enable();
404 return 0;
405}
406EXPORT_SYMBOL_GPL(mnt_clone_write);
407
408/**
Jan Karaeb04c282012-06-12 16:20:35 +0200409 * __mnt_want_write_file - get write access to a file's mount
410 * @file: the file who's mount on which to take a write
411 *
412 * This is like __mnt_want_write, but it takes a file and can
413 * do some optimisations if the file is open for write already
414 */
415int __mnt_want_write_file(struct file *file)
416{
Al Viro83f936c2014-03-14 12:02:47 -0400417 if (!(file->f_mode & FMODE_WRITER))
Jan Karaeb04c282012-06-12 16:20:35 +0200418 return __mnt_want_write(file->f_path.mnt);
419 else
420 return mnt_clone_write(file->f_path.mnt);
421}
422
423/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000424 * mnt_want_write_file - get write access to a file's mount
425 * @file: the file who's mount on which to take a write
426 *
427 * This is like mnt_want_write, but it takes a file and can
428 * do some optimisations if the file is open for write already
429 */
430int mnt_want_write_file(struct file *file)
431{
Jan Karaeb04c282012-06-12 16:20:35 +0200432 int ret;
433
434 sb_start_write(file->f_path.mnt->mnt_sb);
435 ret = __mnt_want_write_file(file);
436 if (ret)
437 sb_end_write(file->f_path.mnt->mnt_sb);
438 return ret;
npiggin@suse.de96029c42009-04-26 20:25:55 +1000439}
440EXPORT_SYMBOL_GPL(mnt_want_write_file);
441
442/**
Jan Karaeb04c282012-06-12 16:20:35 +0200443 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800444 * @mnt: the mount on which to give up write access
445 *
446 * Tells the low-level filesystem that we are done
447 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200448 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800449 */
Jan Karaeb04c282012-06-12 16:20:35 +0200450void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800451{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000452 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500453 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000454 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800455}
Jan Karaeb04c282012-06-12 16:20:35 +0200456
457/**
458 * mnt_drop_write - give up write access to a mount
459 * @mnt: the mount on which to give up write access
460 *
461 * Tells the low-level filesystem that we are done performing writes to it and
462 * also allows filesystem to be frozen again. Must be matched with
463 * mnt_want_write() call above.
464 */
465void mnt_drop_write(struct vfsmount *mnt)
466{
467 __mnt_drop_write(mnt);
468 sb_end_write(mnt->mnt_sb);
469}
Dave Hansen83660252008-02-15 14:37:30 -0800470EXPORT_SYMBOL_GPL(mnt_drop_write);
471
Jan Karaeb04c282012-06-12 16:20:35 +0200472void __mnt_drop_write_file(struct file *file)
473{
474 __mnt_drop_write(file->f_path.mnt);
475}
476
Al Viro2a79f172011-12-09 08:06:57 -0500477void mnt_drop_write_file(struct file *file)
478{
479 mnt_drop_write(file->f_path.mnt);
480}
481EXPORT_SYMBOL(mnt_drop_write_file);
482
Al Viro83adc752011-11-24 22:37:54 -0500483static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800484{
Dave Hansen3d733632008-02-15 14:37:59 -0800485 int ret = 0;
486
Al Viro719ea2f2013-09-29 11:24:49 -0400487 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500488 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000489 /*
490 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
491 * should be visible before we do.
492 */
493 smp_mb();
494
495 /*
496 * With writers on hold, if this value is zero, then there are
497 * definitely no active writers (although held writers may subsequently
498 * increment the count, they'll have to wait, and decrement it after
499 * seeing MNT_READONLY).
500 *
501 * It is OK to have counter incremented on one CPU and decremented on
502 * another: the sum will add up correctly. The danger would be when we
503 * sum up each counter, if we read a counter before it is incremented,
504 * but then read another CPU's count which it has been subsequently
505 * decremented from -- we would see more decrements than we should.
506 * MNT_WRITE_HOLD protects against this scenario, because
507 * mnt_want_write first increments count, then smp_mb, then spins on
508 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
509 * we're counting up here.
510 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100511 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000512 ret = -EBUSY;
513 else
Al Viro83adc752011-11-24 22:37:54 -0500514 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000515 /*
516 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
517 * that become unheld will see MNT_READONLY.
518 */
519 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500520 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Al Viro719ea2f2013-09-29 11:24:49 -0400521 unlock_mount_hash();
Dave Hansen3d733632008-02-15 14:37:59 -0800522 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800523}
Dave Hansen83660252008-02-15 14:37:30 -0800524
Al Viro83adc752011-11-24 22:37:54 -0500525static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800526{
Al Viro719ea2f2013-09-29 11:24:49 -0400527 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500528 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Al Viro719ea2f2013-09-29 11:24:49 -0400529 unlock_mount_hash();
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800530}
531
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100532int sb_prepare_remount_readonly(struct super_block *sb)
533{
534 struct mount *mnt;
535 int err = 0;
536
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100537 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
538 if (atomic_long_read(&sb->s_remove_count))
539 return -EBUSY;
540
Al Viro719ea2f2013-09-29 11:24:49 -0400541 lock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100542 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
543 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
544 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
545 smp_mb();
546 if (mnt_get_writers(mnt) > 0) {
547 err = -EBUSY;
548 break;
549 }
550 }
551 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100552 if (!err && atomic_long_read(&sb->s_remove_count))
553 err = -EBUSY;
554
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100555 if (!err) {
556 sb->s_readonly_remount = 1;
557 smp_wmb();
558 }
559 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
560 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
561 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
562 }
Al Viro719ea2f2013-09-29 11:24:49 -0400563 unlock_mount_hash();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100564
565 return err;
566}
567
Al Virob105e272011-11-24 20:38:33 -0500568static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
Al Viro52ba1622011-11-25 02:25:17 -0500570 kfree(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000571#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500572 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000573#endif
Al Virob105e272011-11-24 20:38:33 -0500574 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575}
576
David Howells8ffcb322014-01-24 12:17:54 +0000577static void delayed_free_vfsmnt(struct rcu_head *head)
578{
579 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
580}
581
Al Viro48a066e2013-09-29 22:06:07 -0400582/* call under rcu_read_lock */
583bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
584{
585 struct mount *mnt;
586 if (read_seqretry(&mount_lock, seq))
587 return false;
588 if (bastard == NULL)
589 return true;
590 mnt = real_mount(bastard);
591 mnt_add_count(mnt, 1);
592 if (likely(!read_seqretry(&mount_lock, seq)))
593 return true;
594 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
595 mnt_add_count(mnt, -1);
596 return false;
597 }
598 rcu_read_unlock();
599 mntput(bastard);
600 rcu_read_lock();
601 return false;
602}
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604/*
Al Viro474279d2013-10-01 16:11:26 -0400605 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400606 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
Al Viro474279d2013-10-01 16:11:26 -0400608struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Al Viro38129a12014-03-20 21:10:51 -0400610 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400611 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Al Viro38129a12014-03-20 21:10:51 -0400613 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400614 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
615 return p;
616 return NULL;
617}
618
619/*
620 * find the last mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400621 * mount_lock must be held.
Al Viro474279d2013-10-01 16:11:26 -0400622 */
623struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
624{
Al Viro38129a12014-03-20 21:10:51 -0400625 struct mount *p, *res;
626 res = p = __lookup_mnt(mnt, dentry);
627 if (!p)
628 goto out;
629 hlist_for_each_entry_continue(p, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -0400630 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
631 break;
632 res = p;
633 }
Al Viro38129a12014-03-20 21:10:51 -0400634out:
Al Viro1d6a32a2014-03-20 20:34:43 -0400635 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636}
637
Ram Paia05964f2005-11-07 17:20:17 -0500638/*
David Howellsf015f1262012-06-25 12:55:28 +0100639 * lookup_mnt - Return the first child mount mounted at path
640 *
641 * "First" means first mounted chronologically. If you create the
642 * following mounts:
643 *
644 * mount /dev/sda1 /mnt
645 * mount /dev/sda2 /mnt
646 * mount /dev/sda3 /mnt
647 *
648 * Then lookup_mnt() on the base /mnt dentry in the root mount will
649 * return successively the root dentry and vfsmount of /dev/sda1, then
650 * /dev/sda2, then /dev/sda3, then NULL.
651 *
652 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500653 */
Al Viro1c755af2009-04-18 14:06:57 -0400654struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500655{
Al Viroc7105362011-11-24 18:22:03 -0500656 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400657 struct vfsmount *m;
658 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000659
Al Viro48a066e2013-09-29 22:06:07 -0400660 rcu_read_lock();
661 do {
662 seq = read_seqbegin(&mount_lock);
663 child_mnt = __lookup_mnt(path->mnt, path->dentry);
664 m = child_mnt ? &child_mnt->mnt : NULL;
665 } while (!legitimize_mnt(m, seq));
666 rcu_read_unlock();
667 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500668}
669
Al Viro84d17192013-03-15 10:53:28 -0400670static struct mountpoint *new_mountpoint(struct dentry *dentry)
671{
Al Viro0818bf22014-02-28 13:46:44 -0500672 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d17192013-03-15 10:53:28 -0400673 struct mountpoint *mp;
Miklos Szeredieed81002013-09-05 14:39:11 +0200674 int ret;
Al Viro84d17192013-03-15 10:53:28 -0400675
Al Viro0818bf22014-02-28 13:46:44 -0500676 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d17192013-03-15 10:53:28 -0400677 if (mp->m_dentry == dentry) {
678 /* might be worth a WARN_ON() */
679 if (d_unlinked(dentry))
680 return ERR_PTR(-ENOENT);
681 mp->m_count++;
682 return mp;
683 }
684 }
685
686 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
687 if (!mp)
688 return ERR_PTR(-ENOMEM);
689
Miklos Szeredieed81002013-09-05 14:39:11 +0200690 ret = d_set_mounted(dentry);
691 if (ret) {
Al Viro84d17192013-03-15 10:53:28 -0400692 kfree(mp);
Miklos Szeredieed81002013-09-05 14:39:11 +0200693 return ERR_PTR(ret);
Al Viro84d17192013-03-15 10:53:28 -0400694 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200695
Al Viro84d17192013-03-15 10:53:28 -0400696 mp->m_dentry = dentry;
697 mp->m_count = 1;
Al Viro0818bf22014-02-28 13:46:44 -0500698 hlist_add_head(&mp->m_hash, chain);
Al Viro84d17192013-03-15 10:53:28 -0400699 return mp;
700}
701
702static void put_mountpoint(struct mountpoint *mp)
703{
704 if (!--mp->m_count) {
705 struct dentry *dentry = mp->m_dentry;
706 spin_lock(&dentry->d_lock);
707 dentry->d_flags &= ~DCACHE_MOUNTED;
708 spin_unlock(&dentry->d_lock);
Al Viro0818bf22014-02-28 13:46:44 -0500709 hlist_del(&mp->m_hash);
Al Viro84d17192013-03-15 10:53:28 -0400710 kfree(mp);
711 }
712}
713
Al Viro143c8c92011-11-25 00:46:35 -0500714static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800716 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
Nick Piggin99b7db72010-08-18 04:37:39 +1000719/*
720 * vfsmount lock must be held for write
721 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800722static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500723{
724 if (ns) {
725 ns->event = ++event;
726 wake_up_interruptible(&ns->poll);
727 }
728}
729
Nick Piggin99b7db72010-08-18 04:37:39 +1000730/*
731 * vfsmount lock must be held for write
732 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800733static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500734{
735 if (ns && ns->event != event) {
736 ns->event = event;
737 wake_up_interruptible(&ns->poll);
738 }
739}
740
Nick Piggin99b7db72010-08-18 04:37:39 +1000741/*
742 * vfsmount lock must be held for write
743 */
Al Viro419148d2011-11-24 19:41:16 -0500744static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
Al Viroa73324d2011-11-24 22:25:07 -0500746 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500747 old_path->mnt = &mnt->mnt_parent->mnt;
748 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500749 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500750 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400751 hlist_del_init_rcu(&mnt->mnt_hash);
Al Viro84d17192013-03-15 10:53:28 -0400752 put_mountpoint(mnt->mnt_mp);
753 mnt->mnt_mp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754}
755
Nick Piggin99b7db72010-08-18 04:37:39 +1000756/*
757 * vfsmount lock must be held for write
758 */
Al Viro84d17192013-03-15 10:53:28 -0400759void mnt_set_mountpoint(struct mount *mnt,
760 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500761 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500762{
Al Viro84d17192013-03-15 10:53:28 -0400763 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500764 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro84d17192013-03-15 10:53:28 -0400765 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
Al Viro3a2393d2011-11-25 03:19:09 -0500766 child_mnt->mnt_parent = mnt;
Al Viro84d17192013-03-15 10:53:28 -0400767 child_mnt->mnt_mp = mp;
Ram Paib90fa9a2005-11-07 17:19:50 -0500768}
769
Nick Piggin99b7db72010-08-18 04:37:39 +1000770/*
771 * vfsmount lock must be held for write
772 */
Al Viro84d17192013-03-15 10:53:28 -0400773static void attach_mnt(struct mount *mnt,
774 struct mount *parent,
775 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
Al Viro84d17192013-03-15 10:53:28 -0400777 mnt_set_mountpoint(parent, mp, mnt);
Al Viro38129a12014-03-20 21:10:51 -0400778 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
Al Viro84d17192013-03-15 10:53:28 -0400779 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500780}
781
782/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000783 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500784 */
Al Viro1d6a32a2014-03-20 20:34:43 -0400785static void commit_tree(struct mount *mnt, struct mount *shadows)
Ram Paib90fa9a2005-11-07 17:19:50 -0500786{
Al Viro0714a532011-11-24 22:19:58 -0500787 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500788 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500789 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500790 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500791
Al Viro0714a532011-11-24 22:19:58 -0500792 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500793
Al Viro1a4eeaf2011-11-25 02:19:55 -0500794 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400795 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500796 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500797
Ram Paib90fa9a2005-11-07 17:19:50 -0500798 list_splice(&head, n->list.prev);
799
Al Viro1d6a32a2014-03-20 20:34:43 -0400800 if (shadows)
Al Viro38129a12014-03-20 21:10:51 -0400801 hlist_add_after_rcu(&shadows->mnt_hash, &mnt->mnt_hash);
Al Viro1d6a32a2014-03-20 20:34:43 -0400802 else
Al Viro38129a12014-03-20 21:10:51 -0400803 hlist_add_head_rcu(&mnt->mnt_hash,
Al Viro0818bf22014-02-28 13:46:44 -0500804 m_hash(&parent->mnt, mnt->mnt_mountpoint));
Al Viro6b41d532011-11-24 23:24:33 -0500805 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800806 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807}
808
Al Viro909b0a82011-11-25 03:06:56 -0500809static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810{
Al Viro6b41d532011-11-24 23:24:33 -0500811 struct list_head *next = p->mnt_mounts.next;
812 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500814 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500816 next = p->mnt_child.next;
817 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 break;
Al Viro0714a532011-11-24 22:19:58 -0500819 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 }
821 }
Al Viro6b41d532011-11-24 23:24:33 -0500822 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823}
824
Al Viro315fc832011-11-24 18:57:30 -0500825static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500826{
Al Viro6b41d532011-11-24 23:24:33 -0500827 struct list_head *prev = p->mnt_mounts.prev;
828 while (prev != &p->mnt_mounts) {
829 p = list_entry(prev, struct mount, mnt_child);
830 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500831 }
832 return p;
833}
834
Al Viro9d412a42011-03-17 22:08:28 -0400835struct vfsmount *
836vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
837{
Al Virob105e272011-11-24 20:38:33 -0500838 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400839 struct dentry *root;
840
841 if (!type)
842 return ERR_PTR(-ENODEV);
843
844 mnt = alloc_vfsmnt(name);
845 if (!mnt)
846 return ERR_PTR(-ENOMEM);
847
848 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500849 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400850
851 root = mount_fs(type, flags, name, data);
852 if (IS_ERR(root)) {
David Howells8ffcb322014-01-24 12:17:54 +0000853 mnt_free_id(mnt);
Al Viro9d412a42011-03-17 22:08:28 -0400854 free_vfsmnt(mnt);
855 return ERR_CAST(root);
856 }
857
Al Virob105e272011-11-24 20:38:33 -0500858 mnt->mnt.mnt_root = root;
859 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500860 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500861 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400862 lock_mount_hash();
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100863 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400864 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -0500865 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400866}
867EXPORT_SYMBOL_GPL(vfs_kern_mount);
868
Al Viro87129cc2011-11-24 21:24:27 -0500869static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500870 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871{
Al Viro87129cc2011-11-24 21:24:27 -0500872 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +0100873 struct mount *mnt;
874 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
David Howellsbe34d1a2012-06-25 12:55:18 +0100876 mnt = alloc_vfsmnt(old->mnt_devname);
877 if (!mnt)
878 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100879
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700880 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +0100881 mnt->mnt_group_id = 0; /* not a peer of original */
882 else
883 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100884
David Howellsbe34d1a2012-06-25 12:55:18 +0100885 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
886 err = mnt_alloc_group_id(mnt);
887 if (err)
888 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
David Howellsbe34d1a2012-06-25 12:55:18 +0100890
Al Virof2ebb3a2014-02-27 09:35:45 -0500891 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
Eric W. Biederman132c94e2013-03-22 04:08:05 -0700892 /* Don't allow unprivileged users to change mount flags */
893 if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
894 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
895
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -0700896 /* Don't allow unprivileged users to reveal what is under a mount */
897 if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
898 mnt->mnt.mnt_flags |= MNT_LOCKED;
899
David Howellsbe34d1a2012-06-25 12:55:18 +0100900 atomic_inc(&sb->s_active);
901 mnt->mnt.mnt_sb = sb;
902 mnt->mnt.mnt_root = dget(root);
903 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
904 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -0400905 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +0100906 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400907 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +0100908
Eric W. Biederman7a472ef2012-07-31 13:13:04 -0700909 if ((flag & CL_SLAVE) ||
910 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +0100911 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
912 mnt->mnt_master = old;
913 CLEAR_MNT_SHARED(mnt);
914 } else if (!(flag & CL_PRIVATE)) {
915 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
916 list_add(&mnt->mnt_share, &old->mnt_share);
917 if (IS_MNT_SLAVE(old))
918 list_add(&mnt->mnt_slave, &old->mnt_slave);
919 mnt->mnt_master = old->mnt_master;
920 }
921 if (flag & CL_MAKE_SHARED)
922 set_mnt_shared(mnt);
923
924 /* stick the duplicate mount on the same expiry list
925 * as the original if that was on one */
926 if (flag & CL_EXPIRE) {
927 if (!list_empty(&old->mnt_expire))
928 list_add(&mnt->mnt_expire, &old->mnt_expire);
929 }
930
Al Virocb338d02011-11-24 20:55:08 -0500931 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100932
933 out_free:
David Howells8ffcb322014-01-24 12:17:54 +0000934 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100935 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +0100936 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937}
938
Al Viro900148d2011-11-25 00:33:11 -0500939static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500940{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100941put_again:
Al Viro48a066e2013-09-29 22:06:07 -0400942 rcu_read_lock();
943 mnt_add_count(mnt, -1);
944 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
945 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -0500946 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100947 }
Al Viro719ea2f2013-09-29 11:24:49 -0400948 lock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +1100949 if (mnt_get_count(mnt)) {
Al Viro48a066e2013-09-29 22:06:07 -0400950 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -0400951 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +1100952 return;
953 }
Al Viro863d6842011-11-25 00:57:42 -0500954 if (unlikely(mnt->mnt_pinned)) {
955 mnt_add_count(mnt, mnt->mnt_pinned + 1);
956 mnt->mnt_pinned = 0;
Al Viro48a066e2013-09-29 22:06:07 -0400957 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -0400958 unlock_mount_hash();
Al Viro900148d2011-11-25 00:33:11 -0500959 acct_auto_close_mnt(&mnt->mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100960 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500961 }
Al Viro48a066e2013-09-29 22:06:07 -0400962 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
963 rcu_read_unlock();
964 unlock_mount_hash();
965 return;
966 }
967 mnt->mnt.mnt_flags |= MNT_DOOMED;
968 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +0930969
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100970 list_del(&mnt->mnt_instance);
Al Viro719ea2f2013-09-29 11:24:49 -0400971 unlock_mount_hash();
Al Viro649a7952013-09-28 12:41:25 -0400972
973 /*
974 * This probably indicates that somebody messed
975 * up a mnt_want/drop_write() pair. If this
976 * happens, the filesystem was probably unable
977 * to make r/w->r/o transitions.
978 */
979 /*
980 * The locking used to deal with mnt_count decrement provides barriers,
981 * so mnt_get_writers() below is safe.
982 */
983 WARN_ON(mnt_get_writers(mnt));
984 fsnotify_vfsmount_delete(&mnt->mnt);
985 dput(mnt->mnt.mnt_root);
986 deactivate_super(mnt->mnt.mnt_sb);
Al Viro48a066e2013-09-29 22:06:07 -0400987 mnt_free_id(mnt);
David Howells8ffcb322014-01-24 12:17:54 +0000988 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500989}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100990
991void mntput(struct vfsmount *mnt)
992{
993 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -0500994 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100995 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -0500996 if (unlikely(m->mnt_expiry_mark))
997 m->mnt_expiry_mark = 0;
998 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100999 }
1000}
1001EXPORT_SYMBOL(mntput);
1002
1003struct vfsmount *mntget(struct vfsmount *mnt)
1004{
1005 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001006 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001007 return mnt;
1008}
1009EXPORT_SYMBOL(mntget);
1010
Al Viro7b7b1ac2005-11-07 17:13:39 -05001011void mnt_pin(struct vfsmount *mnt)
1012{
Al Viro719ea2f2013-09-29 11:24:49 -04001013 lock_mount_hash();
Al Viro863d6842011-11-25 00:57:42 -05001014 real_mount(mnt)->mnt_pinned++;
Al Viro719ea2f2013-09-29 11:24:49 -04001015 unlock_mount_hash();
Al Viro7b7b1ac2005-11-07 17:13:39 -05001016}
Al Viro7b7b1ac2005-11-07 17:13:39 -05001017EXPORT_SYMBOL(mnt_pin);
1018
Al Viro863d6842011-11-25 00:57:42 -05001019void mnt_unpin(struct vfsmount *m)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001020{
Al Viro863d6842011-11-25 00:57:42 -05001021 struct mount *mnt = real_mount(m);
Al Viro719ea2f2013-09-29 11:24:49 -04001022 lock_mount_hash();
Al Viro7b7b1ac2005-11-07 17:13:39 -05001023 if (mnt->mnt_pinned) {
Al Viro863d6842011-11-25 00:57:42 -05001024 mnt_add_count(mnt, 1);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001025 mnt->mnt_pinned--;
1026 }
Al Viro719ea2f2013-09-29 11:24:49 -04001027 unlock_mount_hash();
Al Viro7b7b1ac2005-11-07 17:13:39 -05001028}
Al Viro7b7b1ac2005-11-07 17:13:39 -05001029EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001031static inline void mangle(struct seq_file *m, const char *s)
1032{
1033 seq_escape(m, s, " \t\n\\");
1034}
1035
1036/*
1037 * Simple .show_options callback for filesystems which don't want to
1038 * implement more complex mount option showing.
1039 *
1040 * See also save_mount_options().
1041 */
Al Viro34c80b12011-12-08 21:32:45 -05001042int generic_show_options(struct seq_file *m, struct dentry *root)
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001043{
Al Viro2a32ceb2009-05-08 16:05:57 -04001044 const char *options;
1045
1046 rcu_read_lock();
Al Viro34c80b12011-12-08 21:32:45 -05001047 options = rcu_dereference(root->d_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001048
1049 if (options != NULL && options[0]) {
1050 seq_putc(m, ',');
1051 mangle(m, options);
1052 }
Al Viro2a32ceb2009-05-08 16:05:57 -04001053 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001054
1055 return 0;
1056}
1057EXPORT_SYMBOL(generic_show_options);
1058
1059/*
1060 * If filesystem uses generic_show_options(), this function should be
1061 * called from the fill_super() callback.
1062 *
1063 * The .remount_fs callback usually needs to be handled in a special
1064 * way, to make sure, that previous options are not overwritten if the
1065 * remount fails.
1066 *
1067 * Also note, that if the filesystem's .remount_fs function doesn't
1068 * reset all options to their default value, but changes only newly
1069 * given options, then the displayed options will not reflect reality
1070 * any more.
1071 */
1072void save_mount_options(struct super_block *sb, char *options)
1073{
Al Viro2a32ceb2009-05-08 16:05:57 -04001074 BUG_ON(sb->s_options);
1075 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -08001076}
1077EXPORT_SYMBOL(save_mount_options);
1078
Al Viro2a32ceb2009-05-08 16:05:57 -04001079void replace_mount_options(struct super_block *sb, char *options)
1080{
1081 char *old = sb->s_options;
1082 rcu_assign_pointer(sb->s_options, options);
1083 if (old) {
1084 synchronize_rcu();
1085 kfree(old);
1086 }
1087}
1088EXPORT_SYMBOL(replace_mount_options);
1089
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001090#ifdef CONFIG_PROC_FS
Al Viro0226f492011-12-06 12:21:54 -05001091/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092static void *m_start(struct seq_file *m, loff_t *pos)
1093{
Al Viro6ce6e242012-06-09 01:16:59 -04001094 struct proc_mounts *p = proc_mounts(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Ram Pai390c6842005-11-07 17:17:51 -05001096 down_read(&namespace_sem);
Al Viroc7999c32014-02-27 14:40:10 -05001097 if (p->cached_event == p->ns->event) {
1098 void *v = p->cached_mount;
1099 if (*pos == p->cached_index)
1100 return v;
1101 if (*pos == p->cached_index + 1) {
1102 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1103 return p->cached_mount = v;
1104 }
1105 }
1106
1107 p->cached_event = p->ns->event;
1108 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1109 p->cached_index = *pos;
1110 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111}
1112
1113static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1114{
Al Viro6ce6e242012-06-09 01:16:59 -04001115 struct proc_mounts *p = proc_mounts(m);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001116
Al Viroc7999c32014-02-27 14:40:10 -05001117 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1118 p->cached_index = *pos;
1119 return p->cached_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120}
1121
1122static void m_stop(struct seq_file *m, void *v)
1123{
Ram Pai390c6842005-11-07 17:17:51 -05001124 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125}
1126
Al Viro0226f492011-12-06 12:21:54 -05001127static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001128{
Al Viro6ce6e242012-06-09 01:16:59 -04001129 struct proc_mounts *p = proc_mounts(m);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001130 struct mount *r = list_entry(v, struct mount, mnt_list);
Al Viro0226f492011-12-06 12:21:54 -05001131 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132}
1133
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001134const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 .start = m_start,
1136 .next = m_next,
1137 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001138 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001139};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001140#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001141
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142/**
1143 * may_umount_tree - check if a mount tree is busy
1144 * @mnt: root of mount tree
1145 *
1146 * This is called to check if a tree of mounts has any
1147 * open files, pwds, chroots or sub mounts that are
1148 * busy.
1149 */
Al Viro909b0a82011-11-25 03:06:56 -05001150int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
Al Viro909b0a82011-11-25 03:06:56 -05001152 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001153 int actual_refs = 0;
1154 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001155 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001156 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
Nick Pigginb3e19d92011-01-07 17:50:11 +11001158 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001159 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001160 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001161 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 }
Al Viro719ea2f2013-09-29 11:24:49 -04001164 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
1166 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001167 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
Ian Kente3474a82006-03-27 01:14:51 -08001169 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170}
1171
1172EXPORT_SYMBOL(may_umount_tree);
1173
1174/**
1175 * may_umount - check if a mount point is busy
1176 * @mnt: root of mount
1177 *
1178 * This is called to check if a mount point has any
1179 * open files, pwds, chroots or sub mounts. If the
1180 * mount has sub mounts this will return busy
1181 * regardless of whether the sub mounts are busy.
1182 *
1183 * Doesn't take quota and stuff into account. IOW, in some cases it will
1184 * give false negatives. The main reason why it's here is that we need
1185 * a non-destructive way to look for easily umountable filesystems.
1186 */
1187int may_umount(struct vfsmount *mnt)
1188{
Ian Kente3474a82006-03-27 01:14:51 -08001189 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001190 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001191 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001192 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001193 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001194 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001195 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001196 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197}
1198
1199EXPORT_SYMBOL(may_umount);
1200
Al Viro38129a12014-03-20 21:10:51 -04001201static HLIST_HEAD(unmounted); /* protected by namespace_sem */
Al Viroe3197d82013-03-16 14:35:16 -04001202
Al Viro97216be2013-03-16 15:12:40 -04001203static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204{
Al Virod5e50f72011-11-24 20:58:57 -05001205 struct mount *mnt;
Al Viro38129a12014-03-20 21:10:51 -04001206 struct hlist_head head = unmounted;
Al Viro97216be2013-03-16 15:12:40 -04001207
Al Viro38129a12014-03-20 21:10:51 -04001208 if (likely(hlist_empty(&head))) {
Al Viro97216be2013-03-16 15:12:40 -04001209 up_write(&namespace_sem);
1210 return;
1211 }
1212
Al Viro38129a12014-03-20 21:10:51 -04001213 head.first->pprev = &head.first;
1214 INIT_HLIST_HEAD(&unmounted);
1215
Al Viro97216be2013-03-16 15:12:40 -04001216 up_write(&namespace_sem);
1217
Al Viro48a066e2013-09-29 22:06:07 -04001218 synchronize_rcu();
1219
Al Viro38129a12014-03-20 21:10:51 -04001220 while (!hlist_empty(&head)) {
1221 mnt = hlist_entry(head.first, struct mount, mnt_hash);
1222 hlist_del_init(&mnt->mnt_hash);
Al Viroaba809cf2013-09-28 23:10:55 -04001223 if (mnt->mnt_ex_mountpoint.mnt)
1224 path_put(&mnt->mnt_ex_mountpoint);
Al Virod5e50f72011-11-24 20:58:57 -05001225 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001226 }
1227}
1228
Al Viro97216be2013-03-16 15:12:40 -04001229static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001230{
Al Viro97216be2013-03-16 15:12:40 -04001231 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001232}
1233
Nick Piggin99b7db72010-08-18 04:37:39 +10001234/*
Al Viro48a066e2013-09-29 22:06:07 -04001235 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001236 * namespace_sem must be held for write
Al Viro48a066e2013-09-29 22:06:07 -04001237 * how = 0 => just this tree, don't propagate
1238 * how = 1 => propagate; we know that nobody else has reference to any victims
1239 * how = 2 => lazy umount
Nick Piggin99b7db72010-08-18 04:37:39 +10001240 */
Al Viro48a066e2013-09-29 22:06:07 -04001241void umount_tree(struct mount *mnt, int how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001242{
Al Viro38129a12014-03-20 21:10:51 -04001243 HLIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001244 struct mount *p;
Al Viro38129a12014-03-20 21:10:51 -04001245 struct mount *last = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001246
Al Viro38129a12014-03-20 21:10:51 -04001247 for (p = mnt; p; p = next_mnt(p, mnt)) {
1248 hlist_del_init_rcu(&p->mnt_hash);
1249 hlist_add_head(&p->mnt_hash, &tmp_list);
1250 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001251
Al Viro48a066e2013-09-29 22:06:07 -04001252 if (how)
Al Viro7b8a53f2011-01-15 20:08:44 -05001253 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001254
Al Viro38129a12014-03-20 21:10:51 -04001255 hlist_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001256 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001257 list_del_init(&p->mnt_list);
Al Viro143c8c92011-11-25 00:46:35 -05001258 __touch_mnt_namespace(p->mnt_ns);
1259 p->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04001260 if (how < 2)
1261 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro6b41d532011-11-24 23:24:33 -05001262 list_del_init(&p->mnt_child);
Al Viro676da582011-11-24 21:47:05 -05001263 if (mnt_has_parent(p)) {
Al Viro84d17192013-03-15 10:53:28 -04001264 put_mountpoint(p->mnt_mp);
Al Viroaba809cf2013-09-28 23:10:55 -04001265 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1266 p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint;
1267 p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt;
1268 p->mnt_mountpoint = p->mnt.mnt_root;
1269 p->mnt_parent = p;
Al Viro84d17192013-03-15 10:53:28 -04001270 p->mnt_mp = NULL;
Al Viro7c4b93d2008-03-21 23:59:49 -04001271 }
Al Viro0f0afb12011-11-24 20:43:10 -05001272 change_mnt_propagation(p, MS_PRIVATE);
Al Viro38129a12014-03-20 21:10:51 -04001273 last = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 }
Al Viro38129a12014-03-20 21:10:51 -04001275 if (last) {
1276 last->mnt_hash.next = unmounted.first;
1277 unmounted.first = tmp_list.first;
1278 unmounted.first->pprev = &unmounted.first;
1279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280}
1281
Al Virob54b9be2013-03-16 14:39:34 -04001282static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001283
Al Viro1ab59732011-11-24 21:35:16 -05001284static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285{
Al Viro1ab59732011-11-24 21:35:16 -05001286 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 int retval;
1288
Al Viro1ab59732011-11-24 21:35:16 -05001289 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 if (retval)
1291 return retval;
1292
1293 /*
1294 * Allow userspace to request a mountpoint be expired rather than
1295 * unmounting unconditionally. Unmount only happens if:
1296 * (1) the mark is already set (the mark is cleared by mntput())
1297 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1298 */
1299 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001300 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 flags & (MNT_FORCE | MNT_DETACH))
1302 return -EINVAL;
1303
Nick Pigginb3e19d92011-01-07 17:50:11 +11001304 /*
1305 * probably don't strictly need the lock here if we examined
1306 * all race cases, but it's a slowpath.
1307 */
Al Viro719ea2f2013-09-29 11:24:49 -04001308 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001309 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001310 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001312 }
Al Viro719ea2f2013-09-29 11:24:49 -04001313 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Al Viro863d6842011-11-25 00:57:42 -05001315 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 return -EAGAIN;
1317 }
1318
1319 /*
1320 * If we may have to abort operations to get out of this
1321 * mount, and they will themselves hold resources we must
1322 * allow the fs to do things. In the Unix tradition of
1323 * 'Gee thats tricky lets do it in userspace' the umount_begin
1324 * might fail to complete on the first run through as other tasks
1325 * must return, and the like. Thats for the mount program to worry
1326 * about for the moment.
1327 */
1328
Al Viro42faad92008-04-24 07:21:56 -04001329 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001330 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001331 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 /*
1334 * No sense to grab the lock for this test, but test itself looks
1335 * somewhat bogus. Suggestions for better replacement?
1336 * Ho-hum... In principle, we might treat that as umount + switch
1337 * to rootfs. GC would eventually take care of the old vfsmount.
1338 * Actually it makes sense, especially if rootfs would contain a
1339 * /reboot - static binary that would close all descriptors and
1340 * call reboot(9). Then init(8) could umount root and exec /reboot.
1341 */
Al Viro1ab59732011-11-24 21:35:16 -05001342 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 /*
1344 * Special case for "unmounting" root ...
1345 * we just try to remount it readonly.
1346 */
1347 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001348 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 up_write(&sb->s_umount);
1351 return retval;
1352 }
1353
Al Viro97216be2013-03-16 15:12:40 -04001354 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001355 lock_mount_hash();
Al Viro5addc5d2005-11-07 17:15:49 -05001356 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
Al Viro48a066e2013-09-29 22:06:07 -04001358 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001359 if (!list_empty(&mnt->mnt_list))
Al Viro48a066e2013-09-29 22:06:07 -04001360 umount_tree(mnt, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001362 } else {
1363 shrink_submounts(mnt);
1364 retval = -EBUSY;
1365 if (!propagate_mount_busy(mnt, 2)) {
1366 if (!list_empty(&mnt->mnt_list))
1367 umount_tree(mnt, 1);
1368 retval = 0;
1369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 }
Al Viro719ea2f2013-09-29 11:24:49 -04001371 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001372 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 return retval;
1374}
1375
Al Viro9b40bc92013-02-22 22:45:42 -05001376/*
1377 * Is the caller allowed to modify his namespace?
1378 */
1379static inline bool may_mount(void)
1380{
1381 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1382}
1383
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384/*
1385 * Now umount can handle mount points as well as block devices.
1386 * This is important for filesystems which use unnamed block devices.
1387 *
1388 * We now support a flag for forced unmount like the other 'big iron'
1389 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1390 */
1391
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001392SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
Al Viro2d8f3032008-07-22 09:59:21 -04001394 struct path path;
Al Viro900148d2011-11-25 00:33:11 -05001395 struct mount *mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001397 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001399 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1400 return -EINVAL;
1401
Al Viro9b40bc92013-02-22 22:45:42 -05001402 if (!may_mount())
1403 return -EPERM;
1404
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001405 if (!(flags & UMOUNT_NOFOLLOW))
1406 lookup_flags |= LOOKUP_FOLLOW;
1407
Al Viro197df042013-09-08 14:03:27 -04001408 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 if (retval)
1410 goto out;
Al Viro900148d2011-11-25 00:33:11 -05001411 mnt = real_mount(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001413 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 goto dput_and_out;
Al Viro143c8c92011-11-25 00:46:35 -05001415 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 goto dput_and_out;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001417 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1418 goto dput_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
Al Viro900148d2011-11-25 00:33:11 -05001420 retval = do_umount(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001422 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001423 dput(path.dentry);
Al Viro900148d2011-11-25 00:33:11 -05001424 mntput_no_expire(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425out:
1426 return retval;
1427}
1428
1429#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1430
1431/*
Ram Paib58fed82005-11-07 17:16:09 -05001432 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001434SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435{
Ram Paib58fed82005-11-07 17:16:09 -05001436 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437}
1438
1439#endif
1440
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001441static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001442{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001443 /* Is this a proxy for a mount namespace? */
1444 struct inode *inode = dentry->d_inode;
David Howells0bb80f22013-04-12 01:50:06 +01001445 struct proc_ns *ei;
Eric W. Biederman8823c072010-03-07 18:49:36 -08001446
1447 if (!proc_ns_inode(inode))
1448 return false;
1449
David Howells0bb80f22013-04-12 01:50:06 +01001450 ei = get_proc_ns(inode);
Eric W. Biederman8823c072010-03-07 18:49:36 -08001451 if (ei->ns_ops != &mntns_operations)
1452 return false;
1453
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001454 return true;
1455}
1456
1457static bool mnt_ns_loop(struct dentry *dentry)
1458{
1459 /* Could bind mounting the mount namespace inode cause a
1460 * mount namespace loop?
1461 */
1462 struct mnt_namespace *mnt_ns;
1463 if (!is_mnt_ns_file(dentry))
1464 return false;
1465
1466 mnt_ns = get_proc_ns(dentry->d_inode)->ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08001467 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1468}
1469
Al Viro87129cc2011-11-24 21:24:27 -05001470struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001471 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
Al Viro84d17192013-03-15 10:53:28 -04001473 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001475 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1476 return ERR_PTR(-EINVAL);
1477
1478 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001479 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001480
Ram Pai36341f62005-11-07 17:17:22 -05001481 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001482 if (IS_ERR(q))
1483 return q;
1484
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001485 q->mnt.mnt_flags &= ~MNT_LOCKED;
Al Viroa73324d2011-11-24 22:25:07 -05001486 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
1488 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001489 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001490 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001491 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 continue;
1493
Al Viro909b0a82011-11-25 03:06:56 -05001494 for (s = r; s; s = next_mnt(s, r)) {
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001495 if (!(flag & CL_COPY_UNBINDABLE) &&
1496 IS_MNT_UNBINDABLE(s)) {
1497 s = skip_mnt_tree(s);
1498 continue;
1499 }
1500 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1501 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001502 s = skip_mnt_tree(s);
1503 continue;
1504 }
Al Viro0714a532011-11-24 22:19:58 -05001505 while (p != s->mnt_parent) {
1506 p = p->mnt_parent;
1507 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 }
Al Viro87129cc2011-11-24 21:24:27 -05001509 p = s;
Al Viro84d17192013-03-15 10:53:28 -04001510 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001511 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001512 if (IS_ERR(q))
1513 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04001514 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05001515 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro84d17192013-03-15 10:53:28 -04001516 attach_mnt(q, parent, p->mnt_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04001517 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 }
1519 }
1520 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001521out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04001523 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001524 umount_tree(res, 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001525 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001527 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528}
1529
David Howellsbe34d1a2012-06-25 12:55:18 +01001530/* Caller should check returned pointer for errors */
1531
Al Viro589ff872009-04-18 03:28:19 -04001532struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001533{
Al Virocb338d02011-11-24 20:55:08 -05001534 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001535 namespace_lock();
Al Viro87129cc2011-11-24 21:24:27 -05001536 tree = copy_tree(real_mount(path->mnt), path->dentry,
1537 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001538 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001539 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001540 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001541 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001542}
1543
1544void drop_collected_mounts(struct vfsmount *mnt)
1545{
Al Viro97216be2013-03-16 15:12:40 -04001546 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001547 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001548 umount_tree(real_mount(mnt), 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001549 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04001550 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001551}
1552
Al Viro1f707132010-01-30 22:51:25 -05001553int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1554 struct vfsmount *root)
1555{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001556 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001557 int res = f(root, arg);
1558 if (res)
1559 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001560 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1561 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001562 if (res)
1563 return res;
1564 }
1565 return 0;
1566}
1567
Al Viro4b8b21f2011-11-24 19:54:23 -05001568static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001569{
Al Viro315fc832011-11-24 18:57:30 -05001570 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001571
Al Viro909b0a82011-11-25 03:06:56 -05001572 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001573 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001574 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001575 }
1576}
1577
Al Viro4b8b21f2011-11-24 19:54:23 -05001578static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001579{
Al Viro315fc832011-11-24 18:57:30 -05001580 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001581
Al Viro909b0a82011-11-25 03:06:56 -05001582 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05001583 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001584 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001585 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001586 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001587 return err;
1588 }
1589 }
1590 }
1591
1592 return 0;
1593}
1594
Ram Paib90fa9a2005-11-07 17:19:50 -05001595/*
1596 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001597 * @nd : place the mount tree @source_mnt is attached
1598 * @parent_nd : if non-null, detach the source_mnt from its parent and
1599 * store the parent mount and mountpoint dentry.
1600 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001601 *
1602 * NOTE: in the table below explains the semantics when a source mount
1603 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001604 * ---------------------------------------------------------------------------
1605 * | BIND MOUNT OPERATION |
1606 * |**************************************************************************
1607 * | source-->| shared | private | slave | unbindable |
1608 * | dest | | | | |
1609 * | | | | | | |
1610 * | v | | | | |
1611 * |**************************************************************************
1612 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1613 * | | | | | |
1614 * |non-shared| shared (+) | private | slave (*) | invalid |
1615 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001616 * A bind operation clones the source mount and mounts the clone on the
1617 * destination mount.
1618 *
1619 * (++) the cloned mount is propagated to all the mounts in the propagation
1620 * tree of the destination mount and the cloned mount is added to
1621 * the peer group of the source mount.
1622 * (+) the cloned mount is created under the destination mount and is marked
1623 * as shared. The cloned mount is added to the peer group of the source
1624 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001625 * (+++) the mount is propagated to all the mounts in the propagation tree
1626 * of the destination mount and the cloned mount is made slave
1627 * of the same master as that of the source mount. The cloned mount
1628 * is marked as 'shared and slave'.
1629 * (*) the cloned mount is made a slave of the same master as that of the
1630 * source mount.
1631 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001632 * ---------------------------------------------------------------------------
1633 * | MOVE MOUNT OPERATION |
1634 * |**************************************************************************
1635 * | source-->| shared | private | slave | unbindable |
1636 * | dest | | | | |
1637 * | | | | | | |
1638 * | v | | | | |
1639 * |**************************************************************************
1640 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1641 * | | | | | |
1642 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1643 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001644 *
1645 * (+) the mount is moved to the destination. And is then propagated to
1646 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001647 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001648 * (+++) the mount is moved to the destination and is then propagated to
1649 * all the mounts belonging to the destination mount's propagation tree.
1650 * the mount is marked as 'shared and slave'.
1651 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001652 *
1653 * if the source mount is a tree, the operations explained above is
1654 * applied to each mount in the tree.
1655 * Must be called without spinlocks held, since this function can sleep
1656 * in allocations.
1657 */
Al Viro0fb54e52011-11-24 19:59:16 -05001658static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d17192013-03-15 10:53:28 -04001659 struct mount *dest_mnt,
1660 struct mountpoint *dest_mp,
1661 struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001662{
Al Viro38129a12014-03-20 21:10:51 -04001663 HLIST_HEAD(tree_list);
Al Viro315fc832011-11-24 18:57:30 -05001664 struct mount *child, *p;
Al Viro38129a12014-03-20 21:10:51 -04001665 struct hlist_node *n;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001666 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001667
Al Virofc7be132011-11-25 01:05:37 -05001668 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001669 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001670 if (err)
1671 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04001672 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Al Virof2ebb3a2014-02-27 09:35:45 -05001673 lock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001674 if (err)
1675 goto out_cleanup_ids;
Al Viro909b0a82011-11-25 03:06:56 -05001676 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001677 set_mnt_shared(p);
Al Viro0b1b9012014-03-21 10:14:08 -04001678 } else {
1679 lock_mount_hash();
Ram Paib90fa9a2005-11-07 17:19:50 -05001680 }
Al Viro1a390682008-03-21 20:48:19 -04001681 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001682 detach_mnt(source_mnt, parent_path);
Al Viro84d17192013-03-15 10:53:28 -04001683 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05001684 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001685 } else {
Al Viro84d17192013-03-15 10:53:28 -04001686 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Al Viro1d6a32a2014-03-20 20:34:43 -04001687 commit_tree(source_mnt, NULL);
Ram Pai21444402005-11-07 17:20:03 -05001688 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001689
Al Viro38129a12014-03-20 21:10:51 -04001690 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04001691 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04001692 hlist_del_init(&child->mnt_hash);
Al Viro1d6a32a2014-03-20 20:34:43 -04001693 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1694 child->mnt_mountpoint);
1695 commit_tree(child, q);
Ram Paib90fa9a2005-11-07 17:19:50 -05001696 }
Al Viro719ea2f2013-09-29 11:24:49 -04001697 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10001698
Ram Paib90fa9a2005-11-07 17:19:50 -05001699 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001700
1701 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05001702 while (!hlist_empty(&tree_list)) {
1703 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1704 umount_tree(child, 0);
1705 }
1706 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04001707 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001708 out:
1709 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001710}
1711
Al Viro84d17192013-03-15 10:53:28 -04001712static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04001713{
1714 struct vfsmount *mnt;
Al Viro84d17192013-03-15 10:53:28 -04001715 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04001716retry:
Al Viro84d17192013-03-15 10:53:28 -04001717 mutex_lock(&dentry->d_inode->i_mutex);
1718 if (unlikely(cant_mount(dentry))) {
1719 mutex_unlock(&dentry->d_inode->i_mutex);
1720 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04001721 }
Al Viro97216be2013-03-16 15:12:40 -04001722 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04001723 mnt = lookup_mnt(path);
Al Viro84d17192013-03-15 10:53:28 -04001724 if (likely(!mnt)) {
1725 struct mountpoint *mp = new_mountpoint(dentry);
1726 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04001727 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001728 mutex_unlock(&dentry->d_inode->i_mutex);
1729 return mp;
1730 }
1731 return mp;
1732 }
Al Viro97216be2013-03-16 15:12:40 -04001733 namespace_unlock();
Al Virob12cea92011-03-18 08:55:38 -04001734 mutex_unlock(&path->dentry->d_inode->i_mutex);
1735 path_put(path);
1736 path->mnt = mnt;
Al Viro84d17192013-03-15 10:53:28 -04001737 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04001738 goto retry;
1739}
1740
Al Viro84d17192013-03-15 10:53:28 -04001741static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04001742{
Al Viro84d17192013-03-15 10:53:28 -04001743 struct dentry *dentry = where->m_dentry;
1744 put_mountpoint(where);
Al Viro328e6d92013-03-16 14:49:45 -04001745 namespace_unlock();
Al Viro84d17192013-03-15 10:53:28 -04001746 mutex_unlock(&dentry->d_inode->i_mutex);
Al Virob12cea92011-03-18 08:55:38 -04001747}
1748
Al Viro84d17192013-03-15 10:53:28 -04001749static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750{
Al Viro95bc5f22011-11-25 00:30:56 -05001751 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 return -EINVAL;
1753
Al Viro84d17192013-03-15 10:53:28 -04001754 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001755 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 return -ENOTDIR;
1757
Al Viro84d17192013-03-15 10:53:28 -04001758 return attach_recursive_mnt(mnt, p, mp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759}
1760
1761/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001762 * Sanity check the flags to change_mnt_propagation.
1763 */
1764
1765static int flags_to_propagation_type(int flags)
1766{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001767 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001768
1769 /* Fail if any non-propagation flags are set */
1770 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1771 return 0;
1772 /* Only one propagation flag should be set */
1773 if (!is_power_of_2(type))
1774 return 0;
1775 return type;
1776}
1777
1778/*
Ram Pai07b20882005-11-07 17:19:07 -05001779 * recursively change the type of the mountpoint.
1780 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001781static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001782{
Al Viro315fc832011-11-24 18:57:30 -05001783 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001784 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001785 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001786 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001787 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001788
Al Viro2d92ab32008-08-02 00:51:11 -04001789 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001790 return -EINVAL;
1791
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001792 type = flags_to_propagation_type(flag);
1793 if (!type)
1794 return -EINVAL;
1795
Al Viro97216be2013-03-16 15:12:40 -04001796 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001797 if (type == MS_SHARED) {
1798 err = invent_group_ids(mnt, recurse);
1799 if (err)
1800 goto out_unlock;
1801 }
1802
Al Viro719ea2f2013-09-29 11:24:49 -04001803 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001804 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001805 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04001806 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001807
1808 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04001809 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001810 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001811}
1812
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001813static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1814{
1815 struct mount *child;
1816 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1817 if (!is_subdir(child->mnt_mountpoint, dentry))
1818 continue;
1819
1820 if (child->mnt.mnt_flags & MNT_LOCKED)
1821 return true;
1822 }
1823 return false;
1824}
1825
Ram Pai07b20882005-11-07 17:19:07 -05001826/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 * do loopback mount.
1828 */
Al Viro808d4e32012-10-11 11:42:01 -04001829static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001830 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831{
Al Viro2d92ab32008-08-02 00:51:11 -04001832 struct path old_path;
Al Viro84d17192013-03-15 10:53:28 -04001833 struct mount *mnt = NULL, *old, *parent;
1834 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05001835 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 if (!old_name || !*old_name)
1837 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001838 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 if (err)
1840 return err;
1841
Eric W. Biederman8823c072010-03-07 18:49:36 -08001842 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001843 if (mnt_ns_loop(old_path.dentry))
Eric W. Biederman8823c072010-03-07 18:49:36 -08001844 goto out;
1845
Al Viro84d17192013-03-15 10:53:28 -04001846 mp = lock_mount(path);
1847 err = PTR_ERR(mp);
1848 if (IS_ERR(mp))
Jan Blunck4ac91372008-02-14 19:34:32 -08001849 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001850
Al Viro87129cc2011-11-24 21:24:27 -05001851 old = real_mount(old_path.mnt);
Al Viro84d17192013-03-15 10:53:28 -04001852 parent = real_mount(path->mnt);
Al Viro87129cc2011-11-24 21:24:27 -05001853
Al Virob12cea92011-03-18 08:55:38 -04001854 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001855 if (IS_MNT_UNBINDABLE(old))
Al Virob12cea92011-03-18 08:55:38 -04001856 goto out2;
1857
Al Viro84d17192013-03-15 10:53:28 -04001858 if (!check_mnt(parent) || !check_mnt(old))
Al Virob12cea92011-03-18 08:55:38 -04001859 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001861 if (!recurse && has_locked_children(old, old_path.dentry))
1862 goto out2;
1863
Al Viroccd48bc2005-11-07 17:15:04 -05001864 if (recurse)
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001865 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
Al Viroccd48bc2005-11-07 17:15:04 -05001866 else
Al Viro87129cc2011-11-24 21:24:27 -05001867 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001868
David Howellsbe34d1a2012-06-25 12:55:18 +01001869 if (IS_ERR(mnt)) {
1870 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04001871 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01001872 }
Al Viroccd48bc2005-11-07 17:15:04 -05001873
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001874 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
1875
Al Viro84d17192013-03-15 10:53:28 -04001876 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05001877 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04001878 lock_mount_hash();
Al Viro328e6d92013-03-16 14:49:45 -04001879 umount_tree(mnt, 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001880 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001881 }
Al Virob12cea92011-03-18 08:55:38 -04001882out2:
Al Viro84d17192013-03-15 10:53:28 -04001883 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05001884out:
Al Viro2d92ab32008-08-02 00:51:11 -04001885 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 return err;
1887}
1888
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001889static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1890{
1891 int error = 0;
1892 int readonly_request = 0;
1893
1894 if (ms_flags & MS_RDONLY)
1895 readonly_request = 1;
1896 if (readonly_request == __mnt_is_readonly(mnt))
1897 return 0;
1898
Eric W. Biederman90563b12013-03-22 03:10:15 -07001899 if (mnt->mnt_flags & MNT_LOCK_READONLY)
1900 return -EPERM;
1901
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001902 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05001903 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001904 else
Al Viro83adc752011-11-24 22:37:54 -05001905 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001906 return error;
1907}
1908
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909/*
1910 * change filesystem flags. dir should be a physical root of filesystem.
1911 * If you've mounted a non-root directory somewhere and want to do remount
1912 * on it - tough luck.
1913 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001914static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 void *data)
1916{
1917 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001918 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05001919 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
Al Viro143c8c92011-11-25 00:46:35 -05001921 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 return -EINVAL;
1923
Al Viro2d92ab32008-08-02 00:51:11 -04001924 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 return -EINVAL;
1926
Eric Parisff36fe22011-03-03 16:09:14 -05001927 err = security_sb_remount(sb, data);
1928 if (err)
1929 return err;
1930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001932 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001933 err = change_mount_flags(path->mnt, flags);
Al Viro57eccb82013-02-22 22:49:10 -05001934 else if (!capable(CAP_SYS_ADMIN))
1935 err = -EPERM;
Al Viro4aa98cf2009-05-08 13:36:58 -04001936 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001937 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001938 if (!err) {
Al Viro719ea2f2013-09-29 11:24:49 -04001939 lock_mount_hash();
Al Viro143c8c92011-11-25 00:46:35 -05001940 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1941 mnt->mnt.mnt_flags = mnt_flags;
Al Viro143c8c92011-11-25 00:46:35 -05001942 touch_mnt_namespace(mnt->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04001943 unlock_mount_hash();
Dan Williams0e55a7c2008-09-26 19:01:20 -07001944 }
Al Viro6339dab2013-09-16 22:41:01 -04001945 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 return err;
1947}
1948
Al Virocbbe3622011-11-24 20:01:19 -05001949static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05001950{
Al Viro315fc832011-11-24 18:57:30 -05001951 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001952 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001953 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05001954 return 1;
1955 }
1956 return 0;
1957}
1958
Al Viro808d4e32012-10-11 11:42:01 -04001959static int do_move_mount(struct path *path, const char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960{
Al Viro2d92ab32008-08-02 00:51:11 -04001961 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05001962 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05001963 struct mount *old;
Al Viro84d17192013-03-15 10:53:28 -04001964 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05001965 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 if (!old_name || !*old_name)
1967 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001968 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 if (err)
1970 return err;
1971
Al Viro84d17192013-03-15 10:53:28 -04001972 mp = lock_mount(path);
1973 err = PTR_ERR(mp);
1974 if (IS_ERR(mp))
David Howellscc53ce52011-01-14 18:45:26 +00001975 goto out;
1976
Al Viro143c8c92011-11-25 00:46:35 -05001977 old = real_mount(old_path.mnt);
Al Virofc7be132011-11-25 01:05:37 -05001978 p = real_mount(path->mnt);
Al Viro143c8c92011-11-25 00:46:35 -05001979
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 err = -EINVAL;
Al Virofc7be132011-11-25 01:05:37 -05001981 if (!check_mnt(p) || !check_mnt(old))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 goto out1;
1983
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001984 if (old->mnt.mnt_flags & MNT_LOCKED)
1985 goto out1;
1986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001988 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001989 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
Al Viro676da582011-11-24 21:47:05 -05001991 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05001992 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993
Al Viro2d92ab32008-08-02 00:51:11 -04001994 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1995 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001996 goto out1;
1997 /*
1998 * Don't move a mount residing in a shared parent.
1999 */
Al Virofc7be132011-11-25 01:05:37 -05002000 if (IS_MNT_SHARED(old->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002001 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05002002 /*
2003 * Don't move a mount tree containing unbindable mounts to a destination
2004 * mount which is shared.
2005 */
Al Virofc7be132011-11-25 01:05:37 -05002006 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05002007 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 err = -ELOOP;
Al Virofc7be132011-11-25 01:05:37 -05002009 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05002010 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05002011 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
Al Viro84d17192013-03-15 10:53:28 -04002013 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08002014 if (err)
Ram Pai21444402005-11-07 17:20:03 -05002015 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
2017 /* if the mount is moved, it should no longer be expire
2018 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05002019 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020out1:
Al Viro84d17192013-03-15 10:53:28 -04002021 unlock_mount(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04002024 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04002025 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 return err;
2027}
2028
Al Viro9d412a42011-03-17 22:08:28 -04002029static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2030{
2031 int err;
2032 const char *subtype = strchr(fstype, '.');
2033 if (subtype) {
2034 subtype++;
2035 err = -EINVAL;
2036 if (!subtype[0])
2037 goto err;
2038 } else
2039 subtype = "";
2040
2041 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2042 err = -ENOMEM;
2043 if (!mnt->mnt_sb->s_subtype)
2044 goto err;
2045 return mnt;
2046
2047 err:
2048 mntput(mnt);
2049 return ERR_PTR(err);
2050}
2051
Al Viro9d412a42011-03-17 22:08:28 -04002052/*
2053 * add a mount into a namespace's mount tree
2054 */
Al Viro95bc5f22011-11-25 00:30:56 -05002055static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04002056{
Al Viro84d17192013-03-15 10:53:28 -04002057 struct mountpoint *mp;
2058 struct mount *parent;
Al Viro9d412a42011-03-17 22:08:28 -04002059 int err;
2060
Al Virof2ebb3a2014-02-27 09:35:45 -05002061 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04002062
Al Viro84d17192013-03-15 10:53:28 -04002063 mp = lock_mount(path);
2064 if (IS_ERR(mp))
2065 return PTR_ERR(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002066
Al Viro84d17192013-03-15 10:53:28 -04002067 parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04002068 err = -EINVAL;
Al Viro84d17192013-03-15 10:53:28 -04002069 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04002070 /* that's acceptable only for automounts done in private ns */
2071 if (!(mnt_flags & MNT_SHRINKABLE))
2072 goto unlock;
2073 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d17192013-03-15 10:53:28 -04002074 if (!parent->mnt_ns)
Al Viro156cacb2012-09-21 08:19:02 -04002075 goto unlock;
2076 }
Al Viro9d412a42011-03-17 22:08:28 -04002077
2078 /* Refuse the same filesystem on the same mount point */
2079 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05002080 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04002081 path->mnt->mnt_root == path->dentry)
2082 goto unlock;
2083
2084 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05002085 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04002086 goto unlock;
2087
Al Viro95bc5f22011-11-25 00:30:56 -05002088 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro84d17192013-03-15 10:53:28 -04002089 err = graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04002090
2091unlock:
Al Viro84d17192013-03-15 10:53:28 -04002092 unlock_mount(mp);
Al Viro9d412a42011-03-17 22:08:28 -04002093 return err;
2094}
Al Virob1e75df2011-01-17 01:47:59 -05002095
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096/*
2097 * create a new mount for userspace and request it to be added into the
2098 * namespace's tree
2099 */
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002100static int do_new_mount(struct path *path, const char *fstype, int flags,
Al Viro808d4e32012-10-11 11:42:01 -04002101 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002103 struct file_system_type *type;
Al Viro9b40bc92013-02-22 22:45:42 -05002104 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002106 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002108 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 return -EINVAL;
2110
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002111 type = get_fs_type(fstype);
2112 if (!type)
2113 return -ENODEV;
2114
2115 if (user_ns != &init_user_ns) {
2116 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2117 put_filesystem(type);
2118 return -EPERM;
2119 }
2120 /* Only in special cases allow devices from mounts
2121 * created outside the initial user namespace.
2122 */
2123 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2124 flags |= MS_NODEV;
2125 mnt_flags |= MNT_NODEV;
2126 }
2127 }
2128
2129 mnt = vfs_kern_mount(type, flags, name, data);
2130 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2131 !mnt->mnt_sb->s_subtype)
2132 mnt = fs_set_subtype(mnt, fstype);
2133
2134 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 if (IS_ERR(mnt))
2136 return PTR_ERR(mnt);
2137
Al Viro95bc5f22011-11-25 00:30:56 -05002138 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002139 if (err)
2140 mntput(mnt);
2141 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142}
2143
Al Viro19a167a2011-01-17 01:35:23 -05002144int finish_automount(struct vfsmount *m, struct path *path)
2145{
Al Viro6776db3d2011-11-25 00:22:05 -05002146 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002147 int err;
2148 /* The new mount record should have at least 2 refs to prevent it being
2149 * expired before we get a chance to add it
2150 */
Al Viro6776db3d2011-11-25 00:22:05 -05002151 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002152
2153 if (m->mnt_sb == path->mnt->mnt_sb &&
2154 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002155 err = -ELOOP;
2156 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002157 }
2158
Al Viro95bc5f22011-11-25 00:30:56 -05002159 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002160 if (!err)
2161 return 0;
2162fail:
2163 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002164 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002165 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05002166 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04002167 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002168 }
Al Virob1e75df2011-01-17 01:47:59 -05002169 mntput(m);
2170 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002171 return err;
2172}
2173
David Howellsea5b7782011-01-14 19:10:03 +00002174/**
2175 * mnt_set_expiry - Put a mount on an expiration list
2176 * @mnt: The mount to list.
2177 * @expiry_list: The list to add the mount to.
2178 */
2179void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2180{
Al Viro97216be2013-03-16 15:12:40 -04002181 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00002182
Al Viro6776db3d2011-11-25 00:22:05 -05002183 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002184
Al Viro97216be2013-03-16 15:12:40 -04002185 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002186}
2187EXPORT_SYMBOL(mnt_set_expiry);
2188
2189/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 * process a list of expirable mountpoints with the intent of discarding any
2191 * mountpoints that aren't in use and haven't been touched since last we came
2192 * here
2193 */
2194void mark_mounts_for_expiry(struct list_head *mounts)
2195{
Al Viro761d5c32011-11-24 21:07:43 -05002196 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 LIST_HEAD(graveyard);
2198
2199 if (list_empty(mounts))
2200 return;
2201
Al Viro97216be2013-03-16 15:12:40 -04002202 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002203 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
2205 /* extract from the expiration list every vfsmount that matches the
2206 * following criteria:
2207 * - only referenced by its parent vfsmount
2208 * - still marked for expiry (marked on the last call here; marks are
2209 * cleared by mntput())
2210 */
Al Viro6776db3d2011-11-25 00:22:05 -05002211 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002212 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002213 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002215 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002217 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002218 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002219 touch_mnt_namespace(mnt->mnt_ns);
Al Viro328e6d92013-03-16 14:49:45 -04002220 umount_tree(mnt, 1);
Al Virobcc5c7d2008-03-22 00:21:53 -04002221 }
Al Viro719ea2f2013-09-29 11:24:49 -04002222 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002223 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224}
2225
2226EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2227
2228/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002229 * Ripoff of 'select_parent()'
2230 *
2231 * search the list of submounts for a given mountpoint, and move any
2232 * shrinkable submounts to the 'graveyard' list.
2233 */
Al Viro692afc32011-11-24 21:15:14 -05002234static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002235{
Al Viro692afc32011-11-24 21:15:14 -05002236 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002237 struct list_head *next;
2238 int found = 0;
2239
2240repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002241 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002242resume:
Al Viro6b41d532011-11-24 23:24:33 -05002243 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002244 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002245 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002246
2247 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002248 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002249 continue;
2250 /*
2251 * Descend a level if the d_mounts list is non-empty.
2252 */
Al Viro6b41d532011-11-24 23:24:33 -05002253 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002254 this_parent = mnt;
2255 goto repeat;
2256 }
2257
Al Viro1ab59732011-11-24 21:35:16 -05002258 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002259 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002260 found++;
2261 }
2262 }
2263 /*
2264 * All done at this level ... ascend and resume the search
2265 */
2266 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002267 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002268 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002269 goto resume;
2270 }
2271 return found;
2272}
2273
2274/*
2275 * process a list of expirable mountpoints with the intent of discarding any
2276 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002277 *
Al Viro48a066e2013-09-29 22:06:07 -04002278 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002279 */
Al Virob54b9be2013-03-16 14:39:34 -04002280static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002281{
2282 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002283 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002284
Trond Myklebust5528f9112006-06-09 09:34:17 -04002285 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002286 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002287 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002288 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002289 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05002290 touch_mnt_namespace(m->mnt_ns);
Al Viro328e6d92013-03-16 14:49:45 -04002291 umount_tree(m, 1);
Al Virobcc5c7d2008-03-22 00:21:53 -04002292 }
2293 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002294}
2295
Trond Myklebust5528f9112006-06-09 09:34:17 -04002296/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 * Some copy_from_user() implementations do not return the exact number of
2298 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2299 * Note that this function differs from copy_from_user() in that it will oops
2300 * on bad values of `to', rather than returning a short copy.
2301 */
Ram Paib58fed82005-11-07 17:16:09 -05002302static long exact_copy_from_user(void *to, const void __user * from,
2303 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304{
2305 char *t = to;
2306 const char __user *f = from;
2307 char c;
2308
2309 if (!access_ok(VERIFY_READ, from, n))
2310 return n;
2311
2312 while (n) {
2313 if (__get_user(c, f)) {
2314 memset(t, 0, n);
2315 break;
2316 }
2317 *t++ = c;
2318 f++;
2319 n--;
2320 }
2321 return n;
2322}
2323
Ram Paib58fed82005-11-07 17:16:09 -05002324int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325{
2326 int i;
2327 unsigned long page;
2328 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002329
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 *where = 0;
2331 if (!data)
2332 return 0;
2333
2334 if (!(page = __get_free_page(GFP_KERNEL)))
2335 return -ENOMEM;
2336
2337 /* We only care that *some* data at the address the user
2338 * gave us is valid. Just in case, we'll zero
2339 * the remainder of the page.
2340 */
2341 /* copy_from_user cannot cross TASK_SIZE ! */
2342 size = TASK_SIZE - (unsigned long)data;
2343 if (size > PAGE_SIZE)
2344 size = PAGE_SIZE;
2345
2346 i = size - exact_copy_from_user((void *)page, data, size);
2347 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002348 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 return -EFAULT;
2350 }
2351 if (i != PAGE_SIZE)
2352 memset((char *)page + i, 0, PAGE_SIZE - i);
2353 *where = page;
2354 return 0;
2355}
2356
Vegard Nossumeca6f532009-09-18 13:05:45 -07002357int copy_mount_string(const void __user *data, char **where)
2358{
2359 char *tmp;
2360
2361 if (!data) {
2362 *where = NULL;
2363 return 0;
2364 }
2365
2366 tmp = strndup_user(data, PAGE_SIZE);
2367 if (IS_ERR(tmp))
2368 return PTR_ERR(tmp);
2369
2370 *where = tmp;
2371 return 0;
2372}
2373
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374/*
2375 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2376 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2377 *
2378 * data is a (void *) that can point to any structure up to
2379 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2380 * information (or be NULL).
2381 *
2382 * Pre-0.97 versions of mount() didn't have a flags word.
2383 * When the flags word was introduced its top half was required
2384 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2385 * Therefore, if this magic number is present, it carries no information
2386 * and must be discarded.
2387 */
Al Viro808d4e32012-10-11 11:42:01 -04002388long do_mount(const char *dev_name, const char *dir_name,
2389 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390{
Al Viro2d92ab32008-08-02 00:51:11 -04002391 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 int retval = 0;
2393 int mnt_flags = 0;
2394
2395 /* Discard magic */
2396 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2397 flags &= ~MS_MGC_MSK;
2398
2399 /* Basic sanity checks */
2400
2401 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2402 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403
2404 if (data_page)
2405 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2406
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002407 /* ... and get the mountpoint */
2408 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2409 if (retval)
2410 return retval;
2411
2412 retval = security_sb_mount(dev_name, &path,
2413 type_page, flags, data_page);
Al Viro0d5cadb2013-05-04 14:40:51 -04002414 if (!retval && !may_mount())
2415 retval = -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002416 if (retval)
2417 goto dput_out;
2418
Andi Kleen613cbe32009-04-19 18:40:43 +02002419 /* Default to relatime unless overriden */
2420 if (!(flags & MS_NOATIME))
2421 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002422
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 /* Separate the per-mountpoint flags */
2424 if (flags & MS_NOSUID)
2425 mnt_flags |= MNT_NOSUID;
2426 if (flags & MS_NODEV)
2427 mnt_flags |= MNT_NODEV;
2428 if (flags & MS_NOEXEC)
2429 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002430 if (flags & MS_NOATIME)
2431 mnt_flags |= MNT_NOATIME;
2432 if (flags & MS_NODIRATIME)
2433 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002434 if (flags & MS_STRICTATIME)
2435 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002436 if (flags & MS_RDONLY)
2437 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002438
Al Viro7a4dec52010-08-09 12:05:43 -04002439 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002440 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2441 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002444 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 data_page);
2446 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002447 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002448 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002449 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002451 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 else
Al Viro2d92ab32008-08-02 00:51:11 -04002453 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 dev_name, data_page);
2455dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002456 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 return retval;
2458}
2459
Eric W. Biederman771b1372012-07-26 21:08:32 -07002460static void free_mnt_ns(struct mnt_namespace *ns)
2461{
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002462 proc_free_inum(ns->proc_inum);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002463 put_user_ns(ns->user_ns);
2464 kfree(ns);
2465}
2466
Eric W. Biederman8823c072010-03-07 18:49:36 -08002467/*
2468 * Assign a sequence number so we can detect when we attempt to bind
2469 * mount a reference to an older mount namespace into the current
2470 * mount namespace, preventing reference counting loops. A 64bit
2471 * number incrementing at 10Ghz will take 12,427 years to wrap which
2472 * is effectively never, so we can ignore the possibility.
2473 */
2474static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2475
Eric W. Biederman771b1372012-07-26 21:08:32 -07002476static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002477{
2478 struct mnt_namespace *new_ns;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002479 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002480
2481 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2482 if (!new_ns)
2483 return ERR_PTR(-ENOMEM);
Eric W. Biederman98f842e2011-06-15 10:21:48 -07002484 ret = proc_alloc_inum(&new_ns->proc_inum);
2485 if (ret) {
2486 kfree(new_ns);
2487 return ERR_PTR(ret);
2488 }
Eric W. Biederman8823c072010-03-07 18:49:36 -08002489 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002490 atomic_set(&new_ns->count, 1);
2491 new_ns->root = NULL;
2492 INIT_LIST_HEAD(&new_ns->list);
2493 init_waitqueue_head(&new_ns->poll);
2494 new_ns->event = 0;
Eric W. Biederman771b1372012-07-26 21:08:32 -07002495 new_ns->user_ns = get_user_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002496 return new_ns;
2497}
2498
Al Viro9559f682013-09-28 20:47:57 -04002499struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2500 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002502 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002503 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002504 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04002505 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05002506 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002507 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508
Al Viro9559f682013-09-28 20:47:57 -04002509 BUG_ON(!ns);
2510
2511 if (likely(!(flags & CLONE_NEWNS))) {
2512 get_mnt_ns(ns);
2513 return ns;
2514 }
2515
2516 old = ns->root;
2517
Eric W. Biederman771b1372012-07-26 21:08:32 -07002518 new_ns = alloc_mnt_ns(user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002519 if (IS_ERR(new_ns))
2520 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
Al Viro97216be2013-03-16 15:12:40 -04002522 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002524 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04002525 if (user_ns != ns->user_ns)
Eric W. Biederman132c94e2013-03-22 04:08:05 -07002526 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07002527 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01002528 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04002529 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07002530 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01002531 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 }
Al Virobe08d6d2011-12-06 13:32:36 -05002533 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002534 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535
2536 /*
2537 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2538 * as belonging to new namespace. We have already acquired a private
2539 * fs_struct, so tsk->fs->lock is not needed.
2540 */
Al Viro909b0a82011-11-25 03:06:56 -05002541 p = old;
Al Virocb338d02011-11-24 20:55:08 -05002542 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05002544 q->mnt_ns = new_ns;
Al Viro9559f682013-09-28 20:47:57 -04002545 if (new_fs) {
2546 if (&p->mnt == new_fs->root.mnt) {
2547 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002548 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 }
Al Viro9559f682013-09-28 20:47:57 -04002550 if (&p->mnt == new_fs->pwd.mnt) {
2551 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05002552 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 }
Al Viro909b0a82011-11-25 03:06:56 -05002555 p = next_mnt(p, old);
2556 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002557 if (!q)
2558 break;
2559 while (p->mnt.mnt_root != q->mnt.mnt_root)
2560 p = next_mnt(p, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 }
Al Viro328e6d92013-03-16 14:49:45 -04002562 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002565 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002567 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568
JANAK DESAI741a2952006-02-07 12:59:00 -08002569 return new_ns;
2570}
2571
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002572/**
2573 * create_mnt_ns - creates a private namespace and adds a root filesystem
2574 * @mnt: pointer to the new root filesystem mountpoint
2575 */
Al Viro1a4eeaf2011-11-25 02:19:55 -05002576static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002577{
Eric W. Biederman771b1372012-07-26 21:08:32 -07002578 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002579 if (!IS_ERR(new_ns)) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002580 struct mount *mnt = real_mount(m);
2581 mnt->mnt_ns = new_ns;
Al Virobe08d6d2011-12-06 13:32:36 -05002582 new_ns->root = mnt;
Al Virob1983cd2013-05-04 15:18:53 -04002583 list_add(&mnt->mnt_list, &new_ns->list);
Al Viroc1334492011-11-16 16:12:14 -05002584 } else {
Al Viro1a4eeaf2011-11-25 02:19:55 -05002585 mntput(m);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002586 }
2587 return new_ns;
2588}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002589
Al Viroea441d12011-11-16 21:43:59 -05002590struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2591{
2592 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002593 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002594 struct path path;
2595 int err;
2596
2597 ns = create_mnt_ns(mnt);
2598 if (IS_ERR(ns))
2599 return ERR_CAST(ns);
2600
2601 err = vfs_path_lookup(mnt->mnt_root, mnt,
2602 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2603
2604 put_mnt_ns(ns);
2605
2606 if (err)
2607 return ERR_PTR(err);
2608
2609 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002610 s = path.mnt->mnt_sb;
2611 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002612 mntput(path.mnt);
2613 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002614 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002615 /* ... and return the root of (sub)tree on it */
2616 return path.dentry;
2617}
2618EXPORT_SYMBOL(mount_subtree);
2619
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002620SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2621 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002623 int ret;
2624 char *kernel_type;
Jeff Layton91a27b22012-10-10 15:25:28 -04002625 struct filename *kernel_dir;
Vegard Nossumeca6f532009-09-18 13:05:45 -07002626 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628
Vegard Nossumeca6f532009-09-18 13:05:45 -07002629 ret = copy_mount_string(type, &kernel_type);
2630 if (ret < 0)
2631 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632
Vegard Nossumeca6f532009-09-18 13:05:45 -07002633 kernel_dir = getname(dir_name);
2634 if (IS_ERR(kernel_dir)) {
2635 ret = PTR_ERR(kernel_dir);
2636 goto out_dir;
2637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
Vegard Nossumeca6f532009-09-18 13:05:45 -07002639 ret = copy_mount_string(dev_name, &kernel_dev);
2640 if (ret < 0)
2641 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642
Vegard Nossumeca6f532009-09-18 13:05:45 -07002643 ret = copy_mount_options(data, &data_page);
2644 if (ret < 0)
2645 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646
Jeff Layton91a27b22012-10-10 15:25:28 -04002647 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
Vegard Nossumeca6f532009-09-18 13:05:45 -07002648 (void *) data_page);
2649
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002651out_data:
2652 kfree(kernel_dev);
2653out_dev:
2654 putname(kernel_dir);
2655out_dir:
2656 kfree(kernel_type);
2657out_type:
2658 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659}
2660
2661/*
Al Viroafac7cb2011-11-23 19:34:49 -05002662 * Return true if path is reachable from root
2663 *
Al Viro48a066e2013-09-29 22:06:07 -04002664 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05002665 */
Al Viro643822b2011-11-24 22:00:28 -05002666bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002667 const struct path *root)
2668{
Al Viro643822b2011-11-24 22:00:28 -05002669 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002670 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002671 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002672 }
Al Viro643822b2011-11-24 22:00:28 -05002673 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002674}
2675
2676int path_is_under(struct path *path1, struct path *path2)
2677{
2678 int res;
Al Viro48a066e2013-09-29 22:06:07 -04002679 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002680 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04002681 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05002682 return res;
2683}
2684EXPORT_SYMBOL(path_is_under);
2685
2686/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 * pivot_root Semantics:
2688 * Moves the root file system of the current process to the directory put_old,
2689 * makes new_root as the new root file system of the current process, and sets
2690 * root/cwd of all processes which had them on the current root to new_root.
2691 *
2692 * Restrictions:
2693 * The new_root and put_old must be directories, and must not be on the
2694 * same file system as the current process root. The put_old must be
2695 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2696 * pointed to by put_old must yield the same directory as new_root. No other
2697 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2698 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002699 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2700 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2701 * in this situation.
2702 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 * Notes:
2704 * - we don't move root/cwd if they are not at the root (reason: if something
2705 * cared enough to change them, it's probably wrong to force them elsewhere)
2706 * - it's okay to pick a root that isn't the root of a file system, e.g.
2707 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2708 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2709 * first.
2710 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002711SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2712 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713{
Al Viro2d8f3032008-07-22 09:59:21 -04002714 struct path new, old, parent_path, root_parent, root;
Al Viro84d17192013-03-15 10:53:28 -04002715 struct mount *new_mnt, *root_mnt, *old_mnt;
2716 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 int error;
2718
Al Viro9b40bc92013-02-22 22:45:42 -05002719 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 return -EPERM;
2721
Al Viro2d8f3032008-07-22 09:59:21 -04002722 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 if (error)
2724 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725
Al Viro2d8f3032008-07-22 09:59:21 -04002726 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 if (error)
2728 goto out1;
2729
Al Viro2d8f3032008-07-22 09:59:21 -04002730 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002731 if (error)
2732 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002734 get_fs_root(current->fs, &root);
Al Viro84d17192013-03-15 10:53:28 -04002735 old_mp = lock_mount(&old);
2736 error = PTR_ERR(old_mp);
2737 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04002738 goto out3;
2739
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002741 new_mnt = real_mount(new.mnt);
2742 root_mnt = real_mount(root.mnt);
Al Viro84d17192013-03-15 10:53:28 -04002743 old_mnt = real_mount(old.mnt);
2744 if (IS_MNT_SHARED(old_mnt) ||
Al Virofc7be132011-11-25 01:05:37 -05002745 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2746 IS_MNT_SHARED(root_mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002747 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05002748 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002749 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002750 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2751 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002753 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002754 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 error = -EBUSY;
Al Viro84d17192013-03-15 10:53:28 -04002756 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04002757 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002759 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002760 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002761 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002762 goto out4; /* not attached */
Al Viro84d17192013-03-15 10:53:28 -04002763 root_mp = root_mnt->mnt_mp;
Al Viro2d8f3032008-07-22 09:59:21 -04002764 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002765 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002766 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002767 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002768 /* make sure we can reach put_old from new_root */
Al Viro84d17192013-03-15 10:53:28 -04002769 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002770 goto out4;
Al Viro84d17192013-03-15 10:53:28 -04002771 root_mp->m_count++; /* pin it so it won't go away */
Al Viro719ea2f2013-09-29 11:24:49 -04002772 lock_mount_hash();
Al Viro419148d2011-11-24 19:41:16 -05002773 detach_mnt(new_mnt, &parent_path);
2774 detach_mnt(root_mnt, &root_parent);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002775 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2776 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2777 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2778 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002779 /* mount old root on put_old */
Al Viro84d17192013-03-15 10:53:28 -04002780 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08002781 /* mount new_root on / */
Al Viro84d17192013-03-15 10:53:28 -04002782 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002783 touch_mnt_namespace(current->nsproxy->mnt_ns);
Al Viro719ea2f2013-09-29 11:24:49 -04002784 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04002785 chroot_fs_refs(&root, &new);
Al Viro84d17192013-03-15 10:53:28 -04002786 put_mountpoint(root_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002788out4:
Al Viro84d17192013-03-15 10:53:28 -04002789 unlock_mount(old_mp);
Al Virob12cea92011-03-18 08:55:38 -04002790 if (!error) {
2791 path_put(&root_parent);
2792 path_put(&parent_path);
2793 }
2794out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002795 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002796out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002797 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002799 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802}
2803
2804static void __init init_mount_tree(void)
2805{
2806 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002807 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002808 struct path root;
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002809 struct file_system_type *type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002811 type = get_fs_type("rootfs");
2812 if (!type)
2813 panic("Can't find rootfs type");
2814 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2815 put_filesystem(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 if (IS_ERR(mnt))
2817 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002818
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002819 ns = create_mnt_ns(mnt);
2820 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002823 init_task.nsproxy->mnt_ns = ns;
2824 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825
Al Virobe08d6d2011-12-06 13:32:36 -05002826 root.mnt = mnt;
2827 root.dentry = mnt->mnt_root;
Jan Blunckac748a02008-02-14 19:34:39 -08002828
2829 set_fs_pwd(current->fs, &root);
2830 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831}
2832
Denis Cheng74bf17c2007-10-16 23:26:30 -07002833void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002835 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002836 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837
Al Viro7d6fec42011-11-23 12:14:10 -05002838 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002839 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840
Al Viro0818bf22014-02-28 13:46:44 -05002841 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04002842 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05002843 mhash_entries, 19,
2844 0,
2845 &m_hash_shift, &m_hash_mask, 0, 0);
2846 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
2847 sizeof(struct hlist_head),
2848 mphash_entries, 19,
2849 0,
2850 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851
Al Viro84d17192013-03-15 10:53:28 -04002852 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 panic("Failed to allocate mount hash table\n");
2854
Al Viro0818bf22014-02-28 13:46:44 -05002855 for (u = 0; u <= m_hash_mask; u++)
Al Viro38129a12014-03-20 21:10:51 -04002856 INIT_HLIST_HEAD(&mount_hashtable[u]);
Al Viro0818bf22014-02-28 13:46:44 -05002857 for (u = 0; u <= mp_hash_mask; u++)
2858 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859
Tejun Heo4b93dc92013-11-28 14:54:43 -05002860 kernfs_init();
2861
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002862 err = sysfs_init();
2863 if (err)
2864 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002865 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002866 fs_kobj = kobject_create_and_add("fs", NULL);
2867 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002868 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869 init_rootfs();
2870 init_mount_tree();
2871}
2872
Trond Myklebust616511d2009-06-22 15:09:13 -04002873void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874{
Al Virod498b252010-02-05 02:21:06 -05002875 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002876 return;
Al Viro7b00ed62013-09-16 21:19:20 -04002877 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07002878 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879}
Al Viro9d412a42011-03-17 22:08:28 -04002880
2881struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2882{
Tim Chen423e0ab2011-07-19 09:32:38 -07002883 struct vfsmount *mnt;
2884 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2885 if (!IS_ERR(mnt)) {
2886 /*
2887 * it is a longterm mount, don't release mnt until
2888 * we unmount before file sys is unregistered
2889 */
Al Virof7a99c52012-06-09 00:59:08 -04002890 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07002891 }
2892 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002893}
2894EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002895
2896void kern_unmount(struct vfsmount *mnt)
2897{
2898 /* release long term mount so mount point can be released */
2899 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04002900 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04002901 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07002902 mntput(mnt);
2903 }
2904}
2905EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002906
2907bool our_mnt(struct vfsmount *mnt)
2908{
Al Viro143c8c92011-11-25 00:46:35 -05002909 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05002910}
Eric W. Biederman8823c072010-03-07 18:49:36 -08002911
Eric W. Biederman31515272013-03-15 01:45:51 -07002912bool current_chrooted(void)
2913{
2914 /* Does the current process have a non-standard root */
2915 struct path ns_root;
2916 struct path fs_root;
2917 bool chrooted;
2918
2919 /* Find the namespace root */
2920 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
2921 ns_root.dentry = ns_root.mnt->mnt_root;
2922 path_get(&ns_root);
2923 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2924 ;
2925
2926 get_fs_root(current->fs, &fs_root);
2927
2928 chrooted = !path_equal(&fs_root, &ns_root);
2929
2930 path_put(&fs_root);
2931 path_put(&ns_root);
2932
2933 return chrooted;
2934}
2935
Eric W. Biedermane51db732013-03-30 19:57:41 -07002936bool fs_fully_visible(struct file_system_type *type)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002937{
2938 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2939 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07002940 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002941
Eric W. Biedermane51db732013-03-30 19:57:41 -07002942 if (unlikely(!ns))
2943 return false;
2944
Al Viro44bb4382013-09-16 21:37:36 -04002945 down_read(&namespace_sem);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002946 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07002947 struct mount *child;
2948 if (mnt->mnt.mnt_sb->s_type != type)
2949 continue;
2950
2951 /* This mount is not fully visible if there are any child mounts
2952 * that cover anything except for empty directories.
2953 */
2954 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2955 struct inode *inode = child->mnt_mountpoint->d_inode;
2956 if (!S_ISDIR(inode->i_mode))
2957 goto next;
Eric W. Biederman41301ae2013-11-14 21:22:25 -08002958 if (inode->i_nlink > 2)
Eric W. Biedermane51db732013-03-30 19:57:41 -07002959 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002960 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07002961 visible = true;
2962 goto found;
2963 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002964 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07002965found:
Al Viro44bb4382013-09-16 21:37:36 -04002966 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07002967 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07002968}
2969
Eric W. Biederman8823c072010-03-07 18:49:36 -08002970static void *mntns_get(struct task_struct *task)
2971{
2972 struct mnt_namespace *ns = NULL;
2973 struct nsproxy *nsproxy;
2974
2975 rcu_read_lock();
2976 nsproxy = task_nsproxy(task);
2977 if (nsproxy) {
2978 ns = nsproxy->mnt_ns;
2979 get_mnt_ns(ns);
2980 }
2981 rcu_read_unlock();
2982
2983 return ns;
2984}
2985
2986static void mntns_put(void *ns)
2987{
2988 put_mnt_ns(ns);
2989}
2990
2991static int mntns_install(struct nsproxy *nsproxy, void *ns)
2992{
2993 struct fs_struct *fs = current->fs;
2994 struct mnt_namespace *mnt_ns = ns;
2995 struct path root;
2996
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002997 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -07002998 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
2999 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08003000 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08003001
3002 if (fs->users != 1)
3003 return -EINVAL;
3004
3005 get_mnt_ns(mnt_ns);
3006 put_mnt_ns(nsproxy->mnt_ns);
3007 nsproxy->mnt_ns = mnt_ns;
3008
3009 /* Find the root */
3010 root.mnt = &mnt_ns->root->mnt;
3011 root.dentry = mnt_ns->root->mnt.mnt_root;
3012 path_get(&root);
3013 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3014 ;
3015
3016 /* Update the pwd and root */
3017 set_fs_pwd(fs, &root);
3018 set_fs_root(fs, &root);
3019
3020 path_put(&root);
3021 return 0;
3022}
3023
Eric W. Biederman98f842e2011-06-15 10:21:48 -07003024static unsigned int mntns_inum(void *ns)
3025{
3026 struct mnt_namespace *mnt_ns = ns;
3027 return mnt_ns->proc_inum;
3028}
3029
Eric W. Biederman8823c072010-03-07 18:49:36 -08003030const struct proc_ns_operations mntns_operations = {
3031 .name = "mnt",
3032 .type = CLONE_NEWNS,
3033 .get = mntns_get,
3034 .put = mntns_put,
3035 .install = mntns_install,
Eric W. Biederman98f842e2011-06-15 10:21:48 -07003036 .inum = mntns_inum,
Eric W. Biederman8823c072010-03-07 18:49:36 -08003037};